LN Ch1 16 Incl Appendix and Two Short Notes in One File Vaekstmaster2015
Uploaded by
Ariful AminLN Ch1 16 Incl Appendix and Two Short Notes in One File Vaekstmaster2015
Uploaded by
Ariful AminLecture Notes in Economic Growth
Christian Groth
May 18, 2015
ii
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
Contents
Preface ix
1 Introduction to economic growth 1
1.1 The field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1.1 Economic growth theory . . . . . . . . . . . . . . . . . 1
1.1.2 Some long-run data . . . . . . . . . . . . . . . . . . . . 3
1.2 Calculation of the average growth rate . . . . . . . . . . . . . 4
1.2.1 Discrete compounding . . . . . . . . . . . . . . . . . . 4
1.2.2 Continuous compounding . . . . . . . . . . . . . . . . 6
1.2.3 Doubling time . . . . . . . . . . . . . . . . . . . . . . . 7
1.3 Some stylized facts of economic growth . . . . . . . . . . . . . 7
1.3.1 The Kuznets facts . . . . . . . . . . . . . . . . . . . . 7
1.3.2 Kaldor’s stylized facts . . . . . . . . . . . . . . . . . . 9
1.4 Concepts of income convergence . . . . . . . . . . . . . . . . . 10
1.4.1 convergence vs. convergence . . . . . . . . . . . . . 10
1.4.2 Measures of dispersion . . . . . . . . . . . . . . . . . . 11
1.4.3 Weighting by size of population . . . . . . . . . . . . . 13
1.4.4 Unconditional vs. conditional convergence . . . . . . . 14
1.4.5 A bird’s-eye view of the data . . . . . . . . . . . . . . . 17
1.4.6 Other convergence concepts . . . . . . . . . . . . . . . 17
1.5 Literature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2 Review of technology 23
2.1 The production technology . . . . . . . . . . . . . . . . . . . . 23
2.1.1 A neoclassical production function . . . . . . . . . . . 24
2.1.2 Returns to scale . . . . . . . . . . . . . . . . . . . . . . 27
2.1.3 Properties of the production function under CRS . . . 32
2.2 Technological change . . . . . . . . . . . . . . . . . . . . . . . 35
2.3 The concepts of a representative firm and an aggregate pro-
duction function . . . . . . . . . . . . . . . . . . . . . . . . . . 39
2.4 Long-run vs. short-run production functions* . . . . . . . . . 42
iii
iv CONTENTS
2.5 Literature notes . . . . . . . . . . . . . . . . . . . . . . . . . . 44
2.6 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3 Continuous time analysis 47
3.1 The transition from discrete time to continuous time . . . . . 47
3.1.1 Multiple compounding per year . . . . . . . . . . . . . 47
3.1.2 Compound interest and discounting . . . . . . . . . . . 49
3.2 The allowed range for parameter values . . . . . . . . . . . . . 50
3.3 Stocks and flows . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.4 The choice between discrete and continuous time formulation . 53
3.5 Appendix A: Growth arithmetic in continuous time . . . . . . 54
3.6 Appendix B: Solution formulas for linear differential equations
of first order . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
4 Skill-biased technical change. Balanced growth theorems 57
4.1 The rising skill premium . . . . . . . . . . . . . . . . . . . . . 58
4.1.1 Skill-biased technical change in the sense of Hicks: An
example . . . . . . . . . . . . . . . . . . . . . . . . . . 58
4.1.2 Capital-skill complementarity . . . . . . . . . . . . . . 59
4.2 Balanced growth and constancy of key ratios . . . . . . . . . . 60
4.2.1 The concepts of steady state and balanced growth . . . 61
4.2.2 A general result about balanced growth . . . . . . . . . 62
4.3 The crucial role of Harrod-neutrality . . . . . . . . . . . . . . 64
4.4 Harrod-neutrality and the functional income distribution . . . 68
4.5 What if technological change is embodied? . . . . . . . . . . . 70
4.6 Concluding remarks . . . . . . . . . . . . . . . . . . . . . . . . 72
4.7 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
5 Growth accounting and the concept of TFP: Some warnings 75
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
5.2 TFP level and TFP growth . . . . . . . . . . . . . . . . . . . 75
5.2.1 TFP growth . . . . . . . . . . . . . . . . . . . . . . . . 77
5.2.2 The TFP level . . . . . . . . . . . . . . . . . . . . . . . 78
5.3 The case of Hicks-neutrality* . . . . . . . . . . . . . . . . . . 79
5.4 The case of absence of Hicks-neutrality* . . . . . . . . . . . . 80
5.5 Three warnings . . . . . . . . . . . . . . . . . . . . . . . . . . 82
5.6 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
6 Transitional dynamics. Barro-style growth regressions 87
6.1 Point of departure: the Solow model . . . . . . . . . . . . . . 87
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
CONTENTS v
6.2 Do poor countries tend to approach their steady state from
below? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
6.3 Convergence speed and adjustment time . . . . . . . . . . . . 91
6.3.1 Convergence speed for ̃() . . . . . . . . . . . . . . . . 92
6.3.2 Convergence speed for log ̃() . . . . . . . . . . . . . . 94
6.3.3 Convergence speed for () ∗ () . . . . . . . . . . . . 94
6.3.4 Adjustment time . . . . . . . . . . . . . . . . . . . . . 96
6.4 Barro-style growth regressions . . . . . . . . . . . . . . . . . . 98
6.5 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
7 Michael Kremer’s population-breeds-ideas model 103
7.1 The model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
7.2 Law of motion . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
7.3 The inevitable ending of the Malthusian regime . . . . . . . . 105
7.4 Closing remarks . . . . . . . . . . . . . . . . . . . . . . . . . . 106
7.5 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
7.6 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
8 Choice of social discount rate 109
8.1 Basic distinctions relating to discounting . . . . . . . . . . . . 110
8.1.1 The unit of account . . . . . . . . . . . . . . . . . . . . 110
8.1.2 The economic context . . . . . . . . . . . . . . . . . . 113
8.2 Criteria for choice of a social discount rate . . . . . . . . . . . 114
8.3 Optimal capital accumulation . . . . . . . . . . . . . . . . . . 117
8.3.1 The setting . . . . . . . . . . . . . . . . . . . . . . . . 117
8.3.2 First-order conditions and their economic interpretation 119
8.3.3 The social consumption discount rate . . . . . . . . . . 120
8.4 The climate change problem from an economic point of view . 124
8.4.1 Damage projections . . . . . . . . . . . . . . . . . . . . 124
8.4.2 Uncertainty, risk aversion, and the certainty-equivalent
loss . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
8.4.3 Comparing benefits and costs . . . . . . . . . . . . . . 129
8.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
8.6 Appendix: A closer look at Arrow’s estimate of the certainty
loss . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
8.7 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
9 Human capital, learning technology, and the Mincer equa-
tion 137
9.1 Conceptual issues . . . . . . . . . . . . . . . . . . . . . . . . . 137
9.1.1 Macroeconomic approaches to human capital . . . . . . 138
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
vi CONTENTS
9.1.2 Human capital and the efficiency of labor . . . . . . . . 139
9.2 The life-cycle approach to human capital . . . . . . . . . . . . 142
9.3 Choosing length of education . . . . . . . . . . . . . . . . . . 146
9.3.1 Human wealth . . . . . . . . . . . . . . . . . . . . . . . 146
9.3.2 A perfect credit and life annuity market . . . . . . . . 148
9.3.3 Maximizing human wealth . . . . . . . . . . . . . . . . 149
9.4 Explaining the Mincer equation . . . . . . . . . . . . . . . . . 152
9.5 Some empirics . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
9.6 Literature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
10 Human capital and knowledge creation in a growing econ-
omy 161
10.1 The model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
10.2 Productivity growth along a BGP with R&D . . . . . . . . . . 162
10.2.1 Balanced growth with R&D . . . . . . . . . . . . . . . 163
10.2.2 A precondition for sustained productivity growth when
̄ = 0: population growth . . . . . . . . . . . . . . . . 165
10.2.3 The concept of endogenous growth . . . . . . . . . . . 167
10.3 Permanent level effects . . . . . . . . . . . . . . . . . . . . . . 167
10.4 The case of no R&D . . . . . . . . . . . . . . . . . . . . . . . 168
10.5 Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
10.5.1 The case = 0 . . . . . . . . . . . . . . . . . . . . . . 169
10.5.2 The case of rising life expectancy . . . . . . . . . . . . 170
10.6 Concluding remarks . . . . . . . . . . . . . . . . . . . . . . . . 172
10.7 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
11 AK and reduced-form AK models. Consumption taxation. 175
11.1 General equilibrium dynamics in the simple AK model . . . . 175
11.2 Reduced-form AK models . . . . . . . . . . . . . . . . . . . . 178
11.3 On consumption taxation . . . . . . . . . . . . . . . . . . . . 178
12 Learning by investing: two versions 183
12.1 The common framework . . . . . . . . . . . . . . . . . . . . . 184
12.1.1 The individual firm . . . . . . . . . . . . . . . . . . . . 185
12.1.2 The household . . . . . . . . . . . . . . . . . . . . . . . 186
12.1.3 Equilibrium in factor markets . . . . . . . . . . . . . . 186
12.2 The arrow case: 1 . . . . . . . . . . . . . . . . . . . . . . 187
12.2.1 Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . 187
12.2.2 Two types of endogenous growth . . . . . . . . . . . . 192
12.3 Romer’s limiting case: = 1 = 0 . . . . . . . . . . . . . . . 193
12.3.1 Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . 194
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
CONTENTS vii
12.3.2 Economic policy in the Romer case . . . . . . . . . . . 197
12.4 Appendix: The golden-rule capital intensity in the Arrow case 201
13 Perspectives on learning by doing and learning by investing205
13.1 Learning by doing, learning by using, learning by watching . . 206
13.1.1 The case: 1 in (13.3) . . . . . . . . . . . . . . . . . 208
13.1.2 The case = 1 in (13.3) . . . . . . . . . . . . . . . . . 210
13.2 Disembodied learning by investing . . . . . . . . . . . . . . . . 211
13.2.1 The Arrow case: 1 and ≥ 0 . . . . . . . . . . . . 212
13.2.2 The Romer case: = 1 and = 0 . . . . . . . . . . . . 212
13.2.3 The size of the learning parameter . . . . . . . . . . . 213
13.3 Disembodied vs. embodied technical change . . . . . . . . . . 217
13.3.1 Disembodied technical change . . . . . . . . . . . . . . 217
13.3.2 Embodied technical change . . . . . . . . . . . . . . . 217
13.3.3 Embodied technical change and learning by investing . 219
13.4 Static comparative advantage vs. dynamics of learning by doing*223
13.4.1 A simple two-sector learning-by-doing model . . . . . . 223
13.4.2 A more robust specification . . . . . . . . . . . . . . . 225
13.4.3 Resource curse? . . . . . . . . . . . . . . . . . . . . . . 226
13.5 Robustness issues and scale effects . . . . . . . . . . . . . . . . 227
13.5.1 On terminology . . . . . . . . . . . . . . . . . . . . . . 227
13.5.2 Robustness of simple endogenous growth models . . . . 230
13.5.3 Weak and strong scale effects . . . . . . . . . . . . . . 232
13.5.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . 234
13.6 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 236
13.7 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 239
14 The lab-equipment model 243
14.1 Overview of the economy . . . . . . . . . . . . . . . . . . . . . 244
14.1.1 The sectorial production functions . . . . . . . . . . . . 244
14.1.2 National income accounting . . . . . . . . . . . . . . . 247
14.1.3 The potential for sustained productivity growth . . . . 248
14.2 Households and the labor market . . . . . . . . . . . . . . . . 248
14.3 Firms’ behavior . . . . . . . . . . . . . . . . . . . . . . . . . . 249
14.3.1 The competitive producers of basic goods . . . . . . . . 250
14.3.2 The monopolist suppliers of intermediate goods . . . . 250
14.3.3 R&D firms . . . . . . . . . . . . . . . . . . . . . . . . . 252
14.3.4 Equilibrium in the loan market . . . . . . . . . . . . . 256
14.4 General equilibrium of an economy satisfying (A1) . . . . . . . 258
14.4.1 The balanced growth path . . . . . . . . . . . . . . . . 259
14.4.2 Comparative analysis . . . . . . . . . . . . . . . . . . . 260
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
viii CONTENTS
15 Stochastic erosion of innovators’ monopoly power 263
15.1 The three production sectors . . . . . . . . . . . . . . . . . . . 264
15.2 Temporary monopoly . . . . . . . . . . . . . . . . . . . . . . . 265
15.3 The aggregate production function in equilibrium . . . . . . . 267
15.4 The no-arbitrage condition under uncertainty* . . . . . . . . . 268
15.5 The equilibrium rate of return when R&D is active . . . . . . 271
15.6 Transitional dynamics* . . . . . . . . . . . . . . . . . . . . . . 272
15.7 Long-run growth . . . . . . . . . . . . . . . . . . . . . . . . . 274
15.8 Economic policy . . . . . . . . . . . . . . . . . . . . . . . . . . 275
15.9 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 278
15.10References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 279
16 Natural resources and economic growth 281
16.1 Classification of means of production . . . . . . . . . . . . . . 281
16.2 The notion of sustainable development . . . . . . . . . . . . . 283
16.3 Renewable resources . . . . . . . . . . . . . . . . . . . . . . . 286
16.4 Non-renewable resources: The DHSS model . . . . . . . . . . 291
16.4.1 Input substitution . . . . . . . . . . . . . . . . . . . . 292
16.4.2 Technical progress . . . . . . . . . . . . . . . . . . . . 294
16.4.3 Increasing returns to scale . . . . . . . . . . . . . . . . 297
16.4.4 Summary on the DHSS model . . . . . . . . . . . . . 298
16.4.5 An extended DHSS model . . . . . . . . . . . . . . . . 298
16.5 Endogenous technical change . . . . . . . . . . . . . . . . . . . 299
16.5.1 A two-sector R&D-based model . . . . . . . . . . . . . 299
16.5.2 Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . 300
16.6 Natural resources and the issue of limits to economic growth . 308
16.7 Bibliographic notes . . . . . . . . . . . . . . . . . . . . . . . . 309
16.8 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 309
16.8.1 A. The CES function . . . . . . . . . . . . . . . . . . . 309
16.8.2 B. Balanced growth with an essential non-renewable
resource . . . . . . . . . . . . . . . . . . . . . . . . . . 315
16.9 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 318
A Appendix: Errata 325
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
Preface
This is a collection of earlier separate lecture notes in Economic Growth.
The notes have been used in recent years in the course Economic Growth
within the Master’s Program in Economics at the Department of Economics,
University of Copenhagen.
Compared with the earlier versions of the lecture notes some chapters
have been extended and in some cases divided into several chapters. In
addition, discovered typos and similar have been corrected. In some of the
chapters a terminal list of references is at present lacking.
The lecture notes are in no way intended as a substitute for the currently
applied textbook: D. Acemoglu, Introduction to Modern Economic Growth,
Princeton University Press, 2009. The lecture notes are meant to be read
along with the textbook. Some parts of the lecture notes are alternative
presentations of stuff also covered in the textbook, while many other parts
are complementary in the sense of presenting additional material. Sections
marked by an asterisk, *, are cursory reading.
For constructive criticism I thank Niklas Brønager, class instructor since
2012, and plenty of earlier students. No doubt, obscurities remain. Hence, I
very much welcome comments and suggestions of any kind relating to these
lecture notes.
February 2015
Christian Groth
ix
x PREFACE
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
Chapter 1
Introduction to economic
growth
This introductory lecture note is a refresher on basic concepts.
Section 1.1 defines Economic Growth as a field of economics. In Section
1.2 formulas for calculation of compound average growth rates in discrete and
continuous time are presented. Section 1.3 briefly presents two sets of what
is by many considered as “stylized facts” about economic growth. Finally,
Section 1.4 discusses, in an informal way, the different concepts of cross-
country income convergence. In his introductory Chapter 1, §1.5, Acemoglu1
briefly touches upon these concepts.
1.1 The field
Economic growth analysis is the study of what factors and mechanisms deter-
mine the time path of productivity (a simple index of productivity is output
per unit of labor). The focus is on
• productivity levels and
• productivity growth.
1.1.1 Economic growth theory
Economic growth theory endogenizes productivity growth via considering
human capital accumulation (formal education as well as learning-by-doing)
1
Throughout these lecture notes, “Acemoglu” refers to Daron Acemoglu, Introduction
to Modern Economic Growth, Princeton University Press: Oxford, 2009.
1
2 CHAPTER 1. INTRODUCTION TO ECONOMIC GROWTH
and endogenous research and development. Also the conditioning role of
geography and juridical, political, and cultural institutions is taken into ac-
count.
Although for practical reasons, economic growth theory is often stated
in terms of national income and product account variables like per capita
GDP, the term “economic growth” may be interpreted as referring to some-
thing deeper. We could think of “economic growth” as the widening of the
opportunities of human beings to lead freer and more worthwhile lives.
To make our complex economic environment accessible for theoretical
analysis we use economic models. What is an economic model? It is a way
of organizing one’s thoughts about the economic functioning of a society. A
more specific answer is to define an economic model as a conceptual struc-
ture based on a set of mathematically formulated assumptions which have
an economic interpretation and from which empirically testable predictions
can be derived. In particular, an economic growth model is an economic
model concerned with productivity issues. The union of connected and non-
contradictory models dealing with economic growth and the theorems derived
from these constitute an economic growth theory. Occasionally, intense con-
troversies about the validity of different growth theories take place.
The terms “New Growth Theory” and “endogenous growth theory” re-
fer to theory and models which attempt at explaining sustained per capita
growth as an outcome of internal mechanisms in the model rather than just
a reflection of exogenous technical progress as in “Old Growth Theory”.
Among the themes addressed in this course are:
• How is the world income distribution evolving?
• Why do living standards differ so much across countries and regions?
Why are some countries 50 times richer than others?
• Why do per capita growth rates differ over long periods?
• What are the roles of human capital and technology innovation in eco-
nomic growth? Getting the questions right.
• Catching-up and increased speed of communication and technology dif-
fusion.
• Economic growth, natural resources, and the environment (including
the climate). What are the limits to growth?
• Policies to ignite and sustain productivity growth.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
1.1. The field 3
• The prospects of growth in the future.
The course concentrates on mechanisms behind the evolution of produc-
tivity in the industrialized world. We study these mechanisms as integral
parts of dynamic general equilibrium models. The exam is a test of the ex-
tent to which the student has acquired understanding of these models, is
able to evaluate them, from both a theoretical and empirical perspective,
and is able to use them to analyze specific economic questions. The course
is calculus intensive.
1.1.2 Some long-run data
Let denote real GDP (per year) and let be population size. Then
is GDP per capita. Further, let denote the average (compound) growth
rate of per year since 1870 and let denote the average (compound)
growth rate of per year since 1870. Table 1.1 gives these growth rates
for four countries.
Denmark 2,67 1,87
UK 1,96 1,46
USA 3,40 1,89
Japan 3,54 2,54
Table 1.1: Average annual growth rate of GDP and GDP per capita in percent,
1870—2006. Discrete compounding. Source: Maddison, A: The World Economy:
Historical Statistics, 2006, Table 1b, 1c and 5c.
Figure 1.1 displays the time path of annual GDP and GDP per capita in
Denmark 1870-2006 along with regression lines estimated by OLS (logarith-
mic scale on the vertical axis). Figure 1.2 displays the time path of GDP per
capita in UK, USA, and Japan 1870-2006. In both figures the average annual
growth rates are reported. In spite of being based on exactly the same data
as Table 1.1, the numbers are slightly different. Indeed, the numbers in the
figures are slightly lower than those in the table. The reason is that discrete
compounding is used in Table 1.1 while continuous compounding is used in
the two figures. These two alternative methods of calculation are explained
in the next section.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
4 CHAPTER 1. INTRODUCTION TO ECONOMIC GROWTH
Figure 1.1: GDP and GDP per capita (1990 International Geary-Khamis dollars)
in Denmark, 1870-2006. Source: Maddison, A. (2009). Statistics on World Popu-
lation, GDP and Per Capita GDP, 1-2006 AD, www.ggdc.net/maddison.
1.2 Calculation of the average growth rate
1.2.1 Discrete compounding
Let denote aggregate labor productivity, i.e., ≡ where is employ-
ment. The average growth rate of from period 0 to period with discrete
compounding, is that which satisfies
= 0 (1 + ) = 1 2 , or (1.1)
1 + = ( )1 i.e.,
0
= ( )1 − 1 (1.2)
0
“Compounding” means adding the one-period “net return” to the “principal”
before adding next period’s “net return” (like with interest on interest, also
called “compound interest”). The growth factor 1 + will generally be
less than the arithmetic average of the period-by-period growth factors. To
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
1.2. Calculation of the average growth rate 5
Figure 1.2: GDP per capita (1990 International Geary-Khamis dollars) in UK,
USA and Japan, 1870-2006. Source: Maddison, A. (2009). Statistics on World
Population, GDP and Per Capita GDP, 1-2006 AD, www.ggdc.net/maddison.
underline this difference, 1 + is sometimes called the “compound average
growth factor” or the “geometric average growth factor” and itself then
called the “compound average growth rate” or the “geometric average growth
rate”
Using a pocket calculator, the following steps in the calculation of may
be convenient. Take logs on both sides of (1.1) to get
ln = ln(1 + ) ⇒
0
ln 0
ln(1 + ) = ⇒ (1.3)
ln 0
= antilog( ) − 1. (1.4)
Note that in the formulas (1.2) and (1.4) equals the number of periods
minus 1.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
6 CHAPTER 1. INTRODUCTION TO ECONOMIC GROWTH
1.2.2 Continuous compounding
The average growth rate of , with continuous compounding, is that which
satisfies
= 0 (1.5)
where denotes the Euler number, i.e., the base of the natural logarithm.2
Solving for gives
ln 0 ln − ln 0
= = (1.6)
The first formula in (1.6) is convenient for calculation with a pocket calcula-
tor, whereas the second formula is perhaps closer to intuition. Another name
for is the “exponential average growth rate”.
Again, for discrete time data the in the formula equals the number of
periods minus 1.
Comparing with (1.3) we see that = ln(1 + ) for 6= 0 Yet, by
a first-order Taylor approximation of ln(1 + ) about = 0 we have
= ln(1 + ) ≈ for “small”. (1.7)
For a given data set the calculated from (1.2) will be slightly above the
calculated from (1.6), cf. the mentioned difference between the growth rates
in Table 1.1 and those in Figure 1.1 and Figure 1.2. The reason is that a given
growth force is more powerful when compounding is continuous rather than
discrete. Anyway, the difference between and is usually unimportant.
If for example refers to the annual GDP growth rate, it will be a small
number, and the difference between and immaterial. For example, to
= 0040 corresponds ≈ 0039 Even if = 010, the corresponding is
00953. But if stands for the inflation rate and there is high inflation, the
difference between and will be substantial. During hyperinflation the
monthly inflation rate may be, say, = 100%, but the corresponding will
be only 69%.
Which method, discrete or continuous compounding, is preferable? To
some extent it is a matter of taste or convenience. In period analysis discrete
compounding is most common and in continuous time analysis continuous
compounding is most common.
For calculation with a pocket calculator the continuous compounding for-
mula, (1.6), is slightly easier to use than the discrete compounding formulas,
whether (1.2) or (1.4).
2
Unless otherwise specified, whenever we write ln or log the natural logarithm is
understood.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
1.3. Some stylized facts of economic growth 7
To avoid too much sensitiveness to the initial and terminal observations,
which may involve measurement error or depend on the state of the business
cycle, one can use an OLS approach to the trend coefficient, in the following
regression:
ln = + +
This is in fact what is done in Fig. 1.1.
1.2.3 Doubling time
How long time does it take for to double if the growth rate with discrete
compounding is ? Knowing we rewrite the formula (1.3):
ln 0 ln 2 06931
= = ≈
ln(1 + ) ln(1 + ) ln(1 + )
With = 00187 cf. Table 1.1, we find
≈ 374 years,
meaning that productivity doubles every 374 years.
How long time does it take for to double if the growth rate with con-
tinuous compounding is ? The answer is based on rewriting the formula
(1.6):
ln 0 ln 2 06931
= = ≈
Maintaining the value 00187 also for we find
06931
≈ ≈ 371 years.
00187
Again, with a pocket calculator the continuous compounding formula is
slightly easier to use. With a lower say = 001 we find doubling time
equal to 691 years. With = 007 (think of China since the early 1980’s),
doubling time is about 10 years! Owing to the compounding, exponential
growth is extremely powerful.
1.3 Some stylized facts of economic growth
1.3.1 The Kuznets facts
A well-known characteristic of modern economic growth is structural change:
unbalanced sectorial growth. There is a massive reallocation of labor from
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
8 CHAPTER 1. INTRODUCTION TO ECONOMIC GROWTH
Figure 1.3: The Kuznets facts. Source: Kongsamut et al., Beyond Balanced
Growth, Review of Economic Studies, vol. 68, Oct. 2001, 869-82.
agriculture into industry (manufacturing, construction, and mining) and fur-
ther into services (including transport and communication). The shares of
total consumption expenditure going to these three sectors have moved sim-
ilarly. Differences in the demand elasticities with respect to income seem the
main explanation. These observations are often referred to as the Kuznets
facts (after Simon Kuznets, 1901-85, see, e.g., Kuznets 1957).
The two graphs in Figure 1.3 illustrate the Kuznets facts.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
1.3. Some stylized facts of economic growth 9
1.3.2 Kaldor’s stylized facts
Surprisingly, in spite of the Kuznets facts, the evolution at the aggregate level
in developed countries is by many economists seen as roughly described by
what is called Kaldor’s “stylized facts” (after the Hungarian-British econo-
mist Nicholas Kaldor, 1908-1986, see, e.g., Kaldor 1957, 1961)3 :
1. Real output per man-hour grows at a more or less constant rate
over fairly long periods of time. (Of course, there are short-run fluctuations
superposed around this trend.)
2. The stock of physical capital per man-hour grows at a more or less
constant rate over fairly long periods of time.
3. The ratio of output to capital shows no systematic trend.
4. The rate of return to capital shows no systematic trend.
5. The income shares of labor and capital (in the national account-
ing sense, i.e., including land and other natural resources), respectively, are
nearly constant.
6. The growth rate of output per man-hour differs substantially across
countries.
These claimed regularities do certainly not fit all developed countries
equally well. Although Solow’s growth model (Solow, 1956) can be seen as the
first successful attempt at building a model consistent with Kaldor’s “stylized
facts”, Solow once remarked about them: “There is no doubt that they are
stylized, though it is possible to question whether they are facts” (Solow,
1970). Yet, for instance a relatively recent study by Attfield and Temple
(2010) of US and UK data since the Second World War is not unfavorable
to Kaldor’s “facts”. The sixth Kaldor fact is, of course, generally accepted
as a well documented observation (a nice summary is contained in Pritchett,
1997).
Kaldor also proposed hypotheses about the links between growth in the
different sectors (see, e.g., Kaldor 1967):
a. Productivity growth in the manufacturing and construction sec-
tors is enhanced by output growth in these sectors (this is also known as
Verdoorn’s Law). Increasing returns to scale and learning by doing are the
main factors behind this.
b. Productivity growth in agriculture and services is enhanced by out-
put growth in the manufacturing and construction sectors.
3
Kaldor presented his six regularities as “a stylised view of the facts”.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
10 CHAPTER 1. INTRODUCTION TO ECONOMIC GROWTH
1.4 Concepts of income convergence
The two most popular across-country income convergence concepts are “
convergence” and “ convergence”.
1.4.1 convergence vs. convergence
Definition 1 We say that convergence occurs for a given selection of coun-
tries if there is a tendency for the poor (those with low income per capita or
low output per worker) to subsequently grow faster than the rich.
By “grow faster” is meant that the growth rate of per capita income (or
per worker output) is systematically higher.
In many contexts, a more appropriate convergence concept is the follow-
ing:
Definition 2 We say that convergence, with respect to a given measure of
dispersion, occurs for a given collection of countries if this measure of disper-
sion, applied to income per capita or output per worker across the countries,
declines systematically over time. On the other hand, divergence occurs, if
the dispersion increases systematically over time.
The reason that convergence must be considered the more appropri-
ate concept is the following. In the end, it is the question of increasing
or decreasing dispersion across countries that we are interested in. From a
superficial point of view one might think that convergence implies decreas-
ing dispersion and vice versa, so that convergence and convergence are
more or less equivalent concepts. But since the world is not deterministic,
but stochastic, this is not true. Indeed, convergence is only a necessary,
not a sufficient condition for convergence. This is because over time some
reshuffling among the countries is always taking place, and this implies that
there will always be some extreme countries (those initially far away from
the mean) that move closer to the mean, thus creating a negative correla-
tion between initial level and subsequent growth, in spite of equally many
countries moving from a middle position toward one of the extremes.4 In
this way convergence may be observed at the same time as there is no
4
As an intuitive analogy, think of the ordinal rankings of the sports teams in a league.
The dispersion of rankings is constant by definition. Yet, no doubt there will allways be
some tendency for weak teams to rebound toward the mean and of champions to revert
to mediocrity. (This example is taken from the first edition of Barro and Sala-i-Martin,
Economic Growth, 1995; I do not know why, but the example was deleted in the second
edition from 2004.)
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
1.4. Concepts of income convergence 11
convergence; the mere presence of random measurement errors implies a bias
in this direction because a growth rate depends negatively on the initial mea-
surement and positively on the later measurement. In fact, convergence
may be consistent with divergence (for a formal proof of this claim, see
Barro and Sala-i-Martin, 2004, pp. 50-51 and 462 ff.; see also Valdés, 1999,
p. 49-50, and Romer, 2001, p. 32-34).
Hence, it is wrong to conclude from convergence (poor countries tend
to grow faster than rich ones) to convergence (reduced dispersion of per
capita income) without any further investigation. The mistake is called “re-
gression towards the mean” or “Galton’s fallacy”. Francis Galton was an
anthropologist (and a cousin of Darwin), who in the late nineteenth century
observed that tall fathers tended to have not as tall sons and small fathers
tended to have taller sons. From this he falsely concluded that there was
a tendency to averaging out of the differences in height in the population.
Indeed, being a true aristocrat, Galton found this tendency pitiable. But
since his conclusion was mistaken, he did not really have to worry.
Since convergence comes closer to what we are ultimately looking for,
from now, when we speak of just “income convergence”, convergence is
understood.
In the above definitions of convergence and convergence, respectively,
we were vague as to what kind of selection of countries is considered. In
principle we would like it to be a representative sample of the “population”
of countries that we are interested in. The population could be all countries
in the world. Or it could be the countries that a century ago had obtained a
certain level of development.
One should be aware that historical GDP data are constructed retrospec-
tively. Long time series data have only been constructed for those countries
that became relatively rich during the after-WWII period. Thus, if we as our
sample select the countries for which long data series exist, what is known as
selection bias is involved which generates a spurious convergence. A country
which was poor a century ago will only appear in the sample if it grew rapidly
over the next 100 years. A country which was relatively rich a century ago
will appear in the sample unconditionally. This selection bias problem was
pointed out by DeLong (1988) in a criticism of widespread false interpreta-
tions of Maddison’s long data series (Maddison 1982).
1.4.2 Measures of dispersion
Our next problem is: what measure of dispersion is to be used as a useful
descriptive statistics for convergence? Here there are different possibilities.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
12 CHAPTER 1. INTRODUCTION TO ECONOMIC GROWTH
To be precise about this we need some notation. Let
≡ and
≡
where = real GDP, = employment, and = population. If the focus
is on living standards, is the relevant variable.5 But if the focus is on
(labor) productivity, it is that is relevant. Since most growth models
focus on rather than let os take as our example.
One might think that the standard deviation of could be a relevant
measure of dispersion when discussing whether convergence is present or
not. The standard deviation of across countries in a given year is
v
u
u1 X
≡ t ( − ̄)2 (1.8)
=1
where P
̄ ≡ (1.9)
i.e., ̄ is the average output per worker. However, if this measure were used,
it would be hard to find any group of countries for which there is income
convergence. This is because tends to grow over time for most countries,
and then there is an inherent tendency for the variance also to grow; hence
also the square root of the variance, tends to grow. Indeed, suppose that
for all countries, is doubled from time 1 to time 2 Then, automatically,
is also doubled. But hardly anyone would interpret this as an increase in
the income inequality across the countries.
Hence, it is more adequate to look at the standard deviation of relative
income levels: s
1 X
̄ ≡ ( − 1)2 (1.10)
̄
This measure is the same as what is called the coefficient of variation,
usually defined as
≡ (1.11)
̄
5
Or perhaps better, where ≡ ≡ − − Here, denotes net
interest payments on foreign debt and denotes net labor income of foreign workers in
the country.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
1.4. Concepts of income convergence 13
that is, the standard deviation of standardized by the mean. That the two
measures are identical can be seen in this way:
q P
1 2 s s
( − ̄) 1 X − ̄ 2 1 X
≡ = ( ) = ( − 1)2 ≡ ̄
̄ ̄ ̄ ̄
The point is that the coefficient of variation is “scale free”, which the standard
deviation itself is not.
Instead of the coefficient of variation, another scale free measure is often
used, namely the standard deviation of ln , i.e.,
s
1X
ln ≡ (ln − ln ∗ )2 (1.12)
where P
∗ ln
ln ≡ (1.13)
√
Note that ∗ is the geometric average, i.e., ∗ ≡ 1 2 · · · Now, by a
first-order Taylor approximation of ln around = ̄, we have
1
ln ≈ ln ̄ + ( − ̄)
̄
Hence, as a very rough approximation we have ln ≈ ̄ = though
this approximation can be quite poor (cf. Dalgaard and Vastrup, 2001).
It may be possible, however, to defend the use of ln in its own right to
the extent that tends to be approximately lognormally distributed across
countries.
Yet another possible measure of income dispersion across countries is the
Gini index (see for example Cowell, 1995).
1.4.3 Weighting by size of population
Another important issue is whether the applied dispersion measure is based
on a weighting of the countries by size of population. For the world as a
whole, when no weighting by size of population is used, then there is a slight
tendency to income divergence according to the ln criterion (Acemoglu,
2009, p. 4), where is per capita income (≡ ). As seen by Fig. 4 below,
this tendency is not so clear according to the criterion. Anyway, when
there is weighting by size of population, then in the last twenty years there
has been a tendency to income convergence at the global level (Sala-i-Martin
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
14 CHAPTER 1. INTRODUCTION TO ECONOMIC GROWTH
2006; Acemoglu, 2009, p. 6). With weighting by size of population (1.12) is
modified to
sX
ln ≡ (ln − ln ∗ )2
where
X
= and ln ∗ ≡ ln
1.4.4 Unconditional vs. conditional convergence
Yet another distinction in the study of income convergence is that between
unconditional (or absolute) and conditional convergence. We say that a
large heterogeneous group of countries (say the countries in the world) show
unconditional income convergence if income convergence occurs for the whole
group without conditioning on specific characteristics of the countries. If
income convergence occurs only for a subgroup of the countries, namely those
countries that in advance share the same “structural characteristics”, then
we say there is conditional income convergence. As noted earlier, when we
speak of just income “convergence”, income “ convergence” is understood.
If in a given context there might be doubt, one should of course be explicit
and speak of unconditional or conditional convergence. Similarly, if the
focus for some reason is on convergence, we should distinguish between
unconditional and conditional convergence.
What the precise meaning of “structural characteristics” is, will depend
on what model of the countries the researcher has in mind. According to
the Solow model, a set of relevant “structural characteristics” are: the aggre-
gate production function, the initial level of technology, the rate of technical
progress, the capital depreciation rate, the saving rate, and the population
growth rate. But the Solow model, as well as its extension with human cap-
ital (Mankiw et al., 1992), is a model of a closed economy with exogenous
technical progress. The model deals with “within-country” convergence in
the sense that the model predicts that a closed economy being initially be-
low or above its steady state path, will over time converge towards its steady
state path. It is far from obvious that this kind of model is a good model
of cross-country convergence in a globalized world where capital mobility
and to some extent also labor mobility are important and some countries are
pushing the technological frontier further out, while others try to imitate and
catch up.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
1.4. Concepts of income convergence 15
Dispersion
21000
Dispersion of GDP per capita
18000
Dispersion of GDP per worker
15000
12000
9000
6000
3000
0
1950 1960 1970 1980 1990 2000
Year
Remarks: Germany is not included in GDP per worker. GDP per worker is missing for
Sweden and Greece in 1950, and for Portugal in 1998. The EU comprises Belgium,
Denmark, Finland, France, Greece, Holland, Ireland, Italy, Luxembourg, Portugal, Spain,
Sweden, Germany, the UK and Austria.
Source: Pwt6, OECD Economic Outlook No. 65 1999 via Eco Win and World Bank Global
Development Network Growth Database.
Figure 1.4: Standard deviation of GDP per capita and per worker across 12 EU
countries, 1950-1998.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
16 CHAPTER 1. INTRODUCTION TO ECONOMIC GROWTH
Dispersion
0,4
Dispersion of the log of GDP
0,36 per capita
0,32 Dispersion of the log of GDP per
worker
0,28
0,24
0,2
0,16
0,12
0,08
0,04
0
1950 1960 1970 1980 1990 2000
Year
Remarks: Germany is not included in GDP per worker. GDP per worker is missing for Sweden and
Greece in 1950, and for Portugal in 1998. The EU comprises Belgium, Denmark, Finland, France,
Greece, Holland, Ireland, Italy, Luxemb ourg, Portugal, Spain, Sweden, Germany, the UK and
Austria.
Source: Pwt6, OECD Economic Outlook No. 65 1999 via Eco Win and World Bank Global
Development Network Growth Database.
Figure 1.5: Standard deviation of the log of GDP per capita and per worker across
12 EU countries, 1950-1998.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
1.4. Concepts of income convergence 17
1.4.5 A bird’s-eye view of the data
In the following no serious econometrics is attempted. We use the term
“trend” in an admittedly loose sense.
Figure 1.4 shows the time profile for the standard deviation of itself for
12 EU countries, whereas Figure 1.5 and Figure 1.6 show the time profile
of the standard deviation of log and the time profile of the coefficient of
variation, respectively. Comparing the upward trend in Figure 1.4 with the
downward trend in the two other figures, we have an illustration of the fact
that the movement of the standard deviation of itself does not capture
income convergence. To put it another way: although there seems to be
conditional income convergence with respect to the two scale-free measures,
Figure 1.4 shows that this tendency to convergence is not so strong as to
produce a narrowing of the absolute distance between the EU countries.6
Figure 1.7 shows the time path of the coefficient of variation across 121
countries in the world, 22 OECD countries and 12 EU countries, respectively.
We see the lack of unconditional income convergence, but the presence of con-
ditional income convergence. One should not over-interpret the observation
of convergence for the 22 OECD countries over the period 1950-1990. It is
likely that this observation suffer from the selection bias problem mentioned
in Section 1.4.1. A country that was poor in 1950 will typically have become
a member of OECD only if it grew relatively fast afterwards.
1.4.6 Other convergence concepts
Of course, just considering the time profile of the first and second moments
of a distribution may sometimes be a poor characterization of the evolution
of the distribution. For example, there are signs that the distribution has
polarized into twin peaks of rich and poor countries (Quah, 1996a; Jones,
1997). Related to this observation is the notion of club convergence. If in-
come convergence occurs only among a subgroup of the countries that to
some extent share the same initial conditions, then we say there is club-
convergence. This concept is relevant in a setting where there are multiple
steady states toward which countries can converge. At least at the theoret-
ical level multiple steady states can easily arise in overlapping generations
models. Then the initial condition for a given country matters for which of
these steady states this country is heading to. Similarly, we may say that
conditional club-convergence is present, if income convergence occurs only
6
Unfortunately, sometimes misleading graphs or texts to graphs about across-country
income convergence are published. In the collection of exercises, Chapter 1, you are asked
to discuss some examples of this.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
18 CHAPTER 1. INTRODUCTION TO ECONOMIC GROWTH
Coefficient of variation
0,9
Coefficient of variation for GDP per capita
0,8
Coefficient of variation for GDP per worker
0,7
0,6
0,5
0,4
0,3
0,2
0,1
0
1950 1960 1970 1980 1990 2000
Year
Remarks: Germany is not included in GDP per worker. GDP per worker is missing for Sweden and Greece in 1950,
and for Portugal in 1998. The EU comprises Belgium, Denmark, Finland, France, Greece, Holland, Ireland, Italy,
Luxembourg, Portugal, Spain, Sweden, Germany, the UK and Austria.
Source: Pwt6, OECD Economic Outlook No. 65 1999 via Eco Win and World Bank Global Development Network
Growth Database.
Figure 1.6: Coefficient of variation of GDP per capita and GDP per worker across
12 EU countries, 1950-1998.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
1.5. Literature 19
Coefficient
Coefficient
of variation
of variation
1,2
The world (1960 -88)
1
0,8
0,6 22 OECD countries (1950-90)
0,4
0,2 EU-12 (1960-9 5)
0
1950 1953 1956 1959 1962 1965 1968 1971 1974 1977 1980 1983 1986 1989 1992 1995
Remarks: 'T he world' comprises 1 21 countries (not weighed by size) where complete time series for GDP per cap ita exist.
The OECD coun tries exclu de South Korea, Hungary, Poland, Iceland , Czech Rep., Luxembourg and Mexico.
EU-12 co mprises: Benelux, Germany, France, Italy, Denmark, Ireland, UK, Spain, Portugal og Greece.
Source: Penn World Table 5.6 and OECD Economic Ou tlook, Statistics on Microcomputer Disc, Decemb er 1998.
Figure 1.7: Coefficient of variation of income per capita across different sets of
countries.
for a subgroup of the countries, namely countries sharing similar structural
characteristics (this may to some extent be true for the OECD countries)
and, within an interval, similar initial conditions.
Instead of focusing on income convergence, one could study TFP conver-
gence at aggregate or industry level.7 Sometimes the less demanding concept
of growth rate convergence is the focus.
The above considerations are only of a very elementary nature and are
only about descriptive statistics. The reader is referred to the large existing
literature on concepts and econometric methods of relevance for character-
izing the evolution of world income distribution (see Quah, 1996b, 1996c,
1997, and for a survey, see Islam 2003).
1.5 Literature
Acemoglu, D., 2009, Introduction to Modern Economic Growth, Princeton
University Press: Oxford.
Acemoglu, D., and V. Guerrieri, 2008, Capital deepening and nonbalanced
7
See, for instance, Bernard and Jones 1996a and 1996b.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
20 CHAPTER 1. INTRODUCTION TO ECONOMIC GROWTH
economic growth, J. Political Economy, vol. 116 (3), 467- .
Attfield, C., and J.R.W. Temple, 2010, Balanced growth and the great
ratios: New evidence for the US and UK, Journal of Macroeconomics,
vol. 32, 937-956.
Barro, R. J., and X. Sala-i-Martin, 1995, Economic Growth, MIT Press,
New York. Second edition, 2004.
Bernard, A.B., and C.I. Jones, 1996a, ..., Economic Journal.
- , 1996b, Comparing Apples to Oranges: Productivity Convergence and
Measurement Across Industries and Countries, American Economic
Review, vol. 86 (5), 1216-1238.
Cowell, Frank A., 1995, Measuring Inequality. 2. ed., London.
Dalgaard, C.-J., and J. Vastrup, 2001, On the measurement of -convergence,
Economics letters, vol. 70, 283-87.
Dansk økonomi. Efterår 2001, (Det økonomiske Råds formandskab) Kbh.
2001.
Deininger, K., and L. Squire, 1996, A new data set measuring income in-
equality, The World Bank Economic Review, 10, 3.
Delong, B., 1988, ... American Economic Review.
Handbook of Economic Growth, vol. 1A and 1B, ed. by S. N. Durlauf and
P. Aghion, Amsterdam 2005.
Handbook of Income Distribution, vol. 1, ed. by A.B. Atkinson and F.
Bourguignon, Amsterdam 2000.
Islam, N., 2003, What have we learnt from the convergence debate? Journal
of Economic Surveys 17, 3, 309-62.
Kaldor, N., 1957, A model of economic growth, The Economic Journal, vol.
67, pp. 591-624.
- , 1961, “Capital Accumulation and Economic Growth”. In: F. Lutz, ed.,
Theory of Capital, London: MacMillan.
- , 1967, Strategic Factors in Economic Development, New York State School
of Industrial and Labor Relations, Cornell University.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
1.5. Literature 21
Kongsamut et al., 2001, Beyond Balanced Growth, Review of Economic
Studies, vol. 68, 869-882.
Kuznets, S., 1957, Quantitative aspects of economic growth of nations: II,
Economic Development and Cultural Change, Supplement to vol. 5,
3-111.
Maddison, A., 1982,
Mankiw, N.G., D. Romer, and D.N. Weil, 1992,
Pritchett, L., 1997, Divergence — big time, Journal of Economic Perspec-
tives, vol. 11, no. 3.
Quah, D., 1996a, Twin peaks ..., Economic Journal, vol. 106, 1045-1055.
-, 1996b, Empirics for growth and convergence, European Economic Review,
vol. 40 (6).
-, 1996c, Convergence empirics ..., J. of Ec. Growth, vol. 1 (1).
-, 1997, Searching for prosperity: A comment, Carnegie-Rochester Confer-
ende Series on Public Policy, vol. 55, 305-319.
Romer, D., 2012, Advanced Macroeconomics, 4th ed., McGraw-Hill: New
York.
Sala-i-Martin, X., 2006, The World Distribution of Income, Quarterly Jour-
nal of Economics 121, No. 2.
Solow, R.M., 1970, Growth theory. An exposition, Clarendon Press: Oxford.
Second enlarged edition, 2000.
Valdés, B., 1999, Economic Growth. Theory, Empirics, and Policy, Edward
Elgar.
On measurement problems, see: http://www.worldbank.org/poverty/inequal/methods/index.htm
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
22 CHAPTER 1. INTRODUCTION TO ECONOMIC GROWTH
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
Chapter 2
Review of technology
The aim of this chapter is, first, to introduce the terminology concerning
firms’ technology and technological change used in the lectures and exercises
of this course. At a few points I deviate somewhat from definitions in Ace-
moglu’s book. Section 1.3 can be used as a formula manual for the case of
CRS.
Second, the chapter contains a brief discussion of the somewhat contro-
versial notions of a representative firm and an aggregate production function.
Regarding the distinction between discrete and continuous time analysis,
most of the definitions contained in this chapter are applicable to both.
2.1 The production technology
Consider a two-factor production function given by
= ( ) (2.1)
where is output (value added) per time unit, is capital input per time
unit, and is labor input per time unit ( ≥ 0 ≥ 0). We may think of
(2.1) as describing the output of a firm, a sector, or the economy as a whole.
It is in any case a very simplified description, ignoring the heterogeneity of
output, capital, and labor. Yet, for many macroeconomic questions it may
be a useful first approach. Note that in (2.1) not only but also and
represent flows, that is, quantities per unit of time. If the time unit is one
year, we think of as measured in machine hours per year. Similarly, we
think of as measured in labor hours per year. Unless otherwise specified, it
is understood that the rate of utilization of the production factors is constant
over time and normalized to one for each production factor. As explained
in Chapter 1, we can then use the same symbol, for the flow of capital
services as for the stock of capital. Similarly with
23
24 CHAPTER 2. REVIEW OF TECHNOLOGY
2.1.1 A neoclassical production function
By definition, and are non-negative. It is generally understood that a
production function, = ( ) is continuous and that (0 0) = 0 (no in-
put, no output). Sometimes, when specific functional forms are used to repre-
sent a production function, that function may not be defined at points where
= 0 or = 0 or both. In such a case we adopt the convention that the do-
main of the function is understood extended to include such boundary points
whenever it is possible to assign function values to them such that continuity
is maintained. For instance the function ( ) = + ( + )
where 0 and 0 is not defined at ( ) = (0 0) But by assigning
the function value 0 to the point (0 0) we maintain both continuity and the
“no input, no output” property, cf. Exercise 2.4.
We call the production function neoclassical if for all ( ) with 0
and 0 the following additional conditions are satisfied:
(a) ( ) has continuous first- and second-order partial derivatives sat-
isfying:
0 0 (2.2)
0 0 (2.3)
(b) ( ) is strictly quasiconcave (i.e., the level curves, also called iso-
quants, are strictly convex to the origin).
In words: (a) says that a neoclassical production function has continuous
substitution possibilities between and and the marginal productivities
are positive, but diminishing in own factor. Thus, for a given number of ma-
chines, adding one more unit of labor, adds to output, but less so, the higher
is already the labor input. And (b) says that every isoquant, ( ) = ̄
has a strictly convex form qualitatively similar to that shown in Figure 2.1.1
When we speak of for example as the marginal productivity of labor, it
is because the “pure” partial derivative, = has the denomina-
tion of a productivity (output units/yr)/(man-yrs/yr). It is quite common,
however, to refer to as the marginal product of labor. Then a unit mar-
ginal increase in the labor input is understood: ∆ ≈ ( )∆ =
when ∆ = 1 Similarly, can be interpreted as the marginal productiv-
ity of capital or as the marginal product of capital. In the latter case it is
understood that ∆ = 1 so that ∆ ≈ ( )∆ =
1
For any fixed ̄ ª ≥ 0 the associated isoquant is the level set
{( ) ∈ R+ | ( ) = ̄
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
2.1. The production technology 25
The definition of a neoclassical production function can be extended to
the case of inputs. Let the input quantities be 1 2 and consider
a production function = (1 2 ) Then is called neoclassical if
all the marginal productivities are positive, but diminishing, and is strictly
quasiconcave (i.e., the upper contour sets are strictly convex, cf. Appendix
A).
Returning to the two-factor case, since ( ) presumably depends on
the level of technical knowledge and this level depends on time, we might
want to replace (2.1) by
= ( ) (2.4)
where the superscript on indicates that the production function may shift
over time, due to changes in technology. We then say that (·) is a neoclas-
sical production function if it satisfies the conditions (a) and (b) for all pairs
( ). Technological progress can then be said to occur when, for and
held constant, output increases with
For convenience, to begin with we skip the explicit reference to time and
level of technology.
The marginal rate of substitution Given a neoclassical production
function we consider the isoquant defined by ( ) = ̄ where ̄
is a positive constant. The marginal rate of substitution, , of for
at the point ( ) is defined as the absolute slope of the isoquant at that
point, cf. Figure 2.1. The equation ( ) = ̄ defines as an implicit
function of By implicit differentiation we find ( ) + ( )
= 0 from which follows
( )
≡ − = 0 (2.5)
| =̄ ( )
That is, measures the amount of that can be saved (approxi-
mately) by applying an extra unit of labor. In turn, this equals the ratio
of the marginal productivities of labor and capital, respectively.2 Since
is neoclassical, by definition is strictly quasi-concave and so the marginal
rate of substitution is diminishing as substitution proceeds, i.e., as the labor
input is further increased along a given isoquant. Notice that this feature
characterizes the marginal rate of substitution for any neoclassical production
function, whatever the returns to scale (see below).
¯
2
The subscript ¯ = ̄ in (2.5) indicates that we are moving along a given isoquant,
( ) = ̄ Expressions like, e.g., ( ) or 2 ( ) mean the partial derivative of
w.r.t. the second argument, evaluated at the point ( )
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
26 CHAPTER 2. REVIEW OF TECHNOLOGY
K/L
F (K , L) Y
MRS KL
L
L
Figure 2.1: as the absolute slope of the isoquant.
When we want to draw attention to the dependency of the marginal rate of
substitution on the factor combination considered, we write ( )
Sometimes in the literature, the marginal rate of substitution between two
production factors, and is called the technical rate of substitution (or
the technical rate of transformation) in order to distinguish from a consumer’s
marginal rate of substitution between two consumption goods.
As is well-known from microeconomics, a firm that minimizes production
costs for a given output level and given factor prices, will choose a factor com-
bination such that equals the ratio of the factor prices. If ( )
is homogeneous of degree , then the marginal rate of substitution depends
only on the factor proportion and is thus the same at any point on the ray
= (̄̄) That is, in this case the expansion path is a straight line.
The Inada conditions A continuously differentiable production function
is said to satisfy the Inada conditions 3 if
lim ( ) = ∞ lim ( ) = 0 (2.6)
→0 →∞
lim ( ) = ∞ lim ( ) = 0 (2.7)
→0 →∞
In this case, the marginal productivity of either production factor has no
upper bound when the input of the factor becomes infinitely small. And the
marginal productivity is gradually vanishing when the input of the factor
increases without bound. Actually, (2.6) and (2.7) express four conditions,
which it is preferable to consider separately and label one by one. In (2.6) we
have two Inada conditions for (the marginal productivity of capital),
the first being a lower, the second an upper Inada condition for . And
3
After the Japanese economist Ken-Ichi Inada, 1925-2002.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
2.1. The production technology 27
in (2.7) we have two Inada conditions for (the marginal productivity
of labor), the first being a lower, the second an upper Inada condition for
. In the literature, when a sentence like “the Inada conditions are
assumed” appears, it is sometimes not made clear which, and how many, of
the four are meant. Unless it is evident from the context, it is better to be
explicit about what is meant.
The definition of a neoclassical production function we gave above is quite
common in macroeconomic journal articles and convenient because of its
flexibility. There are textbooks that define a neoclassical production function
more narrowly by including the Inada conditions as a requirement for calling
the production function neoclassical. In contrast, in this course, when in a
given context we need one or another Inada condition, we state it explicitly
as an additional assumption.
2.1.2 Returns to scale
If all the inputs are multiplied by some factor, is output then multiplied by
the same factor? There may be different answers to this question, depending
on circumstances. We consider a production function ( ) where 0
and 0 Then is said to have constant returns to scale (CRS for short)
if it is homogeneous of degree one, i.e., if for all ( ) and all 0
( ) = ( )
As all inputs are scaled up or down by some factor 1, output is scaled up
or down by the same factor.4 The assumption of CRS is often defended by
the replication argument. Before discussing this argument, lets us define the
two alternative “pure” cases.
The production function ( ) is said to have increasing returns to
scale (IRS for short) if, for all ( ) and all 1,
( ) ( )
That is, IRS is present if, when all inputs are scaled up by some factor
1, output is scaled up by more than this factor. The existence of gains by
specialization and division of labor, synergy effects, etc. sometimes speak in
support of this assumption, at least up to a certain level of production. The
assumption is also called the economies of scale assumption.
4
In their definition of a neoclassical production function some textbooks add constant
returns to scale as a requirement besides (a) and (b). This course follows the alternative
terminology where, if in a given context an assumption of constant returns to scale is
needed, this is stated as an additional assumption.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
28 CHAPTER 2. REVIEW OF TECHNOLOGY
Another possibility is decreasing returns to scale (DRS). This is said to
occur when for all ( ) and all 1
( ) ( )
That is, DRS is present if, when all inputs are scaled up by some factor,
output is scaled up by less than this factor. This assumption is also called
the diseconomies of scale assumption. The underlying hypothesis may be
that control and coordination problems confine the expansion of size. Or,
considering the “replication argument” below, DRS may simply reflect that
behind the scene there is an additional production factor, for example land
or a irreplaceable quality of management, which is tacitly held fixed, when
the factors of production are varied.
EXAMPLE 1 The production function
= 0 0 1 0 1 (2.8)
where and are given parameters, is called a Cobb-Douglas production
function. The parameter depends on the choice of measurement units; for
a given such choice it reflects “efficiency”, also called the “total factor pro-
ductivity”. Exercise 2.2 asks the reader to verify that (2.8) satisfies (a) and
(b) above and is therefore a neoclassical production function. The function
is homogeneous of degree + . If + = 1 there are CRS. If + 1
there are DRS, and if + 1 there are IRS. Note that and must
be less than 1 in order not to violate the diminishing marginal productivity
condition. ¤
EXAMPLE 2 The production function
= min( ) 0 0 (2.9)
where and are given parameters, is called a Leontief production function
or a fixed-coefficients production function; and are called the technical
coefficients. The function is not neoclassical, since the conditions (a) and (b)
are not satisfied. Indeed, with this production function the production fac-
tors are not substitutable at all. This case is also known as the case of perfect
complementarity between the production factors. The interpretation is that
already installed production equipment requires a fixed number of workers to
operate it. The inverse of the parameters and indicate the required cap-
ital input per unit of output and the required labor input per unit of output,
respectively. Extended to many inputs, this type of production function is
often used in multi-sector input-output models (also called Leontief models).
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
2.1. The production technology 29
In aggregate analysis neoclassical production functions, allowing substitution
between capital and labor, are more popular than Leontief functions. But
sometimes the latter are preferred, in particular in short-run analysis with
focus on the use of already installed equipment where the substitution pos-
sibilities are limited.5 As (2.9) reads, the function has CRS. A generalized
form of the Leontief function is = min( ) where 0. When
1 there are DRS, and when 1 there are IRS. ¤
The replication argument The assumption of CRS is widely used in
macroeconomics. The model builder may appeal to the replication argument.
To explain the content of this argument we have to first clarify the distinction
between rival and nonrival inputs or more generally the distinction between
rival and nonrival goods. A good is rival if its character is such that one
agent’s use of it inhibits other agents’ use of it at the same time. A pencil
is thus rival. Many production inputs like raw materials, machines, labor
etc. have this property. In contrast, however, technical knowledge like a
farmaceutical formula or an engineering principle is nonrival. An unbounded
number of factories can simultaneously use the same farmaceutical formula.
The replication argument now says that by, conceptually, doubling all the
rival inputs, we should always be able to double the output, since we just
“replicate” what we are already doing. One should be aware that the CRS
assumption is about technology in the sense of functions linking inputs to
outputs − limits to the availability of input resources is an entirely different
matter. The fact that for example managerial talent may be in limited supply
does not preclude the thought experiment that if a firm could double all its
inputs, including the number of talented managers, then the output level
could also be doubled.
The replication argument presupposes, first, that all the relevant inputs
are explicit as arguments in the production function; second, that these are
changed equiproportionately. This, however, exhibits the weakness of the
replication argument as a defence for assuming CRS of our present production
function, (·) One could easily make the case that besides capital and labor,
also land is a necessary input and should appear as a separate argument.6
If an industrial firm decides to duplicate what it has been doing, it needs a
piece of land to build another plant like the first. Then, on the basis of the
replication argument we should in fact expect DRS w.r.t. capital and labor
alone. In manufacturing and services, empirically, this and other possible
5
Cf. Section 2.4.
6
We think of “capital” as producible means of production, whereas “land” refers to
non-producible natural resources, including for example building sites.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
30 CHAPTER 2. REVIEW OF TECHNOLOGY
sources for departure from CRS may be minor and so many macroeconomists
feel comfortable enough with assuming CRS w.r.t. and alone, at least
as a first approximation. This approximation is, however, less applicable to
poor countries, where natural resources may be a quantitatively important
production factor.
There is a further problem with the replication argument. Strictly speak-
ing, the CRS claim is that by changing all the inputs equiproportionately
by any positive factor, which does not have to be an integer, the firm
should be able to get output changed by the same factor. Hence, the replica-
tion argument requires that indivisibilities are negligible, which is certainly
not always the case. In fact, the replication argument is more an argument
against DRS than for CRS in particular. The argument does not rule out
IRS due to synergy effects as size is increased.
Sometimes the replication line of reasoning is given a more subtle form.
This builds on a useful local measure of returns to scale, named the elasticity
of scale.
The elasticity of scale* To allow for indivisibilities and mixed cases (for
example IRS at low levels of production and CRS or DRS at higher levels),
we need a local measure of returns to scale. One defines the elasticity of
scale, ( ) of at the point ( ) where ( ) 0 as
( ) ∆ ( ) ( )
( ) = ≈ evaluated at = 1
( ) ∆
(2.10)
So the elasticity of scale at a point ( ) indicates the (approximate) per-
centage increase in output when both inputs are increased by 1 percent. We
say that ⎧
⎨ 1 then there are locally IRS,
if ( ) = 1 then there are locally CRS, (2.11)
⎩
1 then there are locally DRS.
The production function may have the same elasticity of scale everywhere.
This is the case if and only if the production function is homogeneous. If
is homogeneous of degree then ( ) = and is called the elasticity
of scale parameter.
Note that the elasticity of scale at a point ( ) will always equal the
sum of the partial output elasticities at that point:
( ) ( )
( ) = + (2.12)
( ) ( )
This follows from the definition in (2.10) by taking into account that
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
2.1. The production technology 31
LMC (Y ) LAC (Y )
Y
Y*
Figure 2.2: Locally CRS at optimal plant size.
( )
= ( ) + ( )
= ( ) + ( ) when evaluated at = 1
Figure 2.2 illustrates a popular case from introductory economics, an
average cost curve which from the perspective of the individual firm (or plant)
is U-shaped: at low levels of output there are falling average costs (thus IRS),
at higher levels rising average costs (thus DRS).7 Given the input prices,
and and a specified output level, ̄ we know that the cost minimizing
factor combination (̄ ̄) is such that (̄ ̄) (̄ ̄) = It is
shown in Appendix A that the elasticity of scale at (̄ ̄) will satisfy:
(̄ )
(̄ ̄) = (2.13)
(̄ )
where (̄ ) is average costs (the minimum unit cost associated with
producing ̄ ) and (̄ ) is marginal costs at the output level ̄ . The
in and stands for “long-run”, indicating that both capital and
labor are considered variable production factors within the period considered.
At the optimal plant size, ∗ there is equality between and ,
implying a unit elasticity of scale, that is, locally we have CRS. That the long-
run average costs are here portrayed as rising for ̄ ∗ is not essential
for the argument but may reflect either that coordination difficulties are
inevitable or that some additional production factor, say the building site of
the plant, is tacitly held fixed.
Anyway, we have here a more subtle replication argument for CRS w.r.t.
and at the aggregate level. Even though technologies may differ across
plants, the surviving plants in a competitive market will have the same aver-
age costs at the optimal plant size. In the medium and long run, changes in
7
By a “firm” is generally meant the company as a whole. A company may have several
“manufacturing plants” placed at different locations.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
32 CHAPTER 2. REVIEW OF TECHNOLOGY
aggregate output will take place primarily by entry and exit of optimal-size
plants. Then, with a large number of relatively small plants, each produc-
ing at approximately constant unit costs for small output variations, we can
without substantial error assume constant returns to scale at the aggregate
level. So the argument goes. Notice, however, that even in this form the
replication argument is not entirely convincing since the question of indivis-
ibility remains. The optimal plant size may be large relative to the market
− and is in fact so in many industries. Besides, in this case also the perfect
competition premise breaks down.
2.1.3 Properties of the production function under CRS
The empirical evidence concerning returns to scale is mixed. Notwithstand-
ing the theoretical and empirical ambiguities, the assumption of CRS w.r.t.
capital and labor has a prominent role in macroeconomics. In many con-
texts it is regarded as an acceptable approximation and a convenient simple
background for studying the question at hand.
Expedient inferences of the CRS assumption include:
(i) marginal costs are constant and equal to average costs (so the right-
hand side of (2.13) equals unity);
(ii) if production factors are paid according to their marginal productivi-
ties, factor payments exactly exhaust total output so that pure profits
are neither positive nor negative (so the right-hand side of (2.12) equals
unity);
(iii) a production function known to exhibit CRS and satisfy property (a)
from the definition of a neoclassical production function above, will au-
tomatically satisfy also property (b) and consequently be neoclassical;
(iv) a neoclassical two-factor production function with CRS has always
0 i.e., it exhibits “direct complementarity” between and
;
(v) a two-factor production function known to have CRS and to be twice
continuously differentiable with positive marginal productivity of each
factor everywhere in such a way that all isoquants are strictly convex to
the origin, must have diminishing marginal productivities everywhere.8
8
Proofs of these claims can be found in intermediate microeconomics textbooks and in
the Appendix to Chapter 2 of my Lecture Notes in Macroeconomics.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
2.1. The production technology 33
A principal implication of the CRS assumption is that it allows a re-
duction of dimensionality. Considering a neoclassical production function,
= ( ) with 0 we can under CRS write ( ) = ( 1)
≡ () where ≡ is called the capital-labor ratio (sometimes the cap-
ital intensity) and () is the production function in intensive form (some-
times named the per capita production function). Thus output per unit of
labor depends only on the capital intensity:
≡ = ()
When the original production function is neoclassical, under CRS the
expression for the marginal productivity of capital simplifies:
[()]
( ) = = = 0 () = 0 () (2.14)
And the marginal productivity of labor can be written
[ ()]
( ) = = = () + 0 ()
= () + 0 ()(−−2 ) = () − 0 () (2.15)
A neoclassical CRS production function in intensive form always has a posi-
tive first derivative and a negative second derivative, i.e., 0 0 and 00 0
The property 0 0 follows from (2.14) and (2.2). And the property 00 0
follows from (2.3) combined with
0 () 1
( ) = = 00 () = 00 ()
For a neoclassical production function with CRS, we also have
() − 0 () 0 for all 0 (2.16)
in view of (0) ≥ 0 and 00 0 Moreover,
lim [() − 0 ()] = (0) (2.17)
→0
Indeed, from the mean value theorem9 we know there exists a number ∈
(0 1) such that for any given 0 we have () − (0) = 0 () From this
follows ()− 0 () = (0) ()− 0 () since 0 () 0 () by 00 0.
9
This theorem says that if is continuous in [ ] and differentiable in ( ) then
there exists at least one point in ( ) such that 0 () = ( () − ())( − )
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
34 CHAPTER 2. REVIEW OF TECHNOLOGY
In view of (0) ≥ 0 this establishes (2.16) And from () () − 0 ()
(0) and continuity of follows (2.17).
Under CRS the Inada conditions for can be written
lim 0 () = ∞ lim 0 () = 0 (2.18)
→0 →∞
In this case standard parlance is just to say that “ satisfies the Inada con-
ditions”.
An input which must be positive for positive output to arise is called an
essential input; an input which is not essential is called an inessential input.
The second part of (2.18), representing the upper Inada condition for
under CRS, has the implication that labor is an essential input; but capital
need not be, as the production function () = + (1 + ) 0 0
illustrates. Similarly, under CRS the upper Inada condition for implies
that capital is an essential input. These claims are proved in Appendix C.
Combining these results, when both the upper Inada conditions hold and
CRS obtain, then both capital and labor are essential inputs.10
Figure 2.3 is drawn to provide an intuitive understanding of a neoclassical
CRS production function and at the same time illustrate that the lower Inada
conditions are more questionable than the upper Inada conditions. The left
panel of Figure 2.3 shows output per unit of labor for a CRS neoclassical pro-
duction function satisfying the Inada conditions for . The () in the
diagram could for instance represent the Cobb-Douglas function in Example
1 with = 1− i.e., () = The right panel of Figure 2.3 shows a non-
neoclassical case where only two alternative Leontief techniques are available,
technique 1: = min(1 1 ) and technique 2: = min(2 2 ) In the
exposed case it is assumed that 2 1 and 2 1 (if 2 ≥ 1 at the
same time as 2 1 technique 1 would not be efficient, because the same
output could be obtained with less input of at least one of the factors by
shifting to technique 2). If the available and are such that 1 1
or 2 2 , some of either or respectively, is idle. If, however, the
available and are such that 1 1 2 2 it is efficient to combine
the two techniques and use the fraction of and in technique 1 and the
remainder in technique 2, where = (2 2 − )(2 2 − 1 1 ) In this
way we get the “labor productivity curve” OPQR (the envelope of the two
techniques) in Figure 2.3. Note that for → 0 stays equal to 1 ∞
whereas for all 2 2 = 0 A similar feature remains true, when
we consider many, say alternative efficient Leontief techniques available.
Assuming these techniques cover a considerable range w.r.t. the ratios,
10
Given a Cobb-Douglas production function, both production factors are essential
whether we have DRS, CRS, or IRS.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
2.2. Technological change 35
y y
f '(k0 ) Q
f (k 0 ) y f (k ) R
P
f(k0)-f’(k0)k0
k k
O k0 O B1 / A1 B2 / A2
Figure 2.3: Two labor productivity curves based on CRS technologies. Left: neo-
classical technology with Inada conditions for MPK satisfied; the graphical repre-
sentation of MPK and MPL at = 0 .as 0 (0 ) and ( 0 )− 0 (0 )0 are indicated.
Right: a combination of two efficient Leontief techniques.
we get a labor productivity curve looking more like that of a neoclassical CRS
production function. On the one hand, this gives some intuition of what lies
behind the assumption of a neoclassical CRS production function. On the
other hand, it remains true that for all = 011 whereas
for → 0 stays equal to 1 ∞ thus questioning the lower Inada
condition.
The implausibility of the lower Inada conditions is also underlined if we
look at their implication in combination with the more reasonable upper
Inada conditions. Indeed, the four Inada conditions taken together imply,
under CRS, that output has no upper bound when either input goes to
infinity for fixed amount of the other input (see Appendix C).
2.2 Technological change
When considering the movement over time of the economy, we shall often
take into account the existence of technological change. When technological
change occurs, the production function becomes time-dependent. Over time
the production factors tend to become more productive: more output for
given inputs. To put it differently: the isoquants move inward. When this is
the case, we say that the technological change displays technological progress.
11
Here we assume the techniques are numbered according to ranking with respect to the
size of
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
36 CHAPTER 2. REVIEW OF TECHNOLOGY
Concepts of neutral technological change
A first step in taking technological change into account is to replace (2.1) by
(2.4). Empirical studies typically specialize (2.4) by assuming that techno-
logical change take a form known as factor-augmenting technological change:
= ( ) (2.19)
where is a (time-independent) neoclassical production function, and
are output, capital, and labor input, respectively, at time while and
are time-dependent efficiencies of capital and labor, respectively, reflecting
technological change. In macroeconomics an even more specific form is often
assumed, namely the form of Harrod-neutral technological change.12 This
amounts to assuming that in (2.19) is a constant (which we can then
normalize to one). So only which we will then denote is changing over
time, and we have
= ( ) (2.20)
The efficiency of labor, is then said to indicate the technology level. Al-
though one can imagine natural disasters implying a fall in generally
tends to rise over time and then we say that (2.20) represents Harrod-neutral
technological progress. An alternative name for this is labor-augmenting tech-
nological progress (technological change acts as if the labor input were aug-
mented).
If the function in (2.20) is homogeneous of degree one (so that the
technology exhibits CRS w.r.t. capital and labor), we may write
̃ ≡ = ( 1) = (̃ 1) ≡ (̃ ) 0 0 00 0
where ̃ ≡ ( ) ≡ (habitually called the “effective” capital in-
tensity or, if there is no risk of confusion, just the capital intensity). In
rough accordance with a general trend in aggregate productivity data for
industrialized countries we often assume that grows at a constant rate,
so that in discrete time = 0 (1 + ) and in continuous time = 0
where 0 The popularity in macroeconomics of the hypothesis of labor-
augmenting technological progress derives from its consistency with Kaldor’s
“stylized facts”, cf. Chapter 4.
There exists two alternative concepts of neutral technological progress.
Hicks-neutral technological progress is said to occur if technological develop-
ment is such that the production function can be written in the form
= ( ) (2.21)
12
The name refers to the English economist Roy F. Harrod, 1900−1978.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
2.2. Technological change 37
where, again, is a (time-independent) neoclassical production function,
while is the growing technology level.13 The assumption of Hicks-neutrality
has been used more in microeconomics and partial equilibrium analysis than
in macroeconomics. If has CRS, we can write (2.21) as = ( )
Comparing with (2.19), we see that in this case Hicks-neutrality is equivalent
with = in (2.19), whereby technological change is said to be equally
factor-augmenting.
Finally, in a kind of symmetric analogy with (2.20), Solow-neutral tech-
nological progress14 is often in textbooks presented by a formula like:
= ( ) (2.22)
Another name for the same is capital-augmenting technological progress (be-
cause here technological change acts as if the capital input were augmented).
Solow’s original concept15 of neutral technological change is not well por-
trayed this way, however, since it is related to the notion of embodied tech-
nological change and capital of different vintages, see below.
It is easily shown (Exercise 2.5) that the Cobb-Douglas production func-
tion (2.8) satisfies all three neutrality criteria at the same time, if it satisfies
one of them (which it does if technological change does not affect and ).
It can also be shown that within the class of neoclassical CRS production
functions the Cobb-Douglas function is the only one with this property (see
Exercise 4.? in Chapter 4).
Note that the neutrality concepts do not say anything about the source
of technological progress, only about the quantitative form in which it ma-
terializes. For instance, the occurrence of Harrod-neutrality should not be
interpreted as indicating that the technological change emanates specifically
from the labor input in some sense. Harrod-neutrality only means that tech-
nological innovations predominantly are such that not only do labor and
capital in combination become more productive, but this happens to man-
ifest itself in the form (2.20). Similarly, if indeed an improvement in the
quality of the labor input occurs, this “labor-specific” improvement may be
manifested in a higher or both.
Before proceeding, we briefly comment on how the capital stock,
is typically measured. While data on gross investment, is available in
national income and product accounts, data on usually is not. One ap-
13
The name refers to the English economist and Nobel Prize laureate John R. Hicks,
1904−1989.
14
The name refers to the American economist and Nobel Prize laureate Robert Solow
(1924−).
15
Solow (1960).
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
38 CHAPTER 2. REVIEW OF TECHNOLOGY
proach to the measurement of is the perpetual inventory method which
builds upon the accounting relationship
= −1 + (1 − )−1 (2.23)
Assuming a constant capital depreciation rate backward substitution gives
X
= −1 +(1−) [−2 + (1 − )−2 ] = . . . = (1−)−1 − +(1−) −
=1
(2.24)
Based on a long time series for and an estimate of one can insert these
observed values in the formula and calculate , starting from a rough con-
jecture about the initial value − The result will not be very sensitive to
this conjecture since for large the last term in (2.24) becomes very small.
Embodied vs. disembodied technological progress
There exists an additional taxonomy of technological change. We say that
technological change is embodied, if taking advantage of new technical knowl-
edge requires construction of new investment goods. The new technology is
incorporated in the design of newly produced equipment, but this equipment
will not participate in subsequent technological progress. An example: only
the most recent vintage of a computer series incorporates the most recent
advance in information technology. Then investment goods produced later
(investment goods of a later “vintage”) have higher productivity than in-
vestment goods produced earlier at the same resource cost. Thus investment
becomes an important driving force in productivity increases.
We way formalize embodied technological progress by writing capital ac-
cumulation in the following way:
+1 − = − (2.25)
where is gross investment in period , i.e., = − and measures
the “quality” (productivity) of newly produced investment goods. The rising
level of technology implies rising so that a given level of investment gives
rise to a greater and greater addition to the capital stock, measured
in efficiency units. In aggregate models and are produced with the
same technology, the aggregate production function. From this together with
(2.25) follows that capital goods can be produced at the same minimum
cost as one consumption good. Hence, the equilibrium price, of capital
goods in terms of the consumption good must equal the inverse of i.e.,
= 1 The output-capital ratio in value terms is () =
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
2.3. The concepts of a representative firm and an aggregate production
function 39
Note that even if technological change does not directly appear in the
production function, that is, even if for instance (2.20) is replaced by
= ( ) the economy may experience a rising standard of living when
is growing over time.
In contrast, disembodied technological change occurs when new technical
and organizational knowledge increases the combined productivity of the pro-
duction factors independently of when they were constructed or educated. If
the appearing in (2.20), (2.21), and (2.22) above refers to the total, histor-
ically accumulated capital stock as calculated by (2.24), then the evolution
of in these expressions can be seen as representing disembodied technolog-
ical change. All vintages of the capital equipment benefit from a rise in the
technology level No new investment is needed to benefit.
Based on data for the U.S. 1950-1990, and taking quality improvements
into account, Greenwood et al. (1997) estimate that embodied technological
progress explains about 60% of the growth in output per man hour. So,
empirically, embodied technological progress seems to play a dominant role.
As this tends not to be fully incorporated in national income accounting at
fixed prices, there is a need to adjust the investment levels in (2.24) to better
take estimated quality improvements into account. Otherwise the resulting
will not indicate the capital stock measured in efficiency units.
2.3 The concepts of a representative firm and
an aggregate production function
Many macroeconomic models make use of the simplifying notion of a rep-
resentative firm. By this is meant a fictional firm whose production “rep-
resents” aggregate production (value added) in a sector or in society as a
whole.
Suppose there are firms in the sector considered or in society as a
whole. Let be the production function for firm so that = ( )
where , and are output, capital input, and labor input, respectively,
= 1 2 . Further, let = Σ=1 , = Σ=1 and = Σ=1 .
Ignoring technological change, suppose the aggregate variables are related
through some function, ∗ (·) such that we can write
= ∗ ( )
and such that the choices of a single firm facing this production function
coincide with the aggregate outcomes, Σ=1 , Σ=1 and Σ=1 in the
original economy. Then ∗ ( ) is called the aggregate production function
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
40 CHAPTER 2. REVIEW OF TECHNOLOGY
or the production function of the representative firm. It is as if aggregate
production is the result of the behavior of such a single firm.
A simple example where the aggregate production function is well-defined
is the following. Suppose that all firms have the same production function,
i.e., (·) = (·) so that = ( ) = 1 2 If in addition has
CRS, we then have
= ( ) = ( 1) ≡ ( )
where ≡ Hence, facing given factor prices, cost minimizing firms
will choose
P the same P capital intensity = for all From = then
follows = so that = Thence,
X X X
≡ = ( ) = () = () = ( 1) = ( )
In this (trivial) case the aggregate production function is well-defined and
turns out to be exactly the same as the identical CRS production functions
of the individual firms.
Allowing for the existence of different production functions at firm level,
we may define the aggregate production function as
( ) = max 1 (1 1 ) + · · · + ( )
(1 1 )≥0
X X
s.t. ≤ ≤
Allowing also the existence of different output goods, different capital goods,
and different types of labor makes the issue more intricate, of course. Yet,
if firms are price taking profit maximizers and there are nonincreasing re-
turns to scale, we at least know that the aggregate outcome is as if, for
given prices, the firms jointly maximize aggregate profit on the basis of their
combined production technology (Mas-Colell et al., 1955). The problem is,
however, that the conditions needed for this to imply existence of an ag-
gregate production function which is well-behaved (in the sense of inheriting
simple qualitative properties from its constituent parts) are restrictive.
Nevertheless macroeconomics often treats aggregate output as a single ho-
mogeneous good and capital and labor as being two single and homogeneous
inputs. There was in the 1960s a heated debate about the problems involved
in this, with particular emphasis on the aggregation of different kinds of
equipment into one variable, the capital stock “”. The debate is known
as the “Cambridge controversy” because the dispute was between a group of
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
2.3. The concepts of a representative firm and an aggregate production
function 41
economists from Cambridge University, UK, and a group from Massachusetts
Institute of Technology (MIT), which is located in Cambridge, USA. The for-
mer group questioned the theoretical robustness of several of the neoclassical
tenets, including the proposition that rising aggregate capital intensity tends
to be associated with a falling rate of interest. Starting at the disaggregate
level, an association of this sort is not a logical necessity because, with differ-
ent production functions across the industries, the relative prices of produced
inputs tend to change, when the interest rate changes. While acknowledging
the possibility of “paradoxical” relationships, the latter group maintained
that in a macroeconomic context they are likely to cause devastating prob-
lems only under exceptional circumstances. In the end this is a matter of
empirical assessment.16
To avoid complexity and because, for many important issues in growth
theory, there is today no well-tried alternative, we shall in this course most
of the time use aggregate constructs like “ ”, “”, and “” as simplify-
ing devices, hopefully acceptable in a first approximation. There are cases,
however, where some disaggregation is pertinent. When for example the role
of imperfect competition is in focus, we shall be ready to disaggregate the
production side of the economy into several product lines, each producing its
own differentiated product. We shall also touch upon a type of growth models
where a key ingredient is the phenomenon of “creative destruction” meaning
that an incumbent technological leader is competed out by an entrant with
a qualitatively new technology.
Like the representative firm, the representative household and the aggre-
gate consumption function are simplifying notions that should be applied
only when they do not get in the way of the issue to be studied. The im-
portance of budget constraints may make it even more difficult to aggregate
over households than over firms. Yet, if (and that is a big if) all households
have the same, constant marginal propensity to consume out of income, ag-
gregation is straightforward and the representative household is a meaningful
concept. On the other hand, if we aim at understanding, say, the interaction
between lending and borrowing households, perhaps via financial intermedi-
aries, the representative household is not a useful starting point. Similarly,
if the theme is conflicts of interests between firm owners and employees, the
existence of different types of households should be taken into account.
16
In his review of the Cambridge controversy Mas-Colell (1989) concluded that: “What
the ‘paradoxical’ comparative statics [of disaggregate capital theory] has taught us is
simply that modelling the world as having a single capital good is not a priori justified.
So be it.”
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
42 CHAPTER 2. REVIEW OF TECHNOLOGY
2.4 Long-run vs. short-run production func-
tions*
Is the substitutability between capital and labor the same “ex ante” and “ex
post”? By ex ante is meant “when plant and machinery are to be decided
upon” and by ex post is meant “after the equipment is designed and con-
structed”. In the standard neoclassical competitive setup, of for instance
the Solow or the Ramsey model, there is a presumption that also after the
construction and installation of the equipment in the firm, the ratio of the
factor inputs can be fully adjusted to a change in the relative factor price. In
practice, however, when some machinery has been constructed and installed,
its functioning will often require a more or less fixed number of machine op-
erators. What can be varied is just the degree of utilization of the machinery.
That is, after construction and installation of the machinery, the choice op-
portunities are no longer described by the neoclassical production function
but by a Leontief production function,
= min(̄ ) 0 0 (2.26)
where ̄ is the size of the installed machinery (a fixed factor in the short
run) measured in efficiency units, is its utilization rate (0 ≤ ≤ 1) and
and are given technical coefficients measuring efficiency.
So in the short run the choice variables are and In fact, essentially
only is a choice variable since efficient production trivially requires =
̄ Under “full capacity utilization” we have = 1 (each machine is
used 24 hours per day seven days per week). “Capacity” is given as ̄ per
week. Producing efficiently at capacity requires = ̄ and the marginal
product by increasing labor input is here nil. But if demand, is less than
capacity, satisfying this demand efficiently requires = (̄) 1 and
= As long as 1 the marginal productivity of labor is a constant,
The various efficient input proportions that are possible ex ante may be
approximately described by a neoclassical CRS production function. Let this
function on intensive form be denoted = () When investment is decided
upon and undertaken, there is thus a choice between alternative efficient pairs
of the technical coefficients and in (2.26). These pairs satisfy
() = = (2.27)
So, for an increasing sequence of ’s, 1 2 . . . , . . . , the corresponding
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
2.4. Long-run vs. short-run production functions* 43
pairs are ( ) = (( ) ( )) = 1 2. . . .17 We say that ex ante,
depending on the relative factor prices as they are “now” and are expected
to evolve in the future, a suitable technique, ( ) is chosen from an
opportunity set described by the given neoclassical production function. But
ex post, i.e., when the equipment corresponding to this technique is installed,
the production opportunities are described by a Leontief production function
with ( ) = ( )
In the picturesque language of Phelps (1963), technology is in this case
putty-clay. Ex ante the technology involves capital which is “putty” in the
sense of being in a malleable state which can be transformed into a range of
various machinery requiring capital-labor ratios of different magnitude. But
once the machinery is constructed, it enters a “hardened” state and becomes
”clay”. Then factor substitution is no longer possible; the capital-labor ra-
tio at full capacity utilization is fixed at the level = as in (2.26).
Following the terminology of Johansen (1972), we say that a putty-clay tech-
nology involves a “long-run production function” which is neoclassical and a
“short-run production function” which is Leontief.
In contrast, the standard neoclassical setup assumes the same range of
substitutability between capital and labor ex ante and ex post. Then the
technology is called putty-putty. This term may also be used if ex post there
is at least some substitutability although less than ex ante. At the opposite
pole of putty-putty we may consider a technology which is clay-clay. Here
neither ex ante nor ex post is factor substitution possible. Table 2.1 gives an
overview of the alternative cases.
Table 2.1. Technologies classified according to
factor substitutability ex ante and ex post
Ex post substitution
Ex ante substitution possible impossible
possible putty-putty putty-clay
impossible clay-clay
The putty-clay case is generally considered the realistic case. As time
proceeds, technological progress occurs. To take this into account, we may
replace (2.27) and (2.26) by ( ) = = and = min( ̄ )
respectively. If a new pair of Leontief coefficients, (2 2 ) efficiency-
dominates its predecessor (by satisfying 2 ≥ 1 and 2 ≥ 1 with at
17
The points P and Q in the right-hand panel of Fig. 2.3 can be interpreted as con-
structed this way from the neoclassical production function in the left-hand panel of the
figure.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
44 CHAPTER 2. REVIEW OF TECHNOLOGY
least one strict equality), it may pay the firm to invest in the new technol-
ogy at the same time as some old machinery is scrapped. Real wages tend
to rise along with technological progress and the scrapping occurs because
the revenue from using the old machinery in production no longer covers the
associated labor costs.
The clay property ex-post of many technologies is important for short-run
analysis. It implies that there may be non-decreasing marginal productivity
of labor up to a certain point. It also implies that in its investment decision
the firm will have to take expected future technologies and future factor prices
into account. For many issues in long-run analysis the clay property ex-post
may be less important, since over time adjustment takes place through new
investment.
2.5 Literature notes
As to the question of the empirical validity of the constant returns to scale
assumption, Malinvaud (1998) offers an account of the econometric difficul-
ties associated with estimating production functions. Studies by Basu (1996)
and Basu and Fernald (1997) suggest returns to scale are about constant or
decreasing. Studies by Hall (1990), Caballero and Lyons (1992), Harris and
Lau (1992), Antweiler and Treffler (2002), and Harrison (2003) suggest there
are quantitatively significant increasing returns, either internal or external.
On this background it is not surprising that the case of IRS (at least at in-
dustry level), together with market forms different from perfect competition,
has in recent years received more attention in macroeconomics and in the
theory of economic growth.
Macroeconomists’ use of the value-laden term “technological progress” in
connection with technological change may seem suspect. But the term should
be interpreted as merely a label for certain types of shifts of isoquants in an
abstract universe. At a more concrete and disaggregate level analysts of
course make use of more refined notions about technological change, recog-
nizing for example not only benefits of new technologies, but also the risks,
including risk of fundamental mistakes (think of the introduction and later
abandonment of asbestos in the construction industry).
Informative history of technology is contained in Ruttan (2001) and Smil
(2003). For more general economic history, see, e.g., Clark (2008) and Pers-
son (2010). Forecasts of technological development in the next decades are
contained in, for instance, Brynjolfsson and McAfee (2014).
Embodied technological progress, sometimes called investment-specific
technological progress, is explored in, for instance, Solow (1960), Greenwood
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
2.6. References 45
et al. (1997), and Groth and Wendner (2014). Hulten (2001) surveys the
literature and issues related to measurement of the direct contribution of
capital accumulation and technological change, respectively, to productivity
growth.
Conditions ensuring that a representative household is admitted and the
concept of Gorman preferences are discussed in Acemoglu (2009). Another
useful source, also concerning the conditions for the representative firm to be
a meaningful notion, is Mas-Colell et al. (1995). For general discussions of the
limitations of representative agent approaches, see Kirman (1992) and Galle-
gati and Kirman (1999). Reviews of the “Cambridge Controversy” are con-
tained in Mas-Colell (1989) and Felipe and Fisher (2003). The last-mentioned
authors find the conditions required for the well-behavedness of these con-
structs so stringent that it is difficult to believe that actual economies are
in any sense close to satisfy them. For a less distrustful view, see for in-
stance Ferguson (1969), Johansen (1972), Malinvaud (1998), Jorgenson et al.
(2005), and Jones (2005).
It is often assumed that capital depreciation can be described as geomet-
ric (in continuous time exponential) evaporation of the capital stock. This
formula is popular in macroeconomics, more so because of its simplicity than
its realism. An introduction to more general approaches to depreciation is
contained in, e.g., Nickell (1978).
2.6 References
(incomplete)
Brynjolfsson, E., and A. McAfee, 2014, The Second Machine Age, Norton.
Clark, G., 2008, A Farewell to Alms: A Brief Economic History of the
World, Princeton University Press.
Persson, K. G., 2010, An economic history of Europe. Knowledge, insti-
tutions and growth, 600 to the present, Cambridge University Press: Cam-
bridge.
Ruttan, V. W. , 2001, Technology, Growth, and Development: An Induced
Innovation Perspective, Oxford University Press: Oxford.
Smil, V., 2003, Energy at the crossroads. Global perspectives and uncer-
tainties, MIT Press: Cambridge Mass.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
46 CHAPTER 2. REVIEW OF TECHNOLOGY
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
Chapter 3
Continuous time analysis
Because dynamic analysis is generally easier in continuous time, growth mod-
els are often stated in continuous time. This chapter gives an account of the
conceptual aspects of continuous time analysis. Appendix A considers sim-
ple growth arithmetic in continuous time. And Appendix B provides solution
formulas for linear first-order differential equations.
3.1 The transition from discrete time to con-
tinuous time
We start from a discrete time framework. The run of time is divided into
successive periods of equal length, taken as the time-unit. Let us here index
the periods by = 0 1 2 . Thus financial wealth accumulates according to
+1 − = 0 given,
where is (net) saving in period
3.1.1 Multiple compounding per year
With time flowing continuously, we let () refer to financial wealth at time
Similarly, ( + ∆) refers to financial wealth at time + ∆ To begin with,
let ∆ equal one time unit. Then (∆) equals () and is of the same value
as Consider the forward first difference in ∆() ≡ ( + ∆) − () It
makes sense to consider this change in in relation to the length of the time
interval involved, that is, to consider the ratio ∆()∆ As long as ∆ = 1
with = ∆ we have ∆()∆ = (+1 − )1 = +1 − Now, keep
the time unit unchanged, but let the length of the time interval [ + ∆)
47
48 CHAPTER 3. CONTINUOUS TIME ANALYSIS
approach zero, i.e., let ∆ → 0. When is a differentiable function of , we
have
∆() ( + ∆) − () ()
lim = lim =
∆→0 ∆ ∆→0 ∆
where () often written ̇() is known as the derivative of (·) at the
point Wealth accumulation in continuous time can then be written
̇() = () (0) = 0 given, (3.1)
where () is the saving flow at time . For ∆ “small” we have the approx-
imation ∆() ≈ ̇()∆ = ()∆ In particular, for ∆ = 1 we have ∆()
= ( + 1) − () ≈ ()
As time unit choose one year. Going back to discrete time, if wealth
grows at a constant rate 0 per year, then after periods of length one
year, with annual compounding, we have
= 0 (1 + ) = 0 1 2 . (3.2)
If instead compounding (adding saving to the principal) occurs times a
year, then after periods of length 1 year and a growth rate of per
such period,
= 0 (1 + ) (3.3)
With still denoting time measured in years passed since date 0, we have
= periods. Substituting into (3.3) gives
∙ ¸
1
() = = 0 (1 + ) = 0 (1 + ) where ≡
We keep and fixed, but let → ∞ Thus → ∞ Then, in the limit
there is continuous compounding and it can be shown that
() = 0 (3.4)
where is a mathematical constant called the base of the natural logarithm
and defined as ≡ lim→∞ (1 + 1) ' 2.7182818285....
The formula (3.4) is the continuous-time analogue to the discrete time
formula (3.2) with annual compounding. A geometric growth factor is re-
placed by an exponential growth factor, and this growth factor is valid
for any in the time interval (− 1 2 ) for which the growth rate of equals
the constant ( 1 and 2 being some positive real numbers).
We can also view the formulas (3.2) and (3.4) as the solutions to a differ-
ence equation and a differential equation, respectively. Thus, (3.2) is the so-
lution to the linear difference equation +1 = (1+) , given the initial value
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
3.1. The transition from discrete time to continuous time 49
0 And (3.4) is the solution to the linear differential equation ̇() = ()
given the initial condition (0) = 0 Now consider a time-dependent growth
rate, () The corresponding differential equation is ̇() = ()() and it
has the solution
() = (0) 0 ( ) (3.5)
R
where the exponent, 0 ( ) , is the definite integral of the function ( )
from 0 to The result (3.5) is called the basic accumulation formula in
continuous time and the factor 0 ( ) is called the growth factor or the
accumulation factor.
3.1.2 Compound interest and discounting
Let () denote the short-term real interest rate in continuous time at time .
To clarify what is meant by this, consider a deposit of () euro on a drawing
account in a bank at time . If the general price level in the economy at time
is () euro, the real value of the deposit is () = () () at time
By definition the real rate of return on the deposit in continuous time (with
continuous compounding) at time is the (proportionate) instantaneous rate
at which the real value of the deposit expands per time unit when there is
no withdrawal from the account. Thus, if the instantaneous nominal interest
rate is () we have ̇ () () = () and so, by the fraction rule in continuous
time (cf. Appendix A),
̇() ̇ () ̇ ()
() = = − = () − () (3.6)
() () ()
where () ≡ ̇ () () is the instantaneous inflation rate. In contrast to the
corresponding formula in discrete time, this formula is exact. Sometimes ()
and () are referred to as the nominal and real interest intensity, respectively,
or the nominal and real force of interest.
Calculating the terminal value of the deposit at time 1 0 given its
value at time 0 and assuming no withdrawal in the time interval [0 1 ], the
accumulation formula (3.5) immediately yields
1
()
(1 ) = (0 ) 0
When calculating present values in continuous time analysis, we use com-
pound discounting. We reverse the accumulation formula and go from the
compounded or terminal value to the present value (0 ). Similarly, given a
consumption plan, (())=
1
0
, the present value of this plan as seen from time
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
50 CHAPTER 3. CONTINUOUS TIME ANALYSIS
0 is Z 1
= () − (3.7)
0
presupposing a constant interest rate. Instead of the geometric discount
factor, 1(1 + ) from discrete time analysis, we have here an exponential
discount factor, 1( ) = − and instead of a sum, an integral. When the
interest rate varies over time, (3.7) is replaced by
Z 1
− ( )
= () 0
0
In (3.7) () is discounted by − ≈ (1 + )− for “small”. This might
not seem analogue to the discrete-time discounting in (??) where it is −1
that is discounted by (1 + )− assuming a constant interest rate. When
taking into account the timing convention that payment for −1 in period
− 1 occurs at the end of the period (= time ) there is no discrepancy,
however, since the continuous-time analogue to this payment is ().
3.2 The allowed range for parameter values
The allowed range for parameters may change when we go from discrete time
to continuous time with continuous compounding. For example, the usual
equation for aggregate capital accumulation in continuous time is
̇() = () − () (0) = 0 given, (3.8)
where () is the capital stock, () is the gross investment at time and
≥ 0 is the (physical) capital depreciation rate. Unlike in discrete time, here
1 is conceptually allowed. Indeed, suppose for simplicity that () = 0
for all ≥ 0; then (3.8) gives () = 0 − . This formula is meaningful for
any ≥ 0 Usually, the time unit used in continuous time macro models is
one year (or, in business cycle theory, rather a quarter of a year) and then a
realistic value of is of course 1 (say, between 0.05 and 0.10). However, if
the time unit applied to the model is large (think of a Diamond-style OLG
model), say 30 years, then 1 may fit better, empirically, if the model
is converted into continuous time with the same time unit. Suppose, for
example, that physical capital has a half-life of 10 years. With 30 years as
our time unit, inserting into the formula 12 = −3 gives = (ln 2) · 3 ' 2
In many simple macromodels, where the level of aggregation is high, the
relative price of a unit of physical capital in terms of the consumption good
is 1 and thus constant. More generally, if we let the relative price of the
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
3.3. Stocks and flows 51
capital good in terms of the consumption good at time be () and allow
̇() 6= 0 then we have to distinguish between the physical depreciation
of capital, and the economic depreciation, that is, the loss in economic
value of a machine per time unit. The economic depreciation will be () =
() − ̇() namely the economic value of the physical wear and tear (and
technological obsolescence, say) minus the capital gain (positive or negative)
on the machine.
Other variables and parameters that by definition are bounded from below
in discrete time analysis, but not so in continuous time analysis, include rates
of return and discount rates in general.
3.3 Stocks and flows
An advantage of continuous time analysis is that it forces the analyst to make
a clear distinction between stocks (say wealth) and flows (say consumption
or saving). Recall, a stock variable is a variable measured as a quantity at a
given point in time. The variables () and () considered above are stock
variables. A flow variable is a variable measured as quantity per time unit
at a given point in time. The variables () ̇() and () are flow variables.
One can not add a stock and a flow, because they have different de-
nominations. What is meant by this? The elementary measurement units
in economics are quantity units (so many machines of a certain kind or so
many liters of oil or so many units of payment, for instance) and time units
(months, quarters, years). On the basis of these elementary units we can form
composite measurement units. Thus, the capital stock, has the denomi-
nation “quantity of machines”, whereas investment, has the denomination
“quantity of machines per time unit” or, shorter, “quantity/time”. A growth
rate or interest rate has the denomination “(quantity/time)/quantity” =
“time−1 ”. If we change our time unit, say from quarters to years, the value
of a flow variable as well as a growth rate is changed, in this case quadrupled
(presupposing annual compounding).
In continuous time analysis expressions like ()+() or ()+ ̇() are
thus illegitimate. But one can write ( + ∆) ≈ () + (() − ())∆ or
̇()∆ ≈ (() − ())∆ In the same way, suppose a bath tub at time
contains 50 liters of water and that the tap pours 12 liter per second into the
tub for some time. Then a sum like 50 + 12 (/sec) does not make sense. But
the amount of water in the tub after one minute is meaningful. This amount
would be 50 + 12 · 60 ((/sec)×sec) = 80 . In analogy, economic flow
variables in continuous time should be seen as intensities defined for every
in the time interval considered, say the time interval [0, ) or perhaps
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
52 CHAPTER 3. CONTINUOUS TIME ANALYSIS
s (t )
s(t0 )
t
0 t0 t 0 t
Figure 3.1: With ∆ “small” the integral of () from 0 to 0 +∆ is ≈ the hatched
area.
[0, ∞). For example, when we say that () is “investment” at time ,
this is really a short-hand for “investment intensity” at time . The actual
investment in a time interval [0 0 + ∆) i.e., the invested amount during
R +∆
this time interval, is the integral, 00 () ≈ (0 )∆ Similarly, the flow
of individual saving, () should be interpreted as the saving intensity at
time The actual saving in a time interval [0 0 + ∆) i.e., the saved (or
R +∆
accumulated) amount during this time interval, is the integral, 00 ()
If ∆ is “small”, this integral is approximately equal to the product (0 )· ∆,
cf. the hatched area in Figure 3.1.
The notation commonly used in discrete time analysis blurs the distinc-
tion between stocks and flows. Expressions like +1 = + without further
comment, are usual. Seemingly, here a stock, wealth, and a flow, saving, are
added. In fact, however, it is wealth at the beginning of period and the
saved amount during period that are added: +1 = + · ∆. The tacit
condition is that the period length, ∆ is the time unit, so that ∆ = 1.
But suppose that, for example in a business cycle model, the period length
is one quarter, but the time unit is one year. Then saving in quarter is
= (+1 − ) · 4 per year.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
3.4. The choice between discrete and continuous time formulation 53
3.4 The choice between discrete and contin-
uous time formulation
In empirical economics, data typically come in discrete time form and data
for flow variables typically refer to periods of constant length. One could
argue that this discrete form of the data speaks for discrete time rather than
continuous time modelling. And the fact that economic actors often think
and plan in period terms, may seem a good reason for putting at least mi-
croeconomic analysis in period terms. Nonetheless real time is continuous.
And, as for instance Allen (1967) argued, it can hardly be said that the
mass of economic actors think and plan with one and the same period. In
macroeconomics we consider the sum of the actions. In this perspective the
continuous time approach has the advantage of allowing variation within the
usually artificial periods in which the data are chopped up. And centralized
asset markets equilibrate very fast and respond immediately to new infor-
mation. For such markets a formulation in continuous time seems a better
approximation.
There is also a risk that a discrete time model may generate artificial
oscillations over time. Suppose the “true” model of some mechanism is given
by the differential equation
̇ = −1 (3.9)
The solution is () = (0) which converges in a monotonic way toward 0
for → ∞ However, the analyst takes a discrete time approach and sets up
the seemingly “corresponding” discrete time model
+1 − =
This yields the difference equation +1 = (1 + ) , where 1 + 0 The
solution is = (1 + ) 0 = 0 1 2 As (1 + ) is positive when is
even and negative when is odd, oscillations arise (together with divergence
if −2) in spite of the “true” model generating monotonous convergence
towards the steady state ∗ = 0.
It should be added, however, that this potential problem can always be
avoided within discrete time models by choosing a sufficiently short period
length. Indeed, the solution to a differential equation can always be ob-
tained as the limit of the solution to a corresponding difference equation for
the period length approaching zero. In the case of (3.9) the approximating
difference equation is +1 = (1 + ∆) where ∆ is the period length,
= ∆, and = (∆) By choosing ∆ small enough, the solution comes
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
54 CHAPTER 3. CONTINUOUS TIME ANALYSIS
arbitrarily close to the solution of (3.9). It is generally more difficult to go
in the opposite direction and find a differential equation that approximates
a given difference equation. But the problem is solved as soon as a differ-
ential equation has been found that has the initial difference equation as an
approximating difference equation.
From the point of view of the economic contents, the choice between
discrete time and continuous time may be a matter of taste. Yet, everything
else equal, the clearer distinction between stocks and flows in continuous
time than in discrete time speaks for the former. From the point of view
of mathematical convenience, the continuous time formulation, which has
worked so well in the natural sciences, is preferable. At least this is so in
the absence of uncertainty. For problems where uncertainty is important,
discrete time formulations are easier to work with unless one is familiar with
stochastic calculus.
3.5 Appendix A: Growth arithmetic in con-
tinuous time
Let the variables and be differentiable functions of time Suppose
() () and () are positive for all Then:
̇() ̇() ̇()
PRODUCT RULE () = ()() ⇒ ()
= ()
+ ()
Proof. Taking logs on both sides of the equation () = ()() gives ln ()
= ln () + ln (). Differentiation w.r.t. , using the chain rule, gives the
conclusion. ¤
The procedure applied in this proof is called logarithmic differentiation
w.r.t.
() ̇() ̇() ̇()
FRACTION RULE () = ()
⇒ ()
= ()
− ()
The proof is similar.
̇() ̇()
POWER FUNCTION RULE () = () ⇒ ()
= ()
The proof is similar.
In continuous time these simple formulas are exactly true. In discrete time
the analogue formulas are only approximately true and the approximation
can be quite bad unless the growth rates of and are small.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
3.6. Appendix B: Solution formulas for linear differential equations of first
order 55
3.6 Appendix B: Solution formulas for linear
differential equations of first order
For a general differential equation of first order, ̇() = (() ) with
(0 ) = 0 and where is a continuous function, we have, at least for
in an interval (− +) for some 0
Z
() = 0 + (( ) ) (*)
0
To get a confirmation, calculate ̇() from (*).
For the special case of a linear differential equation of first order, ̇() +
()() = () we can specify the solution. Three sub-cases of rising com-
plexity are:
1. ̇() + () = with 6= 0 and initial condition (0 ) = 0 Solution:
() = (0 − ∗ )−(−0 ) + ∗ where ∗ =
If = 0 we get, directly from (*), the solution () = 0 + 1
2. ̇() + () = () with initial condition (0 ) = 0 Solution:
Z
−(−0 ) −(−0 )
() = 0 + ()(−0 )
0
Special case: () = with 6= − and initial condition (0 ) = 0
Solution:
Z
−(−0 ) −(−0 ) (−0 )
() = 0 + (+)(−0 ) = (0 − )−(−0 ) +
0 + +
3. ̇() + ()() = () with initial condition (0 ) = 0 Solution:
Z
− ( ) − ( ) ( )
() = 0 0 + 0 () 0
0
1
Some non-linear differential equations can be transformed into this simple case. For
simplicity let 0 = 0 Consider the equation ̇() = () 0 0 given, 6= 0 6= 1
(a Bernoulli equation). To find the solution for () let () ≡ ()1− Then, ̇()
= (1 − )()− ̇() = (1 − )()− () = (1 − ) The solution for this is ()
= 0 + (1 − ) where 0 = 01− Thereby the solution for () is () = ()1(1−)
³ ´1(1−)
= 01− + (1 − ) which is defined for −01− ((1 − )
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
56 CHAPTER 3. CONTINUOUS TIME ANALYSIS
Special case: () = 0 Solution:
() = 0 − 0 ( )
Even more special case: () = 0 and () = a constant. Solution:
() = 0 −(−0 )
Remark 1 For 0 = 0 most of the formulas will look simpler.
Remark 2 To check whether a suggested solution is a solution, calculate
the time derivative of the suggested solution and add an arbitrary constant.
By appropriate adjustment of the constant, the final result should be a repli-
cation of the original differential equation together with its initial condition.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
Chapter 4
Skill-biased technical change.
Balanced growth theorems
This chapter is both an alternative and a supplement to the pages 60-64
in Acemoglu, where the concepts of neutral technical change and balanced
growth, including Uzawa’s theorem, are discussed.
Since “neutral” technical change should be seen in relation to “biased”
technical change, Section 1 below introduces the concept of “biased” tech-
nical change. Like regarding neutral technical change, also regarding biased
technical change there exist three different definitions, Hicks’, Harrod’s, and
what the literature has dubbed “Solow’s”. Below we concentrate on Hick’s
definition − with an application to the role of technical change for the evo-
lution of the skill premium. So the focus is on the production factors skilled
and unskilled labor rather than capital and labor. While regarding capital
and labor it is Harrod’s classifications that are most used in macroeconomics,
regarding skilled and unskilled labor it is Hicks’.
The remaining sections discuss the concept of balanced growth and present
three fundamental propositions about balanced growth. In view of the gen-
erality of the propositions, they have a broad field of application. Our propo-
sitions 1 and 2 are slight extensions of part 1 and 2, respectively, of what
Acemoglu calls Uzawa’s Theorem I (Acemoglu, 2009, p. 60). Our Proposi-
tion 3 essentially corresponds to what Acemoglu calls Uzawa’s Theorem II
(Acemoglu, 2009, p. 63).
57
CHAPTER 4. SKILL-BIASED TECHNICAL CHANGE.
58 BALANCED GROWTH THEOREMS
4.1 The rising skill premium
4.1.1 Skill-biased technical change in the sense of Hicks:
An example
Let aggregate output be produced through a differentiable three-factor pro-
duction function ̃ :
= ̃ ( 1 2 )
where is capital input, 1 is input of unskilled labor (also called blue-collar
labor below), and 2 is input of skilled labor. Suppose technological change
is such that the production function can be rewritten
̃ ( 1 2 ) = ( (1 2 )) (4.1)
where the “nested” function (1 2 ) represents input of a “human cap-
ital” aggregate. Let be CRS-neoclassical w.r.t. and and let
be CRS-neoclassical w.r.t. (1 2 ) Finally, let 0. So “technical
change” amounts to “technical progress”.
In equilibrium under perfect competition in the labor markets the relative
wage, often called the “skill premium”, will be
2 2 2 2 (1 2 ) 2 (1 2 1 )
= = = = (4.2)
1 1 1 1 (1 2 ) 1 (1 2 1 )
where we have used Euler’s theorem (saying that if is homogeneous of
degree one in its first two arguments, then the partial derivatives of are
homogeneous of degree zero w.r.t. these arguments).
Time is continuous (nevertheless the time argument of a variable, is in
this section written as a subscript ). Hicks’ definitions are now: If for all
2 1 0
³ ´
2 (12 1 )
1 (12 1 )
2
T 0 then technical change is
constant
1
⎧
⎨ skill-biased in the sense of Hicks,
skill-neutral in the sense of Hicks.(4.3)
⎩
blue collar-biased in the sense of Hicks,
respectively.
In the US the skill premium (measured by the wage ratio for college
grads vis-a-vis high school grads) has had an upward trend since 1950 (see
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
4.1. The rising skill premium 59
for instance Jones and Romer, 2010).1 If in the same period the relative
supply of skilled labor had been roughly constant, by (4.3) in combination
with (4.2), a possible explanation could be that technological change has
been skill-biased in the sense of Hicks. In reality, in the same period also the
relative supply of skilled labor has been rising (in fact even faster than the
skill premium). Since in spite of this the skill premium has risen, it suggests
that the extend of “skill-biasedness” has been even stronger.
We may alternatively put it this way. As the function is CRS-neoclassical
w.r.t. 1 and 2 we have 22 0 and 12 0 cf. Chapter 2. Hence, by
(4.2), a rising 2 1 without technical change would imply a declining skill
premium. That the opposite has happened must, within our simple model,
be due to (a) there has been technical change, and (b) technical change
has favoured skilled labor (which means that technical change has been skill-
biased in the sense of Hicks).
An additional aspect of the story is that skill-biasedness helps explain
the observed increase in the relative supply of skilled labor. If for a constant
relative supply of skilled labor, the skill premium is increasing, this increase
strengthens the incentive to go to college. Thereby the relative supply of
skilled labor (reflecting the fraction of skilled labor in the labor force) tends
to increase.
4.1.2 Capital-skill complementarity
An additional potential source of a rising skill premium is capital-skill com-
plementarity. Let the aggregate production function be
= ̃ ( 1 2 ) = ( 1 1 2 2 ) = (+1 1 ) (2 2 )1− 0 1
where 1 and 2 are technical coefficients that may be rising over time.
In this production function capital and unskilled labor are perfectly substi-
tutable (the partial elasticity of factor substitution between them is +∞) On
the other hand there is direct complementarity between capital and skilled
labor, i.e., 2 (2 ) 0
Under perfect competition the skill premium is
2 2 ( + 1 1 ) (1 − )(2 2 )− 2
= = (4.4)
1 2 ( + 1 1 )−1 1 (2 2 )1−
µ ¶
1 − + 1 1 2
=
2 2 1
1
On the other hand, over the years 1915 - 1950 the skill premium had a downward
trend (Jones and Romer, 2010).
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
CHAPTER 4. SKILL-BIASED TECHNICAL CHANGE.
60 BALANCED GROWTH THEOREMS
Here, if technical change is absent (1 and 2 constant), a rising capital
stock will, for fixed 1 and 2 raise the skill premium.
A more realistic scenario is, however, a situation with an approximately
constant real interest rate, cf. Kaldor’s stylized facts. We have, again by
perfect competition,
µ ¶−1
−1 1− + 1 1
= ( + 1 1 ) (2 2 ) = = + (4.5)
2 2
where is the real interest rate at time and is the (constant) capital
depreciation rate. For = a constant, (4.5) gives
µ ¶− 1−
1
+ 1 1 +
= ≡ (4.6)
2 2
a constant. In this case, (4.4) shows that capital-skill complementarity is
not sufficient for a rising skill premium. A rising skill premium requires that
technical change brings about a rising 2 1 . So again an observed rising
skill premium, along with a more or less constant real interest rate, suggests
that technical change is skill-biased.
We may rewrite (4.6) as
1 1
=−
2 2 2 2
where the conjecture is that 1 1 (2 2 ) → 0 for → ∞ The analysis
suggests the following story. Skill-biased technical progress generates rising
productivity as well as a rising skill premium. The latter induces more and
more people to go to college. The rising level of education in the labor force
raises productivity further. This is a basis for further capital accumulation,
continuing to replace unskilled labor, and so on.
In particular since the early 1980s the skill premium has been sharply
increasing in the US (see Acemoglu, p. 498). This is also the period where
ICT technologies took off.
4.2 Balanced growth and constancy of key ra-
tios
The focus now shifts to homogeneous labor vis-a-vis capital.
We shall state general definitions of the concepts of “steady state” and
“balanced growth”, concepts that are related but not identical. With respect
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
4.2. Balanced growth and constancy of key ratios 61
to “balanced growth” this implies a minor deviation from the way Acemoglu
briefly defines it informally on his page 57. The main purpose of the present
chapter is to lay bare the connections between these two concepts as well
as their relation to the hypothesis of Harrod-neutral technical progress and
Kaldor’s stylized facts.
4.2.1 The concepts of steady state and balanced growth
A basic equation in many one-sector growth models for a closed economy in
continuous time is
̇ = − = − − ≡ − (4.7)
where is aggregate capital, aggregate gross investment, aggregate
output, aggregate consumption, aggregate gross saving (≡ − ), and
≥ 0 is a constant physical capital depreciation rate.
Usually, in the theoretical literature on dynamic models, a steady state is
defined in the following way:
Definition 3 A steady state of a dynamic model is a stationary solution to
the fundamental differential equation(s) of the model.
Or briefly: a steady state is a stationary point of a dynamic process.
Let us take the Solow growth model as an example. Here gross saving
equals where is a constant, 0 1 Aggregate output is given by a
neoclassical production function, with CRS and Harrod-neutral technical
progress: = ( ) = (̃ 1) ≡ (̃) where is the labor
force, is the level of technology, and ̃ ≡ () is the (effective) capital
intensity. Moreover, 0 0 and 00 0 Solow assumes () = (0) and
() = (0) , where ≥ 0 and ≥ 0 are the constant growth rates of the
labor force and technology, respectively. By log-differentiating ̃ w.r.t. 2
we end up with the fundamental differential equation (“law of motion”) of
the Solow model: ·
̃ = (̃) − ( + + )̃ (4.8)
∗
Thus, in the Solow model, a (non-trivial) steady state is a ̃ 0 such that,
·
if ̃ = ̃∗ then ̃ = 0 In passing we note that, by (4.8), such a ̃∗ must
satisfy the equation (̃∗ )̃ ∗ = ( + + ) and in view of 00 0 it is
unique if it exists.
The most common definition in the literature of balanced growth for an
aggregate economy is the following:
2
Or by directly using the fraction rule, see Appendix A to Chapter 3.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
CHAPTER 4. SKILL-BIASED TECHNICAL CHANGE.
62 BALANCED GROWTH THEOREMS
Definition 4 A balanced growth path is a path ( )∞ =0 along which the
quantities and are positive and grow at constant rates (not necessarily
positive and not necessarily the same).
Acemoglu, however, defines (Acemoglu, 2009, p. 57) balanced growth
in the following way: “balanced growth refers to an allocation where output
grows at a constant rate and capital-output ratio, the interest rate, and factor
shares remain constant”. My problem with this definition is that it mixes
growth of aggregate quantities with income distribution aspects (interest rate
and factor income shares). And it is not made clear what is meant by the
output-capital ratio if the relative price of capital goods is changing over
time. So I stick to the definition above which is quite standard and is known
to function well in many different contexts.
Note that in the Solow model (as well as in many other models) we have
that if the economy is in a steady state, ̃ = ̃ ∗ then the economy features
balanced growth. Indeed, a steady state of the Solow model implies by
definition that ̃ ≡ () is constant. Hence must grow at the same
constant rate as namely + In addition, = (̃ ∗ ) in a steady
state, showing that also must grow at the constant rate + And so
must then = (1 − ) So in a steady state of the Solow model the path
followed by ( )∞=0 is a balanced growth path.
As we shall see in the next section, in the Solow model (and many other
models) the reverse also holds: if the economy features balanced growth,
then it is in a steady state. But this equivalence between steady state and
balanced growth does not hold in all models.
4.2.2 A general result about balanced growth
An interesting fact is that, given the dynamic resource constraint (4.7), we
have always that if there is balanced growth with positive gross saving, then
the ratios and are constant (by “always” is meant: indepen-
dently of how saving is determined and of how the labor force and technology
evolve). And also the other way round: as long as gross saving is positive,
constancy of the and ratios is enough to ensure balanced growth.
So balanced growth and constancy of key ratios are essentially equivalent.
This is a very practical general observation. And since Acemoglu does not
state any balanced growth theorem at this general level, we shall do it here,
together with a proof. Letting denote the growth rate of the (positively
valued) variable i.e., ≡ ̇ we claim:
Proposition 1 (the balanced growth equivalence theorem). Let ( )∞
=0
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
4.2. Balanced growth and constancy of key ratios 63
be a path along which , and ≡ − are positive for all ≥ 0
Then, given the accumulation equation (4.7), the following holds:
(i) if there is balanced growth, then = = and the ratios
and are constant;
(ii) if and are constant, then and grow at the same
constant rate, i.e., not only is there balanced growth, but the growth
rates of and are the same.
Proof Consider a path ( )∞ =0 along which , and ≡ −
are positive for all ≥ 0 (i) Assume there is balanced growth. Then, by
definition, and are constant. Hence, by (4.7), we have that =
+ is constant, implying
= (*)
Further, since = +
̇ ̇ ̇
= = + = + = + (by (*))
−
= + = ( − ) + (**)
Now, let us provisionally assume that 6= Then (**) gives
−
= (***)
−
which is a constant since and are constant. Constancy of
requires that = hence, by (***), = 1 i.e., = In view
of = + , however, this outcome contradicts the given condition that
0 Hence, our provisional assumption and its implication, (***), are
falsified. Instead we have = . By (**), this implies = = but
now without the condition = 1 being implied. It follows that and
are constant.
(ii) Suppose and are constant. Then = = , so that
is a constant. We now show that this implies that is constant.
Indeed, from (4.7), = 1 − so that also is constant. It follows
that = = so that is constant. By (4.7),
̇ +
= = +
so that is constant. This, together with constancy of and
implies that also and are constant. ¤
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
CHAPTER 4. SKILL-BIASED TECHNICAL CHANGE.
64 BALANCED GROWTH THEOREMS
Remark. It is part (i) of the proposition which requires the assumption 0
for all ≥ 0 If = 0 we would have = − and ≡ − = hence
= for all ≥ 0 Then there would be balanced growth if the common
value of and had a constant growth rate. This growth rate, however,
could easily differ from that of Suppose = 1− = and =
( and constants). Then we would have = = − +(1−) which
could easily be strictly positive and thereby different from = − ≤ 0 so
that (i) no longer holds. ¤
The nice feature is that this proposition holds for any model for which
the simple dynamic resource constraint (4.7) is valid. No assumptions about
for example CRS and other technology aspects or about market form are
involved. Note also that Proposition 1 suggests a link from balanced growth
to steady state. And such a link is present in for instance the Solow model.
Indeed, by (i) of Proposition 1, balanced growth implies constancy of
which in the Solow model implies that (̃)̃ is constant. In turn, the latter
is only possible if ̃ is constant, that is, if the economy is in steady state.
There exist cases, however, where this equivalence does not hold (some
open economy models and some models with embodied technological change,
see Groth et al., 2010). Therefore, it is recommendable always to maintain
a distinction between the terms steady state and balanced growth.
4.3 The crucial role of Harrod-neutrality
Proposition 1 suggests that if one accepts Kaldor’s stylized facts (see Chapter
1) as a characterization of the past century’s growth experience, and if one
wants a model consistent with them, one should construct the model such
that it can generate balanced growth. For a model to be capable of generating
balanced growth, however, technological progress must be of the Harrod-
neutral type (i.e., be labor-augmenting), at least in a neighborhood of the
balanced growth path. For a fairly general context (but of course not as
general as that of Proposition 1), this was shown already by Uzawa (1961).
We now present a modernized version of Uzawa’s contribution.
Let the aggregate production function be
() = ̃ (() () ) 0 (4.9)
where is a constant that depends on measurement units. The only tech-
nology assumption needed is that ̃ has CRS w.r.t. the first two arguments
(̃ need not be neoclassical for example). As a representation of technical
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
4.3. The crucial role of Harrod-neutrality 65
progress, we assume ̃ 0 for all ≥ 0 (i.e., as time proceeds, un-
changed inputs result in more and more output). We also assume that the
labor force evolves according to
() = (0) (4.10)
where is a constant. Further, non-consumed output is invested and so (4.7)
is the dynamic resource constraint of the economy.
Proposition 2 (Uzawa’s balanced growth theorem) Let = ( () () ())∞ =0 ,
where 0 () () for all ≥ 0 be a path satisfying the capital accumu-
lation equation (4.7), given the CRS-production function (4.9) and the labor
force path in (4.10). Then:
(i) a necessary condition for this path to be a balanced growth path is that
along the path it holds that
() = ̃ (() () ) = ̃ (() ()() 0) (4.11)
where () = with ≡ − ;
(ii) for any 0 such that there is a + + with the property
that the production function ̃ in (4.9) allows an output-capital ratio
equal to at = 0 (i.e., ̃ (1 ̃−1 0) = for some real number ̃ 0),
a sufficient condition for the path P to be a balanced growth path with
output-capital ratio , is that the technology can be written as in (4.11)
with () = .
Proof (i)3 Suppose the path ( () () ())∞
=0 is a balanced growth path.
By definition, and are then constant, so that () = (0) and
() = (0) We then have
()− = (0) = ̃ ((0) (0) 0) = ̃ (()− ()− 0) (*)
where we have used (4.9) with = 0 In view of the precondition that ()
≡ () − () 0 we know from (i) of Proposition 1, that is constant
so that = . By CRS, (*) then implies
() = ̃ (() − () − 0) = ̃ (() ( −) () 0)
We see that (4.11) holds for () = with ≡ −
3
This part draws upon Schlicht (2006), who generalized a proof in Wan (1971, p. 59)
for the special case of a constant saving rate.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
CHAPTER 4. SKILL-BIASED TECHNICAL CHANGE.
66 BALANCED GROWTH THEOREMS
(ii) Suppose (4.11) holds with () = Let 0 be given such that
there is a + + with the property that
̃ (1 ̃−1 0) = (**)
for some constant ̃ 0 Our strategy is to prove the claim in (ii) by con-
struction of a path = ( () () ())∞ =0 which satisfies it. We let
be such that the saving-income ratio is a constant ≡ ( + + ), i.e.,
() − () ≡ () = () for all ≥ 0 Inserting this, together with () =
(̃())()(), where (̃()) ≡ ̃ (̃() 1 0) and ̃() ≡ ()(()())
into (4.7), we get the Solow equation (4.8). Hence ̃() is constant if and
only if ̃() satisfies the equation (̃())̃() = ( + + ) ≡ By (**)
and the definition of the required value of ̃() is ̃ which is thus the
steady-state for the constructed Solow equation. Letting (0) satisfy (0)
= ̃(0) where = (0) we thus have ̃(0) = (0)((0)(0)) = ̃ So
that the initial value of ̃() equals the steady state value. It now follows
that ̃() = ̃ for all ≥ 0 and so ()() = (̃())̃() = (̃)̃ =
for all ≥ 0 In addition, () = (1 − ) () so that () () is constant
along the path By (ii) of Proposition 1 now follows that the path is a
balanced growth path, as was to be proved. ¤
The form (4.11) indicates that along a balanced growth path, technical
progress must be purely “labor augmenting”, that is, Harrod-neutral. It is in
this case convenient to define a new CRS function, by (() ()())
≡ ̃ (() ()() 0) Then (i) of the proposition implies that at least along
the balanced growth path, we can rewrite the production function this way:
() = ̃ (() (0)() ) = (() ()()) (4.12)
where (0) = () = (0) with ≡ −
It is important to recognize that the occurrence of Harrod-neutrality says
nothing about what the source of technological progress is. Harrod-neutrality
should not be interpreted as indicating that the technological progress em-
anates specifically from the labor input. Harrod-neutrality only means that
technical innovations predominantly are such that not only do labor and cap-
ital in combination become more productive, but this happens to manifest
itself at the aggregate level in the form (4.12).4
What is the intuition behind the Uzawa result that for balanced growth to
be possible, technical progress must have the purely labor-augmenting form?
4
For a CRS Cobb-Douglas production function with technological progress, Harrod-
neutrality is present whenever the output elasticity w.r.t capital (often denoted ) is
constant over time.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
4.3. The crucial role of Harrod-neutrality 67
First, notice that there is an asymmetry between capital and labor. Capital
is an accumulated amount of non-consumed output. In contrast, in simple
macro models labor is a non-produced production factor which (at least in
the context of (4.10)) grows in an exogenous way. Second, because of CRS,
the original formulation, (4.9), of the production function implies that
() ()
1 = ̃ ( ) (4.13)
() ()
Now, since capital is accumulated non-consumed output, it tends to inherit
the trend in output such that () () must be constant along a balanced
growth path (this is what Proposition 1 is about). Labor does not inherit the
trend in output; indeed, the ratio () () is free to adjust as time proceeds.
When there is technical progress ( ̃ 0) along a balanced growth path,
this progress must manifest itself in the form of a changing () () in (13.5)
as proceeds, precisely because () () must be constant along the path.
In the “normal” case where ̃ 0 the needed change in () () is a
fall (i.e., a rise in ()()) This is what (13.5) shows. Indeed, the fall in
() () must exactly offset the effect on ̃ of the rising when there is a
fixed capital-output ratio.5 It follows that along the balanced growth path,
()() is an increasing implicit function of If we denote this function
() we end up with (4.12) with specified properties ( and ).
The generality of Uzawa’s theorem is noteworthy. The theorem assumes
CRS, but does not presuppose that the technology is neoclassical, not to
speak of satisfying the Inada conditions.6 And the theorem holds for exoge-
nous as well as endogenous technological progress. It is also worth mentioning
that the proof of the sufficiency part of the theorem is constructive. It pro-
vides a method to construct a hypothetical balanced growth path (BGP from
now).7
A simple implication of the Uzawa theorem is the following. Interpreting
the () in (4.11) as the “level of technology”, we have:
COROLLARY Along a BGP with positive gross saving and the technology
level, () growing at the rate output grows at the rate + while labor
productivity, ≡ and consumption per unit of labor, ≡ grow
at the rate
5
This way of presenting the intuition behind the Uzawa result draws upon Jones and
Scrimgeour (2008).
6
Many accounts of the Uzawa theorem, including Jones and Scrimgeour (2008), presume
a neoclassical production function, but the theorem is much more general.
7
Part (ii) of Proposition 2 is left out in Acemoglu’s book.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
CHAPTER 4. SKILL-BIASED TECHNICAL CHANGE.
68 BALANCED GROWTH THEOREMS
Proof That = + follows from (i) of Proposition 2. As to the growth
rate of labor productivity we have
(0)
= = (0)( −) = (0)
(0)
Finally, by Proposition 1, along a BGP with 0 must grow at the same
rate as ¤
We shall now consider the implication of Harrod-neutrality for the income
shares of capital and labor when the technology is neoclassical and markets
are perfectly competitive.
4.4 Harrod-neutrality and the functional in-
come distribution
There is one facet of Kaldor’s stylized facts we have so far not related to
Harrod-neutral technical progress, namely the long-run “approximate” con-
stancy of both the income share of labor, and the rate of return to
capital. At least with neoclassical technology, profit maximizing firms, and
perfect competition in the output and factor markets, these properties are
inherent in the combination of constant returns to scale, balanced growth,
and the assumption that the relative price of capital goods (relative to con-
sumption goods) is constant over time. The latter condition holds in models
where the capital good is nothing but non-consumed output, cf. (4.7).8
To see this, we start out from a neoclassical CRS production function
with Harrod-neutral technological progress,
() = (() ()()) (4.14)
With () denoting the real wage at time in equilibrium under perfect
competition the labor income share will be
()
()() ()
() 2 (() ()())()()
= = (4.15)
() () ()
In this simple model, without natural resources, (gross) capital income equals
non-labor income, () − ()() Hence, if () denotes the (net) rate of
return to capital at time , then
() − ()() − ()
() = (4.16)
()
8
The reader may think of the “corn economy” example in Acemoglu, p. 28.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
4.4. Harrod-neutrality and the functional income distribution 69
Denoting the (gross) capital income share by () we can write this ()
(in equilibrium) in three ways:
() − ()() (() + )()
() ≡ =
() ()
(() ()()) − 2 (() ()())()() 1 (() ()())()
() = =
() ()
()
()
()
() = (4.17)
()
where the first row comes from (4.16), the second from (4.14) and (4.15), the
third from the second together with Euler’s theorem.9 Comparing the first
and the last row, we see that in equilibrium
()
= () +
()
In this condition we recognize one of the first-order conditions in the rep-
resentative firm’s profit maximization problem under perfect competition,
since () + can be seen as the firm’s required gross rate of return.10
In the absence of uncertainty, the equilibrium real interest rate in the
bond market must equal the rate of return on capital, () And () + can
then be seen as the firm’s cost of disposal over capital per unit of capital per
time unit, consisting of interest cost plus capital depreciation.
Proposition 3 (factor income shares and rate of return under balanced
growth) Let the path (() () ())∞
=0 be a BGP in a competitive economy
with the production function (4.14) and with positive saving. Then, along the
BGP, the () in (4.17) is a constant, ∈ (0 1). The labor income share
will be 1 − and the (net) rate of return on capital will be = − where
is the constant output-capital ratio along the BGP.
Proof By CRS we have () = (() ()()) = ()() (̃() 1)
≡ ()()(̃()) In view of part (i) of Proposition 2, by balanced growth,
()() is some constant, . Since ()() = (̃())̃() and 00 0
this implies ̃() constant, say equal to ̃∗ But ()() = 0 (̃()) which
9
From Euler’s theorem, 1 + 2 = ( ) when is homogeneous of degree
one
10
With natural resources, say land, entering the set of production factors, the formula,
(4.16), for the rate of return to capital should be modified by subtracting land rents from
the numerator.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
CHAPTER 4. SKILL-BIASED TECHNICAL CHANGE.
70 BALANCED GROWTH THEOREMS
then equals the constant 0 (̃∗ ) along the BGP. It then follows from (4.17)
that () = 0 (̃∗ ) ≡ Moreover, 0 1 where 0 follows from
0 0 and 1 from the fact that = = (̃ ∗ )̃∗ 0 (̃∗ ) in view
of 00 0 and (0) ≥ 0 Then, by the first equality in (4.17), ()() ()
= 1 − () = 1 − . Finally, by (4.16), the (net) rate of return on capital is
= (1 − ()() ()) ()() − = − ¤
This proposition is of interest by displaying a link from balanced growth
to constancy of factor income shares and the rate of return, that is, some
of the “stylized facts” claimed by Kaldor. Note, however, that although the
proposition implies constancy of the income shares and the rate of return,
it does not determine them, except in terms of and But both and,
generally, are endogenous and depend on ̃∗ 11 which will generally be
unknown as long as we have not specified a theory of saving. This takes us
to theories of aggregate saving, for example the simple Ramsey model, cf.
Chapter 8 in Acemoglu’s book.
4.5 What if technological change is embod-
ied?
In our presentation of technological progress above we have implicitly as-
sumed that all technological change is disembodied. And the way the propo-
sitions 1, 2, and 3, are formulated assume this.
As noted in Chapter 2, disembodied technological change occurs when new
technical knowledge advances the combined productivity of capital and labor
independently of whether the workers operate old or new machines. Consider
again the aggregate dynamic resource constraint (4.7) and the production
function (4.9):
̇() = () − () (4.18)
() = ̃ (() () ) ̃ 0 (4.19)
Here () − () is aggregate gross investment, () For a given level of ()
the resulting amount of new capital goods per time unit (̇()+()), mea-
sured in efficiency units, is independent of when this investment occurs. It is
thereby not affected by technological progress. Similarly, the interpretation
of ̃ 0 in (4.19) is that the higher technology level obtained as time
proceeds results in higher productivity of all capital and labor. Thus also
11
As to there is of course a trivial exception, namely the case where the production
function is Cobb-Douglas and therefore is a given parameter.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
4.5. What if technological change is embodied? 71
firms that have only old capital equipment benefit from recent advances in
technical knowledge. No new investment is needed to take advantage of the
recent technological and organizational developments.12
In contrast, we say that technological change is embodied, if taking ad-
vantage of new technical knowledge requires construction of new investment
goods. The newest technology is incorporated in the design of newly pro-
duced equipment; and this equipment will not participate in subsequent
technological progress. Whatever the source of new technical knowledge,
investment becomes an important bearer of the productivity increases which
this new knowledge makes possible. Without new investment, the potential
productivity increases remain potential instead of being realized.
As also noted in Chapter 2, we may represent embodied technological
progress (also called investment-specific technological change) by writing cap-
ital accumulation in the following way,
̇() = ()() − () (4.20)
where () is gross investment at time and () measures the “quality”
(productivity) of newly produced investment goods. The increasing level of
technology implies increasing () so that a given level of investment gives
rise to a greater and greater additions to the capital stock, measured
in efficiency units. As in our aggregate framework, capital goods can be
produced at the same minimum cost as one consumption good, we have · =
1 where is the equilibrium price of capital goods in terms of consumption
goods. So embodied technological progress is likely to result in a steady
decline in the relative price of capital equipment, a prediction confirmed by
the data (see, e.g., Greenwood et al., 1997).
This raises the question how the propositions 1, 2, and 3 fare in the case
of embodied technological progress. The answer is that a generalized version
of Proposition 1 goes through. Essentially, we only need to replace (4.7) by
(13.13) and interpret in Proposition 1 as the value of the capital stock,
i.e., we have to replace by ̃ =
But the concept of Harrod-neutrality no longer fits the situation with-
out further elaboration. Hence to obtain analogies to Proposition 2 and
Proposition 3 is a more complicated matter. Suffice it to say that with em-
bodied technological progress, the class of production functions that are con-
sistent with balanced growth is smaller than with disembodied technological
progress.
12
In the standard versions of the Solow model and the Ramsey model it is assumed that
all technological progress has this form - for no other reason than that this is by far the
simplest case to analyze.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
CHAPTER 4. SKILL-BIASED TECHNICAL CHANGE.
72 BALANCED GROWTH THEOREMS
4.6 Concluding remarks
In the Solow model as well as in many other models with disembodied techno-
logical progress, a steady state and a balanced growth path imply each other.
Indeed, they are in that model, as well as many others, two sides of the same
process. There exist exceptions, however, that is, cases where steady state
and a balanced growth are not equivalent (some open economy models and
some models with embodied technical change). So the two concepts should
be held apart.13
Note that the definition of balanced growth refers to aggregate variables.
At the same time as there is balanced growth at the aggregate level, structural
change may occur. That is, a changing sectorial composition of the economy
is under certain conditions compatible with balanced growth (in a generalized
sense) at the aggregate level, cf. the “Kuznets facts” (see Kongsamut et al.,
2001, and Acemoglu, 2009, Chapter 20).
In view of the key importance of Harrod-neutrality, a natural question is:
has growth theory uncovered any endogenous tendency for technical progress
to converge to Harrod-neutrality? Fortunately, in his Chapter 15 Acemoglu
outlines a theory about a mechanism entailing such a tendency, the theory of
“directed technical change”. Jones (2005) suggests an alternative mechanism.
4.7 References
Acemoglu, D., 2009, Introduction to Modern Economic Growth, Princeton
University Press: Oxford.
Barro, R., and X. Sala-i-Martin, 2004, Economic Growth, second edition,
MIT Press: Cambridge (Mass.)
Duffy. J., C. Papageorgiou, and F. Perez-Sebastian, 2004, Capital-Skill
Complementarity? Evidence from a Panel of Countries, The Review of
Economics and Statistics, vol. 86(1), 327-344.
Gordon, R. J., 1990. The Measurement of Durable goods Prices. Chicago
University Press: Chicago.
13
Here we deviate from Acemoglu, p. 65, where he says that he will use the two terms
“interchangingly”. We also deviate from Barro and Sala-i-Martin (2004, pp. 33-34) who
define a steady state as synonymous with a balanced growth path as the latter was defined
above.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
4.7. References 73
Greenwood, J., Z. Hercowitz, and P. Krusell, 1997. Long-Run Implications
of Investment-Specific Technological Change. American Economic Re-
view 87 (3), 342-362.
Groth, C., K.-J. Koch, and Thomas Steger, 2010, When growth is less than
exponential, Economic Theory 44, 213-242.
Groth, C., and R. Wendner, 2014. Embodied Learning by Investing and
Speed of Convergence, J. of Macroeconomics (forthcoming).
Jones, C. I., 2005, The shape of production functions and the direction of
technical change. Quarterly Journal of Economics, no. 2, 517-549.
Jones, C. I., and D. Scrimgeour, 2008, The steady-state growth theorem:
Understanding Uzawa (1961), Review of Economics and Statistics 90
(1), 180-182.
Jones, C. I., and P. M. Romer, 2010, The new Kaldor facts: Ideas, insti-
tutions, population, and human capital, American Economic Journal:
Macroeconomics, vol. 2 (1), 224-245. Cursory.
Kongsamut, P., S. Rebelo, and D. Xie, 2001, Beyond balanced growth.
Review of Economic Studies 48, 869-882.
Perez-Sebastian, F., 2008, “Testing capital-skill complementarity across sec-
tors in a panel of Spanish regions”, WP 2008.
Schlicht, E., 2006, A variant of Uzawa’s theorem, Economics Bulletin 6,
1-5.
Stokey, N.L., 1996, Free trade, factor returns, and factor accumulation, J.
Econ. Growth, vol. 1 (4), 421-447.
Uzawa, H., 1961, Neutral inventions and the stability of growth equilibrium,
Review of Economic Studies 28, No. 2, 117-124.
Wan, H. Y. Jr., 1971, Economic Growth, Harcourt Brace: New York.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
CHAPTER 4. SKILL-BIASED TECHNICAL CHANGE.
74 BALANCED GROWTH THEOREMS
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
Chapter 5
Growth accounting and the
concept of TFP: Some warnings
5.1 Introduction
This chapter discusses the concepts of Total Factor Productivity, TFP, and
TFP growth, and ends up with three warnings regarding uncritical use of
them.
First, however, we should provide a precise definition of the TFP level
which is in fact a tricky concept. Unfortunately, Acemoglu (p. 78) does
not make a clear distinction between TFP level and TFP growth. Moreover,
Acemoglu’s point of departure (p. 77) assumes a priori that the way the pro-
duction function is time-dependent can be represented by a one-dimensional
index, () The TFP concept and the applicability of growth accounting
are, however, not limited to this case.
For convenience, we treat time as continuous (although the timing of the
variables is indicated merely by a subscript).1
5.2 TFP level and TFP growth
Let denote aggregate output (value added in fixed prices) at time in a
sector or the economy as a whole. Suppose is determined by the function
= ( ) (5.1)
1
I thank Niklas Brønager for useful discussions related to this chapter.
75
CHAPTER 5. GROWTH ACCOUNTING AND THE CONCEPT
76 OF TFP: SOME WARNINGS
where is an aggregate input of physical capital and an index of quality-
adjusted labor input.2 The “quality-adjustment” of the input of labor (man-
hours per year) aims at taking educational level and work experience into
account. In fact, both output and the two inputs are aggregates of het-
erogeneous elements. The involved conceptual and measurement difficulties
are huge and there are different opinions in the growth accounting literature
about how to best deal with them. Here we ignore these problems. The
third argument in (5.1) is time, indicating that the production function
(· · ) is time-dependent. Thus “shifts in the production function”, due
to changes in efficiency and technology (“technical change” for short), can
be taken into account. We treat time as continuous and assume that is
a neoclassical production function. When the partial derivative of w.r.t.
the third argument is positive, i.e., 0 technical change amounts
to technical progress. We consider the economy from a purely supply-side
perspective.3
We shall here concentrate on the fundamentals of TFP and TFP growth.
These can in principle be described without taking the heterogeneity and
changing quality of the labor input into account. Hence we shall from now
on ignore this aspect and simplifying assume that labor is homogeneous and
labor quality is constant. So (5.1) is reduced to the simpler case,
= ( ) (5.2)
where is the number of man-hours per year. As to measurement of
, some adaptation of the perpetual inventory method 4 is typically used,
with some correction for under-estimated quality improvements of invest-
ment goods in national income accounting. The output measure is (or at
least should be) corrected correspondingly, also for under-estimated quality
improvements of consumption goods.
2
Natural resources (land, oil wells, coal in the ground, etc.) constitute a third primary
production factor. The role of this factor is in growth accounting often subsumed under
.
3
Sometimes in growth accounting the left-hand side variable, in (5.2) is the gross
product rather than value added. Then non-durable intermediate inputs should be taken
into account as a third production factor and enter as an additional argument of ̃ in
(5.2). Since non-market production is difficult to measure, the government sector is usually
excluded from in (5.2). Total Factor Productivity is by some authors called Multifactor
Productivity and abbreviated MFP.
4
Cf. Section 2.2 in Chapter 2.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
5.2. TFP level and TFP growth 77
5.2.1 TFP growth
The notion of Total Factor Productivity at time TFP is intended to
indicate a level of productivity. Nevertheless there is a tendency in the
literature to evade a direct definition of this level and instead go straight
away to a decomposition of output growth. Let us start the same way here
but not forget to come back to the issue about what can be meant by the
level of TFP.
The growth rate of a variable at time will be denoted . We take
the total derivative w.r.t. in (5.2) to get
̇ = ( )̇ + ( )̇ + ( ) · 1
Dividing through by gives
̇ 1 h i
≡ = ( )̇ + ( )̇ + ( ) · 1
( ) ( ) ( )
= + +
( )
≡ + + (5.3)
( )
where and are shorthands for ( ) ≡ ( )
and ( )
≡ (
( )
)
respectively, that is, the partial output elasticities w.r.t. the
two production factors, evaluated at the factor combination ( ) at time
Finally, ( ) ≡ , that is, the partial derivative w.r.t. the
third argument of the function , evaluated at the point ( )
The equation (5.3) is the basic growth-accounting relation, showing how
the output growth rate can be decomposed into the “contribution” from
growth in each of the inputs and a residual. The TFP growth rate is defined
as the residual
( )
TFP, ≡ − ( + ) = (5.4)
So the TFP growth rate is what is left when from the output growth rate is
subtracted the “contribution” from growth in the factor inputs weighted by
the output elasticities w.r.t. these inputs. This is sometimes interpreted as
reflecting that part of the output growth rate which is explained by technical
progress. One should be careful, however, not to identify a descriptive ac-
counting relationship with deeper causality. Without a complete model, at
most one can say that the TFP growth rate measures that fraction of output
growth that is not directly attributable to growth in the capital and labor
inputs. So:
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
CHAPTER 5. GROWTH ACCOUNTING AND THE CONCEPT
78 OF TFP: SOME WARNINGS
The TFP growth rate can be interpreted as reflecting the “direct
contribution” to current output growth from current technical
change (in a broad sense including learning by doing and organi-
zational improvement).
Let us consider how the actual measurement of TFP, can be carried out.
The output elasticities w.r.t. capital and labor, and will, under
perfect competition and absence of externalities and of increasing returns
to scale, equal the income shares of capital and labor, respectively. Time
series for these income shares and for , and hence also for
and , can be obtained (directly or with some adaptation) from national
income accounts. This allows straightforward measurement of the residual,
TFP, 5
The decomposition in (5.4) was introduced already by Solow (1957). Since
the TFP growth rate appears as a residual, it is sometimes called the Solow
residual. As a residual it may reflect the contribution of many things, some
wanted (current technical innovation in a broad sense including organiza-
tional improvement), others unwanted (such as varying capacity utilization,
omitted inputs, measurement errors, and aggregation bias).
5.2.2 The TFP level
Now let us consider the level of TFP, that “something” for which we have
calculated its growth rate without yet having defined what it really is. But
knowing the growth rate of TFP for all in a certain time interval, we in fact
have a differential equation in the TFP level of the form () = ()()
namely:
(TFP ) = TFP, · TFP
The solution of this simple linear differential equation is6
T F P,
TFP = TFP0 0 (5.5)
For a given initial value TFP0 0 (which may be normalized to 1 if de-
sired), the time path of TFP is determined by the right-hand side of (5.5).
Consequently:
The TFP level at time can interpreted as reflecting the cumula-
tive “direct contribution” to output since time 0 from cumulative
technical change since time 0.
5
Of course, data are in discrete time. So to make actual calculations we have to translate
(5.4) into discrete time. The weights and can then be estimated by two-years
moving averages of the factor income shares as shown in Acemoglu (2009, p. 79).
6
See Appendix B of Chapter 3 in these lecture notes or Appendix B to Acemoglu.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
5.3. The case of Hicks-neutrality* 79
Why do we say “direct contribution”? The reason is that the cumulative
technical change since time 0 may also have an indirect effect on output,
namely via affecting the output elasticities w.r.t. capital and labor, and
Through this channel cumulative technical change affects the role of
input growth for output growth. This possible indirect effect over time of
technical change is not included in the TFP concept.
To clarify the matter we will compare the TFP calculation under Hicks-
neutral technical change with that under other forms of technical change.
5.3 The case of Hicks-neutrality*
In the case of Hicks neutrality, by definition, technical change can be repre-
sented by the evolution of a one-dimensional variable, and the production
function in (5.2) can be specified as
= ( ) = ( ) (5.6)
Here the TFP level is at any time, , identical to the level of if we normalize
the initial values of both and TFP to be the same, i.e., TFP0 = 0 0.
Indeed, calculating the TFP growth rate, (5.4), on the basis of (5.6) gives
( ) ̇ ( ) ̇
TFP, = = = ≡ (5.7)
( )
where the second equality comes from the fact that and are kept fixed
when the partial derivative of w.r.t. is calculated. The formula (5.5) now
gives
TFP = 0 · 0 , =
The nice feature of Hicks neutrality is thus that we can write
( ) ( )
TFP = = = (5.8)
( 0) 0 ( )
using the normalization 0 = 1 That is:
Under Hicks neutrality, current TFP appears as the ratio be-
tween the current output level and the hypothetical output level
that would have resulted from the current inputs of capital and
labor in case of no technical change since time 0.
So in the case of Hicks neutrality the economic meaning of the TFP level
is straightforward. The reason is that under Hicks neutrality the output
elasticities w.r.t. capital and labor, and are independent of technical
change.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
CHAPTER 5. GROWTH ACCOUNTING AND THE CONCEPT
80 OF TFP: SOME WARNINGS
5.4 The case of absence of Hicks-neutrality*
The above very intuitive interpretation of TFP is only valid under Hicks-
neutral technical change. Neither under general technical change nor even
under Harrod- or Solow-neutral technical change (unless the production func-
tion is Cobb-Douglas so that both Harrod and Solow neutrality imply Hicks-
neutrality), will current TFP appear as the ratio between the current output
level and the hypothetical output level that would have resulted from the
current inputs of capital and labor in case of no technical change since time
0.
To see this, let us return to the general time-dependent production func-
tion in (5.2). Let denote the ratio between the current output level at
time and the hypothetical output level, ( 0) that would have ob-
tained with the current inputs of capital and labor in case of no change in
the technology since time 0, i.e.,
( )
≡ (5.9)
( 0)
So can be seen as a factor of joint-productivity growth from time 0 to
time evaluated at the time- input combination.
If this should always indicate the level of TFP at time , the growth
rate of should equal the growth rate of TFP. Generally, it does not,
however. Indeed, defining ( ) ≡ ( 0) by the rule for the time
derivative of fractions7 , we have
̃ ( ) ( )
≡ −
( ) ( )
1 h i
= ( )̇ + ( )̇ + ( ) · 1
1 h i
− ( )̇ + ( )̇
( )
( )
= ( ) + ( ) +
−( ( 0) + ( 0) ) (5.10)
= ( ( ) − ( 0)) + ( ( ) − ( 0)) + TFP,
6 = TFP, generally,
where TFP, is given in (5.4). Unless the partial output elasticities w.r.t.
capital and labor, respectively, are unaffected by technical change, the con-
clusion is that TFP will differ from our defined in (5.9). So:
7
See Appendix A to Chapter 3 of these lecture notes.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
5.4. The case of absence of Hicks-neutrality* 81
In the absence of Hicks neutrality, current TFP does not gener-
ally appear as the ratio between the current output level and the
hypothetical output level that would have resulted from the cur-
rent inputs of capital and labor in case of no technical change
since time 0.
A closer look at vs. TFP
As in (5.9) is the time- output arising from the time- inputs relative to
the fictional time-0 output from the same inputs, we consider along with
TFP as two alternative joint-productivity indices. From (5.10) we see that
TFP, = −( ( ) − ( 0)) −( ( )− ( 0))
So the growth rate of TFP equals the growth rate of the joint-productivity
index corrected for the cumulative impact of technical change since time 0
on the direct contribution to time- output growth from time- input growth.
This impact comes about when the output elasticities w.r.t. capital and la-
bor, respectively, are affected by technical change, that is, when ( )
6= ( 0) and/or ( ) 6= ( 0)
Under Hicks-neutral technical change there will be no correction because
the output elasticities are independent of technical change. In this case TFP
coincides with the index In the absence of Hicks-neutrality the two indices
differ. This is why we in Section 2.2 characterized the TFP level as the
cumulative “direct contribution” to output since time 0 from cumulative
technical change, thus excluding the possible indirect contribution coming
about via the potential effect of technical change on the output elasticities
w.r.t. capital and labor and thereby on the contribution to output from input
growth.
Given that the joint-productivity index is the more intuitive joint-
productivity measure, why is TFP the more popular measure? There are at
least two reasons for this. First, it can be shown that the TFP measure has
more convenient balanced growth properties. Second, is more difficult to
measure. To see this we substitute (5.3) into (5.10) to get
= − ( ( 0) + ( 0) ) (5.11)
( 0)
The relevant output elasticities, ( 0) ≡ ( 0)
and ( 0)
( 0)
≡ are hypothetical constructs, referring to the technology as it
( 0)
was at time 0, but with the factor combination observed at time , not at time
0. The nice thing about the Solow residual is that under the assumptions
of perfect competition and absence of externalities, it allows measurement
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
CHAPTER 5. GROWTH ACCOUNTING AND THE CONCEPT
82 OF TFP: SOME WARNINGS
by using data on prices and quantities alone, that is, without knowledge
of the production function. To evaluate , however, we need estimates
of the hypothetical output elasticities, ( 0) and ( 0) This
requires knowledge about how the output elasticities depend on the factor
combination and time, respectively, that is, knowledge about the production
function.
Now to the warnings concerning application of the TFP measure.
5.5 Three warnings
Balanced growth at the aggregate level, hence Harrod neutrality, seems to
characterize the growth experience of the UK and US over at least a century
(Kongsamut et al., 2001; Attfield and Temple, 2010). At the same time
the aggregate elasticity of factor substitution is generally estimated to be
significantly less than one (see, e.g., Antras, 2004). This amounts to rejection
of the Cobb-Douglas specification of the aggregate production function and
so, at the aggregate level, Harrod neutrality rules out Hicks neutrality.
Warning 1 Since Hicks-neutrality is empirically doubtful at the aggre-
gate level, TFP can often not be identified with the simple intuitive joint-
productivity measure defined in (5.9) above.
Warning 2 When Harrod neutrality obtains, relative TFP growth rates
across sectors or countries can be quite deceptive.
Suppose there are countries and that country has the aggregate pro-
duction function
= () ( ) = 1 2
where () is a neoclassical production function with CRS and is the level
of labor-augmenting technology which, for simplicity, we assume shared by
all the countries (these are open and “close” to each other). So technical
progress is Harrod-neutral. Let the growth rate of be a constant 0
Many models imply that ̃ ≡ ( ) tends to a constant, ̃∗ , in the long
run, which we assume is also the case here. Then, for → ∞ ≡
≡ ̃ where ̃ → ̃∗ and ≡ ≡ ̃ where ̃ → ̃∗ = () (̃∗ );
here () is the production function on intensive form. So in the long run
and tend to = .
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
5.5. Three warnings 83
Formula (5.4) then gives the TFP growth rate of country in the long
run as
TFP = − (∗ + (1 − ∗ ) ) = − − ∗ ( − )
= − ∗ = (1 − ∗ ) (5.12)
where ∗ is the output elasticity w.r.t. capital, ()0 (̃ )̃ () (̃ ) evaluated
at ̃ = ̃∗ Under labor-augmenting technical progress, the TFP growth rate
thus varies negatively with the output elasticity w.r.t. capital (the capital
income share under perfect competition). Owing to differences in product
and industry composition, the countries have different ∗ ’s. In view of (5.12),
for two different countries, and we get
⎧
⎨ ∞ if ∗ ∗
→ 1 if ∗ = ∗ (5.13)
⎩ ∗ ∗
0 if
for → ∞8 Thus, in spite of long-run growth in the essential variable,
being the same across the countries, their TFP growth rates are very
different. Countries with low ∗ ’s appear to be technologically very dynamic
and countries with high ∗ ’s appear to be lagging behind. It is all due to the
difference in across countries; a higher just means that a larger fraction
of = = becomes “explained” by in the growth accounting (5.12),
leaving a smaller residual. And the level of has nothing to do with technical
progress.
We conclude that comparison of TFP levels across countries or time may
misrepresent the intuitive meaning of productivity and technical progress
when output elasticities w.r.t. capital differ and technical progress is Harrod-
neutral (even if technical progress were at the same time Hicks-neutral as is
the case with a Cobb-Douglas specification). It may be more reasonable to
just compare levels of across countries and time.
Warning 3 Growth accounting is - as the name says - just about account-
ing and measurement. So do not confuse growth accounting with causality
in growth analysis. To talk about causality we need a theoretical model sup-
ported by the data. On the basis of such a model we can say that this or that
set of exogenous factors through the propagation mechanisms of the model
cause this or that phenomenon, including economic growth. In contrast, con-
sidering the growth accounting identity (5.3) in itself, none of the terms have
8
If is Cobb-Douglas with output elasticity w.r.t. capital equal to , the result
in (5.12) can be derived more directly by first defining = 1−
, then writing the
production function in the Hicks-neutral form (5.6), and finally use (5.7).
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
CHAPTER 5. GROWTH ACCOUNTING AND THE CONCEPT
84 OF TFP: SOME WARNINGS
priority over the others w.r.t. a causal role. And there are important omitted
variables. There are simple illustrations in Exercises III.1 and III.2.
In a complete model with exogenous technical progress, part of will
be induced by this technical progress. If technical progress is endogenous
through learning by investing, as in Arrow (1962), there is mutual causa-
tion between and technical progress. Yet another kind of model might
explain both technical progress and capital accumulation through R&D, cf.
the survey by Barro (1999).
5.6 References
Antràs, P., 2004, Is the U.S. aggregate production function Cobb-Douglas?
New estimates of the elasticity of substitution, Contributions to Macro-
economics, vol. 4, no. 1, 1-34.
Attfield, C., and J.R.W. temple, 2010, Balanced growth and the great ratios:
New evidence for the US and UK, J. of Macroeconomics, vol. 32, 937-
956.
Barro, R.J., 1999, Notes on growth accounting, J. of Economic Growth, vol.
4 (2), 119-137.
Bernard, A. B., and C. I. Jones, 1996a, Technology and Convergence, Eco-
nomic Journal, vol. 106, 1037-1044.
Bernard, A. B., and C. I. Jones, 1996b, Comparing Apples to Oranges:
productivity convergence and measurement across industries and coun-
tries, American Economic Review, vol. 86, no. 5, 1216-1238.
Greenwood, J., and P. Krusell, 2006, Growth accounting with investment-
specific technological progress: A discussion of two approaches, J. of
Monetary Economics.
Hercowitz, Z., 1998, The ‘embodiment’ controversy: A review essay, J. of
Monetary Economics, vol. 41, 217-224.
Hulten, C.R., 2001, Total factor productivity. A short biography. In: Hul-
ten, C.R., E.R. Dean, and M. Harper (eds.), New Developments in
Productivity Analysis, Chicago: University of Chicago Press, 2001, 1-
47.
Kongsamut, P., S. Rebelo, and D. Xie, 2001, Beyond Balanced Growth,
Review of Economic Studies, vol. 68, 869-882.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
5.6. References 85
Sakellaris, P., and D.J. Wilson, 2004, Quantifying embodied technological
progress, Review of Economic Dynamics, vol. 7, 1-26.
Solow, R.M., 1957, Technical change and the aggregate production function,
Review of Economics and Statistics, vol. 39, 312-20.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
CHAPTER 5. GROWTH ACCOUNTING AND THE CONCEPT
86 OF TFP: SOME WARNINGS
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
Chapter 6
Transitional dynamics.
Barro-style growth regressions
In this chapter we discuss three issues, all of which are related to the transi-
tional dynamics of a growth model:
Do poor countries necessarily tend to approach their steady state from
below?
How fast (or rather how slow) are the transitional dynamics in a growth
model?
What exactly is the theoretical foundation for a Barro-style growth
regression analysis?
The Solow growth model may serve as the analytical point of departure
for the …rst two issues and to some extent also for the third.
6.1 Point of departure: the Solow model
As is well-known, the fundamental di¤erential equation for the Solow model
is
~ = sf (k(t))
k(t) ~ ~
( + g + n)k(t); ~
k(0) = k~0 > 0, (6.1)
~
where k(t) ~
K(t)=(A(t)L(t)); f (k(t)) ~
F (k(t); 1); A(t) = A0 egt ; and
nt
L(t) = L0 e (standard notation). The production function F is neoclassical
with CRS and the parameters satisfy 0 < s < 1 and +g+n > 0: The produc-
tion function on intensive form, f; therefore satis…es f (0) 0; f 0 > 0; f 00 < 0;
and
~ > + g + n > lim f 0 (k):
lim f 0 (k) ~ (A1)
~
k!0 s ~
k!1
83
CHAPTER 6. TRANSITIONAL DYNAMICS. BARRO-STYLE
84 GROWTH REGRESSIONS
ỹ
(δ + g + n)k̃
f k̃
sf k̃
ỹ ∗
k̃
k̃ ∗ k̃0
Figure 6.1: Phase diagram 1.
Then there exists a unique non-trivial steady state, k~ > 0; that is, a unique
positive solution to the equation
sf (k~ ) = ( + g + n)k~ : (6.2)
Furthermore, given an arbitrary k~0 > 0; we have for all t 0;
~ T 0 for k(t)
k(t) ~ S k~ ; (6.3)
respectively. The steady state, k~ ; is thus globally asymptotically stable in the
sense that for all k~0 > 0; limt!1 k(t)
~ = k and this convergence is monotonic
~ ~
(in the sense that k(t) k does not change sign during the adjustment
process).
From now on the dating of k~ is suppressed unless needed for clarity. Figure
6.1 illustrates the dynamics as seen from the perspective of (6.1) (in this and
the two next …gures, x should read g. Figure 6.2 illustrates the dynamics
emerging when we rewrite (6.1) this way:
+ g + n~
k~ = s f (k)
~ k T 0 for k~ S k~ :
s
In Figure 6.3 yet another illustration is exhibited, based on rewriting (6.1)
this way:
k~ ~
f (k)
=s ( + g + n);
k~ k~
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
6.1. Point of departure: the Solow model 85
ỹ
(δ + g + n)
k̃
s
f k̃
ỹ ∗
k̃
k̃ ∗ k̃0
Figure 6.2: Phase diagram 2.
S
K
gk̃
δ+g+n
f (k̃)
s
k̃
k̃
k̃0 k̃ ∗
Figure 6.3: Phase diagram 3.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
CHAPTER 6. TRANSITIONAL DYNAMICS. BARRO-STYLE
86 GROWTH REGRESSIONS
~ k~ is gross saving per unit of capital, S=K (Y C)=K:
where sf (k)=
An important variable in the analysis of the adjustment process towards
steady state is the output elasticity w.r.t. capital:
K @Y k~ 0 ~ ~
= f (k) "(k); (6.4)
Y @K ~
f (k)
~ < 1 for all k~ > 0:
where 0 < "(k)
6.2 Do poor countries tend to approach their
steady state from below?
From some textbooks (for instance Barro and Sala-i-Martin, 2004) one gets
the impression that poor countries tend to approach their steady state from
below. But this is not what the Penn World Table data seems to indicate.
And from a theoretical point of view the size of k~0 relative to k~ is certainly
ambiguous, whether the country is rich or poor. To see this, consider a poor
country with initial e¤ective capital intensity
K0
k~0 :
A 0 L0
Here K0 =L0 will typically be small for a poor country (the country has not
yet accumulated much capital relative to its fast-growing population). The
technology level, A0 ; however, also tends to be small for a poor country.
Hence, whether we should expect k~0 < k~ or k~0 > k~ is not obvious apriori.
Or equivalently: whether we should expect that a poor country’s GDP at an
arbitrary point in time grows at a rate higher or lower than the country’s
steady-state growth rate, g + n; is not obvious apriori.
While Figure 6.3 illustrates the case where the inequality k~0 < k~ holds,
Figure 6.1 and 6.2 illustrate the opposite case. There exists some empirical
evidence indicating that poor countries tend to approach their steady state
from above. Indeed, Cho and Graham (1996) …nd that “on average, countries
with a lower income per adult are above their steady-state positions, while
countries with a higher income are below their steady-state positions”.
The prejudice that poor countries apriori should tend to approach their
steady state from below seems to come from a confusion of conditional and
unconditional convergence. The Solow model predicts - and data supports
- that within a group of countries with similar structural characteristics (ap-
proximately the same f; A0 ; g; s; n; and ); the initially poorer countries will
grow faster than the richer countries. This is because the poorer countries
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
6.3. Within-country convergence speed and adjustment time 87
(small y(0) = f (k~0 )A0 ) will be the countries with relatively small initial
capital-labor ratio, k0 : As all the countries in the group have approximately
the same A0 ; the poorer countries thus have k~0 k0 =A0 relatively small, i.e.,
k~0 < k~ . From y Y =L ~
y~A = f (k)A follows that the growth rate in
output per worker of these poor countries tends to exceed g: Indeed, we have
generally (for instance in the Solow model as well as the Ramsey model)
y_ y~ ~ k~
f 0 (k)
= +g = + g T g for k~ T 0; i.e., for k~ S k~ :
y y~ ~
f (k)
So, within the group, the poor countries tend to approach the steady state,
k~ ; from below.
The countries in the world as a whole, however, di¤er a lot w.r.t. their
structural characteristics, including their A0 : Unconditional convergence is
de…nitely rejected by the data. Then there is no reason to expect the poorer
countries to have k~0 < k~ rather than k~0 > k~ . Indeed, according to the
mentioned study by Cho and Graham (1996), it turns out that the data for
the relatively poor countries favors the latter inequality rather than the …rst.
6.3 Within-country convergence speed and ad-
justment time
Our next issue is: How fast (or rather how slow) are the transitional dynamics
in a growth model? To put it another way: according to a given growth model
with convergence, how fast does the economy approach its steady state? The
answer turns out to be: not very fast - to say the least. This is a rather
general conclusion and is con…rmed by the empirics: adjustment processes
in a growth context are quite time consuming.
In Acemoglu’s textbook we meet the concept of speed of convergence at p.
54 (under an alternative name, rate of adjustment) and p. 81 (in connection
with Barro-style growth regressions). Here we shall go more into detail with
the issue of speed of convergence.
Again the Solow model is our frame of reference. We search for a formula
~
for the speed of convergence of k(t) and y(t)=y (t) in a closed economy de-
scribed by the Solow model. So our analysis is concerned with within-country
convergence: how fast do variables such as k~ and y approach their steady
state paths in a closed economy? The key adjustment mechanism is linked
to diminishing returns to capital (falling marginal productivity of capital)
in the process of capital accumulation. The problem of cross-country con-
vergence (which is what “ convergence” and “ convergence” are about)
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
CHAPTER 6. TRANSITIONAL DYNAMICS. BARRO-STYLE
88 GROWTH REGRESSIONS
is in principle more complex because also such mechanisms as technological
catching-up and cross-country factor movements are involved.
6.3.1 ~
Convergence speed for k(t)
~ to (k(t)
The ratio of k(t) ~ k~ ) 6= 0 can be written
~
k(t) ~
d(k(t) k~ )=dt
= ; (6.5)
~
k(t) k~ ~
k(t) k~
since dk~ =dt = 0: We de…ne the instantaneous speed of convergence at time
~
t as the (proportionate) rate of decline of the distance k(t) k~ at time t
~ 1 Thus,
and we denote it SOCt (k):
d ~
k(t) k~ =dt ~
~ d(k(t) k~ )=dt
SOCt (k) = ; (6.6)
~
k(t) k~ ~
k(t) k~
where the equality sign is valid for monotonic convergence.
~ depends on both the absolute size of the di¤erence k~
Generally, SOCt (k)
~ ~
k at time t and its sign. But if the di¤erence is already “small”, SOCt (k)
will be “almost” constant for increasing t and we can …nd an approximate
~ be de…ned by k~ = sf (k)
measure for it. Let the function '(k) ~ mk~ '(k);~
where m ~ around k~ =
+ g + n: A …rst-order Taylor approximation of '(k)
k~ gives
~
'(k) '(k~ ) + '0 (k~ )(k~ k~ ) = 0 + (sf 0 (k~ ) m)(k~ k~ ):
For k~ in a small neighborhood of the steady state, k~ ; we thus have
k~ = '(k)
~ (sf 0 (k~ ) m)(k~ k~ )
sf 0 (k~ )
= ( 1)m(k~ k~ )
m
k~ f (k~ )
0
= ( 1)m(k~ k~ ) (from (6.2))
~
f (k )
("(k~ ) 1)m(k~ k~ ) (from (6.4)).
1
Synonyms for speed of convergence are rate of convergence, rate of adjustment or
adjustment speed.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
6.3. Within-country convergence speed and adjustment time 89
Applying the de…nition (6.6) and the identity m + g + n; we now get
~
d(k(t) k~ )=dt ~
k(t)
~ =
SOCt (k) = (1 "(k~ ))( +g+n) (k~ ) > 0:
~
k(t) k~ ~
k(t) k~
(6.7)
This result tells us how fast, approximately, the economy approaches its
steady state if it starts “close” to it. If, for example, (k~ ) = 0:02 per year,
~ and k~ vanishes per year. We also see
then 2 percent of the gap between k(t)
that everything else equal, a higher output elasticity w.r.t. capital implies a
lower speed of convergence.
In the limit, for k~ k~ ! 0; the instantaneous speed of convergence
coincides with what is called the asymptotic speed of convergence, de…ned as
~
SOC (k) ~ = (k~ ):
lim SOCt (k) (6.8)
jk~ ~ j!0
k
Multiplying through by (k(t) ~ k~ ); the equation (6.7) takes the form of
a homogeneous linear di¤erential equation (with constant coe¢ cient), x(t) _ =
~
x(t); the solution of which is x(t) = x(0)e t : With x(t) = k(t) k~ and “=”
replaced by “ ”, we get in the present case
~ ~ )t
k(t) k~ ~
(k(0) k~ )e (k
: (6.9)
This is the approximative time path for the gap between k(t) ~ and k~ and
shows how the gap becomes smaller and smaller at the rate (k~ ).
One of the reasons that the speed of convergence is important is that it
indicates what weight should be placed on transitional dynamics of a growth
model relative to the steady-state behavior. The speed of convergence mat-
ters for instance for the evaluation of growth-promoting policies. In growth
models with diminishing marginal productivity of production factors, suc-
cessful growth-promoting policies have transitory growth e¤ects and perma-
nent level e¤ects. Slower convergence implies that the full bene…ts are slower
to arrive.
6.3.2 ~
Convergence speed for log k(t) *
We have found an approximate expression for the convergence speed of k: ~
Since models in empirical analysis and applied theory are often based on log-
linearization, we might ask what the speed of convergence of log k~ is. The
answer is: approximately the same and asymptotically exactly the same as
that of k~ itself! Let us see why.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
CHAPTER 6. TRANSITIONAL DYNAMICS. BARRO-STYLE
90 GROWTH REGRESSIONS
~ around k~ = k~ gives
A …rst-order Taylor approximation of log k(t)
~ 1 ~
log k(t) log k~ + (k(t) k~ ): (6.10)
~
k
By de…nition
~
d(log k(t) log k~ )=dt ~
dk(t)=dt
~ =
SOCt (log k) =
~
log k(t) log k~ ~
k(t)(log ~
k(t) log k~ )
~
dk(t)=dt k~
= ~ ! SOC (k)
SOCt (k) ~ = (k~ )(6.11)
~
k(t)
~
k(t) ~
k ~
k(t)
~
k
~
for k(t) ! k~ ;
where in the second line we have used, …rst, the approximation (6.10), second,
the de…nition in (6.7), and third, the de…nition in (6.8).
So, at least in a neighborhood of the steady state, the instantaneous rate
of decline of the logarithmic distance of k~ to the steady-state value of k~
approximates the instantaneous rate of decline of the distance of k~ itself to
its steady-state value. The asymptotic speed of convergence of log k~ coincides
with that of k~ itself and is exactly (k~ ):
In the Cobb-Douglas case (where "(k~ ) is a constant, say ) it is possible
to …nd an explicit solution to the Solow model, see Acemoglu p. 53 and
Exercise II.2. It turns out that the instantaneous speed of convergence in a
…nite distance from the steady state is a constant and equals the asymptotic
speed of convergence, (1 )( + g + n):
6.3.3 Convergence speed for y(t)=y (t) *
The variable which we are interested in is usually not so much k~ in itself,
but rather labor productivity, y(t) y~(t)A(t): In the interesting case where
g > 0; labor productivity does not converge towards a constant. We therefore
focus on the ratio y(t)=y (t); where y (t) denotes the hypothetical value of
labor productivity at time t; conditional on the economy being on its steady-
state path, i.e.,
y (t) y~ A(t): (6.12)
We have
y(t) y~(t)A(t) y~(t)
= : (6.13)
y (t) y~ A(t) y~
As y~(t) ! y~ for t ! 1; the ratio y(t)=y (t) converges towards 1 for t ! 1:
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
6.3. Within-country convergence speed and adjustment time 91
Taking logs on both sides of (6.13), we get
y(t) y~(t)
log = log = log y~(t) log y~
y (t) y~
1
log y~ + (~ y (t) y ) log y~ (…rst-order Taylor approx. of log y~ )
y~
1 ~
= (f (k(t)) f (k~ ))
f (k~ )
1
(f (k~ ) + f 0 (k~ )(k(t)
~ k~ ) f (k~ )) (…rst-order approx. of f (k))
~
f (k~ )
k~ f 0 (k~ ) k(t)
~ k~ ~
k(t) k~
= "(k~ )
f (k~ ) k~ k~
"(k~ )(log k(t)
~ log k~ ) (by (6.10)). (6.14)
~
Multiplying through by (log k(t) log k~ ) in (6.11) and carrying out the
di¤erentiation w.r.t. time, we …nd an approximate expression for the growth
~
rate of k;
~
dk(t)=dt k~
gk~ (t) ~
SOCt (k)(log ~
k(t) log k~ )
~
k(t) ~
k(t)
! (k~ )(log k(t)
~ log k~ ) for k(t)
~ ! k~ ; (6.15)
where the convergence follows from the last part of (6.11). We now calculate
the time derivative on both sides of (6.14) to get
y(t) y~(t) d~
y (t)=dt
d(log )=dt = d(log )=dt = gy~(t)
y (t) y~ y~(t)
"(k~ )g~ (t) "(k~ ) (k~ )(log k(t)
k
~ log k~ ): (6.16)
from (6.15). Dividing through by log(y(t)=y (t)) in this expression, taking
(6.14) into account, gives
d(log yy(t)
(t)
)=dt d(log yy(t)
(t)
log 1)=dt y
y(t)
= y(t)
SOCt (log ) (k~ ); (6.17)
log log log 1 y
y (t) y (t)
in view of log 1 = 0. So the logarithmic distance of y from its value on the
steady-state path at time t has approximately the same rate of decline as the
logarithmic distance of k~ from k’
~ s value on the steady-state path at time t:
The asymptotic speed of convergence for log y(t)=y (t) is exactly the same
~ namely (k~ ).
as that for k;
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
CHAPTER 6. TRANSITIONAL DYNAMICS. BARRO-STYLE
92 GROWTH REGRESSIONS
What about the speed of convergence of y(t)=y (t) itself? Here the same
principle as in (6.11) applies. The asymptotic speed of convergence for
log(y(t)=y (t)) is the same as that for y(t)=y (t) (and vice versa), namely
(k~ ):
With one year as our time unit, standard parameter values are: g = 0:02;
n = 0:01; = 0:05; and "(k~ ) = 1=3: We then get (k~ ) = (1 "(k~ ))( +g+n)
= 0:053 per year. In the empirical Chapter 11 of Barro and Sala-i-Martin
(2004), it is argued that a lower value of (k~ ); say 0.02 per year, …ts the data
better. This requires "(k~ ) = 0:75: Such a high value of "(k~ ) ( the income
share of capital) may seem di¢ cult to defend. But if we reinterpret K in
the Solow model so as to include human capital (skills embodied in human
beings and acquired through education and learning by doing), a value of
"(k~ ) at that level may not be far out.
6.3.4 Adjustment time
Let ! be the time that it takes for the fraction ! 2 (0; 1) of the initial gap
between k~ and k~ to be eliminated, i.e., ! satis…es the equation
~
k( !) k~ ~ ! ) k~
k(
= =1 !; (6.18)
~
k(0) k~ ~
k(0) k~
where 1 ! is the fraction of the initial gap still remaining at time ! . In
~
(6.18) we have applied that sign(k(t) k~ ) = sign(k(0)
~ k~ ) in view of
monotonic convergence.
By (6.9), we have
~ ~ )
k( !) k~ ~
(k(0) k~ )e (k !
:
In view of (6.18), this implies
~ )
(k
1 ! e !
:
Taking logs on both sides and solving for ! gives
log(1 !)
! : (6.19)
(k~ )
This is the approximate adjustment time required for k~ to eliminate the
fraction ! of the initial distance of k~ to its steady-state value, k~ , when the
adjustment speed (speed of convergence) is (k~ ):
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
6.3. Within-country convergence speed and adjustment time 93
Often we consider the half-life of the adjustment, that is, the time it
takes for half of the initial gap to be eliminated. To …nd the half-life of the
adjustment of k;~ we put ! = 1 in (6.19). Again we use one year as our time
2
unit. With the parameter values from Section 6.3.3, we have (k~ ) = 0:053
per year and thus
log 12 0:69
1 = 13; 1 years.
2 0:053 0:053
As noted above, Barro and Sala-i-Martin (2004) estimate the asymptotic
speed of convergence to be (k~ ) = 0.02 per year. With this value, the
half-life is approximately
log 12 0:69
1 = 34:7 years.
2 0:02 0:02
And the time needed to eliminate three quarters of the initial distance to
steady state, 3=4 ; will then be about 70 years (= 2 35 years, since 1 3=4 =
1 1
2 2
).
Among empirical analysts there is not general agreement about the size of
~
(k ). Some authors, for example Islam (1995), using a panel data approach,
…nd speeds of convergence considerably larger, between 0:05 and 0:09. Mc-
Quinne and Whelan (2007) get similar results. There is a growing realization
that the speed of convergence di¤ers across periods and groups of countries.
Perhaps an empirically reasonable range is 0:02 < (k~ ) < 0:09: Correspond-
ingly, a reasonable range for the half-life of the adjustment will be 7:6 years
< 1 < 34:7 years.
2
Most of the empirical studies of convergence use a variety of cross-country
regression analysis of the kind described in the next section. Yet the theoret-
ical frame of reference is often the Solow model - or its extension with human
capital (Mankiw et al., 1992). These models are closed economy models with
exogenous technical progress and deal with “within-country”convergence. It
is not obvious that they constitute an appropriate framework for studying
cross-country convergence in a globalized world where capital mobility and to
some extent also labor mobility are important and where some countries are
pushing the technological frontier further out, while others try to imitate and
catch up. At least one should be aware that the empirical estimates obtained
may re‡ect mechanisms in addition to the falling marginal productivity of
capital in the process of capital accumulation.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
CHAPTER 6. TRANSITIONAL DYNAMICS. BARRO-STYLE
94 GROWTH REGRESSIONS
6.4 Barro-style growth regressions*
Barro-style growth regression analysis, which became very popular in the
1990s, draws upon transitional dynamics aspects (including the speed of con-
vergence) as well as steady state aspects of neoclassical growth theory (for
instance the Solow model or the Ramsey model).
In his Section 3.2 of Chapter 3 Acemoglu presents Barro’s growth regres-
sion equations in an unconventional form, see Acemoglu’s equations (3.12),
(3.13), and (3.14). The left-hand side appears as if it is just the growth rate
of y (output per unit of labor) from one year to the next. But the true left-
hand side of a Barro equation is the average compound annual growth rate of
y over many years. Moreover, since Acemoglu’s text is very brief about the
formal links to the underlying neoclassical theory of transitional dynamics,
we will spell the details out here.
Most of the preparatory work has already been done above. The point of
departure is a neoclassical one-sector growth model for a closed economy:
~ = s(k(t))f
k(t) ~ ~
(k(t)) ( + g + n)k(t); ~
k(0) = k~0 > 0; given, (6.20)
~
where k(t) K(t)=(A(t)L(t)); A(t) = A0 egt ; and L(t) = L0 ent as above.
~
The Solow model is the special case where the saving-income ratio, s(k(t));
is a constant s 2 (0; 1):
It is assumed that the model, (6.20), generates monotonic convergence,
~ ! k~ > 0 for t ! 1: Applying again a …rst-order Taylor approxi-
i.e., k(t)
~ now may depend
mation, as in Section 3.1, and taking into account that s(k)
on k;~ as for instance it generally does in the Ramsey model, we …nd the
asymptotic speed of convergence for k~ to be
~ = (1
SOC (k) "(k~ ) (k~ ))( + g + n) (k~ ) > 0; (*)
where (k~ ) k~ s0 (k~ )=s(k~ ) is the elasticity of the saving-income ratio w.r.t.
the e¤ective capital intensity, evaluated at k~ = k~ : (In case of the Ramsey
model, one can alternatively use the fact that SOC (k) ~ equals the absolute
value of the negative eigenvalue of the Jacobian matrix associated with the
dynamic system of the model, evaluated in the steady state. For a fully
speci…ed Ramsey model this eigenvalue can be numerically calculated by an
appropriate computer algorithm; in the Cobb-Douglas case there exists even
an explicit algebraic formula for the eigenvalue, see Barro and Sala-i-Martin,
2004). In a neighborhood of the steady state, the previous formulas remain
valid with (k~ ) de…ned as in (*). The asymptotic speed of convergence of for
example y(t)=y (t) is thus (k~ ) as given in (*). For notational convenience,
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
6.4. Barro-style growth regressions* 95
we will just denote it ; interpreted as a derived parameter, i.e.,
= (1 "(k~ ) (k~ ))( + g + n) (k~ ): (6.21)
In case of the Solow model, (k~ ) = 0 and we are back in Section 3.
In view of y(t) y~(t)A(t); we have gy (t) = gy~(t) + g: By (6.16) and the
de…nition of ,
gy (t) g "(k~ ) (log k(t)
~ log k~ ) g (log y(t) log y (t)); (6.22)
where the last approximation comes from (6.14). This generalizes Acemoglu’s
Equation (3.10) (recall that Acemoglu concentrates on the Solow model and
that his k is the same as our k~ ):
With the horizontal axis representing time, Figure 6.4 gives an illustration
of these transitional dynamics. As gy (t) = d log y(t)=dt and g = d log y (t)=dt;
(6.22) is equivalent to
d(log y(t) log y (t))
(log y(t) log y (t)): (6.23)
dt
So again we have a simple di¤erential equation of the form x(t)_ = x(t); the
solution of which is x(t) = x(0)e t : The solution of (6.23) is thus
t
log y(t) log y (t) (log y(0) log y (0))e :
As y (t) = y (0)egt ; this can be written
t
log y(t) log y (0) + gt + (log y(0) log y (0))e : (6.24)
The solid curve in Figure 6.4 depicts the evolution of log y(t) in the case
where k~0 < k~ (note that log y (0) = log f (k~ ) + log A0 ). The dotted curve
exempli…es the case where k~0 > k~ . The …gure illustrates per capita income
convergence: low initial income is associated with a high subsequent growth
rate which, however, diminishes along with the diminishing logarithmic dis-
tance of per capita income to its level on the steady state path.
For convenience, we will from now on treat (6.24) as an equality. Sub-
tracting log y(0) on both sides, we get
t
log y(t) log y(0) = log y (0) log y(0) + gt + (log y(0) log y (0))e
= gt (1 e t )(log y(0) log y (0)):
Dividing through by t > 0 gives
t
log y(t) log y(0) 1 e
=g (log y(0) log y (0)): (6.25)
t t
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
CHAPTER 6. TRANSITIONAL DYNAMICS. BARRO-STYLE
96 GROWTH REGRESSIONS
log y (t )
log y *(t )
g
log A0 log f (k *)
log A0 log f (k0 )
t
0
Figure 6.4: Evolution of log y(t). Solid curve: the case k~0 < k~ . Dotted curve: the
case k~0 > k~ . Stippled line: the steady-state path.
On the left-hand side appears the average compound annual growth rate of
y from period 0 to period t; which we will denote gy (0; t): On the right-hand
side appears the initial distance of log y to its hypothetical level along the
steady state path. The coe¢ cient, (1 e t )=t; to this distance is negative
and approaches zero for t ! 1: Thus (6.25) is a translation into growth
form of the convergence of log yt towards the steady-state path, log yt ; in the
theoretical model without shocks. Rearranging the right-hand side, we get
t t
1 e 1 e
gy (0; t) = g + log y (0) log y(0) b0 + b1 log y(0);
t t
where both the constant b0 g + (1 e t )=t log y (0) and the coe¢ cient
b1 (1 e t )=t are determined by “structural characteristics”. Indeed,
is determined by ; g; n; "(k~ ); and (k~ ) through (6.21), and y (0) is de-
termined by A0 and f (k~ ) through (6.12), where, in turn, k~ is determined
by the steady state condition s(k~ )f (k~ ) = ( + g + n)k~ ; s(k~ ) being the
saving-income ratio in the steady state.
With data for N countries, i = 1; 2;. . . ; N; a test of the unconditional
convergence hypothesis may be based on the regression equation
gyi (0; t) = b0 + b1 log yi (0) + i ; i N (0; 2
); (6.26)
where i is the error term. This can be seen as a Barro growth regression
equation in its simplest form. For countries in the entire world, the theoret-
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
6.4. Barro-style growth regressions* 97
ical hypothesis b1 < 0 is clearly not supported (or, to use the language of
statistics, the null hypothesis, b1 = 0; is not rejected).2
Allowing for the considered countries having di¤erent structural charac-
teristics, the Barro growth regression equation takes the form
gyi (0; t) = b0i + b1 log yi (0) + i ; b1 < 0; i N (0; 2
): (6.27)
In this “…xed e¤ects” form, the equation has been applied for a test of the
conditional convergence hypothesis, b1 < 0; often supporting this hypothesis.
That is, within groups of countries with similar characteristics (like, e.g., the
OECD countries), there is a tendency to convergence.
From the estimate of b1 the implied estimate of the asymptotic speed of
convergence, ; is readily obtained through the formula b1 (1 e t )=t:
1
Even ; and therefore also the slope, b ; does depend, theoretically, on
country-speci…c structural characteristics. But the sensitivity on these do
not generally seem large enough to blur the analysis based on (6.27) which
abstracts from this dependency.
With the aim of testing hypotheses about growth determinants, Barro
(1991) and Barro and Sala-i-Martin (1992, 2004) decompose b0i so as to re‡ect
the role of a set of potentially causal measurable variables,
b0i = 0 + 1 xi1 + 2 xi2 + ... + m xim ;
where the ’s are the coe¢ cients and the x’s are the potentially causal vari-
ables.3 These variables could be measurable Solow-type parameters among
those appearing in (6.20) or a broader set of determinants, including for in-
stance the educational level in the labor force, and institutional variables like
rule of law and democracy. Some studies include the initial within-country
inequality in income or wealth among the x’s and extend the theoretical
framework correspondingly.4
From an econometric point of view there are several problematic features
in regressions of Barro’s form (also called the convergence approach). These
problems are discussed in Acemoglu pp. 82-85.
2
Cf. Acemoglu, p. 16. For the OECD countries, however, b1 is de…nitely estimated to
be negative (cf. Acemoglu, p. 17).
3
Note that our vector is called in Acemoglu, pp. 83-84. So Acemoglu’s is to be
distinquished from our which denotes the asymptotic speed of convergence.
4
See, e.g., Alesina and Rodrik (1994) and Perotti (1996), who argue for a negative
relationship between inequality and growth. Forbes (2000), however, rejects that there
should be a robust negative correlation between the two.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
CHAPTER 6. TRANSITIONAL DYNAMICS. BARRO-STYLE
98 GROWTH REGRESSIONS
6.5 References
Alesina, A., and D. Rodrik, 1994, Distributive politics and economic growth,
Quarterly Journal of Economics, vol. 109, 465-490.
Barro, R. J., 1991, Economic growth in a cross section of countries, Quar-
terly Journal of Economics, vol. 106, 407-443.
Barro, R. J., X. Sala-i-Martin, 1992, Convergence, Journal of Political Econ-
omy, vol. 100, 223-251.
Barro, R., and X. Sala-i-Martin, 2004, Economic Growth. Second edition,
MIT Press: Cambridge (Mass.).
Cho, D., and S. Graham, 1996, The other side of conditional convergence,
Economics Letters, vol. 50, 285-290.
Forbes, K.J., 2000, A reassessment of the relationship between inequality
and growth, American Economic Review, vol. 90, no. 4, 869-87.
Groth, C., and R. Wendner, 2014, Embodied learning by investing and
speed of convergence, Journal of Macroeconomics, vol. 40, 245-269.
Islam, N., 1995, Growth Empirics. A Panel Data Approach, Quarterly
Journal of Economics, vol. 110, 1127-1170.
McQinn, K., K. Whelan, 2007, Conditional Convergence and the Dynamics
of the Capital-Output Ratio, Journal of Economic Growth, vol. 12,
159-184.
Perotti, R., 1996, Growth, income distribution, and democracy: What the
data say, Journal of Economic Growth, vol. 1, 149-188.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
Chapter 7
Michael Kremer’s
population-breeds-ideas model
This chapter relates to Section 4.2 of Acemoglu’s book where two special cases
of the population-breeds-ideas model (Kremer 1993) are presented. Here we
start with a more general version of the model. The point of the model is to
show that under certain conditions, the cumulative and nonrival character of
technical knowledge makes it likely that the Malthusian regime of stagnating
income per capita, close to existence minimum and caused by scarcity of
land, will sooner or later in the historical evolution be surpassed.
7.1 The general model
Suppose a pre-industrial economy can be described by:
Yt = At Lt Z 1 ; > 0; 0 < < 1; (7.1)
A_ t = At Lt ;
"
> 0; " S 0; A0 > 0 given; (7.2)
Yt
Lt = ; y > 0; (7.3)
y
where Y is aggregate output, A the level of technical knowledge, L the la-
bor force (= population), Z the amount of land (…xed), and y subsistence
minimum (so the ' in Acemoglu’s equation (4.2) is simply the inverse of the
subsistence minimum). Both Z and y are considered as constant parameters.
Time is continuous and it is understood that a kind of Malthusian population
mechanism (see below) is operative behind the scene.
The exclusion of capital from the aggregate production function, (7.1),
re‡ects the presumption that capital (tools etc.) is quantitatively of minor
99
CHAPTER 7. MICHAEL KREMER’S
100 POPULATION-BREEDS-IDEAS MODEL
importance in a pre-industrial economy. In accordance with the replication
argument, the production function has CRS w.r.t. the rival inputs, labor and
land. The factor At measures total factor productivity. As the right-hand
side of (7.2) is positive, the technology level, At ; is rising over time (although
far back in time very very slowly). The increase in At per time unit is seen
to be an increasing function of the size of the population. This re‡ects the
hypothesis that population breeds ideas; these are nonrival and enter the
pool of technical knowledge available for society as a whole. Indeed, the
use of an idea by one agent does not preclude others’use of the same idea.
Dividing through by L in (7.1) we see that y Yt =Lt = At (Z=Lt )1 : The
nonrival character is displayed by labor productivity being dependent on the
total stock of knowledge, not on this stock per worker. In contrast, labor
productivity depends on land per worker.
The rate per capita by which population breeds ideas is , A" : In case
" > 0; this rate is an increasing function of the already existing level of
technical knowledge. This case re‡ects the hypothesis that the larger is
the stock of ideas the easier do new ideas arise (perhaps by combination of
existing ideas). The opposite case, " < 0; is the one where “the easiest ideas
are found …rst”or “the low-hanging fruits are picked …rst”.
Equation (7.3) is a shortcut description of a Malthusian population mech-
anism. Suppose the true mechanism is
L_ t = (yt y)Lt T 0 for yt T y; (7.4)
where > 0 is the speed of adjustment, yt is per capita income, and y > 0 is
subsistence minimum. A rise in yt above y will lead to increases in Lt , thereby
generating downward pressure on Yt =Lt and perhaps end up pushing yt below
y: When this happens, population will be decreasing for a while and so return
towards its sustainable level, Yt =y: Equation (7.3) treats this mechanism as if
the population instantaneously adjusts to its sustainable level (as if ! 1).
The model hereby gives a long-run picture, ignoring the Malthusian ups and
downs in population and per capita income about the subsistence minimum.
The important feature is that the technology level, and thereby Yt ; as well
as the sustainable population will be rising over time. This speeds up the
arrival of new ideas and so Yt is raised even faster although per-capita income
remains at its long-run level, y.1
For simplicity, we now normalize the constant Z to be 1.
1
Extending the model with the institution of private ownership and competitive mar-
kets, the absence of a growing standard of living corresponds to the doctrine from classical
economics called the iron law of wages. This is the theory (from Malthus and Ricardo)
that scarce natural resources and the pressure from population growth causes real wages
to remain at subsistende level.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
7.2. Law of motion 101
Comparison with the two special cases in Acemoglu
At pp. 113-14 Acemoglu presents two versions of this framework, both of
which assume = 1 . This assumption is arbitrary; it is included as a
special case in our formulation above. As to the other parameter relating
to the role of knowledge, ", Acemuglu assumes " = 0 in his …rst version of
the framework. This leads to constant population growth but stagnating
standard of living (Acemoglu, p. 113). In his second version, Acemoglu
assumes " = 1: This leads to many centuries of slow but (weakly) accelerating
population growth and then ultimately a “takeo¤”with sustained rise in the
standard of living, to be followed by the “demographic transition” (outside
the model). This latter outcome arises for a much larger set of parameter
values and is therefore theoretically more robust than appears in Acemoglu’s
exposition.
7.2 Law of motion
The dynamics of the model can be reduced to one di¤erential equation, the
law of motion of technical knowledge. By (7.3), Lt = Yt =y = At Lt =y. Con-
sequently L1t = At =y so that
1
Lt = y 1 At1 : (7.5)
Substituting this into (7.2) gives the law of motion of technical knowledge:
1 "+ 1
A_ t = y 1 At ^ At ; (7.6)
where we have de…ned ^ y 1=( 1) and " + =(1 ): As will appear
in the remainder, the “feedback parameter” is of key importance for the
dynamics. We immediately see that if = 1; the di¤erential equation (7.6)
is linear, while otherwise it is nonlinear.
The case = 1 : When = 1; there will be a constant growth rate gA = ^
in technical knowledge. By (7.5), this results in a constant population growth
rate gL = [ =(1 )] ^ ; which is also the growth rate of output in view of
(7.3). By the de…nition of ^ in (7.6), we see that, as expected, the population
and output growth rate is an increasing function of the creativity parameter
and a decreasing function of the subsistence minimum.2
These classical economists did not recognize any tendency to sustained technical progress
and therefore missed the immanent tendency to population growth at the pre-industrial
stage of economic development. Karl Marx was the …rst among the classical economists
to really see and emphasize sustained technical progress.
2
If = 1 as in Acemoglu’s analysis, = 1 requires " = 0; and in this case L and
Y grow at the same rate as knowledge.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
CHAPTER 7. MICHAEL KREMER’S
102 POPULATION-BREEDS-IDEAS MODEL
In this case the economy never leaves the Malthusian regime of a more or
less constant standard of living close to existence minimum. Takeo¤ never
occurs.
The case 6= 1: Then (7.6) can be written
A_ t = ^ At ; (7.7)
which is a nonlinear di¤erential equation in A:3 Let x A1 : Then
x_ t = (1 )At ^ At = (1 )^; (7.8)
a constant. To …nd xt from this, we only need simple integration:
Z t
xt = x0 + x_ d = x0 + (1 ) ^ t:
0
1
As A = x 1 and x0 = A10 ; this implies
1 h i11 1
1
At = xt = A10 + (1 ^
) t =h : (7.9)
i 1
1
A01 ( 1) ^ t
There are now two sub-cases, > 1 and < 1: The latter sub-case leads
to permanent but decelerating growth in knowledge and population and the
Malthusian regime is never transcended (see Exercise III.3). The former
sub-case is the interesting one.
7.3 The inevitable ending of the Malthusian
regime when > 1
Assume > 1: In this case the result (7.9) implies that the Malthusian
regime must come to an end.
Although to begin with, At may grow extremely slowly, the growth in At
will be accelerating because of the positive feedback (visible in (7.2)) from
both rising population and rising At . Indeed, since > 1; the denominator
in (7.9) will be decreasing over time and approach zero in …nite time, namely
as t approaches the …nite value t = A10 =(( 1) ^ ): As an implication, At
goes towards in…nity in …nite time. The stylized graph in Fig. 7.1 illustrates.
The evolution of technical knowledge becomes explosive as t approaches t :
3
The di¤erential equation, (7.7), is a special case of what is known as the Bernoulli
equation. In spite of being a non-linear di¤erential equation, the Bernoulli equation always
has an explicit solution.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
7.3. The inevitable ending of the Malthusian regime when >1 103
A0
t
0 tˆ t*
Figure 7.1: Accelerating growth in A when the feedback parameter exceeds one.
It follows from (7.5) and (7.1) that explosive growth in A implies explosive
growth in L and Y; respectively. The acceleration in the evolution of Y will
sooner or later make Y rise fast enough so that the Malthusian population
mechanism (which for biological reasons has to be slow) can not catch up.
Then, what was in the Malthusian population mechanism, equation (7.4),
only a transitory excess of yt over y, will at some t = t^ < t become a
permanent excess and take the form of sustained growth in yt . This is known
as the takeo¤.
Note that Fig. 7.1 illustrates only what the process (7.7), with > 1; im-
plies as long as it rules, namely that knowledge goes towards in…nity in …nite
time. The process necessarily ceases to rule long before time t is reached,
however. This is because the process presupposes that the Malthusian popu-
lation mechanism keeps track with output growth which at some point before
t becomes impossible because of the acceleration in the latter.
In a neighborhood of this point the takeo¤ will occur, featuring sustained
growth in output per capita. According to equation (7.4) the takeo¤ should
also feature a permanently rising population growth rate. As economic his-
tory has testi…ed, however, along with the rising standard of living the de-
mographics changed radically (in the U.K. during the 19th century). The
demographic transition took place with fertility declining faster than mortal-
ity. This results in completely di¤erent dynamics about which the present
model has nothing to say.4 As to the demographic transition as such, ex-
4
Kremer (1993), however, also includes an extended model taking some of these changed
dynamics into account.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
CHAPTER 7. MICHAEL KREMER’S
104 POPULATION-BREEDS-IDEAS MODEL
planations suggested by economists include: higher real wages mean higher
opportunity costs of raising children instead of producing; reduced use of
child labor; the trade-o¤ between “quality” (educational level) of the o¤-
spring and their “quantity” (Becker, Galor)5 ; skill-biased technical change;
and improved contraception technology.
7.4 Closing remarks
The population-breeds-ideas model is about dynamics in the Malthusian
regime of the pre-industrial epoch. The story told by the model is the follow-
ing. When the feedback parameter, ; is above one, the Malthusian regime
has to come to an end because the battle between scarcity of land (or natural
resources more generally) and technological progress will inevitably be won
by the latter. The reason is the cumulative and nonrival character of tech-
nical knowledge. This nonrivalry implies economies of scale. Moreover, the
stock of knowledge is growing endogenously. This knowledge growth gener-
ates output growth and, through the demographic mechanism (7.3), growth
in the stock of people, which implies a positive feedback to the growth of
knowledge and so on. On top of this, if " > 0; knowledge growth has a direct
positive feedback on itself through (7.2). When the total positiv feedback is
strong enough ( > 1), it generates an explosive process.6
On the basis of demographers’ estimates of the growth in global popu-
lation over most of human history, Kremer (1993) …nds empirical support
for > 1: Indeed, in the opposite case, 1; there would not have been
a rising world population growth rate since one million years B.C. to the
industrial revolution. The data in Kremer (1993, p. 682) indicates that the
world population growth rate has been more or less proportional to the size
of population until recently.
Final remark. In the formulation of the model, I have made one sim-
pli…cation relative to Kremer’s setup. Kremer starts from a slightly more
general ideas-creation equation, namely A_ t = A"t Lt with > 0; while in
our (7.2) we have assumed = 1. If > 1; the ideas-creating brains rein-
force one another. This only forti…es the acceleration in knowledge creation
and thereby “supports”the case > 1.7 If on the other hand 0 < < 1; the
idea-creating brains partly o¤set one another, for instance by simultaneously
coming up with more or less the same ideas (the case of “overlap”). This
generalization does not change the qualitative results. By assuming that the
5
See Acemoglu, Section 21.2.
6
In the appendix the explosion result is considered in a general mathematical context.
7
Kremer’s calibration suggests 6=5:
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
7.5. Appendix 105
number of new ideas per time unit is proportional to the stock of brains,
we have chosen to focus on an intermediate case in order to avoid secondary
factors blurring the main mechanism.
7.5 Appendix
Mathematically, the background for the explosion result is that the solution
to a …rst-order di¤erential equation of the form x(t)
_ = + bx(t)c ; c > 1;
b 6= 0; x(0) = x0 given, is always explosive. Indeed, the solution, x = x(t);
will have the property that x(t) ! 1 for t ! t for some t > 0 where t
depends on the initial conditions; and thereby the solution is de…ned only on
a bounded time interval which depends on the initial condition.
Take the di¤erential equation x(t)
_ = 1 + x(t)2 as an example. As is
well-known, the solution is x(t) = tan t = sin t= cos t, de…ned on the interval
( =2; =2):
7.6 References
Becker, G. S., ...
Galor, O., 2011, Uni…ed Growth Theory, Princeton University Press.
Kremer, M., 1993, Population Growth and Technological Change: One Mil-
lion B.C. to 1990, Quarterly Journal of Economics 108, No. 3.
Møller, N. Framroze, and P. Sharp, 2014, Malthus in cointegration space:
Evidence of a post-Malthusian pre-industrial England, J. of Economic
Growth, vol. 19 (1), 105-140.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
CHAPTER 7. MICHAEL KREMER’S
106 POPULATION-BREEDS-IDEAS MODEL
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
Chapter 8
Choice of social discount rate
With an application to the climate change
problem
A controversial issue within economists’ debate on long-term public in-
vestment and in particular the climate change problem is the choice of the
social discount rate. This choice matters a lot for the present value of a
project which involves costs that begin now and bene…ts that occur only
after many years, say 75-100-200 years from now, as is the case with the
measures against global warming.
Compare the present value of receiving 1000 in‡ation-corrected euros a
hundred years from now under two alternative discount rates, r = 0:07 and
r = 0:01 per year:
r 100 0:9 if r = 0:07;
P V0 = 1000 e =
368 if r = 0:01:
So when evaluated at a 7 percent discount rate the 1000 in‡ation-corrected
euros a hundred years from now are worth less than 1 euro today. But with
a discount rate at 1 percent they are worth 368 euros today.
In this chapter we discuss di¤erent aspects of social discounting, that
is, discounting from a policy maker’s point of view. We shall set up the
theoretical framework around the concept of optimal capital accumulation
described in Acemoglu, Chapter 8, Section 8.3. In the …nal sections we apply
the framework to an elementary discussion of the climate change problem
from an economic perspective.
Unfortunately it is not always recognized that “discount rate”can mean
several di¤erent things. This sometimes leads to serious confusion, even
107
108 CHAPTER 8. CHOICE OF SOCIAL DISCOUNT RATE
within academic debates about policies addressing climate change. We there-
fore start with the ABC of discounting.
A discount rate is an interest rate applied in the construction of a discount
factor. The latter is a factor by which a project’s future costs or bene…ts,
measured in some unit of account, are converted into present equivalents.
Applying a discount factor thus allows economic e¤ects occurring at di¤erent
times to be compared. The lower the discount factor the higher the associated
discount rate.
8.1 Basic distinctions relating to discounting
A basic reason that we have to distinguish between di¤erent types of discount
rates is that there is a variety of possible units of account.
8.1.1 The unit of account
To simplify matters, in this section we assume there is no uncertainty. Future
market interest rates will thus with probability one be equal to the ex ante
expected future interest rates.
Think of period t as running from date t to date t + 1: More precisely,
think of period t as the time interval [t; t + 1) on a continuous time axis
with time unit equal to the period length. With time t thus referring to
the beginning of period t, we speak of “date t” as synonymous with time
t. This timing convention is common in discrete-time growth and business
cycle theory and is convenient because it makes switching between discrete
and continuous time analysis fairly easy.1 Unless otherwise indicated, our
period length, hence our time unit, will be one year.
Money as the unit of account
When the unit of account is money, we talk about a nominal discount rate.
More speci…cally, if the money unit is euro, we talk about an euro discount
rate. Consider a one-period bond promising one euro at date one to the
investor buying the bond at date 0. If the market interest rate is i0 ; the
present value at date 0 of the bond is
1
euro.
1 + i0
1
Note, however, that this timing convention is di¤erent from that in the standard
…nance literature where, for example, Kt would denote the end-of-period t stock that
begins to yield its services next period.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
8.1. Basic distinctions relating to discounting 109
In this calculation the (nominal) discount factor is 1=(1 + i0 ) and tells how
many euro need be invested in the bond at time 0 to obtain 1 euro at time
1. When the interest rate in this way appears as a constituent of a discount
factor, it is called a (nominal) discount rate. Like any interest rate it tells
how many additional units of account (here euros) are returned after one
period of unit length, if one unit of account (one euro) is invested in the
asset at the beginning of the period.2
A payment stream, z0 ; z1 ;. . . ; zt ;. . . , zT ; where zt (? 0) is the net payment
in euro due at the end of period t; has present value (in euro as seen from
the beginning of period 0)
z0 z1 zT 1
P V0 = + + + ; (8.1)
1 + i0 (1 + i0 )(1 + i1 ) (1 + i0 )(1 + i1 ) (1 + iT 1 )
where it is the nominal interest rate in euro on a one-period bond from date
t to date t + 1, t = 0; 1; : : : ; T 1:
_
The average nominal discount rate from date T to date 0 is the number
i 0;T 1 satisfying
_
1=T
1 + i 0;T 1 = ((1 + i0 )(1 + i1 ) (1 + iT 1 )) : (8.2)
The corresponding nominal discount factor is
_ 1
T
(1 + i 0;T 1) = : (8.3)
(1 + i0 )(1 + i1 ) (1 + iT 1)
If i is constant, the average nominal discount rate is of course the same as i
and the nominal discount factor is simply 1=(1 + i)T :
If the stream of z’s in (8.1) represents expected but uncertain dividends
to an investor as seen from date 0, we may ask: What is the relevant discount
rate to be applied on the stream by the investor? The answer is that the
relevant discount rate is that rate of return the investor can obtain generally
on investments with a similar risk pro…le. So the relevant discount rate is
simply the opportunity cost faced by the investor.
In continuous time with continuous compounding the formulas corre-
sponding to (8.1), (8.2), and (8.3) are
Z T Rt
i( )d
P V0 = z(t) e 0 dt; (8.4)
0
RT
_
0
i( )d
i (0; T ) ; and (8.5)
T
_ RT
e i (0;T )T
= e 0 i( )d : (8.6)
2
A discount factor is by de…nition a non-negative number. Hence, a discount rate in
discrete time is by de…nition greater than 1:
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
110 CHAPTER 8. CHOICE OF SOCIAL DISCOUNT RATE
And as above, if i is constant, the nominal discount factor takes the simple
form e iT :
Consumption as the unit of account
When the unit of account is a basket of consumption goods or, for simplicity,
just a homogeneous consumption good, we talk about a consumption discount
rate (or a real discount rate). Let the consumption good’s price in terms
of euros be Pt ; t = 0; 1;. . . , T: A consumption stream c0 ; c1 ;. . . ; ct ;. . . , cT ;
where ct is available at the end of period t; has present value (as seen from
the beginning of period 0)
c0 c1 cT 1
P V0 = + + + : (8.7)
1 + r0 (1 + r0 )(1 + r1 ) (1 + r0 )(1 + r1 ) (1 + rT 1)
Instead of the nominal interest rate, the proper discount rate is now the real
interest rate, rt ; on a one-period bond from date t to date t + 1: Ignoring
indexed bonds, the real interest rate is not directly observable, but can be
calculated in the following way from the observable nominal interest rate it :
Pt 1 (1 + it ) 1 + it
1 + rt = = ;
Pt 1+ t
where Pt 1 is the price (in terms of money) of a period-(t 1) consumption
good paid for at the end of period t 1 (= the beginning of period t) and
t Pt =Pt 1 1 is the in‡ation rate from period t 1 to period t:
The consumption discount factor (or real discount factor) from date t + 1
to date t is 1=(1+rt ): This discount factor tells how many consumption goods’
worth need be invested in the bond at time t to obtain one consumption
good’s worth at time t+1. The stream c0 ; c1 ;. . . ; ct ;. . . , cT could alternatively
represent an income stream measured in current consumption units. Then
the real interest interest rate, rt ; would still be the relevant real discount rate
and (8.7) would give the present real value of the income stream.
The average consumption discount rate and the corresponding consump-
tion discount factor are de…ned in a way analogous to (8.2) and (8.3), respec-
tively, but with it replaced by rt : Similarly for the continuous time versions
(8.4), (8.5), and (8.6).
Utility as the unit of account
Even though “utility”is not a measurable entity but just a convenient math-
ematical devise used to represent preferences, a utility discount rate is in
many cases a meaningful concept.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
8.1. Basic distinctions relating to discounting 111
Suppose intertemporal preferences can be represented by a sum of period
utilities discounted by a constant rate, :
u(c1 ) u(cT 1 )
U (c0 ; c1 ; ; cT 1) = u(c0 ) + + + ; (8.8)
1+ (1 + )T 1
where u( ) is the period utility function. Here appears as a utility dis-
count rate. The associated utility discount factor from date T to date 0 is
1=(1 + )T 1 : We may alternatively write the intertemporal utility function
as U~ (c0 ; c1 ; ; cT 1 ) (1 + ) 1 U (c0 ; c1 ; ; cT 1 ): Then the utility dis-
count factor from date T to date 0 appears instead as 1=(1 + )T ; which
in form corresponds exactly to (8.3); this di¤erence is, however, immaterial,
since U~ ( ) and U ( ) represent the same preferences and will imply the same
choices. In continuous time (with continuous compounding) the “sum” of
discounted utility is Z T
t
U0 = u(c(t))e dt;
0
t
where e is the utility discount factor from time t to time 0.3
8.1.2 The economic context
A cost-bene…t analysis is a systematic investigation of the costs and bene…ts
in a broad sense associated with a given project. The aim is to provide a
rational basis for decision making. In connection with a project by a national
or local government, it is common to add the pre…x “social” and speak of
social cost-bene…t analysis. But in principle the initiator of a cost-bene…t
analysis can be any agent in society.
When the timing of costs and bene…ts are central elements in the project,
we call it an investment project and then the choice of discount rate becomes
important. Along with the unit of account, the economic context of the
investment project to be evaluated matters for this choice. Here is a brief
list of key distinctions:
1. It matters whether the circumstances of relevance for the investment
project are endowed with certainty, computable risk, or non-computable
risk, also called fundamental uncertainty. In the latter case, the prob-
ability distribution is unknown (or scientists deeply disagree about it)
and, typically, the full range of possible outcomes is unknown.
3
Note that a …rst-order Taylor approximation of ex around x = 0 gives ex e0 +
0
e (x 0) = 1 + x for x “small”; hence, x ln(1 + x) for x “small”: Replacing x by and
taking powers, we see the analogy between e t and (1 + ) t : Because of the continuous
compounding, we have e t < (1 + ) t whenever > 0 and t > 0 and the di¤erence
increases with rising t:
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
112 CHAPTER 8. CHOICE OF SOCIAL DISCOUNT RATE
2. Length of the time horizon. Recently several countries have decided
to draw a line between less than vs. more than 30-50 years, choosing
a lower discount rate for years on the other side of the line. This is
in accordance with recommendations from economists and statisticians
arguing that the further ahead in time the discount rate applies, the
smaller should it be. With longer time horizons systematic risk and
fundamental uncertainty, about both the socio-economic environment
as such and the results of the speci…c project, play a larger role, thus
motivating precautionary saving.
3. A single or several di¤erent kinds consumption goods. As we shall
see below, the relevant consumption discount rate in a given context
depends on several factors, including the growth rate of consump-
tion. When fundamentally di¤erent consumption goods enter the util-
ity function - for instance an ordinary produced commodity versus ser-
vices from the eco-system - then a disaggregate setup is needed and the
relevant consumption discount rate may become an intricate matter.
Sterner and Persson (2008) give an introduction to this issue.
4. Private vs. social. Discounting from an individual household’s or …rm’s
point of view, as it occurs in private investment analysis, is one thing.
Discounting from a government’s point of view is another, and in con-
nection with evaluation of government projects we speak of social cost-
bene…t analysis. Here externalities and other market failures should
be taken into account. Whatever the unit of account, a discount rate
applied in social cost-bene…t analysis is called a social discount rate.
5. Micro vs. macro. Social cost-bene…t analysis may be concerned with a
microeconomic project and policy initiatives that involve only marginal
changes. In this case a lot of circumstances are exogenous (like in
partial equilibrium analysis). Alternatively social cost-bene…t analysis
may be concerned with a macroeconomic project and involve over-all
changes. Then more circumstances are endogenous, including possibly
the rate of economic growth and the quality of the natural environment
on a grand scale. In macroeconomic cost-bene…t analysis intra- and
intergenerational ethical issues are thus important. And in connection
with a macroeconomic project also the distinction between a closed and
an open economy, with access to international …nancial markets, comes
in.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
8.2. Criteria for choice of a social discount rate 113
8.2 Criteria for choice of a social discount
rate
There has been some disagreement among both economists and policy makers
about how to discount in social cost-bene…t analysis, in particular when the
economy as a whole and a long time horizon are involved. At one side we
have the descriptive approach to social discounting, sometimes called the
opportunity cost view:
According to this view, even when considering climate change
policy evaluation and caring seriously about future generations,
the average market rate of return, before taxes, is the relevant
discount rate. This is because funds used today to pay the cost of,
say, mitigating greenhouse gas emissions, could be set aside and
invested in other things and thereby accumulate at the market
rate of return for the bene…t of the future generations.
At the other side we …nd a series of opinions that are not easily lumped
together apart from their scepticism about the descriptive approach (in its
narrow sense as de…ned above). These “other views”are commonly grouped
together under the labels normative or prescriptive approach. This termino-
logical partitioning has become standard. With some hesitation we adopt it
here (the reason for the hesitation should become clear below).
One reason that the descriptive approach is by some considered inappro-
priate is the presence of market failures.4 Another is the presence of con‡ict-
ing interests: those people who bene…t may not be the same as those who
bear the costs. And where as yet unborn generations are involved, di¢ cult
ethical and coordination issues arise.
Amartia Sen (1961) pointed at the isolation paradox. Suppose each old
has an altruistic concern for all members of the next generation. Then a
transfer from any member of the old generation to the heir entails an exter-
nality that bene…ts all other members of the old generation. A nation-wide
coordination (political agreement) that internalizes these externalities would
raise intergenerational transfers (bequests etc.) and this corresponds to a
lowering of the intergenerational utility discount rate, , cf. (8.8).
More generally, members of the present generations may be willing to
join in a collective contract of more saving and investment by all, though
unwilling to save more in isolation.
4
Intervening into the debate about the suitable discount rate for climate change
projects, Heal (2008) asks ironically: “Is it appropriate to assume no market failure in
evaluating a consumption discount rate for a model of climate change?”.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
114 CHAPTER 8. CHOICE OF SOCIAL DISCOUNT RATE
Other reasons for a relatively low social discount rate have been pro-
posed. One is based on the super-responsibility argument: the government
has responsibility over a longer time horizon than those currently alive. An-
other is based on the dual-role argument: the members of the currently alive
generations may in their political or public role be more concerned about
the welfare of the future generations than they are in their private economic
decisions.
Critics of the descriptive approach may agree about the relevance of ask-
ing: “To what extent will investments made to reduce greenhouse gas emis-
sions displace investments made elsewhere?”. They may be inclined to add
that there is no guarantee that the funds in question are set aside for invest-
ment bene…tting generations alive two hundred years ahead, say.
Another point against the descriptive approach is that the future damages
of global warming could easily be underestimated. If nothing is done now, the
risk of the damage being irreparable at any cost becomes higher. Applying
the current market rate of return as discount rate for damages occurring
say 200 years from now on may imply that these damages become almost
imperceptible and so action tends to be postponed. This may be problematic
if there is a positive albeit low probability that a tipping point with disastrous
consequences is reached.
The reason for hesitation to lump together these “other views” under
the labels normative or prescriptive approach is that the contraposition of
“descriptive” versus “normative” in this context may be misleading. In the
…nal analysis also the descriptive approach has a normative element namely
the view that the social discount rate ought to be that implied by the market
behavior of the current generations as re‡ected in the current market interest
rate - the alternative is seen as paternalism. Here the other side of the debate
may respond that such “paternalism”need not be illegitimate but rather the
responsibility of democratically elected governments.
Anyway, in practice there seems to be a kind of convergence in the sense
that elements from the descriptive and the prescriptive way of thinking tend
to be combined. Nevertheless, there is considerable diversity across coun-
tries regarding the governments’o¢ cial “social consumption discount rate”
(sometimes just called the “social discount rate”) to be applied for public
investment projects. Even considering only West-European countries and
Western O¤shoots, including the U.S., the range is roughly from 8% to 2%
per year. An increasing fraction of these countries prescribe a lower rate for
bene…ts and costs accruing more than 30-50 years in the future (Harrison,
2010). The Danish Ministry of Finance recently (May 2013) reduced its social
consumption discount rate from 5% per year to 4% per year for the …rst 35
years of the time horizon of the project, 3% for the years in the interval 36 to
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
8.3. Optimal capital accumulation 115
69 years, and 2% for the remainder of the time horizon if relevant.5 Among
economists involved in climate policy evaluation, there is a wide range of
opinions as to what the recommended social discount rate should be (from
1.4% to 8.0%).6 An evaluation of the net worth of the public involvement
in the Danish wind energy sector in the 1990s gives opposite conclusions
depending on whether the discount rate is 5-6% (until recently the o¢ cial
Danish discount rate) or 3-4% (Hansen, 2010).
This diversity notwithstanding, let us consider some examples of social
cost-bene…t problems of a macroeconomic nature and with a long time hori-
zon. Our …rst example will be the standard neoclassical problem of optimal
capital accumulation.
8.3 Optimal capital accumulation
The perspective is that of an ”all-knowing and all-powerful” social planner
facing a basic intertemporal allocation problem in a closed economy: how
much should society save? The point of departure for this problem is the
prescriptive approach. The only discount rate which is decided in advance is
the utility discount rate, . No consumption discount rate is part of either
the objective function or the constraints. Instead, a long-run consumption
discount rate applicable to a class of public investment problems comes out
as a by-product of the steady-state solution to the problem.
8.3.1 The setting
We place our social planner in the simplest neoclassical set-up with exogenous
Harrod-neutral technical change. Uncertainty is ignored. Although time is
continuous, for simplicity we date the variables by sub-indices, thus writing
Yt etc. The aggregate production function is neoclassical and has CRS:
Yt = F (Kt ; Tt Lt ) Tt Lt f (k~t ); (8.9)
where Yt is output, Kt physical capital input, and Lt labor input which
equals the labor force which in turn equals the population and grows at the
constant rate n. The argument in the production function on intensive form
is de…ned by k~t Kt =(Tt Lt ): The factor Tt represents the economy-wide level
of technology and grows exogenously according to
Tt = T0 egt ; (8.10)
5
Finansministeriet (2013) .
6
Harrison (2010).
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
116 CHAPTER 8. CHOICE OF SOCIAL DISCOUNT RATE
where T0 > 0 and g 0 are given constants. Population grows at the
constant rate n 0: Output is used for consumption and investment so that
K_ t = Yt ct Lt Kt ; (8.11)
where ct is per capita consumption and 0 a constant capital depreciation
rate.
The social planner’s objective is to maximize a social welfare function,
W . We assume that this function is time separable with (i) an instantaneous
utility function u(c) with u0 > 0 and u00 < 0 and where c is per capita
consumption; (ii) a constant utility discount rate 0; often named “the
pure rate of time preference”; and (iii) an in…nite time horizon. The social
planner’s optimization problem is to choose a plan (ct )1
t=0 so as to maximize
Z 1
W = u(ct )Lt e t dt s.t. (8.12)
0
ct 0; (8.13)
ct
k~t = f (k~t ) ( + g + n)k~t ; k~0 > 0 given, (8.14)
Tt
k~t 0 for all t 0: (8.15)
Comments
1. If there are technically feasible paths along which the improper integral
W goes to +1, a maximum of W does not exist (in the CRRA case, u(c) =
c1 =(1 ); > 0; this will happen if and only if the parameter condition
n > (1 )g is not satis…ed). By “optimizing”we then mean …nding an
“overtaking optimal”solution or a “catching-up optimal”solution, assuming
one of either exists (cf. Sydsæter et al. 2008).
2. The long time horizon should be seen as involving many successive
and as yet unborn generations. Comparisons across time should primarily
be interpreted as comparisons across generations.
3. The model abstracts from inequality within generations.
4. By weighting per capita utility by Lt and thereby e¤ectively taking
population growth, n; into account, the social welfare function (8.12) respects
the principle of discounted classical utilitarianism. A positive pure rate of
time preference, ; implies discounting the utility of future people just be-
cause they belong to the future. Some analysts defend this discounting of the
future by the argument that it is a typical characteristic of an individual’s
preferences. Others …nd that this is not a valid argument for long-horizon
evaluations because these involve di¤erent persons and even as yet unborn
generations. For example Stern (2007) argues that the only ethically defen-
sible reason for choosing a positive is that there is always a small risk of
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
8.3. Optimal capital accumulation 117
extinction of the human race due to for example a devastating meteorite or
nuclear war. This issue aside, in (8.12) the e¤ective utility discount rate
will be n. This implies that the larger is n, the more weight is assigned
to the future because more people will be available.7 We shall throughout
assume that the size of population is exogenous although this may not ac-
cord entirely well with large public investment projects, like climate change
mitigation, that have implication for health and mortality. With endoge-
nous population very di¢ cult ethical issues arise (Dasgupta (2001), Broome
(2005)).
8.3.2 First-order conditions and their economic inter-
pretation
To characterize the solution to the problem, we use the Maximum Principle.
The current-value Hamiltonian is
h c i
H(k;~ c; ; t) = u(c) + f (k)~ ( + g + n)k~ ;
T
where is the adjoint variable associated with the dynamic constraint (8.14).
An interior optimal path (k~t ; ct )1
t=0 will satisfy that there exists a continuous
function = t such that, for all t 0;
@H
= 0; i.e., u0 (c) = ; and (8.16)
@c T
@H ~ _
= (f 0 (k) g n) = ( n) (8.17)
@ k~
hold along the path and the transversality condition,
lim k~t t e ( n)t
= 0; (8.18)
t!1
is satis…ed.
By taking logs on both sides of (8.16) and di¤erentiating w.r.t. t we get
du0 (ct )=dt u00 (ct ) _t
= c
_ t = g= (f 0 (k~t ) );
u0 (ct ) u0 (ct ) t
where the last equality comes from (8.17). Reordering gives
u00 (ct )
f 0 (k~t ) = + c_t ; (8.19)
u0 (ct )
7
In contrast, the principle of discounted average utilitarianism is characterized by pop-
ulation growth not a¤ecting the e¤ective utility discount rate. This corresponds to elimi-
nating the factor Lt in the integrand in (8.12).
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
118 CHAPTER 8. CHOICE OF SOCIAL DISCOUNT RATE
where the term ( u00 (ct )=u0 (ct )) > 0 indicates the rate of decline in marginal
utility when consumption is increased by one unit. So the right-hand side of
(8.19) exceeds when c_t > 0:
A technically feasible path satisfying both (8.19) and the transversality
condition (12.33) with t = Tt u0 (ct ) will be an optimal path and there are no
other optimal paths.8
The optimality condition (8.19) could of course be written on the stan-
dard Keynes-Ramsey rule form, where c_t =ct is isolated on one side of the
equation. But from the perspective of rates of return, and therefore discount
rates, the form (8.19) is more useful, however. The condition expresses the
general principle that in the optimal plan the marginal unit of per capita out-
put is equally valuable whether used for investment or current consumption.
When used for investment, it gives a rate of return equal to the net marginal
productivity of capital indicated on the left-hand side of (8.19). When used
for current consumption, it raises current utility. Doing this to an extent
just enough so that no further postponement of consumption is justi…ed, the
required rate of return is exactly obtained. The condition (8.19) says that
in the optimal plan the actual marginal rate of return (the left-hand side)
equals the required marginal rate of return (the right-hand side).
Reading the optimality condition (8.19) from the right to the left, there is
an analogy between this condition and the general microeconomic principle
that the consumer equates the marginal rate of substitution, MRS, between
any two consumption goods with the price ratio given from the market.
In the present context the two goods refer to the same consumption good
delivered in two successive time intervals. And instead of a price ratio we
have the marginal rate of transformation, MRT, between consumption in
the two time intervals as given by technology. The analogy is only partial,
however, because this MRT is, from the perspective of the optimizing agent
(the social planner) not a given but is endogenous just as much as the MRS
is endogenous.
8.3.3 The social consumption discount rate
More speci…cally, (8.19) says that the social planner will sacri…ce per capita
consumption today for more per capita consumption tomorrow only up to
the point where this saving for the next generations is compensated by a rate
8
This su¢ ciency and uniqueness follows from Mangasarian’s su¢ ciency theorem and
~ c~): The implied resource allocation
the fact that the Hamiltonian is strictly concave in (k;
will be the same as that of a competitive conomy which has the same technology as
that given in (8.9) and has a representative household that has the same intertemporal
preferences as those of the social planner given in (8.12) (this is the Equivalence theorem).
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
8.3. Optimal capital accumulation 119
of return su¢ ciently above : Naturally, the required compensation is higher,
the faster marginal utility declines with rising consumption, i.e., the larger
is ( u00 =u0 )c.
_ Indeed, every extra unit of consumption handed over to future
generations delivers a smaller and smaller marginal utility to these future
generations. So the marginal unit of investment today is only warranted if
the marginal rate of return is su¢ ciently above ; as indicated by (8.19).
Letting the required marginal rate of return be denoted rtSP and letting
the values of the variables along the optimal time path be marked by a bar,
we can write the right-hand side of (8.19) as
ct
rtSP = + (ct ) ; (8.20)
ct
where (c) cu00 (c)=u0 (c) > 0 (the absolute elasticity of marginal utility of
consumption). For a given (ct ); a higher per capita consumption growth rate
implies a higher required rate of return on marginal saving. In other words,
the higher the standard of living of future generations compared with cur-
rent generations, the higher is the required rate of return on current marginal
saving. Indeed, less should be saved for the future generations. Similarly, for
a given per capita consumption growth rate, ct =ct > 0; the required rate of
return on marginal saving is higher, the larger is (ct ): This is because (ct )
re‡ects aversion towards consumption inequality across time and generations
(in a context with uncertainty (ct ) also re‡ects what is known as the rela-
tive risk aversion, see below). Indeed, (ct ) indicates the percentage fall in
marginal utility when per capita consumption is raised by one percent. So a
higher (ct ) contributes to more consumption smoothing over time.
So far these remarks are only various ways of interpreting an optimality
condition. Worth emphasizing is:
The required marginal rate of return (the right-hand side of (8.20)) at
time t is not something given in advance, but an endogenous and time-
dependent variable which along the optimal path must equal the actual
marginal rate of return (the endogenous rate of return on investment
represented by the left-hand side of equation (8.20)). Indeed, both
the required and the actual marginal rates of return are endogenous
because they depend on the endogenous variables ct and c_t and on
what has been decided up to time t and is re‡ected in the current value
of the state variable, k~t . As we know from phase diagram analysis in
~ c=T ) plane, there are in…nitely many technically feasible paths
the (k;
satisfying the inverted Keynes-Ramsey rule in (8.20) for all t 0: What
is lacking up to now is to select among these paths one that satis…es
the transversality condition (12.33).
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
120 CHAPTER 8. CHOICE OF SOCIAL DISCOUNT RATE
In the present problem the only discount rate which is decided in ad-
vance is the utility discount rate, . No consumption discount rate is
part of either the objective function or the constraints. We shall now
see, however, that a long-run consumption discount rate applicable to
(less-inclusive) public investment problems comes out as a by-product
of the steady-state solution to the problem.
Steady state
To help existence of a steady state we now assume that the instantaneous
utility, u(c); belong to the CRRA family so that (c) = ; a positive constant.
Then
c1 1
; when > 0; 6= 1;
u(c) = 1 (8.21)
ln c; when = 1:
We know that if the parameter condition n > (1 )g holds and f
satis…es the Inada conditions, then there exists a unique path satisfying the
necessary and su¢ cient optimality conditions, including the transversality
condition (12.33). Moreover, this path converges to a balanced growth path
with a constant e¤ective capital-labor ratio, k~ ; satisfying f 0 (k~ ) = + g:
So, at least for the long run, we may replace ct =ct in (8.20) with the constant
rate of exogenous technical progress, g: Then (8.20) reduces to a required
consumption rate of return that is now constant and given by the parameters
in the problem:
rSP = + g: (8.22)
This rSP is the prevalent suggestion for the choice of a social consumption
discount rate. Note that as long as g > 0; rSP will be positive even if
= 0: A higher will imply stronger discounting of additional consumption
in the future because higher means faster decline in the marginal utility of
consumption in response to a given rise in consumption. So with g equal to,
say, 1.5% per year, the social discount rate rSP is in fact more sensitive to
the value of than to the value of : Note also that a higher g raises rsp and
thereby reduces the incentive to save and invest.
Now consider a public investment project with time horizon T ( 1)
which comes at the expense of an investment in capital in the “usual” way
as described above. Suppose the project is “minor” or “local” in the sense
of not a¤ecting the structure of the economy as a whole, like for instance
the long-run per capita growth rate, g: Let the project involve an initial
investment outlay of k0 and a stream of real net revenues, (zt )Tt=0 , assuming
that both costs and bene…ts are measurable in terms of current consumption
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
8.3. Optimal capital accumulation 121
equivalents.9 Letting rSP serve to convert future consumption into current
consumption equivalents, we calculate the present value of the project,
Z T
rSP t
P V0 = k0 + zt e dt:
0
The project is worth undertaking if P V0 > 0:
Limitations of the Ramsey formula rSP = + g
For a closed economy, reasonably well described by the model, it makes
sense to choose the rSP given in (8.22) as discount rate for public investment
projects if the economy is not “far”from its steady state. There are several
cases where modi…cation is needed, however:
1. Assuming the model still describes the economy reasonably well, if
the actual economy is initially “far” from its steady state and T is of
moderate size, g in (8.22) should be replaced by a somewhat larger
value if k~0 < k~ (since in that case c=c
_ > g) and somewhat smaller value
~ ~
if k0 > k (since in that case c=c_ < g):
2. The role of natural resources, especially non-renewable natural resources,
has been ignored. If they are essential inputs, the parameter g needs
reinterpretation and a negative value can not be ruled out apriori. In
that case the social discount rate can in principle be negative.
3. Global problems like the climate change problem has an important in-
ternational dimension. As there is great variation in the standard of
living, c; and to some extent also in g across developed and develop-
ing countries, it might be relevant to include not only a parameter, 1 ;
re‡ecting aversion towards consumption inequality over time and gen-
erations but also a parameter, 2 ; re‡ecting aversion towards spatial
consumption inequality, i.e., inequality across countries.
4. Another limitation of the Ramsey formula (8.22), as it stands, is that
it ignores uncertainty. In particular with a long planning horizon un-
certainty both concerning the results of the investment project and
concerning the socio-economic environment are important and should
of course be incorporated in the analysis.
9
We bypass all the di¢ cult issues involved in converting non-marketed goods like envi-
ronmental qualities, biodiversity, health, and mortality risk etc. into consumption equiv-
alents.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
122 CHAPTER 8. CHOICE OF SOCIAL DISCOUNT RATE
5. Finally, for “large” macroeconomic investment projects, the long-run
technology growth rate may not be given, but dependent on the chosen
policy. In that case, neither g nor rSP are given. This is in fact the typ-
ical situation within the macroeconomic theory of endogenous produc-
tivity growth. Then formulation of a “broader” optimization problem
is necessary. Only parameters like the utility discount rate, ; and the
elasticity of marginal utility of consumption, ; will in this case serve as
points of departure. There are no “correct”values for these parameters
in any objective and unconditional sense. Ethical judgments inevitably
enter.
In connection with the climate change problem we shall in the next section
apply a brief article by Arrow (2007)10 to illustrate at least one way to deal
with the problems 4 and 5.
8.4 The climate change problem from an eco-
nomic point of view
There is now overwhelming agreement among scientists that man-made global
warming is a reality. Mankind faces a truly large-scale economic external-
ity problem with potentially dramatic consequences for global economic and
social development in centuries.
Future economic evolution is uncertain and depends on policies chosen
now. A series of possible “act now”measures has been described in detail in
the voluminous The economics of climate change. The Stern Review, made
by a team of researchers lead by the prominent British economist Nicholas
Stern (Stern 2007).
The just mentioned article by Arrow is essentially a comment on the
Stern Review and on the debate about discount rates it provoked among
climate economists as well as in political circles. It is Arrow’s view that
taking risk aversion properly into account implies that the conclusion of the
Stern Review goes through: Mankind is better o¤ to act now to reduce CO2
emissions, and thereby the “greenhouse e¤ect”, substantially rather than to
risk the consequences of failing to meet this challenge. In many areas of
life, high insurance premia are willingly paid to reduce risks. It is in such a
perspective that part of the costs of mitigation should be seen.
10
Arrow won the Nobel Prize in Economics in 1972.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
8.4. The climate change problem from an economic point of view 123
8.4.1 Damage projections
As asserted by the Stern Review, the CO2 problem is “the greatest and
widest-ranging market failure ever seen”(Stern 2007, p. ). The current level
of CO2 (including other greenhouse gases, in CO2 equivalents) is today (i.e.,
in 2007) about 430 parts per million (ppm), compared with 280 ppm before
the industrial revolution. Under a “business as usual” assumption the level
will likely be around 550 ppm by 2035 and will continue to increase. The
level 550 ppm is almost twice the pre-industrial level, and a level that has
not been reached for several million years.
Most climate change models predict this would be associated with a rise
in temperature of at least two degrees Centigrade, probably more. A contin-
uation of “business as usual”is likely to lead to a trebling of CO2 by the end
of the century and to a 50% likelihood of a rise in temperature of more than
…ve degrees Centigrade. Five degrees Centigrade are about the same as the
increase from the last ice age to the present.
The full consequences of such rises are not known. But drastic negative
e¤ects on agriculture in the heavily populated tropical regions due to changes
in rainfall patterns are certain. The rise in the sea level will wipe out small
island countries, and for example Bangladesh will loose much of its land area.
A reversal of the Gulf Stream is possible, which could cause climate in Europe
to resemble that of Greenland. Tropical storms and other kinds of extreme
weather events will become more frequent. Many glaciers will disappear and
with them, valuable water supplies. After a climate tipping point has been
passed, self-reinforcing processes are activated. An example is that global
warming may thaw permafrost areas in Siberia and thereby release large
amounts of the greenhouse gas methane, thus accelerating global warming.
The really challenging factors are that the emissions of CO2 and other
gases are almost irreversible. They constitute a global negative externality
at a grand scale. The Stern Review assesses that avoiding such an outcome
is possible by a series of concrete measures (carbon taxes, technology policy,
international collective action) aimed at stopping or at least reducing the
emission of green-house gases and mitigate their consequences. Out of the
Stern Review’s suggested range of the estimated costs associated with this,
in his evaluation of an “act now”policy Arrow chooses a cost level of 1% of
GNP every year forever (see below).
According to many observers, postponing action is likely to increase both
risks and costs. The Stern Review suggests that the costs of action now
are less than the costs of inaction because the marginal damages of rising
temperature increase strongly as temperatures rise. In the words by Nobel
Laureate, Joseph Stiglitz: “[The Stern Review] makes clear that the question
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
124 CHAPTER 8. CHOICE OF SOCIAL DISCOUNT RATE
is not whether we can a¤ord to act, but whether we can a¤ord not to act”
(Stiglitz, 2007).
Let us now consider Arrow’s reasoning.
8.4.2 Uncertainty, risk aversion, and the certainty-equivalent
loss
The point of departure for both Arrow and Stern Review is a social (in fact
“global”) welfare function
Z 1 Z 1
( n)t
W = E0 u(ct )e dt = e ( n)t E0 u(ct )dt; (8.23)
0 0
where E0 is the expectation operator conditional on information up time 0,
and where risk aversion is assumed present, i.e., u00 (c) < 0:
Since there is uncertainty about the size of the future damages - reduc-
tions in c -, we follow Arrow’s attempt to convert this uncertainty into a
certainty-equivalent damage. When risk aversion is present, an uncertain
gain is evaluated as being equivalent to a certain gain smaller than the ex-
pected value (the “average”) of the possible outcomes. With the green-house
gas e¤ect, mankind is facing an uncertain damage. This should be evaluated
as being equivalent to a certain loss greater than the expected value of the
possible damages. For the so-called High-climate Scenario (considered by Ar-
row to be the best-substantiated scenario) the Stern Review estimates that
by year 2200 the losses in global GNP per capita, by following a “business as
usual”policy rather than a “mitigation now”policy, have an expected value
of 13.8% of what global GNP per capita would be if green-house gas concen-
tration is prevented from exceeding 550 ppm. The estimated loss distribution
has a 0.05 percentile of about 3% and a 0.95 percentile of about 34%.
Assuming consumption per capita, c; in year 2200 is proportional to GNP
per capita in year 2200, the alternatives for that year thus are:
under “mitigation now”policy (MNP): c = c1 ;
under “business as usual”(BAU): c = (1 X)c1 c0 ;
where c1 is considered given while X is a stochastic variable measuring the
fraction of c1 lost in year 2200 due to the damage occurring under BAU. A
probability density function of X according to the High-climate RScenario is
1
represented by f (x) in Figure 8.1. The expected loss of EX = 0 xf (x)dx
= 0:138 is indicated and so are the 5th and 95th percentiles of 0.03 and 0.34,
respectively.11 The distribution is right-skew.
11
The Stern Review estimates that X < 0 has zero probability.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
8.4. The climate change problem from an economic point of view 125
Figure 8.1: The density function of the per capita consumption loss X in year
2200.
Let x0 denote the certainty-equivalent loss, that is, the number x0 satis-
fying
u((1 x0 )c1 ) = Eu((1 X)c1 ) = Eu(c0 ): (8.24)
This means that an agent with preferences expressed by u is indi¤erent be-
tween facing a certain loss of size x0 or an uncertain loss, X; that has density
function f .
The condition (8.24) is illustrated in Figure 8.2. The density function for
the stochastic BAU consumption level, c0 , is indicated in the lower panel of
the …gure by a reversed coordinate system. If the utility function is speci…ed
and one knows the complete density function, then Eu((1 X)c1 ) is known
and the certainty-equivalent BAU consumption level (1 x0 )c1 , can be read
o¤ the diagram.
The instantaneous utility function, u(c); chosen by Arrow as well as Stern
is of the CRRA form (8.21). Arrow proposes the value 2 for ; while the Stern
Review relies on = 1 which by many critics was considered “too low”from
a descriptive-empirical point of view. As mentioned above, in a context of
uncertainty, not only measures the aversion towards consumption inequality
across time and generations but also the level of relative risk aversion.
The problem now is that the loss density function f (x) is not known.
The Stern Review only reports an estimated mean of 0:138 together with
estimated 5th and 95th percentiles of 0.03 and 0.34, respectively. This does
not su¢ ce for calculation of a good estimate of expected utility, Eu((1
X)c1 ): At best one can give a rough approximation. Arrow’s approach to
this problem is to split the probability mass into two halves and place them
on the 5th and 95th percentiles, respectively, assuming this gives a reasonable
approximation:
Eu((1 X)c1 ) u((1 0:03)c1 )0:5 + u((1 0:34)c1 )0:5: (8.25)
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
126 CHAPTER 8. CHOICE OF SOCIAL DISCOUNT RATE
Figure 8.2: The certanty-equivalent loss, x0 , assuming expected utility is known.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
8.4. The climate change problem from an economic point of view 127
With u(c) given as CRRA, by (8.24) and (8.25) we thus have
[(1 x0 )c1 ]1 [(1 0:03)c1 ]1 [(1 0:34)c1 ]1
0:5 + 0:5;
1 1 1
since the additive constant 1=(1 ) cancels out on both sides. We see that
also c11 =(1 ) cancels out on both sides so that we are left with
(1 x0 ) 1 (1 0:03)1 0:5 + (1 0:34)1 0:5:
With = 2 the approximative estimate of the certainty-equivalent loss is
x^0 = 0:21, that is “about 20%” (of GNP per capita in year 2200) as Arrow
says (Arrow 2007, p. 5).12
Here we shall proceed with this estimate of the certainty-equivalent loss
while in the appendix we raise a little doubt about the unbiasedness of the
estimate. On average the estimated certainty-equivalent loss corresponds to
a decrease of the expected growth rate per year of GNP per capita between
year 2001 and year 2200 from g1 = 1:3% (the base rate of GNP per capita
growth before the damages by further “business as usual”) to g0 = 1:2% per
year.
8.4.3 Comparing bene…t and cost of mitigation now
Avoiding the projected fall in average per capita consumption growth is thus
the bene…t of the “mitigation now” policy while the costs amount to the
above-mentioned 1% of GNP every year forever.
The criterion for assessing whether the “mitigation now”policy is worth
the costs is the “global welfare function” presented in (8.23) above with
instantaneous utility being of CRRA form.13 The Stern Review has been
criticized by several economic analysts for adopting “too low”values of both
the two intergenerational preference parameters, and : As to the rate of
time preference, ; following the “descriptive approach”, these critics argue
that a level about 1-3% per year is better in line with a backward calculation
from observed market rates of return. Anticipating such criticism, the Stern
Review …ghts back by claiming that such high values are not ethically de-
fensible since they amount to discriminating future generations for the only
reason that they belong to the future. As mentioned in Section 8.3.1, Stern
argues that the only ethically defensible reason for choosing a positive is
12
Although the calculation behind these “about 20%”is not directly reported in Arrow’s
brief article, he has in an e-mail to me con…rmed that (8.25) is the applied method.
13
We ignore the minor di¤erence vis-a-vis the Stern Review that it brings in a so-called
scrap value function subsuming discounted utility from year 2200 to in…nity.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
128 CHAPTER 8. CHOICE OF SOCIAL DISCOUNT RATE
that there always is a small risk of extinction of the human race due to for
example a devastating meteorite or nuclear war. Based on this view, Stern
chooses a value of close to zero, namely = 0:001.14
As Arrow argues and as we shall see in a moment, this disagreement as
to the size of is not really crucial given the bene…ts and costs involved.
Regarding ; which measures the (absolute) elasticity of marginal utility of
consumption and thereby aversion to consumption variation, both across
time and across alternative uncertain outcomes, we follow Arrow and let
equal 2 (while, as mentioned, Stern Review has = 1):
The break-even utility discount rate
Assuming balanced growth with some constant productivity growth rate, g;
consumption per capita will also grow at the rate g; i.e., ct = c(0)egt for all
t 0:15 Then
(c(0)egt )1
u(ct ) = ;
1
along the balanced growth path.
Now, based on the certainty-equivalent growth path under “business as
usual”(BAU) policy, the social welfare function then takes the value16
Z 1 Z 1
c(0)1 g0 t 1 ( n)t c(0)1
W0 = (e ) e dt = e[(1 )g0 ( n)]t
dt
1 0 1 0
c(0)1 1
= ;
1 n (1 )g0
where g0 is the (constant) per capita consumption growth rate under BAU
policy.
Let the value of the social welfare function under the “mitigation now”
policy be denoted W1 : According to the numbers mentioned above, the latter
policy involves a cost whereby c(0) is replaced by c(0)0 = 0:99c(0) and a
14
This is in fact a relatively high value of in the sense that it suggests that the
probability of extinction within one hundred years from now is as high as 9.5% (1 P (X <
x)) = 1 e 0:1 = 0:095): But as the Stern Review (p. 53) indicates, the term “extinction”
is meant to include “partial extinction by som exogenous or man-made force which has
little to do with climate change”.
15
To avoid confusion with our previous c0 ; appearing in (8.24), we write initial per capita
consumption as c(0) rather than c0 :
16
The transversality condition holds and the utility integral W0 is convergent if n
> (1 )g0 : In the present case where = 0:001; = 2 and g0 = 0:012; W0 is thus
convergent for n < (1 )g0 = + g0 = 0:013: This inequality seems likely to hold.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
8.4. The climate change problem from an economic point of view 129
bene…t whereby g0 = 0:012 is replaced by g1 = 0:013.17 We get
(0:99c(0))1 1
W1 = :
1 n (1 )g1
Since the bene…ts of the “mitigation now”policy come in the future and
the costs are there from date zero, we have W1 > W0 only if the e¤ective
utility discount rate, n; is below some upper bound. Let us calculate the
least upper bound in this regard. With = 2; we have
1 1 1 1
W1 = (0:99c(0)) > W0 = (c(0))
n + g1 n + g0
1 1
) <
0:99( n + g1 ) n + g0
) 0:99( n + g1 ) > n + g0
) 0:99g1 g0 > 0:01( n)
) 0:00087 > 0:01( n)
) n < 0:087 or n < 8:7% per year.
The break-even point for n at which W1 = W0 is thus 8:7% per year.
As Arrow remarks, “no estimate of the pure rate of time preference even
by those who believe in relatively strong discounting of the future has ever
approached 8.5%”.18 The conclusion is that given the estimated certainty-
equivalent loss, the “mitigation now” policy passes the cost-bene…t test for
any reasonable value of the pure rate of time preference.
It should be mentioned that there has been considerably disagreement
also about other aspects of the Stern Review’s investigation, not the least
the time pro…les for the projected bene…ts and costs.19 So it is fair to say that
“further sensitivity analysis is called for”, as Arrow remarks. He adds: “Still,
I believe there can be little serious argument over the importance of a policy
of avoiding major further increases in combustion by-products”(Arrow 2007,
p. 5)
17
By taking g1 = 0:013 > g0 also after year 2200, we deviate a little from both Arrow
and Stern in a direction favoring the Stern conclusion slightly.
18
Possibly the di¤erence between Arrow’s 8.5% and our result is due to the point men-
tioned in the previous footnote. Another minor di¤erence is that Arrow seemingly takes n
to be zero since he speaks of the “pure rate of time preference” rather than the “e¤ective
rate of time preference”, n.
19
See for example: http://en.wikipedia.org/wiki/Stern_Review#cite_ref-5
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
130 CHAPTER 8. CHOICE OF SOCIAL DISCOUNT RATE
8.5 Conclusion
In his brief analysis of the economics of the climate change problem Arrow
(2007) …nds the fundamental conclusion of the Stern Review justi…ed even
if one, unlike the Stern Review, heavily discounts the utility of future gen-
erations. In addition to discounting, risk aversion plays a key role in the
argument. A signi…cant part of the costs of mitigation is like an insurance
premium society should be ready to pay.
The analysis above took a computable risk approach. For more elaborate
treatments of the uncertainty issues, also involving situations with systematic
uncertainty, about c1 for instance, increasing with the length of the time
horizon, as well as fundamental uncertainty, see the list of references, in
particular the papers by Gollier and Weitzman.
We have been tacit concerning the di¢ cult political economy problems
about how to obtain coordinated international action vis-a-vis global warm-
ing. About this, see, e.g., Gersbach (2008) and Roemer (2010).
Nicholas Stern, who is still involved in climate change research and policy,
said in an interview at the World Economic Forum 2013 in Davos: “Looking
back, I underestimated the risks. The planet and the atmosphere seem to
be absorbing less carbon than we expected, and emissions are rising pretty
strongly. Some of the e¤ects are coming through more quickly than we
thought then”(interview in the Guardian, January 26, 2013).
8.6 Appendix: A closer look at Arrow’s esti-
mate of the certainty-equivalent loss
In this appendix we brie‡y discuss Arrow’s estimate of the certainty-equivalent
loss based on (8.25). The applied procedure would be accurate if the density
function f (x) were symmetric and the utility function u(c) were linear.
So let us …rst consider the case of a linear utility function, u~(c); cf. the
stippled positively sloped line in Figure 8.3. With f (x) symmetric, EX co-
incides with the median of the distribution. Given the estimated 5th and
95th percentiles of 0.03 and 0.34, respectively, we would thus have EX
= (0:34 0:03)=2 = 0:156. So E((1 X)c1 ) = (1 0:156)c1 . In view of
u~(c) being linear, we then get u~((1 0:156)c1 ) = E u~((1 X)c1 ). And for
this case an estimate of the certainty-equivalent loss, x0 ; of course equals
EX = 0:156:
The “true”density function, f (x), is right-skew, however, and has EX =
0:138: In combination with the linear utility function, u~(c); this implies an
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
8.6. Appendix: A closer look at Arrow’s estimate of the certainty-equivalent
loss 131
Figure 8.3: The case of symmetric density. Comparison of linear and strictly
concave utility function.
estimate of x0 equal to 0:138; that is, we get a lower value for the certainty-
equivalent loss than with a symmetric density function.
Now let us consider the “true” utility function, u(c). In Figure 8.3 it is
represented by the solid strictly concave curve u(c0 ). Let us again imagine
for a while that the density function is symmetric. As before, half of the
probability mass would then be to the right of the mean of c0 , (1 0:156)c1 ,
and the other half to the left. The density function might happen to be such
that the expected utility is just the average of utility at the 5th percentile
and utility at the 95th percentile, that is, as if the two halves of the prob-
ability mass were placed at the 5th and 95th percentiles of 0.03 and 0.34,
respectively; if so, the estimated certainty-equivalent loss is the x^0 shown in
Figure 8.3.
This would just be a peculiar coincidence, however. The probability mass
of the symmetric density function could be more, or less, concentrated close
to EX = 0:156: In case it is more concentrated, it is as if the two halves of
the probability mass are placed at the consumption levels (1 0:156 + a)c1
and (1 0:156 a)c1 for some “small” positive a; cf. Figure 8.3. The
corresponding estimate of the certainty-equivalent loss is denoted x^00 in the
…gure and is smaller than x^0 so that the associated c0 is larger than before.
Finally, we may conjecture that allowing for the actual right-skewness
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
132 CHAPTER 8. CHOICE OF SOCIAL DISCOUNT RATE
of the density function will generally tend to diminish the estimate of the
certainty-equivalent loss.
The conclusion seems to be that Arrow’s procedure, as it stands, is ques-
tionable. Or the procedure is based on assumptions about the properties of
the density function not spelled out in the brief article. Anyway, sensitivity
analysis is called for. This could be part of an interesting master’s thesis by
someone better equipped in mathematical statistics than the present author
is.
8.7 References
Arrow, K. J., W. R. Cline, K. G. Mäler, M. Munasinghe, R. Squitieri, and
J. E. Stiglitz, 1996, “Intertemporal Equity, Discounting, and Economic
E¢ ciency”. In: Climate Change 1995. Economic and Social Dimen-
sions of Climate Change: Contribution of Working Group III to the
Second Assessment Report of the Intergovernmental Panel on Climate
Change, ed. by J. P. Bruce, H. Lee, and E. F. Haites. Cambridge, UK:
Cambridge University Press, 125–44.
Arrow, K. J., 2007, Global climate change: A challenge to policy, The
Economist’s Voice, vol. 4 (3), Article 2. Available at
http://www.bepress.com/ev/vol4/iss3/art2/
Arrow, K. J., 2009, A note on uncertainty and discounting in models of
economic growth, J. Risk and Uncertainty, vol. 38, 87-94.
Arrow, K. J., M. L. Cropper, C. Gollier, B. Groom, G. M. Heal, R. G.
Newell, W. D. Nordhaus, R. S. Pindyck, W. A. Pizer, P. R. Portney,
T. Sterner, R. S. J. Tol, and M. L. Weitzman, 2012, “How should
bene…ts and costs be discounted in an intergenerational context? The
views of an expert panel”, Resources for the Future, DP 12-53.
Atkinson, G., S. Dietz, J. Helgeson, C. Hepburn, and H. Sælen, 2009, Sib-
lings, not triplets: Social preferences for risk, inequality and time in
discounting climate change, Economics e-journal, vol. 3.
Broome, J., 2005, Should we value population?, The Journal of Political
Philosophy, vol. 13 (4), 399-413.
Dasgupta, P., 2001, Human well-being and the natural environment, Oxford:
Oxford University Press.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
8.7. References 133
Gersbach, H., 2008, A new way to address climate change: A global refund-
ing system, Economists’Voice, July.
Gollier, C., 2008, Discounting with fat-tailed economic growth, J. Risk and
Uncertainty, vol. 37, 171-186.
Gollier, C., 2011, On the under-estimation of the precautionary e¤ect in
discounting, CESifo WP no. 3536, July.
Goulder, L.H., and R.C. Williams, 2012, The choice of discount rate for
climate change policy evaluation, Resources for the Future, DP 12-43.
Groth, C., and P. Schou, 2007, Growth and non-renewable reources: The
di¤erent roles of capital and resource taxes, Journal of Environmental
Economics and Management, vol. 53, 80-98.
Finansministeriet, 1999, Vejledning udarbejdelse af samfundsøkonomiske kon-
sekvensvurderinger, Kbh.
Finansministeriet, 2013, Ny og lavere samfundsøkonomisk diskonteringsrate,
Faktaark 31. maj 2013.
Hansen, A. C., 2010, Den samfundsmæssige kalkulationsrente, Nationaløkonomisk
Tidsskrift, vol. 148, no. 2, 159-192.
Harrison, M., 2010, Valuing the future: The social discount rate in cost-
bene…t analysis, Productivity Commission, Canberra.
Heal, G. F., 1998, Valuing the Future: Economic Theory and Sustainability,
Columbia University Press, New York.
Heal, G. F., 2008,
Hepburn, C., 2006, Discounting climate change damages: Working note for
the Stern Review, Final Draft.
Lind, R.C., ed., 1982, Discounting for time and risk in energy policy, The
Johns Hopkins University Press, Baltimore.
Mortensen, J. B., and L. H. Pedersen, 2009, Klimapolitik: kortsigtede
omkostninger og langsigtede gevinster, Samfundsøkonomen, Maj, nr.
2, -18.
Møller, F., 2009, Velfærd nu eller fremtiden. Velfærdsøkonomisk og nyt-
teetisk vurdering over tid. Århus Universitetsforlag, Århus.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
134 CHAPTER 8. CHOICE OF SOCIAL DISCOUNT RATE
Nordhaus, W. D., 2007, A review of the Stern Review on the economics
of climate change, Journal of Economic Literature, vol. XLV, No. 3,
686-702.
Pindyck, R.S., 2012, The climate policy dilemma, NBER WP 18205.
Ploeg, F. van der, and C. Withagen, 2012, Is there really a Green Paradox?
J. Environmental Economics and Management, vol. 64 (3), 342-363
Quiggin, J., 2006, Stern and the critics on discounting, ...
Rezai, A., D. K. Foley, and L. Taylor, 2012, Global warming and economic
externalities, Economic Theory, vol. 49, 329-351.
Roemer, J., 2011, The ethics of intertemporal distribution in a warming
planet, Environmental and Resource Economics, vol. 48, 363-390. By
same author, see also http://pantheon.yale.edu/~jer39/climatechange.html
Samfundsøkonomiske analyser 2012, Finansdepartementet, Norges o¤entlige
utredninger NOU 2012: 16. (http://www.regjeringen.no/nb/dep/…n/
dok/nouer/2012/nou-2012-16.html?id=700821).
Sen, A., 1961, On optimizing the rate of saving, Economic Journal 71,
479-496.
Sinn, H.-W., 2007, Public policies against global warming, CESifo WP no.
2087, August.
Stern, Sir Nicholas (2007), The Economics of Climate Change. The Stern
Review, Cambridge, UK. (A report made for the British government.)
Stern, Sir Nicholas (2013), Interview in the Guardian. http://www.theguardian.com/environ
stern-climate-change-davos
Sterner, T., and U. M. Persson, 2008, An even Stern Review. Introducing
relative prices into the discounting debate, Rev. Econ. and Politics,
vol. 2 (1), 61-76.
Stiglitz, J. E., 2007,
Sydsæter, K., P. Hammond, A. Seierstad, and A. Strøm, 2008, Further
Mathematics for Economic Analysis. Prentice Hall: London.
Weitzman, M. L., 2007, A review of the Stern Review on the economics
of climate change, Journal of Economic Literature, vol. XLV, No. 3,
703-724.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
8.7. References 135
Weitzman, M. L., 2008, Why the far-distant future should be discounted at
its lowest possible rate, J. Environmental Economics and Management,
vol. 36 (3), 201-208.
Weitzman, M. L., 2009, On modeling and interpreting the economics of
catastrophic climate change, Review of Economics and Statistics, vol.
91, 1-19.
Weitzman, M. L., 2011, Gamma discounting, American Economic Review,
vol. 91 (1).
Weitzman, M. L., 2012, Rare disasters, tail-hedged investments, and risk-
adjusted discount rates, NBER Working Paper.
Økonomi og Miljø 2010, De Økonomiske Råds formandskab, København.
Økonomi og Miljø 2013, De Økonomiske Råds formandskab, København.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
136 CHAPTER 8. CHOICE OF SOCIAL DISCOUNT RATE
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
Chapter 9
Human capital, learning
technology, and the Mincer
equation
This chapter is meant as a supplement to Acemoglu, §10.1-2 and §11.2. First
an overview of di¤erent approaches to human capital formation in macroeco-
nomics is given. Next we go into detail with one of these approaches, the
life-cycle approach. In Section 9.3 a simple model of the choice of school-
ing length is considered. Finally, Section 9.4 presents the theory behind the
empirical relationship named the Mincer equation.1 In this connection it is
emphasized that the Mincer equation should be seen as an equilibrium rela-
tionship for relative wages at a given point in time rather than as a production
function for human capital.
9.1 Conceptual issues
We de…ne human capital as the stock of productive skills embodied in an
individual. Human capital is thus a production factor, while human wealth
is the present value of expected future labor income (after tax).
Increases in the stock of human capital occurs through formal education
and on-the-job-training. By contributing to the maintenance of life and well-
being, also health care is of importance for the stock of human capital and
the incentive to invest in human capital.
Since human capital is embodied in individuals and can only be used one
place at a time, it is a rival and excludable good. Human capital is thus
very di¤erent from technical knowledge. We think of technical knowledge as
1
After Mincer (1958, 1974).
137
CHAPTER 9. HUMAN CAPITAL, LEARNING TECHNOLOGY,
138 AND THE MINCER EQUATION
a list of instructions about how di¤erent inputs can be combined to produce
a certain output. A principle of chemical engineering is an example of a piece
of technical knowledge. In contrast to human capital, technical knowledge
is a non-rival and only partially excludable good. Competence in applying
technical knowledge is one of the skills that to a larger or smaller extent is
part of human capital.
9.1.1 Macroeconomic approaches to human capital
In the macroeconomic literature there are di¤erent theoretical approaches to
the modelling of human capital. Broadly speaking we may distinguish these
approaches along two “dimensions”: 1) What characteristics of human capi-
tal are emphasized? 2) What characteristics of the decision maker investing
in human capital are emphasized? Combining these two “dimensions”, we
get Table 1.
Table 1. Macroeconomic approaches to the modelling of human capital.
The character of human capital (hc):
The character of the Is hc treated as essentially di¤erent
decision maker from physical capital?
No Yes
Solow-type rule-of-thumb households Mankiw et al. (1992)
In…nitely-lived family “dynasties” Barro&Sala-i-Martin (2004) Lucas (1988)
(the representative agent approach) Dalgaard&Kreiner (2001)
Finitely-lived individuals going through Ben-Porath (1967)
a life cycle (the life cycle approach) Heijdra&Romp (2009)
My personal opinion is that for most issues the approach in the lower-
right corner of Table 1 is preferable, that is, the approach treating human
capital as a distinct capital good in a life cycle perspective. The viewpoint
is:
First, by being embodied in a person and being lost upon death of this
person, human capital is very di¤erent from physical capital. In addition,
investment in human capital is irreversible (can not be recovered). Human
capital is also distinct in view of the limited extend to which it can be used
as a collateral, at least in non-slave societies. Financing an investment in
physical capital, a house for example, by credit is comparatively easy because
the house can serve as a collateral. A creditor can not gain title to a person,
however. At most a creditor can gain title to a part of that person’s future
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
9.1. Conceptual issues 139
earnings in excess of a certain level required for a “normal” or “minimum”
standard of living.
Second, educational investment is closely related to life expectancy and
the life cycle of human beings: school - work - retirement. So a life cycle per-
spective seems the natural approach. Fortunately, convenient macroeconomic
frameworks incorporating life cycle aspects exist in the form of overlapping
generations models (for example Diamond’s OLG model or Blanchard’s con-
tinuous time OLG model).
9.1.2 Human capital and the e¢ ciency of labor
Generally we tend to think of human capital as a combination of di¤erent
skills. Macroeconomics, however, often tries (justi…ed or not) to boil down
the notion of human capital to a one-dimensional entity. So let us imagine
that the current stock of human capital in society is measured by the one-
dimensional index H: With L denoting the size of the labor force, we de…ne
h H=L, that is, h is the average stock of human capital in the labor
force. Further, let the “quality”(or “e¢ ciency”) of this stock in production
be denoted q (under certain conditions this quality might be proxied by the
average real wage per man-hour). Then it is reasonable to link q and h by
some increasing quality function
q = q(h); where q(0) 0; q 0 > 0: (9.1)
Consider an aggregate production function, F~ ; giving output per time
unit at time t as
@ F~
Y = F~ (K; q(h)L; t); > 0; (9.2)
@t
where K is input of physical capital. The third argument of F~ is time, t;
indicating that the production function is time-dependent due to technical
progress.
Generally the analyst would prefer a measure of human capital such that
the quality of human capital is proportional to the stock of human capital,
allowing us to write q(h) = h by normalizing the factor of proportionality
to be 1. The main reason is that an expedient variable representing human
capital in a model requires that the analyst can decompose the real wage
per working hour multiplicatively into two factors, the real wage per unit of
human capital per working hour and the stock of human capital, h. That is,
an expedient human capital concept requires that we can write
w = w^ h; (9.3)
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
CHAPTER 9. HUMAN CAPITAL, LEARNING TECHNOLOGY,
140 AND THE MINCER EQUATION
where w^ is the real wage per unit of human capital per working hour. Indeed,
if have
Y = F~ (K; hL; t); (9.4)
under perfect competition we can write
@Y
w= = F~2 (K; hL; t)h = w^ h:
@L
Under Harrod-neutral technical progress, (9.4) would take the form
Y = F~ (K; hL; t) = F (K; AhL) F (K; EL); (9.5)
where E A h is the “e¤ective”labor input. The proportionality between
E and h will under perfect competition allow us to write
@Y
w= = F~2 (K; EL; t)E = wE E = wE A h = w^ h:
@L
So with the introduction of the technology level, A, an additional decomposi-
tion, w^ = wE A comes in, while the original decomposition in (9.3) remains
valid.
Whether or not the desired proportionality q(h) = h can be obtained
depends on how we model the formation of the “stu¤” h: Empirically it
turns out that treating the formation of human capital as similar to that of
physical capital does not lead to the desired proportionality.
Treating the formation of human capital as similar to that of phys-
ical capital
Consider a model where human capital is formed in a way similar to physi-
cal capital. The Mankiw-Romer-Weil (1992) extension of the Solow growth
model with human capital is a case in point. Non-consumed aggregate output
is split into one part generating additional physical capital one-to-one, while
the other part is assumed to generate additional human capital one-to-one.
Then for a closed economy in continuous time we can write:
Y = C + IK + IH ;
K_ = IK K K; K > 0;
_
H = IH H H; H > 0; (9.6)
where IK and IH denote gross investment in physical and human capital, re-
spectively. This approach essentially assumes that human capital is produced
by the same technology as consumption and investment goods.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
9.1. Conceptual issues 141
Suppose the huge practical measurement problems concerning IH have
been somehow overcome. Then from long time series for IH an index for Ht
can be constructed by the perpetual inventory method in a way similar to the
way an index for Kt is constructed from long time series for IK . Indeed, in
discrete time, with 0 < H < 1; we get, by backward substitution,
Ht+1 = IH;t + (1 H )Ht = IH;t + (1 H ) [IH;t 1 + (1 H )Ht 1 ]
XT
i T +1
= (1 H ) IH;t i + (1 H) Ht T : (9.7)
i=0
From the time series for IH , an estimate of H , and a rough conjecture about
the initial value, Ht T ; we can calculate Ht+1 : The result will not be very
sensitive to the conjectured value of Ht T since for large T the last term in
(9.7) becomes very small.
In principle there need not be anything wrong with this approach. A
snag arises, however, if, without further notice, the approach is combined
with an explicit or implicit postulate that q(h) is proportional to the “stu¤”,
h; brought into being in the way described by (9.6). The snag is that the
empirical evidence does not support this when the formation of human capital
is modelled as in (9.6). This is what, for instance, Mankiw, Romer, and Weil
(1992) …nd in their cross-country regression analysis based on the approach
in equation (9.6). One of their conclusions is that the following production
function for a country’s GDP is an acceptable approximation:
Y = BK 1=3 H 1=3 L1=3 ; (9.8)
where B stands for the total factor productivity of the country and is gener-
ally growing over time.2 De…ning A = B 3=2 and applying that H = hL; we
can write (9.8) on the form
Y = BK 1=3 (hL)1=3 L1=3 = K 1=3 (h1=2 AL)2=3 :
That is, we end up with the form Y = F (K; q(h)AL) where q(h) = h1=2 ; not
q(h) = h. We should thus not expect the real wage to rise in proportion to
h; when h is considered as some “stu¤” formed in a way similar to the way
physical capital is formed.
2
The way Mankiw-Romer-Weil measure IH is indirect and questionable. In addition,
the way they let their measure enter the regression equation has been criticized for con-
founding the e¤ects of the human capital stock and human capital investment, cf. Gemmel
(1996) and Sianesi and Van Reenen (2003). It will take us too far to go into detail with
these problems here.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
CHAPTER 9. HUMAN CAPITAL, LEARNING TECHNOLOGY,
142 AND THE MINCER EQUATION
Before proceeding, a terminological point is in place. Why do we call q(h)
in (9.2) a “quality” function rather than simply a “productivity” function?
The reason is the following. With perfect competition and CRS, in equilib-
0
rium
h the real wagei per man-hour would be w = @Y =@L = F2 (K; Aq(h)L))Aq(h)
= f (k)~ ~ 0 (k)
kf ~ Aq(h); where k~ K=(Aq(h)L): So, with a converging k; ~
the long-run growth rate of the real wage would in continuous time tend to
be
gw = gA + gq :
In this context we are inclined to identify “labor productivity” with Aq(h)
rather than just q(h) and “growth in labor productivity”with gA + gq rather
than just gq : So a distinct name for q seems appropriate and an often used
name is “quality”.
The conclusion so far is that specifying human capital formation as in
(9.6) does not generally lead to a linear quality function. To obtain the
desired linearity we have to specify the formation of human capital in a way
di¤erent from the equation (9.6). This dissociation with the approach (9.6)
applies, of course, also to its equivalent form on a per capita basis,
H_ IH
h_ = ( n)h = ( H + n)h: (9.9)
H L
(In the derivation of (9.9) we have …rst calculated the growth rate of h
H=L; then inserted (9.6), and …nally multiplied through by h.)
9.2 The life-cycle perspective on human cap-
ital
In the life-cycle approach to human capital formation we perceive h as the
human capital embodied in a single individual and lost upon death of this
individual. We study how h evolves over the lifetime of the individual as a
result of both educational investment (say time spent in school) and work
experience. In this way the life-cycle approach recognizes that human capi-
tal is di¤erent from physical capital. By seeing human capital formation as
the result of individual learning, the life-cycle approach opens up for distin-
guishing between the production technologies for human and physical capital.
Thereby the life-cycle approach o¤ers a better chance for obtaining the linear
relationship, q(h) = h:
Let the human capital of an individual of “age” (beyond childhood) be
denoted h . Let the total time available per time unit for study, work, and
leisure be normalized to 1. Let s denote the fraction of time the individual
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
9.2. The life-cycle perspective on human capital 143
spends in school at age : This allows the individual to go to school only part-
time and spend the remainder of non-leisure time working. If ` denotes the
fraction of time spent at work, we have
0 s +` 1:
The fraction of time used as leisure (or child rearing, say) at age is 1 s ` :
If full retirement occurs at age , we have s = ` = 0 for .
We measure age in the same time units as calendar time. As a slight
generalization of Acemoglu’s equation (10.2),3 where leisure is not considered,
we assume that the increase in h per unit of time (age) generally depends on
four variables: current time in school, current time at work, human capital
already obtained, and current calendar time itself, that is,
dh
h_ = G(s ; ` ; h ; t); h0 0 given. (9.10)
d
The function G can be seen as a production function for human capital
in brief a learning technology. The …rst argument of G re‡ects the role of
formal education. Empirically, the primary input in formal education is the
time spent by the students studying; this time is not used in work or leisure
and it thereby gives rise to an opportunity cost of studying.4 The second
argument of G takes work experience into account and the third argument
allows for the already obtained level of human capital to a¤ect the strength of
the in‡uence from s and ` : Finally, the fourth argument, current calendar
time allows for changes over time in the learning technology (organization of
the learning process).
Consider an individual “born”at date v t (v for vintage). If still alive
at time t; the age of this individual is t v. The obtained stock of human
capital at age will be
Z
h = h0 + G(sx ; `x ; hx ; v + x)dx:
0
A basic supposition in the life-cycle approach is that it is possible to specify
the function G such that a person’s time-t human capital embodies a time-t
labor productivity proportional to this amount of human capital and thereby,
under perfect competition, a real wage proportional to this human capital.
3
Acemoglu, 2009, p. 360.
4
We may perceive the costs associated with teachers’ time and educational buildings
and equipment as being either quantitatively negligible or implicit in the function symbol
G:
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
CHAPTER 9. HUMAN CAPITAL, LEARNING TECHNOLOGY,
144 AND THE MINCER EQUATION
Below we consider four speci…cations of the learning technology that one
may encounter in the literature.
EXAMPLE 1 In a path-breaking model by the Israeli economist Ben-Porath
(1967) the learning technology is speci…ed this way:
h_ = g(s h ) h ; g 0 > 0; g 00 < 0; > 0; h0 > 0:
Here time spent in school is more e¢ cient in building human capital the more
human capital the individual has already. Work experience does not add to
human capital formation. The parameter enters to re‡ect obsolescence (due
to technical change) of skills learnt in school.
EXAMPLE 2 Growiec (2010) and Growiec and Groth (2013) study the
aggregate implications of a learning technology speci…ed this way:
h_ = ( s + ` )h ; > 0; 0; h0 > 0: (9.11)
Here measures the e¢ ciency of schooling and the e¢ ciency of work ex-
perience. The e¤ects of schooling and (if > 0) work experience are here
proportional to the level of human capital already obtained by the individ-
ual (a strong assumption which may be questioned).5 The linear di¤erential
equation (9.11) allows an explicit solution,
R
( sx + `x )dx
h = h0 e 0 ; (9.12)
a formula valid as long as the person is alive. This result has some a¢ nity
with the “Mincer equation”, to be considered below.
EXAMPLE 3 Here we consider an individual with exogenous and constant
leisure. Hence time available for study and work is constant and conveniently
normalized to 1 (as if there were no leisure at all). Moreover, in the beginning
of life beyond childhood the individual goes to school full-time in S time units
(years) and thereafter works full-time until death (no retirement). Thus
1 for 0 < S;
s = (9.13)
0 for S:
We further simplify by ignoring the e¤ect of work experience (or we may say
that work experience just o¤sets obsolescence of skills learnt in school). The
learning technology is speci…ed as
h_ = 1
s ; > 0; h0 0, (9.14)
5
Lucas (1988) builds on the case = 0:
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
9.2. The life-cycle perspective on human capital 145
If < 1, it becomes more di¢ cult to learn more the longer you have already
been to school. If > 1, it becomes easier to learn more the longer you have
already been under education.
The speci…cation (9.13) implies that throughout working life the individ-
ual has constant human capital equal to h0 + S : Indeed, integrating (9.14),
we have for t S and until time of death,
Z Z S
h = h0 + h_ x dx = h0 + x 1
dx = h0 + x jS0 = h0 + S : (9.15)
0 0
So the parameter measures the elasticity of human capital w.r.t. the num-
ber of years in school. As brie‡y commented on in the concluding section,
there is some empirical support for the power function speci…cation in (9.15)
and even the hypothesis = 1 may not be rejected.
In Example 1 there is no explicit solution for the level of human capital.
Then the solution can be characterized by phase diagram analysis (as in
Acemoglu, §10.3). In the examples 2 and 3 we can …nd an explicit solution
for the level of human capital. In this case the term “learning technology”is
used not only in connection with the original di¤erential form as in (9.10), but
also for the integrated form, as in (9.12) and (9.15), respectively. Sometimes
the integrated form, like (9.15), is called a schooling technology.
EXAMPLE 4 Here we still assume the setup in (9.13) of Example 3, includ-
ing the absence of both after-school learning and gradual depreciation. But
the right-hand side of (9.14) is generalized to '( )s ; where '( ) is some
positively valued function of age. Then we end up with human capital after
leaving school equal to some increasing function of S :
h = h(S); where h(0) 0; h0 > 0: (9.16)
In cross-section or time series analysis it may be relevant to extend this by
writing h = ah(S); a > 0; the parameter a could then re‡ect quality of
schooling. In the next section we shall focus on the form (9.16).
Before proceeding, let us brie‡y comment on the problem of aggregation
over the di¤erent members of the labor force at a given point in time. In
the aggregate framework of Section 9.1 multiplicity of skill types and job
types is ignored. Human capital is treated as a one-dimensional and additive
production factor. In production functions like (9.4) only aggregate human
capital, H; matters. So output is thought to be the same whether the in-
put is 2 million workers, each with one unit of human capital, or 1 million
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
CHAPTER 9. HUMAN CAPITAL, LEARNING TECHNOLOGY,
146 AND THE MINCER EQUATION
workers, each with 2 units of human capital. In human capital theory this
questionable assumption is called the perfect substitutability assumption or
the e¢ ciency unit assumption (Sattinger, 1980). If we are willing to impose
this assumption, going from micro to macro at a given point in time is con-
ceptually simple. With h denoting individual human capitalR 1 and f (h) being
the density function at a given point in time (so that 0 f (h)dh = 1); we
…nd average
R1 human capital in the labor force at that point in time to be
h = 0 hf (h)dh and aggregate human capital as H = hL; where L is the
size of the labor force. To build a theory of the evolution over time of the
density function, f (h); is, however, a complicated matter. Within as well as
across the di¤erent cohorts there is heterogeneity regarding both schooling
and retirement. And the fertility and mortality patterns are changing over
time.
If we want to open up for a distinction between di¤erent types of jobs
and di¤erent types of labor, say, skilled and unskilled labor, we may replace
the production function (9.4) with
Y = F~ (K; h1 L1 ; h2 L2 ; t); (9.17)
where L1 and L2 indicate man-hours delivered by the two types of workers,
respectively, and h1 and h2 are the associated human capital levels (measured
in e¢ ciency units for each of the two kinds of jobs), respectively. This could
be the basis for studying skill-biased technical change.
Whether or not the aggregate human capital, H; is a useful concept or
not in connection with production can be seen as a question about whether
or not we can rewrite the production function like (9.17) as Y = F (K; H; t),
where H = h1 L1 + h2 L2 : We can, if the two types of labor are perfectly
substitutable, otherwise not.
9.3 Choosing length of education
9.3.1 Human wealth
Consider an individual. We assume, realistically, that expected lifetime of
this individual is …nite while the age at death is stochastic (uncertain) ex
ante. We further assume that independently of the already obtained age, the
probability of surviving x more time units, say years, is P (X > x) = e mx ;
where X is remaining lifetime, a stochastic variable, while m > 0 is the
mortality rate. This mortality rate is, unrealistically, assumed independent
of age (and also independent of calendar time). The mortality rate indicates
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
9.3. Choosing length of education 147
the approximate probability of dying within one year “from now”.6
Consider our individual’s planning as seen from the time, v; of “birth”
(i.e., entering life beyond childhood). Suppose schooling is a full-time activity
and that this person plans to attend school in the …rst S years of life and
after that work full time until death. Let `t v (S) denote the planned supply
of labor (hours per year) to the labor market at time t by this person. As
`t v (S) depends on the stochastic age at death, T; `t v (S) is itself a stochastic
variable with two possible outcomes:
0 when t v + S or t > v + T;
`t v (S) =
` when v + S < t v + T;
where ` > 0 is an exogenous constant (“full-time”working).
Let wt (S) denote the real wage received per working hour delivered at
time t by a person who after S years in school works ` hours per year until
death. This allows us to write the present value as seen from time v of
expected lifetime earnings, i.e., the human wealth, for a person “born” at
time v as
Z v+T
r(t v)
HW (v; S) = 0 + Ev wt (S)`e dt
Zv+S
1
= 0 + Ev wt (S)`t v (S)e r(t v) dt
v+S
Z 1
= Ev (wt (S)`t v (S)e r(t v) )dt;
v+S
R1
as in this context the integration
P1 operator v+S
( )dt acts like a discrete-time
summation operator, t=v : Here we have introduced the risk-free interest
rate r which is the relevant rate of discount for future labor income condi-
6
If T denotes the uncertain age at death (a stochastic variable) and m is a nonnegative
number, the mortality rate (or “hazard rate” of death) at the age ; denoted m( ); is
de…ned as m( ) = lim !0 1 P (T + j T > ):
In the present model this is assumed equal to a constant, m. The unconditional prob-
ability of not reaching age is then P (T ) = 1 e m F ( ): Hence the density
0 m
function is f ( ) = F ( ) = me and P ( < T + ) me m .R So, for = 0;
1
P (0 < T ) m = m if = 1: Life expectancy is E(T ) = 0 me m d
= 1=m: All this is like in the “perpetual-youth” overlapping generations model by Blan-
chard (1985).
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
CHAPTER 9. HUMAN CAPITAL, LEARNING TECHNOLOGY,
148 AND THE MINCER EQUATION
tional on being alive at the moment concerned. Hence,
Z 1
HW (v; S) = wt (S)e r(t v) (` P (T > t v) + 0 P (T t v))dt
Zv+S
1
= wt (S)e r(t v) `e m(t v) dt
Zv+S
1
= wt (S)`e (r+m)(t v) dt: (9.18)
v+S
In writing the present value of the expected stream of labor income this
way, we have assumed that:
A1 The risk-free interest rate, r; is constant over time.
A2 There is no educational fee.
We now introduce two additional assumptions:
A3 Labor e¢ ciency (human capital) of a person with S years of schooling
is h(S); h0 > 0, so that
wt (S) = w^t h(S);
where w^t is the real wage per unit of human capital per working hour
at time t.7
A4 Owing to Harrod-neutral technical progress at a constant rate g 2
[0; r + m) 0, w^t = w ^0 egt : So technical progress makes a given h
more and more productive (there is direct complementarity between
the technology level and human capital as in (9.5) above).
Given A3 and A4, we get from (9.18) the expected “lifetime earnings”
conditional on a schooling level S :
Z 1
HW (v; S) = w^t h(S)`e (r+m)(t v) dt (9.19)
v+S
Z 1
gv
= w^0 e h(S)` e[g (r+m)](t v) dt
v+S
[g (r+m)](t ) 1
gv e e[g (r+m)]S
= w^0 e h(S)` = w^0 eg h(S)` :
g (r + m) +S r+m g
From now on we chose measurement units such that the “normal” working
time per year is 1 rather than `:
The next question is: how do students make a living while studying?
7
Cf. Example 4 of Section 9.2.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
9.3. Choosing length of education 149
9.3.2 A perfect credit and life annuity market
Assuming the students are born with no …nancial wealth and themselves have
to …nance their costs of living, they have to borrow while studying. Later in
life, when they receive an income, they repay the loans with interest.
In this context we shall introduce the simplifying assumption of a perfect
credit and life annuity market. The …nancial sector will be unwilling to o¤er
the students loans at the going risk-free interest rate, r. Indeed, a creditor
faces the risk that the student dies before having paid o¤ the debt including
the compound interest. Given the described constant mortality rate and
given existence of a perfect credit and life insurance market, it can be shown8
that the equilibrium interest rate on student loans is the “actuarial rate”,
r + m: (This result presupposes that the insurance companies have negligible
administration costs.)
If the individual later in life, after having paid o¤ the debt and obtained a
positive net …nancial position, places the savings on life annuity accounts in
life insurance companies, the actuarial rate, r+m; will also be the equilibrium
rate of return received (until death) on these deposits. At death the liability
of the insurance company is cancelled.
The advantage of saving in life annuities (at least for people without a
bequest motive) is that life annuities imply a transfer of income from after
time of death to before time of death by o¤ering a higher rate of return than
risk-free bonds, but only until the depositor dies. As mentioned, at that time
the total deposit is automatically transferred to the insurance company in
return for the high annuity payouts while the depositor was alive.9
9.3.3 Maximizing human wealth
Suppose that neither the educational process itself nor the resulting stock
of human capital enter the utility function (no “joy of going to school”, no
“joy of being a learned person”). In this perspective human capital is only
an investment good (not also a durable consumption good).10
8
See Lecture notes in macroeconomics, Chapter 12.
9
Whatever name is in practice used for the real world’s private pension arrangements,
including labor market pension arrangements, many of them have such life annuity ingre-
dients.
Owing to asymmetric information and related credit market imperfections, in real world
situations such loan contracts are rare. This is a reason for public sector intervention in
the provision of loans to students. These di¢ culties are ignored by the present model. In
Acemoglu, pp. 761-764, credit market imperfections are considered.
10
For a broader conception of human capital, see for instance Sen (1997).
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
CHAPTER 9. HUMAN CAPITAL, LEARNING TECHNOLOGY,
150 AND THE MINCER EQUATION
If moreover there is no utility from leisure, the educational decision can
be separated from whatever plan for the time path of consumption and sav-
ing through life the individual may decide (cf. the Separation Theorem in
Acemoglu, §10.1). That is, the only incentive for acquiring human capital is
to increase the human wealth HW ( ; S) given in (9.19).
An interior solution to the problem maxS HW (v; S) satis…es the …rst-
order condition:
@HW w^0
(v; S) = h0 (S)e[g (r+m)]S
h(S)e[g (r+m)]S
(r + m g)
@S r+m g
h0 (S)
= HW (v; S) (r + m g) = 0; (9.20)
h(S)
from which follows
h0 (S)
=r+m g r~: (9.21)
h(S)
We call (9.21) the schooling …rst-order condition and r~ the e¤ective dis-
count rate for the schooling decision. In the optimal plan this equals the
e¤ective discount rate appearing on the right-hand side of (9.21), namely the
interest rate adjusted for (a) the approximate probability of dying within
a year from “now”, 1 e m m; and (b) wage growth due to technical
progress. The trade-o¤ faced by the individual is the following: increasing S
by one year results in a higher level of human capital (higher future earning
power) but postpones by one year the time when earning an income begins.
The e¤ective interest cost is diminished by g; re‡ecting the fact that the real
wage per unit of human capital will grow by the rate g from the current year
to the next year.
The intuition behind the …rst-order condition (9.21) is perhaps easier to
grasp if we put g on the left-hand-side and multiply by w^t in the numerator
as well as the denominator. Then the condition looks like a standard no-
arbitrage condition:
w^t h0 (S) + w^t gh(S)
= r + m:
w^t h(S)
On the the left-hand side we have the actual approximate net rate of return
obtained by “investing one more year” in education. In the numerator we
have the direct increase in wage income by increasing S by one unit plus the
gain arising from the fact that human capital, h(S); is worth more in earnings
capacity one year later due to technical progress. In the denominator we
have the educational investment made by letting the obtained human capital,
h(S), “stay”one more year in school instead of at the labor market. Indeed,
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
9.3. Choosing length of education 151
w^t h(S) is the size of that investment in the sense of the opportunity cost of
staying in school one more year.
In an optimal plan the actual net rate of return on the marginal invest-
ment equals the required rate of return, r + m: The required rate of return is
what could be obtained by the alternative strategy, which is to leave school
already after S years and then invest the …rst years’s labor income in life
annuities paying the net rate of return, r + m; per year until death. That is,
the …rst-order condition can be seen as a no-arbitrage equation. (As is usual,
our interpretation treats marginal changes as if they were discrete.)
Suppose S = S > 0 satis…es the …rst-order condition (9.21). To check
the second-order condition, we consider
@ 2 HW
(v; S )
@S 2
@HW h0 (S ) h(S )h00 (S ) h0 (S )2
= (v; S ) (r + m g) + HW (v; S )
@S h(S ) h(S )2
S S
h0 (S )
h00 (S ) h(S )
h0 (S ) 0
= HW (v; S ) h (S ); (9.22)
S h(S )
since the …rst term on the right-hand side in the second row vanishes due to
(9.21) being satis…ed at S = S . The second-order condition, @ 2 HW=@S 2 < 0
at S = S holds if and only if the elasticity of h w.r.t. S exceeds that of
h0 w.r.t. S at S = S : A su¢ cient but not necessary condition for this is
that h00 0: Anyway, since HW (v; S) is a continuous function of S; if there
is a unique S > 0 satisfying (9.21), and if @ 2 HW=@S 2 < 0 holds for this
S ; then this S is the unique optimal length of education for the individual.
If individuals are alike in the sense of having the same innate abilities and
facing the same schooling technology h( ), they will all choose S :
EXAMPLE 5 Suppose h(S) = S , > 0; as in Example 3, but with h0 = 0.
Then the …rst-order condition (9.21) gives a unique solution S = =(r + m
g); and the second-order condition (9.22) holds for all > 0: More sharply
decreasing returns to schooling (smaller ) shortens the optimal time spent
in school as does of course a higher e¤ective discount rate, r + m g:
Consider two countries, one rich (industrialized) and one poor (agricul-
tural). With one year as the time unit, let the parameter values be as in the
…rst four columns in the table below. The resulting optimal S for each of the
countries is given in the last column.
r m g S
rich country 0:6 0.06 0.01 0.02 12.0
poor country 0:6 0.12 0.02 0.00 4.3
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
CHAPTER 9. HUMAN CAPITAL, LEARNING TECHNOLOGY,
152 AND THE MINCER EQUATION
Figure 9.1: The semi-log schooling-wage relationship for …xed t. Di¤erent coun-
tries. Source: Krueger and Lindahl (2001).
The di¤erence in S is due to r and m being higher and g lower in the poor
country.
The above example follows a short note by Jones (2007) entitled “A sim-
ple Mincerian approach to endogenizing schooling”. The term “Mincerian
approach” should here be interpreted in a broad sense as more or less syn-
onymous with “life-cycle approach”.
Often in the macroeconomic literature, however, the term “Mincerian
approach”is identi…ed with an exponential speci…cation of the learning tech-
nology:
h(S) = h(0)e S ; > 0: (9.23)
This exponential form can at the formal level be seen as resulting from a
combination of equation (9.11) from Example 2 and equation (9.13) from
Example 3.11 The sole basis for an exponential relationship is empirical
11
One should be aware, however, that the present simple framework does not really em-
brace an exponential speci…cation of h. Indeed, the second-order condition (9.22) implied
by the “perpetual youth” assumption of age-independent mortality and no retirement, is
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
9.4. Explaining the Mincer equation 153
cross-sectional evidence on relative wages at a given point in time, cf. Figure
??. This is not the same as providing empirical support for an exponential
production function for human capital. As brie‡y commented in the con-
cluding section, there seems to be little empirical support for an exponential
production function. And in fact, as we shall now see, Mincer’s microeco-
nomic explanation of the exponential relationship (cf. Mincer, 1958, 1974)
has nothing to do with a speci…c production function for human capital.
9.4 Explaining the Mincer equation
In Mincer’s theory behind the observed exponential relationship called the
Mincer equation, there is no role at all for any speci…c schooling technology,
h( ); leading to a unique solution, S : The essential point is that the empir-
ical Mincer equation is based on heterogeneity in the jobs o¤ered to people
(di¤erent educational levels not being perfectly substitutable). An exponen-
tial relationship where people, in spite of being alike ex ante, choose di¤erent
educational levels ex post can then arise through the equilibrium forces of
supply and demand in the job markets.
Imagine, …rst, a case where all individuals have in fact chosen the same
educational level, S ; because they are ex ante alike and all face the same
arbitrary human capital production function, h(S); satisfying (9.22). Then
jobs that require other educational levels will go un…lled and so the job mar-
kets will not clear. The forces of excess demand and excess supply will then
tend to generate an educational wage pro…le di¤erent from the one presumed
in (9.19), that is, di¤erent from w^t h(S). Sooner or later an equilibrium edu-
cational wage pro…le tends to arise such that people are indi¤erent as to how
much schooling they choose, thereby allowing market clearing. This requires
a wage pro…le, wt (S); such that a marginal condition analogue to (9.21) holds
for all S for which there is a positive amount of labor traded in equilibrium,
say all S 2 0; S :
dwt (S)=dS
=r+m g r~ for all S 2 0; S : (9.24)
wt (S)
It is here assumed, in the spirit of assumption A4 above, that technical
progress implies that wt (S) for …xed S grows at the rate g; i.e., wt (S) =
incompatible with the strong convexity implied by the exponential function. Of course,
this must be seen as a limitation of the “perpetual youth” setup (where there is no con-
clusive upper bound for anyone’s lifetime) rather than a reason for rejecting apriori the
exponential speci…cation (9.23).
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
CHAPTER 9. HUMAN CAPITAL, LEARNING TECHNOLOGY,
154 AND THE MINCER EQUATION
w0 (S)egt ; for all S 2 0; S : The equation (9.24) is a linear di¤erential equa-
tion for wt w.r.t. S; de…ned in the interval 0 S S: And the function
wt (S); where t is …xed, is then unknown solution to this di¤erential equation.
That is, we have a di¤erential equation of the form dx(S)=dS = r~x(S): This
is a di¤erential equation where the unknown function x(S) is a function of
schooling length rather than calendar time. The solution is x(S) = x(0)er~S :
Replacing the function x( ) with the function wt ( ); we thus have the solution
wt (S) = wt (0)er~S : (9.25)
Note that in the previous section, in the context of (9.21), we required
the proportionate marginal return to schooling to equal r~ only for a speci…c
S; i.e.,
d(w^t h(S))=dS h0 (S)
= =r+m g r~ for S = S : (9.26)
w^t h(S) h(S)
This is only a …rst-order condition assumed to hold at some point, S : It will
generally not be a di¤erential equation the solution of which gives a Min-
cerian exponential relationship. A di¤erential equation requires a derivative
relationship to hold not only at one point, but in an interval for the indepen-
dent variable (S in (9.24)). Indeed, in (9.24) we require the proportionate
marginal return to schooling to equal r~ in a whole interval of schooling lev-
els. Otherwise, with heterogeneity in the jobs o¤ered there could not be
equilibrium.12
Returning to (9.25), by taking logs on both sides, we get
log wt (S) = log wt (0) + r~S; (9.27)
which is the Mincer equation on log-linear form.
Empirically, the Mincer equation does surprisingly well, cf. Figure ??.13
Note that (9.25) also yields a theory of how the “Mincerian slope”, ; in
(9.23) is determined, namely as the mortality- and growth-corrected real
interest rate, r~. The evidence for this part of the theory is more scarce.
Given the equilibrium educational wage pro…le, wt (S); the human wealth
12
As I see it, Acemoglu (2009, p. 362) makes the logical error of identifying a …rst-order
condition, (9.26), with a di¤erential equation, (9.24).
13
The slopes are in the interval (0:05; 0:15).
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
9.5. Some empirics 155
of an individual “born”at time 0 can be written
Z 1 Z 1
r~S (r+m)t r~S
HW0 = wt (0)e e dt = e w0 (0)egt e (r+m)t dt
S S
Z 1 1
e[g (r+m)]t
= w0 (0)er~S e[g (r+m)]t dt = w0 (0)er~S
S g (r + m) S
w0 (0)
= ; (9.28)
r+m g
since r~ r+m g: In view of the adjustment of the S-dependent wage levels,
in equilibrium the human wealth of the individual is thus independent of S
(within an interval) according to the Mincerian theory. Indeed, the essence
of Mincer’s theory is that if one level of schooling implies a higher human
wealth than the other levels of schooling, the number of individuals choosing
that level of schooling will rise until the associated wage has been brought
down so as to be in line with the human wealth associated with the other
levels of schooling. Of course, such adjustment processes must in practice be
quite time consuming and can only be approximative.14
In this context, the original schooling technology, h( ); for human capital
formation has lost any importance. It does not enter human wealth in a
long-run equilibrium in the disaggregate model where human wealth is simply
given by (9.28). In this equilibrium people have di¤erent S’s and the received
wage of an individual per unit of work has no relationship with the human
capital production function, h( ); by which we started in this section.
Although there thus exists a microeconomic theory behind a Mincerian
relationship, this theory gives us a relationship for relative wages in a cross-
section at a given point in time. It leaves open what an intertemporal pro-
duction function for human capital, relating educational investment, S; to a
resulting level, h, of labor e¢ ciency in a macroeconomic setting, looks like.
Besides, the Mincerian slope, r~; is a market price, not an aspect of schooling
technology.
9.5 Some empirics
In their cross-country regression analysis de la Fuente and Domenech (2006)
…nd a relationship essentially like that in Example 3 with = 1.15
14
Who among the ex ante similar individuals ends up with what schooling level is inde-
terminate in this setup.
15
The authors …nd that the elasticity of GDP w.r.t. average years in school in the labor
force is at least 0.60. The empirical macroeconomic literature typically measures S as the
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
CHAPTER 9. HUMAN CAPITAL, LEARNING TECHNOLOGY,
156 AND THE MINCER EQUATION
Similarly, the cross-country study, based on calibration, by Bills and
Klenow (2000) as well as the time series study by Cervelatti and Sunde
(2010) favor the hypothesis of diminishing returns to schooling. According
to this, the linear term, r~S; in the exponent in (9.23) should be replaced by
a strictly concave function of S: These …ndings are in accordance with the
results by Psacharopoulus (1994).
For S > 0; the power function in Example 5 can be written h = S = e ln S
and is thus in better harmony with the data than the exponential function
(9.23). A parameter indicating the quality of schooling may be added: h =
ae ln S ; where a > 0 may be a function of the teacher-pupil ratio, teaching
materials per student etc. See Caselli (2005).
Outlook
Models based on the life-cycle approach to human capital typically conclude
that education is productivity enhancing, i.e., education has a level e¤ect on
income per capita but is not a factor which in itself can explain sustained per
capita growth, cf. Exercise V.7 and V.8. A more plausible main driving factor
behind growth seems rather to be technological innovations. A higher level
of per capita human capital may temporarily raise the speed of innovations,
however.
Final remark
Our formulation of the schooling length decision problem in Section 9.3 con-
tained several simplications so that we ended up with a static maximization
problem in Section 9.3.3. More general setups lead to truly dynamic human
capital accumulation problems.
This chapter considered human capital as a productivity-enhancing fac-
tor. There is a complementary perspective on human capital, namely the
Nelson-Phelps hypothesis about the key role of human capital for technology
adoption and technological catching up, see Acemoglu, §10.8, and Exercise
Problem V.3.
average number of years of schooling in the working-age population, taken for instance
from the Barro and Lee (2001) dataset. This means that complicated aggregation issues,
arising from cohort heterogeneity and from the fact that individual human capital is lost
upon death, are bypassed. For discussion, see Growiec and Groth (2013).
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
9.6. Literature 157
9.6 Literature
Barro, R.J., and J. Lee, 2001, International data on educational attainment,
Oxford Economic Papers, vol. 53 (3), 541-563.
Barro, R.J., and X. Sala-i-Martin, 2004, Economic Growth, 2nd edition,
MIT Press.
Ben-Porath, Y., 1967, The production of human capital and the life cycle
of earnings, Journal of Political Economy 75 (4), 352-365.
Bils, M., and P. J. Klenow, 2000. Does schooling cause growth? American
Economic Review, 90 (5), 1160-1183.
Blanchard, O., 1985, J. Political Economy,
Caselli, F., 2005, Accounting for cross-country income di¤erences. In: Hand-
book of Economic Growth, vol. IA.
Cervellati, M., and U. Sunde, 2010, Longevity and lifetime labor supply:
Evidence and implications revisited, WP.
Cohen, D., and M. Soto, 2007, Growth and human capital: good data, good
results, Journal of Economic Growth 12, 51-76.
Cunha, F., J.J. Heckman, L. Lochner, and D.V. Masterov, 2006, Interpret-
ing the evidence on life cycle skill formation, Handbook of the Eco-
nomics of Education, vol. 1, Amsterdam: Elsevier.
Dalgaard, C.-J., and C.-T. Kreiner, 2001, Is declining productivity in-
evitable? J. Econ. Growth, vol. 6 (3), 187-203.
de la Fuente, A., and R. Domenech, 2006, Human capital in growth re-
gressions: How much di¤erence does quality data make, Journal of the
European Economic Association 4 (1), 1-36.
Gemmell, N., 1996, , Oxford Bulletin of Economics and Statistics 58,
9-28.
Growiec, J., 2010, Human capital, aggregation, and growth, Macroeconomic
Dynamics 14, 189-211.
Growiec, J., and C. Groth, 2013, On aggregating human capital across het-
erogeneous cohorts, Working Paper, http://www.econ.ku.dk/okocg/Forside/Publications.htm
Hall, R., and C.I. Jones, 1999, , Quarterly Journal of Economics.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
CHAPTER 9. HUMAN CAPITAL, LEARNING TECHNOLOGY,
158 AND THE MINCER EQUATION
Hanushek, E.A., and L. Woessmann, 2012, Do better schools lead to more
gorwth? Cognitive skills, economic outcomes, and causation, J. Econ.
Growth, vol. 17, 267-321.
Hazan, M., 2009, Longevity and lifetime labor supply: Evidence and impli-
cations, Econometrica 77 (6), 1829-1863.
Heckman, J.J., L.J. Lochner, and P.E. Todd, 2003, Fifty years of Mincer
earnings regressions, NBER WP 9732.
Heijdra, B.J., and W.E. Romp, 2009, Journal of Economic
Dynamics and Control 33, 725-744.
Hendry, D., and H. Krolzig, 2004, We ran one regression, Oxford Bulletin
of Economics and Statistics 66 (5), 799-810.
Jones, B.,
Jones, C. I., 2005, Handbook of Economic Growth,
Jones, C. I., 2007, A simple Mincerian approach to endogenizing schooling,
WP.
Krueger, A. B., and M. Lindahl, 2001. Education for growth: Why and for
whom? Journal of Economic Literature, 39, 1101-1136.
Lucas, R.E., 1988, On the mechanics of economic development, Journal of
Monetary Economics 22, 3-42.
Lucas, R.E., 1993, Making a miracle, Econometrica 61, 251-272.
Mankiw, G., 1995,
Mankiw, G., D. Romer, and D. Weil, 1992, QJE.
Miles, D., 1999, Modelling the impact of demographic change upon the
economy, Economic Journal 109, 1-36.
Mincer, J., 1958,
Mincer, J., 1974, Schooling, Experience, and Earnings, New York: NBER
Press,
Ortigueira, S., 2003, Equipment prices, human capital and economic growth,
Journal of Economic Dynamics and Control 28, 307-329.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
9.6. Literature 159
Psacharopoulus, 1994,
Rosen, S., 1976. A Theory of Life Earnings, JPE 84 (4), S45-S67.
Rosen, S., 2008. Human capital. In: The New Palgrave Dictionary of
Economics, 2nd ed., ed. by S. N. Durlauf and L. E. Blume, available
at
http://www.econ.ku.dk/english/libraries/links/
Rosenzweig, M.R., 2010, Microeconomic approaches to development: School-
ing, learning, and growth, Economic Growth Center Discussion Paper
No. 985, Yale University.
Sattinger, M., 1980, Capital and the Distribution of Labor Earnings, North-
Holland: Amsterdam.
Sheshinski, E., 2007, The Economic Theory of Annuities, Princeton: Prince-
ton University Press.
Sheshinski, E., 2009, Uncertain longevity and investment in education, WP,
The Hebrew University of Jerusalem, August.
Sianesi, B., and J. Van Reenen, 2003, The returns to education: Macroeco-
nomics, Journal of Economic Surveys 17 (2), 157-200.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
CHAPTER 9. HUMAN CAPITAL, LEARNING TECHNOLOGY,
160 AND THE MINCER EQUATION
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
Chapter 10
Knowledge creation and human
capital in a growing economy
As a follow-up on the concept of a schooling technology presented in Chapter
9, Section 9.2, the present chapter considers aspects of the interplay between
physical capital, human capital, and knowledge creation in a simple balanced
growth framework. The aim is to get the structure of an economy with a
central role for both R&D and human capital “right”. We focus in this
chapter on technically feasible time paths, that is, time paths that a feasible
from the point of view of technology and initial resources. More precisely, we
focus on technically feasible paths that are consistent with balanced growth.
Institutions and incentives that may be needed for the economy to realize
such a path is not the issue here.
10.1 The model
We consider a closed economy with education and two production sectors,
manufacturing and R&D. Time is continuous. Postponing the modeling of
education a little, at the aggregate level we have:
Yt = Tt Kt (ht LY t )1 ; 0 < < 1; (10.1)
K_ t = Yt ct Nt Kt ; 0; K0 > 0 given, (10.2)
Tt = At ; > 0; (10.3)
_ '
At = At ht LAt ; > 0; ' < 1; A0 > 0 given, (10.4)
0 < LY t LY t + LAt = Lt , (10.5)
where Yt is manufacturing output (the value of which is less than GN P when
LAt > 0), Tt is total factor productivity (TFP), Kt is physical capital input,
161
CHAPTER 10. KNOWLEDGE CREATION AND HUMAN CAPITAL
162 IN A GROWING ECONOMY
ht is average human capital in the labor force, LY t and LAt are inputs of
labor in manufacturing and R&D, respectively, Ct is aggregate consumption,
At is the stock of technical knowledge, and Lt is aggregate labor input, all at
time t. The size of population is denoted Nt and so per capita consumption
is ct Ct =Nt :
Comments: As to (10.5), ht LAt is the total input of human capital per
time unit in R&D and A't is the productivity of this input at the aggregate
level. The parameter ' measures the elasticity of research productivity w.r.t.
the level of the available stock of technical knowledge. The case 0 < ' < 1
represents the “standing on the shoulders” case where knowledge creation
becomes easier the more knowledge there is already. In contrast, the case
' < 0 represents the “…shing out” case, also called the “easiest inventions
are made …rst” case. This would re‡ect that it becomes more and more
di¢ cult to create the next advance in technical knowledge. Note also that
the productivity of man-hours (LAt ) in R&D depends on the level of human
capital, ht : As to (10.5), the strict and weak inequalities are motivated by the
view that for the system to be economically viable, there must be activity in
the Y -sector whereas it is of interest to allow for and compare the cases
LAt > 0 and LAt = 0 (active versus passive R&D sector).
The population growth rate is assumed constant:
Nt = N0 ent ; n 0; N0 > 0 given. (10.6)
We assume a stationary age distribution in the population. Although details
about schooling are postponed, we already here assume that schooling and
retirement are consistent with the labor force being a constant fraction of
the population:
Lt = (1 )Nt ; (10.7)
where 2 (0; 1). Then, by (10.6) follows
Lt = L0 ent ; n 0; L0 > 0. (10.8)
We let the growth rate at time t of a variable x > 0 be denoted gxt . When
writing just gx , without the time index t, it is understood that the growth
rate of x is constant over time.
10.2 Productivity growth along a BGP with
R&D
Let us …rst …nd an expression for the TFP growth rate. By log-di¤erentiation
w.r.t. t in (10.1), we have
gY t = gT t + gKt + (1 )(ght + gLY t ): (10.9)
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
10.2. Productivity growth along a BGP with R&D 163
The current TFP growth rate is thus
gT F P t gY t ( gKt + (1 )(ght + gLY t )) = gT t = gAt ; (10.10)
where the last equality follows from (10.3). By (10.4), we get
A_ t
gAt = A't 1 ht LAt = 0, with > if and only if LAt > 0: (10.11)
At
We shall …rst consider the case of active R&D:
ASSUMPTION (A1): LAt > 0 for all t 0:
This assumption implies gAt > 0 and so the growth rate of gAt is well-de…ned.
By log-di¤erentiation w.r.t. t in (10.11) we have
g_ At
= (' 1)gAt + ght + gLA t : (10.12)
gAt
10.2.1 Balanced growth with R&D
In the present context we de…ne a balanced growth path (BGP) as a path
along which gY t ; gCt ; gKt ; gAt ; and ght are constant (not necessarily equal
and not necessarily positive). With y denoting per capita manufacturing
output, i.e., y Y =L; let us …nd the growth rate of y in balanced growth
with active R&D. We introduce the following additional assumptions:
ASSUMPTION (A2): The economy follows a BGP.
ASSUMPTION (A3): Yt ct Nt > 0 for all t 0:
By imposing (A3), we rule out the degenerate case where gK = :
Along a BGP, by de…nition, gAt is a constant, gA . Since thereby g_ A = 0;
solving for gA in (10.12) gives
ght + gLA t
gA = > 0; (10.13)
1 '
where the positivity is due to the assumption (A1). For the formula (10.13)
to be consistent with balanced growth, gLA t must be a constant, gLA ; since
otherwise gA and ght could not both be constant as they must in balanced
growth, by de…nition. Moreover, we must have gLA = n: To see this, imagine
that gLA < n. Then, in order for the growth rate of the sum LY t + LAt
to accord with (10.8), we would need gLY t > n forever, which would imply
LY t + LAt > Lt sooner or later. This is a contradiction. And if instead we
imagine that gLA > n while still being constant, we would, at least after
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
CHAPTER 10. KNOWLEDGE CREATION AND HUMAN CAPITAL
164 IN A GROWING ECONOMY
some time, have LAt > Lt ; again a contradiction. We conclude that gLA = n:
For LY t + LAt to accord with (10.8), it then follows that also gLY t must be
a constant, gLY ; and equal to n: We have hereby proved that along a BGP
with R&D,
gLA = gLY = n: (10.14)
It follows that LA =L is constant along a BGP with R&D.
Given the accumulation equation (10.2) and the assumption (A3), it fol-
lows by the Balanced Growth Equivalence Theorem of Chapter 4 that
gC = gY = gK
along a BGP. From (10.9), together with (10.7) and the de…nition c C=N ,
then follows that along a BGP,
gc = gy = gY n = gT + gK + (1 )(gh + n) n
= gT + (gK n) + (1 )gh = gT + gk + (1 )gh ; (10.15)
where the last equality comes from k K=LY and gLY = n: As gK = gY and
gLY = gL ; we have gk = gy : Then (10.15) gives
gT gA
gy = + gh = + gh ; (10.16)
1 1
in view of (10.10).
Education
Let the time unit be one year. Suppose an individual “born”at time v (v for
“vintage”) spends the …rst S years of life in school and then enters the labor
market with a human capital equal to h(S); where h0 > 0. We ignore the
role of teachers and schooling equipment in the formation of human capital.
The role of work experience for human capital later in life is likewise ignored.
Moreover, we assume that S is the same for all members of a given cohort
and also until further notice the same across cohorts. So
h = h(S); h0 > 0: (10.17)
After leaving school, individuals work full-time until either death before
age R or retirement at age R where R > S; of course; life expectancy is
assumed the same for all cohorts. Assuming a stationary age distribution in
the population, we see that in (10.7) represents the constant fraction of
the population consisting of people either below age S, i.e., under education,
or above age R; i.e., retired people ( will be an increasing function of S and
a decreasing function of R):1
1
A complete model would treat S as endogenous in general equilibrium. In a partial
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
10.2. Productivity growth along a BGP with R&D 165
Sustained productivity growth along a BGP
It follows that average human capital is constant. Thus gh = 0 and (10.16)
reduces to
gA
gy = > 0; (10.18)
1
In equation (10.15) productivity growth, gy ; is decomposed into a con-
tribution from technical change, a contribution from “capital deepening”
(growth in k), and a contribution from human capital growth if any. As long
as S in (10.17) is assumed constant over time, there is no human capital
growth. So we can re-write (10.15):
gy = gT + gk = gT F P + gk ; (10.19)
in view of gT F P = gT = gA from (10.10). This equation decomposes the
productivity growth rate into a direct contribution from technical change and
a direct contribution from capital deepening. Digging deeper, (10.18) tells
us that both these direct contributions rest on sustained knowledge growth.
The correct interpretation of (10.19) is that it just displays the two factors
behind the current increase in y; while (10.18) takes into account that both
TFP growth and capital deepening are in a long-run perspective themselves
driven by knowledge growth.
10.2.2 A precondition for sustained productivity growth
when gh = 0: population growth
We saw that along a BGP with R&D, gLA = n: By (10.13) and gh = 0 then
follows that along a BGP with R&D,
n
gA = > 0: (10.20)
1 '
From this inequality we see that existence of a BGP with R&D requires
ASSUMPTION (A4): n > 0
to hold.
equilibrium analysis one could possibly use an approach similar to the one in Chapter 9,
Section 9.3. We shall not enter into that, however, because the next step, determination
of the real rate of interest in general equilibrium, is a complex problem and requires a
lot of additional speci…cations of households’characteristica and market structure. Fortu-
nately, it is not necessary to determine S as long as the focus is only on determining the
productivity growth rate along a BGP.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
CHAPTER 10. KNOWLEDGE CREATION AND HUMAN CAPITAL
166 IN A GROWING ECONOMY
On the basis of (A4) and (10.18) we …nally conclude that
n
gy = > 0: (10.21)
(1 ')(1 )
Here we have taken into account that also knowledge growth is endogenous
in that it is determined by allocation of resources (research workers) to R&D
activity. The result (10.21) tells us that not only is population growth neces-
sary for sustained productivity growth but productivity growth is faster the
faster is population growth.
Why does population growth ultimately help productivity growth (at
least in this model)? The explanation is that productivity growth is driven
by knowledge creation. Knowledge is a nonrival good its use by one agent
does not, in itself, limit its simultaneous use by other agents. The value of
a piece of technical knowledge a technical idea is proportional to the
number of users. Considering the producible T in (10.1) as an additional
production factor along with capital and labor, (10.1) displays increasing
returns to scale in manufacturing w.r.t. these three production factors. Al-
though there are diminishing marginal returns to capital, there are increasing
returns to scale w.r.t. capital, labor, and the accumulative technology level.
For the increasing returns to unfold in the long run, growth in the labor force
(hence in population) is needed. Growth in the labor force and T not only
counterbalances the falling marginal productivity of capital,2 but actually
upholds sustained per capita growth the more so the faster is population
growth.
The growth-promoting role of the exogenous rate of population growth
re‡ects the presence of what is called a weak scale e¤ect in the model. A scale
e¤ect is said to be present in an economic system if there is an advantage of
scale measured by population size. This advantage of scale is in the present
case due to the productivity-enhancing role of a nonrival good, technical
knowledge, that is produced by the research workers in the idea-creating
R&D sector. Thereby higher population growth results in higher per capita
growth in the long run. On the other hand, a large population is not in itself,
when ' < 1; su¢ cient to generate sustained positive per capita growth. This
is why we talk of a weak scale e¤ect. In contrast, what is known as a strong
scale e¤ect (associated with the case ' 1) is present if a larger population
as such (without population growth) would be enough to generate higher per
capita growth in the long run.
In view of cross-border di¤usion of ideas and technology, the result (10.21)
should not be seen as a prediction about individual countries. It should
2
This counter-balancing role re‡ects the direct complementarity between the produc-
tion factors in (10.1).
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
10.2. Productivity growth along a BGP with R&D 167
rather be seen as pertaining to larger regions, nowadays probably the total
industrialized part of the world. So the single country is not the relevant
unit of observation and cross-country regression analysis thereby not the
right framework for testing such a link from n to gy :
The reason that in (10.21), a higher promotes productivity growth
is that indicates the sensitivity (elasticity) of TFP w.r.t. accumulative
knowledge. Indeed, the larger is ; the larger is the percentage increase
in manufacturing output that results from a one-percentage increase in the
stock of knowledge.
The intuition behind the growth-enhancing role of in (10.21) follows
from (10.1) which indicates that measures the elasticity of manufacturing
output w.r.t. another accumulative input, physical capital. The larger is ,
the larger is the percentage increase in manufacturing output resulting from
a one-percentage increase in the stock of capital.
Finally, the intuition behind the growth-enhancing role of ' in (10.21)
can be obtained from the equation (10.4) which describes the creation of
new knowledge. The equation shows that the larger is ', the larger is the
percentage increase in the time-derivative of technical knowledge resulting
from a one-percentage increase in the stock of knowledge.
10.2.3 The concept of endogenous growth
The above analysis provides an example of endogenous growth in the sense
that the positive sustained per capita growth rate is generated through an
economic mechanism within the model, allocation of resources to R&D; by
an “economic mechanism” is meant a process involving economic decisions,
either directly or indirectly. This is in contrast to the Solow or standard
Ramsey model where technical progress is exogenous, given as manna from
heaven.
There are basically two types of endogenous growth. One is called semi-
endogenous growth and is present if growth is endogenous but a positive
per capita growth rate can not be sustained in the long run without the
support from growth in some exogenous factor (for example growth in the
labor force). As n > 0 is needed for sustained per capita growth in the above
model, growth is here driven by R&D in a semi-endogenous way.
The other type of endogenous growth is called fully endogenous growth
and occurs if the long-run growth rate of Y =L is positive without the support
from growth in any exogenous factor (for example growth in the labor force).
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
CHAPTER 10. KNOWLEDGE CREATION AND HUMAN CAPITAL
168 IN A GROWING ECONOMY
10.3 Permanent level e¤ects
In the result (10.21), there is no trace of the size of the fraction, LA =L; of
the labor force allocated to R&D. This is due to the assumption that ' < 1.
This assumption implies diminishing marginal productivity of knowledge in
the creation of new knowledge. Indeed, when ' < 1; @ A=@A _ = 'A' 1 hLA
is a decreasing function of the stock of knowledge already obtained. A shift
of LA =L to a higher level can temporarily generate faster knowledge growth
and thereby faster productivity growth, but due to the diminishing marginal
productivity of knowledge in the creation of new knowledge, in the long run
gA and gy will be back at their balanced-growth level given in (10.20) and
(10.21), respectively.
It can be shown, however, that a marginally higher LA =L generally has
a permanent level e¤ect, that is, a permanent e¤ect on y along a BGP. If
initially LA =L is “small”, this level e¤ect tends to be positive. This is like
in the Solow growth model where a shift to a higher saving-income ratio, s;
has a temporary positive growth e¤ect and a permanent positive level e¤ect
on y. In contrast to the Solow model, however, if LA =L is already “large”,
the level e¤ect on y of a marginal increase in LA =L may be negative. This is
because Y is produced by LY = (1 LA =L)L; not L:3
Human capital a¤ects the productivity of man-hours in both manufac-
turing and R&D. As mentioned we treat the number of years in school and
average human capital, h, as exogenous. Then it is straightforward to study
the comparative-dynamic e¤ect of a higher level of average human capital,
h: In the present model there will be a permanent level e¤ect on y but no
permanent growth e¤ect.
Yet, a complicating aspect is that, given the model, a higher value of h
will cost a higher number of years in school, i.e., a higher S: A higher S
implies that a smaller fraction of the population will be in the labor force, cf.
(10.7) where is an increasing function of S. This implies that there is no
longer a one-to-one relationship between a positive level e¤ect on y Y =L
and a positive level e¤ect on per capita consumption, c C=N = (C=Y )
(Y =L) (L=N ): We will not go into detail with this kind of trade-o¤ here.
10.4 The case of no R&D
As an alternative to (A1) we now consider the case of no R&D:
ASSUMPTION (A5): LAt = 0 for all t 0:
3
In Exercise VII.7 you are asked to analyze this kind of problems in a more precise
way.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
10.5. A look ahead 169
Under this assumption the whole labor force is employed in manufac-
turing, i.e., LY t = Lt for all t 0: There is no growth in knowledge and
therefore no TFP growth. Whether n > 0 or n = 0; along a BGP satisfy-
ing (A3), (10.19) is still valid but reduces to gy = gk : At the same time,
however, the Balanced Growth Equivalence Theorem of Chapter 4 says that
along a BGP satisfying (A3), gY = gK ; which implies gy = gk : As 2 0; 1);
we have thus reached a contradiction unless gy = gk = 0:
So, as expected, without technological progress there can not exist sus-
tained per capita growth. To put it di¤erently, along a BGP we necessarily
have gY = gK = gC = n; where C cN:
10.5 A look ahead
Given the prospect of non-increasing population in the world economy al-
ready within a century from now (United Nations, 2013), the prospect of
sustained per capita growth in the world economy in the very long run may
seem bleak according to the model. Let us take a closer look at the issue.
10.5.1 The case n = 0
Suppose n = 0 in the above model and return to the assumption (A1). As
gLA can no longer be a positive constant, gA and gy can no longer be positive
constants. Hence balanced growth with gy > 0 is impossible. Does this imply
that there need be economic stagnation in the sense of gy = 0? No, what is
ruled out is that yt = y0 egy t is impossible for any constant gy > 0: So it is
exponential growth that is impossible.
Still paths along which yt ! 1 and ct ! 1 for t ! 1 are techni-
cally feasible. Along such paths, gy and gc will be positive forever, but with
limt!1 gy = 0 and limt!1 gc = 0: To see this, suppose LAt = LA ; a positive
constant less than L; where L is the constant labor force which is propor-
tional to the constant population. Suppose further, for simplicity, that h is
can be considered exogenous. Then, from (10.4) and (10.17) follows
A_ t = A't hLA A't ; hLA :
This Bernoulli di¤erential equation has the solution4
1 1
At = A10 '
+ (1 ') t 1 '
A10 '
+ (1 ') hLA t 1 '
! 1 for t ! 1:
(10.22)
4
See Section 7.2 of Chapter 7.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
CHAPTER 10. KNOWLEDGE CREATION AND HUMAN CAPITAL
170 IN A GROWING ECONOMY
The stock of knowledge thus follows what is known as a quasi-arithmetic
growth path a form of less-than-exponential growth. The special case
' = 0 leads to simple arithmetic growth: At = A0 + (1 ') hLA t: In
case 0 < ' < 1; At features more-than-arithmetic growth and in case ' < 0;
At features less-than-arithmetic growth. It can be shown that with a social
welfare function of the standard Ramsey type, cf. Chapter 8, the social
planner’s solution converges, for t ! 1, toward a path where also Kt ; Yt ; yt ;
and ct feature quasi-arithmetic growth.5
10.5.2 The case of rising life expectancy
There is another demographic aspect of potential importance for future pro-
ductivity growth, namely the prospect of increasing schooling length in the
wake of an increasing life expectancy.
Over the past 30-40 years average years of schooling have tended to grow
arithmetically at a rate of about 0:8 years per decade in the EU as a whole,
compared to 0.7 years in the US (Montanino et al. 2004). A central fac-
tor behind this development is the rising life expectancy due to improved
income, salubrity, nutrition, sanitation, and medicine. Increased life ex-
pectancy heightens the returns to education. In the …rst half of the twentieth
century life expectancy in the US improved at a rate of four years per decade.
In the second half the rate has been smaller, but still close to two years per
decade (Arias, 2004). Oeppen and Vaupel (2002) report that since 1840 fe-
male life expectancy in the record-holding country in the world has steadily
increased by almost a quarter of a year per year. To what extent such devel-
opments may continue is not clear. But at least for a long time to come we
may expect growth in life expectancy and thereby also in educational invest-
ment because of the lengthening of the recovery period for that investment.
Increasing schooling length introduces heterogeneity w.r.t. individual hu-
man capital into the model. In a cross-section of workers at a given point in
time the workers’h becomes a decreasing function of age. And increasing life
expectancy changes the aggregate growth process for population and labor
force. This takes us somewhat outside the above analytical framework with a
stationary age structure and no schooling heterogeneity. Yet let us speculate
a little.
Suppose the schooling technology can be presented by a power function:6
h = h(S) = S ; > 0: (10.23)
5
Groth et al. (2010).
6
There is some empirical support for this hypothesis, cf. Section 9.5 of Chapter 9.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
10.5. A look ahead 171
Let every member of cohort v 0 spend S(v) years in school, thereby leaving
school with human capital h(v) = S(v) : Then the growth rate of h of the
cohort just leaving school is
dh(v)=dv S(v) 1 S 0 (v) S 0 (v)
= = :
h(v) S(v) S(v)
Assume sustained arithmetic growth in schooling length takes place due to
a steadily rising life expectancy. Then
S(v) = S0 + v; S0 0; > 0: (10.24)
Hence,
S 0 (v)
= ! 0 for v ! 1:
S(v) S0 + v
On this background the projection will be that also average human capital,
ht ; will be growing over time but at a rate, gh ; approaching 0 for t ! 1: This
gives no chance that the gh in the formula (10.13) can avoid approaching nil.
So our model rules out exponential per capita growth in the long run when
n = 0 and h(v) = S(v) :
As a thought experiment, suppose instead that the schooling technology
is exponential:
h = h(S) = e S(v) ; > 0: (10.25)
Then the growth rate of the human capital of the cohort just leaving school
is
dh(v)=dv e S(v) s0 (v)
= = S 0 (v) > 0: (10.26)
h(v) e S(v)
Assume arithmetic growth in life expectancy as well as schooling length,
the latter following (10.24). Then S 0 (v) = ; a positive constant. My
conjecture is that also average human capital, ht ; will in this case under
certain conditions grow at the constant rate, ; at least approximately (I
have not made the required demographic calculus).
Let us try some numbers. Suppose life expectancy in modern times
steadily increases by years per year and let schooling time and retirement
age be constant fractions of life expectancy. Let the schooling time fraction
be denoted !: Then S 0 (v) = = ! and gh = S 0 (v) = ! . With = 0:2;
! = 0:2; and = 0:10; we get gh = 0:004:7 Suppose n = 0:005 and ' = 0:5
7
As reported by Krueger and Lindahl (2001), in cross-section studies is usually
estimated to be in the range (0:05; 0:15):
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
CHAPTER 10. KNOWLEDGE CREATION AND HUMAN CAPITAL
172 IN A GROWING ECONOMY
(as suggested by Jones, 1995).8
Along a BGP with R&D we then have, by (10.13),
gh + n 0:004 + 0:005
gA = = = 0:018:
1 ' 0:5
In case =1 , (10.16) thus yields
gy = gA + gh = 0:018 + 0:004 = 0:022:
If instead n = 0 (in accordance with the long-run projection) and LAt =
LA 2 (0; L); we get along a BGP with R&D
gh 0:004
gy = + gh = + 0:004 = 0:012:
1 ' 1 '
In spite of n = 0; the thought experiment (10.25) thus leads to a non-
negligible level of exponential growth. With the compounding e¤ects of
exponential growth it is certainly substantial. I call it a “thought experi-
ment” because the empirical foundation of the exponential human capital
production function (10.25) is weak if not non-existing.
10.6 Concluding remarks
In a semi-endogenous growth setting we have considered human capital for-
mation and knowledge creating R&D. The latter is ultimately the factor
driving productivity growth unless one is willing to make very strong assump-
tions about the human capital production function. Technical knowledge is
capable of performing this role because it is a nonrival good and is “in…-
nitely expansible”, as emphasized by Paul Romer (1990) and Danny Quah
(1996). Contrary to this, in Lucas (1988) the distinction between technical
knowledge and human capital is not emphasized and it is the accumulation
of human capital that is driving long-run productivity growth.9
8
n = 0:005 per year may seem a low number for the empirical growth rate of research
labor (scientists and engineers) in the US and other countries over the last century. On
the other hand, for simplicity our model has ignored the likely duplication externality
due to overlap in R&D at the economy-wide level. Taking that overlap into account, we
should replace hLA in (10.4) by (hLA )1 ; and n in (10.20) by (1 )n; where 2 (0; 1)
measures the extent of duplication. Jones (1995) suggests = 0:5:
9
Although distinguishing between human capital and knowledge creation, the approach
by Dalgaard and Kreiner (2001) is very di¤erent from the one we have followed above and
has a¢ nity partly with Lucas and partly with Mankiw et al. (1992).
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
10.7. References 173
In the above analysis we have ignored the role of scarce natural resources
for limits to growth. We will come back to this issue in chapters 13 and 16.
We have ruled out ' = 1 because in combination with n > 0 it would
tend to generate a forever growing productivity growth rate, a feature not
in accordance with the actual economic evolution of the industrialized world
over the last century. We have ruled out ' > 1 because in combination
even with n = 0; it would tend to generate economic “explosion” in a very
dramatic and implausible sense: in…nite output in …nite time! Jones (2005)
argues that the empirical evidence speaks for ' < 1 in modern times.
The above analysis simply tells us what the growth rate must be in the
long run provided that the system considered converges to balanced growth.
On the other hand, speci…cation of the market structure and the household
sector, including demography and preferences, will be needed if we want to
study the adjustment processes outside balanced growth or determine an
equilibrium real interest rate or similar.
It is due to the semi-endogenous growth setting (the ' < 1 assumption)
that one can …nd the long-run per capita growth rate from knowledge of
technology parameters and the rate of population growth alone. How the
market structure and the household sector are described, is immaterial for
the long-run growth rate. These things will in the long run have “only”level
e¤ects.
Only if economic policy a¤ects the technology parameters or the popu-
lation growth rate, will it be able to a¤ect the long-run growth rate. Still,
economic policy can temporarily a¤ect economic growth and in this way af-
fect the level of the long-run growth path.
10.7 References
Arias, E., 2004, United States Life Tables 2004, National Vital Statistics
Reports, vol. 56, no. 9.
Dalgaard, C.-J., and C.-T. Kreiner, 2001, Is declining productivity in-
evitable? J. Econ. Growth, vol. 6 (3), 187-203.
Groth, C., K.-J. Koch, and T. M. Steger, 2010, When economic growth is
less than exponential, Economic Theory, vol. 44 (2), 213-242.
Jones, C.I., 1995, , J. Political Economy, vol.
Jones, C.I., 2005, Handbook of Economic Growth, vol. 1B, Elsevier.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
CHAPTER 10. KNOWLEDGE CREATION AND HUMAN CAPITAL
174 IN A GROWING ECONOMY
Krueger, A. B., and M. Lindahl, 2001. Education for growth: Why and for
whom? Journal of Economic Literature, 39, 1101-1136.
Lucas, R.E., 1988, On the mechanics of economic development, Journal of
Monetary Economics 22, 3-42.
Mankiw, G., D. Romer, and D. Weil, 1992, QJE.
Montanino, A., B. Przywara, and D. Young, 2004, Investment in education:
The implications for economic growth and public …nances, European
Commission Economic Paper, No. 217, November.
Oeppen, J., and J. W. Vaupel, 2002, Broken Limits to Life Expectancy,
Science Magazine, vol. 296, May 10, 1029-1031.
Quah, D.T., 1996, The Invisible Hand and the Weightless Economy, Occa-
sional Paper 12, Centre for Economic Performance, LSE, London, May
1996.
Romer, P., 1990, Endogenous technological change, J. Political Economy,
vol. 98 (5), S71-S102.
United Nations, 2013, World Population Prospects. The 2012 Revision.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
Chapter 11
AK and reduced-form AK
models. Consumption taxation.
In his Chapter 11 Acemoglu discusses simple fully-endogenous growth models
in the form of Ramsey-style AK and reduced-form AK models, respectively.
The name “AK”refers to a special feature of the aggregate production func-
tion, namely the absence of diminishing returns to capital. We present the
AK story within a Ramsey (i.e., representative agent) framework. A charac-
teristic result from AK models is that they have no transitional dynamics.
With the aim of synthesizing the formal characteristics of such models,
this lecture note gives an account of the common features of AK models
(Section 11.1) and reduced-form AK models (Section 11.2), respectively. Fi-
nally, for later application we discuss in Section 11.3 conditions under which
consumption taxation is not distortionary.
11.1 General equilibrium dynamics in the sim-
ple AK model
In the simple AK model (Acemoglu, Ch. 11.1) we consider a fully automa-
tized economy where the aggregate production function is
Y (t) = AK(t); A > 0: (11.1)
Thus there are constant returns to capital, not diminishing returns, and
labor is no longer a production factor. This section provides a detailed proof
that when we embed this technology in a Ramsey framework with perfect
competition, the model generates balanced growth from the beginning. So
there will be no transitional dynamics.
175
CHAPTER 11. AK AND REDUCED-FORM AK
176 MODELS. CONSUMPTION TAXATION
We consider a closed economy with perfect competition and no govern-
ment sector. The dynamic resource constraint for the economy is
_
K(t) = Y (t) c(t)L(t) K(t) = AK(t) c(t)L(t) K(0) > 0 given,
K(t);
(11.2)
where L(t) is the population size. After having found the equilibrium interest
rate to be r = A ; we …nd the equilibrium growth rate of per capita
consumption to be
c(t)
_ 1 1
= (r ) (A ) gc ; (11.3)
c(t)
a constant. To ensure positive growth we impose the parameter restriction
A > : (A1)
And to ensure boundedness of discounted utility (and thereby a possibility
of satisfying the transversality condition of the representative household) we
impose the additional parameter restriction:
n > (1 )gc : (A2)
Reordering gives
r = gc + > gc + n; (11.4)
where the equality is due to (16.27).
Solving the linear di¤erential equation (16.27) gives
c(t) = c(0)egc t ; (11.5)
where c(0) is unknown so far (because c is not a predetermined variable).
We shall …nd c(0) by appealing to the household’s transversality condition,
(r n)t
lim a(t)e = 0; (TVC)
t!1
where a(t) is per capita …nancial wealth at time t: Recalling the No-Ponzi-
Game condition,
lim a(t)e (r n)t 0; (NPG)
t!1
we see that the transversality condition is equivalent to the No-Ponzi-Game
condition being not over-satis…ed.
De…ning k(t) K(t)=L(t); the dynamic resource constraint, (11.2), is in
per-capita terms
_
k(t) = (A n)k(t) c(0)egc t ; k(0) > 0 given, (11.6)
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
11.1. General equilibrium dynamics in the simple AK model 177
where we have inserted (12.24). The solution to this linear di¤erential equa-
tion is (cf. Appendix to Chapter 3)
c(0) c(0)
k(t) = k(0) e(r n)t
+ egc t ; r A : (11.7)
r n gc r n gc
In our closed-economy framework with no public debt, a(t) = k(t): So the
question is: When will the time path (11.7) satisfy (TVC) with a(t) = k(t)?
To …nd out, we multiply by the discount factor e (r n)t on both sides of (11.7)
to get
c(0) c(0)
k(t)e (r n)t = k(0) + e (r gc n)t :
r n gc r n gc
Thus, in view of the assumption (A2), (11.4) holds and thereby the last term
on the right-hand side vanishes for t ! 1: Hence
c(0)
lim k(t)e (r n)t = k(0) :
t!1 r n gc
From this we see that the representative household satis…es (TVC) if and
only if it chooses
c(0) = (r n gc )k(0): (11.8)
This is the equilibrium solution for the household’s chosen per capita con-
sumption at time t = 0. If the household instead had chosen c(0) < (r n
gc )k(0); then limt!1 k(t)e (r n)t > 0 and so the household would not satisfy
(TVC) but instead be over-saving. And if it had chosen c(0) > (r n gc )k(0);
then limt!1 k(t)e (r n)t < 0 and so the household would be over-consuming
and violate (NPG) (hence also (TVC)).
Substituting the solution for c(0) into (11.7) gives the evolution of k(t)
in equilibrium,
c(0)
k(t) = egc t = k(0)egc t :
r n gc
So from the beginning k grows at the same constant rate as c. Since per
capita output is y Y =L = Ak; the same is true for per capita output.
Hence, from start the system is in balanced growth (there is no transitional
dynamics).
The AK model features one of the simplest kinds of endogenous growth
one can think of. Growth is endogenous in the model in the sense that there
is positive per capita growth in the long run, generated by an internal mecha-
nism in the model (not by exogenous technology growth). The endogenously
determined capital accumulation constitutes the mechanism through which
sustained per capita growth is generated. It is because the net marginal pro-
ductivity of capital is assumed constant and, according to (A1), higher than
the rate of impatience, ; that capital accumulation itself is so powerful.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
CHAPTER 11. AK AND REDUCED-FORM AK
178 MODELS. CONSUMPTION TAXATION
11.2 Reduced-form AK models
The models known as reduced-form AK models are a generalization of the
simple AK model considered above. In contrast to the simple AK model,
where only physical capital is an input, a reduced-form AK model assumes
a technology involving at least two di¤erent inputs. Yet it is possible that
in general equilibrium the aggregate production function ends up implying
proportionality between output and some measure of “broad capital”, i.e.,
~
Y (t) = B K(t);
where B is some endogenously determined positive constant and K(t) ~ is
“broad capital”. If in addition the real interest rate in general equilibrium
ends up being a constant, the model is called a reduced-form AK model. In the
simple AK model constancy of average productivity of capital is postulated
from the beginning. In the reduced-form AK models the average productivity
of capital becomes and remains endogenously constant over time.
In for instance the “AK model with physical and human capital”in Ace-
moglu, Ch. 11.2, along the balanced growth path (obtained after an initial
phase with specialization in either physical or human capital accumulation)
we have1
Y (t) = F (K(t); h(t)L(t)) = f (k^ )h(t)L(t) = f (k^ )H(t): (11.9)
De…ne
~
K(t) K(t) + H(t) = “broad capital”.
Then
K(t)
~
K(t) ( + 1)H(t) = (k^ + 1)H(t):
H(t)
Isolating H(t) and inserting into (11.9) gives
1 ~ ~
Y (t) = K(t) B K(t):
k^ + 1
So at the abstract level it is conceivable that “broad capital”, de…ned as the
sum of physical and human capital, can be meaningful. Empirically, however,
there is no basis for believing this concept of “broad capital to be useful, cf.
Exercises V.4 and V.5.
Anyway, a reduced-form AK model ends up with quite similar aggregate
relations as those in the simple AK model. Hence the solution procedure
1
The mentioned initial phase is left unnoticed in Acemoglu. In our notation k K=L
and k^ K=H; while Acemoglu’s text has k K=H:
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
11.3. On consumption taxation 179
to …nd the equilibrium path (see Chapter 12) is quite similar to that in the
simple AK model above. Again there will be no transitional dynamics.
The nice feature of AK models is that they provide very simple theoretical
examples of endogenous growth. The problematic feature is that they may
simplify the technology description too much and at best constitute knife-
edge cases. More about this in Chapter 13.
11.3 On consumption taxation
As a preparation for the discussion shortly in this course of …scal policy in
relation to economic growth, we shall here try to clarify an aspect of con-
sumption taxation. This is the question: is a consumption tax distortionary
- always? never? sometimes?
The answer is the following.
1. Suppose labor supply is elastic (due to leisure entering the utility
function). Then a consumption tax (whether constant or time-dependent) is
generally distortionary (not neutral). This is because it reduces the e¤ective
opportunity cost of leisure by reducing the amount of consumption forgone
by working one hour less. Indeed, the tax makes consumption goods more
expensive and so the amount of consumption that the agent can buy for
the hourly wage becomes smaller. The substitution e¤ect on leisure of a
consumption tax is thus positive, while the income and wealth e¤ects will be
negative. Generally, the net e¤ect will not be zero, but it can be of any sign;
it may be small in absolute terms.
2. Suppose labor supply is inelastic (no trade-o¤ between consumption
and leisure). Then, at least in the type of growth models we consider in this
course, a constant (time-independent) consumption tax acts as a lump-sum
tax and is thus non-distortionary. If the consumption tax is time-dependent,
however, a distortion of the intertemporal aspect of household decisions tends
to arise.
To understand answer 2, consider a Ramsey household with inelastic labor
supply. Suppose the household faces a time-varying consumption tax rate
t > 0: To obtain a consumption level per time unit equal to ct per capita,
the household has to spend
ct = (1 + t )ct
units of account (in real terms) per capita. Thus, spending ct per capita per
time unit results in the per capita consumption level
1
ct = (1 + t) ct : (11.10)
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
CHAPTER 11. AK AND REDUCED-FORM AK
180 MODELS. CONSUMPTION TAXATION
In order to concentrate on the consumption tax as such, we assume the
tax revenue is simply given back as lump-sum transfers and that there are
no other government activities. Then, with a balanced government budget,
we have
x t L t = t ct L t ;
where xt is the per capita lump-sum transfer, exogenous to the household,
and Lt is the size of the representative household.
Assuming CRRA utility with parameter > 0, the instantaneous per
capita utility can be written
c1t 1 (1 + t)
1 1
ct 1
u(ct ) = = :
1 1
In our standard notation the household’s intertemporal optimization prob-
lem, in continuous time, is then to choose (ct )1
t=0 so as to maximize
Z 1
(1 + t ) 1 ct1 1 ( n)t
U0 = e dt s.t.
0 1
ct 0;
a_ t = (rt n)at + wt + xt ct ; a0 given,
R1
(rs n)ds
lim at e 0 0:
t!1
From now, we let the timing of the variables be implicit unless needed for
clarity. The current-value Hamiltonian is
1 1
(1 + ) c 1
H= + [(r n)a + w + x c] ;
1
where is the co-state variable associated with …nancial per capita wealth,
a: An interior optimal solution will satisfy the …rst-order conditions
@H 1 1
= (1 + ) c = 0; so that (1 + ) c = ; (11.11)
@c
@H _ +(
= (r n) = n) ; (11.12)
@a
and a transversality condition which amounts to
R1
(rs n)ds
lim at e 0 = 0: (11.13)
t!1
We take logs in (11.11) to get
( 1) log(1 + ) log c = log :
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
11.3. On consumption taxation 181
Di¤erentiating w.r.t. time, taking into account that = t; gives
_ c _
( 1) = = r:
1+ c
By ordering, we …nd the growth rate of consumption spending,
c 1 _
= r+( 1) :
c 1+
Using (11.10), this gives the growth rate of consumption,
c_ c _ 1 _ _ 1 _
= = r+( 1) = (r ):
c c 1+ 1+ 1+ 1+
Assuming …rms maximize pro…t under perfect competition, in equilibrium
the real interest rate will satisfy
@Y
r= : (11.14)
@K
But the e¤ective real interest rate, r^; faced by the consuming household, is
_
r^ = r Q r for _ R 0;
1+
respectively. If for example the consumption tax is increasing, then the e¤ec-
tive real interest rate faced by the consumer is smaller than the market real
interest rate, given in (11.14), because saving implies postponing consump-
tion and future consumption is more expensive due to the higher consumption
tax rate.
The conclusion is that a time-varying consumption tax rate is distor-
tionary. It implies a wedge between the intertemporal rate of transformation
faced by the consumer, re‡ected by r^; and the intertemporal rate of transfor-
mation available in the technology of society, indicated by r in (11.14). On
the other hand, if the consumption tax rate is constant, the consumption
tax is non-distortionary when there is no utility from leisure.
A remark on tax smoothing
In models with transitional dynamics it is often so that maintaining constant
tax rates is inconsistent with maintaining a balanced government budget. Is
the implication of this that we should recommend the government to let tax
rates be continually adjusted so as to maintain a forever balanced budget?
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
CHAPTER 11. AK AND REDUCED-FORM AK
182 MODELS. CONSUMPTION TAXATION
No! As the above example as well as business cycle theory suggest, maintain-
ing tax rates constant (“tax smoothing”), and thereby allowing government
de…cits and surpluses to arise, will generally make more sense. In itself, a
budget de…cit is not worrisome. It only becomes worrisome if it is not accom-
panied later by su¢ cient budget surpluses to avoid an exploding government
debt/GDP ratio to arise. This requires that the tax rates taken together
have a level which in the long run matches the level of government expenses.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
Chapter 12
Learning by investing: two
versions
This lecture note is a supplement to Acemoglu, §11.4-5, where only Paul
Romer’s version of the learning-by-investing hypothesis is presented.
The learning-by-investing model, sometimes called the learning-by-doing
model, is one of the basic complete endogenous growth models. By “com-
plete” is meant that the model specifies not only the technological aspects of
the economy but also the market structure and the household sector, includ-
ing household preferences. As in much other endogenous growth theory the
modeling of the household sector follows Ramsey and assumes the existence
of a representative infinitely-lived household. Since this results in a simple
determination of the long-run interest rate (the modified golden rule), the
analyst can in a first approach concentrate on the main issue, technological
change, without being detracted by aspects secondary to this issue.
In the present model learning from investment experience and diffusion
across firms of the resulting new technical knowledge (positive externalities)
play a key role.
There are two popular alternative versions of the model. The distinguish-
ing feature is whether the learning parameter (see below) is less than one or
equal to one. The first case corresponds to (a simplified version of) a model
by Nobel laureate Kenneth Arrow (1962). The second case has been drawn
attention to by Paul Romer (1986) who assumes that the learning parameter
equals one. These two contributions start out from a common framework
which we now present.
183
CHAPTER 12. LEARNING BY INVESTING:
184 TWO VERSIONS
12.1 The common framework
We consider a closed economy with firms and households interacting under
conditions of perfect competition. Later, a government attempting to inter-
nalize the positive investment externality is introduced.
Let there be firms in the economy ( “large”). Suppose they all have
the same neoclassical production function, with CRS. Firm no. faces the
technology
= ( ) = 1 2 (12.1)
where the economy-wide technology level is an increasing function of so-
ciety’s previous experience, proxied by cumulative aggregate net investment:
µZ ¶
= = 0 ≤ 1 (12.2)
−∞
P
where is aggregate net investment and = 1
The idea is that investment − the production of capital goods − as an
unintended by-product results in experience or what we may call on-the-job
learning. Experience allows producers to recognize opportunities for process
and quality improvements. In this way knowledge is achieved about how to
produce the capital goods in a cost-efficient way and how to design them so
that in combination with labor they are more productive and better satisfy
the needs of the users. Moreover, as emphasized by Arrow,
“each new machine produced and put into use is capable of chang-
ing the environment in which production takes place, so that
learning is taking place with continually new stimuli” (Arrow,
1962).2
The learning is assumed to benefit essentially all firms in the economy.
There are knowledge spillovers across firms and these spillovers are reason-
ably fast relative to the time horizon relevant for growth theory. In our
macroeconomic approach both and are in fact assumed to be exactly
1
With arbitrary units of measurement for labor and output the hypothesis is =
0 In (12.2) measurement units are chosen such that = 1.
2
Concerning empirical evidence of learning-by-doing and learning-by-investing, see
Chapter 13. The citation of Arrow indicates that it was rather experience from cumu-
lative gross investment he had in mind as the basis for learning. Yet the hypothesis in
(12.2) is the more popular one - seemingly for no better reason than that it leads to
simpler dynamics. Another way in which (12.2) deviates from Arrow’s original ideas is
by assuming that technical progress is disembodied rather than embodied, an important
distinction to which we return in Chapter 13.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
12.1. The common framework 185
the same for all firms in the economy. That is, in this specification the firms
producing consumption-goods benefit from the learning just as much as the
firms producing capital-goods.
The parameter indicates the elasticity of the general technology level,
with respect to cumulative aggregate net investment and is named the
“learning parameter”. Whereas Arrow assumes 1 Romer focuses on the
case = 1 The case of 1 is ruled out since it would lead to explosive
growth (infinite output in finite time) and is therefore not plausible.
12.1.1 The individual firm
In the simple Ramsey model we assumed that households directly own the
capital goods in the economy and rent them out to the firms. When dis-
cussing learning-by-investment, it somehow fits the intuition better if we
(realistically) assume that the firms generally own the capital goods they
use. They then finance their capital investment by issuing shares and bonds.
Households’ financial wealth then consists of these shares and bonds.
Consider firm There is perfect competition in all markets. So the firm
is a price taker. Its problem is to choose a production and investment plan
which maximizes the present value, of expected future cash-flows. Thus
the firm chooses ( )∞
=0 to maximize
Z ∞
0 = [ ( ) − − ] − 0
0
subject to ̇ = − Here and are the real wage and gross
investment, respectively, at time , is the real interest rate at time and
≥ 0 is the capital depreciation rate. Rising marginal capital installation
costs and other kinds of adjustment costs are assumed minor and can be
ignored. It can be shown that in this case the firm’s problem is equivalent
to maximization of current pure profits in every short time interval. So, as
hitherto, we can describe the firm as just solving a series of static profit
maximization problems.
We suppress the time index when not needed for clarity. At any date firm
maximizes current pure profits, Π = ( ) − ( + ) − This
leads to the first-order conditions for an interior solution:
Π = 1 ( ) − ( + ) = 0 (12.3)
Π = 2 ( ) − = 0
Behind (12.3) is the presumption that each firm is small relative to the econ-
omy as a whole, so that each firm’s investment has a negligible effect on
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
CHAPTER 12. LEARNING BY INVESTING:
186 TWO VERSIONS
the economy-wide technology level . Since is homogeneous of degree
one, by Euler’s theorem,3 the first-order partial derivatives, 1 and 2 are
homogeneous of degree 0. Thus, we can write (12.3) as
1 ( ) = + (12.4)
where ≡ . Since is neoclassical, 11 0 Therefore (12.4) deter-
mines uniquely. From (12.4) follows that the chosen capital-labor ratio,
will be the same for all firms, say ̄
12.1.2 The household
The representative household (or family dynasty) has = 0 members
each of which supplies one unit of labor inelastically per time unit, ≥ 0.
The household has CRRA instantaneous utility with parameter 0 The
pure rate of time preference is a constant, . The flow budget identity in per
head terms is
̇ = ( − ) + − 0 given,
where is per head financial wealth. The NPG condition is
lim − 0 ( −) ≥ 0
→∞
The resulting consumption-saving plan implies that per head consumption
follows the Keynes-Ramsey rule,
̇ 1
= ( − )
and the transversality condition that the NPG condition is satisfied with
strict equality. In general equilibrium of our closed economy without natural
resources and government debt, will equal
12.1.3 Equilibrium in factor markets
P P
In equilibrium = and = where and are the avail-
able
P amounts P of capital
P and labor, respectively (both pre-determined). Since
=
= ̄ = ̄ the chosen capital intensity, satisfies
= ̄ = ≡ = 1 2 (12.5)
3
Recall that a function ( ) defined in a domain is homogeneous of degree if for
all ( ) in ( ) = ( ) for all 0 If a differentiable function ( ) is
homogeneous of degree then (i) 10 ( ) + 20 ( ) = ( ) and (ii) the first-order
partial derivatives, 10 ( ) and 20 ( ) are homogeneous of degree − 1.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
12.2. The arrow case: 1 187
As a consequence we can use (12.4) to determine the equilibrium interest
rate:
= 1 ( ) − (12.6)
That is, whereas in the firm’s first-order condition (12.4) causality goes from
to in (12.6) causality goes from to Note also that in our closed
economy with no natural resources and no government debt, will equal
The implied aggregate production function is
X X X X
= ≡ = ( ) = ( ) (by (12.1) and (12.5))
X
= ( ) = ( ) = ( ) = ( ) (by (12.2)), (12.7)
where we have several times used that is homogeneous of degree one.
12.2 The arrow case: 1
The Arrow case is the robust case where the learning parameter satisfies
0 1 The method for analyzing the Arrow case is analogue to that
used in the study of the Ramsey model with exogenous technical progress.
In particular, aggregate capital per unit of effective labor, ̃ ≡ () is a
key variable. Let ̃ ≡ () Then
( )
̃ = = (̃ 1) ≡ (̃) 0 0 00 0 (12.8)
We can now write (12.6) as
= 0 (̃ ) − (12.9)
where ̃ is pre-determined.
12.2.1 Dynamics
From the definition ̃ ≡ () follows
·
̃ ̇ ̇ ̇ ̇ ̇
= − − = − − (by (12.2))
̃
− − ̃ − ̃ − ̃
= (1 − ) − = (1 − ) − where ̃ ≡ ≡
̃
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
CHAPTER 12. LEARNING BY INVESTING:
188 TWO VERSIONS
Multiplying through by ̃ we have
·
̃ = (1 − )( (̃) − ̃) − [(1 − ) + ] ̃ (12.10)
In view of (12.9), the Keynes-Ramsey rule implies
̇ 1 1³ 0 ´
≡ = ( − ) = (̃) − − (12.11)
Defining ̃ ≡ now follows
̃ ̇ ̇ ̇ ̇ ̇ − − ̇
= − = − = − = − (̃ − ̃ − ̃)
̃ ̃
1 0
= ( (̃) − − ) − (̃ − ̃ − ̃)
̃
Multiplying through by ̃ we have
∙ ¸
· 1 0
̃ = ( (̃) − − ) − ((̃) − ̃ − ̃) ̃ (12.12)
̃
The two coupled differential equations, (12.10) and (12.12), determine
the evolution over time of the economy.
Phase diagram
·
Figure 12.1 depicts the phase diagram. The ̃ = 0 locus comes from (12.10),
which gives
·
̃ = 0 for ̃ = (̃) − ( + )̃ (12.13)
1−
·
where we realistically may assume that + (1 − ) 0 As to the ̃ = 0
locus, we have
· ̃ 0
̃ = 0 for ̃ = (̃) − ̃ − ( (̃) − − )
̃
= (̃) − ̃ − ≡ (̃) (from (12.11)). (12.14)
·
Before determining the slope of the ̃ = 0 locus, it is convenient to consider
the steady state, (̃∗ ̃∗ ).
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
12.2. The arrow case: 1 189
c
c 0
A
c*
E
k 0
B
k
0 k* kGR k
Figure 12.1: Phase diagram for the Arrow model.
Steady state
In a steady state ̃ and ̃ are constant so that the growth rate of as well
as equals ̇ + i.e.,
̇ ̇ ̇ ̇
= = + = +
Solving gives
̇ ̇
= =
1−
Thence, in a steady state
̇
= −= −= ≡ ∗ and (12.15)
1− 1−
̇ ̇
= = = ∗ (12.16)
1−
The steady-state values of and ̃ respectively, will therefore satisfy, by
(12.11),
∗ = 0 (̃∗ ) − = + ∗ = + (12.17)
1−
To ensure existence of a steady state we assume that the private marginal
product of capital is sufficiently sensitive to capital per unit of effective labor,
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
CHAPTER 12. LEARNING BY INVESTING:
190 TWO VERSIONS
from now called the “capital intensity”:
lim 0 (̃) + + lim 0 (̃) (A1)
̃→0 1 − ̃→∞
The
transversality condition of the representative household is that lim→∞
0 ( −) = 0 where is per capita financial wealth. In general equi-
−
librium = ≡ ̃ where in steady state grows according to (12.16).
Thus, in steady state the transversality condition can be written
∗ ∗ +)
lim ̃∗ ( − = 0 (TVC)
→∞
For this to hold, we need
∗ ∗ + = (12.18)
1−
by (12.15). In view of (12.17), this is equivalent to
− (1 − ) (A2)
1−
which we assume satisfied.·
As to the slope of the ̃ = 0 locus we have, from (12.14),
1 00 (̃) 1
0 (̃) = 0 (̃) − − (̃ + ) 0 (̃) − − (12.19)
since 00 0 At least in a small neighborhood of the steady state we can
sign the right-hand side of this expression. Indeed,
1 1
0 (̃∗ )−− ∗ = +∗ − ∗ = + − = −−(1−) 0
1− 1− 1−
(12.20)
by (12.15) and (A2). So, combining with (12.19), we conclude that 0 (̃∗ ) 0
By continuity, in a small neighborhood of the steady state, 0 (̃) ≈ 0 (̃∗ ) 0
·
Therefore, close to the steady state, the ̃ = 0 locus is positively sloped, as
indicated in Figure 12.1.
Still, we have to check the following question: In a neighborhood of the
· ·
steady state, which is steeper, the ̃ = 0 locus or the ̃ = 0 locus? The slope
of the latter is 0 (̃) − − (1 − ) from (12.13) At the steady state this
slope is
1
0 (̃∗ ) − − ∗ ∈ (0 0 (̃∗ ))
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
12.2. The arrow case: 1 191
· ·
in view of (12.20) and (12.19). The ̃ = 0 locus is thus steeper. So, the ̃ = 0
·
locus crosses the ̃ = 0 locus from below and can only cross once.
The assumption (A1) ensures existence of a ̃∗ 0 satisfying (12.17). As
Figure 12.1 is drawn, a little more is implicitly assumed namely that there
exists a ̂ 0 such that the private net marginal product of capital equals
the steady-state growth rate of output, i.e.,
̇ ̇ ̇
0 (̂) − = ( )∗ = ( )∗ + = += (12.21)
1− 1−
·
where we have used (12.16). Thus, the tangent to the ̃ = 0 locus at ̃ = ̂
is horizontal and ̂ ̃∗ as indicated in the figure.
Note, however, that ̂ is not the golden-rule capital intensity. The latter
is the capital intensity, ̃ at which the social net marginal product of
capital equals the steady-state growth rate of output (see Appendix). If ̃
exists, it will be larger than ̂ as indicated in Figure 12.1. To see this, we
now derive a convenient expression for the social marginal product of capital.
From (12.7) we have
= 1 (·) + 2 (·) −1 = 0 (̃) + 2 (·) ( −1 ) (by (12.8))
= 0 (̃) + ( (·) − 1 (·)) −1 (by Euler’s theorem)
0 0 −1
= (̃) + ((̃) − (̃)) (by (12.8) and (12.2))
(̃) − ̃ 0 (̃)
= 0 (̃) + ((̃) −1 − 0 (̃)) = 0 (̃) + 0 (̃)
̃
in view of ̃ = ( ) = 1− −1 and (̃)̃− 0 (̃) 0 As expected, the
positive externality makes the social marginal product of capital larger than
the private one. Since we can also write = (1 − ) 0 (̃) + (̃)̃
we see that is (still) a decreasing function of ̃ since both 0 (̃) and
(̃)̃ are decreasing in ̃ So the golden rule capital intensity, ̃ will be
that capital intensity which satisfies
à !∗
0
(̃ ) − ̃ (̃ ) ̇
0 (̃ ) + − = =
̃ 1−
To ensure there exists such a ̃ we strengthen the right-hand side inequal-
ity in (A1) by the assumption
à !
0
(̃) − ̃ (̃)
lim 0 (̃) + + (A3)
̃→∞ ̃ 1−
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
CHAPTER 12. LEARNING BY INVESTING:
192 TWO VERSIONS
00
This, together with (A1) and 0, implies existence of a unique ̃ , and
in view of our additional assumption (A2), we have 0 ̃∗ ̂ ̃ as
displayed in Figure 12.1.
Stability
The arrows in Figure 12.1 indicate the direction of movement, as determined
by (12.10) and (12.12)). We see that the steady state is a saddle point. The
dynamic system has one pre-determined variable, ̃ and one jump variable, ̃
The saddle path is not parallel to the jump variable axis. We claim that for a
given ̃0 0 (i) the initial value of ̃0 will be the ordinate to the point where
the vertical line ̃ = ̃0 crosses the saddle path; (ii) over time the economy
will move along the saddle path towards the steady state. Indeed, this time
path is consistent with all conditions of general equilibrium, including the
transversality condition (TVC). And the path is the only technically feasible
path with this property. Indeed, all the divergent paths in Figure 12.1 can
be ruled out as equilibrium paths because they can be shown to violate the
transversality condition of the household.
In the long run and ≡ ≡ ̃ = (̃∗ ) grow at the rate (1−)
which is positive if and only if 0 This is an example of endogenous growth
in the sense that the positive long-run per capita growth rate is generated
through an internal mechanism (learning) in the model (in contrast to exoge-
nous technology growth as in the Ramsey model with exogenous technical
progress).
12.2.2 Two types of endogenous growth
As also touched upon in Chapter 10, it is useful to distinguish between two
types of endogenous growth. Fully endogenous growth occurs when the long-
run growth rate of is positive without the support from growth in any
exogenous factor (for example exogenous growth in the labor force); the
Romer case, to be considered in the next section, provides an example. Semi-
endogenous growth occurs if growth is endogenous but a positive per capita
growth rate can not be maintained in the long run without the support from
growth in some exogenous factor (for example growth in the labor force).
Clearly, in the Arrow version of learning by investing, growth is “only” semi-
endogenous. The technical reason for this is the assumption that the learning
parameter, is below 1 which implies diminishing marginal returns to cap-
ital at the aggregate level. As a consequence, if and only if 0 do we
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
12.3. Romer’s limiting case: = 1 = 0 193
have ̇ 0 in the long run.4 In line with this, ∗ 0
The key role of population growth derives from the fact that although
there are diminishing marginal returns to capital at the aggregate level, there
are increasing returns to scale w.r.t. capital and labor. For the increasing
returns to be exploited, growth in the labor force is needed. To put it differ-
ently: when there are increasing returns to and together, growth in the
labor force not only counterbalances the falling marginal product of aggre-
gate capital (this counter-balancing role reflects the direct complementarity
between and ), but also upholds sustained productivity growth via the
learning mechanism.
Note that in the semi-endogenous growth case, ∗ = (1 − )2 0
for 0 That is, a higher value of the learning parameter implies higher
per capita growth in the long run, when 0. Note also that ∗ = 0
= ∗ that is, in the semi-endogenous growth case, preference parameters
do not matter for the long-run per capita growth rate. As indicated by
(12.15), the long-run growth rate is tied down by the learning parameter,
and the rate of population growth, Like in the simple Ramsey model,
however, it can be shown that preference parameters matter for the level of
the growth path. For instance (12.17) shows that ̃∗ 0 so that more
patience (lower ) imply a higher ̃∗ and thereby a higher = (̃∗ )
This suggests that although taxes and subsidies do not have long-run
growth effects, they can have level effects.
12.3 Romer’s limiting case: = 1 = 0
We now consider the limiting case = 1 We should think of it as a thought
experiment because, by most observers, the value 1 is considered an unrealis-
tically high value for the learning parameter. Moreover, in combination with
0 the value 1 will lead to a forever rising per capita growth rate which
does not accord the economic history of the industrialized world over more
than a century. To avoid a forever rising growth rate, we therefore introduce
the parameter restriction = 0
The resulting model turns out to be extremely simple and at the same
time it gives striking results (both circumstances have probably contributed
to its popularity).
First, with = 1 we get = and so the equilibrium interest rate is,
4
Note, however, that the model, and therefore (12.15), presupposes ≥ 0 If 0
then would tend to be decreasing and so, by (12.2), the level of technical knowledge
would be decreasing, which is implausible, at least for a modern industrialized economy.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
CHAPTER 12. LEARNING BY INVESTING:
194 TWO VERSIONS
by (12.6),
= 1 ( ) − = 1 (1 ) − ≡ ̄
where we have divided the two arguments of 1 ( ) by ≡ and
again used Euler’s theorem. Note that the interest rate is constant “from the
beginning” and independent of the historically given initial value of 0 .
The aggregate production function is now
= ( ) = (1 ) constant, (12.22)
and is thus linear in the aggregate capital stock.5 In this way the general neo-
classical presumption of diminishing returns to capital has been suspended
and replaced by exactly constant returns to capital. Thereby the Romer
model belongs to the class of reduced-form AK models, that is, models where
in general equilibrium the interest rate and the aggregate output-capital ratio
are necessarily constant over time whatever the initial conditions.
The method for analyzing an AK model is different from the one used for
a diminishing returns model as above.
12.3.1 Dynamics
The Keynes-Ramsey rule now takes the form
̇ 1 1
= (̄ − ) = (1 (1 ) − − ) ≡ (12.23)
which is also constant “from the beginning”. To ensure positive growth, we
assume
1 (1 ) − (A1’)
And to ensure bounded intertemporal utility (and thereby a possibility of
satisfying the transversality condition of the representative household), it is
assumed that
(1 − ) and therefore + = ̄ (A2’)
Solving the linear differential equation (12.23) gives
= 0 (12.24)
where 0 is unknown so far (because is not a predetermined variable). We
shall find 0 by applying the households’ transversality condition
lim −̄ = lim −̄ = 0 (TVC)
→∞ →∞
5
Acemoglu, p. 400, writes this as = ˜()
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
12.3. Romer’s limiting case: = 1 = 0 195
F1 (1, L )
F ( K , L)
F (1, L)
F (1, L)
K
1
Figure 12.2: Illustration of the fact that for given, (1 ) 1 (1 )
First, note that the dynamic resource constraint for the economy is
̇ = − − = (1 ) − −
or, in per-capita terms,
̇ = [ (1 ) − ] − 0 (12.25)
In this equation it is important that (1 ) − − 0 To understand
this inequality, note that, by (A2’), (1 ) − − (1 ) − − ̄ =
(1 ) − 1 (1 ) = 2 (1 ) 0 where the first equality is due to ̄
= 1 (1 ) − and the second is due to the fact that since is homogeneous
of degree 1, we have, by Euler’s theorem, (1 ) = 1 (1 ) · 1 + 2 (1 )
1 (1 ) in view of (A1’). The key property (1 ) − 1 (1 ) 0 is
illustrated in Figure 12.2.
The solution of a general linear differential equation of the form ̇() +
() = with 6= − is
() = ((0) − )− + (12.26)
+ +
Thus the solution to (12.25) is
0 0
= (0 − )( (1)−) + (12.27)
(1 ) − − (1 ) − −
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
CHAPTER 12. LEARNING BY INVESTING:
196 TWO VERSIONS
To check whether (TVC) is satisfied we consider
0 0
−̄ = (0 − )( (1)−−̄) + (−̄)
(1 ) − − (1 ) − −
0
→ (0 − )( (1)−−̄) for → ∞
(1 ) − −
since ̄ by (A2’). But ̄ = 1 (1 ) − (1 ) − and so (TVC) is
only satisfied if
0 = ( (1 ) − − )0 (12.28)
If 0 is less than this, there will be over-saving and (TVC) is violated ( −̄ →
∞ for → ∞ since = ). If 0 is higher than this, both the NPG and
(TVC) are violated ( −̄ → −∞ for → ∞).
Inserting the solution for 0 into (12.27), we get
0
= = 0
(1 ) − −
that is, grows at the same constant rate as “from the beginning” Since
≡ = (1 ) the same is true for Hence, from start the system is
in balanced growth (there is no transitional dynamics).
This is a case of fully endogenous growth in the sense that the long-run
growth rate of is positive without the support by growth in any exogenous
factor. This outcome is due to the absence of diminishing returns to aggregate
capital, which is implied by the assumed high value of the learning parameter.
But the empirical foundation for this high value is weak, to say the least, cf.
Chapter 13. A further drawback of this special version of the learning model
is that the results are non-robust. With slightly less than 1, we are back
in the Arrow case and growth peters out, since = 0 With slightly above
1, it can be shown that growth becomes explosive: infinite output in finite
time!6
The Romer case, = 1 is thus a knife-edge case in a double sense. First,
it imposes a particular value for a parameter which apriori can take any value
within an interval. Second, the imposed value leads to non-robust results;
values in a hair’s breadth distance result in qualitatively different behavior
of the dynamic system.
Note that the causal structure in the long run in the diminishing returns
case is different than in the AK-case of Romer. In the diminishing returns
case the steady-state growth rate is determined first, as ∗ in (12.15), then
∗ is determined through the Keynes-Ramsey rule and, finally, is de-
termined by the technology, given ∗ In contrast, the Romer case has
6
See Appendix B in Chapter 13.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
12.3. Romer’s limiting case: = 1 = 0 197
and directly given as (1 ) and ̄ respectively. In turn, ̄ determines the
(constant) equilibrium growth rate through the Keynes-Ramsey rule
12.3.2 Economic policy in the Romer case
In the AK case, that is, the fully endogenous growth case, we have
0 and 0 Thus, preference parameters matter for the long-run
growth rate and not “only” for the level of the upward-sloping time path
of per capita output. This suggests that taxes and subsidies can have long-
run growth effects. In any case, in this model there is a motivation for
government intervention due to the positive externality of private investment.
This motivation is present whether 1 or = 1 Here we concentrate on
the latter case, for no better reason than that it is simpler. We first find the
social planner’s solution.
The social planner
Recall that by a social planner we mean a fictional ”all-knowing and all-
powerful” decision maker who maximizes an objective function under no
other constraints than what follows from technology and initial resources.
The social planner faces the aggregate production function (12.22) or, in per
capita terms, = (1 ) The social planner’s problem is to choose ( )∞
=0
to maximize
Z ∞ 1−
0 = − s.t.
0 1 −
≥ 0
̇ = (1 ) − − 0 0 given, (12.29)
≥ 0 for all 0 (12.30)
The current-value Hamiltonian is
1−
( ) = + ( (1 ) − − )
1−
where = is the adjoint variable associated with the state variable, which
is capital per unit of labor. Necessary first-order conditions for an interior
optimal solution are
= − − = 0, i.e., − = (12.31)
= ( (1 ) − ) = −̇ + (12.32)
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
CHAPTER 12. LEARNING BY INVESTING:
198 TWO VERSIONS
We guess that also the transversality condition,
lim − = 0 (12.33)
→∞
must be satisfied by an optimal solution.7 This guess will be of help in finding
a candidate solution. Having found a candidate solution, we shall invoke a
theorem on sufficient conditions to ensure that our candidate solution is
really an optimal solution.
Log-differentiating w.r.t. in (12.31) and combining with (12.32) gives
the social planner’s Keynes-Ramsey rule,
̇ 1
= ( (1 ) − − ) ≡ (12.34)
We see that This is because the social planner internalizes the
economy-wide learning effect associated with capital investment, that is, the
social planner takes into account that the “social” marginal product of capital
is = (1 ) 1 (1 ) To ensure bounded intertemporal utility we
sharpen (A2’) to
(1 − ) (A2”)
To find the time path of , note that the dynamic resource constraint (12.29)
can be written
̇ = ( (1 ) − ) − 0
in view of (12.34). By the general solution formula (12.26) this has the
solution
0 0
= (0 − )( (1)−) + (12.35)
(1 ) − − (1 ) − −
In view of (12.32), in an interior optimal solution the time path of the adjoint
variable is
= 0 −[( (1)−−]
where 0 = −
0 0 by (12.31) Thus, the conjectured transversality condi-
tion (12.33) implies
lim −( (1)−) = 0 (12.36)
→∞
7
The proviso implied by saying “guess” is due to the fact that optimal control theory
does not guarantee that this “standard” transversality condition is necessary for optimality
in all infinite horizon optimization problems.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
12.3. Romer’s limiting case: = 1 = 0 199
where we have eliminated 0 To ensure that this is satisfied, we multiply
from (12.35) by −( (1)−) to get
0 0
−( (1)−) = 0 − + [ −( (1)−)]
(1 ) − − (1 ) − −
0
→ 0 − for → ∞
(1 ) − −
since, by (A2”), + = (1 ) − in view of (12.34). Thus,
(12.36) is only satisfied if
0 = ( (1 ) − − )0 (12.37)
Inserting this solution for 0 into (12.35), we get
0
= = 0
(1 ) − −
that is, grows at the same constant rate as “from the beginning” Since
≡ = (1 ) the same is true for Hence, our candidate for the social
planner’s solution is from start in balanced growth (there is no transitional
dynamics).
The next step is to check whether our candidate solution satisfies a set of
sufficient conditions for an optimal solution. Here we can use Mangasarian’s
theorem which, applied to a problem like this, with one control variable and
one state variable, says that the following conditions are sufficient:
(a) Concavity: The Hamiltonian is jointly concave in the control and state
variables, here and .
(b) Non-negativity: There is for all ≥ 0 a non-negativity constraint on
the state variable; and the co-state variable, is non-negative for all
≥ 0.
(c) TVC: The candidate solution satisfies the transversality condition lim→∞
− = 0 where − is the discounted co-state variable.
In the present case we see that the Hamiltonian is a sum of concave
functions and therefore is itself concave in ( ) Further, from (12.30) we see
that condition (b) is satisfied. Finally, our candidate solution is constructed
so as to satisfy condition (c). The conclusion is that our candidate solution
is an optimal solution. We call it the SP allocation.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
CHAPTER 12. LEARNING BY INVESTING:
200 TWO VERSIONS
Implementing the SP allocation in the market economy
Returning to the market economy, we assume there is a policy maker, say
the government, with only two activities. These are (i) paying an investment
subsidy, to the firms so that their capital costs are reduced to
(1 − )( + )
per unit of capital per time unit; (ii) financing this subsidy by a constant
consumption tax rate
Let us first find the size of needed to establish the SP allocation. Firm
now chooses such that
| fixed = 1 ( ) = (1 − )( + )
By Euler’s theorem this implies
1 ( ) = (1 − )( + ) for all
so that in equilibrium we must have
1 ( ) = (1 − )( + )
where ≡ which is pre-determined from the supply side. Thus, the
equilibrium interest rate must satisfy
1 ( ) 1 (1 )
= − = − (12.38)
1− 1−
again using Euler’s theorem.
It follows that should be chosen such that the “right” arises. What is
the “right” ? It is that net rate of return which is implied by the production
technology at the aggregate level, namely − = (1 )− If we can
obtain = (1 )− then there is no wedge between the intertemporal rate
of transformation faced by the consumer and that implied by the technology.
The required thus satisfies
1 (1 )
= − = (1 ) −
1−
so that
1 (1 ) (1 ) − 1 (1 ) 2 (1 )
=1− = =
(1 ) (1 ) (1 )
In case = ( )1− 0 1 = 1. . . this gives = 1 −
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
12.4. Appendix: The golden-rule capital intensity in the Arrow case 201
It remains to find the required consumption tax rate The tax revenue
will be and the required tax revenue is
T = ( + ) = ( (1 ) − 1 (1 )) =
Thus, with a balanced budget the required tax rate is
T (1 ) − 1 (1 ) (1 ) − 1 (1 )
= = = 0 (12.39)
(1 ) − −
where we have used that the proportionality in (12.37) between and
holds for all ≥ 0 Substituting (12.34) into (12.39), the solution for can
be written
[ (1 ) − 1 (1 )] 2 (1 )
= =
( − 1)( (1 ) − ) + ( − 1)( (1 ) − ) +
The required tax rate on consumption is thus a constant. It therefore does
not distort the consumption/saving decision on the margin, cf. Chapter 11.
It follows that the allocation obtained by this subsidy-tax policy is the SP
allocation. A policy, here the policy ( ) which in a decentralized system
induces the SP allocation, is called a first-best policy.
12.4 Appendix: The golden-rule capital in-
tensity in the Arrow case
In our discussion of the Arrow model in Section 12.2 (where 0 1)
we claimed that the golden-rule capital intensity, ̃ will be that effective
capital-labor ratio at which the social net marginal product of capital equals
the steady-state growth rate of output. In this respect the Arrow model with
endogenous technical progress is similar to the standard neoclassical growth
model with exogenous technical progress.
The claim corresponds to a very general theorem, valid also for models
with many capital goods and non-existence of an aggregate production func-
tion. This theorem says that the highest sustainable path for consumption
per unit of labor in the economy will be that path which results from those
techniques which profit maximizing firms choose under perfect competition
when the real interest rate equals the steady-state growth rate of GNP (see
Gale and Rockwell, 1975).
To prove our claim, note that in steady state, (12.14) holds whereby
consumption per unit of labor (here the same as per capita consumption in
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
CHAPTER 12. LEARNING BY INVESTING:
202 TWO VERSIONS
view of = labor force = population) can be written
∙ ¸
≡ ̃ = (̃) − ( + )̃
1−
∙ ¸³ ´
∗
= (̃) − ( + )̃ 0 1− (by = )
1− 1−
∙ ¸³ ´
1 1−
= (̃) − ( + )̃ (̃0 ) 1− 1− (from ̃ = = also for = 0)
1−
∙ ¸
= (̃) − ( + )̃ ̃ 1− 0 1− 1− ≡ (̃)0 1− 1−
1−
defining (̃) in the obvious way.
We look for that value of ̃ at which this steady-state path for is at the
highest technically feasible level. The positive coefficient, 0 1− 1− , is the
only time dependent factor and can be ignored since it is exogenous. The
problem is thereby reduced to the static problem of maximizing (̃) with
respect to ̃ 0 We find
∙ ¸ ∙ ¸
0 0
(̃) = (̃) − ( + ) ̃ 1− + (̃) − ( + )̃ ̃ 1− −1
1− 1− 1−
" Ã ! #
(̃)
= 0 (̃) − ( + )+ − ( + ) ̃ 1−
1− ̃ 1− 1−
" #
( ̃) ̃ 1−
= (1 − ) 0 (̃) − (1 − ) − + − ( + )
̃ 1− 1−
" #
0 (̃) ̃ 1− ̃ 1−
= (1 − ) (̃) − + − ≡ (̃) (12.40)
̃ 1− 1− 1−
defining (̃) in the obvious way. The first-order condition for the problem,
0 (̃) = 0 is equivalent to (̃) = 0 After ordering this gives
(̃) − ̃ 0 (̃)
0 (̃) + − = (12.41)
̃ 1−
We see that
0 (̃) R 0 for (̃) R 0
respectively. Moreover,
(̃) − ̃ 0 (̃)
0 (̃) = (1 − ) 00 (̃) − 0
̃2
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
12.4. Appendix: The golden-rule capital intensity in the Arrow case 203
in view of 00 0 and (̃)̃ 0 (̃) So a ̃ 0 satisfying (̃) = 0 is the
unique maximizer of (̃) By (A1) and (A3) in Section 12.2 such a ̃ exists
and is thereby the same as the ̃ we were looking for.
The left-hand side of (12.41) equals the social marginal product of capital
and the right-hand side equals the steady-state growth rate of output. At
̃ = ̃ it therefore holds that
à !∗
̇
− =
This confirms our claim in Section 12.2 about ̃ .
Remark about the absence of a golden rule in the Romer model. In the
Romer model the golden rule is not a well-defined concept for the following
reason. Along any balanced growth path we have from (12.29),
̇ 0
≡ = (1 ) − − = (1 ) − −
0
because (= ) is by definition constant along a balanced growth path,
whereby also must be constant. We see that is decreasing linearly
from (1 ) − to − when 0 0 rises from nil to (1 ) So choosing
among alternative technically feasible balanced growth paths is inevitably a
choice between starting with low consumption to get high growth forever or
starting with high consumption to get low growth forever. Given any 0 0
the alternative possible balanced growth paths will therefore sooner or later
cross each other in the ( ln ) plane. Hence, for the given 0 there exists no
balanced growth path which for all ≥ 0 has higher than along any other
technically feasible balanced growth path. So no golden rule path exists.
This is a general property of AK and reduced-form AK models.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
CHAPTER 12. LEARNING BY INVESTING:
204 TWO VERSIONS
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
Chapter 13
Perspectives on learning by
doing and learning by investing
This chapter adds some theoretical and empirical perspectives to the dis-
cussion in Chapter 12 and in Acemoglu, Chapter 11 and 12. The contents
are:
1. Learning by doing, learning by using, learning by watching
2. Disembodied learning by investing
3. Disembodied vs. embodied technical change
4. Static comparative advantage vs. dynamics of learning by doing*
5. Robustness and scale effects
(a) On terminology
(b) Robustness of simple endogenous growth models
(c) Weak and strong scale effects
(d) Discussion
Sections and sub-sections marked by an asterisk are only cursory reading.
The growth rate of any time-dependent variable 0 is written ≡ ̇
In this chapter the economy-wide technology level at time is denoted
rather than
205
CHAPTER 13. PERSPECTIVES ON LEARNING BY DOING
206 AND LEARNING BY INVESTING
13.1 Learning by doing, learning by using,
learning by watching
The term learning by doing refers to the hypothesis that accumulated work
experience, including repetition of the same type of action, improves workers’
productivity and adds to technical knowledge. In connection with training
in applying new production equipment sometimes the related term learning
by using is appropriate. In a broader context the literature sometimes refers
to spillover effects as learning by watching.
A learning-by-doing model typically combines an aggregate CRS produc-
tion function,
= ( ) (13.1)
with a learning function, for example,
̇ = 0 0 ≤ 1 (13.2)
where is a learning parameter and is a constant that, depending on the
value of and the complete model in which (13.2) is embedded, is either an
unimportant constant that depends only on measuring units or a parameter
of importance for the productivity level or even the productivity growth
rate. In Section 13.4 below, on the resource curse problem, we consider a
two-sector model where each sector’s productivity growth is governed by such
a relationship.1
Another learning hypothesis is of the form
̇ = 0 0 given, 0 ≤ 1 0 (13.3)
Here both and are learning parameters, reflecting the elasticities of learn-
ing w.r.t. the technology level and labor hours, respectively. The higher the
number of human beings involved in production and the more time they
spend in production, the more experience is accumulated. Sub-optimal in-
gredients in the production processes are identified and eliminated. The
experience and knowledge arising in one firm or one sector is speedily dif-
fused to other firms and other sectors in the economy (knowledge spillovers
or learning by watching), and as a result the aggregate productivity level is
increased.2
1
In his Chapter 20, Section 20.4, on industrialization and structural change Acemoglu
considers a model with two sectors, an agrarian and a manufacturing sector, where in the
latter learning by doing in the form (13.2) with = 1 plays an important role.
2
Diffusion of proficiency also occurs via apprentice-master relationships.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
13.1. Learning by doing, learning by using, learning by watching 207
Since hours spent, is perhaps a better indicator for “new experience”
than output, specification (13.3) may seem more appealing than specifi-
cation (13.2). So this section concentrates on (13.3).
If the labor force is growing should be assumed strictly less than one,
because with = 1 there would be a built-in tendency to forever faster
growth, which does not seem plausible. In fact, 0 can not be ruled
out; that would reflect that learning becomes more and more difficult (“the
easiest ideas are found first”). On the other hand, the case of “standing on
the shoulders” is also possible, that is, the case 0 ≤ 1, which is the case
where new learning becomes easier, the more is learnt already.
In “very-long-run” growth theory concerned with human development in
an economic history perspective, the in (13.3) has been replaced simply
by the size of population in the relevant region (which may be considerably
larger than a single country). This is the “population breeds ideas” view,
cf. Kremer (1993). Anyway, many simple models consider the labor force
to be proportional to population size, and then it does not matter whether
we use the learning-by-doing interpretation or the population-breeds-ideas
interpretation.
The so-called Horndal effect (reported by Lundberg, 1961) was one of
the empirical observations motivating the learning-by-doing idea in growth
theory:
“The Horndal-iron works in Sweden had no new investment (and therefore
presumably no significant change in its methods of production) for a
period of 15 years, yet productivity (output per man-hour) rose on the
average close to 2 % per annum. We find again steadily increasing
performance which can only be imputed to learning from experience”
(here cited after Arrow, 1962).
Similar patterns of on-the-job productivity improvements have been ob-
served in ship-building, airframe construction, and chemical industries. On
the other hand, within a single production line there seems to be a tendency
for this kind of productivity increases to gradually peter out, which suggests
0 in (13.3). We may call this phenomenon “diminishing returns in the
learning process”: the potential for new learning gradually evens out as more
and more learning has already taken place. But new products are continu-
ously invented and the accumulated knowledge is transmitted, more or less,
to the production of these new products that start on a “new learning curve”,
along which there is initially “a large amount to be learned”.3 This combi-
3
A learning curve is a graph of estimated productivity (or its inverse, cf. Fig. 13.1 or
Fig. 13.2 below) as a function of cumulative output or of time passed since production of
the new product began at some plant.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
CHAPTER 13. PERSPECTIVES ON LEARNING BY DOING
208 AND LEARNING BY INVESTING
nation of qualitative innovation and continuous productivity improvement
through learning may at the aggregate level end up in a ≥ 0 in (13.3).
In any case, whatever the sign of at the aggregate level, with 1
this model is capable of generating sustained endogenous per capita growth
(without “growth explosion”) if the labor force is growing at a rate
0. Indeed, as in Chapter 12, there are two cases that are consistent with
a balanced growth path (BGP for short) with positive per capita growth,
namely the case 1 combined with 0 and the case = 1 combined
with = 0
We will show this for a closed economy with = 0 ≥ 0 and with
capital accumulation according to
̇ = − = − − 0 0 given (13.4)
13.1.1 The case: 1 in (13.3)
Let us first consider the growth rate of ≡ along a BGP. There are two
steps in the calculation of this growth rate.
Step 1. Given (13.4), from basic balanced growth theory (Chapter 4) we
know that along a BGP with positive gross saving, not only are, by definition,
and constant, but they are also the same, so that is constant
over time. Owing to the CRS assumption, (13.1) implies that
1 = ( ) (13.5)
When is constant, ≡ must be constant, whereby
= = − (13.6)
a constant.
Step 2. Dividing through by in (13.3), we get
̇
≡ = −1
Taking logs gives log = log + ( − 1) log + log And taking the time
derivative on both sides of this equation leads to
̇
= ( − 1) + (13.7)
In view of being constant along a BGP, we have ̇ = 0 and so (13.7)
gives
=
1−
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
13.1. Learning by doing, learning by using, learning by watching 209
presupposing 1 Hence, by (13.6),
=
1−
Under the assumption that 0 this per capita growth rate is positive,
whatever the sign of . Given the growth rate is an increasing function
of both learning parameters. Since a positive per capita growth rate can in
the long run be maintained only if supported by 0 this is an example of
semi-endogenous growth (as long as is exogenous).
This model thus gives growth results somewhat similar to the results
in Arrow’s learning-by-investing model, cf. Chapter 12. In both models the
learning is an unintended by-product of the work process and construction of
investment goods, respectively. And both models assume that knowledge is
non-appropriable (non-exclusive) and that knowledge spillovers across firms
are fast (in the time perspective of growth theory). So there are positive
externalities which may motivate government intervention.
Methodological remark: Different approaches to the calculation
of long-run growth rates Within this semi-endogenous growth case, de-
pending on the situation, different approaches to the calculation of long-run
growth rates may be available. In Chapter 12, in the analysis of the Arrow
case 1 the point of departure in the calculation was the steady state
property of Arrow’s model that ̃ ≡ ( ) is a constant. But this point of
departure presupposes that we have established a well-defined steady state
in the sense of a stationary point of a complete dynamic system (which in
the Arrow model consists of two first-order differential equations in ̃ and ̃
respectively), usually involving also a description of the household sector.
In the present case we are not in this situation because we have not
specified how the saving in (13.4) is determined. This explains why above
(as well as in Chapter 10) we have taken another approach to the calculation
of the long-run growth rate. We simply assume balanced growth and ask
what the growth rate must then be. If the technologies in the economy
are such that per capita growth in the long run can only be due to either
exogenous productivity growth or semi-endogenous productivity growth, this
approach is usually sufficient to determine a unique growth rate.
Note also, however, that this latter feature is in itself an interesting and
useful result (as exemplified in Chapter 10). It tells us what the growth
rate must be in the long run provided that the system converges to balanced
growth. The growth rate will be the same, independently of the market
structure and the specification of the household sector, that is, it will be the
same whether, for example, there is a Ramsey-style household sector or an
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
CHAPTER 13. PERSPECTIVES ON LEARNING BY DOING
210 AND LEARNING BY INVESTING
overlapping generations set-up.4 And at least in the first case the growth
rate will be the same whatever the size of the preference parameters (the
rate of time preference and the elasticity of marginal utility of consumption).
Moreover, only if economic policy affects the learning parameters or the pop-
ulation growth rate (two things that are often ruled out inherently by the
setup), will the long-run growth rate be affected. Still, economic policy can
temporarily affect economic growth and in this way affect the level of the
long-run growth path.
13.1.2 The case = 1 in (13.3)
With = 1 in (13.3), the above growth rate formulas are no longer valid.
But returning to (13.3), we have = . Then, unless = 0 the growth
rate of will tend to rise forever, since we have = 0 → ∞ for
0.
So we will assume = 0 Then = 0 for all implying = 0 for
all . Since both and 0 are exogenous, it is as if the rate of technical
progress, were exogenous. Yet, technical progress is generated by an
internal mechanism. If the government by economic policy could affect or
0 also would be affected. In any case, under balanced growth, (13.5)
holds again and so = must be constant. This implies =
= 0 0 Consequently, positive per capita growth can be maintained
forever without support of growth in any exogenous factor, that is, growth
is fully endogenous.
As in the semi-endogenous growth case we can here determine the growth
rate along a BGP independently of how the household sector is described.
And preference parameters do not affect the growth rate. The fact that this
is so even in the fully-endogenous growth case is due to the “law of motion”
of technology making up a subsystem that is independent of the remainder of
the economic system. This is a special feature of the “growth engine” (13.3).
Although it is not a typical ingredient of endogenous growth models, this
growth engine can not be ruled out apriori. The simple alternative, (13.2), is
very different in that the endogenous aggregate output, , is involved. We
return to (13.2) in Section 13.4 below.
Before proceeding, a brief remark on the explosive case 1 in (13.2)
or (13.3) is in place. If we imagine 1 growth becomes explosive in
the extreme sense that output as well as productivity, hence also per capita
consumption, will tend to infinity in finite time. This is so even if = 0
4
Specification of these things is needed if we want to study the transitional dynamics:
the adjustment processes outside balanced growth/steady state, including the question of
convergence to balanced growth/steady state.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
13.2. Disembodied learning by investing 211
The argument is based on the mathematical fact that, given a differential
equation ̇ = where 1 and 0 0 the solution has the property
that there exists a 1 0 such that → ∞ for → 1 . For details, see
Appendix B.
13.2 Disembodied learning by investing
In the above framework the work process is a source of learning whether it
takes place in the consumption or capital goods sector. This is learning by
doing in a broad sense. If the source of learning is specifically associated
with the construction of capital goods, the learning by doing is often said to
be of the form of learning by investing. Why in the headline of this section
we have added the qualification “disembodied”, will be made clear in Section
13.3. Another name for learning by investing is investment-specific learning
by doing.
The prevalent view in the empirical literature seems to be that learning
by investing is the most important form of learning by doing; ship-building
and airframe construction are prominent examples. To the extent that the
construction of capital equipment is based on more complex and involved
technologies than is the production of consumer goods, we are also, intu-
itively, inclined to expect that the greatest potential for productivity in-
creases through learning is in the investment goods sector.5
In the simplest version of the learning-by-investment hypothesis, (13.3)
above is replaced by
µZ ¶
= = 0 ≤ 1 (13.8)
−∞
where is aggregate net investment. This is the hypothesis that the economy-
wide technology level is an increasing function of society’s previous ex-
perience, proxied by cumulative aggregate net investment.6 The Arrow and
Romer models, as described in Chapter 12, correspond to the cases 0 1
and = 1 respectively.
In this framework, where the “growth engine” depends on capital accu-
mulation, it is only in the Arrow case that we can calculate the per-capita
5
After the information-and-communication technology (ICT) revolution, where a lot of
technically advanced consumer goods have entered the scene, this traditional presumption
may be less compelling.
6
Contrary to the dynamic learning-by-doing specification (13.3), there is here no good
reason for allowing 0
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
CHAPTER 13. PERSPECTIVES ON LEARNING BY DOING
212 AND LEARNING BY INVESTING
growth rate along a BGP without specifying anything about the household
sector.
13.2.1 The Arrow case: 1 and ≥ 0
We may apply the same two steps as in Section 13.1.1. Step 1 is then an
exact replication of step 1 above. Step 2 turns out to be even simpler than
above, because (13.8) immediately gives log = log so that = ,
which substituted into (13.6) yields
= = = − = −
From this follows, first,
= (13.9)
1−
and, second,
=
1−
Alternatively, we may in this case condense the two steps into one by
rewriting (13.5) in the form
= (1 ) = (1 −1 )
by (13.8). Along the BGP, since is constant, so must the second argu-
ment, −1 , be. It follows that
( − 1) + = 0
thus confirming (13.9).
Whatever the approach to the calculation, the per capita growth rate is
here tied down by the size of the learning parameter and the growth rate of
the labor force.
13.2.2 The Romer case: = 1 and = 0
In the Romer case, however, the growth rate along a BGP cannot be de-
termined until the saving behavior in the economy is modeled. Indeed, the
knife-edge case = 1 opens up for many different per capita growth rates
under balanced growth. Which one is “selected” by the economy depends on
how the household sector is described.
For a Ramsey setup with = 0 the last part of Chapter 12 showed how the
growth rate generated by the economy depends on the rate of time preference
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
13.2. Disembodied learning by investing 213
Figure 13.1: Man-hours per vessel against cumulative number of vessels completed
to date in shipyard 1 and shipyard 2, respectively. Log-log paper. Source: Searle
(1945).
and the elasticity of marginal utility of consumption of the representative
household. Growth is here fully-endogenous in the sense that a positive per
capita growth rate can be maintained forever without the support by growth
in any exogenous factor. Moreover, according to this model, economic policy
that internalizes the positive externality in the system can raise not only the
productivity level, but also the long-run productivity growth rate.
13.2.3 The size of the learning parameter
What is from an empirical point of view a plausible value for the learning
parameter, ? This question is important because quite different results
emerge depending on whether is close to 1 or considerably lower (fully-
endogenous growth versus semi-endogenous growth). At the same time the
question is not easy to answer because in the models is a parameter that is
meant to reflect the aggregate effect of the learning going on in single firms
and spreading across firms and industries.
Like Lucas (1993), we will consider the empirical studies of on-the-job
productivity increases in ship-building by Searle (1945) and Rapping (1965).
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
CHAPTER 13. PERSPECTIVES ON LEARNING BY DOING
214 AND LEARNING BY INVESTING
Figure 13.2: Average man-hours (over ten shipyards) per vessel gainst calendar
time. Four different vessel types. Source: Searle (1945).
Both studies used data on the production of different types of cargo vessels
during the second world war. Figures 1 and 2 are taken from Lucas’ review
article, Lucas (1993), but the original source is Searle (1945). For the vessel
type called “Liberty Ships” Lucas cites the observation by Searle (1945):
“the reduction in man-hours per ship with each doubling of cumulative
output ranged from 12 to 24 percent.”
Let us try to connect this observation to the learning parameter in
Arrow’s and Romer’s framework. We begin by considering firm which
operates in the investment goods sector. We imagine that firm ’s equipment
is unchanged during the observation period (as is understood in the above
citation as well as the citation from Arrow (1962) in Section 13.1). Let firm
’s current output and employment be and respectively. The current
labor productivity is then = Let the firm’s cumulative output
be denoted This cumulative output is a part of cumulative investment
in society. At the micro-level the learning-by-investing hypothesis is the
hypothesis that labor productivity is an increasing function of the firm’s
cumulative output,
In figures 1 and 2 the dependent variable is not directly labor productivity,
but its inverse, namely the required man-hours per unit of output, =
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
13.2. Disembodied learning by investing 215
= 1 Figure 13.1 suggests a log-linear relationship between this
variable and the cumulative output:
log = − log (13.10)
That is, as cumulative output rises, the required man-hours per unit of output
declines over time in this way:
=
Equivalently, labor productivity rises over time in this way:
1
= = −
So, specifying the relationship by a power function, as in (13.8), makes sense.
Now, let = 1 be a fixed point in time. Then, (13.10) becomes
log 1 = − log 1
Let 2 be the later point in time where cumulative output has been doubled.
Then at time 2 the required man-hours per unit of output has declined to
log 2 = − log 2 = − log(21 )
Hence,
log 1 − log 2 = − log 1 + log(21 ) = log 2 (13.11)
Lucas’ citation above from Searle amounts to a claim that
1 − 2
012 024 (13.12)
1
By a first-order Taylor approximation we have log 2 ≈ log 1 + (2 −
1 )1 . Hence, (1 − 2 )1 ≈ log 1 − log 2 Substituting this
into (13.12) gives, approximately,
012 log 1 − log 2 024
Combining this with (13.11) gives 012 log 2 024 so that
012 024
017 = = 035
log 2 log 2
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
CHAPTER 13. PERSPECTIVES ON LEARNING BY DOING
216 AND LEARNING BY INVESTING
Rapping (1965) finds by a more rigorous econometric approach to be
in the vicinity of 0.26 (still ship building). Arrow (1962) and Solow (1997)
refer to data on airframe building. This data roughly suggests = 13
How can this be translated into a guess on the “aggregate” learning pa-
rameter in (13.8)? This is a complicated question and the subsequent
remarks are very tentative. First of all, the potential for both internal and
external learning seems to vary a lot across different industries. Second, the
amount of spillovers can not simply be added to the above, since they are
already partly included in the estimate of Even theoretically, the role of
experience in different industries cannot simply be added up because to some
extent there is redundancy due to overlapping experience and sometimes the
learning in other industries is of limited relevance. Given that we are inter-
ested in an upper bound for a “guestimate” is that the spillovers matter
for the final at most the same as from ship building so that ≤ 27
On the basis of these casual considerations we claim that a much higher
than about 23 may be considered fairly implausible. This speaks for the
Arrow case of semi-endogenous growth rather than the Romer case of fully-
endogenous growth, at least as long as we think of learning by investing
as the sole source of productivity growth. Another point is that to the
extent learning is internal and at least temporarily appropriable, we should
expect at least some firms to internalize the phenomenon in its optimizing
behavior (Thornton and Thompson, 2001). Although the learning is far from
fully excludable, it takes time for others to discover and imitate technical
and organizational improvements. Many simple growth models ignore this
and treat all learning by doing and learning by investing as a 100 percent
externality, which seems an exaggeration.
A further issue is to what extent learning by investing takes the form of
disembodied versus embodied technical change. This is the topic of the next
section.
7
For more elaborate studies of empirical aspects of learning by doing and learning by
investing, see Irwin and Klenow (1994), Jovanovic and Nyarko (1995), and Greenwood and
Jovanovic (2001). Caballero and Lyons (1992) find clear evidence of positive externalities
across US manufacturing industries. Studies finding that the quantitative importance of
spillovers is significantly smaller than required by the Romer case include Englander and
Mittelstadt (1988) and Benhabib and Jovanovic (1991). See also the surveys by Syverson
(2011) and Thompson (2012).
Although in this lecture note we focus on learning as an externality, there exists studies
focusing on internal learning by doing, see, e.g., Gunn and Johri, 2011.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
13.3. Disembodied vs. embodied technical change 217
13.3 Disembodied vs. embodied technical change
Arrow’s and Romer’s models build on the idea that the source of learning
is primarily experience in the investment goods sector. Both models assume
that the learning, via knowledge spillovers across firms, provides an engine
of productivity growth in essentially all sectors of the economy. And both
models (Arrow’s, however, only in its simplified version, which we considered
in Chapter 12, not in its original version) assume that a firm can benefit from
recent technical advances irrespective of whether it buys new equipment or
just uses old equipment. That is, the models assume that technical change
is disembodied.
13.3.1 Disembodied technical change
Disembodied technical change occurs when new technical knowledge advances
the combined productivity of capital and labor independently of whether the
workers operate old or new machines. Consider again (13.1) and (13.3).
When the appearing in (13.1) refers to the total, historically accumu-
lated capital stock, then the interpretation is that the higher technology
level generated in (13.3) or (13.8) results in higher productivity of all labor,
independently of the vintage of the capital equipment with which this labor
is combined. Thus also firms with old capital equipment benefit from re-
cent advances in technical knowledge. No new investment is needed to take
advantage of the recent technological and organizational developments.
Examples of this kind of productivity increases include improvement in
management and work practices/organization and improvement in account-
ing.
13.3.2 Embodied technical change
In contrast, we say that technical change is embodied, if taking advantage of
new technical knowledge requires construction of new investment goods. The
newest technology is incorporated in the design of newly produced equipment;
and this equipment will not participate in subsequent technical progress. An
example: only the most recent vintage of a computer series incorporates the
most recent advance in information technology. Then investment goods pro-
duced later (investment goods of a later “vintage”) have higher productivity
than investment goods produced earlier at the same resource cost. Whatever
the source of new technical knowledge, investment becomes an important
bearer of the productivity increases which this new knowledge makes pos-
sible. Without new investment, the potential productivity increases remain
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
CHAPTER 13. PERSPECTIVES ON LEARNING BY DOING
218 AND LEARNING BY INVESTING
potential instead of being realized.8
One way to formally represent embodied technical progress is to write
capital accumulation in the following way,
̇ = − (13.13)
where is gross investment at time and measures the “quality” (produc-
tivity) of newly produced investment goods. The rising level of technology
implies rising so that a given level of investment gives rise to a greater and
greater addition to the capital stock, measured in efficiency units. Even
if technical change does not directly appear in the production function, that
is, even if for instance (13.1) is replaced by = ( ) the economy may
in this manner still experience a rising standard of living.
Figure 13.3: Relative price of equipment and quality-adjusted equipment
investment-to-GNP ratio. Source: Greenwood, Hercowitz, and Krusell (1997).
Embodied technical progress is likely to result in a steady decline in the
price of capital equipment relative to the price of consumption goods. This
prediction is confirmed by the data. Greenwood et al. (1997) find for the
U.S. that the relative price, of capital equipment has been declining at
8
The concept of embodied technical change was introduced by Solow (1960). The notion
of Solow-neutral technical change is related to embodied technical change and capital of
different vintages.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
13.3. Disembodied vs. embodied technical change 219
an average rate of 003 per year in the period 1950-1990, cf. the “Price”
curve in Figure 13.3.9 As the “Quantity” curve in Figure 13.3 shows, over
the same period there has been a secular rise in the ratio of new equipment
investment (in efficiency units) to GNP; note that what in the figure is called
the “investment-to-GNP Ratio” is really “quality-adjusted investment-to-
GNP Ratio”, not the usual investment-income ratio, .
Moreover, the correlation between de-trended and de-trended
is −046 Greenwood et al. interpret this as evidence that technical advances
have made equipment less expensive, triggering increases in the accumulation
of equipment both in the short and the long run. The authors also estimate
that embodied technical change explains 60% of the growth in output per
man hour.
13.3.3 Embodied technical change and learning by in-
vesting
Whether technological progress is disembodied or embodied says nothing
about whether its source is exogenous or endogenous. Indeed, the increases
of in (13.13) may be modeled as exogenous or endogenous. In the latter
case, a popular hypothesis is that the source is learning by investing. This
learning may take the form (13.8) above. In that case the experience that
matter for learning is cumulative net investment.
An alternative hypothesis is:
µZ ¶
= 0 ≤ ̄ (13.14)
−∞
where is gross investment at time Here the experience that matter has
its basis in cumulative gross investment. An upper bound, ̄ for the learning
parameter is introduced to avoid explosive growth. The hypothesis (13.14)
seems closer to both intuition and the original ideas of Arrow:
“Each new machine produced and put into use is capable of
changing the environment in which production takes place, so
that learning is taking place with continually new stimuli” (Ar-
row, 1962).
Contrary to the integral based on net investment in (13.8), the integral
in the learning hypothesis (13.14) does not allow an immediate translation
9
The relative price index in Fig. 13.3 is based on the book by R. Gordon (1990), which
is an attempt to correct previous price indices for equipment by better taking into account
quality improvements in new equipment.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
CHAPTER 13. PERSPECTIVES ON LEARNING BY DOING
220 AND LEARNING BY INVESTING
into an expression in terms of the accumulated capital stock. Instead a new
state variable, cumulative gross investment, enter the system and opens up
for richer dynamics.
We may combine (13.14) with an aggregate Cobb-Douglas production
function,
= 1−
(13.15)
Then the upper bound for the learning parameter in (13.14) is ̄ = (1 −
).10
The case (1 − )
Suppose (1 − ) Using (13.14) together with (13.13), (16.21), and
= − one finds under balanced growth with = constant and
0 1
(1 − )(1 + )
= (13.16)
1 − (1 + )
= (13.17)
1+
1
= (13.18)
1+
= = − = (13.19)
1 − (1 + )
cf. Appendix A. We see that 0 if and only if 0 So growth is here
semi-endogenous.
Let us assume there is perfect competition in all markets. Since capital
goods can be produced at the same minimum cost as one consumption good,
the equilibrium price, of capital goods in terms of the consumption good
must equal the inverse of that is, = 1 With the consumption good
being the numeraire, let the rental rate in the market for capital services be
denoted and the real interest rate in the market for loans be denoted
10
An alternative to the specification of embodied learning by gross investment in (13.14)
is
µZ ¶̃ −
= 0 ̃ ≤ ̃
−∞
implying that it is cumulative quality-adjusted gross investment that matters, cf. Green-
wood and Jovanovic (2001). If combined with the production function (16.21) the appro-
−
priate upper bound on the learning parameter, ̃ is ̃ = 1 −
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
13.3. Disembodied vs. embodied technical change 221
Ignoring uncertainty, we have the no-arbitrage condition
− ( − ̇ )
= (13.20)
where − ̇ is the true economic depreciation of the capital good per time
unit. Since = 1 (13.17) and (13.16) indicate that along a BGP the
relative price of capital goods will be declining according to
(1 − )
= − 0
1 − (1 + )
Note that along the BGP. Is this a violation of Proposition 1 of
Chapter 4? No, that proposition presupposes that capital accumulation oc-
curs according to the standard equation (13.4), not (13.13). And although
differs from the output-capital ratio in value terms, () is constant
along the BGP. In fact, the BGP complies entirely with Kaldor’s stylized
facts if we interpret “capital” as the value of capital, .
The formulas (13.16) and (13.19) display that (1 + ) 1 is needed
to avoid a forever rising growth rate if 0. This inequality is equivalent
to (1 − ) and confirms that the upper bound, ̄ in (13.14) equals
(1 − ) With = 13 this upper bound is 2 The bound is thus no longer
1 as in the simple learning-by-investing model of Section 13.2. The reason is
twofold, namely partly that now is formed via cumulative gross investment
instead of net investment, partly that the role of is to strengthen capital
formation rather than the efficiency of production factors in aggregate final
goods produce.
When = 0 the system can no longer generate a constant positive per
capita growth rate (exponential growth). Groth et al. (2010) show, however,
that the system is capable of generating quasi-arithmetic growth. This class of
growth processes, which fill the whole range between exponential growth and
complete stagnation, was briefly commented on in Section 10.5 of Chapter
10.
∗
The case = (1 − ) and = 0
When = (1 − ) we have (1 + ) = 1 and so the growth formulas
(13.16) and (13.19) no longer hold. But the way that (13.17) and (13.18)
are derived (see Appendix A) ensures that these two equations remain valid
along a BGP. Given = (1 − ) (13.17) can be written = (1 − )
which is equivalent to
= 1−
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
CHAPTER 13. PERSPECTIVES ON LEARNING BY DOING
222 AND LEARNING BY INVESTING
along a BGP ( is some positive constant to be determined).
To see whether a BGP exists, note that (13.14) implies
̇ −1 −(1−)
= = = = −(1−) − = −(1−) −
considering a BGP with = constant. Substituting (16.21) into this,
we get
= −(1−) − 1− = −(1−) 1− (13.21)
If = 0, the right-hand side of (13.21) is constant and so is = (1 − )
by (13.17), and = = (1 − ) by (13.19)
If 0 at the same time as = (1 − ) however, there is a tendency
to a forever rising growth rate in , hence also in and . No BGP exists
in this case.
Returning to the case where a BGP exists, a striking feature revealed by
(13.21) is that the saving rate, matters for the growth rate of hence also
for the growth rate of and respectively, along a BGP. As in the Romer
case of the disembodied learning-by-investing model, the growth rates along
a BGP cannot be determined until the saving behavior in the economy is
modeled.
So the considered knife-edge case, = (1 − ) combined with = 0
opens up for many different per capita growth rates under balanced growth.
Which one is “selected” by the economy depends on how the household sec-
tor is described. In a Ramsey setup with = 0 one can show that the
growth rate under balanced growth depends negatively on the rate of time
preference and the elasticity of marginal utility of consumption of the repre-
sentative household. And not only is growth in this case fully endogenous in
the sense that a positive per capita growth rate can be maintained forever
without the support by growth in any exogenous factor. An economic pol-
icy that subsidizes investment can generate not only a transitory rise in the
productivity growth rate, but also a permanently higher productivity growth
rate.
In contrast to the Romer (1986) model, cf. Section 13.2.2 above, we do
not here end up with a reduced-form AK model. Indeed, we end up with a
model with transitional dynamics, as a consequence of the presence of two
state variables, and
If instead 1(1 + ) we get a tendency to explosive growth − infinite
output in finite time − a not plausible scenario, cf. Appendix B.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
13.4. Static comparative advantage vs. dynamics of learning by doing* 223
13.4 Static comparative advantage vs. dy-
namics of learning by doing*
In this section we will briefly discuss a development economics perspective
of the above learning-based growth models.
More specifically we will take a look at the possible “conflict” between
static comparative advantage and economic growth. The background to this
possible “conflict” is the dynamic externalities inherent in learning by doing
and learning by investing.11
13.4.1 A simple two-sector learning-by-doing model
We consider an isolated economy with two production sectors, sector 1 and
sector 2, each producing its specific consumption good. Labor is the only
input and aggregate labor supply is constant. There are many small firms
in the two sectors. Aggregate output in the sectors are:
1 = 1 1 (13.22)
2 = 2 2 (13.23)
where
1 + 2 =
There are sector-specific learning-by-doing externalities in the following form:
̇1 = 1 1 1 ≥ 0 (13.24)
̇2 = 2 2 2 ≥ 0 (13.25)
Although not visible in our aggregate formulation, there are substantial
knowledge spillovers across firms within the sectors. Across sectors, spillovers
are assumed negligible.
Assume firms maximize profits and that there is perfect competition in the
goods and labor markets. Then, prices are equal to the (constant) marginal
costs. Let the relative price of sector 2-goods in terms of sector-1 goods be
called (i.e., we use sector-1 goods as numeraire). Let the hourly wage in
terms of sector-1 goods be In general equilibrium with production in both
sectors we then have
1 = 2 =
11
Krugman (1987), Lucas (1988, Section 5).
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
CHAPTER 13. PERSPECTIVES ON LEARNING BY DOING
224 AND LEARNING BY INVESTING
saying that the value of the (constant) marginal productivity of labor in each
sector equals the wage. Hence,
2 1
= 1 or = (13.26)
1 2
saying that the relative price of the two goods is inversely proportional to
the relative labor productivities in the two sectors. The demand side, which
is not modelled here, will of course play a role for the final allocation of labor
to the two sectors.
Taking logs in (13.26) and differentiating w.r.t. gives
̇ ̇1 ̇2 1 1 2 2
= − = − = 1 1 − 2 2
1 2 1 2
using (13.24) and (13.25). Thus,
̇ = (1 1 − 2 2 )
Assume sector 2 (say some industrial activity) is more disposed to learning-
by-doing than sector 1 (say mining) so that 2 1 Consider for simplicity
the case where at time 0 there is symmetry in the sense that 10 = 20
Then, the relative price of sector-2 goods in terms of sector-1 goods will,
at least initially, tend to diminish over time. The resulting substitution ef-
fect is likely to stimulate demand for sector-2 goods. Suppose this effect is
large enough to ensure that 2 = 2 2 never becomes lower than 1 1 2
that is, 2 2 ≥ 1 1 for all Then the scenario with ̇ ≤ 0 is sustained
over time and the sector with highest growth potential remains a substantial
constituent of the economy. This implies sustained economic growth in the
aggregate economy.
Now, suppose the country considered is a rather backward, developing
country which until time 0 has been a closed economy (very high tariffs etc.).
Then the country decides to open up for free foreign trade. Let the relative
world market price of sector 2-goods be ̄ which we for simplicity assume is
constant At time 0 there are two alternative possibilities to consider:
0
Case 1: ̄ 120
(world-market price of good 2 higher than the opportu-
nity cost of producing good 2). Then the country specializes fully in sector-2
goods. Since this is the sector with a high growth potential, economic growth
is stimulated. The relative productivity level 1 2 decreases so that the
scenario with ̄ 1 2 remains. A virtuous circle of dynamics of learning
by doing is unfolded and high economic growth is sustained.
0
Case 2: ̄ 120
(world-market price of good 2 lower than the opportunity
cost of producing good 2). Then the country specializes fully in sector-1
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
13.4. Static comparative advantage vs. dynamics of learning by doing* 225
goods. Since this is the sector with a low growth potential, economic growth
is impeded or completely halted. The relative productivity level 1 2 does
not decrease. Hence, the scenario with ̄ 1 2 sustains itself and persists.
Low or zero economic growth is sustained. The static comparative advantage
in sector-1 goods remains and the country is locked in low growth.
·
If instead ̄ is time-dependent, suppose ̄ 0 (by similar arguments as
for the closed economy). Then the case 2 scenario is again self-sustaining.
The point is that there may be circumstances (like in case 2), where
temporary protection for a backward country is growth promoting (this is a
specific kind of “infant industry” argument).
13.4.2 A more robust specification
The way (13.24) and (13.25) are formulated, we have
̇1
= 1 1 (13.27)
1
̇2
= 2 2 (13.28)
2
by (13.22) and (13.23). Thus, the model implies scale effects on growth, that
is, strong scale effects.
An alternative specification introduces limits to learning-by-doing in the
following way:
̇1 = 1 11 1 1
̇2 = 2 22 2 1
Then (13.27) and (13.28) are replaced by
̇1
= 1 11 −1 11 (13.29)
1
̇2
= 2 22 −1 22 (13.30)
2
Now the problematic strong scale effect has disappeared. At the same time,
since 1 − 1 0 and 2 − 1 0 (13.29) and (13.30) show that growth peters
out as long as the “diminishing returns” to learning-by-doing are not offset
by an increasing labor force or an additional source (outside the model) of
technical progress. If 0 we get sustained growth of the semi-endogenous
type as in the Arrow model of learning-by-investing.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
CHAPTER 13. PERSPECTIVES ON LEARNING BY DOING
226 AND LEARNING BY INVESTING
Yet the analysis may still be a basis for an “infant industry” argument. If
the circumstances are like in case 2, temporary protection may help a back-
ward country to enter a higher long-run path of evolution. Stiglitz underlines
South Korea as an example:
What matters is dynamic comparative advantage, or comparative
advantage in the long run, which can be shaped. Forty years ago,
South Korea had a comparative advantage in growing rice. Had
it stuck to that strength, it would not be the industrial giant that
it is today. It might be the world’s most efficient rice grower, but
it would still be poor (Stiglitz, 2012, p. 2).
This point is related to two different aspects of technical knowledge. On
the one hand, technical knowledge is a nonrival good and this non-rivalness
speaks for openness, thereby improving conditions for knowledge spillovers
and learning from other countries. On the other hand, the potential for
knowledge accumulation and internal learning by doing is different in dif-
ferent production sectors. And some sectors with a lot of internal learning
potential and economies of scale never gets started unless to begin with they
are protected from foreign competition.
13.4.3 Resource curse?
The analysis also suggests a mechanism that, along with others, may help
explaining what is known as the resource curse problem. This problem refers
to the paradox that being abundant in natural resources may sometimes seem
a curse for a country rather than a blessing. At least quite many empirical
studies have shown a negative correlation between resource abundance and
economic growth (see, e.g., Sachs and Warner 1995, Gylfason et al., 1999).
The mechanism behind this phenomenon could be the following. Consider
a mining country with an abundance of natural resources in the ground.
Empirically, growth in total factor productivity in mining activity is relatively
low. Interpreting this as reflecting a relatively low learning potential, the
mining sector may be represented by sector 1 above. Given the abundance
of natural resources, 10 is likely to be high relative to the productivity in
the manufacturing sector, 20 So the country is likely to be in the situation
described as case 2. As a result, economic growth may never get started.
The basic problem here is, however, not of an economic nature in a narrow
sense, but rather of an institutional character. Taxation on the natural re-
source and use of the tax revenue for public investment in growth promoting
factors (infrastructure, health care, education, R&D) or directly in the sector
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
13.5. Robustness issues and scale effects 227
with high learning potential can from an economic point of view circumvent
the curse to a blessing. It is not the natural resources as such, but rather
barriers of a political character, conflicts of interest among groups and social
classes, even civil war over the right to exploit the resources, or dominance
by foreign superpowers, that may be the obstacles to a sound economic de-
velopment (Mehlum et al., 2002). An additional potential obstacle is related
to the possible response of a country’s real exchange rate, and therefore its
competitiveness, to a new discovery of natural resources in a country.12
Summing up: Discovery of a valuable mineral in the ground in a country
with weak institutions may, through corruption etc. have adverse effects on
resource allocation and economic growth in the country. But: “Resources
should be a blessing, not a curse. They can be, but it will not happen on its
own. And it will not happen easily” (Stiglitz, 2012, p. 2).
13.5 Robustness issues and scale effects
First some words about terminology.
13.5.1 On terminology
How terms like “endogenous growth” and “semi-endogenous growth” are de-
fined varies in the literature. Recalling the notation ≡ and ≡ ̇
in this course we use the definitions:
Endogenous growth is present if there is a positive long-run per capita
growth rate (i.e., 0) and the source of this is some internal mecha-
nism in the model (so that exogenous technology growth is not needed).
Fully-endogenous growth (sometimes called strictly endogenous growth) is
present if there is a positive long-run per capita growth rate and this
occurs without the support by growth in any exogenous factor (for
example exogenous growth in the labor force).
An example: the Romer version of the model of learning by investing
features fully endogenous growth. The technical reason for this is the as-
sumption that the learning parameter, is such that there are constant
returns to capital at the aggregate level. We get 0 constant, and, in a
Ramsey set-up, results like 0 and 0, that is, preference
12
Ploeg (2011) provides a survey over different theories related to the resource curse
problem. See also Ploeg and Venables (2012) and Stiglitz (2012).
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
CHAPTER 13. PERSPECTIVES ON LEARNING BY DOING
228 AND LEARNING BY INVESTING
parameters matter for long-run growth. This suggests, at least at the theoret-
ical level, that taxes and subsidies, by affecting incentives, may have effects
on long-run growth (cf. Chapter 12). On the other hand, a fully-endogenous
growth model need not have this implication. We saw an example of this in
Section 13.1, where the “law of motion” of technology makes up a subsystem
that is independent of the remainder of the economic system.
In any case, fully-endogenous growth is technologically possible if and
only if there are non-diminishing returns (at least asymptotically) to the
producible inputs in the growth-generating sector(s), also called the “growth
engine”. The growth engine in an endogenous growth model is defined as
the set of input-producing sectors or activities using their own output as
input. This set may consist of only one sector such as the manufacturing
sector in the simple AK model, the educational sector in the Lucas (1988)
model, or the R&D sector in the Romer (1990) model. A model is capable
of generating fully-endogenous growth if the growth engine has CRS w.r.t.
producible inputs.
No argument like the replication argument for CRS w.r.t. the rival in-
puts exists regarding CRS w.r.t. the producible inputs. This is one of the
reasons that also another kind of endogenous growth is often considered in
the literature. This takes us to “semi-endogenous growth”.
Semi-endogenous growth is present if growth is endogenous but a posi-
tive long-run per capita growth rate can not be sustained without the
support by growth in some exogenous factor (for example exogenous
growth in the labor force).
For example, the Arrow model of learning by investing features semi-
endogenous growth. The technical reason for this is the assumption that
the learning parameter, is less than 1 which implies diminishing marginal
returns to capital at the aggregate level. Along a BGP we get
= = = (13.31)
1−
If and only if 0 can a positive constant be maintained forever. When
the learning mechanism is assisted by population growth, it is strong enough
to over time endogenously maintain a constant average productivity of cap-
ital. The key role of population growth derives from the fact that at the
aggregate level there are increasing returns to scale w.r.t. capital and labor.
For the increasing returns to be sufficiently exploited to generate exponen-
tial growth, population growth is needed.13 Note that in this case
13
Of course the model shifts from featuring “semi-” to featuring “fully-endogenous”
growth if the model is extended with an internal mechanism determining the population
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
13.5. Robustness issues and scale effects 229
= 0 = that is, preference parameters do not matter for long-run
growth (only for the level of the growth path). This suggests that taxes and
subsidies do not have long-run growth effects. Yet, in Arrow’s model and
similar semi-endogenous growth models economic policy can have important
long-run level effects.
Strangely enough, some textbooks (for example Barro and Sala-i-Martin,
2004) do not call much attention to the distinction between fully-endogenous
growth and semi-endogenous growth. Rather, they tend to use the term
endogenous growth as synonymous with what we here call fully-endogenous
growth. But there is certainly no reason to rule out apriori the parameter
cases corresponding to semi-endogenous growth.
In the Acemoglu textbook (Acemoglu, 2009, p. 448) “semi-endogenous
growth” is defined or characterized as endogenous growth where the long-
run per capita growth rate of the economy “does not respond to taxes or
other policies”. As an implication, endogenous growth which is not semi-
endogenous is in Acemoglu’s text implicitly defined as endogenous growth
where the long-run per capita growth rate of the economy does respond to
taxes or other policies.
We have defined the distinction between “semi-endogenous growth” and
“fully-endogenous growth” in a different way. In our terminology, this dis-
tinction does not coincide with the distinction between policy-dependent
and policy-invariant growth. Indeed, in our terminology positive per capita
growth may rest on an “exogenous source” in the sense of deriving from
exogenous technical progress and yet the long-run per capita growth rate
may be policy-dependent. In Chapter 16 we will see an example in connec-
tion with the Dasgupta-Heal-Solow-Stiglitz model, also known as the DHSS
model.
There also exist models that according to our definition feature semi-
endogenous growth and yet the long-run per capita growth rate is policy-
dependent (Cozzi, 1997; Sorger, 2010). Similarly, there exist models that
according to our definition feature fully-endogenous growth and yet the long-
run per capita growth rate is policy-invariant (Section 13.1.2 above shows
an example).
A word of warning before proceeding. The distinction between an ex-
ogenous or endogenous per capita growth rate is only meaningful within a
given meta-theoretical framework. It is always possible to make the meta-
theoretical framework so “broad” that the per capita growth rate must be
considered endogenous within this framework. From the perspective of so-
ciety as a whole we can imagine many different political and institutional
growth rate. Jones (2003) provides such a model.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
CHAPTER 13. PERSPECTIVES ON LEARNING BY DOING
230 AND LEARNING BY INVESTING
structures − as witnessed by long-run historical evolution − some of which
clearly are less conducive to economic growth than others.
13.5.2 Robustness of simple endogenous growth mod-
els
The series of learning-based growth models considered above illustrate the
fact that endogenous growth models with exogenous population typically
exist in two varieties or cases. One is the fully-endogenous growth case where
a particular value is imposed on a key parameter. This value is such that
there are constant returns (at least asymptotically) to producible inputs in
the growth engine of the economy.14 In the “corresponding” semi-endogenous
growth case, the key parameter is allowed to take any value in an open
interval. The endpoint of this interval appears as the “knife-edge” value
assumed in the fully-endogenous growth case.
Although the two varieties build on qualitatively the same mathematical
model of a certain growth mechanism (say, learning by doing or research and
development), the long-run results turn out to be very sensitive to which
of the two cases is assumed. In the fully-endogenous growth case a posi-
tive per-capita growth rate is maintained forever without support of growth
in any exogenous factor. In the semi-endogenous growth case, the growth
process needs “support” by some growing exogenous factor in order for sus-
tained growth to be possible. The established terminology is somewhat se-
ductive here. “Fully endogenous” sounds as something going much deeper
than “semi-endogenous”. But nothing of that sort should be implied. It is
just a matter of different parameter values.
As Solow (1997, pp. 7-8) emphasizes in connection with learning-by-
investing models (with constant population), the Romer case with = 1 is a
very special case, indeed an “extreme case, not something intermediate”. A
value of slightly above 1 leads to explosive growth: infinite output in finite
time even when = 0.15 And a value of slightly below 1 leads to growth
petering out in the long run even when = 0.
Whereas the strength of the semi-endogenous growth case is its theo-
retical and empirical robustness, the convenience of the fully-endogenous
14
Suppose our CRS aggregate production function is = + 1− 0
0 0 1 we have ≡ = + , where ≡ We then get = +−1
→ for → ∞ that is, the output-capital ratio converges to a positive constant when
the capital-labor ratio goes to infinity. We then say that asymptotically there are CRS
w.r.t. the producible inputs, here just In this kind of “asymptotic” AK models the
force of diminishing returns to capital ultimately becomes negligible.
15
A demonstration is in Appendix B.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
13.5. Robustness issues and scale effects 231
growth case is that it has much simpler dynamics. Then the question arises
to what extent a fully-endogenous growth model can be seen as a useful ap-
proximation to its semi-endogenous growth “counterpart”. Imagine that we
contemplate applying the fully-endogenous growth case as a basis for making
forecasts or for policy evaluation in a situation where the “true” case is the
semi-endogenous growth case. Then we would like to know: Are the impulse-
response functions generated by a shock in the fully-endogenous growth case
an acceptable approximation to those generated by the same shock in the
corresponding semi-endogenous growth case for a sufficiently long time hori-
zon to be of interest?16 The answer is “yes” if the critical parameter has a
value “close” to the knife edge value and “no” otherwise. How close it need
be, depends on circumstances. My own tentative impression is that usually
it is “closer” than what the empirical evidence warrants.
Even if a single growth-generating mechanism, like learning by doing,
does not in itself seem strong enough to generate a reduced-form AK model
(the fully-endogenous growth case), there might exist complementary factors
and mechanisms that in total could generate something close to a reduced-
form AK model. The time-series test by, for instance, Jones (1995b) and
Romero-Avila (2006), however, reject this.17
Comment on “growth petering out” when = 0 The above-mentioned
“petering out” of long-run growth in the semi-endogenous case when = 0
takes different forms in different models. When exponential growth cannot
be sustained in a model, sometimes it remains true that → ∞ for → ∞,
and sometimes instead complete stagnation results. In the present context,
where we focus on learning, it is the source of learning that matters. Sup-
pose that, as in the simple Arrow version ( 1) of learning-by-investing
in Section 13.2.1 above (and in Chapter 12), learning is associated with net
investment, then = 0 will lead to complete stagnation in the sense that
there is an upper bound on that is never transcended. The productivity-
driving factor, net investment, dries out. Even if there is an incentive to
maintain the capital stock, this does not require positive net investment and
so learning tends to stop. The productivity-driving factor, net investment,
dries out.
When learning is associated with gross investment, however, learning
continues because even when net investment is vanishing, gross investment
remains positive because there is generally an incentive to maintain the cap-
16
Obviously, the ultimate effects of the shock tend to be very different in the two models.
17
There is a longstanding discussion about this and similar time-series econometric is-
sues, see the course website under Supplementary Material.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
CHAPTER 13. PERSPECTIVES ON LEARNING BY DOING
232 AND LEARNING BY INVESTING
ital stock. Thereby sustained learning is generated. In turn, this tends to
induce more investment than needed to replace wear and tear and so cap-
ital accumulates, although at a declining rate. The diminishing marginal
returns to capital are countervailed by the rising productivity of investment
goods due to learning. We get permanent though diminishing growth, that
is, → ∞ for → ∞ at the same time as → 0 but 0 remains true.
Arithmetic growth, = 0 + with 0 is an example. More generally,
as mentioned in Section 13.3.3, quasi-arithmetic growth tends to arise.
It is similar in the learning-by-doing examples of sections 13.1 and 13.4,
where learning is simply associated with producing. Learning continues even
if the capital stock is just upheld.
Another issue is whether there exist factors that in spite of = 0 (or,
to be more precise, in spite of → 0 as projected by the United Nations
to happen within a century from now (United Nations, 2013)) may replace
the growth-supporting role of population growth under semi-endogenous pa-
rameter conditions like 1. In Section 10.5 of Chapter 10 we indicated
scepticism that human capital accumulation would be able to do that. But
both urbanization and the evolution of information and communication tech-
nologies seem likely for a long time to at least help in that direction.
13.5.3 Weak and strong scale effects
Romer’s learning-by-investing hypothesis (where the learning parameter equals
1) implies a problematic (strong) scale effect. When embedded in a Ramsey
set-up, the model generates a time path along which
1
= = = (1 (1 ) − − )
From this follows not only standard results for fully-endogenous growth mod-
els, such as
0 0
but also18
1
= 12 (1 ) 0 (13.32)
This is because in this model the rate of return, 1 (1 ) − depends (posi-
tively) on Interpreting the size (“scale”) of the economy as measured by
18
Here we use that a neoclassical production function ( ) with CRS satisfies the
“direct complementarity condition” 12 0
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
13.5. Robustness issues and scale effects 233
the size, of the labor force, we call such an effect a scale effect. To distin-
guish it from another kind of scale effect, it is useful to name it a scale effect
on growth or a strong scale effect.
Scale effects can be of a less dramatic form. In this case we speak of a
scale effect on levels or a weak scale effect. This form arises when the learning
parameter is less than 1. We thus see from (13.31) that in Arrow’s model
of learning-by-investing the steady state growth rate is independent of the
size of the economy. Consequently, in Arrow’s model there is no strong scale
effect. There is, however, a (positive) scale effect on levels in the sense that
along a steady state growth path,
0 (13.33)
0
This says the following. Suppose we consider two closed economies char-
acterized by the same parameters, including the same 19 The economies
differ only w.r.t. initial size of the labor force. Suppose both economies are
in steady state. Then, according to (13.33), the economy with the larger
labor force has, for all larger output per unit of labor. The background is
the positive relationship between the labor efficiency index, and aggregate
cumulative (net) investment,
=
which is due to learning and knowledge spillovers across firms. Thus, a given
level of per capita investment increases labor productivity more in a larger
economy (where ̇ will be larger) than in a smaller economy.
More generally, the fundamental background is that technical knowledge
is a non-rival good − its use by one firm does not (in itself) limit the amount
of knowledge available to other firms.20 In a large economic system, say an
integrated set of open economies, more people benefit from a given increase
in knowledge than in a small economic system. At the same time the per
capita cost of creating the increase in knowledge is less in the large system
than in the small system.
To prove (13.33), note that along a steady state path
≡ ̃ = ̃ ∗ = (̃∗ ) = (̃∗ ) (13.34)
where
≡ ̃ = ̃∗ = ̃∗
19
Remember that in contrast to the Romer model, Arrow’s model allows 0
20
By patent protection, secrecy, and copyright some aspects of technical knowledge are
sometimes partially and temporarily excludable, but that is another matter. Excludability
is ignored in our simple learning-by-doing and learning-by-investing models.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
CHAPTER 13. PERSPECTIVES ON LEARNING BY DOING
234 AND LEARNING BY INVESTING
Solving this equation for gives
= (̃∗ )1(1−) = (̃∗ 0 )1(1−)
Substituting this into (13.34), we get
= (̃∗ )(̃∗ 0 )(1−) (13.35)
from which follows
[(1−)]−1
= (̃∗ )(̃∗ )(1−) 0 = 0 (13.36)
0 1− 1 − 0
since ̃∗ is independent of 0 This confirms (13.33). The scale effect on
also gives scope for higher per capita consumption the higher is 0
The scale effect on levels displayed by (13.36) is increasing in the learn-
ing parameter everything else equal. When = 1 the scale effect is so
powerful that it is transformed into a scale effect on the growth rate.
13.5.4 Discussion
Are there good theoretical and/or empirical reasons to believe in the existence
of (positive) scale effects on levels or perhaps even on growth in the long run?
Let us start with some theoretical considerations.
Theoretical aspects
From the point of view of theory, we should recognize the likelihood that
offsetting forces are in play. On the one hand, there is the problem of limited
natural resources. For a given level of technology, if there are CRS w.r.t.
capital, labor, and land (or other natural resources), there are diminishing
returns to capital and labor taken together. In this Malthusian perspective,
an increased scale (increased population) results, everything else equal, in
lower rather than higher per capita output, that is, a negative scale effect
should be expected.
On the other hand, there is the anti-Mathusian view that repeated im-
provements in technology tend to overcome, or rather more than overcome,
this Malthusian force, if appropriate socio-economic conditions are present.
Here the theory of endogenous technical change comes in by telling us that
a large population may be good for technical progress if the institutions in
society are growth-friendly. A larger population breeds more ideas, the more
so the better its education is; a larger population also promotes division of la-
bor and larger markets. This helps the creation of new technologies or, from
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
13.5. Robustness issues and scale effects 235
the perspective of an open economy, it helps the local adoption of already
existing technologies outside the country. In a less spectacular way it helps
by furthering day-by-day productivity increases due to learning by doing and
learning by watching. The non-rival character of technical knowledge is an
important feature behind all this. It implies that output per capita depends
on the total stock of ideas, not on the stock per person. This implies −
everything else equal − an advantage of scale.
In the models considered so far in this course, natural resources and the
environment have been more or less ignored. Here only a few remarks about
this limitation. The approach we have followed is intended to clarify certain
mechanisms − in abstraction from numerous things. The models in focus
have primarily been about aspects of an industrialized economy. Yet the
natural environment is always a precondition. A tendency to positive scale
effects on levels may be more or less counteracted by congestion and aggra-
vated environmental problems ultimately caused by increased population and
a population density above some threshold.
What can we say from an empirical point of view?
Empirical aspects
First of all we should remember that in view of cross-border diffusion of ideas
and technology, a positive scale effect (whether weak or strong) should not be
seen as a prediction about individual countries, but rather as pertaining to
larger regions, nowadays probably the total industrialized part of the world.
So cross-country regression analysis is not the right framework for testing
for scale effects, whether on levels or the growth rate. The relevant scale
variable is not the size of the country, but the size of a larger region to which
the country belongs, perhaps the whole world; and multivariate time series
analysis seems the most relevant approach.
Since in the last century there has been no clear upward trend in per
capita growth rates in spite of a growing world population (and also a growing
population in the industrialized part of the world separately), most econo-
mists do not believe in strong scale effects. But on the issue of weak scale
effects the opinion is definitely more divided.
Considering the very-long run history of population and per capita income
of different regions of the world, there clearly exists evidence in favour of
scale effects (Kremer, 1993). Whether advantages of scale are present also
in a contemporary context is more debated. Recent econometric studies
supporting the hypothesis of positive scale effects on levels include Antweiler
and Trefler (2002) and Alcalá and Ciccone (2004). Finally, considering the
economic growth in China and India since the 1980s, we must acknowledge
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
CHAPTER 13. PERSPECTIVES ON LEARNING BY DOING
236 AND LEARNING BY INVESTING
that this impressive performance at least does not speak against the existence
of positive scale effects on levels.
Acemoglu seems to find positive scale effects on levels plausible at the
theoretical level (pp. 113-114). At the same time, however, later in his book
he seems somewhat skeptical as to the existence of empirical support for this.
Indeed, with regard to the fact that R&D-based theoretical growth models
tend to generate at least weak scale effects, Acemoglu claims: “It is not clear
whether data support these types of scale effects” (Acemoglu, 2009, p. 448).
My personal view on the matter is that we should, of course, recognize
that offsetting forces, coming from our finite natural environment, are in play
and that a lot of uncertainty is involved. Nevertheless it seems likely that at
least up to a certain point there are positive scale effects on levels.
Policy implications If this holds true, it supports the view that inter-
national economic integration is generally a good idea. The concern about
congestion and environmental problems, in particular global warming, should
probably, however, preclude recommending governments and the United Na-
tions to try to promote population growth.
Moreover, it is important to remember the distinction between the global
and the local level. The in the formula (13.31) refers to a much larger
region than a single country; we may refer to this region as “the set of
knowledge-producing countries in the world”. No recommendation of higher
population growth in a single country is implied by this theoretical formula.
When discussing economic policy from the perspective of a single country, all
aspects of relevance in the given local context should be incorporated. For a
developing country with limited infrastructure and weak educational system,
family-planning programs and similar may in many cases make sense from
both a social and a productivity point of view (cf. Dasgupta, 1995).
13.6 Appendix
A. Balanced growth in the embodied technical change model with
investment-specific learning
In this appendix the results (13.16), (13.17), (13.18), and (13.19) are derived.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
13.6. Appendix 237
The model is:
= 1− 0 1 (13.37)
= − (13.38)
̇ = − (13.39)
µZ ¶
= 0 ≤ ̄ (13.40)
−∞
= 0 ≥ 0 (13.41)
Consider a BGP. By definition, and then grow at constant rates, not
necessarily positive. With = constant and 0 1 (13.37) gives
= = + (1 − ) (13.42)
a constant. By (13.39), = − showing that is constant along
a BGP. Hence,
+ = (13.43)
and so also must be constant. From (13.40) follows that = −1
Taking logs in this equation and differentiating w.r.t. gives
̇ 1
= − + = 0
in view of constancy of Substituting into (13.43) yields (1 + ) =
which combined with (13.42) gives
(1 − )(1 + )
=
1 − (1 + )
which is (13.16). In view of = = = ( + ) = (1 + ) the
results (13.17), (13.18), and (13.19) immediately follow.
B. Big bang a hair’s breadth from the AK
Here we shall prove the statement in Section 13.5.2: a hair’s breadth from
the AK assumption the technology is so productive as to generate infinite
output in finite time.
The simple AK model as well as reduced-form AK models end up in an
aggregate production function
=
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
CHAPTER 13. PERSPECTIVES ON LEARNING BY DOING
238 AND LEARNING BY INVESTING
We ask the question: what happens if the exponent on is not exactly 1,
but slightly above. For simplicity, let = 1 and consider
= = 1 + ' 0
Our claim is that if 1 a constant saving rate, will generate infinite
and in finite time.
We embed the technology in a Solow-style model with = = 0 and
get:
̇ ≡ = 0 1 (0) = 0 0 given (13.44)
We see that not only is ̇ 0 for all ≥ 0 but ̇ is increasing over time
since is increasing. So, for sure, → ∞ but how fast?
One way of answering this question exploits the fact that ̇ = is
a Bernouilli equation and can be solved by considering the transformation
= 1− as we do in Chapter 7 and Exercise III.3. Closely related to
that method is the approach below, which may have the advantage of being
somewhat more transparent and intuitive.
To find out, note that (13.44) is a separable differential equation which
implies
− =
By integration,
Z Z
−
= + C ⇒
−+1
= + C (13.45)
1−
where C is some constant, determined by the initial condition (0) = 0 For
= 0 (13.45) gives C = 0−+1 (1 − ) Consequently, the solution = ()
satisfies
0 1− ()1−
− = (13.46)
−1 −1
As increases, the left-hand side of this equation follows suit since ()
increases and 1 There is a ̄ ∞ such that when → ̄ from below,
() → ∞ Indeed, by (13.46) we see that such a ̄ must be the solution to
the equation µ 1− ¶
0 ()1−
lim − = ̄
()→∞ −1 −1
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
13.7. References 239
Since µ ¶
0 1− ()1− 0 1−
lim − =
()→∞ −1 −1 −1
we find
1 0 1−
̄ =
−1
To get an idea about the implied order of magnitude, let the time unit be
one year and = 01 0 0 = 01− = 2 and = 105 Then ̄ = 400 years.
So the Big Bang ( = ∞) would occur in 400 years from now if = 105
As Solow remarks (Solow 1994), this arrival to the Land of Cockaigne
would imply the “end of scarcity”, a very optimistic perspective.
In a discrete time setup we get an analogue conclusion. With airframe
construction in mind let us imagine that the learning parameter is slightly
above 1. Then we must accept the implication that it takes only a finite
number of labor hours to produce an infinite number of airframes. This is
because, given the (direct) labor input required to produce the ’th in a
sequence of identical airframes is proportional to − the total labor input
required to produce the first airframesPis proportional to 11 +12 +13
+ + 1 Now, the infinite series ∞
=1 1 converges if 1 As a
consequence only a finite amount of labor is needed to produce an infinite
number of airframes. “This seems to contradict the whole idea of scarcity”,
Solow observes (Solow 1997, p. 8).
13.7 References
Antweiler and Trefler, 2002, , AER.
Alcalá and Ciccone, 2004, , QJE.
Arrow, K. J., 1962. The Economic Implications of Learning by Doing.
Review of Economic Studies 29, 153-73.
Benhabib. J., and B. Jovanovic, 1991. Externalities and growth accounting.
American Economic Review 81 (1), 82-113.
Boucekkine, R., F. del Rio, and O. Licandro, 2003. Embodied Technological
Change, Learning-by-doing and the Productivity Slowdown. Scandina-
vian Journal of Economics 105 (1), 87-97.
Cozzi, 1997, , Journal of Economic Growth.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
CHAPTER 13. PERSPECTIVES ON LEARNING BY DOING
240 AND LEARNING BY INVESTING
DeLong, B. J., and L. H. Summers, 1991. Equipment Investment and Eco-
nomic Growth. Quarterly Journal of Economics 106, 445-502.
Englander, A., and A. Mittelstadt, 1988. Total factor productivity: Macro-
economic and structural aspects of the slowdown. OECD Economic
Studies, No. 10, 8-56.
Gordon, R. J., 1990. The Measurement of Durable goods Prices. Chicago
University Press: Chicago.
Greenwood, J., Z. Hercowitz, and P. Krusell, 1997. Long-Run Implications
of Investment-Specific Technological Change. American Economic Re-
view 87 (3), 342-362.
Greenwood, J., and B. Jovanovic, 2001. Accounting for growth. In: New
Developments in Productivity Analysis, ed. by C. R. Hulten, E. R.
Dean, and M. J. Harper, NBER Studies in Income and Wealth, Chicago:
University of Chicago Press.
Groth, C., T. M. Steger, and K.-J. Koch, 2010. When economic growth is
less than exponential, Economic Theory 44, 213-242.
Groth, C., and R. Wendner, 2014. Embodied learning by investing and
speed of convergence, Journal of Macroeconomics, forthcoming.
Gunn and Johri, 2011, , Review of Economic Dynamics, 992-101.
Ha, J., and P. Howitt, 2007, Accounting for trends in productivity and R&D:
A Schumpeterian critique of semi-endogenous growth theory, Journal
of Money, Credit, and Banking, vol. 39, no. 4, 733-774.
Hercowitz, Z., 1998. The ’embodiment’ controversy: A review essay. Jour-
nal of Monetary Economics 41, 217-224.
Hulten, C. R., 1992. Growth accounting when technical change is embodied
in capital. American Economic Review 82 (4), 964-980.
Irwin, D.A., and P. J. Klenow, 1994, Learning-by-doing spillovers in the
semi-conductor industry, Journal of Political Economy 102 (6), 1200-
1227.
Jones, C. I., 1994. Economic Growth and the Relative Price of Capital.
Journal of Monetary Economics 34, 359-382.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
13.7. References 241
Jones, C. I., 1995a, R&D-based models of economic growth, Journal of
Political Economy.
Jones, C. I., 1995b. Time series tests of endogenous growth models. Quar-
terly Journal of Economics, 110 (2), 495-525.
Jones, C. I., 1999, Growth: With or without scale effects , American Eco-
nomic Review, vol. 89, Papers and proceedings, May, 139-144.
Jones, C. I., 2003, Population and ideas: A theory of endogenous growth,.
In: ...
Jones, C. I., 2005. Growth and ideas. In: Handbook of Economic Growth,
vol. 1B, ed. by P. Aghion and S. N. Durlauf, Elsevier: Amsterdam,
1063-1111.
Jovanovic, B., and Nyarko (1995), Empirical learning curves, Brookings
Papers on Economic Activity (Micro), no. 1.
Klenow, P. J., and Rodriguez-Clare, A., 2005. Externalities and growth.
In: Handbook of Economic Growth, vol. 1A, ed. by P. Aghion and S.
N. Durlauf, Elsevier: Amsterdam.
Kremer, M., 1993, , QJE.
Krugman, P., 1987.
Levine and Renelt, 1992, , AER.
Lucas, R. Jr., 1988.
Lucas, R. Jr., 1993. Making a miracle, Econometrica.
Mehlum, H., K. Moene, and R. Torvik, 2002, Institutions and the resource
curse, WP, Oslo.
Pack, 1994, , J. Econ. Perspectives.
Ploeg, R. van der, 2011, Natural resources: Curse or blessing? Journal of
Economic Literature, vol. 49 (2), 366-420.
Rapping, 1965,
Razin, A., and E. Sadka, 1995, Population Economics, MIT Press.
Romer, P., 1986.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
CHAPTER 13. PERSPECTIVES ON LEARNING BY DOING
242 AND LEARNING BY INVESTING
Romer, P., 1990,
Romero-Avila, D., 2006, Can the AK model be rescued? New evidence from
unit root tests with good size and power, Topics in Macroeconomics,
vol. 6 (1), Article 3.
Sachs, J. D., and A. M. Warner, 1995. Natural resource abundance and
economic growth, NBER WP # 5398.
Searle, 1945.
Solow, R. M., 1960. Investment and technical progress. In: K. J. Arrow,
S. Karlin, and P. Suppes, eds., Mathematical Methods in the Social
Sciences, Stanford: Stanford University Press, pp. 89-104.
Solow, R.M., 1994, ....., J. Econ. Perspectives.
Solow. R. M., 1997, Learning from ‘Learning by Doing’, Stanford.
Sorger, G., 2010, , Economica.
Thornton and Thompson, 2001, Learning from experience and learning from
others: An exploration of learning and spillovers in wartime shipbuild-
ing, AER, Dec.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
Chapter 14
The lab-equipment model
In the learning-by-doing and learning-by-investing models of chapters 12 and
13, technical progress comes as a by-product of the production activity and
is considered an externality. This is just one mechanism behind technical
progress. Another branch of growth theory focuses on technical progress as
evolving from purposeful decisions by firms in search of monopoly profits on
innovations. This branch of growth theory is called innovation-based growth
theory.
Recall the definition of technical knowledge as a list of instructions about
how different inputs can be combined to produce a certain output. For ex-
ample it could be a principle of chemical engineering. Such a list or principle
can be copied on the blackboard, in books, in journals, on floppy disks etc.
and can, by its nature, be available and used over and over again at arbitrar-
ily many places at the same time. Thus, technical knowledge is a non-rival
good.1 At least temporarily, however, new technical knowledge may be tem-
porarily excludable by patents, secrecy, or copyright so that the innovator
can maintain a monopoly on the commercial use of new technical knowledge
for some time.
The lab-equipment model (based on Paul Romer, AER 1987) is the sim-
plest model within the class of models focusing on horizontal innovations.
This term refers to inventions of new types of goods, i.e., new “technical
designs” in the language of Romer. The present model considers invention of
new technical designs for input goods, but a more general framework would
include new types of consumption goods as well.2 The rising number of vari-
1
Even though a particular medium on which a copy of a list of inctructions is placed
is a rival good, it can usually be reproduced at very low cost in comparison with the cost
of making additions to the stock of technical knowledge.
2
For a model where the new goods are new consumption goods, see Acemoglu, Chapter
13, Section 13.4.
243
244 CHAPTER 14. THE LAB-EQUIPMENT MODEL
eties of goods contributes to productivity via increased division of labor and
specialization in society. Thus this class of models is known as “increasing-
variety models”.
In Acemoglu’s Chapter 13, Section 13.1, the lab-equipment model is pre-
sented in a version containing two knife-edge conditions in the form of ar-
bitrary parameter links. In the present text we present the lab-equipment
model without these parameter links. In addition, the presentation below
goes more into detail with the national income aspects of the model and
with the interaction between the financing needs of R&D labs and the saving
by the households.
14.1 Overview of the economy
We consider a closed market economy. The activities in the economy can be
subdivided into three sectors:
1. The basic-goods sector which operates under conditions of perfect com-
petition and free entry.
2. The specialized intermediate-goods sector which operates under condi-
tions of monopolistic competition and barriers to entry.
3. The R&D sector inventing new technical designs and operating under
conditions of perfect competition and free entry.
All produced goods are non-durable goods. There is no physical capital
(durable produced means of production) in the economy. All firms are profit
maximizers.
14.1.1 The sectorial production functions
In the basic-goods sector, sector 1, firms combine labor and different inter-
mediate goods to produce a homogeneous output good. The representative
firm in the sector has the production function
à !
X
= 1− 0 0 1 (14.1)
=1
where is output in the sector, is a positive constant, is input of
intermediate good ( = 1 2 ) is the number of different types
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
14.1. Overview of the economy 245
of intermediate goods available at time and is labor input.3 To avoid
arbitrary parameter links, we do not introduce Acemoglu’s assumption that
the technical coefficient happens to equal 1(1 − ) Sector 1 is the only
sector that uses labor.
Basic goods have three alternative uses. They can be used a) for consump-
tion, ; b) as raw material, to be converted into specialized intermediate
goods (in Danish “halvfabrikata”); and c) as investment, in R&D. Hence,
= + + (14.2)
In the specialized intermediate-goods sector, sector 2, at time there are
monopoly firms, each of which supplies a particular already invented
intermediate good. Once the technical design for intermediate good has
been invented in sector 3 (see below), the inventor takes out (free of charge)
a perpetual patent on the commercial use of this design and enters sector
2 as an innovator. Given the technical design, the innovator can instantly
transform a certain number of basic goods into a proportional number of
intermediate goods of the invented specialized kind. Specifically, at every
time it takes units of the basic good to supply units of intermediate
good :
units of the basic good y units of intermediate good (14.3)
where is a positive constant. We may think of the new technical design
as a computer code which, once in place, just requires pressing a key on
a computer in order activate the desired number of transformations. The
computer cost is negligible and the transformation requires no labor.
Thus, is both the marginal and the average cost of supplying the inter-
mediate good . This transformation technology applies to all intermediate
goods, = 1 2 , and all . Hence, the in (14.2) satisfies
X
≡ ≡ (14.4)
=1
where is the total supply of intermediate goods, all of which are used up
in the production of basic goods. Apart from introducing a specific symbol,
for this total supply of intermediate goods, our notation is the same as
Acemoglu’s, Chapter 13. Yet, to help intuition, we think of variety as some-
thing discrete rather than a continuum and use summation across varieties
as in (14.1) and (14.4) whereas Acemoglu’s uses integrals.
3
By an “intermediate good” is meant a non-durable means of production (like materials
and energy) used up in the single production process while "capital” means a durable
means of production (like a machine).
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
246 CHAPTER 14. THE LAB-EQUIPMENT MODEL
The model gives a “truncated” picture of the R&D sector, sector 3, as
fictional research labs that transform incoming basic goods (now considered
as R&D “equipment”) into a random stream of research successes. A re-
search success is an invention of a technical design (blueprint) for making
a new specialized intermediate good. There is free entry to R&D activity.
The uncertainty associated with R&D is “ideosyncratic” (unsystematic, di-
versifiable) and the economy is “large”. On average it takes an input flow of
1 units of the basic good, and nothing else, to obtain one successful R&D
outcome (an invention) per time unit. By the law of large numbers, the
aggregate number of new technical designs (inventions) in the economy per
time unit equals the expected number. With time continuous and ignoring
indivisibilities,4 we can therefore write
̇ ≡ = 0 constant, (14.5)
where, as noted above, is the aggregate research input per time unit and
is “research productivity”. Since the payoff to the outlay, on R&D comes
in the future, this outlay makes up an investment. Although the invested
basic goods are non-durable goods, the resulting new technical knowledge is
durable.
At first sight this whole production setup may seem peculiar. In sector 2
as well as sector 3, parts of the output from sector 1 is used as input to be
transformed into specialized intermediate goods and new technical designs,
respectively. But there is no labor input in sector 2 and sector 3. Formulating
the three kinds of production in the economy in this manner is a convenient
way of saving notation and is typical in this type of models.5 A more realistic
full-fledged description of the production structure would start with a pro-
duction function, with both labor and intermediate goods as inputs, in each
sector. Then an assumption could be imposed that the production functions
are the same, apart from allowing the total factor productivity to vary across
the sectors (only if 1 = = 1 would the total factor productivities be the
same). Setting the model up that way would fit intuition better but would
also require a more cumbersome notation. Anyway, the conclusions would
not be changed.
Before considering agents’ behavior, it may be clarifying to do a little
national income accounting.
4
Conceptually, is a discrete variable taking values in {1 2 }. Yet, for “large”
it is usually acceptable to smooth out as a continuous and differentiable function of
5
At the same time it is the lack of direct research labor in sector 3 that motivates the
term “lab-equipment model”. And it is the multi-faceted use of output from sector 1 that
motivates the term “basic goods”.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
14.1. Overview of the economy 247
14.1.2 National income accounting
The production side Using the basic good as our unit of account, all the
specialized intermediate goods will in equilibrium have the same price (see
Section 14.3.2). We therefore have:
value added in sector 1 = − (14.6)
value added in sector 2 = −
value added in sector 3 = ̇ −
where is the market value of an innovation and turns out to be independent
of time. The aggregate value added, or net national product, is
= − + − + ̇ −
= − + − + ̇ − = − (14.7)
where the last equality comes from ̇ − = 0 in equilibrium due to the
way sector 3 is described. Since there is no capital that depreciates in the
economy, gross national product and net national product are the same.
Notice that the production function for is a production function neither
for nor even for value added in sector 1, but simply for the quantity
of produced goods in that sector. It is typical for a multi-sector model with
non-durable intermediate goods that the production functions in the different
sectors do not describe value added in the sector but the produced quantity.
The income side There are two kinds of income in the economy, wage
income and profits. The time- real wage per unit of labor is denoted
and the profit per time unit earned by each monopoly firm in sector 2 is
denoted (in equilibrium it turns out to be the same for all the monopoly
firms). Profits are immediately paid out to the share owners. Owing to
perfect competition and CRS in both sector 1 and sector 3, there is no profit
generated in these sectors. The income side of NNP is thereby
= +
since the number of monopoly firms is Aggregate income is used for
consumption and saving,
+ = +
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
248 CHAPTER 14. THE LAB-EQUIPMENT MODEL
The uses of NNP By (14.7) and (14.4), final output can be written
= − = − = + (14.8)
that is, as the sum of aggregate consumption and investment. Aggregate
saving is
= + − = − =
by (14.8), reflecting that aggregate saving in a closed economy equals aggre-
gate investment, the R&D expense, .
14.1.3 The potential for sustained productivity growth
Already the production function (14.1) conveys the basic idea of an “increasing-
variety model”. In equilibrium we get = for all since the intermediate
goods enter symmetrically in this production function and end up having the
same price (see below). Thereby, (14.1) becomes
= 1−
= ( ) 1− 1−
≡ ( )
where is the total input of intermediate goods and 2 0. We see that
| =const. = 2 ( ) 0
This says that for a given total input, of intermediate goods, and a
given the higher the number of varieties (with which follows a lower of
each intermediate since is given), the more productive is this total input
of intermediate goods. “Variety is productive”. There are “gains to division
of labor and specialization in society”. The number of input varieties,
can thus be interpreted as a measure of the level of productivity-enhancing
knowledge.6 Note also that the function displays a form of increasing
returns to scale with respect to three “inputs”: intermediate goods,
variety, and labor,
14.2 Households and the labor market
There are households, all alike, with infinite horizon and preference para-
meters 0 and . Each household supplies inelastically one unit of labor
6
There exists a related class of models where growth (measured in terms of produced
economic value) is driven by increasing variety of consumption goods rather than increasing
variety of input goods. These models are sometimes called “love of variety” models. See
Acemoglu, Section 13.4.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
14.3. Firms’ behavior 249
per time unit. Let denote per capita consumption A household
chooses a plan ( )∞
=0 to maximize
Z ∞
1− −
0 = s.t.
0 1−
≥ 0
̇ = + − 0 given,
lim − 0
≥ 0 (14.9)
→∞
where equals per capita financial wealth. In equilibrium
=
because the only asset with market value in the economy is equity shares in
the monopoly firms the value of which equals the market value per technical
design multiplied by the number of technical designs available. As accounted
for in Section 14.3.3, the risk-averse households (00 0) can fully diversify
any risk so as to obtain the rate of return, with certainty on all their
saving.
The first-order conditions for the consumption-saving problem lead to the
Keynes-Ramsey rule
̇ 1
= ( − ) (14.10)
The necessary transversality condition is that the No-Ponzi-Game condition
(14.9) is satisfied with equality.
The labor market
There is perfect competition and complete real wage flexibility in the labor
market. For every the supply of labor is a constant. The demand for
labor, comes from the basic-goods sector (as the two other sectors do not
use labor). In equilibrium,
= (14.11)
14.3 Firms’ behavior
To save notation, in the description below, we take (14.11) for granted.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
250 CHAPTER 14. THE LAB-EQUIPMENT MODEL
14.3.1 The competitive producers of basic goods
At every the representative firm in the basic-goods sector maximizes profit
under perfect competition:
à !
X X
max Π = 1−
− − . (14.12)
1 2
=1 =1
The first-order conditions are, for every
Π = − = − = 0 (14.13)
and
Π = − = (1 − )−
− = 0 = 1 2
This gives the demand for intermediate good :
µ ¶−1
−1
= = [(1 − )]1 = 1 2
(1 − )
(14.14)
The price elasticity of demand, E for intermediate good is thus −1
This reflects that the elasticity of substitution between the specialized inter-
mediate goods in (14.12) is 1(1 − (1 − )) = 1 This elasticity is above 1.
Hence, while the specialized intermediate goods are not perfect substitutes,
they are sufficiently substitutable for a monopolistic competition equilibrium
in sector 2 to exist, as we shall now see.
14.3.2 The monopolist suppliers of intermediate goods
In principle the decision problem of monopolist is the following. Subject to
the demand function (14.14), a price and quantity path ( )∞ = should
be chosen so as to maximize the value of the firm (the present value of future
cash flows): Z ∞
= −
(14.15)
where is the profit at time
= ( − ) (14.16)
and where the discount discount rate is the risk-free interest rate.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
14.3. Firms’ behavior 251
Since there is in this intertemporal problem no interdependence across
time, the problem reduces to a series of static problems, one for each :
max = ( − )
s.t. (14.14).
To solve for , we could substitute the constraint into the expression for ,
take the derivative w.r.t. and then equalize the result to zero.
Alternatively, we may use the rule that the profit maximizing price of
a monopolist is the price at which marginal revenue equals marginal cost,
= This is the more intuitive route we will take. We have
(= total revenue) = = ( )
where ( ) denotes the maximum price at which the amount can be sold.
Thus, by the product rule,
= = ( ) + 0 ( ) = (1 + E )
µ ¶ µ ¶
1 1
≡ 1 + = 1 + = (1 − )
E −1
from (14.14). Marginal cost is = So the profit maximizing price is
= ≡ (14.17)
1−
Owing to monopoly power, the price is above ; the mark-up factor (or
“degree of monopoly”) is 1(1 − ) As expected, a lower absolute price
elasticity of demand, 1 results in a higher mark-up.
Since the elasticity of demand w.r.t. the price is independent of the
quantity demanded and since is constant, the chosen price is time inde-
pendent. Moreover the price is the same for all = 1 2 . Substitution
into (14.14), (14.16), and (14.15), gives
µ ¶1
(1 − )2
= ≡ for all (14.18)
= ( − ) = ( − ) = ≡ for all and
(14.19)
1− 1−
Z ∞
Z ∞
= − = − ≡ for all (14.20)
respectively. We see that all the monopoly firms sell the same quantity
earn the same profit, and have the same market value, In addition,
(14.18) and (14.19) show that and are constant over time. We will soon
see that so is
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
252 CHAPTER 14. THE LAB-EQUIPMENT MODEL
The reduced-form aggregate production function in the economy
Note that although we have skipped the two arbitrary parameter links, =
1(1 − ) and = 1 − applied by Acemoglu, the resulting expressions
for and are tractable anyway.7 So is the implied result for gross
output in the basic-goods sector:
µ ¶ 1−
1− (1 − )2
= = ≡ ̂ (14.21)
where we have inserted (14.18) into (14.1) and defined
µ ¶ 1−
(1 − )2
̂ ≡
The value added in the sector is
− = ̂ − = (̂ − )
where and are constants given in (14.17) and (14.18), respectively.
So both gross and net output in the basic-goods sector are proportional
to the number of intermediate-goods varieties (in some sense an index of the
endogenous level of technical knowledge in society). Moreover, a similar pro-
portionality will hold for the net national product, . Indeed, according
to (14.8),
= − = ̂ − = (̂ − ) (14.22)
This is a first signal that the model is likely to end up as a reduced-form AK
model with (“knowledge capital”) acting as the capital variable.
Now to the R&D firms of sector 3.
14.3.3 R&D firms
In Section 1.1 we expressed the aggregate number of new technical designs
(inventions) per time unit this way:
̇ ≡ = 0 constant, (*)
7
With his two parameter links Acemoglu obtains = (1 − )−2 from which follows
the simple formulas = and = for all and all Although these formulas
are, of course, simpler, they are “dangerous” when one wants to calculate, for instance,
in order to assess the effect of a rise in (the output elasticity w.r.t. labor) on
the monopoly profit .
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
14.3. Firms’ behavior 253
where is the R&D investment (in terms of basic goods) and is “research
productivity”. What is the microeconomic story behind this?
There is a “large” number of R&D labs and free entry and exit. All R&D
labs operate under the same conditions with regard to “research technol-
ogy”. The following simplifying assumptions are made. The random R&D
outcomes are:
(i) uncorrelated across time (no memory),
(ii) uncorrelated across the R&D labs,
(iii) uncorrelated with any variable in the economy, and
(iv) there is no overlap in research.
The “no memory” assumption, (i), ignores learning over time within the
lab which seems a quite drastic assumption; indeed, innovation should be
considered a cumulative process. Assumption (ii) seems drastic as well, since
some learning across R&D labs is likely. In combination, the assumptions
(i), (ii), and (iii) sum up to what is called “ideosyncratic” uncertainty. The
“no overlap” assumption, (iv), amounts to assuming that inventions can go
in so many directions that the likelihood of different research labs chasing
and making the same invention is negligible. So we can find the aggregate in-
crease in “knowledge” simply by summing the contributions by the individual
research labs.
The “research technology”
The “research technology” faced by the individual R&D labs can be described
as a Poisson process. The expected number of successful research outcomes
(inventions) per time unit is proportional to the flow input of basic goods
into the lab.
Consider an arbitrary R&D lab, at time = 1 2 where is
“large”. Let be the amount of basic goods the lab devotes to research per
time unit. There is an instantaneous success arrival rate, per unit invested
such that, given the research flow the success arrival rate (= expected
number of inventions per time unit) at time , is
= 0 (14.23)
The Poisson parameter, , measures “research productivity”. The interpre-
tation of is that if denotes the number of success arrivals in the time
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
254 CHAPTER 14. THE LAB-EQUIPMENT MODEL
interval ( + ∆] then
( | ∆)
= lim (14.24)
∆→0 ∆
where is the conditional expectation operator at time .
At the aggregate level, since, by assumption, there is no overlap in re-
search,
³P ¯ ´
P ¯
( ) ∆ X ( | ∆)
∆ ( )
=1
= ≈ =
∆ ∆ ∆
∆
Appealing to the law of large numbers, we replace “≈” by “=” ignore indi-
visibilities, and take limits:
∆ X ( | ∆) X X
̇ = lim = lim = = =
∆→0 ∆ ∆→0 ∆
(14.25)
which is (*). The third equality in (14.25) comes from (14.24), the fourth
from (14.23), and the last from the definition of aggregate R&D input, .
The financing of R&D
There is a time lag of random length between a research lab’s outlay on R&D
and the arrival of a successful research outcome, an invention. During this
period, which in principle has no upper bound, the R&D lab is incurring
sunk costs and has no revenue at all. R&D is thus risky and continuous
refinancing is needed until the research is successful.
Under certain conditions, the required financing of R&D will nevertheless
be available. To clarify this, consider first the situation ex post a successful
research outcome. When a successful research outcome arrives, the inventor
takes out (free of charge) a perpetual patent on the commercial use of the
invention. This gives the invention the market value, the same for all
research labs, cf. (14.20). The inventor can realize this market value either
by licensing the right to use the invention commercially or by directly herself
entering sector 2 as a monopolist supplier of the new good made possible by
the invention. To fix ideas, we assume the latter always takes place.
Now consider the situation ex ante an R&D investment is decided.
CLAIM 1 Given the market value, of an invention, the expected payoff
per time unit per unit of basic goods invested in R&D is
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
14.3. Firms’ behavior 255
Proof Consider an arbitrary R&D lab The probability of a successful re-
search outcome in a “small” time interval ( + ∆] is approximately ∆
And the probability that more than one successful research outcome arrives
in the time interval is negligible. We thus have
(R&D payoff | ∆) ≈ ∆ + 0 · (1 − ∆) = ∆ (14.26)
Substituting (14.23) into this and dividing through by ∆ gives
(R&D payoff | ∆) ∆
≈ =
∆ ∆
Letting ∆ → 0 “≈” can in the limit be replaced by “=”, thus confirming
the claim. ¤
Considering equilibrium in the loan market, we have:
P
CLAIM 2 Let = (i) In any equilibrium in the loan market, whether
with = 0 or 0 we have
≤ 1 (14.27)
(ii) In any equilibrium in the loan market where 0 we have
= 1 (14.28)
Proof. (i) Suppose that, contrary to (14.27), we have 1 By Claim 1,
the expected R&D payoff per time unit per unit cost of R&D is then higher
than the R&D cost and so expected pure profit by doing R&D is positive.
The flow demand for finance to R&D firms will therefore be unbounded.
The flow supply of finance, ultimately coming from household saving, is,
however, bounded and thus there is excess demand for funds and thereby
not equilibrium.8 Thus 1 can be ruled out as an equilibrium and this
leaves (14.27) as the only possible state in an equilibrium.
(ii) Consider an equilibrium with 0 Since it is an equilibrium,
(14.27) must hold. By way of contradiction, let us imagine there is strict
inequality in (14.27). Then all R&D firms will choose = 0 and we reach
the conclusion that = 0 thus contradicting that 0 So there can
not be strict inequality in (14.27) and we are left with (14.28) as the only
possible state in an equilibrium with 0 ¤
8
For the sake of intuition, allow disequilibrium to exist in the very short run. Then
the excess demand for funds drives the interest rate, up, thus lowering (cf. (14.20))
until = 1
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
256 CHAPTER 14. THE LAB-EQUIPMENT MODEL
It follows from Claim 2 that when the market value of inventions satisfy
(14.28), the cost of doing R&D is on average exactly covered by the expected
payoff. In return for putting one unit of account at the disposal of a research
lab, the household gets a payoff of if the research turns out to be successful
and zero otherwise. In expected value the payoff per time unit is one unit
of account. It is as if the household buys a lottery ticket offered by the
R&D lab to finance its current R&D costs. The lottery prize consists of
shares of stock giving the right to the future monopoly profits if the current
research is successful within one time unit. The lottery is “fair” because the
cost of participating equals the expected payoff. In spite of being risk averse
(00 () 0) the households are willing to participate because the uncertainty
is “ideosyncratic” and the economy is “large”. This allows the households to
avoid the risk by spreading their investment over a variety of R&D labs, i.e.,
by diversifying their investment.
14.3.4 Equilibrium in the loan market
What must the size of the equilibrium real interest rate, in the loan market
be? This rate must satisfy the following no-arbitrage relation vis-a-vis the
instantaneous rate of return on shares in sector-2 firms supplying specialized
intermediate goods:
+
= (14.29)
where is the constant dividend (assuming all profit is paid out to the
share owners) and is the capital gain (positive or negative) on holding
shares. As an implication of Claim 2, in an equilibrium with 0 the
market value of any invention is
= 1 ≡
a constant. So = 0 and (14.29) simplifies to
= = ≡ (14.30)
1
where is determined by (14.19). That is, along an equilibrium path with
0 the interest rate is constant and determined by (14.30).
To ensure that 0 and thereby positive growth is present in the
economy, we need that the parameters are such that households do save.
In view of the Keynes-Ramsey rule, this requires which in turn, by
(14.30), requires a sufficiently high research productivity
(A1)
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
14.3. Firms’ behavior 257
What ensures that household saving and R&D investment match each
other? Let aggregate financial wealth at time be denoted A Then, in an
equilibrium with 0,
1
≡ A = =
In view of ̇ = from (14.5), we therefore have
1 1
Ȧ = ̇ = ̇ = = (14.31)
By definition, households’ aggregate saving, , equals the increase in finan-
cial wealth per time unit, i.e., = Ȧ 9 Substituting this into (14.31), we
see that the investment, and saving, are two sides of the same coin
when the interest rate takes the equilibrium value in (14.30), and full
employment, as in (14.11), is ensured through real wage flexibility.
To understand that there are neither losers nor winners in this saving-
investment process, it may help intuition to imagine that all the saving, ∆
in a short time interval ( + ∆] first goes to large mutual funds (that have
no administrative costs). These mutual funds instantly use the receipts to
buy lottery tickets offered by R&D labs to cover current R&D costs. For the
mutual funds taken together this involves an exchange of the outlay ∆ for
shares giving the right to the future monopoly profits associated with those
research labs that turn out to be successful in the time interval considered.
By the law of large numbers the inventions by these labs have exactly the
same value as the outlay. Indeed, by (14.31), we have
̇ ∆ = ∆
From then on, holding diversified shares in the monopolies supplying the
newly invented intermediate goods gives the normal rate of return in the
economy, A fraction of the R&D labs have not been successful in the time
interval considered (and the financing to them has thereby been lost). But
others have been successful and made an invention. The unequal occurrence
of failures and successes across the many different R&D labs is neutralized
when it comes to the payout to the customers, i.e., the households who have
deposits in the mutual funds.
As an alternative financing setup, suppose that the R&D labs offer project
contracts of the following form. A contract stipulates that the investor pays
9
In this model households’ gross saving equals their net saving since there are no assets
that depreciate.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
258 CHAPTER 14. THE LAB-EQUIPMENT MODEL
the lab 1 units of account per time unit until a successful research outcome
arrives. The corresponding liability of the lab is, when achieving success and
becoming an entrepreneur in sector 2, to let the subsequent permanent profit
stream earned on the invention go to the investor. By Claim 1, such R&D
contracts have no market value. But after a successful R&D outcome there
is a capital gain in the sense that the contracts become shares in the hands of
the investors giving permanent dividends equal to per time unit and thus
having a market value equal to = 1 forever.
Note that as the model is formulated, there is no value added in the R&D
sector, as was also mentioned in connection with (14.7) in Section 14.1.2.
Instead, the value that at the aggregate level comes out as ̇ is just a
cost free one-to-one instantaneous transformation of which is a part of
the value added created in the basic-goods sector. It is ultimately this value
added that households’ saving pays for.
14.4 General equilibrium of an economy sat-
isfying (A1)
The assumption (A1) ensures a research productivity high enough to provide
a rate of return exceeding the rate of time preference and thereby induce the
household saving needed for R&D investment, to be positive. And from
(14.30) we know that along an equilibrium path with 0 and therefore
̇ 0, the interest rate (= the rate of return in the economy) is a constant,
Then the Keynes-Ramsey rule, (14.10), yields
̇ 1 1
= ( − ) = ( − ) ≡ (14.32)
where is given (14.19). To ensure that the path considered with ̇ 0 is
really capable of being an equilibrium path, we need the parameter restriction
1
(1 − ) = (1 − ) ( − ) (A2)
since otherwise the transversality condition of the household could not be
satisfied.10
From (14.21) and (14.22) we know that along an equilibrium path, gross
as well as net output in the basic-goods sector are proportional to the stock
of “knowledge capital”, Moreover, the analysis of the previous section
10
Another aspect of this is that (A2) ensures that the utility integral 0 is bounded and
thereby allows maximization in the first place.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
14.4. General equilibrium of an economy satisfying (A1) 259
shows that the preliminary national income accounting sketched in Section
14.1.2 is correct. Hence, by (14.22), also the aggregate value added in the
economy as a whole, NNP, is proportional to Indeed,
= − = ̂ − = (̂ − ) ≡ ̄
So the model does indeed belong to the class of reduced-form AK models.
14.4.1 The balanced growth path
From the general theory of reduced-form AK models with Ramsey house-
holds, we know that the “capital” variable of the model, here “knowledge
capital”, will grow at the same constant rate as per capita consumption
already from the beginning. In the present case the latter growth rate is
given by (14.32). And
̇ = = ( − ) = (̄ − ) (14.33)
so that µ ¶
̇
≡ = ̄ −
As = this implies
= (̄ − ) ,
for all ≥ 0 Hence, the until now unknown initial per capita consumption is
0
0 = (̄ − )
Labour productivity can be defined as
≡ = ̄ (14.34)
hence = =
Thus the model generates fully endogenous balanced growth and there are
no transitional dynamics. What makes fully endogenous growth possible is,
as usual, that the “growth engine” of the economy features constant returns
to scale w.r.t. producible inputs. Generally, as defined in Chapter 13.5, the
growth engine of a model is the set of input-producing sectors using their own
output as an input. After having derived the aggregate production function
in sector 1 as expressed in (14.21), sector 2 can be considered integrated in
sector 1. On this basis, sector 1 and sector 3 constitute the growth engine in
the model. Basic goods, = ++ and technical knowledge, represented
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
260 CHAPTER 14. THE LAB-EQUIPMENT MODEL
by the number, of varieties of intermediate goods, are the two kinds of
producible inputs. Sector 1 delivers the input flow to itself and the input
flow to sector 3. And sector 3 delivers the input “knowledge capital”,
to sector 1. The production functions (14.21) and (14.5) show that there
are constant returns to scale w.r.t. these two producible inputs. This is the
reason that fully endogenous growth is generated. In view of the absence
of transitional dynamics, the model can be classified as a reduced-form two-
sector AK model.
14.4.2 Comparative analysis
Given the per capita growth rate in (14.32), we have:
= −1 0. Higher impatience ⇒ lower propensity to save
⇒ less investment in R&D.
0. Higher desire for consumption smoothing ⇒ attempt to
transform some of the higher future consumption possibility into higher con-
sumption today ⇒ lower saving ⇒ less investment in R&D.
0. Higher factor productivity ⇒ higher return on saving ⇒
more saving at the aggregate level (the negative substitution effect and wealth
effect on consumption dominates the positive income effect) ⇒ more invest-
ment in R&D. As usual the constant need not have a narrow technical
interpretation. It can reflect the quality of the institutions in society (rule of
law etc.) and the level of “social capital”.11
0 Higher production costs of the specialized intermediate goods
⇒ higher production costs for basic goods ⇒ higher R&D costs ⇒ less in-
vestment in R&D.12
0. Higher R&D productivity results in more R&D investment
and higher growth.
= () 0. A larger population implies lower per
capita cost, associated with producing a given amount of new technical
knowledge which in turn improves productivity for all members of society.
This is an implication of knowledge being a nonrival good. In a larger soci-
ety, with larger markets, the incentive to do R&D is therefore higher. In the
present version of the R&D model the result is a higher growth rate perma-
11
By social capital is meant society’s stock of social networks and shared norms that
support and maintain confidence, credibility, trust, and trustworthiness.
12
Acemoglu’s Equation (13.20), p. 439, entails that = 0 This is due to the
arbitrary parameter link = 1 − This link implies that the effect of increasing the
production costs of specialized intermediate goods is not separated from the effect of
increasing the elasticity of output of basic goods w.r.t. input of intermediate goods, cf.
(14.1).
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
14.4. General equilibrium of an economy satisfying (A1) 261
nently. This is a manifestation of the controversial strong scale effect (scale
effect on growth), typical for the “first-generation” innovation-based growth
models with fully endogenous growth. This strong scale effect, as well as
the fully endogenous growth property, is due to a “hidden” knife-edge con-
dition in the specification of the “growth engine”, cf. the general discussion
in Chapter 13.5 and Exercise VII.5.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
262 CHAPTER 14. THE LAB-EQUIPMENT MODEL
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
Chapter 15
Stochastic erosion of
innovators’ monopoly power
In this chapter we extend the lab-equipment model of Chapter 14 by adding
stochastic erosion of innovators’ monopoly power. The motivation is the
following.
The model of Chapter 14 assumed that the innovator had perpetual
monopoly over the production and sale of the new type of intermediate good.
In practice, by legislation patents are of limited duration, 15-20 years. More-
over, it may be difficult to codify exactly the technical aspects of innovations,
hence not even within such a limited period do patents give 100% effective
protection. While the pharmaceutical industry rely quite much on patents,
in many other branches innovative firms use other protection strategies such
as concealment of the new technical design. In ICT industries copyright to
new software plays a significant role. Still, whatever the protection strategy
used, imitators sooner or later find out how to make very close substitutes.
To better accommodate these facts, the present chapter sets up a lab-
equipment model where competition in the supply of specialized intermediate
goods is more intense than in Chapter 14. For convenience we name the
model of Chapter 14 Model I. Compared with that model the only difference
in the new model is that the duration of monopoly power over the commercial
use of an invention is limited and uncertain. We name the resulting model
Model II. The notation is the same as in Model I. The analysis is related
to the brief discussion of the issues in Acemoglu’s Chapter 13.1.6 and in
particular to his Exercise 13.13.
First a recapitulation of the technological aspects of the economy.
263
CHAPTER 15. STOCHASTIC EROSION OF INNOVATOR’S
264 MONOPOLY POWER
15.1 The three production sectors
The technology of the economy is the same as in Model I. In the basic-goods
sector (sector 1) firms combine labor and different intermediate goods to
produce a homogeneous output good. The representative firm in the sector
has the production function
à !
X
= 1− 0 0 1 (15.1)
=1
where , and denote output of the firm, labor input, and input of
intermediate good , respectively, where = 1 2 . This sector, as well
as the labor market, operate under perfect competition.
The aggregate output of basic goods is used partly for replacing the basic
goods, used in the production of intermediate goods used up in the pro-
duction of basic goods, partly for consumption, and partly for investment
in R&D, . Hence, we have
= + + (15.2)
In the intermediate-goods sector, sector 2, at time there are monopoly
firms, each of which supplies a particular already invented intermediate good.
Once the technical design for intermediate good has been invented in sector
3, the inventor enters sector 2 as an innovator. Given the technical design,
the innovator can instantly transform a certain number of basic goods into a
proportional number of intermediate goods of the invented specialized kind.
That is,
it takes units of the basic good to supply units of intermediate good
(15.3)
where is a positive constant. The transformation requires no labor. Thus,
is both the marginal and the average cost of supplying the intermediate
good . This transformation technology applies to all intermediate goods,
= 1 2 , and all . Hence, the in (15.2) satisfies
X
≡ ≡ (15.4)
=1
where is the total supply of intermediate goods, all of which are used up
in the production of basic goods.
For a limited period after the invention has been made, through secrecy
or imperfect patenting the inventor maintains monopoly power over the com-
mercial use of the invention. The length of this period is uncertain, see below.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
15.2. Temporary monopoly 265
In the R&D sector, sector 3, new “technical designs” (blueprints) for
making new specialized intermediate goods are invented. The uncertainty
associated with R&D is “ideosyncratic”. On average it takes an input of
1 units of the basic good, and nothing else, to obtain one successful R&D
outcome (an invention) per time unit. There is free entry to the R&D activ-
ity. Ignoring indivisibilities, the aggregate number of new technical designs
(inventions) in the economy per time unit is
̇ ≡ = 0 constant, (15.5)
where, as noted above, is the aggregate R&D investment in terms of basic
goods delivered to sector 3 per time unit. As also noted above, after an
invention has been made, the inventor enters sector 2 as an innovator and
begins supplying the new intermediate good to firms in sector 1.
15.2 Temporary monopoly
To begin with the innovator has a monopoly over the production and sale
of the new intermediate good. This may be in the form of a more or less
effective patent (free of charge) or copyright to software or simply by secrecy
and concealment of the new technical design. But sooner or later imitators
find out how to make very close substitutes. There is uncertainty as to how
long the monopoly position of an innovator lasts.
We assume the cessation of monopoly power follows a Poisson process
with an exogenous “arrival” rate 0 the same for all monopolies.1 The
“event” which “arrives” sooner or later is “exposure to unbounded competi-
tion”. Independently of how long the monopoly position for firm has been
maintained, the probability that it breaks down in the next time interval of
length ∆ is approximately · ∆ for ∆ “small”. Equivalently, if denotes
the remaining lifetime of the monopoly status of intermediate good , then
the probability that is − for all 0 Further, the expirations of
the different monopoliesR ∞are stochastically independent. The expected dura-
−
tion of a monopoly is 0 = 1 We shall refer to the parameter
as the Poisson expiration rate.
An investor (household) who contemplates to finance the R&D activity
of a prospective innovator now faces a double risk, first the risk that the
R&D is unsuccessful for a long time, second the risk that, when finally it
is successful, the monopoly profits on the resulting innovation will only last
1
This approach builds on Barro and Sala-i-Martin (1995).
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
CHAPTER 15. STOCHASTIC EROSION OF INNOVATOR’S
266 MONOPOLY POWER
for a short time. The model assumes, however, that all uncertainty is idio-
syncratic, that is, the stochastic events that an R&D lab is successful in a
certain time interval and that an innovator looses her monopoly position in a
certain time interval are uncorrelated across R&D labs, innovators, and time
and are in fact not correlated with anything in the economy. Assuming a
“large” number of both R&D labs and intermediate-goods firms still being
monopolies, investors can eliminate any risk by diversifying their investment
as described in Chapter 14. Of course, this whole setup is an abstraction and
can at best be considered a benchmark case.
As labor supply is a constant, clearing in the labor market implies
= We insert this into the production function (15.1) of the repre-
sentative firm in sector 1. Maximizing profit, at time this firm then de-
−1
mands ( ) = ((1 − ))1 units of intermediate good per time
unit, = 1 2 . As long as innovator is still a monopolist, she faces
this downward-sloping demand curve with price elasticity −1 and sets the
price, such that = (marginal revenue = marginal cost). With
the basic good as our numeraire, this amounts to
1
(1 − ) =
1
Solving for we get
1
= (1 + markup) · = ≡
1−
Thereby, as long as innovator is still a monopolist, the sales of intermediate
good is
µ ¶1
1 −1 (1 − )2
( ) = () = ((1 − )) = ≡ () (15.6)
for = 1 2 . We shall refer to () as the monopoly supply of a specific
intermediate good.
The corresponding total revenue per time unit is ((1 − )) · () and
the total cost is · () . The earned profit per time unit is thus
= ( − ) () = ( − )() = () ≡ () (15.7)
1− 1−
for = 1 2 . The formulas for () and () are the same as those for
and respectively, in Model I, cf. Chapter 14.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
15.3. The aggregate production function in equilibrium 267
As described above, however, sooner or later innovator loses the monopoly.
Thereafter intermediate good is supplied under conditions of perfect com-
petition and its price, is driven down to the competitive market price
level = marginal cost = . Since marginal cost, is also average cost, the
profit vanishes. The aggregate sales of intermediate good now supplied by
many competitors, are
µ ¶1
(1 − )
( ) = () = ≡ (1 − )−1 () ≡ () () (15.8)
where () will be referred to as the competitive supply of a specific interme-
diate good. The inequality in (15.8) follows from 0 1 Economically,
the inequality in (15.8) reflects that the demand depends negatively on the
price, which is lower under perfect competition.
To summarize: In view of production and cost symmetry, each intermedi-
ate good supplied under monopolistic conditions is supplied in the amount,
() and each intermediate good supplied under competitive conditions is
supplied in the larger amount, () . That is,
½
() if is still a monopoly,
= () (15.9)
if is no longer a monopoly,
where () and () are given in (15.6) and (15.8), respectively.
15.3 The aggregate production function in equi-
librium
Substituting (15.9) into (15.1), we can write output in sector 1 as
h i
() ()
= (() )1− + (() )1− (15.10)
()
where is the number of intermediate good types that at time are still
()
supplied under monopolistic conditions and is the number of intermedi-
ate good types that have become competitive. For each we have
() ()
= + (15.11)
There are now two state variables in the model. There is therefore scope for
transitional dynamics, as we shall see soon.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
CHAPTER 15. STOCHASTIC EROSION OF INNOVATOR’S
268 MONOPOLY POWER
With the help of (15.11) and (15.8), we may rewrite (15.10):
h i
() ()
= ( − )(() )1− + (1 − )−(1−) (() )1−
h i
() () −(1−)
= − + (1 − ) (() )1−
h i
()
= + ((1 − )−(1−) − 1) (() )1−
" #
()
= 1 + ((1 − )−(1−) − 1) (() )1− (15.12)
()
Aggregate output is seen to depend on If the dynamics are such that
()
tends to a positive constant, then will tend to be proportional to
the produced “input”, since () is a constant, cf. (15.6). Therefore,
the model is likely capable of generating fully endogenous growth, driven by
R&D. We come back to this below.
()
In the case of universal and perpetual monopoly power, = 0 and
()
so (15.12) reduces to = (() )1− ≡ which is the equilibrium
output of basic goods in Model I. Substituting this into the expression (15.12),
we see that
" #
()
() ()
= 1 + ((1 − )−(1−) − 1) (15.13)
()
While in Model I, the Poisson expiration rate, is nil, hence = 0 for
()
all here wqe have 0 so that 0 This means that a fraction of
the intermediate goods are supplied at a price equal to marginal cost thus
inducing efficient use of these. Thereby productivity is enhanced and we
()
get as shown by (15.12) (where (1 − )−(1−) 1 in view of
0 1)
15.4 The no-arbitrage condition under uncer-
tainty*
All uncertainty is assumed to be ideosyncratic. By diversified investment in
R&D lotteries and the stock market, the risk-averse households can therefore
eliminate any risk and obtain the risk-free rate of return, with certainty.
The appropriate discount rate for calculating the present value of expected
future profits in any monopoly is this risk-free rate, Consequently, ruling
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
15.4. The no-arbitrage condition under uncertainty* 269
out speculative bubbles, the market value of monopoly at time is
Z ∞
= ( )− (15.14)
where is the time- profit flow, now a stochastic variable as seen from
time :
½ ()
if firm is still a monopolist at time
=
0 otherwise.
Expected profit flow at time as seen from time is
( ) = () −( −) + 0 · (1 − −( −) ) = () −( −) (15.15)
Substituting into (15.14), we get
Z ∞
()
= − ( +)
≡ (15.16)
This market value is the same for all intermediate goods which at time
still retain monopoly. The expression (15.16) gives the market value in a
certainty-equivalent form. On the one hand the integral in (15.16) “treats”
the monopoly profit stream as if it were perpetual, on the other hand this
future potential profit is discounted at an effective discount rate, +
taking into account the probability, −( −) that at time the ability to
earn this profit has disappeared. The in (15.16) is an observable variable
given that the firm is still a monopoly (otherwise it has market value equal to
nil). The uncertainty is about profits in the future and the discount rate for
these equals the risk-free interest rate plus a risk premium, here equal to
which is the approximate conditional probability that the monopoly status
breaks down in the time interval ( + 1] given it is retained up to time
2
At this point we face the question: how is the risk-free interest rate,
determined? To approach an answer, it is useful to derive the no-arbitrage
condition which is implicit in (15.14). It may help intuition to think of as
the interest rate on a market for safe loans.
By differentiating (15.16) w.r.t. using Leibniz’s’ formula,3 we get
(+)
() + ̇
= + (15.17)
2
This is known from the theory of a Poisson process.
3
See Appendix A.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
CHAPTER 15. STOCHASTIC EROSION OF INNOVATOR’S
270 MONOPOLY POWER
(+)
where ̇ is the conditional capital gain, that is, the increase per time
unit in the market value of the monopoly firm at time , conditional on its
monopoly position remaining in place also in the next moment. This formula
equalizes the instantaneous conditional rate of return per time unit on shares
in monopoly firms to the risk-free interest rate plus a premium reflecting the
risk that the monopoly position expires within the next instant.
Alternatively we may derive the no-arbitrage condition (15.17) without
appealing to Leibniz’s’ formula (which may not be part of the reader’s stan-
dard math toolbox). This alternative approach has the advantage of being
more intuitive. Let
≡ the firm’s earnings in the time interval ( + ∆), given that the
firm is still a monopolist at time .
There will be no opportunities for arbitrage if the expected instantaneous
unconditional rate of return per time unit on shares in the monopoly firm
equals the required rate of return which is the risk-free interest rate, This
amounts to the condition
lim∆→0 (∆ ∆)
= (15.18)
The firm’s earnings in the time interval ( + ∆] is approximately ∆
This is a stochastic variable and its expected value as seen from time is
(+)
( ∆) ≈ ∆(− ) + (1 − ∆)( () + ̇ )∆ (15.19)
Indeed, is the capital loss in case the monopoly position ceases and ∆ is
the approximate probability that this event occurs within the time interval
( + ∆], given that at time it has not yet occurred. Similarly, 1 − ∆ is
the approximate probability that a monopoly position retained up to time
(+)
remains in force at least up to time + ∆ And ( () + ̇ )∆ is the total
return in that case. Now, (15.19) can be written:
(+) (+)
( ∆) ≈ −∆ + ( () + ̇ )∆ − ( () + ̇ 2
)(∆)(15.20)
(+) (+)
= ( () + ̇ − )∆ − ( () + ̇ )(∆)2 ⇒
( ∆) (+) (+)
≈ () + ̇ − − (() + ̇ )∆
∆
(+)
→ () + ̇ − for ∆ → 0
Hence, the condition (15.18) implies the no-arbitrage condition
lim∆→0 (∆ ∆) (+)
() + ̇ −
= = (15.21)
Reordering, we see that this is the same condition as (15.17).
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
15.5. The equilibrium rate of return when R&D is active 271
15.5 The equilibrium rate of return when R&D
is active
At the aggregate level, by the law of large numbers, the cost of making ̇
inventions per time unit at time is = ̇ basic goods per time unit.
The expected cost per invention is thus 1. An equilibrium with active
R&D therefore requires4
= 1 ≡ (15.22)
So the market value of a monopoly firm is constant as long as the monopoly
(+)
position is upheld. The conditional capital gain, ̇ is therefore zero,
whereby substituting (15.22) into (15.21) and applying (15.7) yields
= () − = () − ≡ ∗ ≡ () − () ≡ () (15.23)
1−
where () is the equilibrium interest rate that would apply in case of per-
petual monopoly as in Model I.
Like in Model I, the equilibrium interest rate in Model II is thus from the
beginning a constant, ∗ . In view of 0 (15.23) shows that ∗ ()
Because of the limited duration of monopoly power in our present model, the
expected rate of return on investing in R&D is smaller than in the case of no
erosion of monopoly power as in Model I.
The description of the household sector is as in Model I, except that now
per capita financial wealth is
()
=
() ()
Indeed, now only = − firms have positive market value, namely
the firms that supply intermediate goods under monopolistic conditions. The
households’ first-order condition lead to the Keynes-Ramsey rule
̇ 1 ∗ 1
= ( − ) = (() − − ) ≡ ∗ () (15.24)
()
where is the per capita consumption growth rate from Model I, the case
of perpetual monopoly.
In order to have a model with growth, we assume parameters are such
that ∗ 0 In addition, to avoid unbounded utility and help fulfillment of
4
A more detailed argument is given in Chapter 14.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
CHAPTER 15. STOCHASTIC EROSION OF INNOVATOR’S
272 MONOPOLY POWER
the households’ transversality condition, we assume (1 − )∗ These two
conditions amount to the parameter restrictions
() + and (A1)
1
(1 − )∗ = (1 − ) (() − − ) (A2)
respectively, where () ≡ (1 − )−1 () 0 (from (15.7)), with () ≡
1
(((1 − )2 )) 0 (from (15.6)). The set of parameter combinations
satisfying these two conditions is not empty. Indeed, for arbitrary values of
and the parameters entering () choose for instance = 0 and 0 so
that (A1) is satisfied. Then ∗ 0 and (A2) is satisfied for any 1
(A1) requires that the “growth engine” of the economic system, as deter-
mined in particular by and , is “powerful enough” for growth to arise.
Below we return to what exactly constitutes the growth engine in this model.
Suffice it to say here that increases in and augment the strength of
the growth engine (thereby making (A1) more likely to hold) while a rise in
reduces the strength of the growth engine (thereby making (A1) less likely
to hold).5
15.6 Transitional dynamics*
Given that cessations of individual monopolies follow the assumed indepen-
dent Poisson processes with expiration rate , the aggregate number of tran-
sitions per time unit from monopoly to competitive status follow a Poisson
()
process with arrival rate The expected number of transitions per time
unit from monopoly to competitive status is then
() ()
̇ =
()
Assuming is “large”, the difference between actual and expected tran-
sitions per time unit will be negligible (by the law of large numbers), and we
simply write
() () ()
̇ = = ( − ) (15.25)
Let the fraction of intermediate goods supplied under competitive con-
()
ditions be denoted ≡ ( for “share”) and let ≡ ̇ for any
positively-valued variable Then, taking logs and differentiating w.r.t.
5
In view of () ≡ (1−)−1 ((1−)2 )1 −1 this role of is due to 1−1
0.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
15.6. Transitional dynamics* 273
we get
()
−
= () − = ()
− = (−1
− 1) −
− ( + ) R 0
= −1 for Q (15.26)
+
where the second equality is implied by (15.25).
The general law of movement of is given by (15.5), which, together
with (15.2) and (15.13) and the definition ̃ ≡ implies that
n o
() ()
̇ = = ( − − ) = − ( () + () ) −
(Ã ! )
()
() () ()
= 1 + ((1 − )−(1−) − 1) − (( − )() + () ) −
( )
()
= (1 + ((1 − )−(1−) − 1) ) − () + (() − () ) ) − ̃
©¡ ¢
= 1 + ((1 − )−(1−) − 1) (() )1− − ()
£ ¤ ª
+ () − (1 − )−1 () ) − ̃ (by (15.8))
©
= (() )1− − ()
£ ¤ ª
+ ((1 − )−(1−) − 1)(() )1− − ((1 − )−1 − 1)() − ̃
≡ (1 + 2 − ̃ )
where the constants 1 and 2 are implicitly defined. The growth rate of
can thus be written
= (1 + 2 − ̃ ) (15.27)
We now construct the implied dynamic system in the endogenous vari-
ables and ̃ From (15.26) follows ̇ = − ( + ) which combined
with (15.27) yields
̇ = − ( + (1 + 2 − ̃ )) (15.28)
·
Similarly, from ̃ ≡ follows ̃ ̃ = − = ∗ − by (15.24). So,
·
̃ = (∗ − (1 + 2 − ̃ )) ̃ (15.29)
The differential equations (15.28) and (15.29) constitute a dynamic system
with two endogenous variables, and ̃ the first of which is predetermined
while the second is a jump variable (forward-looking variable).
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
CHAPTER 15. STOCHASTIC EROSION OF INNOVATOR’S
274 MONOPOLY POWER
15.7 Long-run growth
·
In a steady state (̇ = 0 = ̃ ), by definition of ̃ we must have =
where = ∗ cf. (15.24). In steady state, therefore, = ∗ Consequently,
in view of (15.26), the steady-state value of is
∗ = (15.30)
+ ∗
Finally, the steady-state value of ̃ is ̃∗ = (1 + 2 ∗ − ∗ )
In the steady state there is balanced growth in the sense that
()
, and grow at the same constant rate as namely the rate ∗ given in
(15.24). This follows from the constancy of ̃ and (≡ () ) in steady state
together with the expression (15.13) for the aggregate production function
in sector 1.
Moreover, the total supply of intermediate goods per time unit in the
steady state is
() () () ()
= () + () = ( − )() + ()
£ ¤
= (1 − ∗ )() + ∗ ()
So, in the steady state, is proportional to . And by (15.2), the delivery
of basic goods to sector 2 is = per time unit, which is thus also
proportional to in the steady state. Hence, in steady state both and
grow at the same rate as the rate ∗
The same is true for the R&D investment. Indeed, in steady state,
∗
= ̇ = As shown in Appendix B, where also a phase diagram is
sketched, the steady state is a saddle point. An only half-finished dynamic
()
analysis in that appendix suggests that for any given initial 0 0 ∈ (0 1)
there exists a unique solution to the model and it converges to the steady
state for → ∞, that is, saddle-point stability prevails.
So also Model II generates fully endogenous growth. The long-run per
capita growth rate equals ∗ defined in (15.24). What makes fully endoge-
nous growth possible is again that the “growth engine” of the economy fea-
tures constant returns to scale w.r.t. producible inputs. Recall that the
growth engine of a model is defined as the set of input-producing sectors
using their own output as input. The present model can be reduced to two
()
sectors that make up the growth engine. Indeed, the factor in the
aggregate production function of sector 1 given in (15.13) can be written
³ ´1−
() ()
= ( () )1− = (15.31)
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
15.8. Economic policy 275
where () is the total input of intermediates in the production of ba-
()
sic goods in case of universal monopoly power as in Model I, and is
the corresponding required input of basic goods as raw material in its own
production, cf. (15.4). In this way sector 2 can be considered integrated
in sector 1. On this basis, the so delineated sector 1, together with sec-
tor 3, constitutes the growth engine of the present model. Basic goods,
= + + and technical knowledge, represented by the number, of
varieties of intermediate goods, are the two kinds of output that enter their
own production as inputs. Sector 1 delivers the input flow to itself and
the input flow to sector 3. And sector 3 delivers the input flow to sector
() ()
1. The production functions (15.13) (with = ∗ and written as
in (15.31)) and (15.5) show that in steady state there are constant returns
to scale w.r.t. these two producible inputs. It is this property that generates
fully endogenous growth in the model.
The long-run per capita growth rate depends on those parameters that
also appear in Model I in qualitatively the same way as in that model, see
Chapter 14. In Model II, however, the long-run per capita growth rate is
smaller than in Model I with perpetual monopolies, cf. (15.24). This is
due to the new parameter, the Poisson expiration rate Indeed, (15.24)
indicates that a larger i.e., a smaller expected duration, 1 of the status
as a monopolist, implies a lower per capita growth rate, ∗ The reason is that
the erosion of monopoly power implies less protection of private ownership
of the inventions. This reduces the private profitability of R&D and thereby
the incentive to do R&D.
15.8 Economic policy
At the theoretical level the analysis exposes the presence of static and dy-
namic distortions. Compared with perpetual monopoly, erosion of monopoly
power mitigates the static inefficiency problem arising from prices above mar-
ginal cost, as described in Section 15.3. But erosion of monopoly power ag-
gravates the underinvestment in R&D and thereby the dynamic distortion
in the system. In this way long-run growth (within these multiple-sector
AK-style models) is reduced even more, relative to the social optimum, than
in the case of perpetual monopolies.
At the empirical level, for instance Jones and Williams (1998) estimate
that R&D investment in the U.S. economy is only about a fourth of the social
optimum. So government intervention seems definitely motivated.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
CHAPTER 15. STOCHASTIC EROSION OF INNOVATOR’S
276 MONOPOLY POWER
A social planner
Letting () denote the growth rate under perpetual monopoly as in Model
I, we have µ ¶
∗ () 1 ()
= − (15.32)
1−
where () is the competitive supply of each intermediate-good type, defined
in (15.8), and is the optimal growth rate from the point of view of an
“all-knowing and all-powerful” social planner with the same criterion function
as that of the representative household. The first inequality in (15.32) was
shown above and the second is shown in Exercise VII.4. While the formal
derivation of the social planner’s solution is dealt with in that exercise, here
we shall consider the issues in more intuitive terms.
The first policy problem is that in the market economy, the invented
specialized intermediate goods are, at least to begin with, priced above the
private marginal cost, which is also the social marginal cost. Consequently,
under laissez faire, these goods are not supplied and used up to the point
where their marginal productivity equals their social marginal cost. A “free”
potential productivity gain is left unexploited in the economy.
A second problem is that this “static distortion” leads to a “dynamic
distortion”. Indeed, the fact that “too little” of the specialized intermediate
goods is demanded means that the market for each variety is “too small”.
This results in too little profits to the suppliers of these goods, hence too
little market value of inventions, that is, too little remuneration of the R&D
activity. Consequently, there is too little incentive to do R&D, and even the
growth rate () in model I ends up smaller than the social optimum. On
top of this comes in Model II that the imperfect protection of innovations
reduces the incentive to do R&D further, and the growth rate ends up even
lower than in Model I.
Returning to the static distortion, from the social planner’s point of view
the aggregate production function in the basic-goods sector can in the re-
duced form be written
= ( )1− = −(1−) 1−
which is analogue to (15.31). Given = + + for fixed the social
planner wants to choose the “raw material” input so as to maximize what
is left for final use, + i.e., consumption plus investment. The first-order
condition is
( − )
= (1 − )−(1−) − − 1 = 0
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
15.8. Economic policy 277
and obviously 2 ( − )(2 ) 0 Solving for gives
µ ¶1
(1 − )
= = ()
where the last equality comes from (15.8). This shows that society should, as
expected, supply each of the intermediate good types in the competitive
()
amount () rather than supply of them in the amount () () as in
Model II or, even worse, supply all intermediate good types in the amount
() as in Model I.
Policy instruments
To counteract the monopolist price distortion and encourage demand for
monopolized intermediate goods, a subsidy at constant rate to purchases of
monopolized intermediate goods will work. By setting = the monopoly
pricing is exactly neutralized from the point of view of the buyer who will
have to pay (1 − ) = (1 − )(1 − ) = which is the marginal cost of
supplying the good. This solves the static efficiency problem.
In Model I, solving this problem can be shown to automatically solve,
indirectly, also the dynamic efficiency problem. In Model II, solving the static
efficiency problem will also encourage R&D but, because of the imperfect
protection of innovations, not to the extent needed to get the optimal (first-
best) solution. A second policy instrument is needed. A direct stimulus in
the form of a subsidy to R&D investment is called for.
By comparing with the social planner’s allocation, it is possible to find
exact formulas for this R&D subsidy rate as well as non-distortionary financ-
ing such that the social planner’s allocation is implemented in a decentralized
way. Taxation on consumption and labor income are workable in these mod-
els.
Dilemmas in the design of patent systems
There are many dilemmas regarding how to design patent systems. Model
II above illustrates one of them, namely the question what the period length
of patents should be. The inverse of can be interpreted as a measure of
the average duration of patents. A larger (shorter duration) reduces static
inefficiency in an economy described by Model II but it also aggravates the
underinvestment in R&D and thereby increases the dynamic inefficiency in
the economy. We could more generally interpret as reflecting strictness of
antitrust policy and the conclusion would be similar.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
CHAPTER 15. STOCHASTIC EROSION OF INNOVATOR’S
278 MONOPOLY POWER
Going outside the present specific model, there are many further aspects
to take into account, e.g., spill-over effects of R&D and intensional knowledge
sharing, which we shall not consider here. A survey is contained in Hall and
Harhoff (2012). We end this chapter by a citation from Wikipedia (07-05-
2015):
Legal scholars, economists, scientists, engineers, activists, poli-
cymakers, industries, and trade organizations have held differing
views on patents and engaged in contentious debates on the sub-
ject. Recent criticisms primarily from the scientific community
focus on the core tenet of the intended utility of patents, as now
some argue they are retarding innovation. Critical perspectives
emerged in the nineteenth century, and recent debates have dis-
cussed the merits and faults of software patents, nanotechnology
patents and biological patents. These debates are part of a larger
discourse on intellectual property protection which also reflects
differing perspectives on copyright.
15.9 Appendix
A. Deriving (15.17) on the basis of Leibniz’s formula
We shall apply Leibniz’s formula 6 which says:
Z ()
() = ( ) =
()
Z ()
0 0 0 ( )
() = (() ) () − (() ) () +
()
In the present case we have from (15.16), = () (), where
Z ∞
() = − ( +)
whereby () = ∞ and () = , so that 0 () = 0 and 0 () = 1 We get
(+)
̇ = () 0 () that is,
(+)
Z ∞
̇ 0 − ( +)
()
= () = 0 − + − ( +) ( + )
= −1 + ( + ) () = −1 + ( + ) ()
6
For details, see for instance Sydsæter et al. (2008).
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
15.10. References 279
Reordering gives
(+)
() + ̇
= +
which is the no-arbitrage condition (15.17).
B. Stability analysis
The Jacobian matrix, evaluated in the steady state, is
" #
̇ ̇̃
∗ = · ·
̃ ̃̃ |(̃)=(∗ ̃∗ )
∙ ¸
−( + + 2 ∗ ) ∗
=
−2 ̃∗ ̃∗
The determinant of this matrix is
det ∗ = −( + + 2 ∗ )̃∗ + ∗ 2 ̃∗ = −( + )̃∗ 0
Hence, the eigenvalues are of opposite sign and the steady state is a saddle
point. A possible configuration of the phase diagram is sketched in Fig. 15.1.
In the steady state the TVC of the households is satisfied in that
() () ()
∗
∗ ∗ − −∗
− = − =− =
∗
(1 − ) −∗ (1 − ∗ )0 −∗
= = → 0 for → ∞
since ∗ ≡ () − so that (A2) combined with (15.24) implies ∗ ∗ The
TVC is therefore also satisfied along the unique converging path.
15.10 References
Barro, R. J., and X. Sala-i-Martin, 1995, Economic Growth. Second
edition, MIT Press: New York, 2004.
Hall, B.H., and D. Harhoff, 2012, Recent research on the economics of
patents, Annual Review of Economics, 4, 18.1-18.25.
Jones, C.I., and Williams, 1998, ... , Quarterly Journal of Economics.
Sydsæter, K., P. Hammond, A. Seierstad, and A. Strøm, 2008, Further
Mathematics for Economic Analysis, vol. II, Prentice-Hall: London.
Tirole, J., 1988, The Theory of Industrial Organization, MIT Press: New
York.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
CHAPTER 15. STOCHASTIC EROSION OF INNOVATOR’S
280 MONOPOLY POWER
c
s 0
c 0
c * E
s0 s
s* 1
Figure 15.1: Phase diagram.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
Chapter 16
Natural resources and
economic growth
In this course, up to now, the relationship between economic growth and the
earth’s finite natural resources has been touched upon in connection with: the
discussion of returns to scale (Chapter 2), the transition from a pre-industrial
to an industrial economy (in Chapter 7), the environmental problem of global
warming (Chapter 8), and the resource curse (in Chapter 13.4.3). In a more
systematic way the present chapter reviews how natural resources, including
the environment, relate to economic growth.
The contents are:
• Classification of means of production.
• The notion of sustainable development.
• Renewable natural resources.
• Non-renewable natural resources and exogenous technology growth.
• Non-renewable natural resources and endogenous technology growth.
• Natural resources and the issue of limits to economic growth.
The first two sections aim at establishing a common terminology for the
discussion.
16.1 Classification of means of production
We distinguish between different categories of production factors. First two
broad categories:
281
CHAPTER 16. NATURAL RESOURCES AND ECONOMIC
282 GROWTH
1. Producible means of production, also called man-made inputs.
2. Non-producible means of production.
The first category includes:
1.1 Physical inputs like processed raw materials, other intermediate goods,
machines, and buildings.
1.2 Human inputs of a produced character in the form of technical knowl-
edge (available in books, USB sticks etc.) and human capital.
The second category includes:
2.1 Human inputs of a non-produced character, sometimes called “raw la-
bor”.1
2.2 Natural resources. By definition in limited supply on this earth.
Natural resources can be sub-divided into:
2.2.1 Renewable resources, that is, natural resources the stock of which can
be replenished by a natural self-regeneration process. Hence, if the
resource is not over-exploited, it can be sustained in a more or less
constant amount. Examples: ground water, fertile soil, fish in the sea,
clean air, national parks.
2.2.2 Non-renewable resources, that is, natural resources which have no nat-
ural regeneration process (at least not within a relevant time scale).
The stock of a non-renewable resource is thus depletable. Examples:
fossil fuels, many non-energy minerals, virgin wilderness and endan-
gered species.
The climate change problem due to “greenhouse gasses” can be seen as
belonging to somewhere between category 2.2.1 or 2.2.2 in that the quality
of the atmosphere has a natural self-regeneration ability, but the speed of
regeneration is very low.
Given the scarcity of natural resources and the pollution problems caused
by economic activity, key issues are:
1
Outside a slave society, biological reproduction is usually not considered as part of the
economic sphere of society even though formation and maintainance of raw labor requires
child rearing, health, food etc. and is thus conditioned on economic circumstances.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
16.2. The notion of sustainable development 283
a. Is sustainable development possible?
b. Is sustainable economic growth (in a per capita welfare sense) possible?
c. How should a better “thermometer” for the evolution of the economy
than measurement of GNP be designed?
But first: what does “sustainable” and “sustainability” really mean”?
16.2 The notion of sustainable development
The basic idea in the notion of sustainable development is to emphasize
intergenerational responsibility. The Brundtland Commission (1987) defined
sustainable development as “development that meets the needs of present
generations without compromising the ability of future generations to meet
theirs”.
In more standard economic terms we may define sustainable economic
development as a time path along which per capita “welfare” (somehow mea-
sured) remains non-decreasing across generations forever. An aspect of this is
that current economic activities should not impose significant economic risks
on future generations. The “forever” in the definition can not, of course, be
taken literally, but as equivalent to “for a very long time horizon”. We know
that the sun will eventually (in some billion years) burn out and consequently
life on earth will become extinct.
Our definition emphasizes welfare, which should be understood in a broad
sense, that is, as more or less synonymous with “quality of life”, “living condi-
tions”, or “well-being” (the term used in Smulders, 1995). What may matter
is thus not only the per capita amount of marketable consumption goods,
but also fundamental aspects like health, life expectancy, and enjoyment of
services from the ecological system: capability to lead a worthwhile life.
To make this more specific, consider preferences as represented by the
period utility function of a “typical individual”. Suppose two variables enter
as arguments, namely consumption, of a marketable produced good and
some measure, of the quality of services from the eco-system. Suppose
further that the period utility function is of CES form:2
£ ¤1
( ) = + (1 − ) 0 1 1 (16.1)
The parameter is called the substitution parameter. The elasticity of substi-
tution between the two goods is = 1(1 − ) 0 a constant. When → 1
2
CES stands for Constant Elasticity of Substitution.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
CHAPTER 16. NATURAL RESOURCES AND ECONOMIC
284 GROWTH
(from below), the two goods become perfect substitutes (in that → ∞).
The smaller is the less substitutable are the two goods. When 0 we
have 1 and as → −∞ the indifference curves become near to right
angled.3 According to many environmental economists, there are good rea-
sons to believe that 1, since water, basic foodstuff, clean air, absence of
catastrophic climate change, etc. are difficult to replace by produced goods
and services. In this case there is a limit to the extent to which a rising ,
obtainable through a rising per capita income, can compensate for falling
At the same time the techniques by which the consumption good is cur-
rently produced may be “dirty” and thereby cause a falling . An obvious
policy response is the introduction of pollution taxes that give an incentive
for firms (or households) to replace these techniques (or goods) with cleaner
ones. For certain forms of pollution (e.g., sulfur dioxide, SO2 in the air) there
is evidence of an inverted U-curve relationship between the degree of pollu-
tion and the level of economic development measured by GDP per capita −
the environmental Kuznets curve.4
So an important element in sustainable economic development is that the
economic activity of current generations does not spoil the environmental
conditions for future generations. Living up to this requirement necessitates
economic and environmental strategies consistent with the planet’s endow-
ments. This means recognizing the role of environmental constraints for eco-
nomic development. A complicating factor is that specific abatement policies
vis-a-vis particular environmental problems may face resistance from interest
groups, thus raising political-economics issues.
As defined, a criterion for sustainable economic development to be present
is that per capita welfare remains non-decreasing across generations. A sub-
category of this is sustainable economic growth which is present if per capita
welfare is growing across generations. Here we speak of growth in a welfare
sense, not in a physical sense. Permanent exponential per capita output
3
By L’Hôpital’s rule for “0/0” it follows that, for fixed and
£ ¤1
lim + (1 − ) = 1−
→06=0
So the Cobb-Douglas function, which has elasticity of substitution between the goods equal
to 1, is an intermediate case, corresponding to = 0. More technical details in Appendix
A, albeit from the perspective of production rather than preferences.
4
See, e.g., Grossman and Krueger (1995). Others (e.g., Perman and Stern, 2003) claim
that when paying more serious attention to the statistical properties of the data, the
environmental Kuznets curve is generally rejected. Important examples of pollutants ac-
companied by absence of an environmental Kuznets curve include waste storage, reduction
of biodiversity, and emission of CO2 to the atmosphere. A very serious problem with the
latter is that emissions from a single country is spread all over the globe.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
16.2. The notion of sustainable development 285
growth in a physical sense is of course not possible with limited natural re-
sources (matter or energy). The issue about sustainable growth is whether,
by combining the natural resources with man-made inputs (knowledge, hu-
man capital, and physical capital), an output stream of increasing quality,
and therefore increasing economic value, can be maintained. In modern times
capabilities of many digital electronic devices provide conspicuous examples
of exponential growth in quality (or efficiency). Think of processing speed,
memory capacity, and efficiency of electronic sensors. What is known as
Moore’s Law is the rule of thumb that there is a doubling of the efficiency
of microprocessors within every 18 months. The evolution of the internet
has provided much faster and widened dissemination of information and fine
arts.
Of course there are intrinsic difficulties associated with measuring sustain-
ability in terms of well-being. There now exists a large theoretical and applied
literature dealing with these issues. A variety of extensions and modifications
of the standard national income accounting GNP has been developed under
the heading Green NNP (green net national product). An essential feature
in the measurement of Green NNP is that from the conventional GDP (which
essentially just measures the level of economic activity) is subtracted the de-
preciation of not only the physical capital but also the environmental assets.
The latter depreciate due to pollution, overburdening of renewable natural
resources, and depletion of reserves of non-renewable natural resources.5 In
some approaches the focus is on whether a comprehensive measure of wealth
is maintained over time. Along with reproducible assets and natural assets
(including the damage to the atmosphere from “greenhouse gasses”), Arrow
et al. (2012) include health, human capital, and “knowledge capital” in their
measure of “wealth”. They apply this measure in a study of the United
States, China, Brazil, India, and Venezuela over the period 1995-2000. They
find that all five countries over this period satisfy the sustainability criterion
of non-decreasing wealth in this broad sense. Indeed the wealth measure
referred to is found to be growing in all five countries − in the terminol-
ogy of the field positive “genuine saving” has taken place.6 Note that it is
sustainability that is claimed, not optimality.
5
The depreciation of these environmental and natural assets is evaluated in terms of
the social planner’s shadow prices. See, e.g., Heal (1998), Weitzman (2001, 2003), and
Stiglitz et al. (2010).
6
Of course, many measurement uncertainties and disputable issues of weighting are
involved; brief discussions, and questioning, of the study are contained in Solow (2012),
Hamilton (2012), and Smulders (2012). Regarding Denmark 1990-2009, a study by Lind
and Schou (2013), along lines somewhat similar to those of Arrow et al. (2012), also
suggests sustainability to hold.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
CHAPTER 16. NATURAL RESOURCES AND ECONOMIC
286 GROWTH
In the next two sections we will go more into detail with the challenge
to sustainability and growth coming from renewable and non-renewable re-
sources, respectively. We shall primarily deal with the issues from the point
of view of technical feasibility of non-decreasing, and possibly rising, per-
capita consumption. Concerning questions about appropriate institutional
regulation the reader is referred to the specialized literature.
We begin with renewable resources.
16.3 Renewable resources
A useful analytical tool is the following simple model of the stock dynamics
associated with a renewable resource.
Let ≥ 0 denote the stock of the renewable resource at time (so in this
chapter is not our symbol for saving). Then we may write
̇ ≡ = − = ( ) − 0 0 given, (16.2)
where is the self-regeneration of the resource per time unit and ≥ 0
is the resource extraction (and use) per time unit at time . If for instance
the stock refers to the number of fish in the sea, the flow represents the
number of fish caught per time unit. And if, in a pollution context, the stock
refers to “cleanness” of the air in cities, measures, say, the emission of
sulfur dioxide, SO2 , per time unit. The self-regenerated amount per time
unit depends on the available stock through the function ( ) known as a
self-regeneration function.7
Let us briefly consider the example where stands for the size of a fish
population in the sea. Then the self-regeneration function will have a bell-
shape as illustrated in the upper panel of Figure 16.1. Essentially, the self-
regeneration ability is due to the flow of solar energy continuously entering
the the eco-system of the earth. This flow of solar energy is constant and
beyond human control.
The size of the stock at the lower intersection of the () curve with the
horizontal axis is (0) ≥ 0 Below this level, even with = 0 there are too
few female fish to generate offspring, and the population necessarily shrinks
and eventually reaches zero. We may call (0) the minimum sustainable
stock.
At the other intersection of the () curve with the horizontal axis, ̄(0)
represents the maximum sustainable stock. The eco-system cannot support
7
The equation (16.2) also covers the case where represents the stock of a non-
renewable resource if we impose () ≡ 0 i.e., there is no self-regeneration.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
16.3. Renewable resources 287
MSY
G( S )
G (S ) R
O S
S (0) SMSY S (0)
S
G (S ) R
O S
SMSY S (R )
G ( S ) MSY
R
Figure 16.1: The self-generation function (upper panel) and stock dynamics for a
given = ̄ ∈ (0 ] (lower panel).
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
CHAPTER 16. NATURAL RESOURCES AND ECONOMIC
288 GROWTH
further growth in the fish population. The reason may be food scarcity,
spreading of diseases because of high population density, and easiness for
predators to catch the considered fish species and themselves expand. Pop-
ular mathematical specifications of (·) include the logistic function ()
= (1 − ) where 0 0 and the quasi-logistic function () =
(1 − )( − 1) where also 0 In both cases ̄(0) = but (0)
equals 0 in the first case and in the second.
The value indicated on the vertical axis in the upper panel, equals
max (). This value is thus the maximum sustainable yield per time unit.
This yield is sustainable from time 0 , provided the fish population is at time
0 at least of size = arg max () which is that value of where ()
= The size, of the fish population is consistent with maintaining
the harvest per time unit forever in a steady state.
The lower panel in Figure 16.1 illustrates the dynamics in the ( ̇)
plane, given a fixed rate of resource extraction = ̄ ∈ (0 ]. The
arrows indicate the direction of movement in this plane. In the long run, if
= ̄ for all the stock will settle down at the size ̄(̄) The stippled
curve in the upper panel indicates () − ̄ which is the same as ̇ in the
lower panel when = ̄. The stippled curve in the lower panel indicates the
dynamics in case = . In this case the steady-state stock, ̄( ) =
, is unstable. Indeed, a small negative shock to the stock will not lead
to a gradual return but to a self-reinforcing reduction of the stock as long as
the extraction = is maintained.
Note that is an ecological maximum and not necessarily in any
sense an economic optimum. Indeed, since the search and extraction costs
may be a decreasing function of the fish density in the sea, hence of the
stock of fish, it may be worthwhile to increase the stock beyond , thus
settling for a smaller harvest per time unit. Moreover, a fishing industry
cost-benefit analysis may consider maximization of the discounted expected
aggregate profits per time unit, taking into account the expected evolution
of the market price of fish, the cost function, and the dynamic relationship
(16.2).
In addition to its importance for regeneration, the stock, may have
amenity value and thus enter the instantaneous utility function. Then again
some conservation of the stock over and above may be motivated.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
16.3. Renewable resources 289
A dynamic model with a renewable resource and focus on technical
feasibility Consider a simple model consisting of (16.2) together with
= ( ) ≥ 0
̇ = − − ≥ 0 0 0 given,
= 0 ≥ 0 0 0 given, (16.3)
where is aggregate output and , and are inputs of capital, la-
bor, and a renewable resource, respectively, per time unit at time Let the
aggregate production function, be neoclassical8 with constant returns to
scale w.r.t. and The assumption ≥ 0 represents exogenous
technical progress. Further, is aggregate consumption (≡ where
is per capita consumption) and denotes a constant rate of capital depreci-
ation. There is no distinction between employment and population, . The
population growth rate, is assumed constant.
Is sustainable economic development in this setting technically feasible?
By definition, the answer will be yes if non-decreasing per capita consumption
can be sustained forever. From economic history we know of examples of
“tragedy of the commons”, like over-grazing of unregulated common land. As
our discussion is about technical feasibility, we assume this kind of problem
is avoided by appropriate institutions.
Suppose the use of the renewable resource is kept constant at a sustainable
level ̄ ∈ (0 ). To begin with, suppose = 0 so that = for all
≥ 0 Assume that at = ̄ the system is “productive” in the sense that
lim ( ̄ 0) lim ( ̄ 0) (A1)
→0 →∞
This condition is satisfied in Figure 16.2 where the value ̄ has the property
(̄ ̄ 0) = ̄ Given the circumstances, this value is the least upper
bound for a sustainable capital stock in the sense that
if ≥ ̄ we have ̇ 0 for any 0;
if 0 ̄ we have ̇ = 0 for = ( ̄ 0) − 0
For such a illustrated in Figure 16.2, a constant = ( ̄ 0) is main-
tained forever which implies non-decreasing per-capita income, ≡ ,
forever. So, in spite of the limited availability of the natural resource, a non-
decreasing level of consumption is technically feasible even without technical
progress. A forever growing level of consumption will, of course, require
sufficient technical progress capable of substituting for the natural resource.
8
That is, marginal productivities of the production factors are positive, but diminishing,
and the upper contour sets are strictly convex.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
CHAPTER 16. NATURAL RESOURCES AND ECONOMIC
290 GROWTH
Y
K
F ( K , L, R , 0)
O K
K K
Figure 16.2: Sustainable consumption in the case of = 0 and no technical progress
( and ̄ fixed).
Now consider the case 0 and assume CRS w.r.t. and In
view of CRS, we have
̄
1 = ( ) (16.4)
Along a balanced growth path with positive gross saving (if it exists) we
know that and must be constant, cf. the balanced growth equiva-
lence theorem of Chapter 4. Maintaining (= ( )( )) constant
along such a path, requires that is constant and thereby that grows
at the rate But then ̄ will be declining over time. To compensate
for this in (16.4), sufficient technical progress is necessary. This necessity of
course is present, a fortiori, for sustained growth in per-capita consumption
to occur.
As technical progress in the far future is by its very nature uncertain and
unpredictable, there can be no guarantee for sustained per capita growth if
there is sustained population growth.
Pollution As hinted at above, the concern that certain production meth-
ods involve pollution is commonly incorporated into economic analysis by
subsuming environmental quality into the general notion of renewable re-
sources. In that context in (16.2) and Figure 16.1 will represent the “level
of environmental quality” and will be the amount of dirty emissions per
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
16.4. Non-renewable resources: The DHSS model 291
time unit. Since the level of the environmental quality is likely to be an
argument in both the utility function and the production function, again
some limitation of the “extraction” (the pollution flow) is motivated. Pol-
lution taxes may help to encourage abatement activities and make technical
innovations towards cleaner production methods more profitable.
16.4 Non-renewable resources: The DHSS model
Whereas extraction and use of renewable resources can be sustained at a
more or less constant level (if not too high), the situation is different with
non-renewable resources. They have no natural regeneration process (at least
not within a relevant time scale) and so continued extraction per time unit
of these resources will inevitably have to decline and approach zero in the
long run.
To get an idea of the implications, we will consider the Dasgupta-Heal-
Solow-Stiglitz model (DHSS model) from the 1970s.9 The production side of
the model is described by:
= ( ) ≥ 0 (16.5)
̇ = − − ≥ 0 0 0 given, (16.6)
̇ = − ≡ − 0 0 given, (16.7)
= 0 ≥ 0 (16.8)
The new element is the replacement of (16.2) with (16.7), where is the
stock of the non-renewable resource (e.g., oil reserves), and is the depletion
rate. Since we must have ≥ 0 for all there is a finite upper bound on
cumulative resource extraction:
Z ∞
≤ 0 (16.9)
0
Since the resource is non-renewable, no re-generation function appears in
(16.7). Uncertainty is ignored and the extraction activity involves no costs.10
As before, there is no distinction between employment and population, .
The model was formulated as a response to the pessimistic Malthusian
views expressed in the book The Limits to Growth written by MIT ecolo-
gists Meadows et al. (1972).11 Stiglitz, and fellow economists, asked the
9
See, e.g., Stiglitz, 1974.
10
This simplified description of resource extraction is the reason that it is common
to classify the model as a one-sector model, notwithstanding there are two productive
activities in the economy, manufacturing and resource extraction.
11
An up-date came in 2004, see Meadows at al. (2004).
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
CHAPTER 16. NATURAL RESOURCES AND ECONOMIC
292 GROWTH
question: what are the technological conditions needed to avoid falling per
capita consumption in the long run in spite of the inevitable decline in the
use of non-renewable resources? The answer is that there are three ways in
which this decline in resource use may be counterbalanced:
1. input substitution;
2. resource-augmenting technical progress;
3. increasing returns to scale.
Let us consider each of them in turn (although in practice the three
mechanisms tend to be intertwined).
16.4.1 Input substitution
By input substitution is here meant the gradual replacement of the input of
the exhaustible natural resource by man-made input, capital. Substitution
of fossil fuel energy by solar, wind, tidal and wave energy resources is an
example. Similarly, more abundant lower-grade non-renewable resources can
substitute for scarce higher-grade non-renewable resources - and this will
happen when the scarcity price of these has become sufficiently high. A
rise in the price of a mineral makes a synthetic substitute cost-efficient or
lead to increased recycling of the mineral. Finally, the composition of final
output can change toward goods with less material content. Overall, capital
accumulation can be seen as the key background factor for such substitution
processes (though also the arrival of new technical knowledge may be involved
- we come back to this).
Whether capital accumulation can do the job depends crucially on the
degree of substitutability between and To see this, let the produc-
tion function be a three-factor CES production function. Suppressing the
explicit dating of the variables when not needed for clarity, we have.
¡ ¢1
= 1 + 2 + 3 1 2 3 0 1 +2 +3 = 1 1 6= 0
(16.10)
We omit the the time index on and when not needed for clarity.
The important parameter is the substitution parameter. Let denote the
cost to the firm per unit of the resource flow and let ̂ be the cost per unit of
capital (generally, ̂ = + where is the real rate of interest). Then ̂
is the relative factor price, which may be expected to increase as the resource
becomes more scarce. The elasticity of substitution between and can be
measured by [()( ̂)] ( ̂ )() evaluated along an isoquant
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
16.4. Non-renewable resources: The DHSS model 293
curve, i.e., the percentage rise in the - ratio that a cost-minimizing firm
will choose in response to a one-percent rise in the relative factor price, ̂
Since we consider a CES production function, this elasticity is a constant
= 1(1 − ) 0 Indeed, the three-factor CES production function has the
property that the elasticity of substitution between any pair of the three
production factors is the same.
First, suppose 1 i.e., 0 1 Then, for fixed and →
¡ ¢1
1 + 2 0 when → 0 In this case of high substitutability the
resource is seen to be inessential in the sense that it is not necessary for a
positive output. That is, from a production perspective, conservation of the
resource is not vital.
Suppose instead 1 i.e., 0 Although increasing when decreases,
output per unit of the resource flow is then bounded from above. Conse-
quently, the finiteness of the resource inevitably implies doomsday sooner or
later if input substitution is the only salvage mechanism. To see this, keeping
and fixed, we get
∙ ¸1
− 1
= ( ) = 1 ( ) + 2 ( ) + 3 → 3 1 for → 0
(16.11)
since 0 Even if and are increasing, lim→0 = lim→0 ( )
1
= 3 · 0 = 0 Thus, when substitutability is low, the resource is essential
in the sense that output is nil in the absence of the resource.
What about the intermediate case = 1? Although (16.10) is not de-
fined for = 0 using L’Hôpital’s rule (as for the two-factor function, cf.
¡ ¢1
Appendix), it can be shown that 1 + 2 + 3 → 1 2 3
for → 0 In the limit a three-factor Cobb-Douglas function thus appears.
This function has = 1 corresponding to = 0 in the formula = 1(1−)
The circumstances giving rise to the resource being essential thus include the
Cobb-Douglas case = 1
The interesting aspect of the Cobb-Douglas case is that it is the only
case where the resource is essential while at the same time output per unit
of the resource is unbounded from above (since = 1 2 3 −1 → ∞
for → 0).12 Under these circumstances it was considered an open question
whether non-decreasing per capita consumption could be sustained. There-
fore the Cobb-Douglas case was studied intensively. For example, Solow
(1974) showed that if = = 0, then a necessary and sufficient condition
that a constant positive level of consumption can be sustained is that 1 3
12
To avoid misunderstanding: by “Cobb-Douglas case” we refer to any function where
enters in a “Cobb-Douglas fashion”, i.e., any function like = ̃ ( )1−3 3
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
CHAPTER 16. NATURAL RESOURCES AND ECONOMIC
294 GROWTH
This condition in itself seems fairly realistic, since, empirically, 1 is many
times the size of 3 (Nordhaus and Tobin, 1972, Neumayer 2000). Solow
added the observation that under competitive conditions, the highest sus-
tainable level of consumption is obtained when investment in capital exactly
equals the resource rent, · This result was generalized in Hartwick
(1977) and became known as Hartwick’s rule. If there is population growth
( 0) however, not even the Cobb-Douglas case allows sustainable per
capita consumption unless there is sufficient technical progress, as equation
(16.15) below will tell us.
Neumayer (2000) reports that the empirical evidence on the elasticity of
substitution between capital and energy is inconclusive. Ecological econo-
mists tend to claim the poor substitution case to be much more realistic
than the optimistic Cobb-Douglas case, not to speak of the case 1 This
invites considering the role of technical progress.
16.4.2 Technical progress
Solow (1974) and Stiglitz (1974) analyzed the theoretical possibility that
resource-augmenting technological change can overcome the declining use of
non-renewable resources that must be expected in the future. The focus is
not only on whether a non-decreasing consumption level can be maintained,
but also on the possibility of sustained per capita growth in consumption.
New production techniques may raise the efficiency of resource use. For
example, Dasgupta (1993) reports that during the period 1900 to the 1960s,
the quantity of coal required to generate a kilowatt-hour of electricity fell
from nearly seven pounds to less than one pound.13 Further, technological
developments make extraction of lower quality ores cost-effective and make
more durable forms of energy economical. On this background we incorporate
resource-augmenting technical progress at the rate 3 and also allow labor-
augmenting technical progress at the rate 2 So the CES production function
now reads ³ ´ 1
= 1 + 2 (2 ) + 3 (3 ) (16.12)
where 2 = 2 and 3 = 3 considering 2 ≥ 0 and 3 0 as exogenous
constants. If the (proportionate) rate of decline of is kept smaller than
3 then the “effective resource” input is no longer decreasing over time. As
a consequence, even if 1 (the poor substitution case), the finiteness of
nature need not be an insurmountable obstacle to non-decreasing per capita
consumption.
13
For a historical account of energy technology, see Smil (1994).
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
16.4. Non-renewable resources: The DHSS model 295
Actually, a technology with 1 needs a considerable amount of resource-
augmenting technical progress to obtain compliance with the empirical fact
that the income share of natural resources has not been rising (Jones, 2002).
When 1 market forces tend to increase the income share of the factor
that is becoming relatively more scarce. Empirically, and have
increased systematically. However, with a sufficiently increasing 3 , the in-
come share need not increase in spite of 1 Compliance with
Kaldor’s “stylized facts” (more or less constant growth rates of and
and stationarity of the output-capital ratio, the income share of labor,
and the rate of return on capital) can be maintained with moderate labor-
augmenting technical change (2 growing over time). The motivation for not
allowing a rising 1 and replacing in (16.12) by 1 is that this would be
at odds with Kaldor’s “stylized facts”, in particular the absence of a trend
in the rate of return to capital.
With 3 2 + we end up with conditions allowing a balanced growth
path (BGP for short), which we in the present context, with an essential
resource, define as a path along which the quantities and are
positive and change at constant proportionate rates (some or all of which
may be negative). Given (16.12), it can be shown that along a BGP with
positive gross saving, (2 ) is constant and so = 2 (hence also
= 2 )14 There is thus scope for a positive if 0 2 3 −
Of course, one thing is that such a combination of assumptions allows for
an upward trend in per capita consumption - which is what we have seen
since the industrial revolution. Another thing is: will the needed assump-
tions be satisfied for a long time in the future? Since we have considered
exogenous technical change, there is so far no hint from theory. But, even
taking endogenous technical change into account, there will be many uncer-
tainties about what kind of technological changes will come through in the
future and how fast.
Balanced growth in the Cobb-Douglas case
The described results go through in a more straightforward way in the Cobb-
Douglas case. So let us consider this. A convenience is that capital-augmenting,
labor-augmenting, and resource-augmenting technical progress become indis-
tinguishable and can thus be merged into one technology variable, the total
factor productivity :
= 1 2 3 1 2 3 0 1 + 2 + 3 = 1 (16.13)
14
See Appendix B.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
CHAPTER 16. NATURAL RESOURCES AND ECONOMIC
296 GROWTH
where we assume that is growing at some constant rate 0. This,
together with (16.6) - (16.8), is now the model under examination.
Log-differentiating w.r.t. time in (16.13) yields the growth-accounting
relation
= + 1 + 2 + 3 (16.14)
In Appendix B it is shown that along a BGP with positive gross saving the
following holds:
(i) = = ≡ + ;
(ii) = ≡ ̇ ≡ − ≡ −̄ = constant 0;
(iii) nothing of the resource is left unutilized forever.
With constant depletion rate, denoted ̄ along the BGP, (16.14)
thus implies
1
= = ( − 3 − 3 ̄) (16.15)
1 − 1
since 1 + 2 − 1 = −3
Absent the need for input of limited natural resources, we would have
3 = 0 and so = (1 − 1 ) But with 3 0 the non-renewable resource
is essential and implies a drag on per capita growth equal to 3 (+̄)(1−1 ).
We get 0 if and only if 3 ( + ̄) (where, the constant depletion
rate, ̄ can in principle, from a social point of view, be chosen very small if
we want a strict conservation policy).
It is noteworthy that in spite of per-capita growth being due to exoge-
nous technical progress, (16.15) shows that there is scope for policy affecting
the long-run per-capita growth rate to the extent that policy can affect the
depletion rate in the opposite direction.15
“Sustained growth” in and should not be understood in a narrow
physical sense. As alluded to earlier, we have to understand broadly
as “produced means of production” of rising quality and falling material
intensity; similarly, must be seen as a composite of consumer “goods”
with declining material intensity over time (see, e.g., Fagnart and Germain,
2011). This accords with the empirical fact that as income rises, the share
of consumption expenditures devoted to agricultural and industrial products
declines and the share devoted to services, hobbies, sports, and amusement
increases. Although “economic development” is perhaps a more appropri-
ate term (suggesting qualitative and structural change), we retain standard
terminology and speak of “economic growth”.
In any event, simple aggregate models like the present one should be
seen as no more than a frame of reference, a tool for thought experiments.
15
Cf. Section 13.5.1 of Chapter 13.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
16.4. Non-renewable resources: The DHSS model 297
At best such models might have some validity as an approximate summary
description of a certain period of time. One should be aware that an economy
in which the ratio of capital to resource input grows without limit might
well enter a phase where technological relations (including the elasticity of
factor substitution) will be very different from now. For example, along any
economic development path, the aggregate input of non-renewable resources
must in the long run asymptotically approach zero. From a physical point of
view, however, there must be some minimum amount of the resource below
which it can not fulfil its role as a productive input. Thus, strictly speaking,
sustainability requires that in the “very long run”, non-renewable resources
become inessential.
A backstop technology We end this sub-section by a remark on a rather
different way of modeling resource-augmenting technical change. Dasgupta
and Heal (1974) present a model of resource-augmenting technical change,
considering it not as a smooth gradual process, but as something arriving in a
discrete once-for-all manner with economy-wide consequences. The authors
envision a future major discovery of, say, how to harness a lasting energy
source such that a hitherto essential resource like fossil fuel becomes inessen-
tial. The contour of such a backstop technology might be currently known, but
its practical applicability still awaits a technological breakthrough. The time
until the arrival of this breakthrough is uncertain and may well be long. In
Dasgupta, Heal and Majumdar (1977) and Dasgupta, Heal and Pand (1980)
the idea is pursued further, by incorporating costly R&D. The likelihood of
the technological breakthrough to appear in a given time interval depends
positively on the accumulated R&D as well as the current R&D. It is shown
that under certain conditions an index reflecting the probability that the
resource becomes unimportant acts like an addition to the utility discount
rate and that R&D expenditure begins to decline after some time. This is an
interesting example of an early study of endogenous technological change.16
16.4.3 Increasing returns to scale
The third circumstance that might help overcoming the finiteness of nature
is increasing returns to scale. For the CES function with poor substitution
( 1), however, increasing returns to scale, though helping, are not by
themselves sufficient to avoid doomsday. For details, see, e.g., Groth (2007).
16
A similar problem has been investigated by Kamien and Schwartz (1978) and Just et
al. (2005), using somewhat different approaches.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
CHAPTER 16. NATURAL RESOURCES AND ECONOMIC
298 GROWTH
16.4.4 Summary on the DHSS model
Apart from a few remarks by Stiglitz, the focus of the fathers of the DHSS
model is on constant returns to scale; and, as in the simple Solow and Ram-
sey growth models, only exogenous technical progress is considered. For our
purposes we may summarize the DHSS results in the following way. Non-
renewable resources do not really matter seriously if the elasticity of substi-
tution between them and man-made inputs is above one. If not, however,
then:
(a) absent technical progress, if = 1 sustainable per capita consump-
tion requires 1 3 and = 0 = ; otherwise, declining per capita
consumption is inevitable and this is definitely the prospect, if 1;
(b) on the other hand, if there is enough resource-augmenting and labor-
augmenting technical progress, non-decreasing per capita consumption
and even growing per capita consumption may be sustained;
(c) population growth, implying more mouths to feed from limited nat-
ural resources, exacerbates the drag on growth implied by a declining
resource input; indeed, as seen from (16.15), the drag on growth is
3 ( + )(1 − 1 ) along a BGP
16.4.5 An extended DHSS model
The above discussion of sustainable economic development in the presence
of non-renewable resources was carried out on the basis of the original DHSS
model with only capital, labor, and a non-renewable resource as inputs.
In practice the issues of input substitution and technological change are
to a large extent interweaved into the question of substitutability of non-
renewable with renewable resources. A more natural point of departure for
the discussion may therefore be an extended DHSS model of the form:
= ( ) ≥ 0
̇ = − − ≥ 0 0 0 given,
̇ = ( ) − 0 0 given,
̇ = − 0 0 given,
Z ∞
≤ 0
0
where is input of the renewable resource and the corresponding stock,
while is input of the non-renewable resource to which corresponds the
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
16.5. Endogenous technical change 299
stock . Only the non-renewable resource is subject to the constraint of a
finite upper bound on cumulative resource extraction.
Within such a framework a more or less gradual transition from use of
non-renewable energy forms to renewable energy forms (hydro-power, wind
energy, solar energy, biomass, and geothermal energy), likely speeded up
learning by doing as well as R&D, can be studied (see for instance Tahvonen
and Salo, 2001).
16.5 Endogenous technical change
The obvious next step is to examine how endogenizing technical change may
throw new light on the issues relating to non-renewable resources, in partic-
ular the visions (b) and (c). Because of the non-rival character of techni-
cal knowledge, endogenizing knowledge creation may have profound impli-
cations, in particular concerning point (c). Indeed, the relationship between
population growth and economic growth may be circumvented when endoge-
nous creation of ideas (implying a form of increasing returns to scale) is
considered.
16.5.1 A two-sector R&D-based model
We shall look at the economy from the perspective of a fictional social plan-
ner who cares about finding a resource allocation so as to maximize the
intertemporal utility function of a representative household subject to tech-
nical feasibility as given from the initial technology and initial resources.
In addition to cost-free resource extraction, there are two “production”
sectors, the manufacturing sector and the R&D sector. In the manufacturing
sector the aggregate production function is
= 0 + + + = 1 (16.16)
where is output of manufacturing goods, while , , and are inputs
of capital, labor, a non-renewable resource, and land (a renewable resource),
respectively, per time unit at time The amount of land is an exogenous
constant. Total factor productivity is where the variable is assumed
proportional to the stock of technical knowledge accumulated through R&D
investment. Due to this proportionality we can simply identify with the
stock of knowledge at time .
Aggregate manufacturing output is used for consumption, investment,
in physical capital, and investment, in R&D,
+ + =
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
CHAPTER 16. NATURAL RESOURCES AND ECONOMIC
300 GROWTH
Accumulation of capital occurs according to
̇ = − = − − − ≥ 0
0 0 given,
(16.17)
where is the (exogenous) rate of depreciation (decay) of capital.
In the R&D sector additions to the stock of technical knowledge are cre-
ated through R&D investment, :
̇ = − ≥ 0 0 0 given (16.18)
We allow for a positive depreciation rate, to take into account the possi-
bility that as technology advances, old knowledge becomes obsolete and then
over time gradually becomes useless in production.
Extraction of the non-renewable resource is again given by
̇ = − ≡ − 0 0 given, (16.19)
where is the stock of the non-renewable resource (e.g., oil reserves) and
is the depletion rate. Since we must have ≥ 0 for all there is a finite
upper bound on cumulative resource extraction:
Z ∞
≤ 0 (16.20)
0
Finally, population (= labor force) grows according to
= 0 ≥ 0 0 0 given.
Uncertainty is ignored and the extraction activity involves no costs.
We see that the setup is elementarily related to the simple lab-equipment
model (Chapter 14) since by investing part of the manufacturing output
new knowledge is directly generated without intervention by researchers and
similar.17 Note also that there are no intertemporal knowledge-spillovers.
16.5.2 Analysis
We now skip the explicit dating of the variables where not needed for clarity.
The model has three state variables, the stock, of physical capital, the
stock, of non-renewable resources, and the stock, of technical knowl-
edge. To simplify the dynamics, we will concentrate on the special case
17
An interpretation is that part of the activity in the manufacturing sector is directly
R&D activity using the same technology (production function) as is used in the production
of consumption goods and capital goods.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
16.5. Endogenous technical change 301
Figure 16.3: Initial complete specialization followed by balanced growth.
= = In this case, as we shall see, after an initial adjustment pe-
riod, the economy behaves in many respects similarly to a reduced-form AK
model.
Let us first consider efficient paths, i.e., paths such that aggregate con-
sumption can not be increased in some time interval without being decreased
in another time interval. The net marginal productivities of and are
equal if and only if − = − i.e., if and only if
=
The initial stocks, 0 and 0 are historically given. Suppose 0 0
as in Figure 16.3. Then, initially, the net marginal productivity of capital is
larger than that of knowledge, i.e., capital is relatively scarce. An efficient
economy will therefore for a while invest only in capital, i.e., there will be a
phase where = 0 This phase of complete specialization lasts until =
a state reached in finite time, say at time ̄, cf. the figure. Hereafter,
there is investment in both assets so that their ratio remains equal to the
efficient ratio forever. Similarly, if initially 0 0 then there will
be a phase of complete specialization in R&D, and after a finite time interval
the efficient ratio = is achieved and maintained forever.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
CHAPTER 16. NATURAL RESOURCES AND ECONOMIC
302 GROWTH
For ̄ it is thus as if there were only one kind of capital, which we
may call “broad capital” and define as ̃ = + = ( + ) Indeed,
substitution of = and = ̃( + ) into (16.16) gives
= ̃ + ≡ ̃ ̃ ̃ ≡ + (16.21)
( + )+
so that ̃ + + 1 Further, adding (16.18) and (16.17) gives
·
̃ = ̇ + ̇ = − − ̃ (16.22)
where is per capita consumption.
We now proceed with a model based on broad capital, using (16.21),
(16.22) and the usual resource depletion equation (16.19). Essentially, this
model amounts to an extended DHSS model allowing increasing returns to
scale, thereby offering a simple framework for studying endogenous growth
with essential non-renewable resources.
We shall focus on questions like:
1 Is sustainable development (possibly even growth) possible within the
model?
2 Can the utilitarian principle of discounted utility maximizing possibly
clash with a requirement of sustainability? If so, under what condi-
tions?
3 How can environmental policy be designed so as to enhance the prospects
of sustainable development or even sustainable economic growth?
Balanced growth
Log-differentiating (16.21) w.r.t. gives the “growth-accounting equation”
= ̃̃ + + (16.23)
Hence, along a BGP we get
(1 − ̃) + = (̃ + − 1) (16.24)
Since 0 it follows immediately that:
Result (i) A BGP has 0 if and only if
(̃ + − 1) 0 or ̃ 1 (16.25)
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
16.5. Endogenous technical change 303
Proof. Since 0 (*) implies (1 − ̃) (̃ + − 1) Hence, if
0 either ̃ 1 or (̃ ≤ 1 and (̃ + − 1) 0) This proves “only
if”. The “if” part is more involved but follows from Proposition 2 in Groth
(2004). ¤
Result (i) tells us that endogenous growth is theoretically possible, if there
are either increasing returns to the capital-cum-labor input combined with
population growth or increasing returns to capital (broad capital) itself. At
least one of these conditions is required in order for capital accumulation to
offset the effects of the inescapable waning of resource use over time. Based
on Nordhaus (1992), ≈ 02 ≈ 06 ≈ 01 and ≈ 01 seem reasonable.
Given these numbers,
(i) semi-endogenous growth requires ( + + − 1) 0 hence 020;
(ii) fully endogenous growth requires + 1 hence 080
We have defined fully endogenous growth to be present if the long-run
growth rate in per capita output is positive without the support of growth in
any exogenous factor. Result (i) shows that only if ̃ 1 is fully endogenous
growth possible. Although the case ̃ 1 has potentially explosive effects
on the economy, if ̃ is not too much above 1, these effects can be held back
by the strain on the economy imposed by the declining resource input.
In some sense this is “good news”: fully endogenous steady growth is the-
oretically possible and no knife-edge assumption is needed. As we have seen
in earlier chapters, in the conventional framework without non-renewable re-
sources, fully endogenous growth requires constant returns to the producible
input(s) in the growth engine. In our one-sector model the growth engine
is the manufacturing sector itself, and without the essential non-renewable
resource, fully endogenous growth would require the knife-edge condition
̃ = 1 (̃ being above 1 is excluded in this case, because it would lead to
explosive growth in a setting without some countervailing factor). When
non-renewable resources are an essential input in the growth engine, a drag
on the growth potential is imposed. To be able to offset this drag, fully
endogenous growth requires increasing returns to capital.
The “bad news” is, however, that even in combination with essential non-
renewable resources, an assumption of increasing returns to capital seems too
strong and too optimistic. A technology having ̃ just slightly above 1 can
sustain any per capita growth rate − there is no upper bound on .18 This
appears overly optimistic.
This leaves us with semi-endogenous growth as the only plausible form
of endogenous growth (as long as is not endogenous). Indeed, Result (i)
18
This is shown in Groth (2004).
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
CHAPTER 16. NATURAL RESOURCES AND ECONOMIC
304 GROWTH
indicates that semi-endogenous growth corresponds to the case 1− ̃ ≤ 1
In this case sustained positive per capita growth driven by some internal
mechanism is possible, but only if supported by 0 that is, by growth in
an exogenous factor, here population size.
Growth policy and conservation
Result (i) is only about whether the technology as such allows a positive per
capita growth rate or not. What about the size of the growth rate? Can
the growth rate temporarily or perhaps permanently be affected by economic
policy? The simple growth-accounting relation (16.24) immediately shows:
Result (ii) Along a BGP, policies that decrease (increase) the depletion
rate (and only such policies) will increase (decrease) the per capita
growth rate (here we presuppose ̃ 1 the plausible case).
This observation is of particular interest in view of the fact that chang-
ing the perspective from exogenous to endogenous technical progress implies
bringing a source of numerous market failures to light. On the face of it,
the result seems to run against common sense. Does high growth not im-
ply fast depletion (high )? Indeed, the answer is affirmative, but with the
addition that exactly because of the fast depletion such high growth will
only be temporary − it carries the seeds to its own obliteration. For faster
sustained growth there must be sustained slower depletion. The reason for
this is that with protracted depletion, the rate of decline in resource input
becomes smaller. Hence, so does the drag on growth caused by this decline.
As a statement about policy and long-run growth, (ii) is a surprisingly
succinct conclusion. It can be clarified in the following way. For policy to
affect long-run growth, it must affect a linear differential equation linked to
the basic goods sector in the model. In the present framework the resource
depletion relation,
̇ = −
is such an equation. In balanced growth = − ≡ − is constant,
so that the proportionate rate of decline in must comply with, indeed be
equal to, that of Through the growth accounting relation (16.23), given
this fixes and ̃ (equal in balanced growth), hence also = − .
The conventional wisdom in the endogenous growth literature is that
interest income taxes impede economic growth and investment subsidies pro-
mote economic growth. Interestingly, this need not be so when non-renewable
resources are an essential input in the growth engine (which is here the man-
ufacturing sector itself). At least, starting from a Cobb-Douglas aggregate
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
16.5. Endogenous technical change 305
production function as in (16.16), it can be shown that only those policies
that interfere with the depletion rate in the long run (like a profits tax on
resource-extracting companies or a time-dependent tax on resource use) can
affect long-run growth. It is noteworthy that this long-run policy result holds
whether 0 or not and whether growth is exogenous, semi-endogenous
or fully endogenous.19 The general conclusion is that with non-renewable
resources entering the growth-generating sector in an essential way, conven-
tional policy tools receive a different role and there is a role for new tools
(affecting long-run growth through affecting the depletion rate).20
Introducing preferences
To be more specific we introduce household preferences and a social planner.
The resulting resource allocation will coincide with that of a decentralized
competitive economy if agents have perfect foresight and the government has
introduced appropriate subsidies and taxes. As in Stiglitz (1974a), let the
utilitarian social planner choose a path ( )∞
=0 so as to optimize
Z ∞ 1−
0 = − 0 (16.26)
0 1 −
subject to the constraints given by technology, i.e., (16.21), (16.22), and
(16.19), and initial conditions. The parameter 0 is the (absolute) elas-
ticity of marginal utility of consumption (reflecting the strength of the desire
for consumption smoothing) and is a constant rate of time preference.21
Using the Maximum Principle, the first-order conditions for this problem
lead to, first, the social planner’s Keynes-Ramsey rule,
1 1
= ( − − ) = (̃ − − ) (16.27)
̃ ̃
second, the social planner’s Hotelling rule,
()
= ( − ) = (̃ − ) (16.28)
̃ ̃
The Keynes-Ramsey rule says: as long as the net return on investment in
capital is higher than the rate of time preference, one should let current be
19
This is a reminder that the distinction between fully endogenous growth and semi-
endogenous growth is not the same as the distinction between policy-dependent and policy-
invariant growth.
20
These aspects are further explored in Groth and Schou (2006).
21
For simplicity we have here ignored (as does Stiglitz) that also environmental quality
should enter the utility function.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
CHAPTER 16. NATURAL RESOURCES AND ECONOMIC
306 GROWTH
low enough to allow positive net saving (investment) and thereby higher con-
sumption in the future. The Hotelling rule is a no-arbitrage condition saying
that the return (“capital gain”) on leaving the marginal unit of the resource
in the ground must equal the return on extracting and using it in produc-
tion and then investing the proceeds in the alternative asset (reproducible
capital).22
After the initial phase of complete specialization described above, we
have, due to the proportionality between and ̃ that = =
̃ = ̃ ̃ Notice that the Hotelling rule is independent of prefer-
ences; any path that is efficient must satisfy the Hotelling rule (as well as
the exhaustion condition lim→∞ () = 0).
Using the Cobb-Douglas specification, we may rewrite the Hotelling rule
as − = ̃ ̃ − Along a BGP = = + and = − so
that the Hotelling rule combined with the Ramsey rule gives
(1 − ) + = − (16.29)
This linear equation in and combined with the growth-accounting
relationship (16.24) constitutes a linear two-equation system in the growth
rate and the depletion rate. The determinant of this system is ≡ 1 − ̃ −
+ We assume 0 which seems realistic and is in any case necessary
(and sufficient) for stability.23 Then
(̃ + + − 1) −
= and (16.30)
[(̃ + − 1) − ] + (1 − ̃)
= (16.31)
To ensure boundedness from above of the utility integral (16.26) we need
the parameter restriction − (1 − ) which we assume satisfied for
as given in (16.30).
Interesting implications are:
22
After Hotelling (1931), who considered the rule in a market economy. Assuming
perfect competition, the real resource price is = and the real rate of interest is
= − . Then the rule takes the form ̇ = . If there are extraction costs at
rate ( ) then the rule takes the form ̇ − = , where is the price of
the unextracted resource (whereas = + ).
It is another matter that the rise in resource prices and the predicted decline in resource
use have not yet shown up in the data (Krautkraemer 1998, Smil 2003); this may be due
to better extraction technology and discovery of new deposits. But in the long run, if
non-renewable resources are essential, this tendency inevitably will be reversed.
23
As argued above, ̃ 1 seems plausible. Generally, is estimated to be greater than
one (see, e.g., Attanasio and Weber 1995); hence 0 The stability result as well as
other findings reported here are documented in Groth and Schou (2002).
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
16.5. Endogenous technical change 307
Result (iii) If there is impatience ( 0), then even when a non-negative
is technically feasible (i.e., (16.25) satisfied), a negative can be
optimal and stable.
Result (iv) Population growth is good for economic growth. In its absence,
when 0 we get 0 along an optimal BGP; if = 0 = 0
when = 0.
Result (v) There is never a scale effect on the growth rate.
Result (iii) reflects that utility discounting and consumption smoothing
weaken the “growth incentive”.
Result (iv) is completely contrary to the conventional (Malthusian) view
and the learning from the DHSS model. The point is that two offsetting
forces are in play. On the one hand, higher means more mouths to feed
and thus implies a drag on per capita growth (Malthus). On the other hand,
a growing labour force is exactly what is needed in order to exploit the
benefits of increasing returns to scale (anti-Malthus). And at least in the
present framework this dominates the first effect. This feature might seem
to be contradicted by the empirical finding that there is no robust correlation
between and population growth in cross-country regressions (Barro and
Sala-i-Martin 2004, Ch. 12). However, the proper unit of observation in this
context is not the individual country. Indeed, in an internationalized world
with technology diffusion a positive association between and as in (16.30)
should not be seen as a prediction about individual countries, but rather as
pertaining to larger regions, perhaps the global economy. In any event, the
second part of Result (iv) is a dismal part − in view of the projected long-run
stationarity of world population (United Nations 2005).
A somewhat surprising result appears if we imagine (unrealistically) that
̃ is sufficiently above one to make a negative number. If population
growth is absent, 0 is in fact needed for 0 along a BGP However,
0 implies instability. Hence this would be a case of an instable BGP
with fully endogenous growth.24
As to Result (v), it is noteworthy that the absence of a scale effect on
growth holds for any value of ̃ including ̃ = 125
24
Thus, if we do not require 0 in the first place, (iv) could be reformulated as:
existence of a stable optimal BGP with 0 requires 0. This is not to say that
reducing from positive to zero renders an otherwise stable BGP instable. Stability-
instability is governed solely by the sign of Given 0 letting decrease from a level
above the critical value, (̃ + + − 1) given from (16.30), to a level below, changes
from positive to negative, i.e., growth comes to an end.
25
More commonplace observations are that increased impatience leads to faster depletion
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
CHAPTER 16. NATURAL RESOURCES AND ECONOMIC
308 GROWTH
A pertinent question is: are the above results just an artifact of the very
simplified reduced-form AK-style set-up applied here? The answer turns out
to be no. For models with a distinct research technology and intertemporal
knowledge spillovers, this is shown in Groth (2007).
16.6 Natural resources and the issue of limits
to economic growth
Two distinguished professors were asked by a journalist: Are there limits to
economic growth?
The answers received were:26
Clearly YES:
• A finite planet!
• The amount of cement, oil, steel, and water that we can use is limited!
Clearly NO:
• Human creativity has no bounds!
• The quality of wine, TV transmission of concerts, computer games, and
medical treatment knows no limits!
An aim of this chapter has been to bring to mind that it would be strange
if there were no limits to growth. So a better question is:
What determines the limits to economic growth?
The answer suggested is that these limits are determined by the capability
of the economic system to substitute limited natural resources by man-made
goods the variety and quality of which are expanded by creation of new ideas.
In this endeavour frontier countries, first the U.K. and Europe, next the
United States, have succeeded at a high rate for two and a half century. To
what extent this will continue in the future nobody knows. Some economists,
e.g. Gordon (2012), argue there is an enduring tendency to slowing down of
innovation and economic growth (the low-hanging fruits have been taken).
and lower growth (in the plausible case ̃ 1) Further, in the log-utility case ( = 1) the
depletion rate equals the effective rate of impatience, − .
26
Inspired by Sterner (2008).
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
16.7. Bibliographic notes 309
Others, e.g. Brynjolfsson and McAfee (2014), disagree. They reason that the
potentials of information technology and digital communication are on the
verge of the point of ubiquity and flexible application. For these authors the
prospect is “The Second Machine Age” (the title of their book), by which
they mean a new innovative epoch where smart machines and new ideas are
combined and recombined - with pervasive influence on society.
16.7 Bibliographic notes
It is not always recognized that the research of the 1970s on macro implica-
tions of essential natural resources in limited supply already laid the ground-
work for a theory of endogenous and policy-dependent growth with natural
resources. Actually, by extending the DHSS model, Suzuki (1976), Robson
(1980) and Takayama (1980) studied how endogenous innovation may affect
the prospect of overcoming the finiteness of natural resources.
Suzuki’s (1976) article contains an additional model, involving a resource
externality. Interpreting the externality as a “greenhouse effect”, Sinclair
(1992, 1994) and Groth and Schou (2006) pursue this issue further. In the
latter paper a configuration similar to the model in Section 16.5 is stud-
ied. The source of increasing returns to scale is not intentional creation of
knowledge, however, but learning as a by-product of investing as in Arrow
(1962a) and Romer (1986). Empirically, the evidence furnished by, e.g., Hall
(1990) and Caballero and Lyons (1992) suggests that there are quantitatively
significant increasing returns to scale w.r.t. capital and labour or external
effects in US and European manufacturing. Similarly, Antweiler and Trefler
(2002) examine trade data for goods-producing sectors and find evidence for
increasing returns to scale.
Concerning Result (i), note that if some irreducibly exogenous element
in the technological development is allowed in the model by replacing the
constant in (16.21) by where ≥ 0, then (16.25) is replaced by +
(̃+ −1) 0 or ̃ 1 Both Stiglitz (1974a, p. 131) and Withagen (1990,
p. 391) ignore implicitly the possibility ̃ 1 Hence, from the outset they
preclude fully endogenous growth.
16.8 Appendix
16.8.1 A. The CES function
The CES (Constant Elasticity of Substitution) function is used in consumer
theory as a specification of preferences and in production theory as a specifi-
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
CHAPTER 16. NATURAL RESOURCES AND ECONOMIC
310 GROWTH
cation of a production function. Here we consider it as a production function.
It can be shown27 that if a neoclassical production function with CRS has
a constant elasticity of (factor) substitution different from one, it must be of
the form £ ¤1
= + (1 − ) (16.32)
where and are parameters satisfying 0, 0 1 and 1
6= 0 This function has been used intensively in empirical studies and
is called a CES production function. For a given choice of measurement
units, the parameter reflects efficiency and is thus called the efficiency
parameter. The parameters and are called the distribution parameter
and the substitution parameter, respectively. The restriction 1 ensures
that the isoquants are strictly convex to the origin. Note that if 0
the right-hand side of (16.32) is not defined when either or (or both)
equal 0 We can circumvent this problem by extending the domain of the
CES function and assign the function value 0 to these points when 0.
Continuity is maintained in the extended domain.
By taking partial derivatives in (16.32) and substituting back we get
µ ¶1− µ ¶1−
= and = (1 − ) (16.33)
£ ¤1 £ ¤1
where = + (1 − )− and = + 1 − The mar-
ginal rate of substitution of for therefore is
1 − 1−
= = 0
Consequently,
1−
= (1 − )−
where the inverse of the right-hand side is the value of Substitut-
ing these expressions into the general definition of the elasticity of substitution
between capital and labor, evaluated at the point ( )
()
()
̃( ) = =
(16.34)
| =̄ | =̄
gives
1
̃( ) = ≡ (16.35)
1−
27
See, e.g., Arrow et al. (1961).
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
16.8. Appendix 311
confirming the constancy of the elasticity of substitution, given (16.33). Since
1 0 always A higher substitution parameter, results in a higher
elasticity of substitution, And ≶ 1 for ≶ 0 respectively.
Since = 0 is not allowed in (16.32), at first sight we cannot get = 1
from this formula. Yet, = 1 can be introduced as the limiting case of (16.32)
when → 0 which turns out to be the Cobb-Douglas function. Indeed, one
can show28 that, for fixed and
£ ¤1
+ (1 − ) → 1− for → 0
By a similar procedure as above we find that a Cobb-Douglas function always
has elasticity of substitution equal to 1; this is exactly the value taken by
in (16.35) when = 0. In addition, the Cobb-Douglas function is the only
production function that has unit elasticity of substitution everywhere.
Another interesting limiting case of the CES function appears when, for
fixed and we let → −∞ so that → 0 We get
£ ¤1
+ (1 − ) → min( ) for → −∞ (16.36)
So in this case the CES function approaches a Leontief production function,
the isoquants of which form a right angle, cf. Figure 16.4. In the limit there
is no possibility of substitution between capital and labor. In accordance
with this the elasticity of substitution calculated from (16.35) approaches
zero when goes to −∞
Finally, let us consider the “opposite” transition. For fixed and we
let the substitution parameter rise towards 1 and get
£ ¤1
+ (1 − ) → [ + (1 − )] for → 1
Here the elasticity of substitution calculated from (16.35) tends to ∞ and
the isoquants tend to straight lines with slope −(1 − ) In the limit,
the production function thus becomes linear and capital and labor become
perfect substitutes.
Figure 16.4 depicts isoquants for alternative CES production functions
and their limiting cases. In the Cobb-Douglas case, = 1 the horizontal
and vertical asymptotes of the isoquant coincide with the coordinate axes.
When 1 the horizontal and vertical asymptotes of the isoquant belong
to the interior of the positive quadrant. This implies that both capital and
labor are essential inputs. When 1 the isoquant terminates in points
28
For proofs of this and the further claims below, see Appendix E of Chapter 4 in Groth
(2014).
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
CHAPTER 16. NATURAL RESOURCES AND ECONOMIC
312 GROWTH
1
1
K 0
0 1
L
L
Figure 16.4: Isoquants for the CES production function for alternative values of
= 1(1 − ).
on the coordinate axes. Then neither capital, nor labor are essential inputs.
Empirically there is not complete agreement about the “normal” size of the
elasticity of factor substitution for industrialized economies. The elasticity
also differs across the production sectors. A recent thorough econometric
study (Antràs, 2004) of U.S. data indicate the aggregate elasticity of substi-
tution to be in the interval (05 10).
The CES production function in intensive form
Dividing through by on both sides of (16.32), we obtain the CES production
function in intensive form,
1
≡ = ( + 1 − ) (16.37)
where ≡ . The marginal productivity of capital can be written
£ ¤ 1− ³ ´1−
= = + (1 − )− =
which of course equals in (16.33). We see that the CES function
violates either the lower or the upper Inada condition for , depending
on the sign of Indeed, when 0 (i.e., 1) then for → 0 both
and approach an upper bound equal to 1 ∞ thus violating the
lower Inada condition for (see the right-hand panel of Figure 2.3 in
Chapter 2). It is also noteworthy that in this case, for → ∞, approaches
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
16.8. Appendix 313
y y
A 1/
A 1/
1/
A(1 )
A(1 )1/
k k
0 (0 1) 0 ( 1)
Figure 16.5: The CES production function for 1 (left panel) and 1 (right
panel).
an upper bound equal to (1 − )1 ∞. These features reflect the low
degree of substitutability when 0
When instead 0 there is a high degree of substitutability ( 1).
Then, for → ∞ both and → 1 0 thus violating the upper
Inada condition for (see the right-hand panel of Figure 16.5). It is
also noteworthy that for → 0, approaches a positive lower bound equal
to (1 − )1 0. Thus, in this case capital is not essential. At the same
time → ∞ for → 0 (so the lower Inada condition for the marginal
productivity of capital holds).
The marginal productivity of labor is
= = (1 − )( + 1 − )(1−) ≡ ()
from (16.33).
Since (16.32) is symmetric in and we get a series of symmetric results
£ ¤1
by considering output per unit of capital as ≡ = + (1 − )()
In total, therefore, when there is low substitutability ( 0) for fixed input
of either of the production factors, there is an upper bound for how much
an unlimited input of the other production factor can increase output. And
when there is high substitutability ( 0) there is no such bound and an
unlimited input of either production factor take output to infinity.
The Cobb-Douglas case, i.e., the limiting case for → 0 constitutes in
several respects an intermediate case in that all four Inada conditions are
satisfied and we have → 0 for → 0 and → ∞ for → ∞
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
CHAPTER 16. NATURAL RESOURCES AND ECONOMIC
314 GROWTH
Generalizations
The CES production function considered above has CRS. By adding an elas-
ticity of scale parameter, , we get the generalized form
£ ¤
= + (1 − ) 0 (16.38)
In this form the CES function is homogeneous of degree For 0 1
there are DRS, for = 1 CRS, and for 1 IRS. If 6= 1 it may be
£ ¤1
convenient to consider ≡ 1 = 1 + (1 − ) and ≡
= 1 ( + 1 − )1
The elasticity of substitution between and is = 1(1 − ) whatever
the value of So including the limiting cases as well as non-constant returns
to scale in the “family” of production functions with constant elasticity of
substitution, we have the simple classification displayed in Table 16.1.
Table 16.1 The family of production functions
with constant elasticity of substitution.
=0 01 =1 1
Leontief CES Cobb-Douglas CES
Note that only for ≤ 1 is (16.38) a neoclassical production function.
This is because, when 1 the conditions 0 and 0 do not
hold everywhere.
We may generalize further by assuming there are inputs, in the amounts
1 2 Then the CES production function takes the form
£ ¤ X
= 1 1 + 2 2 + 0 for all = 1 0
(16.39)
In analogy with (16.34), for an -factor production function the partial elas-
ticity of substitution between factor and factor is defined as
( )
=
| =̄
where it is understood that not only the output level but also all , 6= , ,
are kept constant. Note that = In the CES case considered in (16.39),
all the partial elasticities of substitution take the same value, 1(1 − )
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
16.8. Appendix 315
16.8.2 B. Balanced growth with an essential non-renewable
resource
The production side of the DHSS model with CES production function is
described by:
= ̃ ( ) ̃ ≥ 0 (16.40)
̇ = − − ≥ 0 0 0 given, (16.41)
̇ = − ≡ − 0 0 given, (16.42)
= 0 ≥ 0 (16.43)
Z ∞
≤ 0 (16.44)
0
We will assume that the non-renewable resource is essential, i.e.,
= 0 implies = 0 (16.45)
From now we omit the dating of the time-dependent variables where not
needed for clarity. Recall that in the context of an essential non-renewable
resource, we define a balanced growth path (BGP for short) as a path along
which the quantities and are positive and change at constant
proportionate rates (some or all of which may be negative).
Lemma 1 Along a BGP the following holds: (a) = 0; (b) (0) =
− (0) and
lim = 0 (16.46)
→∞
Proof Consider a BGP. (a) From (16.42), = −; differentiating with
respect to time gives
̇ = −( − ) = 0
by definition of a BGP.
R ∞ Hence, R∞
= since 0 by definition. For any
constant we have 0 = 0 0 If ≥ 0 (16.44) would thus be
violated. Hence, 0 (b) With = 0 in (16.42), we get ̇0 0 = −0 0
= the last equality following from (a). Hence, 0 = − 0 Finally, the
solution to (16.42) can be written = 0 Then, since is a negative
constant, → 0 for → ∞ ¤
Define
≡ ≡ and ≡ (16.47)
We may write (16.42) as
= − − (16.48)
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
CHAPTER 16. NATURAL RESOURCES AND ECONOMIC
316 GROWTH
Similarly, by (16.42),
≡ − (16.49)
Lemma 2 Along a BGP, = = − 0 is constant and = . If gross
saving is positive in some time interval, we have along the BGP in addition
that = , both constant, and that and are constant.
Proof Consider a BGP. Since is constant by definition of a BGP, must
also be constant in view of (16.49). Then, by Lemma 1, = = − is
constant and 0. Differentiating in (16.48) with respect to gives ̇
= ̇ − ̇ = ( − ) −( − ) = 0 since is constant along a BGP.
Dividing through by which is positive along a BGP, and reordering gives
− = ( − ) (16.50)
But this is a contradiction unless = ; indeed, if 6= then −
6= − at the same time as = → 0 if and
= → ∞ if both cases being incompatible with (16.50) and the
presumed constancy of and hence constancy of both − and
− So = along a BGP. Suppose gross saving is positive in some
time interval and that at the same time 6= = then (16.50) implies
≡ 1, i.e., = for all or gross saving = 0 for all a contradiction.
Hence, = = It follows by (16.47) that and are constant. ¤
Consider the case where the production function is neoclassical with CRS,
and technical progress is labor- and resource-augmenting:
= ( 2 3 ) (16.51)
2 = 2 , 2 ≥ 0 3 = 3 3 ≥ 0
Let ̂ ≡ 2 and ̂ ≡ 3 Let ̂ and ̂ denote the output elasticities
w.r.t. ̂ and ̂ i.e.,
2 3
≡ ̂ ≡ ̂ ≡
(2 ) (3 )
Differentiating in (16.51) w.r.t. and dividing through by (as in growth-
accounting), we then have
̇
≡ = + ̂ ( 2 + ) + ̂ ( 3 + )
= + ̂ ( 2 + ) + (1 − − ̂ )( 3 + ) (16.52)
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
16.8. Appendix 317
the last equality being implied by the CRS property in (16.51).
Suppose the economy follows a BGP with positive gross saving. Then,
by Lemma 2, = and = − 0 Hence, (16.52) can be written
(1 − )( − ( 3 − )) = ̂ ( 2 + − ( 3 − )) (16.53)
Consider the special case where is CES:
³ ´1 X
= 1 + 2 (2 ) + 3 (3 ) 1 2 3 0 = 1 1
(16.54)
As we know from Appendix A, for = 0 the CES formula can be interpreted
as the Cobb-Douglas formula (16.13). Applying (16.33) from Appendix A,
the output elasticities w.r.t. ̂ and ̂ are
µ ¶− µ ¶− µ ¶−
= 1 ̂ = 2 and ̂ = 3
2 3
(16.55)
respectively.
Lemma 3 Let ≡ and ≡ Given (16.43) and (16.54), along a
BGP with positive gross saving, and ̂ are constant, and = = 2
In turn, such a BGP exists if and only if
= 3 − ( 2 + ) 0 (16.56)
Proof Consider a BGP with positive gross saving. By Lemma 2, 0 = −
is constant and ≡ is constant, hence so is The left hand side of
(16.53) is thus constant and so must the right-hand side therefore be. Suppose
that, contrary to (16.56), 6= 3 −( 2 +). Then constancy of the right-hand
of (16.53) requires that ̂ is constant. In turn, by (16.55), this requires that
(2 ) ≡ 2 is constant. Consequently, = 2 = where the second
equality is implied by the claim in Lemma 2 that along a BGP with positive
gross saving in some time interval, = As is constant, it follows
that = ≡ + = 2 + Inserting this into (16.52) and rearranging,
we get
(1 − − ̂ )( 2 + ) = (1 − − ̂ )( 3 − )
where the last equality follows from = −. Isolating gives the equality
in (16.56). Thereby, our assumption 6= 3 − ( 2 + ) leads to a contradic-
tion. Hence, given (16.43) and (16.54), if a BGP with positive gross saving
exists, then = 3 − ( 2 + ) 0. This shows the necessity of (16.56).
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
CHAPTER 16. NATURAL RESOURCES AND ECONOMIC
318 GROWTH
The sufficiency of (16.56) follows by construction, starting by fixing and
thereby − in accordance with (16.56) and moving “backward”, showing
consistency with (16.52) for = = 2 + . ¤
16.9 References
Antràs, 2004, ...
Arrow, K. J., 1962, The economic implications of learning by doing, Review
of Economic Studies 29, 153-73.
Arrow, K. J., P. Dasgupta, L. Goulder, G. Daily, P. Ehrlich, G. Heal, S.
Levin, K.-G. Mäler, S. Schneider, D. Starrett, and B. Walker, 2004,
Are we consuming too much?, J. of Economic Perspectives 18, no. 3,
147-172.
Arrow, K.J, P. Dasgupta, L. Goulder, K. Mumford, and K.L.L. Oleson,
2012, Sustainability and the measurement of wealth, Environment and
Development Economics 17, no. 3, 317—353.
Brock, W. A., and M. Scott Taylor, 2005. Economic Growth and the Envi-
ronment: A review of Theory and Empirics. In: Handbook of Economic
Growth, vol. 1.B, ed. by ...
Brundtland Commission, 1987, ...
Brynjolfsson, E., and A. McAfee, 2014, The Second Machine Age, Norton.
Caballero, R. J., and R. K. Lyons, The case of external economies. In:
Political Economy, Growth, and Business Cycles, ed. by A. Cukierman,
Z. Hercowitz and L. Leiderman, MIT Press, Cambridge, MA, 1992, pp.
117-139.
Dasgupta, P., 1993, Natural resources in an age of substitutability. In:
Handbook of Natural Resource and Energy Economics, vol. 3, ed. by
A. V. Kneese and J. L. Sweeney, Elsevier: Amsterdam, 1111-1130.
Dasgupta, P., and G. M. Heal, 1974, The optimal depletion of exhaustible
resources, Review of Economic Studies, vol. 41, Symposium, 3-28.
Dasgupta, P., and G. M. Heal, 1979, Economic Theory and Exhaustible
Resources, Cambridge University Press, Cambridge.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
16.9. References 319
Dasgupta, P., G. M. Heal and M. K. Majumdar, 1977, Resource deple-
tion and research and development. In: Frontiers of Quantitative Eco-
nomics, vol. IIIB, ed. by M. D. Intrilligator, North-Holland, Amster-
dam, 483-505.
Dasgupta, P., G. M. Heal and A. K. Pand, 1980, Funding research and
development, Applied Mathematical Modeling 4, 87-94.
Fagnart, J.-F., and M. Germain, 2011, Quantitative vs. qualitative growth
with recyclable resources, Ecological Economics, vol. 70, 929-941.
Gordon, R.J., 2012, Is US economic growth over? Faltering innovation
confronts the six headwinds, NBER WP no. 18315.
Grimaud, A., and L. Rougé, 2003, Non-renewable resources and growth
with vertical innovations: optimum, equilibrium and economic policies,
Journal of Environmental Economics and Management 45, 433-453.
Grossman, G.M., and A.B. Krueger, 1995, , QJE, vol. 90 (2).
Groth, C., 2004, Strictly endogenous growth with non-renewable resources
implies an unbounded growth rate, Topics in Macroeconomics 4 (1),
1-13.
Groth, C., 2007, A New-Growth Perspective on Non-renewable Resources.
In: Sustainable Resource Use and Economic Dynamics, ed. by L.
Bretschger and S. Smulders, Springer Verlag: Dordrecht (available at
the course website).
Groth, C., 2014, Lecture notes in macroeconomics (mimeo), Department of
Economics, University of Copenhagen.
Groth, C., and P. Schou, 2002, Can nonrenewable resources alleviate the
knife-edge character of endogenous growth? Oxford Economic Papers
54, 386-411.
−− , 2006, Growth and non-renewable resources: The different roles of
capital and resource taxes, Journal of Environmental Economics and
Management, forthcoming.
Hall, R. E., 1990, Invariance properties of Solow’s residual. In: Growth/Productivity/Un-
employment. Essays to Celebrate Bob Solow’s Birthday, ed. by P.
Diamond, MIT Press, Cambridge (Mass.), 71-112.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
CHAPTER 16. NATURAL RESOURCES AND ECONOMIC
320 GROWTH
Hamilton, K., 2012, Comments on Arrow et al., ‘Sustainability and the
measurement of wealth’, Environment and Development Economics 17,
no. 3, 356 - 361.
Handbook of Natural Resource and Energy Economics, vol. III, ed. by A.
V. Kneese and J. L. Sweeney, Elsevier: Amsterdam, 1993.
Hartwick, J. M., 1977, Intergenerational equity and the investing of rents
from exhaustible resources, American Economic Review 67, 972-974.
Heal, G., 1998. Valuing the Future. Economic Theory and Sustainability.
Columbia University Press: New York.
Hotelling, H., 1931, The economics of exhaustible resources, Journal of Po-
litical Economy 39, 137-175.
Jones, C., 2002, Introduction to Economic Growth, 2. ed., W.W. Norton:
New York.
Kamien, M. I., and N. L. Schwartz, 1978, Optimal exhaustible resource de-
pletion with endogenous technical change, Review of Economic Studies
45 (1), 179-196.
Krautkraemer, J., 1985, Optimal growth, resource amenities and the preser-
vation of natural environments, Review of Economic Studies 52, 153-
170.
−− , 1998, Nonrenewable resource scarcity, Journal of Economic Literature
36 (4), 2065-2107.
Lind, S., og P. Schou, P., 2013, Ægte opsparing, Arbejdspapir 2013:1, De
Økonomiske Råds Sekretariat, København.
Meadows, D. H., D. Meadows, J. Randers, and W. W. Behrens III, 1972,
The Limits to Growth, Universe Books: New York.
Meadows, D. H., D. L. Meadows and J. Randers, 1992, Beyond the Limits,
Chelsea Green Publ. Com., Post Mills.
Meadows, D. H., D. Meadows, and J. Randers, 2004, Limits to Growth: The
30-Year Update,...
Neumayer, E., 2000, Scarce or abundant? The economics of natural resource
availability, Journal of Economic Surveys 14 (3), 307-335.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
16.9. References 321
Nordhaus, W., 1969, An economic theory of technological change, American
Economic Review 59 (2), 18-28.
Nordhaus, W. , 1992, Lethal Model 2: The limits to growth revisited, Brook-
ings Papers on Economic Activities, (2), 1-43.
Nordhaus, W.D., 1999, Quality change in price indices, J. of Economic
Perspectives, vol. 12 (1), 59-68.
Nordhaus, W., and J. Tobin, 1972, Is growth obsolete? In: Economic
Growth: Fiftieth Anniversary Colloquium V, Columbia University Press,
New York. Reprinted in: J. Tobin, Essays in Economics. Theory and
Policy, The MIT Press, Cambridge (Mass.) 1985, 360-450.
Popp, D., 2002, Induced innovation and energy prices, American Economic
Review 92 (1), 160-180.
Robson, A. J., 1980, Costly innovation and natural resources, International
Economic Review 21, 17-30.
Schou, P., 2000, Polluting nonrenewable resources and growth, Environ-
mental and Resource Economics 16, 211-27.
−− , 2002, When environmental policy is superfluous: Growth and polluting
resources, Scandinavian Journal of Economics 104, 605-620.
Seierstad, A., and K. Sydsaeter, 1987, Optimal Control Theory with Eco-
nomic Applications, North Holland, Amsterdam.
Sinclair, P., 1994, On the optimum trend of fossil fuel taxation, Oxford
Economic Papers 46, 869-877.
Smil, V., 2003, Energy at the crossroads. Global perspectives and uncertain-
ties, Cambridge Mass.: MIT Press.
Smulders, S., 1995, Entropy, environment, and endogenous growth, Inter-
national Tax and Public Finance, vol. 2, 319-340.
Smulders, S., 1999,...
Smulders, S., 2012, An arrow in the Achilles’ heel of sustainability and
wealth accounting, Environment and Development Economics 17, no.
3, 368-372.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
CHAPTER 16. NATURAL RESOURCES AND ECONOMIC
322 GROWTH
Smulders, S., and M. de Nooij, 2003, The impact of energy conservation on
technology and economic growth, Resource and Energy Economics 25,
59-79.
Solow, R.M., 1973, Is the end of the world at hand? Challenge, vol. 16, no.
1, 39-50.
Solow, R. M., 1974a, Intergenerational equity and exhaustible resources,
Review of Economic Studies 41, Symposium Issue, 29-45.
−− , 1974b, The economics of resources or the resources of economics,
American Economic Review 64, Papers and Proceedings, 1-14.
Solow, R.M., 2012, A few comments on ‘Sustainability and the measurement
of wealth, Environment and Development Economics 17, no. 3, 354-
355.
Squires and Vestergaard, 2011,
Sterner, T., 2008, The costs and benefits of climate policy: The impact of
the discount rate and changing relative prices. Lecture at Miljøøkonomisk
konference 2008 in Skodsborg, see http://www.dors.dk/graphics/Synkron-
Library/Konference%202008/
Keynote/Sterner_Copenhagen_SEPT08.pdf
Stiglitz, J., 1974a, Growth with exhaustible natural resources: Efficient and
optimal growth paths, Review of Economic Studies 41, Symposium
Issue, 123-137.
−− , 1974b, Growth with exhaustible natural resources: The competitive
economy, Review of Economic Studies 41, Symposium Issue, 139-52.
Stiglitz, J., A. Sen, and J.-P. Fitoussi, 2009, Report by the Commission
on the Measurement of Economic Performance and Social Progress,
www.stiglitz-sen-fitoussi.fr.
Stiglitz, J., A. Sen, and J.-P. Fitoussi, 2010, Mismeasuring Our Lives. Why
the GDP Doesn’t Add Up, The New Press: New York.
Suzuki, H., 1976, On the possibility of steadily growing per capita consump-
tion in an economy with a wasting and non-replenishable resource, Re-
view of Economic Studies 43, 527-35.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
16.9. References 323
Tahvonen, O., and S. Salo, 2001, Economic growth and transitions between
renewable and nonrenewable energy resources, European Economic Re-
view 45, 1379-1398.
Takayama, A., 1980, Optimal technical progress with exhaustible resources.
In: Exhaustible Resources, Optimality, and Trade, ed. by M. C. Kemp
and N. V. Long, North-Holland, Amsterdam, 95-110.
United Nations, 2005, World Population Prospects. The 2004 Revision,
Population Division, New York.
van Kooten, G. C., and E. H. Bulte, 2000. The Economics of Nature.
Managing Biological Assets. Blackwell: Oxford.
Weitzman, M., 2001, A contribution to the theory of welfare accounting,
Scandinavian J. of Economics 103 (1), 1-23.
Weitzman, M., 2003, Income, Wealth, and the Maximum Principle, Harvard
Univ. Press: Cambridge (Mass.).
Withagen, C., 1990, Topics in resource economics. In: Advanced Lectures
in Quantitative Economics, ed. by F. van der Ploeg, Academic Press,
London, 381-419.
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
CHAPTER 16. NATURAL RESOURCES AND ECONOMIC
324 GROWTH
c Groth, Lecture notes in Economic Growth, (mimeo) 2015.
°
Appendix
Errata to lecture notes
(as updated 18/5 2015)
18/2 2015
LN 4, p. 65, first line of (ii): "g > 0" should read "g ≥ 0".
16/3
LN 8, p. 120, eq. (8.21): delete "- 1" on the right-hand side of the equation (to obtain
consistence with the formulation at p. 116 (the middle) and p. 127-29.
29/3
LN 9, p. 140, line 2: "if have" should read "if we can write".
- , - , line 4: "we can write" should read "we have".
- , p. 147, footnote 6, line 1-2: "and m is a nonnegative number" should be deleted.
30/3
LN 9, p. 150, line 14: Delete the complete line, that is, delete "effective discount rate
appearing on the right-hand side of (9.21), namely the".
- , p. 153, line 1-2: "Figure ??" should read "Figure 1".
- , p. 154, line 3: "is then unknown" should read "is the unknown".
- - , line 5 from below: "Figure ??" should read "Figure 1".
20/4
LN 11, p. 178, line 9-8 from below: "...+1)H(t)." should read "...+1)H(t) along the
balanced gorwth path."
- - , line 7 from below: "Y(t) = ... " should read "Y(t) = f(k_hat*) ...".
4/5
LN 14, p. 248, in the formula at the middle of the page, beta should be replaced by 1-beta and
1-beta should be replaced by beta.
- , p. 253, line 5 from below: "the research flow z_jt" should read "the research
investment z_jt".
- , p. 256, line 3: "putting one unit of account at the disposal" should read "putting
finance at the disposal".
- , - , line 4: "the household gets" should read "the households get".
- , - , line 5: "the payoff per time unit" should read "the payoff per time unit per
unit of account invested".
- , - , line 6: "the household buys a lottery ticket" should read "the households
buy lottery tickets".
9/5
LN 14, p. 256, right before Section 14.3.4 begins addition of the following paragraph may help
understanding what is meant by "loan market" and "interest rate" in the model:
1
We see that the obligation of the "borrowing" R&D lab to pay back the principal (possibly with
interest) to the investors (the households) is conditional on R&D success. In case of R&D success,
the investors get the gross return V_(t), to be shared between them in proportion to their investment.
Although such a contract does not fit the notion of a loan in everyday language, in the economics
vocabulary it is considered a specific form of a loan and the market for such contracts a specific
kind of a "loan market". By diversification, the households can with certainty get the expected rate
of return, r_(t), in this market on their saving. We label this rate of return "the interst rate" because
it is as if there were a market for safe loans with r_(t) as the interest rate.
"Letting g ( m) denote" should read "With gc( m) denoting".
LN 15, p. 276, line 2:
- - , line 3, eq. (15.32):
g ( m) by gc( m) .
Replace
- - , line 13 from below: Replace
g ( m) by gc( m) .
18/5
LN 16, p. 315, line 1: Delete "with CES production function".
2
142 CHAPTER 4. A GROWING ECONOMY
At least in these countries, therefore, the potential coordination failure laid
bare by OLG models does not seem to have been operative in practice.
4.4 The functional distribution of income
.....Text to be inserted
The neoclassical theory
.....Text to be inserted
How the labor income share depends on the capital-labor ratio
To begin with we ignore technological progress and write aggregate output
as Y = F (K; L); where F is neoclassical with CRS. From Euler’s theorem
follows that F (K; L) = F1 K + F2 L = f 0 (k)K + (f (k) kf 0 (k))L; where
k K=L: In equilibrium under perfect competition we have
Y = r^K + wL;
where r^ = r + is the cost per unit of capital input and w is the real wage,
i.e., the cost per unit of labor input. The labor income share is
w=^
r
wL f (k) kf 0 (k) w(k) wL k
= SL(k) = = ;
Y f (k) f (k) r^K + wL 1 + w=^
k
r
where the function SL( ) is the share of labor function and w=^ r is the factor
price ratio.
Suppose that capital tends to grow faster than labor so that k rises over
time. Unless the production function is Cobb-Douglas, this will under perfect
competition a¤ect the labor income share. But apriori it is not obvious in
what direction. If the proportionate rise in the factor price ratio w=^ r is
greater (smaller) than that in k; then SL goes up (down). Indeed, if we let
E`x g(x) denote the elasticity of a function g(x) w.r.t. x; then
w
SL0 (k) R 0 for E`k R 1;
r^
respectively.
Usually, however, the inverse elasticity is considered, namely E`w=^r k: This
elasticity, which indicates how sensitive the cost minimizing capital-labor
ratio, k; is to a given factor price ratio w=^r, coincides with the elasticity of
c Groth, Lecture notes in macroeconomics, (mimeo) 2015.
4.4. The functional distribution of income 143
factor substitution (for a general de…nition, see below). The latter is often
denoted : Since in the CRS case, will be a function of only k; we write
E`w=^r k = (k): We therefore have
SL0 (k) R 0 for (k) Q 1;
respectively. If F is Cobb-Douglas, i.e., Y = K L1 ; 0 < < 1; we have
(k) 1; cf. the next section. In this case variation in k does not change
the labor income share under perfect competition. Empirically there is not
complete agreement about the “normal”size of the elasticity of factor substi-
tution for industrialized economies, but the bulk of studies seems to conclude
with (k) < 1 (see Section 4.5).
Now, let us add Harrod-neutral technical progress to the discussion. So
we write aggregate output as Y = F (K; T L); where F is neoclassical with
CRS, and T = Tt = T0 (1 + g)t . Then the labor income share is
wL w=T w~
= :
Y Y =(T L) y~
The above formulas still hold if we replace k by k~ K=(T L) and w by w~
w=T: While k empirically is clearly growing, k~ k=T is not necessarily so
because also T is increasing.
As we have seen, Kaldor’s stylized facts essentially means that, apart from
short-run ‡uctuations, k~ and therefore also r^ and the labor income share
tend to be constant over time, independently of the sign of (k) ~ 1. Given
the production function f , the elasticity of substitution between capital and
labor does not depend on the presence or absence of Harrod-neutral technical
progress, but only on the function itself. This is because under Harrod-
neutrality, the technology level T only appears as a multiplicative factor
to L; whereby T cancels out in the calculation of the elasticity of factor
substitution.
As alluded to earlier, there are empiricists who reject Kaldor’s “facts”
as a general tendency. For instance Piketty (2014) claims that the e¤ective
capital-labor ratio k~ has an upward trend, temporarily braked by two world
wars and the Great Depression in the 1930s. If so, the sign of (k) ~ 1
becomes decisive for in what direction wL=Y will move. Piketty interprets
the econometric literature as favoring (k) ~ > 1; which means there should
be downward pressure on wL=Y . This source behind a falling wL=Y can
be questioned, however. Indeed, (k) ~ > 1 contradicts the more general
empirical view referred to above. According to Summers (2014), Piketty’s
interpretation relies on con‡ating gross and net returns to capital.
c Groth, Lecture notes in macroeconomics, (mimeo) 2015.
144 CHAPTER 4. A GROWING ECONOMY
Immigration
~
Here is another example that illustrates the importance of the size of (k):
Consider an economy with perfect competition and a given aggregate cap-
ital stock K and technology level T (entering the production function in
the labor-augmenting way as above). Suppose that for some reason, im-
migration, say, aggregate labor supply, L; shifts up and full employment is
maintained by the needed real wage adjustment. In what direction will ag-
gregate labor income wL = w( ~ L then change? The e¤ect of the larger
~ k)T
L is to some extent o¤set by a lower w brought about by the lower e¤ective
~ k~ = kf
capital-labor ratio. Indeed, in view of dw=d ~ 00 (k)
~ > 0; we have k~ #
implies w # for …xed T: So we cannot apriori sign the change in wL: The
following relationship can be shown (Exercise 4.??), however:
@(wL) ~
(k)
= (1 )w R 0 for (k)
~ Q (k);
~ (4.26)
@L ~
(k)
respectively, where a(k)~ ~ 0 (k)=f
kf ~ (k)~ is the output elasticity w.r.t. capital
which under perfect competition equals the gross capital income share. It
follows that the larger L will not be fully o¤set by the lower w as long as the
~ exceeds the gross capital income share,
elasticity of factor substitution, (k);
~
(k). This condition seems con…rmed by most of the empirical evidence (see,
e.g., Antras 2004 and Chirinko 2008).
The elasticity of factor substitution*
We shall here discuss the concept of elasticity of factor substitution at a
more general level. Fig. 4.6 depicts an isoquant, F (K; L) = Y ; for a given
neoclassical production function, F (K; L); which need not have CRS. Let
M RS denote the marginal rate of substitution of K for L; i.e., M RS =
FL (K; L)=FK (K; L):9 At a given point (K; L) on the isoquant curve, M RS
is given by the absolute value of the slope of the tangent to the isoquant at
that point. This tangent coincides with that isocost line which, given the
factor prices, has minimal intercept with the vertical axis while at the same
time touching the isoquant. In view of F ( ) being neoclassical, the isoquants
are by de…nition strictly convex to the origin. Consequently, M RS is rising
along the curve when L decreases and thereby K increases. Conversely, we
can let M RS be the independent variable and consider the corresponding
point on the indi¤erence curve, and thereby the ratio K=L, as a function of
9
When there is no risk of confusion as to what is up and what is down, we use M RS
as a shorthand for the more correct M RSKL .
c Groth, Lecture notes in macroeconomics, (mimeo) 2015.
Short Note 2 Economic Growth
16.05.2015 Christian Groth
Saddle-point stability
A concept which perplexes many is the concept of saddle-point stability.
Consider a two-dimensional dynamic system (two coupled …rst-order di¤erential equa-
tions with two endogenous time-dependent variables). Suppose the system has a steady
state which is a saddle point (which is the case if and only if the two eigenvalues of the
associated Jacobi matrix are of opposite sign). Then, so far, either presence or absence
of saddle-point stability is possible. And which of the two cases occur can not be diag-
nosed from the two di¤erential equations in isolation. One has to consider the boundary
conditions.
Here is a complete de…nition of (local) saddle-point stability. A steady state of a
two-dimensional dynamic system is (locally) saddle-point stable if:
1. the steady state is a saddle point;
2. one of the two endogenous variables is predetermined while the other is a jump
variable;
3. the saddle path is not parallel to the jump variable axis; and
4. there is a boundary condition on the system such that the diverging paths are ruled
out as solutions.
Thus, to establish saddle-point stability all four properties must be veri…ed. If for
instance point 1 and 2 hold but, contrary to point 3, the saddle path is parallel to the
jump variable axis, then saddle-point stability does not obtain. Indeed, given that the
predetermined variable initially deviated from its steady-state value, it would not be
possible to …nd any initial value of the jump variable such that the solution of the system
would converge to the steady state for t ! 1:
You might also like
- Macroeconomics-A Growth Theory ApproachNo ratings yetMacroeconomics-A Growth Theory Approach94 pages
- Intro To Economic Growth - Daron Acemoglu PDFNo ratings yetIntro To Economic Growth - Daron Acemoglu PDF453 pages
- Real Macroeconomic Theory Per Krussel March 2014No ratings yetReal Macroeconomic Theory Per Krussel March 2014277 pages
- Contents 28-09-2014-MasterBook-pdf-M2-2014-2No ratings yetContents 28-09-2014-MasterBook-pdf-M2-2014-214 pages
- MACROECONOMICS For PHD (Dynamic, International, Modern, Public Project)No ratings yetMACROECONOMICS For PHD (Dynamic, International, Modern, Public Project)514 pages
- Exogenous Change and Growth of Knowledgechapter1No ratings yetExogenous Change and Growth of Knowledgechapter131 pages
- Advanced Macroeconomics Chapter 1: Exogenous Growth and Accumulation of KnowledgeNo ratings yetAdvanced Macroeconomics Chapter 1: Exogenous Growth and Accumulation of Knowledge31 pages
- Econ 702 Macroeconomic Theory Spring 2016No ratings yetEcon 702 Macroeconomic Theory Spring 201659 pages
- [Springer Texts in Business and Economics] Sibabrata Das, Alex Mourmouras, Peter Rangazas - Economic Growth and Development_ A Dynamic Dual Economy Approach (2024, Springer) - libgen.liNo ratings yet[Springer Texts in Business and Economics] Sibabrata Das, Alex Mourmouras, Peter Rangazas - Economic Growth and Development_ A Dynamic Dual Economy Approach (2024, Springer) - libgen.li444 pages
- Macroeconomic Theory I - Vollrath - Lecture NotesNo ratings yetMacroeconomic Theory I - Vollrath - Lecture Notes138 pages
- (Springer Texts in Business and Economics) Sibabrata Das, Alex Mourmouras, Peter C. Rangazas (Auth.)-Economic Growth and Development_ a Dynamic Dual Economy Approach-Springer International PublishingNo ratings yet(Springer Texts in Business and Economics) Sibabrata Das, Alex Mourmouras, Peter C. Rangazas (Auth.)-Economic Growth and Development_ a Dynamic Dual Economy Approach-Springer International Publishing278 pages
- 22nd April 2020 - Response of An LTI System To Complex ExponentialsNo ratings yet22nd April 2020 - Response of An LTI System To Complex Exponentials17 pages
- 18.02.01.008 Design Studio LLL Small Cafe DesignNo ratings yet18.02.01.008 Design Studio LLL Small Cafe Design28 pages
- Solution Manual for Statistics for Criminology and Criminal Justice, 3rd Edition, Jacinta M. Gau, - 2025 Scribd Download Full Chapters100% (7)Solution Manual for Statistics for Criminology and Criminal Justice, 3rd Edition, Jacinta M. Gau, - 2025 Scribd Download Full Chapters43 pages
- Circuit Descriptions - JS330/360: Section E - HydraulicsNo ratings yetCircuit Descriptions - JS330/360: Section E - Hydraulics1 page
- Welcome To: Metacognitive Training For Depression (D-MCT)No ratings yetWelcome To: Metacognitive Training For Depression (D-MCT)55 pages
- Tarun Lambhate: Professional Traits - An OverviewNo ratings yetTarun Lambhate: Professional Traits - An Overview4 pages
- VEHICLE ID: A 701309 Marketing Type: Fm13 64R REG NO: B6G842 VIN: YV2JS02D3BA701309 IMPORTER: 6785 Country of Operation: PERNo ratings yetVEHICLE ID: A 701309 Marketing Type: Fm13 64R REG NO: B6G842 VIN: YV2JS02D3BA701309 IMPORTER: 6785 Country of Operation: PER5 pages
- A Case Based Approach to Pacemakers ICDs and Cardiac Resynchronization Questions for Examination Review and Clinical Practice Volume 1 1st, revised Edition Paul A. Friedman Et Al. (Eds.) 2024 scribd download100% (2)A Case Based Approach to Pacemakers ICDs and Cardiac Resynchronization Questions for Examination Review and Clinical Practice Volume 1 1st, revised Edition Paul A. Friedman Et Al. (Eds.) 2024 scribd download81 pages
- @aakashallen: Estimation of closest distance of approach (derivation) of α - particleNo ratings yet@aakashallen: Estimation of closest distance of approach (derivation) of α - particle25 pages
- Lecture Slides Jackson NetworksOnline Week2 SlidesNo ratings yetLecture Slides Jackson NetworksOnline Week2 Slides117 pages
- Calculating Opening Balance of A Month Using Mysql - Stack OverflowNo ratings yetCalculating Opening Balance of A Month Using Mysql - Stack Overflow1 page
- Foreign Exchange Transaction Process FlowNo ratings yetForeign Exchange Transaction Process Flow4 pages
- Application & Installation Manual: Cat C4.4 Acert, C6.6 ACERT, and C7.1 ACERT Tier 4 Interim, Stage IIIB EnginesNo ratings yetApplication & Installation Manual: Cat C4.4 Acert, C6.6 ACERT, and C7.1 ACERT Tier 4 Interim, Stage IIIB Engines208 pages
- Modular Robot Unit - Characterisation, Design and RealisationNo ratings yetModular Robot Unit - Characterisation, Design and Realisation64 pages
