Lecture Notes on Macroeconomics

Karl Whelan

University College Dublin

December 18, 2014

Contents

I The IS-MP-PC Model 6

1 Introducing the IS-MP-PC Model 7

2 Analysing the IS-MP-PC Model 37

3 The Taylor Principle 64

4 The Zero Lower Bound and the Liquidity Trap 78

II Rational Expectations 103

5 Rational Expectations and Asset Prices 104

6 Consumption and Asset Pricing 145

7 Exchange Rates, Interest Rates and Expectations 171

1
8 Sticky Prices and the Phillips Curve 184

9 Investment With Adjustment Costs 198

III Long-Run Growth 208

10 Growth Accounting 209

11 The Solow Model 228

12 Endogenous Technological Change 266

13 Cross-Country Technology Diffusion 283

14 Institutions and Efficiency 297

IV Growth and Resources 306

15 The Malthusian Model 307

16 Malthus and the Environment 331

2
Forward
This is a collection of lecture notes that I have used over a number of years teaching Advanced

Macroeconomics to final year undergraduates at University College Dublin. Some of the

material has also been used to teach first-term Masters students. The material is not intended

to be a comprehensive book on advanced undergraduate macroeconomics but simply reflects

my own interests and preferences and is being made available in this format because some

people may find it useful.

Over the years, I have made most of these notes available on my own website but they

tend to go up and down on the website each year depending on when I am teaching a course.

The material also changes from year to year, so some topics that had been previously covered

disappear from my site. Because I know that some people are interested in the notes and

use them to assist with their own teaching or learning, I have decided to make the full set

available permanently in book form.

A few observations on the notes. The first two parts of the book deal with issues related to

short-term macroeconomic fluctuations. Part 1 presents an adapted version of the Keynesian

IS-LM model featuring monetary policy rules. The approach taken is largely borrowed from

Carl Walshs 2002 paper Teaching Inflation Targeting: An Analysis for Intermediate Macro

in the Journal of Economic Education but the analysis here applies the model to a much

wider range of issues. There are also similarities with David Romers manuscript, Short-Run

3
Fluctuations, particularly in the treatment of the zero bound on interest rates but there are

also some important differences.1

My intention with this material is to teach what are probably the key insights of New

Keynesian economicsthat fiscal and monetary policy can be effective in influencing output

in the short-run but not the long-run, that the process by which the public formulates inflation

expectations is crucial, the advantages to having tough central bankers, that monetary policy

rules can be stabilising if they satisfy the Taylor principle, and that some of these key insights

are reversed if the economy is in a liquidity trapwithout relying on the particular restrictive

assumption of rational expectations.

Part 2, in contrast, explores rational expectations in some detail. The unifying theme of

this section is the importance of understanding first-order stochastic differences equations; how

to solve them for forward-looking solutions and how to relate these solutions to observable

data. These methods are used to explore a range of standard macroeconomic topics with

some chapters (such as those on consumption and asset prices) more focused on reviewing the

empirical evidence than others. In terms of pedagogy, my approach is to explain the important

role that rational expectations plays in modern macroeconomic modelling but to encourage

students to realise that empirical testing tends to uncover weaknesses in this approach.

The final two parts of the book cover long-run growth theory. The first few topics (growth

acccounting, Solow and Romer models) are fairly standard while the remaining topics (insti-

tutions, technology diffusion, Malthusian dynamics and growth and resources) partly reflect

my own interests.

Beamer slides are available for almost all the chapters in this book but I do not have time
1
Romers manuscript is available at http://eml.berkeley.edu/ dromer/

4
to adapt my current set of slides to precisely cover all the topics in the book. Anybody who

may wish to get teaching slides to accompanying teaching some of the chapters in this book

is free to email me at [email protected] and Ill help out if I can.

5
 Part I

The IS-MP-PC Model

6
Chapter 1

Introducing the IS-MP-PC Model

I am assuming that everyone reading these notes has seen the IS-LM and AS-AD models. In

the first part of this course, we are going to revisit some of the ideas from those models and

expand on them in a number of ways:

Rather than the traditional LM curve, we will describe monetary policy in a way that

is more consistent with how it is now implemented, i.e. we will assume the central bank

follows a rule that dictates how it sets nominal interest rates. We will focus on how the

properties of the monetary policy rule influence the behaviour of the economy.

We will have a more careful treatment of the roles played by real interest rates.

We will focus more on the role of the publics inflation expectations.

We will model the zero lower bound on interest rates and discuss its implications for

policy.

Our model is going to have three elements to it:

A Phillips Curve describing how inflation depends on output.

7
 An IS Curve describing how output depends upon interest rates.

A Monetary Policy Rule describing how the central bank sets interest rates depend-

ing on inflation and/or output.

Putting these three elements together, I will call it the IS-MP-PC model (i.e. The Income-

Spending/Monetary Policy/Phillips Curve model). I will describe the model with equations.

I will also merge together the second two elements (the IS curve and the monetary policy

rule) to give a new IS-MP curve that can be combined with the Phillips curve to use graphs

to illustrate the models properties.1

Model Element One: The Phillips Curve

Perhaps the most common theme in economics is that you cant have everything you want. Life

is full of trade-offs, so that if you get more of one thing, you have to have less of another thing.

In macroeconomics, there are important trade-offs facing governments when they implement

policy. One of these relates to a trade-off between desired outcomes for inflation and output.

What form does this relationship take? Back when macroeconomics was a relatively young

discipline, in 1958, a study by the LSEs A.W. Phillips seemed to provide the answer. Phillips

documented a strong negative relationship between wage inflation and unemployment: Low

unemployment was associated with high inflation, presumably because tight labour markets

stimulated wage inflation. Figure 1.1 shows one of the graphs from Phillipss paper illustrating

the kind of relationship he found.

1
The model presented here is basically an adapted version of the model presented in Carl Walshs 2002 paper

Teaching Inflation Targeting: An Analysis for Intermediate Macro in the Journal of Economic Education.

8
 In 1960, a paper by MIT economists Robert Solow and Paul Samuelson (both of whom

would go on to win the Nobel prize in economics for other work) replicated these findings

for the US and emphasised that the relationship also worked for price inflation. Figure 1.2

reproduces the graph in their paper describing the relationship they found. The Phillips curve

quickly became the basis for the discussion of macroeconomic policy decisions. Economists

advised that governments faced a tradeoff: Lower unemployment could be achieved, but only

at the cost of higher inflation.

However, Milton Friedmans 1968 presidential address to the American Economic Asso-

ciation produced a well-timed and influential critique of the thinking underlying the Phillips

curve. Friedman pointed out that it was expected real wages that affected wage bargaining.

If low unemployment means workers have a strong bargaining position, then high nominal

wage inflation on its own is not good enough: They want nominal wage inflation greater than

price inflation.

Friedman argued that if policy-makers tried to exploit an apparent Phillips curve tradeoff,

then the public would get used to high inflation and come to expect it. Inflation expectations

would move up and the previously-existing tradeoff between inflation and output would disap-

pear. In particular, he put forward the idea that there was a natural rate of unemployment

and that attempts to keep unemployment below this level could not work in the long run.

9
Evidence on the Phillips Curve

Monetary and fiscal policies in the 1960s were very expansionary around the world, partly

because governments following Phillips curve recipes chose booming economies with low

unemployment at the expense of somewhat higher inflation.

At first, the Phillips curve seemed to work: Inflation rose and unemployment fell. However,

as the public got used to high inflation, the Phillips tradeoff got worse. By the late 1960s

inflation was rising even though unemployment had moved up. Figure 1.3 shows the US time

series for inflation and unemployment. This stagflation combination of high inflation and high

unemployment got even worse in the 1970s. This was exactly what Friedman had predicted.

Today, the data no longer show any sign of a negative relationship between inflation and

unemployment. If fact, if you look at the scatter plot of US inflation and unemployment data

shown in Figure 1.4 , the correlation is positive: The original formulation of the Phillips curve

is widely agreed to be wrong.

10
Figure 1.1: One of A. W. Phillipss Graphs

11
Figure 1.2: Solow and Samuelsons Description of the Phillips Curve

12
Figure 1.3: The Evolution of US Inflation and Unemployment

US Inflation and Unemployment, 1955-2013

12

10

0
1955 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010
Inflation Unemployment

13
 Figure 1.4: The Failure of the Original Phillips Curve

US Inflation and Unemployment, 1955-2013

Inflation is the Four-Quarter Percentage Change in GDP Deflator
11

10

8
Unemployment

3
0 2 4 6 8 10 12
Inflation

14
Equations: Variables, Parameters and All That

We will use both graphs and equations to describe the models in this class. Now I know many

students dont like equations and believe they are best studiously avoided. However, that

wont be a good strategy for doing well in this course, so I strongly encourage you to engage

with the technical material in this class. It isnt as hard is it might look to start with.

Variables and Coefficients: The equations in this class will generally have a certain format.

They will often look a bit like this.

yt = + xt (1.1)

There are two types of objects in this equation. First, there are the variables, yt and xt .

These will correspond to economic variables that we are interested in (inflation or GDP for

example). We interpret yt as meaning the value that the variable y takes during the time

period t). For most models in this course, we will treat time as marching forward in discrete

intervals, i.e. period 1 is followed by period 2, period t is followed by period t + 1 and so on.

Second, there are the parameters or coefficients. In this example, these are given by and

. These are assumed to stay fixed over time. There are usually two types of coefficients:

Intercept terms like that describe the value that series like yt will take when other variables

all equal zero and coefficients like that describe the impact that one variable has on another.

In this case, if is a big number, then a change in the variable xt has a big impact on yt while

if is small, it will have a small impact.

Some of you are probably asking what those squiggly shapes and are. They

are Greek letters. While its not strictly necessary to use these shapes to represent model

parameters, its pretty common in economics. So let me introduce them: is alpha (Al-Fa),

is beta (Bay-ta), is gamma, is delta, is theta (Thay-ta) and naturally enough is pi.

15
Dynamics: One of the things we will be interested in is how the variables we are looking at

will change over time. For example, we will have equations along the lines of

yt = yt1 + xt (1.2)

Reading this equation, it says that the value of y at time t will depend on the value of x at

time t and also on the value that y took in the previous period i.e. t 1. By this, we mean

that this equation holds in every period. In other words, in period 2, y depends on the value

that x takes in period 2 and also on the value that y took in period 1. Similarly, in period 3,

y depends on the value that x takes in period 3 and also on the value that y took in period

2. And so on.

Subscripts and Superscripts: When we write yt , we mean the value that the variable y takes

at time t. Note that the t here is a subscript it goes at the bottom of the y. Some students

dont realise this is a subscript and will just write yt but this is incorrect (it reads as though

the value t is multiplying y which is not whats going on).

We will also sometimes put indicators above certain variables to indicate that they are

special variables. For example, in the model we present now, you will see a variable written

as te which will represent the publics expectation of inflation. In the model, t is inflation at

time t and the e above the in te is there to signify that this is not inflation itself but rather

it is the publics expectation of it.

16
Our Version of the Phillips Curve

We will use both graphs and equations to describe the various elements of our model. Our first

element is an expectations-augmented Phillips curve which we will formulate as a relationship

in which inflation depends on inflation expectations, the gap between output and its natural

level and a temporary inflationary shock. Our equation for this is the following:

t = te + (yt yt ) + t (1.3)

In this equation represents inflation and by t we mean inflation at time t. The equation

states that inflation at time t depends on three factors:

1. Inflation Expectations, te : This termwhich puts an e superscript above the t

represents the publics inflation expectations at time t. We have put a time subscript

on this variable because the publics expectations may change over time. Note that

a 1 point increase in inflation expectations raises inflation by exactly 1 point. This

is because we are assuming that people bargain over real wages and higher expected

inflation translates one-for-one into their wage bargaining, which in turn is passed into

price inflation.

2. The Output Gap, (yt yt ): This is the gap between GDP at time t, as represented

by yt , and what we will term the natural level of output, which we term yt . This

is the level of output at time t that would be consistent with unemployment equalling

its natural rate. (Note we are describing inflation as being dependent on the output

gap rather than the gap between unemployment and its natural rate because this would

require adding an extra element to the model, i.e. the link between unemployment and

output). We would expect this natural level of output to gradually increase over time as

17
 productivity levels improve. The coefficient (pronounced gamma) describes exactly

how much inflation is generated by a 1 percent increase in the gap between output and

its natural rate.

3. Temporary Inflationary Shocks, t : No model in economics is perfect. So while

inflation expectations and the output gap may be key drivers of inflation, they wont

capture all the factors that influence inflation at any time. For example, supply shocks

like a temporary increase in the price of imported oil can drive up inflation for a while.

To capture these kinds of temporary factors, we include an inflationary shock term,

t . ( is a Greek letter pronounced epsilon). The superscript indicates that this is

the inflationary shock (this will distinguish it from the output shock that we will also

add to the model) and the t subscript indicates that these shocks change over time.

The Phillips Curve Graph

Figure 1.5 shows how to describe our Phillips curve equation in a graph. The graph shows a

positive relationship between inflation and output. The key points to notice are the markings

on the two axes indicating what happens when output is at its natural rate. This graph

illustrates the case when there are no temporary inflationary shocks so t = 0. In this case,

the Phillips curve is just

t = te + (yt yt ) (1.4)

So when yt = yt we get t = te . This is a key aspect of this graph. If you are asked to draw

this graph in the final exam and you just draw an upward-sloping curve without describing

the key points on the inflation and output axes, you wont score many points.

18
 The curve can move up or down depending on what happens to the inflationary shocks, t ,

and with inflation expectations. Figure 1.6 illustrates what happens when there is a positive

inflationary shock so that t goes from being zero to being positive. In this case, even when

output equals its natural level (i.e. yt = yt ) we still get inflation being above its expected

level. Figure 1.7 illustrates how the curve changes when expected inflation rises from 1 to

2 . The whole curve shifts upwards because of the increase in expected inflation.

19
 Figure 1.5: The Phillips Curve with t = 0

Inflation

Output

20
Figure 1.6: The Phillips Curve as we move from t = 0 to t > 0

(An Aggregate Supply Shock)

Inflation PC ( > 0)

PC ( =0)

Output

21
Figure 1.7: The Phillips Curve as we move from te = 1 to te = 2

Inflation PC ( )

PC ( )

Output

22
Model Element Two: The IS Curve

The second element of the model is one that should be familiar to you: An IS curve relating

output to interest rates. The higher interest rates are, the lower output is. However, I want

to stress something here that is often not emphasised in introductory treatments of the IS

curve. The IS relationship is between output and real interest rates, not nominal rates. Real

interest rates adjust the headline (nominal) interest rate by subtracting off inflation.

Think for a second about why it is that real interest rates are what matters. You know

already that high interest rates discourage aggregate demand by reducing consumption and

investment spending. But what is a high interest rate? Suppose I told you the interest rate

was 10 percent. Is this a high interest rate?

You might be inclined to say, Yes, this is a high interest rate but the answer is that it

really depends on inflation. Consider the decision to save for tomorrow or spend today. The

argument for saving is that it can allow you to consume more tomorrow. If the real interest

rate is positive, then this means that you will be able to purchase more goods and services

tomorrow with the money that you set aside. For instance, if the interest rate if 5% but

inflation is 2%, then youll be able to buy 3% more stuff next year because you saved. In

constrast, if the interest rate if 5% but inflation is 8%, then youll be able to buy 3% less stuff

next year even though you have saved your money and earned interest. For these reasons, it

is the real interest rate that determines whether consumers think saving is attractive relative

to spending.

The same logic applies to firms thinking about borrowing to make investment purchases.

If inflation is 10%, then a firm can expect that its prices (and profits) will be increasing by

that much over the next year and a 10% interest rate wont seem so high. But if prices are

23
falling, then a 10% interest rate on borrowings will seem very high.

With these ideas in mind, our version of the IS curve will be the following:

yt = yt (it t r ) + yt (1.5)

Expressed in words, this equation states that the gap between output and its natural rate

(yt yt ) depends on two factors:

1. The Real Interest Rate: The nominal interest rate at time t is represented by it ,

so the real interest rate is it t . The coefficient (pronounced alpha) describes

the effect of a one point increase in the real interest rate on output. The equation has

been constructed in a particular way so that it explicitly defines the real interest rate

at which output will, on average, equal its natural rate. This rate can be termed the

natural rate of interest a term first used by early 20th century Swedish economist

Knut Wicksell. Specifically, we can see from the equation that if yt = 0 then a real

interest rate of r will imply yt = yt .

2. Aggregate Demand Shocks, yt : The IS curve is an even more threadbare model of

output than the Phillips curve model is of inflation. Many other factors beyond the

real interest rate influence aggregate spending decisions. These include fiscal policy,

asset prices and consumer and business sentiment. We will model all of these factors

as temporary deviations from zero of an aggregate demand shock, yt . Note that this

shock has a superscript y to distinguish it from the aggregate supply shock t that

moves the Phillips curve up and down.

24
Model Element Three: A Monetary Policy Rule

Thus far, our model has described how inflation depends on output and how output depends

on interest rates. We can complete the model by describing how interest rates are determined.

Traditionally, in the IS-LM model, this is where the LM curve is introduced. The LM

curve links the demand for the real money stock with nominal interest rates and output, with

a relationship of the form

mt
= it + yt (1.6)
pt
For a given stock of money (mt ) and a given level of prices (pt ), this relationship can be

re-arranged to give a positive relationship between output and interest rates of the form
!
1 mt
yt = + it (1.7)
pt

This positive relationship between output and interest rates is combined with the negative

relationship between these variables in the IS curve to determine unique values for output and

interest rates, something that can be illustrated in a graph with an upward-sloping LM curve

and a downward-sloping IS curve. Monetary policy is then described as taking the form of

the central bank adjusting the money supply mt in a way that sets the position of the LM

curve. The determination of prices is usually described separately in an AS-AD model.

We will not be using the LM curve, for three reasons.

1. Realism 1: In practice, no modern central bank implements its monetary policy by

setting a specified level of the monetary base. Instead, they use their power to supply

potentially unlimited amounts of liquidity to set short-term interest rates to equal a

target level. The supply of base money ends up being whatever emerges from enforcing

the interest rate target. This approach which has been the practice at all the major

25
 central banks for at least 30 years makes the LM curve (and the money supply) of

secondary interest when thinking about core macroeconomic issues. Our approach will

be to describe how the central bank sets interest rates and we wont focus on the money

supply.

2. Realism 2: The traditional approach is for IS-LM to describe the determination of

output and interest rates. Then a separate AS-AD model is used to describe the deter-

mination of prices (and thus, implicitly, inflation). However, the reality is that rather

than being determined independently of inflation, most modern central banks set inter-

est rates with a very close eye on inflationary developments. A model that integrates

the determination of inflation with the setting of monetary policy is thus more realistic.

3. Simplicity: In simplifying the determination of output, inflation and interest rates

down to a single model, this approach is also simpler than one that requires two different

sets of graphs.

We will consider two different types of monetary policy rules that may be followed by the

central bank in our model. The first one is a simple one in which the central bank adjusts its

interest rate in line with inflation with the goal of meeting an inflation target. Specifically,

the first rule we will consider is the following one:

it = r + + (t ) (1.8)

In English, the rule can be intepreted as follows: The central bank adjusts the nominal interest

rate, it , upwards when inflation, t , goes up and downwards when inflation goes down (we

are assuming that > 0) and it does so in a way that means when inflation equals a target

level, , chosen by the central bank, real interest rates will be equal to their natural level.

26
 Thats a bit of a mouthful, so lets see that this is the case and then try to understand

why the monetary policy rule would take this form. First, note what the nominal interest

rate will be if inflation equals its target level (i.e. t = ). The term after the final plus sign

in equation (1.8) will equal zero and the nominal interest rate will be

it = r + (1.9)

In this case, because t = , we can also write this as

i t = r + t (1.10)

So the real interest rate will be

i t t = r (1.11)

Now think about why a rule of this form might be a good idea. Suppose the central bank

has a target inflation rate of that it wants to achieve. Ideally, it would like the public to

understand that this is the likely inflation rate that will occur, so that te = . If that can be

achieved, then the Phillips curve (equation 1.3) tells us that, on average, inflation will equal

provided we have yt = yt . And the IS curve tells us that, on average, we will have yt = yt

when it t = r . So this is a policy that can help to enforce an average inflation rate equal

to the central banks desired target, provided the public adjusts its inflationary expectations

accordingly.

27
The IS-MP Curve

Thats the model. It consists of three equations. Lets pull them together in one place. They

are the Phillips curve:

t = te + (yt yt ) + t (1.12)

The IS curve:

yt = yt (it t r ) + yt (1.13)

And the monetary policy rule:

it = r + + (t ) (1.14)

Now you may recall that I had promised a graphical representation of this model. However,

this is a system of three variables which makes it hard to express on a graph with two axes.

To make the model easier to analyse using graphs, we are going to reduce it down to a system

with two main variables (inflation and output). We can do this because the monetary policy

rule makes interest rates are a function of inflation, so we can substitute this rule into the

IS curve to get a new relationship between output and inflation that we will call the IS-MP

curve.

We do this as follows. Replace the term it in equation (1.13) with the right-hand-side of

equation (1.14) to get

yt = yt [r + + (t )] + (t + r ) + yt (1.15)

Now multiply out the terms in this equation to get

yt = yt r (t ) + t + r + yt (1.16)

28
We can bring together the two terms being multiplied by on its own, and cancel out the

terms in r to get

yt = yt (t ) + (t ) + yt (1.17)

which simplifies to

yt = yt ( 1) (t ) + yt (1.18)

This is the IS-MP curve. It combines the information in the IS curve and the MP curve into

one relationship.

The IS-MP Curve Graph

What would this curve look like on a graph? It turns out it depends especially on the value of

, the parameter that describes how the central bank reacts to inflation. The IS-MP curve

says that an extra unit of inflation implies a change of ( 1) in output. Is this positive

or negative? Well we are assuming that > 0 (we put a negative sign in front of this when

determining the effect of real interest rates on output) so this combined coefficient will be

negative if 1 > 0.

In other words, the IS-MP curve will have a negative slope (slope downwards) provided

the central bank reacts to inflation so that > 1. The explanation for this is as follows.

An increase in inflation of x will lead to an increase in nominal interest rates of x so real

interest rates change by ( 1) x. This means that if > 1 an increase in inflation leads

to higher real interest rates and, via the IS curve relation, to lower output. So if > 1

then the IS-MP curve can be depicted as a downward-sloping curve. In contrast, if < 1,

then an increase in inflation leads to lower real interest rates and higher output, implying an

upward-sloping IS-MP curve.

29
 For now, we will assume that > 1 so that we have a downward-sloping IS-MP curve

but we will revisit this issue after a few more lectures. Figure 1.8 thus shows what the IS-MP

curve looks like when the aggregate demand shock yt = 0. Again, the key thing to notice is

the value of inflation that occurs when output equals its natural rate. When yt = yt we get

t = . As with the Phillips curve, if you are asked to draw this graph in the final exam and

you just draw an downward-sloping curve without describing the key points on the inflation

and output axes, you wont score many points. Figure 1.9 shows how the IS-MP curve shifts

to the right if there is a positive value of yt corresponding to a positive shock to aggregate

demand.

30
 Figure 1.8: The IS-MP Curve with yt = 0

Inflation

Output

31
Figure 1.9: The IS-MP curve as we move from yt = 0 to yt > 0

(An Aggregate Demand Shock)

Inflation

IS-MP ( > 0)

Output

IS-MP ( =0)

32
The Full Model

The full IS-MP-PC model can be illustrated in the traditional fashion as a graph with one

curve that slopes upwards (the Phillips curve) and one that slopes downwards (the IS-MP

curve provided we assume that > 1.) Figure 1.10 provides the simplest possible example

of the graph. This is the case where both the temporary shocks, t and yt equal zero and the

publics expectation of inflation is equal to the central banks inflation target. Note that I have

labelled the PC and IS-MP curves to explicitly indicate what the expected and target rates

of inflation are and it will be a good idea for you to do the same when answering questions

about this model.

In the next set of notes, we will analyse this model in depth, examining what happens

when various types of events occur and focusing carefully on how inflation expectations change

over time.

33
Figure 1.10: The IS-MP-PC Model When Expected Inflation Equals

the Inflation Target

Inflation
IS-MP (

PC ( )

Output

34
A More Complicated Monetary Policy Rule: The Taylor Rule

Before moving on to analyse the model in more depth, I want to describe the more complex

version of the monetary policy rule that I alluded to earlier. This rule takes a form that is

associated with Stanford economist John Taylor. In a famous paper published in 1993 called

Discretion Versus Policy Rules in Practice (you will find a link on the class webpage) Taylor

noted that Federal Reserve policy in the few years before his paper was published seemed to

be characterised by a rule in which interest rates were adjusted in response to both inflation

and the gap between output and an estimated trend.

Within our model structure, we can amend our monetary policy rule to be more like this

Taylor rule if we make it take the following form:

it = r + + (t ) + y (yt yt ) (1.19)

It turns out that the properties of the IS-MP-PC model dont really change if we adopt

this more complicated monetary policy rule. If we substitute the expression for the nominal

interest rate in (4.13) into the IS curve equation (4.6), we get

yt = yt [r + + (t ) + y (yt yt )] + (t + r ) + yt (1.20)

This can be re-arranged as follows (canceling out the terms involving r ):

yt yt = y (yt yt ) (t ) + t + yt (1.21)

Bringing together all the terms involving the output gap yt yt , we get

(1 + y ) (yt yt ) = (t ) + (t ) + yt (1.22)

Which can be expressed as

( 1) 1
yt yt = (t ) + yt (1.23)
1 + y 1 + y

35
This equation shows us that broadening the monetary policy rule to incorporate interest rates

responding to the output gap doesnt change the essential form of the IS-MP curve. As long

as > 1, the curve will slope downwards and will feature t = when yt = yt and there

are no inflationary shocks. So while the addition of an output gap response to the monetary

policy rule changes the coefficients of the IS-MP model a bit, it doesnt change the underlying

economics. In the analysis in the next sets of notes, we will stick with the model that uses

the basic inflation targeting monetary policy rule.

36
Chapter 2

Analysing the IS-MP-PC Model

In our first chapter, we introduced the IS-MP-PC model. We will move on now to examining

its properties.

Inflation Expectations and the Inflation Target

Lets start by repeating a graph from the last time. Figure 2.1 shows the simplest possible

example of the model. This is the case where both the temporary shocks, t and yt equal zero

and the publics expectation of inflation equals the central banks inflation target. Specifically,

the graph shows a case where the publics expectation of inflation te = 1 and the central

banks inflation target is = 1 . With no temporary shocks, the value of output consistent

with t = 1 for the IS-MP curve is yt . Similarly, the value of output consistent with t = 1

for the PC curve is also yt . So the model generates an outcome where t = 1 and yt = yt .

Now consider a case in which the publics inflation expectations shift to being higher than

the central banks target rate. Figure 2.2 illustrates this case. It shows the PC curve shifting

upwards to the red line. This position of this red line is determined by the new higher level

37
of expected inflation. Specifically, the publics inflation expectations are now determined by

te = . Note that is the higher level of inflation noted on the y-axis and that this level is

consistent with yt = yt in the new Phillips curve described by the red line.

The outcomes for inflation and output of the increase in inflation expectations are described

by the intersection of the new red PC line and the old unchanged IS-MP curve. The actual

outcome for inflation (denoted as 2 on the graph) ends up being higher than the central

banks inflation target but lower than the publics inflationary expectations. Output ends up

being lower than its natural rate (consistent with a slump or perhaps a full-blown recession)

because the higher level of inflation leads the central bank to raise real interest rates which

reduces output.

When studying this graph, its important to understand the various markings on the curves

and the axes. If I ask you on the final exam to illustrate the impact of an increase in inflation

expectations using this model, my preference would be to see the various assumptions about

inflation targets and inflation expectations explictly marked out, rather than just a graph that

shows one curve has shifted upwards.

Can We Learn More?

Figure 2.2 is a good example of how we can use graphs to illustrate a models properties. It

gets the basic story across as to what happens when inflation expectations rise above target

when the central bank is pursuing a monetary policy rule that increases real rates in response

to higher inflation.

Still, one could look to dig a bit deeper. The inflation outcome as drawn in Figure 2.2 is

slightly more than halfway towards the publics inflation expectations relative to the central

38
banks inflation target. But what actually determines this outcome? In other words, what de-

termines how far from away target inflation will move when the publics inflation expectations

change? How much does it depend on the monetary policy rule? How much does it depend

on other aspects of the model, like the impact of real interest rates on output and the impact

of output on inflation? It would be tricky to get these answers from a graph. However, using

the equations underlying the model, we can get a full solution that fully answers all these

questions.

39
Figure 2.1: The IS-MP-PC Model When Expected Inflation Equals

the Inflation Target

Inflation
IS-MP (

PC ( )

Output

40
Figure 2.2: The IS-MP-PC Model When Expected Inflation Rises

Above the Inflation Target

PC (
Inflation
IS-MP (
)

PC ( )

Output

41
The IS-MP-PC Model Solution for Inflation

Lets repeat the equations describing our two curves as presented in our last set of notes. The

PC curve is

t = te + (yt yt ) + t (2.1)

And the IS-MP curve is

yt = yt ( 1) (t ) + yt (2.2)

Taking all the other elements of the model as given, we can view this as two linear equations

in the two variables t and yt . These equations can be solved to give solutions that describe

how these two variables depend on all the other elements of the model.

This can be done as follows. First, we will derive a complete expression for the behaviour

of inflation and then derive an expression for output. We derive the expression for inflation by

starting with the Phillips curve and replacing the output gap term yt yt with the variables

that the IS-MP curve tells us determines this gap. This gives us the following equation:

t = te + [ ( 1) (t ) + yt ] + t (2.3)

Adding the term ( 1) t to both sides we get

[1 + ( 1)] t = te + ( 1) + yt + t (2.4)

Now dividing each side by 1 + ( 1), we get that inflation is determined by

yt + t
! !
1 e ( 1)
t = + + (2.5)
1 + ( 1) t 1 + ( 1) 1 + ( 1)

There are a lot of symbols in this equation, which make it a bit hard to read. One way to

simplify it is to take the term multiplying inflation expectations and denote it by a single

42
symbol. In this case, we will denote it by the Greek letter (theta, pronounced thay-ta

with a soft th). So, we define this as

1
= (2.6)
1 + ( 1)

Having done this, we can re-write the equation for inflation as

t = te + (1 ) + (yt + t ) (2.7)

This equation shows that, apart from the shocks to output and inflation (the (yt + t )

terms) inflation is a weighted average of the publics inflation expectations and the central

banks inflation target i.e. it must lie between these two values as long as 0 < < 1 (which

it should be). What determines whether inflation depends more on the publics expectations

or the central banks target? In other words, what determines the value of ? Three different

factors determine this value.

1. : This is the parameter that determines how inflation changes when output changes.

The central bank can only influence inflation by influencing output. If the effect of

output on inflation gets bigger, then the central banks inflation target will have more

influence on outcome for inflation.

2. : This is the parameter that determines how output changes when output interest

rates. If the effect of interest rates on output gets bigger, then the central banks

inflation target will have more influence on the outcome for inflation.

3. : Lets continue to assume > 1 (well return to this in the next set of notes). Then

as gets bigger, the central bank is reacting more to inflation being above its target

level. So this parameter getting bigger means less weight on inflation expectations in

43
 determining the outcome for inflation and more weight on the central banks inflation

target.

The IS-MP-PC Model Solution for Output

Next we provide an expression for output. Looking at the IS-MP curve, we see that the output

gap depends on how far inflation is from the central banks target as well as the supply shock

term t . We can use the equation determining inflation, equation (2.7), to get an expression

for the gap between inflation and the target level. Subtract from either side of equation

(2.7) and you get

t = (te + t + yt ) (2.8)

We can now replace the t on the right-hand-side of the IS-MP curve, equation (4.7),

with the right-hand-side of the equation above. This gives

yt = yt ( 1) (te + t + yt ) + yt (2.9)

which can be simplified to

yt = yt ( 1) (te + t ) + (1 ( 1)) yt (2.10)

This equation tells us that whether output is above or below target depends upon the gap

between expected inflation and the inflation target as well as on the two temporary shocks

yt and t . Provided we have the usual condition that > 1, the combined coefficient

( 1) is negative. This means that increases in the publics inflation expectations

relative to the inflation target end up having a negative effect on output. Inflationary supply

shocks (positive values for t ) also have a negative effect on output while positive aggregate

demand shocks (yt > 0) have a positive effect on output.

44
 How far does output fall short of its natural rate when inflation expectations rise above

the central banks target? The coefficient determining this is ( 1). Remembering

that the size of depends positively on , and , we can say that the same three factors

determine the size of this effect. In other words, the larger the impact of output on inflation,

the larger the impact of interest rates on output and the larger the central bank response to

inflation, the larger the shortfall in output will be when inflation expectations rise above the

central banks target.

While the calculations on these pages may seem difficult, they illustrate that a formal

mathematical solution can sometimes give us a much more complete insight into the properties

of a model than graphs. While graphs are often useful at illustrating a particular feature of a

model, they also often fall short of explainnig the full properties of a model.

How Do Inflation Expectations Change?

Lets go back to Figure 2 now. We have seen that after the publics inflation expectations

rise, the result is a fall in output below its natural rate and in increase inflation, though this

increase is smaller than had been expected by the public. What happens next? How does the

publics expectation of inflation change at this point?

Friedmans 1968 paper The Role of Monetary Policy suggested that people gradually adapt

their expectations based on past outcomes for inflation. Consider now a simple model of this

idea of adaptive expectations by assuming that, each period, the expected level of inflation

is simply equal to the level that prevailed last period. Formally, this can be written as

te = t1 (2.11)

45
Under this formulation of expectations, the Phillips curve becomes

t = t1 + (yt yt ) + t (2.12)

Note that if we subtract t1 from both sides of this equation, it becomes

t t1 = (yt yt ) + t (2.13)

In other words, there should be a positive relationship between the change in inflation and

the output gap. There are various methods for measuring output gaps but one quick and easy

method is to use the unemployment rate as an indication of what the output might be. If

unemployment is high, then output is likely to be below its natural rate so the output gap is

negative. In contrast, a low unemployment rate is an indicator that the output gap is likely to

be positive. So if the adaptive expectations formulation of the Phillips curve was correct, then

we would expect to see a negative relationship in the data between the change in inflation

and the unemployment rate.

Figure 2.3 uses the same US quarterly data that we used for Figure 1.4 in the previous

chapter. That figure showed that there was very little relationship between the level of the

unemployment rate and the level of inflation. Figure 2.3 shows a scatter plot of datapoints

on the change in inflation (measured as the four quarter percentage change in the price level

minus the percentage change in the price level over the preceding four quarters) and the

unemployment rate. In contrast to the basic Phillips curve, this adaptive-expectations-style

Phillips curve shows a clear and strong negative relationship between the change in inflation

and the unemployment rate.

46
Figure 2.3: Evidence for Adaptive Inflation Expectations

Changes in US Inflation and Unemployment, 1955-2013

Change in Inflation is Four-Quarter Inflation Relative to a Year Earlier
11

10

8
Unemployment

2
-7.5 -5.0 -2.5 0.0 2.5 5.0 7.5 10.0
Change in Inflation

47
 These results suggest that the adaptive expectations approach appears to provide a reason-

able model of how people formulate inflation expectations. That said, people are unlikely to

simply use mechanical formulae to arrive at their expectations and one can imagine conditions

in which peoples inflation expectations could radically depart from what had happened in the

past e.g. the appointment of a new central bank governor with a different approach to infla-

tion, the adoption of a new currency or other major events. Lets examine for now, however,

how the IS-MP-PC model behaves when people have adaptive inflationary expectations.

Adjustment of Inflation Expectations

After inflation expectations moved up to , the outcome was that inflation moved from 1

(which is the central banks inflation target) to 2 . If people follow adaptive expectations then

the next period, they will set e = 2 . Figure 2.4 shows what happens after this. The PC

curve moves back downwards and inflation moves down to a lower level, denoted on this graph

by 3 . Figure 2.5 indicates how the process plays out. If the public has adaptive expectations,

then inflation and output gradually converge back to the point where output is at its natural

rate and inflation equals the central banks target rate.

Here we have illustrated the implications of an increase in inflation expectations away

from the central banks inflation target. But if the public has adaptive expectations, how

could inflation expectations just jump upwards? Rather than a random unexplained increase

in inflation expectations, the more likely explanation for the Phillips curve shifting upwards

because is temporary supply shocks, i.e. t is positive for a number of periods. Under adaptive

expectations, the public becomes used to higher inflation and so the Phillips curve will remain

above its long-run position even after the temporary supply shock has been reversed.

48
Figure 2.4: Inflation Expectations Adjusting Back Downwards

PC (
Inflation
IS-MP (
) PC ( )

PC ( )

Output

49
Figure 2.5: Inflation and Output Adjust Back to Starting Position

PC (
Inflation
IS-MP (
)

PC ( )

Output

50
A Temporary Aggregate Demand Shock

Having looked at what happens under adaptive expectations when the Phillips curve shifts,

lets consider what happens when we have a temporary shock to aggregate demand, so yt

takes a different value from zero, which means a shift in the IS-MP curve. In Figures 2.6 to

2.9, we illustrate a case where there is a shift towards a positive value of yt for a couple of

periods but then it shifts back to zero.

Figure 2.6 shows the immediate impact of a positive aggregate demand shock. Output

and inflation both go up with inflation reaching the point denoted as 2 in the figure. If the

public has adaptive expectations, then this results in an increase in inflation expectations the

following period. Figure 2.7 shows what happens when the aggregate demand shock persists

but inflation expectations move up to match the previous periods inflation rate. The inflation

rate now rises again to 3 . Figure 2.8 shows how this triggers a further increase in inflation

in the next period as inflation expectations move up from 2 to 3 .

Figure 2.9 shows what happens if the aggregate demand shock then reverses itself in the

next period. The IS-MP curve shifts back to its original position but the Phillips curve

remains elevated. The result is a nasty combination of high inflation and output below its

natural rate. Figure 2.9 contains arrows showing the full set of movements generated by this

aggregate demand shock:

An increase in output and inflation as the shock hits.

A further increase in inflation as inflation expectations adjust upwards, accompanied by

a decline in output.

A decline in output and inflation as the shock disappears.

51
 A further decline in inflation accompanied by an increase in output as inflationary

expectations gradually return to the central banks target.

This chart shows that when the public has adaptive inflation expectations, temporary

positive aggregate demand shocks lead to counter-clockwise loops on graphs that have output

on the x-axis and inflation on the y-axis.

It turns out that much of the data on inflation and output correspond to these kinds of

movements. Figure 2.10 is borrowed from notes on Stanford economist Charles I. Joness

website. They show the data from US on inflation and an estimated output gap from 1960 to

1983. The figure shows a number of periods of clear counter-cyclical movements. Figure 2.11

shows the same data from 1983-2009. This figure also shows some evidence counter-cylical

loops, thought the movements are smaller than the for the pre-1983 period.

52
 Figure 2.6: A Temporary Aggregate Demand Shock (yt > 0)

IS-MP2 (

Inflation
IS-MP1 (

PC2 ( )

Output

53
 Figure 2.7: Inflation Expectations Adjust Upwards to 2

IS-MP3 ( PC3 ( )

Inflation
IS-MP1 (

PC1 ( )

Output

54
Figure 2.8: Inflation Expectations Adjust Upwards Further to 3

IS-MP4 (
PC4 ( )

Inflation
IS-MP1 (

PC1 ( )

Output

55
Figure 2.9: Reversal of Aggregate Demand Shock Leads to Recession

With High Inflation

PC5 ( )

Inflation
IS-MP5 (

PC1 ( )

Output

56
Figure 2.10: From Chad Joness Notes: US Inflation-Output Loops

1960-1983

57
Figure 2.11: From Chad Joness Notes: US Inflation-Output Loops

1983-2009

58
What if Inflation Expectations Dont Adjust?

The evidence presented in Figure 3 suggests that adaptive expectations seems to be a rea-

sonable model for how people have formulated their expectations of inflation. And it can be

argued that it is a fairly convincing model of how people behave: Most people dont have the

time or knowledge to fully understand exactly whats going in the economy and anticipating

that last years conditions provide a guide to what will happen this year probably works well

enough for most people. Indeed, if the value of is relatively high, then inflation will only

change slowly under adaptive expectations, making the adaptive expectations assumption

more accurate.

All that said, it is also possible to imagine situations in which the publics inflation ex-

pectations are not formed adaptively. For example, if the public believes that the central

bank will always act to return inflation quickly towards its target, then they may assume that

deviations from the target will be temporary.

Figure 2.12 shows how the economy reacts to a temporary positive demand shock when in-

flation expectations dont change. The outcome here is much nicer than the counter-clockwise

cycle described in Figure 9. There is no recession at any point, just a short period of output

being above its natural rate and inflation being above its target, followed by a return to their

starting levels.

59
Figure 2.12: Adjustment if Inflation Expectations Dont Change

IS-MP2 (

Inflation
IS-MP1 (

PC ( )

Output

60
Implications for Monetary Policy

The previous examples provide food for thought about what kind of monetary policy we would

like a central bank to implement. Do we want a soft central bank that limits the increase in

real interest rates when inflation rises to protect the economy and which isnt too concerned

about getting inflation back to target quickly? Or do we want a tough central bank that

raises interest rates aggressively and is very concerned about getting inflation back to target?

Suppose we wish to avoid large recessions if possible. You might imagine the soft central

bank would be more likely to deliver this. However, our model says the exact opposite:

Recessions are smaller and the economy less volatile when the central bank acts aggressively

and is committed to returning inflation to its target rate.

We can see this in two results from the model. First, the economy responds better to shocks

if the central bank reacts strongly to changes in inflation. We can see this from equations (2.7)

and (2.10). Equations (2.6) and (2.7) together tell us that the more aggressively the central

bank responds to inflation (the higher is) the smaller is the increase in inflation in response

to a rise in inflation expectations. Equation (2.10) then tells us this more aggressive approach

means a smaller fall in output, i.e. a small recession. The process of a gradual return to the

natural level of output and the central banks target inflation rate will also be faster if the

original changes were smaller. So recessions are smaller and shorter with an aggressive central

bank.

Second, the more people believe that a central bank is maintaining its low inflation target,

the less likely the economy is to go through boom-bust cycles. We can see this by comparing

the dynamics from Figure 9 where inflation expectations shift over time (perhaps because the

public believes the central bank is willing to be flexible about its target) and Figure 12, which

61
shows what happens when inflation expectations do not change after an expansionary shock.

Implications for Central Bank Institutional Design

These results predict that we get better outcomes if we have a tough central bank that

acts aggressively against inflation and which the public believes is committed to keeping the

economy near its inflation target.

How can this outcome be achieved? The academic literature on this topic has suggested

a number of different ways:

1. Political Independence: A central bank that plans for the long-term (and does not

worry about economic performance during election years) is more likely to stick to a

commitment to low inflation. So, independence from political control is an important

way to reassure the public about the banks credibility.

2. Conservative Central Bankers: If the central banker is known to really dislike

inflationand the public believes this, the economy gets closer to the ideal low in-

flation outcome even without commitment. So the government may choose to appoint

a central banker who is more inflation-averse than they are.

3. Consequence for Bad Inflation Outcomes: Introducing laws so that bad things

happen to the central bankers when inflation is high is one way to make the public

believe the they will commit to a low inflation rate.

These ideas have had a considerable influence on the legal structure of central banks around

the world over the past few decades:

62
1. Political Independence: There has been a substantial move around the world towards

making central banks more independent. Close to home, the Bank of England was made

independent in 1997 (previously the Chancellor of the Exchequer had set interest rates)

and the ECB/Eurosystem is highly independent from political control.

2. Conservative Central Bankers: All around the world, central bankers talk much

more now about the evils of inflation and the benefits of price stability. This may be

because they believe this to be the case. But there is also a marketing element. Perhaps

they can face a better macroeconomic tradeoff if the public believes the central banks

commitment to low inflation.

3. Consequence for Bad Inflation Outcomes: In tandem with the move towards

increased independence, many central banks now have legally imposed inflation targets

and various types of bad things happen to the chief central banker when the inflation

target is not met. For instance, the Governor of the Bank of England has to write a

letter to the Chancellor explaining why the target was not met. The Bank of Englands

inflation targeting handbook provides lots of information on the inflation targeting

regimes adopted around the world over the past 25 years.

63
Chapter 3

The Taylor Principle

Up to now, we have maintained the assumption that the central bank reacts to a change in

inflation by implementing a bigger change in interest rates. In terms of the equation for our

monetary policy rule, this means we are assuming > 1. With this assumption, real interest

rates go up when inflation rises and go down when inflation falls. For this reason, our IS-MP

curve slopes downwards: Along this curve, higher inflation means lower output. Because John

Taylors original proposed rule had the feature that > 1, the idea that monetary policy

rules should have this feature has become known as the Taylor Principle. In these notes, we

discuss why policy rules should satisfy the Taylor principle.

Three Different Cases

Recall from our last set of notes that inflation in the IS-MP-PC model is given by

t = te + (1 ) + (yt + t ) (3.1)

where
!
1
= (3.2)
1 + ( 1)

64
Under adaptive expectations te = t1 and the model can be re-written as

t = t1 + (1 ) + (yt + t ) (3.3)

The value of turns out to be crucial to the behaviour of inflation and output in this

model. We can describe three different cases depending on the value of .

Case 1: > 1

If the Taylor principle is satisfied, then ( 1) > 0. That value being positive means that

1 + ( 1) > 1. The parameter is calculated by dividing 1 by this amount so this gives

us a value of that is positive but less than one. So > 1 translates into the case 0 < < 1.

 
1
Case 2: 1
< < 1

As we reduce below one, ( 1) becomes negative, meaning ( 1) < 0 and 1 +

( 1) < 1. The parameter is calculated by dividing 1 by this amount so this gives us

a value of that is greater than one. As falls farther below one, gets bigger and bigger
 
1
and heads towards infinity as approaches 1
(this is the value of that makes the

denominator in the formula equal zero). As long we assume that stays above this level,

we will get a value of that is positive and greater than one.

 
1
Case 3: 0 < < 1

This produces a pathological case in which 1 + ( 1) < 0 so the value of becomes

negative, meaning an increase in inflation expectations actually reduces inflation. We are not

going to consider this case.

65
Macro Dynamics and Difference Equations

These calculations tell us that as long as the Taylor principle is satisfied, the value of lies

between zero and one but that if slips below one, then becomes greater than one. It

turns out this is a very important distinction. To understand the difference between these

two cases, we need to explain a little bit about difference equations.

A difference equation is a formula that generates a sequence of numbers. In economics,

these sequences can be understood as a pattern over time for a variable of interest. After

supplying some starting values, the difference equation provides a sequence explaining how

the variable changes over time. For example, consider a case in which the first value for a

series is z1 = 1 and then zt follows a difference equation

zt = zt1 + 2 (3.4)

This will give z2 = 3, z3 = 5, z4 = 7 and so on. More relevant to our case is the multiplicative

model

zt = bzt1 (3.5)

For a starting value of z1 = x, this difference equation delivers a sequence of values that looks

like this: x, xb, xb2 , xb3 , xb4 .....

Note that if b is between zero and one, then this sequence converges to zero over time no

matter what value x takes but if b > 1, the sequence will explode off towards either plus or

minus infinity depending on whether the initial value was positive or negative. The same logic

prevails if we add a constant term to the difference equation. Consider this equation:

zt = a + bzt1 (3.6)

If b is between zero and one, then no matter what the starting value is, the sequence converges

66
 a
over time to 1b
but if b > 1, the sequence explodes towards infinity. Similarly, if we add

random shocks to the modelmaking it what is known as a first-order autoregressive or AR(1)

modelthe key thing remains the value of b. If the model is

zt = a + bzt1 + t (3.7)

where t is a series of independently drawn zero-mean random shocks, then the presence of

the shocks will mean the series wont simply converge to a constant or steadily explode. But

as long as we have 0 < b < 1 then the series will tend to oscillate above and below the average
a
value of 1b
while if b > 1 the series will tend to explode to infinity over time.

The Taylor Principle and Macroeconomic Stability

These considerations explain why the Taylor principle is so important. If > 1 then inflation

dynamics in the IS-MP-PC model can be described by an AR(1) model with a coefficient on

past inflation that is between zero and one (the in equation 3.3 plays the role of the coefficient

b in the models just considered.) So a policy rule that satisfies the Taylor principle produces a

stable time series for inflation under adaptive expectations. And because output depends on

the gap between inflation expectations and the central banks inflation target, stable inflation

translates into stable output.

In contrast, once falls below 1, the coefficient on past values of inflation in equation

(3.3) becomes greater than one and the coefficient on the inflation target becomes negative.

In this case, any change in inflation produces a greater change in the same direction next

period and inflation ends up exploding off to either plus or minus infinity. Similarly output

either collapses or explodes.

Why does matter so much for macroeconomic stability? Obeying the Taylor principle

67
means that shocks that boost inflation (whether they be supply or demand shocks) raise real

interest rates (because nominal rates go up by more than inflation does) and thus reduce

output, which contains the increase in inflation and keeps the economy stable. In contrast,

when the falls below 1, shocks that raise inflation result in lower real interest rates and

higher output which further fuels the initial increase in inflation (similarly declines in inflation

are further magnified). This produces an unstable explosive spiral.

You might be tempted to think that the arguments in favour of obeying the Taylor principle

as explained here depends crucially on the assumption of adaptive expectations but this isnt

the case. Even before assuming adaptive expectations, from equation (3.1) we can see that

when > 1, the coefficient on the central banks inflation target is negative. So if you

introduced a more sophisticated model of expectations formation, the public would realise

that the central banks inflation target doesnt have its intended influence on inflation and so

there would no reason to expect this value of inflation to come about. But if people know

that changes in expected inflation are translated more than one-for-one into changes in actual

inflation then this could produce self-fulfilling inflationary spirals, even if the public had a

more sophisticated method of forming expectations than the adaptive one employed here.

Graphical Representation

We can use graphs to illustrate the properties of the IS-MP-PC model when the Taylor

principle is not obeyed. Recall that the IS-MP curve is given by this equation

yt = yt ( 1) (t ) + yt (3.8)

The slope of the curve depends on whether or not > 1. In our previous notes, we assumed

> 1 and so the slope ( 1) < 0, meaning the IS-MP curve slopes down. With

68
 < 1, the IS-MP curve slopes up. Figure 3.1 illustrates the IS-MP-PC model in this case

under the assumption that te = = 1 , i.e. that the public expects inflation to equal the

central banks target.

One technical point about this graph is worth noting. I have drawn the upward-sloping

IS-MP curve as a steeper line than the upward-sloping Phillips curve. On the graph as weve
1
drawn it in inflation-output space, the slope of this curve is (1 )
while the slope of the
1
Phillips curve is . One can show that the condition that (1 )
> is the same as showing

that > 0, i.e. that we are ruling out values of associated with the strange third case

noted above.

Now consider what happens when there is an increase in inflation expectations when

falls below one. Figure 3.2 shows a shift in the Phillips curve due to inflation expectations

increasing from 1 to h (You can see that the value of inflation on the red Phillips curve

when yt = yt is t = h ). Notice now that, because the IS-MP curve is steeper than the

Phillips curve, inflation increases above h to take the higher value of 2 . Inflation overshoots

the publics expected value.

Figure 3.3 shows what happens next if the public have adaptive expectations. In this next

period, we have te = 2 and inflation jumps all the way up to the even higher value of 3 . We

wont show any more graphs but the process would continue with inflation increasing every

period. These figures thus show graphically what weve already demonstrated with equations.

The IS-MP-PC model generates explosive dynamics when the monetary policy rule fails to

obey the Taylor principle.

69
  
1
Figure 3.1: The IS-MP-PC Model when 1 < < 1

Inflation IS-MP (

PC ( )

Output

70
  
e 1
Figure 3.2: An Increase in when 1 < < 1

Inflation IS-MP ( PC2 ( )

PC ( )

Output

71
  
1
Figure 3.3: Explosive Dynamics when 1 < < 1

PC3 ( )
Inflation IS-MP (

PC ( )

Output

72
An Increase in the Inflation Target

Figure 3.4 illustrates what happens in the IS-MP-PC model when the central bank changes

its inflation target. The increase in the inflation target shifts the IS-MP curve upwards i.e.

each level of output is associated with a higher level of inflation. However, because the IS-MP

curve is steeper than the Phillips curve, the outcome is a reduction in inflation. Output also

falls.

Even though this is exactly what our earlier equations predicted (the coefficient on the

inflation target is 1 which is negative in this case) this seems like a very strange outcome.

The central bank sets a higher inflation target and then inflation falls. Why is this?

The answer turns out to reflect the particular form of the monetary policy rule that we

are using. This rule is as follows:

it = r + + (t ) (3.9)

You might expect that a higher inflation target would lead to the central bank setting a lower

interest rate, i.e. they ease up to allow the economy to expand and let inflation move higher.

However, if you look closely at this formula, you can see that an increase in the inflation target

actually leads to a higher interest rates when < 1.

This can be explained as follows. The inflation target appears twice in equation (3.9). It

appears in brackets as part of the inflation gap term t which is muliplied by . If

this was the only place that it appeared, then indeed a higher inflation target would lead to

lower interest rates. However, the first part of rule relates to setting the interest rate so that

when inflation equalled its target, real interest rates would equal their natural rate r . The

rule is set on the basis that if inflation is going to be higher on average, then the nominal

interest rate also needs to be higher if real interest rates are to remain unchanged (this is

73
commonly called the Fisher effect of inflation on interest rates).

Putting these two effects together, we see that an increases of x in the inflation target

raises the nominal interest rate by x due to the real interest rate component and reduces it

by x due to inflation now falling below target. If < 1 then the higher inflation target

results in higher interest rates and thus lower output. This is the pattern shown in Figure

3.4.

74
  
1
Figure 3.4: An Increase in when 1 < < 1

IS-MP (
Inflation
IS-MP (

PC ( )

Output

75
Evidence on Monetary Policy Rules and Macroeconomic Stability

Is there any evidence that obeying the Taylor principle provides greater macroeconomic sta-

bility? Some economists believe there is.

An important paper on this topic was Monetary Policy Rules and Macroeconomic Sta-

bility: Evidence and Some Theory by Richard Clarida, Jordi Gali and Mark Gertler. These

economists reported that estimated policy rules for the Federal Reserve appeared to show a

change after Paul Volcker was appointed Chairman in 1979. They estimated that the post-

1979 monetary policy appeared consistent with a rule in which the coefficient on inflation

that was greater than one while the pre-1979 policy seemed consistent with a rule in which

this coefficient was less than one. They also introduce a small model in which the public

adopts rational expectations (more on what this means later) and show that failure to obey

the Taylor principle can lead to the economy generating cycles based on self-fulfilling fluctu-

ations. They argue that failure to obey the Taylor principle could have been responsible for

the poor macroeconomic performance, featuring the stagflation combination of high inflation

and recession, during the 1970s in the US.

There are a number of differences between the approach taken in Clarida, Gali, Gertler

paper and these notes (in particular, their estimated policy rule is a forward-looking one in

which policy reacts to expected future values of inflation and output) and the econometrics

are perhaps more advanced than you have seen but its still a pretty readable paper and a

nice example of policy-relevant macroeconomic research.

That said, this being economics, there have been some dissenting voices on Clarida, Gali

and Gertlers conclusions. In particular, there is the research of Athanasios Orphanides.1

1
Athanasios Orphanides (2004) Monetary Policy Rules, Macroeconomic Stability, and Inflation: A View

76
Orphanides is critical of Taylor rule regressions that use measures of the output gap that

are based on detrending data from the full sample. This includes information that wasnt

available to policy-makers when they were formulating policy in real time and so perhaps it

is unfair to describe them as reacting to these estimates.

This point is particularly relevant for assessing monetary policy prior to 1979. During

the 1970s, growth rates for major international economies slowed considerably. Policy-makers

thought their economies were falling far short of its potential level. In retrospect it is clear

that potential output growth rates were falling and true output gaps were small. Replacing

the full-sample outgap estimates with Using real-time estimates that were available to the

Fed at the time, Orphanides reports regressions which suggest that the 1970s Fed obeyed the

Taylor rule with respect to reacting to inflation and that their mistake was over-reacting to

inaccurate estimates of the output gap.

from the Trenches Journal of Money, Credit and Banking, vol. 36(2).

77
Chapter 4

The Zero Lower Bound and the

Liquidity Trap
Up to now, we have assumed that the central bank in our model economy sets its interest rate

according to a specific policy rule. Whatever the rule says the interest rate should be, the

central banks sets that interest rate. But what if the rule predicts the central bank should set

interest rates equal to a negative value? Will they?

Well lets step back a minute and think about what the interest rate it in our model

actually means. This interest rate appears in the IS curve because we think higher real

interest rates discourage consumption and investment spending by the private sector. So

the relevant interest rates here are interest rates that private sector agents borrow at. And

indeed, if you look at the interest rates that central banks target, they are generally short-term

money market interest rates on transactions featuring private sector agents on both sides of

the deal. For example, the Federal Reserve targets the federal funds rate, which is the rate at

which banks borrow and lend short-term funds to and from each other. The ECB also targets

money market rates in the Euribor market.

78
 With this in mind, consider why these private sector interest rates are unlikely to ever

be negative. If I loan you $100 and only get $101 back next period, I havent earned much

interest but at least I earned some. A negative interest rate would mean me loaning you $100

and getting back less than that next year. Why would I do that? Since money maintains its

nominal face value, Id be better off just the keep the money in my bank account or under a

mattress.

With this in mind, we are going to adapt our model to take into account that there are

times when the central bank would like to set it below zero but is not able to do so.

The Zero Lower Bound

When will the zero lower bound become a problem for a central bank? In our IS-MP-

PC model, it depends on the form of the monetary policy rule. Up to now, we have been

considering a monetary policy rule of the form

it = r + + (t ) (4.1)

This rule sees the nominal interest rate adjusted upwards and downwards as inflation changes.

So the lower bound problem occurs when inflation goes below some critical value. This

value might be negative, so it may occur when there is deflation, meaning prices are falling.

Amending our model to remove the possibility that interest rates could become negative, our

new monetary policy rule is

it = Maximum [r + + (t ) , 0] (4.2)

Because the intended interest rate of the central bank declines with inflation, this means that

there is a particular inflation rate, ZLB , such that if t < ZLB then the interest rate will

equal zero. So what determines this specific value, ZLB that triggers the zero lower bound?

79
 Algebraically, we can characterise ZLB as satisfying

 
r + + ZLB = 0 (4.3)

This can be re-arranged as

ZLB = r (4.4)

which can be solved to give

r
!
1
ZLB
= (4.5)

Equation (4.5) tells us that three factors determine the value of inflation at which the

central bank sets interest rates equal to zero.

1. The inflation target: The higher the inflation target is, then the higher is the level

of inflation at which a central bank will be willing to set interest rates equal to zero.

2. The natural rate of interest: A higher value of r , the natural real interest rate,

lowers the level of inflation at which a central bank will be willing to set interest rates

equal to zero. An increase in this rate makes central bank raise interest rates and so

they will wait until inflation goes lower than previously to set interest rates to zero.

3. The responsiveness of monetary policy to inflation: Increases in raise the

coefficient on in this formula, increasing the first term and it makes the second term

(which has a negative sign) smaller. Both effects mean a higher translates into a

higher value for ZLB . Central banks that react more aggressively against inflation will

wait for inflation to reach lower values before they are willing to set interest rates to

zero.

80
The IS-MP Curve and the Zero Lower Bound

Given this characterisation of when the zero lower bound kicks in, we need to re-formulate the

IS-MP curve. Once inflation falls below ZLB , the central bank cannot keep cutting interest

rates in line with its monetary policy rule. Recalling that the IS curve

yt = yt (it t r ) + yt (4.6)

We had previously derived the IS-MP curve by substituting in the monetary policy rule

formula (4.1) for it term. This gave us the IS-MP curve as:

yt = yt ( 1) (t ) + yt (4.7)

However, when t ZLB we need to substitute in zero instead of the negative value that the

monetary policy rule would predict. So the IS-MP curve becomes

yt ( 1) (t ) + yt when t > ZLB



yt = (4.8)
yt
yt ZLB


+ r + t + when t

The effect of inflation on output in this revised IS-MP curve changes when inflation moves

below ZLB . Above ZLB , higher values of inflation are associated with lower values of output.

Below ZLB , higher values of inflation are associated with higher values of output. Graphically,

this means the IS-MP curve shifts from being downward-sloping to being upward-sloping when

inflation falls below ZLB . Figure 4.1 provides an example of how this looks.

Equation (4.8) also explains the conditions under which the zero lower bound is likely

to be relevant. If there are no aggregate demand shocks, so yt = 0, then the zero lower

bound is likely to kick in at a point where output is above its natural rate; this is the case

illustrated in Figure 4.1. But this combination of high output and low inflation is unlikely to

be an equilibrium in the model unless the public expects very low inflation or deflation so the

81
Phillips curve intersects the IS-MP curve along the section that has output above its natural

rate and inflation below ZLB .

However, if we have a large negative aggregate demand shock, so that yt < 0, then it is

possible to have output below its natural rate and inflation falling below ZLB . As illustrated

in Figure 4.2, this situation is more likely to be an equilibrium (i.e. this position for the IS-MP

curve is more likely to intersect with the Phillips curve) even if inflation expectations are close

to the inflation target.

82
 Figure 4.1: The IS-MP Curve with the Zero Lower Bound

IS-MP ( =0)

Inflation

Output

83
 Figure 4.2: A Negative Aggregate Demand Shock

IS-MP ( =0)

Inflation
PC ( )
IS-MP ( < 0)

Output

84
The Liquidity Trap

When inflation falls below the lower bound, output is determined by

yt = yt + r + t + yt (4.9)

Inflation is still determined by the Phillips curve

t = te + (yt yt ) + t (4.10)

Using the expression for the output gap when the zero lower bound limit has been reached

from equation (4.9) we get an expression for inflation under these conditions as follows

t = te + (r + t + yt ) + t (4.11)

This can be re-arranged to give

1 1
t = te + r + yt +  (4.12)
1 1 1 1 t

1
The coefficient on expected inflation, 1
is greater than one. So, just as with the Taylor

principle example from the last notes, changes in expected inflation translate into even bigger

changes in actual inflation. As we discussed the last time, this leads to unstable dynamics.

Because these dynamics take place only when inflation has fallen below the zero lower bound,

the instability here relates to falling inflation expectations, leading to further declines in

inflation and further declines in inflation expectations. Because output depends positively

on inflation when the zero-bound constraint binds, these dynamics mean falling inflation (or

increasing deflation) and falling output.

This position in which nominal interest rates are zero and the economy falls into a defla-

tionary spiral is known as the liquidity trap. Figures 4.3 and 4.4 illustrate how the liquidity

85
trap operates in our model. Figure 4.3 shows how a large negative aggregate demand shock

can lead to interest rates hitting the zero bound even when expected inflation is positive.

Figure 4.4 illustrates how expected inflation has a completely different effect when the

zero lower bound has been hit. It shows a fall in expected inflation after the negative demand

shock (this example isnt adaptive expectations because I havent drawn inflation expectations

falling all the way to the deflationary outcome graphed in Figure 4.3). In our usual model

set-up, a fall in expected inflation raises output. However, once at the zero bound, a fall in

expected inflation reduces output, which further reduces inflation.

86
 Figure 4.3: Equilibrium At the Lower Bound

IS-MP ( =0)

Inflation
PC ( )
IS-MP ( < 0)

Output

87
 Figure 4.4: Falling Expected Inflation Worsens Slump

IS-MP ( =0)

Inflation
PC ( )
IS-MP ( < 0)

PC ( )

Output

88
The Liquidity Trap with a Taylor Rule

For the simple monetary policy rule that we have been using, the zero lower bound is hit for

a particular trigger level of inflation. Plugging in reasonable parameter values into equation

(4.5) this trigger value will most likely be negative. In other words, with the monetary policy

rule that we have been using, the zero lower bound will only be hit when there is deflation.

However, if we have a different monetary policy rule this result can be overturned. For

example, remember the Taylor-type rule we considered in the first set of notes

it = r + + (t ) + y (yt yt ) (4.13)

Incorporating the zero lower bound, this would be adapted to be

it = Maximum [r + + (t ) + y (yt yt ) , 0] (4.14)

For this rule, the zero lower bound is hit when

r + + (t ) + y (yt yt ) = 0 (4.15)

This condition can be re-written as

(t ) + y (yt yt ) = r (4.16)

In other words, there are a series of different combinations of inflation gaps and output gaps

that can lead to monetary policy hitting the zero lower bound. For example, if yt = yt the

lower bound will be hit at the value of inflation given by equation (4.5), i.e. the level we have

defined as ZLB . In contrast, inflation could equal its target level but policy would hit the
r +
zero bound if output fell as low as yt y
.

Graphically, we can represent all the combinations of output and inflation that produce

zero interest rates under the Taylor rule as the area under a downward-sloping line in Inflation-

Output space. Figure 4.5 gives an illustration of what this area would look like. We showed

89
in the first set of notes that when we are above the zero bound, the IS-MP curve under the

Taylor rule is of the same downward-sloping form as under our simple inflation targeting rule.

At the zero bound, the arguments weve already presented here also apply so that the IS-MP

curve becomes upward sloping.

Figure 4.6 illustrates two different cases of IS-MP curves when monetary policy follows

a Taylor rule. The right-hand curve corresponds to the case yt = 0 (no aggregate demand

shocks) and this curve only interests with the zero bound area when there is a substantial

deflation. In contrast, the left-hand curve corresponds to the case in which yt is highly negative

(a large negative aggregate demand shocks) and this curve interests with the zero bound area

even at levels of inflation that are positive and arent much below the central banks target.

90
 Figure 4.5: Zero Bound is Binding in Blue Triangle Area

Inflation

Output
Zero Lower
Bound Area

91
 Figure 4.6: Zero Bound Can Be Hit With Positive Inflation

IS-MP ( =0)

Inflation IS-MP ( < 0)

Output

92
The Liquidity Trap: Reversing Conventional Wisdom

An important aspect of this model of the liquidity trap is it shows that some of the predictions

that our model made (and which are now part of the conventional wisdom among monetary

policy makers) do not hold when the economy is in a liquidity trap.

Up to now we have seen that as long as the central bank maintains its inflation targets,

then the model with adaptive expectations predicts that deviations of the publics inflation

expectations from this target will be temporary and the economy will tend to converge back

towards its natural level of output. However, once interest rates have hit the zero bound,

this is no longer the case. Instead, the adaptive expectations model predicts the economy can

spiral into an ever-declining slump.

Similarly, our earlier model predicted that a strong belief from the public that the central

bank would keep inflation at target was helpful in stabilising the economy. However, once

you reach the zero bound, convincing the public to raise its inflation expectations (perhaps

by announcing a higher target for inflation) is helpful.

How to Get Out of the Liquidity Trap?

The most obvious way that a liquidity trap can end is if there is a positive aggregate demand

shock that shifts the IS-MP curve back upwards so that the intersection with the Phillips

curve occurs at levels of output and inflation that gets the economy out of the liquidity trap.

However, in reality, liquidity traps have often occurred during periods when there are

ongoing and persistent slumps in aggregate demand. For example, after decades of strong

growth, the Japanese economy went into a slump during the 1990s. Housing prices crashed

and businesses and households were hit with serious negative equity problems. This type

93
of balance sheet recession doesnt necessarily reverse itself quickly. The result in Japan

has been a persistent deflation since the mid-1990s (see Figure 4.7) combined with a weak

economy. The Bank of Japan has set short-term interest rates close to zero throughout this

period (see Figure 4.8).

Given that economies in liquidity traps tend not to self correct with positive aggregate

demand shocks from the private sector, governments can try to boost the economy by us-

ing fiscal policy to stimulate aggregate demand. Japan has used fiscal stimulus on various

occasions with limited success.

What about monetary policy? With its policy interest rates at zero, can a central bank

do any more to boost the economy? Debates on this topic have focused on two areas.

The first area relates to the fact that while the short-term interest rates that are controlled

by central banks may be zero, that doesnt mean the longer-term rates that many people

borrow at will equal zero. By signalling that they intend to keep short-term rates low for a

long period of time and perhaps by directly intervening in the bond market (i.e. quantitiative

easing) central banks can attempt to lower these longer-term rates.

The second area relates to inflation expectations. Our model tells us that output can be

boosted when the economy is in a liquidity trap by raising inflation expectations. This acts

to raise inflation (or reduce deflation) and this reduces real interest rates and boost output.

As an academic and during his early years as a member of the Federal Reserve Board of

Governors (prior to becoming Chairman) Ben Bernanke advocated that the Bank of Japan

should attempt to raise inflation expectations by committing to having a period of inflation

above their target level of 1%. In a 2003 speech titled Some Thoughts on Monetary Policy

in Japan Bernanke said:

94
 What I have in mind is that the Bank of Japan would announce its intention to

restore the price level (as measured by some standard index of prices, such as the

consumer price index excluding fresh food) to the value it would have reached if,

instead of the deflation of the past five years, a moderate inflation of, say, 1 percent

per year had occurred.

The Bank of Japan did not take Bernankes advice. More recently, however, under pressure

from the new Japanese government, the Bank of Japan have changed their inflation target from

1% per year to 2% per year. Though not as radical as the steps recommended by Bernanke

and others, it is clear that the current Japanese authorities recognise the importance of raising

inflation expectations.

95
Figure 4.7: Consumer Price Index in Japan

96
Figure 4.8: Short-Term Interest Rates in Japan

97
Are the US and Euro Area in Liquidity Traps?

In recent years, the US economy has been in a position that, in some ways, resembles the

position of the Japanese economy during its long liquidity trap period. The economic recov-

ery that started in 2009 has been very weak by historical standards and unemployment has

remained fairly high. In response to this weakness, the Federal Reserve has kept its policy

rate (the federal funds rate) at close to zero since late 2008.

Has the US been in a liquidity trap and should the Fed have been considering more radical

policies such as signalling its intent to allow a temporary rise in inflation? Once Ben Bernanke

became Chairman of the Fed, he was not as keen to implement the ideas he recommended as

an academic. One argument that Bernanke advanced against providing price level guidance

was that the US was not in a liquidity trap because inflation is still positive. However, as

weve seen above, for a central bank that responds to deviations of output from its natural

rate (and clearly the Fed does this) then you can have liquidity-trap conditions even with

positive inflation. The key feature of the liquidity trap is zero short-term rates, not deflation.

And this feature has held in the U.S. for a number of years.

Why did Bernanke not adopt the policy he had recommended to the Japanese? The

explanation seems to be that he was concerned that non-standard policies will undermine the

Feds longer-term credibility. At his April 2012 press conference, he said:

I guess the question is, does it make sense to actively seek a higher inflation rate

in order to achieve a slightly increased reductiona slightly increased pace of

reduction in the unemployment rate? The view of the Committee is that that

would be very reckless. We havewe, the Federal Reservehave spent 30 years

building up credibility for low and stable inflation, which has proved extremely

98
 valuable in that weve been be able to take strong accommodative actions in the

last four or five years to support the economy without leading to an unanchoring

of inflation expectations or a destabilization of inflation. To risk that asset for

what I think would be quite tentative and perhaps doubtful gains on the real side

would be, I think, an unwise thing to do.

This suggests that Chairman Bernanke was still more focused on the benefits of well-anchored

low inflation expectations during normal times than on the potential benefits of non-standard

policies in getting the economy out of a slump.

Nobel prize winner Paul Krugman, Bernankes former colleague at Princeton, was critical

of Bernankes unwillingness to attempt to raise inflationary expectations, most notably in an

article titled Chairman Bernanke should listen to Academic Bernanke.1 . From his research

on Japan in the late 1990s, Krugman has discussed the tension that central bankers feel

when in a liquidity trap. When up against the zero bound, they might like to raise inflation

expectations but then they are concerned that this could make inflation go higher than they

would like. The publics awareness that the central bank will clamp down on inflation if the

economy picks up then prevents there being a sufficient increase in inflation rates to get the

economy out of the liquidity trap. Krugman thus stresses the need for central banks facing a

liquidity trap to commit to being irresponsible as a way out of these slumpscommit to a

temporary period of inflation being higher than you would normally like. But central bankers

are a conservative crowd and even temporary irresponsibility does not come easy to them.

The Fed has adopted a number of new policies in recent years such as quantitative easing
1
This article is at http://www.nytimes.com/2012/04/29/magazine/chairman-bernanke-should-listen-to-

professor-bernanke.html

99
and enhanced forward guidance in which they signal that rates will stay low for a long

period. More recently, they have discussed conditions under which they will reduce their QE

purchases and outlined the conditions under which they will keep rates at zero. So there

are signs that the Fed is willing to be flexible. Time will tell if the debate about price-

level targeting or raising the target inflation rate produces a change in policy at the Fed.

With many unconventional tools having been introduced over the past few years, it will be

interesting to see if any of the leading central banks will consider moving away from narrow

inflation targeting regimes.

Short-term interest in the euro area are also now close to zero. While Mario Draghi argues

there is no cause for concern because there is no deflation as of yet, inflation is extremely low

and the ECB has effectively run out of room for influencing the economy through the regular

interest rate channel. So the euro area also appears to resemble the liquidity trap situation

we have described here. Given the ECBs legal requirement to place price stability above all

other goals, they seem very unlikely to adopt the kind of policies that can take an economy

out of a liquidity trap.

100
Figure 4.9: The Fed Has Been at the Lower Bound Since 2008

101
Figure 4.10: The ECB Has Also Hit the Lower Bound

102
 Part II

Rational Expectations

103
Chapter 5

Rational Expectations and Asset

Prices
One of the things weve focused on is how people formulate expectations about inflation. We

put forward one model of how these expectations were formulated, an adaptive expectations

model in which people formulated their expectations by looking at past values for a series.

Over the next few weeks, we will look at an alternative approach that macroeconomists call

rational expectations. This approach is widely used in macroeconomics and we will cover

its application to models of asset prices, consumption and other macroeconomic variables.

Rational Expectations and Macroeconomics

Almost all economic transactions rely crucially on the fact that the economy is not a one-

period game. In the language of macroeconomists, most economic decisions have an in-

tertemporal element to them. Consider some obvious examples:

We accept cash in return for goods and services because we know that, in the future,

104
 this cash can be turned into goods and services for ourselves.

You dont empty out your bank account today and go on a big splurge because youre

still going to be around tommorrow and will have consumption needs then.

Conversely, sometimes you spend more than youre earning because you can get a bank

loan in anticipation of earning more in the future, and paying the loan off then.

Similarly, firms will spend money on capital goods like trucks or computers largely in

anticipation of the benefits they will bring in the future.

Another key aspect of economic transactions is that they generally involve some level of

uncertainty, so we dont always know whats going to happen in the future. Take two of the

examples just given. While it is true that one can accept cash in anticipation of turning it

into goods and services in the future, uncertainty about inflation means that we cant be sure

of the exact quantity of these goods and services. Similarly, one can borrow in anticipation

of higher income at a later stage, but few people can be completely certain of their future

incomes.

For these reasons, people have to make most economic decisions based on their subjective

expectations of important future variables. In valuing cash, we must formulate an expectation

of future values of inflation; in taking out a bank loan, we must have some expectation of our

future income. These expectations will almost certainly turn out to have been incorrect to

some degree, but one still has to formulate them before making these decisions.

So, a key issue in macroeconomic theory is how people formulate expectations of economic

variables in the presence of uncertainty. Prior to the 1970s, this aspect of macro theory

was largely ad hoc. Different researchers took different approaches, but generally it was

105
assumed that agents used some simple extrapolative rule whereby the expected future value

of a variable was close to some weighted average of its recent past values. However, such

models were widely criticised in the 1970s by economists such as Robert Lucas and Thomas

Sargent. Lucas and Sargent instead promoted the use of an alternative approach which they

called rational expectations. This approach had been introduced in an important 1961

paper by John Muth.

The idea that agents expectations are somehow rational has various possible interpreta-

tions. However, when Muths concept of rational expectations meant two very specific things:

They use publicly available information in an efficient manner. Thus, they do not make

systematic mistakes when formulating expectations.

They understand the structure of the model economy and base their expectations of

variables on this knowledge.

To many economists, this is a natural baseline assumption: We usually assume agents

behave in an optimal fashion, so why would we assume that the agents dont understand

the structure of the economy, and formulate expectations in some sub-optimal fashion. That

said, rational expectations models generally produce quite strong predictions, and these can

be tested. Ultimately, any assessment of a rational expectations model must be based upon

its ability to fit the relevant macro data.

How We Will Describe Expectations

We will start with some terminology to explain how we will represent expectations. Suppose

our model economy has a uncertainty so that people do not know what is going to happen in

106
the future. Then we will write Et Zt+2 to mean the expected value the agents in the economy

have at time t for what Z is going to be at time t + 2. In other words, we assume people have

a distribution of potential outcomes for Zt+2 and Et Zt+2 is mean of this distribution.

It is important to realise that Et is not a number that is multiplying Zt+2 . Instead, it is a

qualifier explaining that we are dealing with peoples prior expectations of a Zt+2 rather than

the actual realised value of Zt+2 itself.

Throughout these notes, we will use some basic properties of the expected value of dis-

tributions. Specifically, we will use the fact that expected values of distributions is what is

known as a linear operator. What is meant by that is that

Et (Xt+k + Yt+k ) = Et Xt+k + Et Yt+k (5.1)

Some examples of this are the following. The expected value of five times a series equals five

times the expected value of the series

Et (5Xt+k ) = 5Et (Xt+k ) (5.2)

And the expected value of the sum of two series equals the sum of the two expected values.

Et (Xt+k + Yt+k ) = Et Xt+k + Et Yt+k (5.3)

We will use these properties a lot, so I wont be stopping all the time to explain that they are

being used.

107
Asset Prices

The first class of rational expectations models that we will look relate to the determination of

asset prices. Asset prices are an increasingly important topic in macroeconomics. Movements

in asset prices affect the wealth of consumers and businesses and have an important influence

on spending decisions. In addition, while most of the global recessions that preceded the year

2000 were due to boom and bust cycles involving inflation getting too high and central banks

slowing the economy to constrain it, the most recent two global recessionsthe dot com

recession of 2000/01 and the great recession of 2008/09were triggered by big declines in

asset prices following earlier large increases. A framework for discussing these movements is

thus a necessary part of any training in macroeconomics.

In these notes, we will start by considering the case of an asset that can be purchased

today for price Pt and which yields a dividend of Dt . While this terminology obviously fits

with the asset being a share of equity in a firm and Dt being the dividend payment, it could

also be a house and Dt could be the net return from renting this house out, i.e. the rent minus

any costs incurred in maintenance or managment fees. If this asset is sold tomorrow for price

Pt+1 , then it generates a rate of return on this investment of

Dt + Pt+1
rt+1 = (5.4)
Pt

This rate of return has two components, the first reflects the dividend received during the

period the asset was held, and the second reflects the capital gain (or loss) due to the price

of the asset changing from period t to period t + 1. This can also be written in terms of the

so-called gross return which is just one plus the rate of return.

Dt + Pt+1
1 + rt+1 = (5.5)
Pt

108
A useful re-arrangement of this equation that we will repeatedly work with is the following:

Dt Pt+1
Pt = + (5.6)
1 + rt+1 1 + rt+1

Asset Prices with Rational Expectations and Constant Expected Returns

We will now consider a rational expectations approach to the determination of asset prices.

Rational expectations means investors understand equation (5.6) and that all expectations of

future variables must be consistent with it. This implies that

" #
Dt Pt+1
Et Pt = Et + (5.7)
1 + rt+1 1 + rt+1

where Et means the expectation of a variable formulated at time t. The stock price at time t

is observable to the agent so Et Pt = Pt , implying

" #
Dt Pt+1
Pt = Et + (5.8)
1 + rt+1 1 + rt+1

A second assumption that we will make for the moment is that the expected return on

assets equals some constant value for all future periods, unrelated to the dividend process.

Et rt+k = r k = 1, 2, 3, ..... (5.9)

One way to think of this is that there is a required return, determined perhaps by the rate

of return available on some other asset, which this asset must deliver. With this assumption

in hand and assuming that everyone knows the value of Dt , equation (5.8) can be re-written

as
Dt Et Pt+1
Pt = + (5.10)
1+r 1+r

109
The Repeated Substitution Method

Equation (5.10) is a specific example of what is known as a first-order stochastic difference

equation.1 Because such equations occur commonly in macroeconomics, it will be useful to

write down the general approach to solving these equations, rather than just focusing only on

our current asset price example. In general, this type of equation can be written as

yt = axt + bEt yt+1 (5.11)

Its solution is derived using a technique called repeated substitution. This works as follows.

Equation (9.1) holds in all periods, so under the assumption of rational expectations, the

agents in the economy understand the equation and formulate their expectation in a way that

is consistent with it:

Et yt+1 = aEt xt+1 + bEt Et+1 yt+2 (5.12)

Note that this last term (Et Et+1 yt+2 ) should simplify to Et yt+2 : It would not be rational if

you expected that next period you would have a higher or lower expectation for yt+2 because

it implies you already have some extra information and are not using it. This is known as the

Law of Iterated Expectations. Using this, we get

Et yt+1 = aEt xt+1 + bEt yt+2 (5.13)

Substituting this into the previous equation, we get

yt = axt + abEt xt+1 + b2 Et yt+2 (5.14)

Repeating this method by substituting in for Et yt+2 , and then Et yt+3 and so on, we get a

general solution of the form

yt = axt + abEt xt+1 + ab2 Et xt+2 + .... + abN 1 Et xt+N 1 + bN Et yt+N (5.15)
1
Stochastic means random or incorporating uncertainty. It applies to this equation because agents do not

actually know Pt+1 but instead formulate expectations of it.

110
which can be written in more compact form as
N 1
bk Et xt+k + bN Et yt+N
X
yt = a (5.16)
k=0

For those of you unfamiliar with the summation sign terminology, summation signs work like

this
2
X
zk = z0 + z1 + z2 (5.17)
k=0
X3
zk = z0 + z1 + z2 + z3 (5.18)
k=0
X4
zk = z0 + z1 + z2 + z3 + z4 (5.19)
k=0

and so on.

The Dividend-Discount Model

Comparing equations (5.10) and (9.1), we can see that our asset price equation is a specific

case of the first-order stochastic difference equation with

y t = Pt (5.20)

xt = Dt (5.21)
1
a = (5.22)
1+r
1
b = (5.23)
1+r

This implies that the asset price can be expressed as follows

N 1  k+1 N
1 1
X 
Pt = Et Dt+k + Et Pt+N (5.24)
k=0 1+r 1+r

Another assumption usually made is that this final term tends to zero as N gets big:
N
1

lim Et Pt+N = 0 (5.25)
N 1+r

111
What is the logic behind this assumption? One explanation is that if it did not hold then

we could set all future values of Dt equal to zero, and the asset price would still be positive.

But an asset that never pays out should be inherently worthless, so this condition rules this

possibility out. With this imposed, our solution becomes

 k+1
X 1
Pt = Et Dt+k (5.26)
k=0 1+r

This equation, which states that asset prices should equal a discounted present-value sum of

expected future dividends, is known as the dividend-discount model.

Explaining the Solution Without Equations

The repeated substitution solution is really important to understand so let me try to explain

it without equations. Suppose I told you that the right way to price a stock was as follows.

Todays stock price should equal todays dividend plus half of tomorrows expected

stock price.

Now suppose its Monday. Then that means the right formula should be

Mondays stock price should equal Mondays dividend plus half of Tuesdays ex-

pected stock price.

It also means the following applies to Tuesdays stock price

Tuesdays stock price should equal Tuesdays dividend plus half of Wednesdays

expected stock price.

If people had rational expectations, then Mondays stock prices would equal

112
 Mondays dividend plus half of Tuesdays expected dividend plus one-quarter of

Wednesdays expected stock price

And being consistent about itfactoring in what Wednesdays stock price should beyoud

get the price being equal to

Mondays dividend plus half of Tuesdays expected dividend plus one-quarter of

Wednesdays expected dividend plus one-eigth of Thursdays expected dividend

and so on.

This is the idea being captured in equation (5.26).

Constant Expected Dividend Growth: The Gordon Growth Model

A useful special case that is often used as a benchmark for thinking about stock prices is the

case in which dividend payments are expected to grow at a constant rate such that

Et Dt+k = (1 + g)k Dt (5.27)

In this case, the dividend-discount model predicts that the stock price should be given by

k
Dt X 1+g

Pt = (5.28)
1 + r k=0 1 + r

Now, remember the old multiplier formula, which states that as long as 0 < c < 1, then

1
1 + c + c2 + c3 + .... = ck =
X
(5.29)
k=0 1c

This geometric series formula gets used a lot in modern macroeconomics, not just in examples
1+g
involving the multiplier. Here we can use it as long as 1+r
< 1, i.e. as long as r (the expected

113
return on the stock market) is greater than g (the growth rate of dividends). We will assume

this holds. Thus, we have

Dt 1
Pt = (5.30)
1 + r 1 1+g
1+r
Dt 1+r
= (5.31)
1 + r 1 + r (1 + g)
Dt
= (5.32)
rg

When dividend growth is expected to be constant, prices are a multiple of current dividend

payments, where that multiple depends positively on the expected future growth rate of

dividends and negatively on the expected future rate of return on stocks. This formula is

often called the Gordon growth model, after the economist that popularized it.2 It is often

used as a benchmark for assessing whether an asset is above or below the fair value implied

by rational expectations. Valuations are often expressed in terms of dividend-price ratios, and

the Gordon formula says this should be

Dt
=rg (5.33)
Pt

Allowing for Variations in Dividend Growth

A more flexible way to formulate expectations about future dividends is to assume that divi-

dends fluctuate around a steady-growth trend. An example of this is the following

Dt = c(1 + g)t + ut (5.34)

ut = ut1 + t (5.35)
2
The formula appeared in Myron Gordons 1962 book The Investment, Financing and Valuation of the

Corporation.

114
These equations state that dividends are the sum of two processes: The first grows at rate

g each period. The second, ut , measures a cyclical component of dividends, and this follows

what is known as a first-order autoregressive process (AR(1) for short). Here t is a zero-mean

random shock term. Over large samples, we would expect ut to have an average value of

zero, but deviations from zero will be more persistent the higher is the value of the parameter

We will now derive the dividend-discount models predictions for stock prices when divi-

dends follow this process. The model predicts that

 k+1
1  
Et c(1 + g)t+k + ut+k
X
Pt = (5.36)
k=0 1+r

Lets split this sum into two. First the trend component,
 k+1 
1 c(1 + g)t X 1+g k
  
Et c(1 + g)t+k
X
= (5.37)
k=0 1+r 1 + r k=0 1 + r
c(1 + g)t 1
= (5.38)
1 + r 1 1+g
1+r
c(1 + g)t 1+r
= (5.39)
1 + r 1 + r (1 + g)
c(1 + g)t
= (5.40)
rg

Second, the cyclical component. Because E(t+k ) = 0, we have

Et ut+1 = Et (ut + t+1 ) = ut (5.41)

Et ut+2 = Et (ut+1 + t+2 ) = 2 ut (5.42)

Et ut+k = Et (ut+k1 + t+k ) = k ut (5.43)

So, this second sum can be written as

 k+1 k
1 ut X
X 
Et ut+k = (5.44)
k=0 1+r 1 + r k=0 1 + r

115
 ut 1
= (5.45)
1 + r 1 1+r
ut 1+r
= (5.46)
1+r1+r
ut
= (5.47)
1+r

Putting these two sums together, the stock price at time t is

c(1 + g)t ut
Pt = + (5.48)
rg 1+r

In this case, stock prices dont just grow at a constant rate. Instead they depend positively on

the cyclical component of dividends, ut , and the more persistent are these cyclical deviations

(the higher is), the larger is their effect on stock prices. To give a concrete example, suppose

r = 0.1. When = 0.9 the coefficient on ut is

1 1
= =5 (5.49)
1+r 1.1 0.9

But if = 0.6, then the coefficient falls to

1 1
= =2 (5.50)
1+r 1.1 0.6

Note also that when taking averages over long periods of time, the u components of

dividends and prices will average to zero. Thus, over longer averages the Gordon growth

model would be approximately correct, even though the dividend-price ratio isnt always

constant. Instead, prices would tend to be temporarily high relative to dividends during

periods when dividends are expected to grow at above-average rates for a while, and would

be temporarily low when dividend growth is expected to be below average for a while. This

is why the Gordon formula is normally seen as a guide to long-run average valuations rather

than a prediction as to what the market should be right now.

116
Unpredictability of Stock Returns

The dividend-discount model has some very specific predictions for how stock prices should

change over time. It implies that the change in prices from period t to period t + 1 should be

 k+1  k+1
X 1 X 1
Pt+1 Pt = Et Dt+k+1 Et Dt+k (5.51)
k=0 1+r k=0 1+r

Taking away the summation signs and writing this out in long form, it looks like this

2 3
" #
1 1 1
    
Pt+1 Pt = Dt+1 + Et+1 Dt+2 + Et+1 Dt+3 + ....
1+r 1+r 1+r
" 2 3 #
1 1 1
  
Dt + Et Dt+1 + Et Dt+2 + ... (5.52)
1+r 1+r 1+r

We can re-arrange this equation in a useful way by grouping together each of the two terms

that involve Dt+1 , Dt+2 , Dt+3 and so on. (There is only one term involving Dt .) This can be

written as follows

2
" #
1 1 1
    
Pt+1 Pt = Dt + Dt+1 Et Dt+1
1+r 1+r 1+r
" 2 3 #
1 1

+ Et+1 Dt+2 Et Dt+2 +
1+r 1+r
" 3 4 #
1 1

+ Et+1 Dt+3 Et Dt+3 + .... (5.53)
1+r 1+r

This equation explains three reasons why prices change from period Pt to period Pt+1 .

Pt+1 differs from Pt because it does not take into account Dt this dividend has been

paid now and has no influence any longer on the price at time t + 1. This is the first

term on the right-hand side above.

Pt+1 applies a smaller discount rate to future dividends because have moved forward one
   2
1 1
period in time, e.g. it discounts Dt+1 by 1+r
instead of 1+r
.

117
 People formulate new expectations for the future path of dividends e.g. Et Dt+2 is gone

and has been replaced by Et+1 Dt+2

In general, the first few items above should not be too important. A single dividend payment

being made shouldnt have too much impact on a stocks price and the discount rate shouldnt
   2
1 1
change too much over a single period (e.g. if r is relatively small, then 1+r
and 1+r

shouldnt be too different.) This means that changing expectations about future dividends

should be the main factor driving changes in stock prices.

In fact, it turns out there is a very specific result linking the behaviour of stock prices with

changing expectations. Ultimately, it is not stock prices, per se, that investors are interested

in. Rather, they are interested in the combined return incorporating both price changes and

dividend payments, as described by equation (5.4). It turns out that movements in stock

returns are entirely driven by changes in dividend expectations.

With a number of lines of algebra (described in an appendix) equation (5.53) can be

re-expressed as

 k
" #
X 1
Pt+1 Pt = Dt + rPt + (Et+1 Dt+k Et Dt+k ) (5.54)
k=1 1+r

Recalling the definition of the one-period return on a stock from equation (1), this return can

be written as
P  1 k
Dt + Pt+1 k=1 (E
1+r t+1 Dt+k Et Dt+k )
rt+1 = =r+ (5.55)
Pt Pt

This is a very important result. It tells us that, if the dividend-discount model is correct,

then the rate of return on stocks depends on how people change their minds about what they

expect to happen to dividends in the future: The Et+1 Dt+k Et Dt+k terms on the right-hand

118
side of equation (5.55) describe the difference between what people expected at time t + 1 for

the dividend at time t + k and what they expected for this same dividend payment at time t.

Importantly, if we assume that people formulate rational expectations, then the return on

stocks should be unpredictable. This is because, if we could tell in advance how people were

going to change their expectations of future events, then that would mean people have not

been using information in an efficient manner. So, with rational expectations, the term in the

summation sign in equation (5.55) must be zero on average and must reflect news that could

not have been forecasted at time t. So the only thing determining changes in stock returns in

the innately unforecastable process of people incorporating completely new information.

One small warning about this result. It is often mis-understood as a prediction that stock

prices (rather than stock returns) should be unpredictable. This is not the case. The series

that should be unpredictable is the total stock return including the dividend payment. Indeed,

the model predicts that a high dividend payments at time t lowers stock prices at time t + 1.

Consider for example a firm that promises to make a huge dividend payment next month

but says they wont make any payments after this for a long time. In that case, we would

expect the price of the stock to fall after the dividend is payment. This shows that, even

with rational expectations, stock prices movements can sometimes be predictable. Because

dividend payments are only made on an occasional basis, this prediction can be tested and

various studies have indeed found so-called ex-dividend effects whereby a stock price falls

after a dividend is paid.

119
Evidence on Predictability and Efficient Markets

The theoretical result that stock returns should be unpredictable was tested in a series of

empirical papers in the 1960s and 1970s, most notably by University of Chicago professor

Eugene Fama and his co-authors. Famas famous 1970 paper Efficient Capital Market: A

Review of Theory and Empirical Work reviewd much of this work. This literature came to

a clear conclusion that stock returns did seem to be essentially unpredictable. The idea that

you could not make easy money by timing the market entered public discussion with Burton

Malkiels famous 1973 book, A Random Walk Down Wall Street being particularly influential.

A random walk is a series whose changes cannot be forecasted and rational expectations

implies that changes in the cumulative return on a stock is unforecastable.

The work of Fama and his co-authors was very important in establishing key facts about

how financial markets work. One downside to this research, though, was the introduction

of a terminology that proved confusing. Famas 1970 paper describes financial market as

being efficient if they fully reflect all available information. In general, the researchers

contributing to this literature concluded financial markets were efficient because stock returns

were difficult to forecast. However, this turned out to be a bit of a leap. It is certainly true

that if stock prices incorporate all available information in the rational manner described

above, then returns should be hard to forecast. But the converse doesnt necessarily apply:

Showing that it was difficult to forecasting stock returns turned out to not be the same thing

as proving that stock markets were efficient.

120
Robert Shiller on Excess Volatility

The idea that financial markets were basically efficient was widely accepted in the economics

profession by the late 1970s. Then, a Yale economist in his mid-thirties, Robert Shiller,

dropped something of a bombshell on the finance profession. Shiller showed that the dividend-

discount model beloved of finance academics completely failed to match the observed volatility

of stock prices.3 Specifically, stock prices were much more volatile than could be justified by

the model.

To understand Shillers basic point, we need to take a step back and think about some

basic concepts relating to the formulation of expectations. First note that the ex post outcome

for any variable can be expressed as the sum of its ex ante value expected by somebody and

the unexpected component (i.e. the amount by which that persons expectation was wrong).

This can be written in a formula as

Xt = Et1 Xt + t (5.56)

From statistics, we know that the variance of the sum of two variables equals the sum of their

two variances plus twice their covariance. This means that the variance of Xt can be described

by

Var (Xt ) = Var (Et1 Xt ) + Var (t ) + 2Cov (Et1 Xt , t ) (5.57)

Now note that this last covariance termbetween the surprise element t and the ex-

ante expectation Et1 Xt should equal zero if expectations are fully rational. If there was

a correlationfor instance, so that a low value of the expectation tended to imply a high

value for the errorthen this would means that you could systematically construct a better
3
Do Stock Prices Move Too Much to be Justified by Subsequent Changes in Dividends? American

Economic Review, June 1981

121
forecast once you had seen the forecast that was provided. For example, if a low forecasted

value tended to imply a positive error then you could construct a better forecast by going for

a higher figure. But this contradicts the idea that investors have rational expectations and

thus use all information efficiently.

So, if expectations are rational, then we have

Var (Xt ) = Var (Et1 Xt ) + Var (t ) (5.58)

The variance of the observed series must equal the variance of the ex ante expectation plus

the variance of the unexpected component. Provided there is uncertainty, so there is some

unexpected component, then we must have

Var (Xt ) > Var (Et1 Xt ) (5.59)

In other words, the variance of the ex post outcome should be higher than the variance of ex

ante rational expectation.

This reasoning has implications for the predicted volatility of stock prices. Equation (5.26)

says that stock prices are an ex ante expectation of a discount sum of future dividends. Shillers

observation was that rational expectations should imply that the variance of stock prices be

less than the variance of the present value of subsequent dividend movements:
"  k+1 #
X 1
Var(Pt ) < Var Dt+k (5.60)
k=0 1+r
A check on this calculation, using a wide range of possible values for r, reveals that this

inequality does not hold: Stocks are actually much more volatile than suggested by realized

movements in dividends.4
4
While technically, the infinite sum of dividends cant be calculated because we dont have data going past

the present, Shiller filled in all terms after the end of his sample based on plausible assumptions, and the

results are not sensitive to these assumptions.

122
 Figure 5.1 on the next page reproduces the famous graph from Shillers 1981 paper showing

actual stock prices (the sold line) moving around much more over time than his discounted

outcome of dividends series (the dashed line).

Figure 5.1: Shillers 1981 Chart Illustrating Excess Volatility

123
Longer-Run Predictability

We saw earlier that the dividend-discount model predicts that when the ratio of dividends to

prices is low, this suggests that investors are confident about future dividend growth. Thus,

a low dividend-price ratio should help to predict higher future dividend growth. Shillers

volatility research pointed out, however, that there appears to be a lot of movements in

stock prices that never turn out to be fully justified by later changes dividends. In fact,

later research went a good bit further. For example, Campbell and Shiller (2001) show that

over longer periods, dividend-price ratios are of essentially no use at all in forecasting future

dividend growth.5 In fact, a high ratio of prices to dividends, instead of forecasting high

growth in dividends, tends to forecast lower future returns on the stock market alebit with a

relatively low R-squared. See Figure 5.2.

This last finding seems to contradict Famas earlier conclusions that it was difficult to

forecast stock returns but these results turn out to be compatible with both those findings

and the volatility results. Famas classic results on predictability focused on explaining short-

run stock returns e.g. can we use data from this year to forecast next months stock returns?

However, the form of predictability found by Campbell and Shiller (and indeed a number of

earlier studies) related to predicting average returns over multiple years. It turns out the

an inability to do find short-run predictability is not the same thing as an inability to find

longer-run predictability.

To understand this, we need to develop some ideas about forecasting time series. Consider

a series that follows the following AR(1) time series process:

yt = yt1 + t (5.61)
5
NBER Working Paper No. 8221.

124
where t is a random and unpredictable noise process with a zero mean. If = 1 then the

change in the series is

yt yt1 = t (5.62)

so the series is what is what we described earlier as a random walk process whose changes

cannot be predicted. Suppose, however, that was close to but a bit less than one, say

= 0.99. The change in the series would now be given by

yt yt1 = 0.01yt1 + t (5.63)

Now suppose you wanted to assess whether you could forecast the change in the series based

on last periods value of the series. You could run a regression of the change in yt on last

periods value of the series. The true coefficient in this relationship is -0.01 with the t being

the random error. This coefficient of -0.01 is so close to zero that you will probably be unable

to reject that the true coefficient is zero unless you have far more data than economists usually

have access to.

But what if you were looking at forecasting changes in the series over a longer time-horizon?

To understand why this might be different, we can do another repeated substitution trick.

The series yt depends on its lagged value, yt1 and a random shock. But yt1 in turn depended

on yt2 and another random shock. And yt2 in turn depended on yt3 and another random

shock. And so on. Plugging in all of these substitutions you get the following.

yt = yt1 + t

= 2 yt2 + t + t1

= 3 yt3 + t + t1 + 2 t2

N 1
= N ytN + k tk
X
(5.64)
k=0

125
Now suppose you wanted to forecast the change in yt over N periods with the value of the

series from N periods ago. This change can be written as

  N 1
N
k tk
X
yt ytN = 1 ytN + (5.65)
k=0

Again the change in yt over this period can be written as a function of a past value of the

series and some random noise. The difference in this case is that the coefficient on the lagged

value doesnt have to be small anymore even if had a near-random walk series. For example,

suppose = 0.99 and N = 50 so were looking at the change in the series over 50 periods.

In this case, the coefficient is (0.9950 1) = 0.4. For this reason, regressions that seek to

predict combined returns over longer periods have found statistically significant evidence of

predictability even though this evidence cannot be found for predicting returns over shorter

periods.

It is very easy to demonstrate this result using any software that can generate random

numbers. For example, in an appendix at the back of the notes I provide a short programme

written for the econometric package RATS. The code is pretty intuitive and could be repeated

for lots of other packages. The programme generates random AR(1) series with = 0.99 by

starting them off with a value of zero and then drawing random errors to generate full time

series. Then regressions for sample sizes of 200 are run to see if changes in the series over one

period, twenty periods and fifty periods can be forecasted by the relevant lagged values. This

is done 10,000 times and the average t-statistics from these regressions are calculated.

Table 1 shows the results. The average t-statistic for the one-period forecasting regression

is -1.24, not high enough to reject the null hypothesis that there is no forecasting power. In

contrast, the average t-statistic for the 20-period forecasting regression is -5.69 and the average

t-statistic for the 50-period forecasting regression is -8.95, so you can be very confident that

126
there is statistically significant forecasting power over these horizons.

The intuition behind these results is fairly simple. Provided is less than one in absolute

value, AR(1) series are what is known as mean-stationary. In other words, they tend to revert

back to their average value. In the case of the yt series here, this average value is zero. The

speed at which you can expect them to return to this average value will be slow if is high but

they will eventually return. So if you see a high value of yt , you cant really be that confident

that it will fall next period but you can be very confident that it will eventually tend to fall

back towards its average value of zero.

Pulling these ideas together to explain the various stock price results, suppose prices were

given by
 k+1
" #
X 1
Pt = Et Dt+k + ut (5.66)
k=0 1+r
where

ut = ut1 + t (5.67)

with being close to one and t being an unpredictable noise series. This model says that

stock prices are determined by two elements. The first is the rational dividend-discount price

and the second is a non-fundamental AR(1) element reflecting non-rational market sentiment.

The latter could swing up and down over time as various fads and manias affect the market.

In this case, statistical research would generate three results:

1. Short-term stock returns would be very hard to forecast. This is partly because of the

rational dividend-discount element but also because changes in the non-fundamental

element are hard to forecast over short-horizons.

2. Longer-term stock returns would have a statistically significant forecastable element,

127
 though with a relatively low R-squared. This is because the fundamental element that

accounts for much of the variation cannot be forecasted while you can detect a statisti-

cally significant forecastable element for the non-fundamantal component.

3. Stock prices would be more volatile than predicted by the dividend-discount model,

perhaps significantly. This is because non-fundamental series of the type described here

can go through pretty long swings which adds a lot more volatility than the dividend-

discount model would predict.

This suggests a possible explanation for the behaviour of stock prices. On average, they

appear to be determined by something like the dividend-discount model but they also have

a non-fundamental component that sees the market go through temporary (but potentially

long) swings in which it moves away from the values predicted by this model.

128
Figure 5.2: Campbell and Shillers 2001 Chart

129
Table 1: Illustrating Long-Run Predictability

130
Example: The Prognosis for U.S. Stock Prices

These results would suggest that it may be possible to detect whether a stock market is over-

valued or under-valued and thus to forecast its future path. Lets consider an example and

look at the current state of the U.S. stock market as measured by the S&P 500 which is a

broad measure of large-capitalisation stocks.

Figure 5.3 shows that the U.S. market has been on a tear over the past few years and has

moved well past previous historical highs. The last two times the market expanded rapidly

ended in big crashes. Will this end the same way?

Figure 5.4 shows the ratio of dividends to prices for the S&P 500 index of US stocks over

the period since the second world war. The measure of dividends used in the numerator is

based on the average value of dividends over a twelve month period to smooth out volatile

month-to-month movements. The chart shows that the dividend-price ratio for this index, at

about 2 percent over the past few years, is very low by historical standards. In other words,

prices are very high relative to dividends, which is normally a bad sign.

Still, comparisons with the long-run historical average might be a bad idea. The average

value of this ratio over the period since 1945 is 3.3 percent. The only point since the mid-1990s

that the ratio has exceeded that historical average was a brief period in early 2009 due to the

plunge in stock prices after the Lehman Brothers bankruptcy and the emergence of the worst

recession since the Great Depression.

One reason for this change is that many firms have moved away from paying dividends. In

more recent years, it is particularly clear that dividends have been low relative to how much

companies can afford to pay. The ratio of dividends to earnings for S&P 500 firms is about

half its historical level. There are two reasons for this. There has been a long-run trend of

131
moving away from paying dividends and towards using earnings to fund share repurchases.

This increases the average value of outstanding shares (each of them can get a higher share

of future dividend payments) without the shareholders explicitly receiving dividend income

at present (which would be taxable). In recent years, the reduction in dividend payout rates

may be related to corporate deleveraging, as US firms pay down the large amounts of debt

built up during the period prior to the financial crisis and build up cash buffers as a protection

against the kind of problems that many first ran into during the great recession of 2008/09.

Because of these factors, many analysts instead look at the ratio of total corporate earnings

to prices. Figure 5.4 compares this to the dividend-price ratio. This ratio of average earnings

over the previous twelve months to prices was about 5.1 percent in June 2014. This series had

been is right in line with its historical average of about 6.9 percent in late 2011 but over the

past couple of years it has fallen behind it. So this series suggests that stocks are perhaps a

bit over-valued relative to historical norms but not as much as the dividend-price ratio.

A final consideration, however, is the discount rate being used to value stocks. We know

from the Gordon growth model that one reason stock prices might be high relative to dividends

(or earnings) is that the expected rate of return r may be low. Assuming the required rate

of return on stocks reflects some premium over safe investments such government bonds, this

could provide another explanation for high stock price valuations. Figure 6 shows that real

interest rates on US Treasury bonds are at historically low levels (this series is the yield on

ten-year Treasury bonds minus inflation over the previous year). If these rates are being used

as a benchmark for calculating the required rate of return on stocks, then one might expect

stock prices to be high relative to dividends.

This shows that figuring out whether stocks are under- or over-valued is rarely easy. My

132
assessment is they are probably about in line with reasonable valuations but I may well prove

to be wrong.

Figure 5.3: The S&P 500 Index of U.S. Stock Prices

2250

2000

1750

1500

1250

1000

750

500

250

0
1950 1960 1970 1980 1990 2000 2010

133
 Figure 5.4: Dividend-Price Ratio for S&P 500

0.07

0.06

0.05

0.04

0.03

0.02

0.01
1950 1960 1970 1980 1990 2000 2010

134
Figure 5.5: Dividend-Price and Earnings-Price Ratios for S&P 500

0.175

0.150

0.125

0.100

0.075

0.050

0.025

0.000
1950 1960 1970 1980 1990 2000 2010
Earnings/Price Dividends/Price

135
 Figure 5.6: Real 10-Year Treasury Bond Rate

16

14

12

10

0
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010

136
Time-Varying Expected Returns

I have suggested an alternative explanation of the facts to the dividend-discount model

one in which stock prices are also determined by a temporary but volatile non-fundamental

component. However, the last observation in the previous section about discount rates

suggests another way to mend the dividend-discount model and perhaps explain the

extra volatility that affects stock prices: Change the model to allow for variations in expected

returns. Consider the finding that a high value of the dividend-price ratio predicts poor future

stock returns. Shiller suggests that this is due to temporary irrational factors gradually

disappearing. But another possibility is that that the high value of this ratio is rationally

anticipating low future returns.

We can reformulate the dividend-discount model with time-varying returns as follows. Let

Rt = 1 + rt (5.68)

Start again from the first-order difference equation for stock prices
Dt Pt+1
Pt = + (5.69)
Rt+1 Rt+1
where Rt+1 is the return on stocks in period t + 1. Moving the time-subscripts forward one

period, this implies

Dt+1 Pt+2
Pt+1 = + (5.70)
Rt+2 Rt+2
Substitute this into the original price equation to get
!
Dt 1 Dt+1 Pt+2
Pt = + +
Rt+1 Rt+1 Rt+2 Rt+2
Dt Dt+1 Pt+2
= + + (5.71)
Rt+1 Rt+1 Rt+2 Rt+1 Rt+2
Applying the same trick to substitute for Pt+2 we get
Dt Dt+1 Dt+2 Pt+3
Pt = + + + (5.72)
Rt+1 Rt+1 Rt+2 Rt+1 Rt+2 Rt+3 Rt+1 Rt+2 Rt+3

137
The general formula is
N 1
Dt+k Pt+N
X
Pt = k+1 + N

(5.73)
Q Q
k=0 Rt+m Rt+m
m=1 m=1
h
Q
where xi means the product of x1 , x2 .... xh . Again setting the limit of the t + N term to
n=1

zero and taking expectations, we get a version of the dividend-discount model augmented to

account for variations in the expected rate of return.

D
X t+k
Pt = Et k+1 (5.74)


Q
k=0 Rt+m
m=1

This equation gives a potential explanation for the failure of news about dividends to explain

stock price fluctuations. Stock prices depend positively on expected future dividends. But

they also depend negatively on the Rt+k values which measure the expected future return on

stocks. So perhaps news about future stock returns explains movements in stock prices: When

investors learn that future returns are going to be lower, this raises current stock prices.

In 1991, Eugene Fama provided his updated overview of the literature on the predictability

of stock returns. By this point, Fama accepted the evidence on long-horizon predictability and

had contributed to this literature. However, Fama and French (1988) put forward predictable

time-variation in expected returns as the likely explanation for this result.6 This explanation

has also been promoted by leading modern finance economists such as Harvards John Camp-

bell and Chicagos John Cochrane and is currently the leading hypothesis for reconciling the

evidence on stock prices movements with rational expectations.7

6
Eugene Fama and Kenneth French (1988). Dividend Yields and Expected Stock Returns Journal of

Financial Economics, Volume 22, pages 3-25.

7
Campbells 1991 paper, A Variance Decomposition for Stock Returns, Economic Journal, provides a

nice framework for understanding this theory.

138
What About Interest Rates?

Changing interest rates on bonds are the most obvious source of changes in expected returns

on stocks. Up to now, we only briefly discussed what determines the rate of return that

investors require to invest in the stock market, but it is usually assumed that there is an

arbitrage equation linking stock and bond returns, so that

Et rt+1 = Et it+1 + (5.75)

In other words, next periods expected return on the market needs to equal next periods

expected interest rate on bonds, it+1 , plus a risk premium, , which we will assume is constant.

Are interest rates the culprit accounting for the volatility of stock prices? They are cer-

tainly a plausible candidate. Stock market participants spend a lot of time monitoring the Fed

and the ECB and news interpreted as implying higher interest rates in the future certainly

tends to provoke declines in stock prices. Perhaps surprisingly, then, Campbell and Shiller

(1988) showed that this type of equation still doesnt help that much in explaining stock mar-

ket fluctuations.8 Their methodology involved plugging in forecasts for future interest rates

and dividend growth into the right-hand-side of (5.74) and checking how close the resulting

series is to the actual dividend-price ratio. They concluded that expected fluctuations in in-

terest rates contribute little to explaining the volatility in stock prices. A study co-authored

by Federal Reserve Chairman Ben Bernanke examining the link between monetary policy and

the stock market came to the same conclusions.9

8
The Dividend-Price Ratio and Expectations of Future Dividends and Discount Factors, Review of Fi-

nancial Studies, Autumn 1988.

9
See What Explains the Stock Markets Reaction to Federal Reserve Policy?, by Ben Bernanke and

Kenneth Kuttner, Journal of Finance, June 2005.

139
Time-Varying Risk Premia or Behavioural Finance?

So, changes in interest rates do not appear to explain the volatility of stock market fluctuations.

The final possible explanation for how the dividend-discount model may be consistent with the

data is that changes in expected returns do account for the bulk of stock market movements,

but that the principal source of these changes comes, not from interest rates, but from changes

in the risk premium that determines the excess return that stocks must generate relative to

bonds: The in equation (5.75) must be changing over time. According to this explanation,

asset price booms are often driven by investors being willing to take risks and receive a

relatively low compensation for them (when investors are risk-on in the commonly-used

market terminology) while busts often happen when investors start to demand higher risk

premia (when they are risk-off).

One problem with this conclusion is that it implies that, most of the time, when stocks are

increasing it is because investors are anticipating lower stock returns at a later date. However,

the evidence that we have on this seems to point in the other direction. For example, surveys

have shown that even at the peak of the most recent bull market, average investors still

anticipate high future returns on the market.

If one rejects the idea that, together, news about dividends and news about future returns

explain all of the changes in stock prices, then one is forced to reject the rational expecta-

tions dividend-discount model as a complete model of the stock market. What is missing

from this model? Many believe that the model fails to take into account of various human

behavioural traits that lead people to act in a manner inconsistent with pure rational expec-

tations. Economists like Shiller point to the various asset price bubbles of the past twenty

years such as the dot-com boom and bust and the rise and fall in house prices in countries

140
like the U.S. and Ireland, as clear evidence that investors go through periods of irrational ex-

huberance which sees asset prices become completely detached from the fundamental values

suggested by reasonable applications of the dividend-discount model.

Indeed, the inability to reconcile aggregate stock price movements with rational expecta-

tions is not the only well-known failure of modern financial economics. For instance, there

are many studies documenting the failure of rational optimisation-based models to explain

various cross-sectional patterns in asset returns, e.g. why the average return on stocks ex-

ceeds that on bonds by so much, or discrepancies in the long-run performance of small- and

large-capitalisation stocks. Eugene Fama is the author of a number of famous papers with

Kenneth French that have demonstrated these discrepancies though he interprets these results

as most likely due to a rational pricing of the risk associated with certain kinds of assets.

For many, the answers to these questions lie in abandoning the pure rational expecta-

tions, optimising approach. Indeed, the field of behavioural finance is booming, with various

researchers proposing all sorts of different non-optimising models of what determines asset

prices.10 That said, at present, there is no clear front-runner alternative behavioural-finance

model of the determination of aggregate stock prices.

As a final note on this topic, in an interesting development, the 2013 Nobel Prize for

Economics was shared by Shiller and Fama along with Lars-Peter Hansen who has also worked

on pricing of financial assets. This provoked a widespread discussion of the contributions of

these economists and the current state of play of the finance profession.

10
The papers presented at the bi-annual NBER workshop on behavioural finance give a good flavour of this

work. See the link on the webpage.

141
Appendix 1: Proof of Equation (5.54)

We start by repeating equation (5.53):

2
" #
1 1 1
    
Pt+1 Pt = = Dt + Dt+1 Et Dt+1
1+r 1+r 1+r
" 2 3 #
1 1

+ Et+1 Dt+2 Et Dt+2 +
1+r 1+r
" 3 4 #
1 1

+ Et+1 Dt+3 Et Dt+3 + ....
1+r 1+r

This shows that the change in stock prices is determined by a term relating to this periods

dividends dropping out and then a whole bunch of terms that involve period t + 1 and

period t expectations of future dividends. To be able to pull all the terms for each Dt+k
 k
1
together, we both add and subtract a set of terms of the form 1+r
Et Dt+k . The equation

then looks like this

1
 
Pt+1 Pt = = [Dt+1 Et Dt+1 ]
1+r
2
1

+ [Et+1 Dt+2 Et Dt+2 ]
1+r
3
1

+ [Et+1 Dt+3 Et Dt+3 ] + ....
1 + r
1
 
Dt
1 + r
1 1
  
+ 1 Et Dt+1
1+r 1+r
2
1 1
 
+ 1 Et Dt+2
1+r 1+r
3
1 1
 
+ 1 Et Dt+3 + ... (5.76)
1+r 1+r

The sequence summarised on the first three lines of equation (5.76) can be described using a

summation sign as
 k
" #
X 1
(Et+1 Dt+k Et Dt+k ) (5.77)
k=1 1+r

142
This is an infinite discounted sum of changes to peoples expectations about future dividends.

The sequence summarised on the last three lines of equation (5.76) can be simplified to be

r 1 r
       
(1 + r) Pt Dt = rPt Dt (5.78)
1+r 1+r 1+r

Using these two simplifications, equation (5.76) can be re-written as

k
" #
r 1 1
   X
Pt+1 Pt = Dt Dt + rPt + (Et+1 Dt+k Et Dt+k )
1+r 1+r k=1 1+r
 k
" #
X 1
= Dt + rPt + (Et+1 Dt+k Et Dt+k ) (5.79)
k=1 1+r

which is the equation we were looking for: So the return on stocks can be written as
P  1 k
Dt + Pt+1 k=1 (E 1+r t+1 Dt+k Et Dt+k )
rt+1 = =r+ (5.80)
Pt Pt

143
Appendix 2: Programme For Return Predictability Results

Below is the text of a programme to generate the return predictability results reported in

Table 1. The programme is written for the econometric package RATS but a programme of

this sort could be written for any package that has a random number generator.

allocate 10000
set y = 0
set tstats_1lag = 0
set tstats_20lag = 0
set tstats_50lag = 0
do k = 1,10000
set y 2 300 = 0.99*y{1} + %ran(1)
set dy = y - y{1}
set dy20 = y - y{20}
set dy50 = y - y{50}
linreg(noprint) dy 101 300
# y{1}
comp tstats_1lag(k) = %tstats(1)
linreg(noprint) dy20 101 300
# y{20}
comp tstats_20lag(k) = %tstats(1)
linreg(noprint) dy50 101 300
# y{50}
comp tstats_50lag(k) = %tstats(1)
end do k
stats tstats_1lag
stats tstats_20lag
stats tstats_50lag

144
Chapter 6
Consumption and Asset Pricing
Elementary Keynesian macro theory assumes that households make consumption decisions

based only on their current disposable income. In reality, of course, people have to base their

spending decisions not just on todays income but also on the money they expect to earn in

the future. During the 1950s, important research by Ando and Modigliani (the Life-Cycle

Hypothesis) and Milton Friedman (the Permanent Income Hypothesis) presented significant

evidence that people plan their expenditures in system pattern, smoothing consumption over

time even when their incomes fluctuated.

In these notes, we will use the techniques developed in the last topic to derive a rational

expectations version of the Permanent Income Hypothesis. We will use this model to illustrate

some pitfalls in using econometrics to assess the effects of policy changes. We will discuss

empirical tests of this model and present some more advanced topics. In particular, we will

discuss the link between consumption spending and the return on various financial assets.

145
The Household Budget Constraint

We start with an identity describing the evolution of the stock of assets owned by households.

Letting At be household assets, Yt be labour income, and Ct stand for consumption spending,

this identity is

At+1 = (1 + rt+1 ) (At + Yt Ct ) (6.1)

where rt+1 is the return on household assets at time t + 1. Note that Yt is labour income

(income earned from working) not total income because total income also includes the capital

income earned on assets (i.e. total income is Yt + rt+1 At .) Note, we are assuming that Yt is

take-home labour income, so it can considered net of taxes.

As with the equation for the return on stocks, this can be written as a first-order difference

equation in our standard form

At+1
At = Ct Yt + (6.2)
1 + rt+1

We will assume that agents have rational expectations. Also, in this case, we will assume that

the return on assets equals a constant, r. This implies

1
A t = C t Yt + Et At+1 (6.3)
1+r

Using the same repeated substitution methods as before this can be solved to give

X Et (Ct+k Yt+k )
At = (6.4)
k=0 (1 + r)k
Note that we have again imposed the condition that the final term in our repeated substitution
Et At+k
(1+r)k
goes to zero as k gets large. Effectively, this means that we are assuming that people

consume some of their capital income (i.e. that assets are used to finance a level of consumption

Ct that is generally larger than labour income Yt ). If this is the case, then this term tends to

zero.

146
 One way to understand this equation comes from re-writing it as

X Et Ct+k X Et Yt+k
k = A t + k (6.5)
k=0 (1 + r) k=0 (1 + r)

This is usually called the intertemporal budget constraint. It states that the present value sum

of current and future household consumption must equal the current stock of financial assets

plus the present value sum of current and future labour income.

A consumption function relationship can be derived from this equation by positing some

theoretical relationship between the expected future consumption values, Et Ct+k , and the

current value of consumption. This is done by appealing to the optimising behaviour of the

consumer.

Piketty and r > g

Some of you may be aware of Thomas Pikettys now infamous book Capital in the Twenty

First Century. If youre not, scoll down a few pages to check him out. Perhaps Pikettys

most famous conjecture is there is a natural tendency in capitalist economies for wealth to

accumulate faster than income. This conjecture can be understood on the basis of the simple

budget identity we are working with here.

Consider the simple version of our budget constraint with a constant return on assets

At+1 = (1 + r) (At + Yt Ct ) (6.6)

If assets grew at rate r or faster, then this would likely mean they were growing faster than

GDP, because r is generally higher than GDP growth. So what is the growth rate of the stock

of assets? We can calculate the change in assets as

At+1 At = rAt + (1 + r) (Yt Ct ) (6.7)

147
So the growth rate of assets is given by

At+1 At (1 + r) (Yt Ct )
=r+ (6.8)
At At

This means the growth rate of assets equals r plus an additional term that will be positive

as long as Yt > Ct i.e. as long as labour income is greater than consumption. So this tells us

that the growth rate of assets equals r plus a term that depends upon whether consumption

is greater than or less than labour income. If consumption is less than labour income, assets

grow at a rate that is greater than r while they will grow at a rate slower than r if consumption

is greater than labour income.

Piketty bases his ideas about the tendency for wealth to rise faster than income on the

fact that the rate of return on assets r has tended historically to be higher than the growth

rate of GDP. If we observed Yt > Ct , then assets would grow at a rate greater than r and

so this would generally also be higher than the growth rate of GDP. In general, however, we

probably dont expect consumption to be greater than labour income. If the income people

earn from their assets doesnt ever boost their consumption spending, then what is the point

of it? And indeed, the data generally show that consumption is greater than labour income,

so that people consume some of their capital income (i.e. their income from assets) and total

assets should generally grow at a rate that is less than r. Still, Piketty points out that it is

possible for people to consume some of their capital income and still have assets growing at a

rate smaller than r but greater than g.1

Under what conditions will assets grow at a faster rate than the growth rate of GDP, which
1
For example, page 564: If r > g, it suffices to reinvest a fraction of the return on capital equal to the
growth rate g and consume the rest (r g).

148
Piketty terms g? Our previous equation tells us this happens when

(1 + r) (Yt Ct )
g<r+ (6.9)
At

This can be re-arranged to give

C t Yt rg
< (6.10)
At (1 + r)
So assets will grow faster than incomes if the amount of peoples capital income that they

consume (i.e. the amount they consume above their labour income) as a share of total assets

is below the specific value on the right-hand-side.

Is there any result in economics that leads us to believe that this last inequality should

generally hold? Not to my knowledge. In this sense, Piketty perhaps overstates the extent

to which, on its own, the fact that r > g is a fundamental force for divergence. What is

required for assets to steadily grow relative to income is not only this condition but also an

additional, relatively arbitrary, restriction on how much people can consume and this latter

condition may or may not hold at various times. However, what can be said is that during

periods of high returns on capital, when the gap between r and g is particularly high, then

the bigger the right-hand-side of equation (6.10) will be and it is perhaps more likely that the

condition above will be held.

Most likely, however, the key empirical developments that Pikettys book focuses onrising

assets relative to income and growing inequality of wealthare being driven by other forces

that are making the income distribution more unequal and reducing the share of income going

to workers rather than being related to some innate law of capitalism that drives wealth up

at faster pace than incomes.

149
Figure 6.1: Thomas Piketty

150
Optimising Behaviour by the Consumer

We will assume that consumers wish to maximize a welfare function of the form

!k
X 1
W = U (Ct+k ) (6.11)
k=0 1+

where U (Ct ) is the instantaneous utility obtained at time t, and is a positive number that

describes the fact that households prefer a unit of consumption today to a unit tomorrow.

If the future path of labour income is known, consumers who want to maximize this welfare

function subject to the constraints imposed by the intertemporal budget constraint must solve

the following Lagrangian problem:

!k "
#
X 1 Yt+k
X X Ct+k
L (Ct, Ct+1 , .....) = U (Ct+k ) + At + k k (6.12)
k=0 1+ k=0 (1 + r) k=0 (1 + r)

For every current and future value of consumption, Ct+k , this yields a first-order condition of

the form
!k
1
U 0 (Ct+k ) =0 (6.13)
1+ (1 + r)k
For k = 0, this implies

U 0 (Ct ) = (6.14)

For k = 1, it implies
!
0 1+
U (Ct+1 ) = (6.15)
1+r
Putting these two equations together, we get the following relationship between consumption

today and consumption tomorrow:

!
0 1+r
U (Ct ) = U 0 (Ct+1 ) (6.16)
1+

When there is uncertainty about future labour income, this optimality condition can just be

re-written as
!
0 1+r
U (Ct ) = Et [U 0 (Ct+1 )] (6.17)
1+

151
This implication of the first-order conditions for consumption is sometimes known as an Euler

equation.

In an important 1978 paper, Robert Hall proposed a specific case of this equation.2 Halls

special case assumed that

U (Ct ) = aCt + bCt2 (6.18)

r = (6.19)

In other words, Hall assumed that the utility function was quadratic and that the real interest

rate equalled the household discount rate. In this case, the Euler equation becomes

a + 2bCt = Et [a + 2bCt+1 ] (6.20)

which simplifies to

Ct = Et Ct+1 (6.21)

This states that the optimal solution involves next periods expected value of consumption

equalling the current value. Because, the Euler equation holds for all time periods, we have

Et Ct+k = Et Ct+k+1 k = 1, 2, 3, ..... (6.22)

So, we can apply repeated iteration to get

Ct = Et (Ct+k ) k = 1, 2, 3, ... (6.23)

In other words, all future expected values of consumption equal the current value. Because it

implies that changes in consumption are unpredictable, this is sometimes called the random

walk theory of consumption.

2
Stochastic Implications of the Life-Cycle Permanent Income Hypothesis: Theory and Evidence, Journal
of Political Economy, December 1978.

152
The Rational Expectations Permanent Income Hypothesis

Halls random walk hypothesis has attracted a lot of attention in its own right, but rather

than focus on what should be unpredictable (changes in consumption), we are interested in

deriving an explicit formula for what consumption should equal.

To do this, insert Et Ct+k = Ct into the intertemporal budget constraint, (6.5), to get

X Ct X Et Yt+k
k = At + k (6.24)
k=0 (1 + r) k=0 (1 + r)

Now we can use the geometric sum formula to turn this into a more intuitive formulation:

X 1 1 1+r
k = 1 = (6.25)
k=0 (1 + r) 1 1+r r

So, Halls assumptions imply the following equation, which we will term the Rational Expec-

tations Permanent Income Hypothesis:

r r X Et Yt+k
Ct = At + (6.26)
1+r 1 + r k=0 (1 + r)k

This equation is a rational expectations version of the well-known permanent income hypoth-

esis (I will use the term RE-PIH below) which states that consumption today depends on a

persons expected lifetime sequence of income.

Lets look at this equation closely. It states that the current value of consumption is driven

by three factors:

The expected present discounted sum of current and future labour income.

The current value of household assets. This wealth effect is likely to be an important

channel through which financial markets affect the macroeconomy.

153
 r
The expected return on assets: This determines the coefficient, 1+r
, that multiplies both

assets and the expected present value of labour income. In this model, an increase in

this expected return raises this coefficient, and thus boosts consumption.

A Concrete Example: Constant Expected Growth in Labour Income

This RE-PIH model can be made more concrete by making specific assumptions about ex-

pectations concerning future growth in labour income. Suppose, for instance, that households

expect labour income to grow at a constant rate g in the future:

Et Yt+k = (1 + g)k Yt (6.27)

This implies
k
r rYt X 1+g

Ct = At + (6.28)
1+r 1 + r k=0 1 + r
As long as g < r (and we will assume it is) then we can use the geometric sum formula to

simplify this expression


1+g k 1
X 
= (6.29)
k=0 1+r 1 1+g
1+r
1+r
= (6.30)
rg
This implies a consumption function of the form
r r
Ct = At + Yt (6.31)
1+r rg
Note that the higher is expected future growth in labour income g, the larger is the coefficient

on todays labour income and thus the higher is consumption.

The Lucas Critique

The fact that the coefficients of so-called reduced-form relationships, such as the consumption

154
function equation (6.31), depend on expectations about the future is an important theme in

modern macroeconomics. In particular, in a famous 1976 paper, rational expectations pioneer

Robert Lucas pointed out that the assumption of rational expectations implied that these

coefficients would change if expectations about the future changed.3 In our example, the

MPC from current income will change if expectations about future growth in labour income

change.

Lucass paper focused on potential problems in using econometrically-estimated reduced-

form regressions to assess the impact of policy changes. He pointed out that changes in

policy may change expectations about future values of important variables, and that these

changes in expectations may change the coefficients of reduced-form relationships. This type

of problem can limit the usefulness for policy analysis of reduced-form econometric models

based on historical data. This problem is now known as the Lucas critique of econometric

models.

To give a specific example, suppose the government is thinking of introducing a temporary

tax cut on labour income. As noted above, we can consider Yt to be after-tax labour income,

so it would be temporarily boosted by the tax cut. Now suppose the policy-maker wants an

estimate of the likely effect on consumption of the tax cut. They may get their economic

advisers to run a regression of consumption on assets and after-tax labour income. If, in the

past, consumers had generally expected income growth of g, then the econometric regressions
r
will report a coefficient of approximately rg
on labour income. So, the economic adviser

might conclude that for each extra dollar of labour income produced by the tax cut, there will
r
be an increase in consumption of rg
dollars.
3
Robert Lucas, Econometric Policy Evaluation: A Critique, Carnegie-Rochester Series on Public Policy,
Vol. 1, pages 19-46, 1976.

155
 However, if households have rational expectations and operate according to equation (6.26)

then the true effect of the tax cut could be a lot smaller. For instance, if the tax cut is only

expected to boost this periods income, and to disappear tomorrow, then each dollar of tax cut
r
will produce only 1+r
dollars of extra consumption. The difference between the true effect and

the economic advisors supposedly scientific regression-based forecast could be substantial.

For instance, plugging in some numbers, suppose r = 0.06 and g = 0.02. In this case, the
.06
economic advisor concludes that the effect of a dollar of tax cuts is an extra 1.5 (= .06.02 )
.06
dollars of consumption. In reality, the tax cut will produce only an extra 0.057 (= 1.06 ) dollars

of extra consumption. This is a big difference.

The Lucas critique has played an important role in the increased popularity of rational

expectations economics. Examples like this one show the benefit in using a formulation such

as equation (6.26) that explicitly takes expectations into account, instead of relying only on

reduced-form econometric regressions.

Implications for Fiscal Policy: Ricardian Equivalence

Like households, governments also have budget constraints. Here we consider the implications

of these constraints for consumption spending in the Rational Expectations Permanent Income

Hypothesis. First, let us re-formulate the household budget constraint to explicitly incorporate

taxes. Specifically, lets write the period-by-period constraint as

At+1 = (1 + r) (At + Yt Tt Ct ) (6.32)

where Tt is the total amount of taxes paid by households. Taking the same steps as before,

156
we can re-write the intertemporal budget constraint as

X Et Ct+k X Et (Yt+k Tt+k )
k = A t + (6.33)
k=0 (1 + r) k=0 (1 + r)k

Now lets think about the governments budget constraint. The stock of public debt, Dt

evolves over time according to

Dt+1 = (1 + r) (Dt + Gt Tt ) (6.34)

where Gt is government spending and Tt is tax revenue. Applying the repeated-substitution

method we can obtain an intertemporal version of the governments budget constraint.

X Et Tt+k X Et Gt+k
k = D t + k (6.35)
k=0 (1 + r) k=0 (1 + r)

This states that the present discounted value of tax revenue must equal the current level of

debt plus the present discounted value of government spending. In other words, in the long-

run, the government must raise enough tax revenue to pay off its current debts as well as its

current and future spending.

Consider the implications of this result for household decisions. If households have ra-

tional expectations, then they will understand that the governments intertemporal budget

constraint, equation (6.35), pins down the present value of tax revenue. In this case, we can

substitute the right-hand-side of (6.35) into the household budget constraint to replace the

present value of tax revenue. Doing this, the household budget constraint becomes

X Et Ct+k X Et (Yt+k Gt+k )
k = A t Dt + (6.36)
k=0 (1 + r) k=0 (1 + r)k

Consider now the implications of this result for the impact of a temporary cut in taxes. Before,

we had discussed how a temporary cut in taxes should have a small effect. This equation gives

us an even more extreme result unless governments plan to change the profile of government

157
spending, then a cut to taxes today has no impact at all on consumption spending. This is

because households anticipate that lower taxes today will just trigger higher taxes tomorrow.

This result that rational expectations implied that a deficit-financed cut in taxes should

have no impact on consumption was first presented by Robert Barro in a famous 1974 paper.4

It was later pointed out that some form of this result was alluded to in David Ricardos writings

in the nineteenth century. Economists love fancy names for things, so the result is now often

referred to as Ricardian equivalence.

Evidence on the RE-PIH

There have been lots of macroeconomic studies on how well the RE-PIH fits the data. One

problem worth noting is that there are some important measurement issues when attempting

to test the theory. In particular, the models assumption that consumption expenditures only

yield a positive utility flow in the period in which the money is spent clearly does not apply

to durable goods, such as cars or computers, which yield a steady flow of utility. For this

reason, most empirical research has focused only on spending on nondurables (e.g. food) and

services, with a separate literature focusing on spending on consumer durables.

There are various reasons why the RE-PIH may not hold. Firstly, it assumes that it

is always feasible for households to smooth consumption in the manner predicted by the

theory. For example, even if you anticipate earning lots of money in the future and would

like to have a high level of consumption now, you may not be able to find a bank to fund

a lavish lifestyle now based on your promises of future millions. These kinds of liquidity

constraints may make consumption spending more sensitive to their current incomes than
4
Robert Barro (1974). Are Government Bonds Net Wealth? Journal of Political Economy, Volume
82(6).

158
the RE-PIH predicts. Secondly, people may not have rational expectations and may not plan

their spending decisions in the calculating optimising fashion assumed by the theory.

Following Halls 1978 paper, the 1980s saw a large amount of research on whether the RE-

PIH fitted the data. The most common conclusion was that consumption was excessively

sensitive to disposable income. In particular, changes in consumption appear to be more

forecastable than they should be if Halls random walk idea was correct. Campbell and Mankiw

(1990) is a well-known paper that provides a pretty good summary of these conclusions.5

They present a model in which a fraction of the households behave according to the RE-

PIH while the rest simply consume all of their current income. They estimate the fraction

of non-PIH consumers to be about a half. A common interpretation of this result is that

liquidity constraints have an important impact on aggregate consumption. (A byproduct of

this conclusion would be that financial sector reforms that boost access to credit could have

an important impact on consumption spending.)

Evidence on Ricardian Equivalence

There is also a large literature devoted to testing the Ricardian equivalence hypothesis. In

addition to the various reasons the RE-PIH itself may fail, there are various other reasons

why Ricardian equivalence may not hold. Some are technical points. People dont actually

live forever (as we had assumed in the model) and so they may not worry about future tax

increases that could occur after they have passed away; taxes take a more complicated form

than the simple lump-sum payments presented above; the interest rate in the governments

budget constraint may not be the same as the interest rate in the households constraint.
5
John Campbell and Gregory Mankiw (1990). Permanent Income, Current Income, and Consumption,
Journal of Business and Economic Statistics

159
(You can probably think of a few more.) More substantively, people may often be unable

to tell whether tax changes are temporary or permanent. Most of the macro studies on this

topic (in particular those that use Vector Autoregressions) tend to find the effects of fiscal

policy are quite different from the Ricardian equivalence predictions. Tax cuts and increases

in government spending tend to boost the economy.

Perhaps the most interesting research on this area has been the use of micro data to ex-

amine the effect of changes in taxes that are explicitly predictable and temporary. One recent

example is the paper by Parker, Souleles, Johnson and Robert McClelland which examines

the effect of tax rebates provided to U.S. taxpayers in 2008.6 This programme saw the U.S.

government send once-off payments to consumers in an attempt to stimulate the economy.

Since these payments were being financed by expanding the government deficit, Ricardian

equivalence predicts that consumers should not have responded. Parker et al, however, found

the opposite using data from the Consumer Expenditure Survey. A quick summary:

We find that, on average, households spent about 12-30% (depending on the speci-

fication) of their stimulus payments on nondurable expenditures during the three-

month period in which the payments were received. Further, there was also a

substantial and significant increase in spending on durable goods, in particular

vehicles, bringing the average total spending response to about 50-90% of the

payments.

You might suspect that these results are driven largely by liquidity constraints but the

various microeconomic studies that have examined temporary fiscal policy changes have not
6
Consumer Spending and the Economic Stimulus Payments of 2008. American Economic Review, 103(6),
October 2013.

160
always been consistent with this idea. For example, research by Parker (1999) showed the even

relatively high-income consumers seemed to spend more in response to transitory changes in

their social security taxes (which stop at a certain point in the year when workers reach a max-

imum threshold point) while Souleles (1999) found excess sensitivity results for consumer

spending after people received tax rebate cheques.7 These results show excess sensitivity even

among groups of consumers that are unlikely to be liquidity constrained.

At the same time, this doesnt mean that households go on a splurge every time they

get a large payment. For example, Hsieh (2003) examines how people in Alaska responded

to large anticipated annual payments that they received from a state fund that depends

largely on oil revenues.8 Unlike the evidence on temporary tax cuts, Hsieh finds that Alaskan

households respond to these payments in line with the predictions of the Permanent Income

Hypothesis, smoothing out their consumption over the year. One possible explanation is that

these large and predictable payments are easier for people to understand and plan around and

the consequences of spending them too quickly more serious than smaller once-off federal tax

changes. There is clearly room for more research in this important area.

Precautionary Savings

I want to return to a subtle point that was skipped over earlier. If we keep the assumption

r = , then the consumption Euler equation is

U 0 (Ct ) = Et [U 0 (Ct+1 )] (6.37)

You might think that this equation is enough to deliver the property of constant expected
7
Jonathan Parker. The Reaction of Household Consumption to Predictable Changes in Social Security
Taxes, American Economic Review, Vol 89 No 4, September 1999. Nicholes Souleles. The Response of
Household Consumption to Income Tax Refunds, American Economic Review, Vol 89 No 4, September 1999
8
Chang-Tai Hsieh. Do Consumers React to Anticipated Income Changes? Evidence from the Alaska
Permanent Fund American Economic Review, March 2003.

161
consumption. We generally assume declining marginal utility, so function U 0 is monotonically

decreasing. In this case, surely the expectation of next periods marginal utility being the

same as this periods is the same as next periods expected consumption level being the same

as this periods.

The problem with this thinking is the Et here is a mathematical expectation, i.e. a

weighted average over a set of possible outcomes. And for most functions F generally

E(F (X)) 6= F (E(X)). In particular, for concave functionsfunctions like utility functions

which have negative second derivativesa famous result known as Jensens inequality states

that E(F (X)) < F (E(X)). This underlies the mathematical formulation of why people are

averse to risk: The average utility expected from an uncertain level of consumption is less

than from the sure thing associated with obtaining the average level of consumption. The

sign of the Jensens inequality result is reversed for concave functions, i.e. those with positive

second derivatives.

In this example, we are looking at the properties of Et [U 0 (Ct+1 )]. Whether or not marginal

utility is concave or convex depends on its second derivative, so it depends upon the third

derivative of the utility function U 000 . Most standard utility functions have positive third

derivatives implying convex marginal utility and thus Et [U 0 (Ct+1 )] > U 0 (Et Ct+1 ). What we

can see now is why the quadratic utility function was such a special case. Because this function

has U 000 = 0, its marginal utility is neither concave or convex and the Jensen relationship is an

equality. So, in this very particular case, the utility function displays certainty equivalence:

The uncertain outcome is treated the same way is if people were certain of achieving the

average value of consumption.

162
 Heres a specific example of when certainty equivalence doesnt hold.9 Suppose consumers

have a utility function of the form

1
U (Ct ) = exp (Ct ) (6.38)

where exp is the exponential function. This implies marginal utility of the form

U 0 (Ct ) = exp (Ct ) (6.39)

In this case, the Euler equation becomes

exp (Ct ) = Et (exp (Ct+1 )) (6.40)

Now suppose the uncertainty about Ct+1 is such that it is perceived to have a normal distri-

bution with mean Et (Ct+1 ) and variance 2 . A useful result from statistics is that if a variable

X is normally distributed has mean and variance 2 :

 
X N , 2 (6.41)

then it can be shown that

2
!
E (exp(X)) = exp + (6.42)
2
In our case, this result implies that
!
V ar (Ct+1 )
Et (exp (Ct+1 )) = exp Et (Ct+1 ) + (6.43)
2
2 2
!

= exp Et (Ct+1 ) + (6.44)
2

So, the Euler equation can be written as

2 2
!
exp (Ct ) = exp Et (Ct+1 ) + (6.45)
2
9
This particular example was first presented by Ricardo Caballero (1990), Consumption Puzzles and
Precautionary Savings Journal of Monetary Economics, Volume 25, pages 113-136.

163
Taking logs of both sides this becomes

2 2
Ct = Et (Ct+1 ) + (6.46)
2

which simplifies to
2
Et (Ct+1 ) = Ct + (6.47)
2
Even though expected marginal utility is flat, consumption tomorrow is expected to be higher

than consumption today. Thus, uncertainty induces an upward tilt to the consumption

profile. And this upward tilt has an affect on todays consumption: We cannot sustain higher

consumption tomorrow without having lower consumption today.

Indeed, it turns out that this result allows us to calculate exactly what the effect of

uncertainty is on consumption today. The Euler equation implies that

k 2
Et (Ct+k ) = Ct + (6.48)
2

Inserting this into the intertemporal budget constraint, we get

X Ct 2 X k X Et Yt+k
k + k = At + k (6.49)
k=0 (1 + r) 2 k=1 (1 + r) k=0 (1 + r)

It can be shown (mainly by repeatedly using the well-known geometric sum formula) that

X k 1+r
k = (6.50)
k=1 (1 + r) r2

So, the intertemporal budget constraint simplifies to

X Ct 1 + r 2 X Et Yt+k
k + = A t + k (6.51)
k=0 (1 + r) r2 2 k=0 (1 + r)

and taking the same steps as before, consumption today is

r r X Et Yt+k 2
Ct = At + (6.52)
1+r 1 + r k=0 (1 + r)k 2r

164
 2
This is exactly as before apart from an additional precautionary savings term
2r
. The

more uncertainty there is, the more lower the current level of consumption will be.

This particular result obviously relies on very specific assumptions about the form of the

utility function and the distribution of uncertain outcomes. However, since almost all utility

function feature positive third derivatives, the key property underlying the precautionary

savings resultmarginal utility averaged over the uncertain outcomes being higher than at

the average level of consumption-will generally hold. It is an important result because some

of the more important changes in the savings rate observed over time appear consistent with

this type of precautionary savings behaviour. So, for example, during the global financial

crisis, when there was so much uncertainty about how long the recession would last and what

impact it would have, it is very likely that this greater uncertainty depressed consumption.

Incorporating Time-Varying Asset Returns

One simplification that we have made up to now is that consumers expect a constant return

on assets. Here, we allow expected asset returns to vary. The first thing to note here is that

one can still obtain an intertemporal budget constraint via the repeated substitution method.

This now takes the form

X Et Ct+k X Et Yt+k
! = At + ! (6.53)
k+1 k+1
k=0 Q k=0 Q
(1 + rt+m ) (1 + rt+m )
m=1 m=1

h
Q
where xi means the product of x1 , x2 .... xh . The steps to derive this are identical to the
n=1

steps used to derive equation (71) in the previous set of notes (Rational Expectations and

Asset Prices).

The optimisation problem of the consumer does not change much. This problem now has

165
the Lagrangian


!k
1 Et Yt+k Et Ct+k

X X X
L (Ct, Ct+1 , ....) = U At
(Ct+k )+ + ! !
k=0 1+ k=0
k+1
Q k=0
k+1
Q

(1 + rt+m ) (1 + rt+m )

m=1 m=1

And instead of the simple Euler equation (6.17), we get

" ! #
0 1 + rt+1
U (Ct ) = Et U 0 (Ct+1 ) (6.54)
1+

or, letting

Rt = 1 + rt (6.55)

we can re-write this as

" ! #
0 Rt+1
U (Ct ) = Et U 0 (Ct+1 ) (6.56)
1+

Consumption and Rates of Return on Assets

Previously, we had used an equation like this to derive the behaviour of consumption, given an

assumption about the determination of asset returns. However, Euler equations have taken on

a double role in modern economics because they are also used to consider the determination

of asset returns, taking the path of consumption as given. The Euler equation also takes on

greater importance than it might seem based on our relatively simple calculations because,

once one extends the model to allow the consumer to allocate their wealth across multiple

asset types, it turns out that equation (6.56) must hold for all of these assets. This means

that for a set of different asset returns Ri,t , we must have

" ! #
0 Ri,t+1
U (Ct ) = Et U 0 (Ct+1 ) (6.57)
1+

for each of the assets.

166
 So, for example, consider a pure risk-free asset that pays a guaranteed rate of return next

period. The nearest example in the real-world is a short-term US treasury bill. Because there

is no uncertainty about this rate of return, call it Rf,t , these terms can be taken outside the

expectation term, and the Euler equation becomes

Rf,t+1
U 0 (Ct ) = Et [U 0 (Ct+1 )] (6.58)
1+

So, the risk-free interest rate should be determined as

(1 + ) U 0 (Ct )
Rf,t+1 = (6.59)
Et [U 0 (Ct+1 )]

To think about the relationship between risk-free rates and returns on other assets, it is

useful to use a well-known result from statistical theory, namely

E (XY ) = E(X)E(Y ) + Cov(X, Y ) (6.60)

The expectation of a product of two variables equals the product of the expectations plus the

covariance between the two variables. This allows one to re-write (6.57) as

1
U 0 (Ct ) = [Et (Ri,t+1 ) Et (U 0 (Ct+1 )) + Cov (Ri,t+1 , U 0 (Ct+1 ))] (6.61)
1+

This can be re-arranged to give

(1 + ) U 0 (Ct ) Cov (Ri,t+1 , U 0 (Ct+1 ))

= E t (Ri,t+1 ) + (6.62)
Et [U 0 (Ct+1 )] Et [U 0 (Ct+1 )]

Note now that, by equation (6.67), the left-hand-side of this equation equals the risk-free rate.

So, we have
Cov (Ri,t+1 , U 0 (Ct+1 ))
Et (Ri,t+1 ) = Rf,t+1 (6.63)
Et [U 0 (Ct+1 )]

This equation tells us that expected rate of return on risky assets equals the risk-free rate

minus a term that depends on the covariance of the risky return with the marginal utility of

167
consumption. This equation is known as the Consumption Capital Asset Pricing Model or

Consumption CAPM, and it plays an important role in modern finance. Most asset returns

depend on payments generated by the real economy and so they are procyclicalthey are

better in expansions than during recessions. However, the usual assumption of diminishing

marginal utility implies that U 0 depends negatively on consumption. This means that the

covariance term is negative for assets whose returns are positively correlated with consumption

and these assets will have a higher rate of return than the risk free rate. Indeed, the higher

the correlation of the asset return with consumption, the higher will be the expected return.

Underlying this behaviour is the fact that consumers would like to use assets to hedge

against consumption variations. Given two assets that have the same rate of return, a risk-

averse consumer would prefer to have one that was negatively correlated with consumption

than one that is positively correlated with consumption. For investors to be induced into hold-

ing both assets, the rate of return on the asset with a positive correlation with consumption

needs to be higher.

Puzzles: Equity Premium and Risk-Free Rate

In theory, the consumption CAPM should be able to explain to us why some assets, such

as stocks, tend to have such high returns while others, such as government bonds, have such

low returns. However, it turns out that it has great difficulty in doing so. In the US, the

average real return on stocks over the long run has been about six percent per year while the

average return on Treasury bonds has been about one percent per year. In theory, this could

be explained by the positive correlation between stock returns and consumption. In practice,

this is not so easy. Most studies use simple utility functions such as the Constant Relative

168
Risk Aversion (CRRA) preferences

1
U (Ct ) = Ct1 (6.64)
1

so marginal utility is

U 0 (Ct ) = Ct (6.65)

In this case, the consumption-CAPM equation becomes

 

Cov Ri,t+1 , Ct+1
Et (Ri,t+1 ) = Rf,t+1
h i (6.66)
Et Ct+1
For values of considered consistent with standard estimates of risk aversion, this covariance

on the right-hand side is not nearly big enough to justify the observed equity premium. It

requires values such as = 25, which turns out to imply people are incredibly risk averse: For

instance, it implies they are indifferent between a certain 17 percent decline in consumption

and 50-50 risk of either no decline or a 20 percent decline. One way to explain this finding

is as follows. In practice, consumption tends to be quite smooth over the business cycle (our

earlier model helps to explain why) so for standard values of , marginal utility doesnt change

that much over the cycle and one doesnt need to worry too much equities being procyclical.

However, if is very very high, then the gap between marginal utility in booms and recessions

is much bigger: Marginal utility is really high in recessions and consumers really want an asset

that pays off then. This leads to a high equity premium.

One route that doesnt seem to work is arguing that people really are that risk averse, i.e.

that = 25 somehow is a good value. The reason for this is that this value of would imply

a much higher risk-free rate than we actually see. Plugging the CRRA utility function into

the equation for the risk free rate

(1 + ) Ct
Rf,t+1 = h

i (6.67)
Et Ct+1

169
Neglecting uncertainty about consumption growth, this formula implies that on average, the

risk-free rate should be

Rf = (1 + ) (1 + gC ) (6.68)

where gC is the growth rate of consumption. Plugging in the average growth rate of con-

sumption, a value of = 25 would imply a far higher risk-free rate than we actually see on

government bonds.

There is now a very large literature dedicated to solving the equity premium and risk-free

rate puzzles, but as of yet there is no agreed best solution.10

10
The paper that started this whole literature is Rajnish Mehra and Edward Prescott, The Equity Premium:
A Puzzle Journal of Monetary Economics, 15, 145-161. For a review, see Narayana Kocherlakota, The
Equity Premium: Its Still a Puzzle Journal of Economic Literature, 34, 42-71.

170
Chapter 7
Exchange Rates, Interest Rates and
Expectations
Our next example the role of expectations in macroeconomics is an important one: The link

between interest rates and exchange rates and the behaviour of flexible exchange rates.

Why Exchange Rates Matter

Why do exchange rates matter? Consider the Euro-Pound exchange rate, so that 1 = X.

Now suppose X goes up, so the Euro is worth more relative to the pound. What will happen

to exports from Ireland to the UK and imports to Ireland from the UK?

1. Exports: For each pound in sterling revenues that an Irish firm earns, they now get less

revenue in euros unless they increase their UK price. Because most of their costs (in

particular wages) will be denominated in euros, this means that exporting will become

less profitable at prevailing prices. Irish firms may react to this by increasing the price

they charge in the UK: This will reduce demand for their product, so exports will

still decline. Alternatively, some firms that feel they cannot raises prices to restore

profitability may simply exit from exporting. Between these two mechanisms, an increase

in the value of the euro relative to the pound will reduce Irish exports to the UK.

171
 2. Imports: Because the value of the euro has increased, UK firms will get more sterling

revenues from exporting to Ireland at the same prices, so UK firms that hadnt previously

been exporting to Ireland may start to do so. Alternatively, UK firms already exporting

to Ireland may decide to lower their euro-denominated prices in Ireland and increase

their market share while still getting the same sterling revenue per unit. Either way,

imports to Ireland from the UK will increase.

So while an increase in the value of a countrys currency may sound like a good thing,

it tends to reduce exports, increase imports, and thus reduce the countrys real GDP. In

contrast, a depreciation of the currency boosts exports and has a positive effect on economic

growth. For these reasons, a depreciation of the currency is often welcome in a recession and

the absence of this tool when the exchange rate is fixed is often pointed to as a downside of

such regimes.

That said, exchange rate depreciation has its downsides also:

1. Inflation: Depreciation tends to make imports more expensive and so add to inflation.

This is one reason why central bankers tend to say they favour a strong currencythey

are indicating their preference for low inflation. For small open economies that import

a lot, the inflationary effects of depreciation are much bigger.

2. Temporary Boost: The boost to growth from a devaluation is often temporary. Over

time, the increase in import prices may feed through to higher wages and this gradually

erodes the competitive benefits from devaluation. The more open an economy is, the

stronger the subsequent erosion of the competitive improvement.

172
Free Movement of Capital: Uncovered Interest Parity

Consider the case where there is free mobility of capital: In other words, people can move

money from one country to another immediately and without incurring any fees or taxes.

Specifically, consider the case where money can flow easily between the US and the Euro area.

Suppose now that investors can buy either US or European risk-free one-period bonds.

European bonds have an interest rate of iE US

t and US bonds have an interest rate of it . Let et

represent the amount of dollars that can be obtained in exchange for one Euro: Currently et

is about 1.30.

Now lets think about about the return to a US investor who wants to invest $1 in a Euro-

denominated bond at time t and then convert the money back into dollars at time t + 1. They
1
do this as follows. First, they exchange their $1 for for et
and use this money to buy a European

bond worth. The bond pays an interest rate of iE

t and then next period the US investor

1+iE
  
Et et+1
exchanges their et
t
back into dollars, so they expect to end up with $ 1 + iE
t et
.

If we abstract from risk aversion (the exchange rate movement is presumably uncertain)

then the US investor will be indifferent between this buy-European-bond-and-swap-back-into-

dollars strategy and purchasing a US bond as long as

 E
t et+1
 
1+ iE
t = 1 + iUt S (7.1)
et

An alternative expression for this is

 Et et+1 et
 
1 + iE
t 1+ = 1 + iUt S (7.2)
et

which can be re-written as

Et et+1 et Et et+1 et
 
1+ iE
t + + iE
t = 1 + iUt S (7.3)
et et

173
Subtracting the 1 from each side, we get

Et et+1 et Et et+1 et
 
iE
t + + iE
t = iUt S (7.4)
et et
Et et+1 et
Since both iE
t and et
are going to be relatively small, the product of them will usually

be close to zero, so the condition for the investor to be indifferent between the two investment

strategies is
Et et+1 et
iE
t + iUt S (7.5)
et
This conditionwhich says that the foreign interest rate plus the expected percentage change

in the value of the foreign currency should equal the domestic interest rateis known as the

Uncovered Interest Parity condition.

Why should we expect this condition to hold? Why would be expect investors to be

indifferent between US and European bonds? Well, suppose it turned out that the European

bonds offered a better deal than the US bonds: The combination of interest rate and expected

exchange rate appreciation makes the rate of return on European bonds better than that on

US bonds. Well, if there is perfect capital mobility, then this would mean that there would be

a rush for investors to purchase European bonds rather than US bonds. European institutions

who borrow via selling these bonds (governments, highly rated corporations) would figure out

that they could borrow at a lower interest rate and still find investors willing to buy their

bonds as well as US bonds. By this logic, deviations from Uncovered Interest Parity (UIP)

should be temporary with borrowers adjusting the interest rates on their bonds to ensure that

investors are indifferent between various international investments.

Note that it states that if European interest rates are lower than US rates, then the Euro

must be expected to appreciate. This might seem counter-intuitive: Before reading this, you

174
might expect the country that has higher interest rates to be the one with an appreciating

currency. More on this below.

The Trilemma of International Finance

If the UIP relationship approximately holds, then this has important implications for the

links between a countrys choice of exchange rate regime and its choice of monetary policy.

Specifically, if UIP holds, then it is not possible to have all three of the following:

1. Free capital mobility (money moving freely in and out of the country).

2. A fixed exchange rate.

3. Independent monetary policy.

You can have any two of these three things, but not the third:

1. You can have free capital mobility and a fixed exchange rate (so that Et et+1 = et ) but

then your interest rates must equal those of the area you have fixed exchange rates

against (iUt S = iE
t ). For example, Ireland had a fixed exchange rate with the UK for

many years and interest rates here were the same as in the UK.

2. You can have free capital mobility and set your own monetary policy (iUt S 6= iE
t ) but

then your exchange rate cannot simply be fixed (so that Et et+1 6= et ). For example, in

the UK, the Bank of England sets short-term interest rates and the sterling exchange

rate fluctuates freely in financial markets.

3. You can set your own monetary policy and fix your exchange rate against another coun-

try, but then you must intervene in capital markets to prevent people talking advantage

175
 of investment arbitrage opportunities. For example, China has a fixed exchange rate

with the US dollar and also sets its own monetary policy but it does not allow free

movement of capital.

This idea that you can only have two from three of free capital mobility, a fixed exchange

rate and independent monetary policy is commonly known as the trilemma of international

finance.

Flexible Exchange Rates Under Capital Mobility

Lets think about how exchange rates should behave free under capital mobility. Recall our

example involving US and European bonds. The condition for the expected return on the two

investments to be the same was

 E
t et+1
 
1 + iE
t = 1 + iUt S (7.6)
et

You may have thought at this point that you had escaped from first-order stochastic difference

equations. Unfortunately not. Equation (7.6) isnt a linear first-order stochastic difference

equation of the type that we have studies up to know. However, if we take logs, it becomes

   
log 1 + iE
t + Et log et+1 log et = log 1 + iUt S (7.7)

This is a linear stochastic difference equation describing the properties of the log of the ex-

change rate. It can be re-arranged to be in our more familiar format as

   
log et = log 1 + iE
t log 1 + iUt S + Et log et+1 (7.8)

Going back to our description of first-order stochastic difference equations, this is an another

example of one these equations of the form yt = axt + bEt yt+1 , this time with yt = log et ,

176
    
xt = log 1 + iE
t log 1 + iUt S , a = b = 1. If we apply the repeated substitution technique

to this equation, we get

h    i
Et log 1 + iE US
X
log et = t+k log 1 + it+k (7.9)
k=0

It turns out, however, that this is not the only possible solution. To see this, note that for

any arbitrary number log e we could re-arrange equation (7.8) as

   
log et log e = log 1 + iE
t log 1 + iUt S + Et log et+1 log e (7.10)

In other words, because the coefficient on the expected future exchange rate equals one (be-

cause the b = 1) then the repeated substitution method works not just for et but for any et e

where e is any arbitrary number. So, the general solution is

h    i
Et log 1 + iE US
X
log et = log e + t+k log 1 + it+k (7.11)
k=0

where the theory does not predict what the value of e is. Because the natural log function

has the property that log (1 + x) x, we can simplify this to read

 
Et iE US
X
log et = log e + t+k it+k (7.12)
k=0

We can make a number of points about this equation.

UIP tells us something about the dynamics of the exchange rate but it does not make

definitive predictions about the level an exchange rate should be at, i.e. it does not pin

down a unique value of e. Other theories, such as Purchasing Power Parity (the idea

that exchange rates should adjust so each type of currency has equivalent purchasing

power) do make such predictions, though they dont work very well in practice.

177
 This unexplained e can be seen as a sort of long-run equilibrium exchange rate because

this is the rate that holds when the average interest rate on European bonds in the

future equals the average interest rate on US bonds.

The model predicts that deviations from the long-run exchange rate e are determined

by expectations that interest rates will differ across areas. In this example, the euro will

be higher than e if people expect European interest rates to be higher in the future than

US rates.

The model explains the slightly puzzling result we discussed earlier: That higher interest rates

in Europe imply the euro is expected to depreciate. Suppose in period t 1, Euro and US

interest rates were equal to each other and expected to stay that way. Equation (7.12) implies

that under these circumstances we would have log et1 = log e. Now suppose that, in period

t, Euro interest rates unexpectedly went above US interest rates just for one period. What

would happen? The Euro must end up back at e (because interest rates in the two areas are

going to equal each other after period t) and the Euro must also be expected to depreciate

(because of the higher current interest rate in Euro).

So, in response to the surprise temporary increase in European interest rates, the Euro

immediately jumps upwards and then depreciates back to e. This conforms with our intuition

that higher European interest rates should make the Euro more attractive.

UIP and Exchange Rate Volatility

During the period after the second world war up to the 1970s, most of the worlds economies

operated the so-called Bretton Woods system of quasi-fixed exchange rates. The 1970s saw

the widespread introduction of market-determined flexible exchange rates. Prior to the intro-

178
duction of this system, advocates of market-based flexible exchange rates had predicted that

rates would change very little over time.

The truth turned out to be the opposite: Exchange rates change by very large amounts

on a daily, weekly, monthly basis. See Figure 8.1 which shows the Euro-dollar exchange rate.

It also gone through big swings: Reaching lows of 0.8 in 2000 and highs of 1.6 in 2008. In

addition, there are often large day to day movements where the exchange rate will go up or

down by one or two percent.

The model just developedcombining the UIP with rational expectationshelps to ex-

plain why exchange rates are so volatile. Using equation (7.12) for the level of exchange rates,

we can derive the change in the exchange rate at time t as

   
Et iE US
Et1 iE US
X X
log et = t+k it+k t+k it+k (7.13)
k=0 k=1

We will simplify this a bit via a slightly dodgy bit of terminology, meaning that we will write

(Et Et1 ) xt+k to mean Et xt+k Et1 xt+k , i.e. this means the change between time t 1 and

time t in what people expect xtk to be. Given this, we can re-write the previous equation as

 
log et = iUt1
S
iE (Et Et1 ) iE US
X
t1 + t+k it+k (7.14)
k=0

This equation tells us a lot about how exchange rates should behave if investors have

rational expectations. Exchange rate changes reflect not only the expected change due to past

interest rate differentials expiring (the iUt1

S
iE
t1 term); they also reflect unexpected changes

in the projected path of future interest rate differentials. This means that all information

that affects expectations of future Euro-area and US interest rates feed directly into todays

exchange rate. Because interest rates are set by central banks in response to developments in

179
the macroeconomy, this means that exchange rates should react to all types of macroeconomic

news.

Figure 7.1: Daily Data on the Euro-Dollar Exchange Rate

180
Problems for the UIP-Rational Expectations Theory

The UIP theory helps to explain a number of important aspects of the behaviour of exchange

rates. However, there have been many examples of where the theory just outlined does not

seem to work well. Indeed, quite commonly, there have been examples where the theory

predicts for an extended period of time that a currency depreciation or appreciation should

be expected, when what actually happens is the opposite.

One potential explanation for this apparent failure that could still be consistent with

the model is that Et et+1 et is not the same as et+1 et : The mathematical expectation of

something and its actual outcome can sometimes differ from each other for quite a while. This

is sometimes called the Peso problem. Sometimes interest rates in developing economies (such

as Mexico, after which the term is named) are high because markets think there is a probability

(perhaps a small probability) that a large depreciation may be coming. Just because the

depreciation doesnt happen during a particular sample doesnt mean the expectation was

unreasonable or that it wont be correct at some point.

But evidence also seems to exist of more systematic errors for the UIP theory. Take one

example. For most of the last decade, Japanese interest rates were well below European levels

for most of this decade. The UIP-Rational Expectations approach would have predicted that

the Yen should have been appreciating against the Euro: In fact, the opposite happened

systematically from 2001 to 2008. See Figure ??. Many traders systematically exploited this,

borrowing at low interest rates in Yen, using the funds to buy Euro bonds that yielded higher

interest rates and then repaying their debts in depreciated Yenthe so-called Yen carry trade.

That said, as Figure 2 also shows, the carry trade unwound itself fairly spectacularly in 2008.

The leading explanations for the apparent failures of the UIP-RE theory involve introduc-

181
ing risk aversion (we have assumed investors are risk-neutral) and home-bias (the preference

for assets denominated in your home currency). For instance, in relation to the theorys fail-

ure to explain the Yen carry trade period, its worth noting that many Japanese investors

have a strong preference for Yen-denominated assets and dont want to take on the extra

currency-related risk of investing in dollar or euro-denominated assets.

These kinds of preferences may lead to short-term violations of the stronger predictions

of the UIP-RE theory. However, they will not allow countries to escape from the restrictions

of the Trilemma: A country that attempts to adopt a systematically different interest rate

policy than another country simply will not be able to have a fixed exchange rate with that

country unless it imposes capital controls.

182
Figure 7.2: Daily Data on the Euro-Yen Exchange Rate

183
Chapter 8
Sticky Prices and the Phillips Curve

One of the important themes of macroeconomics is that the behaviour of prices was crucial in

determining how the macro-economy responded to shocks. In the IS-LM model, we needed to

assume that prices were sticky in the short-run to obtain real effects for fiscal and monetary

policy but we assumed that prices were flexible in the long-run so that the economy returned

to its full employment level over time. In the IS-MP-PC theory, we formalised this idea a bit

more: This model featured prices that adjusted over time in response to the real economy

according to a Phillips curve.

In these notes, we will return to the topic of price setting and the relationship over time

between inflation and the business cycle. We will emphasise the role of price flexibility and

expectations.

Evidence on Price Stickiness

When we discussed IS-LM, we assumed that the price level did not keep moving to constantly

equate GDP with the level of output consistent with a natural rate of unemployment. Instead,

we assumed that prices only changed gradually over time in response to the real economy.

184
The idea that prices may be sticky has a long history in Keynesian macroeconomics but,

until recent decades, there was comparatively little evidence on the extent to which prices

changed over time.

This has changed since the statistical agencies have made available the micro-data that

underlie Consumer Price Indices. To construct CPIs, these agencies collect large numbers of

quotes of prices on individual items (e.g. they can tell you the price in April of a bottle of

Heinz ketchup at a particular store). These individual price quote data can be used to assess

how often individual prices are changed.

Studies of this type now exist for a large number of countries. For example, Bils and

Klenow 2004 paper provided evidence for consumer prices in the United States.1 An important

finding from this research is that the data show a very wide range of the frequency with which

different prices change. Figure 8.1 shows a histogram from Bils and Klenows paper showing

the distribution of the percentage probability that any price changes in a month. These vary

from prices that only have a one percent probability of changing each month (Coin-operated

apparel laundry and dry cleaning) to those that have an 80 percent probability of changing

each month (gasoline).

The table on the following page shows the median price duration is about four months. In

other words, half of the prices quoted in the CPI index change more than every four months,

while the other half change less than every four months. Research for the euro area has shown

that price durations are even longer in Europe. For example, Alvarez et al (2006) report a
2
median price duration for the euro area of 10.6 months.
1
Mark Bils and Peter Klenow (2004). Some Evidence on the Importance of Sticky Prices Journal of
Political Economy, Volume 112, Number 5.
2
Luis Alvarez, Emmanuel Dhyne, Marco Hoeberichts, Claudia Kwapil, Herve Le Bihan. Patrick Lunne-
mann, Fernando Martins, Roberto Sabbatini, Harald Stahl, Philip Vermeulen and Jouko Vilmunen (2006).
Sticky Prices in the Euro Area: A Summary of New Micro-Evidence Journal of the European Economic

185
Figure 8.1: The Distribution of Monthly Percent Probability of
Price Changes

Association, Volume 4(2-3), pages 575-584.

186
Bils and Klenow Evidence on Price Durations

187
New Classical and New Keynesian Macroeconomics

After Milton Friedmans critique of the Phillips curve, macroeconomists began to pay more

attention to the question of how expectations were formed. In particular, a number of papers

by Robert Lucas and Thomas Sargent introduced rational expectations into macroeconomic

modelling. These early papers tended to assume that prices were perfectly flexible, which

limited the ability of fiscal and monetary policy to influence output. This school of thought

became labelled New Classical economics.

In a number of famous New Classical papers, Robert Lucas argued that monetary policy

could still have short-run effects even if prices were flexible and people had rational expec-

tations. Lucass model relied on the idea that firms had a difficulty in the short-run distin-

guishing between movements in their prices and movements in the overall price levels. For

this reason, an increase in the money supply that provoked an increase in prices could, in the

short-run, provoke higher output because firms may believe this is increasing their relative

price and making production more profitable. Lucas emphasised, however, that once people

had rational expectations, the impact of policy on output could only be short-lived. In par-

ticular, he stressed that only unpredictable fiscal and monetary policies would have an impact

because people with rational expectations would anticipate the impact of predictable policy

on the price level.

Once we allow prices to be sticky, however, these points no longer hold. Because some prices

will not change even after the government changes fiscal or monetary policy, these policies

will have the traditional short-run impacts described in the IS-LM model even if people have

rational expectations. There are lots of different ways of formulating the idea that prices

may be sticky. Some of the best known formulations were those introduced in papers in

188
the late seventies by John Taylor and Stanley Fischer.3 These papers assumed that only a

certain fraction of firms set prices each period but those who did change their prices would set

them in an optimal manner using rational expectations. This work, which combined rational

expectations with sticky prices, invented what is now known as New Keynesian economics.

Pricing a la Calvo

The New Keynesian literature contains a number of different formulations of sticky prices.

For the rest of these notes, we will use a formulation of sticky prices known as Calvo pricing,

after the economist who first introduced it.4 Though not the most realistic formulation of

sticky prices, it turns out to provide analytically convenient expressions, and has implications

that are very similar to those of more realistic (but more complicated) formulations.

The form of price rigidity faced by the Calvo firm is as follows. Each period, only a random

fraction (1) of firms are able to reset their price; all other firms keep their prices unchanged.

When firms do get to reset their price, they must take into account that the price may be

fixed for many periods. We assume they do this by choosing a log-price, zt , that minimizes

the loss function

 2
()k Et zt pt+k
X
L(zt ) = (8.1)
k=0

where is between zero and one, and pt+k is the log of the optimal price that the firm would

set in period t + k if there were no price rigidity.

This expression probably looks a bit intimidating, so its worth discussing it a bit to explain

what it means. The loss function has a number of different elements:

3
Stanley Fischer (1977), Long-Term Contracts, Rational Expectations, and the Optimal Money Supply
Rule, Journal of Political Economy, 85, 191-205, and John Taylor (1979), Staggered Wage Setting in a
Macro Model, American Economic Review, Papers and Proceedings, Vol. 69, 108-113.
4
Guillermo Calvo, Staggered Contracts in a Utility-Maximizing Framework Journal of Monetary Eco-
nomics, September 1983.

189
  2
The term Et zt pt+k describes the expected loss in profits for the firm at time t + k

due to the fact that it will not be able to set a frictionless optimal price that period.

This quadratic function is intended just as an approximation to some more general profit

function. What is important here is to note that because the firm may be stuck with

the price zt for some time, it will lose profits relative to what it would have been able

to obtain if there were no price rigidities.

The summation
P
shows that the firm considers the implications of the price set today
k=0

for all possible future periods.

However, the fact that < 1 implies that the firm places less weight on future losses

than on todays losses. A dollar today is worth more than a dollar tomorrow because it

can be re-invested. By the same argument, a dollar lost today is more important than

a dollar lost tomorrow.

Future losses are actually discounted at rate ()k , not just k . This is because the firm

only considers the expected future losses from the price being fixed at zt . The chance

that the price will be fixed until t + k is k . So the period t + k loss is weighted by this

probability. There is no point in the firm worrying too much about losses that might

occur from having the wrong price far off in the future, when it is unlikely that the price

will remained fixed for that long.

190
The Optimal Reset Price

After all that, the actual solution for the optimal value of zt , (i.e. the price chosen by the firms

who get to reset) is quite simple. Each of the terms featuring the choice variable zt that is,
 2
each of the zt pt+k termsneed to be differentiated with respect to zt and then the sum

of these derivatives is set equal to zero. This means

 
L0 (zt ) = 2 ()k Et zt pt+k = 0
X
(8.2)
k=0

Separating out the zt terms from the pt+k terms, this implies
" #
k
()k Et pt+k
X X
() zt = (8.3)
k=0 k=0

Now, we can use our old pal the geometric sum formula to simplify the left side of this

equation. In other words, we use the fact that

1
()k =
X
(8.4)
k=0 1

to re-write the equation as

zt
()k Et pt+k
X
= (8.5)
1 k=0
implying a solution of the form

()k Et pt+k
X
zt = (1 ) (8.6)
k=0

Stated in English, all this equation says is that the optimal solution is for the firm to set its

price equal to a weighted average of the prices that it would have expected to set in the future

if there werent any price rigidities. Unable to change price each period, the firm chooses to

try to keep close on average to the right price.

And what is this frictionless optimal price, pt ? We will assume that the firms optimal

pricing strategy without frictions would involve setting prices as a fixed markup over marginal

191
cost:

pt = + mct (8.7)

Thus, the optimal reset price can be written as

()k Et ( + mct+k )
X
zt = (1 ) (8.8)
k=0

The New-Keynesian Phillips Curve

Now, we can show how to derive the behaviour of aggregate inflation in the Calvo economy.

The aggregate price level in this economy is just a weighted average of last periods aggregate

price level and the new reset price, where the weight is determined by :

pt = pt1 + (1 ) zt , (8.9)

This can be re-arranged to express the reset price as a function of the current and past

aggregate price levels

1
zt = (pt pt1 ) (8.10)
1
Now, lets examine equation (8.8) for the optimal reset price again. We have shown that the

first-order stochastic difference equation

yt = axt + bEt yt+1 (8.11)

can be solved to give

bk Et xt+k
X
yt = a (8.12)
k=0

Examining equation (8.8), we can see that zt must obey a first-order stochastic difference

equation with

yt = zt (8.13)

192
 xt = + mct (8.14)

a = 1 (8.15)

b = (8.16)

In other words, we can write the reset price as

zt = Et zt+1 + (1 ) ( + mct ) (8.17)

Substituting in the expression for zt in equation (8.10) we get

1
(pt pt1 ) = (Et pt+1 pt ) + (1 ) ( + mct ) (8.18)
1 1

After a bunch of re-arrangements, this equation can be shown to imply

(1 ) (1 )
t = Et t+1 + ( + mct pt ) (8.19)

where t = pt pt1 is the inflation rate.

This equation is known as the New-Keynesian Phillips Curve. It states that inflation is a

function of two factors:

Next periods expected inflation rate, Et t+1 .

The gap between the frictionless optimal price level + mct and the current price level

pt . Another way to state this is that inflation depends positively on real marginal cost,

mct pt .

Why is real marginal cost a driving variable for inflation? Firms in the Calvo model would

like to keep their price as a fixed markup over marginal cost. If the ratio of marginal cost to

193
price is getting high (i.e. if mct pt is high) then this will spark inflationary pressures because

those firms that are re-setting prices will, on average, be raising them.

Real Marginal Cost and Output

For simplicity, we will denote the deviation of real marginal cost from its frictionless level of

as

mcrt = + mct pt (8.20)

so we can write the NKPC as

(1 ) (1 ) r
t = Et t+1 + mct (8.21)

One problem with attempting to implement this model empirically, is that we dont actually

observe data on real marginal cost. National accounts data contain information on the factors

that affect average costs such as wages, but do not tell us about the cost of producing an

additional unit of output. That said, it seems very likely that marginal costs are procyclical,

and more so than prices. When production levels are high relative to potential output, there

is more competition for the available factors of production, and this leads to increases in

real costs, i.e. increases in the costs of the factors over and above increases in prices. Some

examples of the procyclicality of real marginal costs are fairly obvious. For example, the

existence of overtime wage premia generally means a substantial jump in the marginal cost of

labour once output levels are high enough to require more than the standard workweek.

For these reasons, many researchers implement the NKPC using a measure of the output

gap (the deviation of output from its potential level) as a proxy for real marginal cost. In

other words, they assume a relationship such as

mcrt = yt (8.22)

194
where yt is the output gap. This implies a New-Keynesian Phillips curve of the form

t = Et t+1 + yt (8.23)

where
(1 ) (1 )
= (8.24)

And this approach can be implemented empirically using various measures for estimating

potential output.5

The Asset-Price-Like Behaviour of NKPC Inflation

The New-Keynesian approach assumes that firms have rational expectations. Thus, we can

apply the repeated substitution method to equation (8.23) to arrive at

k Et yt+k
X
t = (8.25)
k=0

Inflation today depends on the whole sequence of expected future output gaps. Thus, the

NKPC sees inflation as behaving according to the classic asset-price logic that we saw with

the dividend-discount stock price model.

The NKPC and the Lucas Critique

The vast majority of macroeconomists now accept Friedmans critique of the original Phillips

curve. Thus, it is widely accepted that inflation expectations will move upwards over time if

output remains above its potential level, and that there is little or scope for policy-makers to

choose a tradeoff between inflation and output. However, as we discussed in earlier lecture
5
Roberts (1995) shows that a number of other models of sticky prices also imply a formulation for inflation
similar to the New Keynesian Phillips curve.

195
notes, there is empirical evidence for a relationship of the form

t = t1 + ut (8.26)

So there is a relationship between the change in inflation and the level of unemployment. In

this formulation, the lagged inflation term reflects how last periods level of inflation changes

peoples expectations and so feeds into todays inflation. This so-called accelerationist Phillips

curve fits the data quite well (or, more precisely, empirical approaches based on a weighted

average of past inflation rates, not just last periods, fit the data well) and comes with its own

well-known terminology. Specifically, economists often speak of the so-called NAIRUthe

non-accelerating inflation rate of unemployment. This is the inflation rate consistent with

constant inflation and it is defined implicitly by

u = 0 u = (8.27)

Empirical estimates of the NAIRU are often invoked in real-world policy discussions, with the

policy recommendations made on the basis of whether unemployment is above or below this

NAIRU level.6

The NKPC model provides a different view of this empirical relationship. While advocates

of the NKPC will concede that the accelerationist model, equation (8.26), fits the data reason-

ably well, they view this as a so-called reduced-form relationship, not a structural relationship.

In other words, if the true model is

t = Et t+1 + yt (8.28)

then equation (8.26) might have a good statistical fit because t1 is likely to be correlated

with Et t+1 . However, they would warn policy-makers not to rely on this relationship, because
6
Note though the NAIRU terminology is actually a misnomer. If unemployment is below u , then inflation
will be increasing, but not accelerating. The price level is what will be accelerating. Perhaps the NAIRU
should be changed to the NAPLRU, but this isnt so catchy so the slipped derivative is probably here to
stay.

196
changes in policy may produce a break the correlation between Et t+1 and t1 and at this

point the statistical accelerationst Phillips curve will break down.

The NKPC and Disinflation

The NKPC also has important implications for how a government can approach reducing

inflation. Consider again the accelerationist Phillips curve, equation (8.26). The fact that

inflation depends on its own lagged values in this formulation means then it would be very

difficult to reduce inflation quickly without a significant increase in unemployment. So, this

Phillips curve suggests that gradualist policies are the best way to reduce inflation.

But the implications of the NKPC are completely different. There may be a statistical

relationship between current and lagged inflation but the NKPC says that there is no structural

relationship at all. Thus, there is no need for gradualist policies to reduce inflation. According

to the NKPC, low inflation can be achieved immediately by the central bank announcing (and

the public believing) that it is committing itself to eliminating positive output gaps in the

future: This can be seen from equation (8.25).

Whether the empirical evidence fits with the NKPCs predictions is open for debate. For

example, there has been plenty of evidence that reductions in inflation do tend to be costly

in terms of lost output and high unemployment. Some, however, have put this down to the

failure of governments and central banks to credibly convince the public of their commitment

to lower inflation rates.

197
Chapter 9
Investment With Adjustment Costs
In the previous chapters, we have seen a number examples of forward-looking first-order

stochastic difference equations of the form

yt = axt + bEt yt+1 (9.1)

The solution that we have derived has been of the form

bk Et xt+k
X
yt = a (9.2)
k=0

so that yt is a completely forward-looking variable. Note that this means that yt does not

depend at all on its own past values. We will now turn to an example which does not

correspond to this case.

Specifically, we will look at a theory of the determination of the capital stock (and thus

investment). Empirical studies show that the capital stock does not change very much from

period to period. Economists usually rationalise this by assuming that there are some form of

adjustment costs that prevent firms from changing their capital stock too quickly. In this

chapter, we will consider a model of investment with adjustment costs, show that it implies

a second-order stochastic difference equation, and examine the methods used to solve these

types of equations.

198
The Firms Problem

Consider now the following model of firm investment. We will assume that, each period,

there is a level of the log of the capital stock, kt , that the firm would choose if there were

no adjustment costs. We will call this the frictionless optimal capital stock. With adjustment

costs the firm has to choose a planned sequence of capital stocks Et {kt, kt+1, kt+2 , .....} minimise

the following loss function

"  #
2
m 2
X
L (kt , kt+1 , tt+2 , ...) = Et kt+m kt+m + (kt+m kt+m1 ) (9.3)
m=0

This might look a bit intimidating but its not too complicated:

 2

Firstly, for each period, t + m, there is a term kt+m kt+m that describes the loss in

profits suffered by the firm from not having its capital stock equated with the frictionless

optimal level.

Secondly, there is a term (kt+m kt+m1 ) which describes the concept of adjustment

costs formally: Ceteris paribus changes in the capital stock have a negative effect on

firm profits.

The reason we are assuming that kt is actually the log of the stock, as opposed to

the stock itself, is that this way these losses can be viewed in percentage terms: It is

the percentage gap between capital and its frictionless optimal that matters and also

the percentage change in the stock. This makes more sense than levels of these gaps

mattering because economic growth will make levels of these variables grow over time.

Finally, the parameter is a discount rate less than one, which tells us that firms care

more about profits today than profits tomorrow.

199
 This loss function can be re-written as
 2 
L (kt , kt+1 , tt+2 , ...) = (kt kt )2 + (kt kt1 )2 + Et
kt+1 kt+1 + (kt+1 kt )2
 2 
+2 Et
kt+2 kt+2 + (kt+2 kt+1 )2 + .... (9.4)

An optimal plan is arrived at by differentiating this with respect to each of the capital stock

terms kt+m and setting these derivatives equal to zero. Consider first differentiating with

respect to kt . This gives

2 (kt kt ) + 2 (kt kt1 ) 2Et (kt+1 kt ) = 0 (9.5)

Again, try differentiating with respect to kt+1 . This gives

h   i

Et 2 kt+1 kt+1 + 2 (kt+1 kt ) 22 (kt+2 kt+1 ) = 0 (9.6)

This is the exact same as the previous first-order condition, only shifted forward one period.

In fact one can show that all of the FOCs describing the optimal dynamics of the capital are

consistent with the same second-order stochastic difference equation

Et [(kt kt ) + (kt kt1 ) (kt+1 kt )] = 0 (9.7)

Drawing terms together, this gives

Et kt+1 + (1 + + ) kt kt1 = kt (9.8)

which can be re-written as

1 1 1 1
 
Et kt+1 1 + + kt + kt1 = kt (9.9)

Because the maximum difference between time subscripts is two, this is a second-order stochas-

tic difference equation. There are two different methods that are commonly used to solve

200
equations of this form. I will discuss the so-called factorization method. For completeness, I

have also attached the derivation of the solution using the other method known as the method

of undetermined coefficients, but you can ignore this if you wish.

Lag Operators

The factorization method makes use what are known as lag and forward operators. These

are commonly used in calculations relating to time series, and they work as follows. The lag

operator turns a variable dated time t into a variable dated time t 1:

Lyt = yt1 (9.10)

Lag operators can be multiplied and added just like normal variables. So, for instance, one

can write

Lk yt = ytk (9.11)

The forward operator has the reverse effect of the lag operator

F k yt = yt+k (9.12)

Lag and forward operators also obey a form of the geometric sum formula. Recall that for

1 < < 1, we have

1
m =
X
(9.13)
m=0 1
Recall also that if 1 < < 1 and

yt = Et yt+1 + xt (9.14)

then the solution is

m Et xt+k
X
yt = (9.15)
m=0

201
Equation (9.14) can be re-written as
" #
1
yt = Et xt (9.16)
1 F

So equation (9.15) means that

1
mF m
X
= (9.17)
1 F m=0

The same applies for lag operators

1
m Lm
X
= (9.18)
1 L m=0

To verify that this is the case, note that if

yt = yt1 + xt (9.19)

then one can apply repeated substitution to re-write this as

yt = xt + xt1 + 2 xt2 + 3 xt3 + ..... (9.20)

Armed with this knowledge of lag and forward operators we can solve the second-order stochas-

tic difference equation using the factorization method.

Solution via Factorization

This method first re-writes equation (9.9) in terms of lag and forward operators. Written this

way it is
1 1 1 1
    
Et F 1+ + + L kt = kt (9.21)

Next, the method re-expresses the left-hand-side in terms of a quadratic equation in F mul-

tiplied by L:
1 1 1 1
    
Et F2 1 + + F+ Lkt = kt (9.22)

202
Now, you may recall that polynominals of the form

g(x) = x2 + bx + c (9.23)

can be re-written in terms of their roots as

g(x) = (x 1 ) (x 2 ) (9.24)

where

1 + 2 = b (9.25)

1 2 = c (9.26)

In this case, one can show that the polynomial

1 1 1
 
x2 1 + + x+ (9.27)

1
has two roots such that one root () is between zero and one while the other equals
. This

means that the optimality condition for the capital stock can be re-expressed as

1 1
   
Et (F ) F Lkt = kt (9.28)

 
1
Dividing across by F
, this becomes
" #
1 1
Et [(F ) Lkt ] = Et 1 kt (9.29)
F

Now we can use the properties of lag operators just derived to show that

1
()k F k
X
1 = = (9.30)
F
1 F k=0

So, the capital stock process has a solution of the form

" #

()n kt+n

X
kt = kt1 + Et (9.31)
n=0

203
Note now how adding adjustment costs changes the solution for a rational expectations model.

This produces a second-order difference equation, and the solution is no longer completely

forward-looking. Instead, the capital stock has a forward-looking component, which is a

geometric discounted sum, but it also has a backward-looking component, whereby it depends

on its own lagged value.

An Example: Investment, Output, and the Cost of Capital

The model can be fleshed out by stating what are the determinants of the frictionless optimal

capital stock. For instance, if the production function was of the Cobb-Douglas form, then

the optimal capital stock would take the form

Yt
Kt = (9.32)
Ct

where Yt is output and Ct is the cost of capital. Using lower-case letters to denote logs, this

can be written as

kt = yt ct (9.33)

So, the capital stock is determined by

" #

()n (yt+n ct+n )
X
kt = kt1 + Et (9.34)
n=0

Now assume that output and the cost of capital both follow AR(1) processes

yt = y yt1 + yt (9.35)

ct = c ct1 + ct (9.36)

The infinite sum component of the solution can now be written as

" #
n n
X X
Et () yt+n = (y ) yt
n=0 n=0

204
 1
= yt (9.37)
1 y

while

1
()n ct+n =
X
Et ct (9.38)
n=0 1 c
So, the capital stock process is

1 1
kt = kt1 + yt ct (9.39)
1 y 1 c

This gives us a reduced-form relationship between the capital stock, the lagged capital

stock, output and the cost of capital.

Note that the magnitudes of the coefficients on output and the cost of capital depend

positively on the persistence of these variables. If y is close to one, then the coefficient on

output will be high, with the same applying for c and the cost of capital. One example of

an application of this type of reasoning is Tevlin and Whelan (2003).1 This paper presents

regressions of equipment investment on output and the cost of capital. It reports much

larger coefficients on the cost of capital for investment in computers than for non-computing

equipment, and uses a model of this sort to provide an explanation. The cost of capital for

computing equipment is largely determined by very persistent shocks that tend to produce

ever-decreasing computer prices. In contrast, for non-computing equipment, the cost of capital

depends on a set of less persistent variables such as interest rates and tax incentives. This

suggests that the cost of capital should have a smaller coefficient in a regression for the non-

computing capital stock.

1
Stacey Tevlin and Karl Whelan. Explaining the Investment Boom of the 1990s, Journal of Money,
Credit, and Banking, Volume 35, pages 1-22, 2003.

205
Appendix: The Undetermined Coefficients Method

The other method used to solve these models starts by assuming that one knows the form of

the solution. So, one guesses that the solution is of the form
" #

n2 kt+n
X
kt = 1 kt1 + Et
n=0

From there, one goes on to figure out a unique set of values for 1 , 2 and that are consistent

with this equation, and with the optimality conditions for the capital stock. In this case
" #

n2 kt+n+1
X
Et kt+1 = 1 kt + Et
n=0

So, we have
" " ##

n2 kt+n+1 + (1 + + ) kt kt1 = kt
X
1 kt + Et
n=0
" #
kt
n2 kt+n+1
X
(1 + + 1 ) kt = kt1 + + Et
n=0

This can be re-written as

kt
" #

n2 kt+n+1
X
kt = kt1 + + Et
(1 + + 1 ) (1 + + 1 ) (1 + + 1 ) n=0

So, one can begin to make inferences about the coefficients:

1 =
(1 + + 1 )
1 1
= =
(1 + + 1 )
2 = = 1

The solution is

" #

()n kt+n

X
kt = kt1 + Et (9.40)
n=0

where solves

(1 + + ) = (9.41)

206
This can be re-written as
1 1 1
 
2
1+ + + =0 (9.42)

so the solution is the same as that derived from the factorization method above.

Personally, I am less fond of this method because it involves guessing the form of the

solution, which is a bit of a cheat, because it is still quite algebra-intensive, and because it

becomes impractical to apply once one moves to higher-order difference equations. In contrast,

the factorization method can be used to characterize the solutions of difference equations of

any order.

207
 Part III
Long-Run Growth

208
Chapter 10
Growth Accounting
The chapters in this section will focus on what is known as growth theory. Unlike most

of macroeconomics, which concerns itself with what happens over the course of the business

cycle (why unemployment or inflation go up or down during expansions and recessions), this

branch of macroeconomics concerns itself with what happens over longer periods of time. In

particular, it looks at the question What determines the growth rate of the economy over

the long run and what can policy measures do to affect it? As we will also discuss, this is

related to the even more fundamental question of what makes some countries rich and others

poor.

In this set of notes, we will cover what is known as growth accounting a technique for

explaining the factors that determine growth.

Production Functions

The usual starting point for growth accounting is the assumption that total real output in an

economy is produced using an aggregate production function technology that depends on the

total amount of labour and capital used in the economy. For illustration, assume that this

takes the form of a Cobb-Douglas production function:

Yt = At Kt Lt (10.1)

209
where Kt is capital input and Lt is labour input. Note that an increase in At results in

higher output without having to raise inputs. Macroeconomists usually call increases in At

technological progress and often refer to this as the technology term. As such, it is easy

to imagine increases in At to be associated with people inventing new technologies that allow

firms to be more productive. Ultimately, however, At is simply a measure of productive

efficiency and it may go up or down for other reasons, e.g. with the imposition or elimination

of government regulations. Because an increase in At increases the productiveness of the other

factors, it is also sometimes known as Total Factor Productivity (TFP), and this is the term

most commonly used in empirical papers that attempt to calculate this series.

Usually, we will be more interested in the determination of output per person in the econ-

omy, rather than total output. Output per person is often labelled productivity by economists

with increases in output per worker called productivity growth. Productivity is obtained by

dividing both sides of equation (10.1) by Lt to get

Yt
= At Kt L1
t (10.2)
Lt
which can be re-arranged to give

Yt Kt

= At L+1
t (10.3)
Lt Lt
This equation shows that there are three potential ways to increase productivity:

Technological progress: Improving the efficiency with which an economy uses its inputs,

i.e. increases in At .

Capital deepening (i.e. increases in capital per worker)

Increases in the number of workers: Note that this only adds to growth if + > 1, i.e.

if there are increasing returns to scale. Most growth theories assumes constant returns

210
 to scale: A doubling of inputs produces a doubling of outputs. If a doubling of inputs

manages to more than double outputs, you could argue that the efficiency of production

has improved and so perhaps this should be considered an increase in A rather than

something that stems from higher inputs. If, there are constant returns to scale, then

+ 1 = 0 and this term disappears and production function can be written as


Yt Kt

= At (10.4)
Lt Lt

The Determinants of Growth

Lets consider what determines growth with a constant returns to scale Cobb-Douglas pro-

duction function (so = 1 )

Yt = At Kt L1
t (10.5)

and lets assume that time is continuous. In other words, the time element t evolves smoothly

instead of just taking integer values like t = 1 and t = 2.

How do we characterise how this economy grows over time? Lets denote the growth rate

of Yt by GYt . This can be defined as

1 dYt
GYt = (10.6)
Yt dt

In other words, the growth rate at any point in time is the change in output (the derivative
dYt
of output with respect to time, dt
) divided by the level of output. We can characterise

the growth rate of Yt as a function of the growth rates of labour, capital and technology by

differentiating the right-hand-side of equation (11.35) with respect to time. Before we do this,

you should recall the product rule for differentiation, i.e. that

dAB dA dB
=B +A (10.7)
dx dx dx
211
For products of three variables (like we have in this case) this implies

dABC dA dB dC
= BC + AC + AB (10.8)
dx dx dx dx

In our case, we have

dYt dAt Kt L1
t dAt dK dL1
= = Kt Lt1 + At Lt1 t + At Kt t (10.9)
dt dt dt dt dt

We can use the chain rule to calculate the terms involving the impact of changes in capital

and labour inputs:

dKt dKt dKt dKt
= = Kt1 (10.10)
dt dKt dt dt
dL1
t dLt1 dLt dLt
= = (1 ) L
t (10.11)
dt dLt dt dt
Plugging these formulae into the right places in equation (10.9) we get

dYt dAt dKt dLt

= Kt L1
t + At Kt1 Lt1 + (1 ) At Kt L
t (10.12)
dt dt dt dt

The growth rate of output is calculated by dividing both sides of this by Yt which is the same

as dividing by At Kt L1
t .

Kt L1 At Kt1 Lt1 At Kt L
! ! !
1 dYt t dAt dKt t dLt
= 1
+ + (1 ) (10.13)
Yt dt At Kt Lt dt At Kt Lt1 dt
At Kt Lt1
dt

Cancelling the various terms that appear multiple times in the terms inside the brackets and

we get
1 dYt 1 dAt 1 dKt 1 dLt
= + + (1 ) (10.14)
Yt dt At dt Kt dt Lt dt
This can written in more intuitive form as

GYt = GA K L
t + Gt + (1 ) Gt (10.15)

The growth rate of output equals the growth rate of the technology term plus a weighted

average of capital growth and labour growth, where the weight is determined by the parameter

212
. This is the key equation in growth accounting studies. These studies provide estimates of

how much GDP growth over a certain period comes from growth in the number of workers,

how much comes from growth in the stock of capital and how much comes from improvements

in Total Factor Productivity.

One can also show that the growth rate of output per worker is the growth rate of output

minus the growth in the number of workers, so this is determined by

 
GYt GLt = GA K L
t + Gt Gt (10.16)

This is a re-statement in growth rate terms of our earlier decomposition of output growth

into technological progress and capital deepening when the production function has constant

returns to scale.

It is good to understand how equation (11.36) was derived but, more generally, it is useful

to know how to derive growth rates of Cobb-Douglas-style variables.

For example, remember the production function is Yt = At Kt L1

t . The reason an

increase of x percent in At translates into an increase of x percent in output is because

At multiplies the other terms.

In contrast, Kt is taken to the power of . An increase in Kt , say by replacing it with

(1 + x) Kt is equivalent to multiplying the existing level of output by (1 + x) . Because

is assumed to be less than one, this is a smaller increase than comes from increasing

At by a factor of (1 + x).

To understand why a 1% increase in both Kt and Lt leads to a 1% increase in output,

note that if we multiplied both the inputs in Kt L1

t by (1 + x), we would get

At ((1 + x) Kt ) ((1 + x) Lt )1 = (1 + x) (1 + x)1 At Kt L1

t = (1 + x) At Kt L1
t

213
How to Calculate the Sources of Growth: Solow (1957)

For most economies, we can calculate GDP, as well as the number of workers and also get some

estimate of the stock of capital (this last is a bit trickier and usually relies on assumptions

about how investment cumulates over time to add to the stock of capital.) We dont directly

observe the value of the Total Factor Productivity term, At . However, if we knew the value

of the parameter , we could figure out the growth rate of TFP from the following equation

based on re-arranging (11.36)

GA Y K L
t = Gt Gt (1 ) Gt (10.17)

But where would we get a value of from? In a famous 1957 paper, MIT economist Robert

Solow pointed out that we could arrive at an estimate of by looking at the shares of GDP

paid to workers and to capital.1

To see how this method works, consider the case of a perfectly competitive firm that is

seeking to maximise profits. Suppose the firm sells its product for a price Pt (which it has no

control over), pays wages to its workers of Wt and rents its capital for a rental rate of Rt (this

last assumptionthat the firm rents its capitalisnt important for the points that follow

but it makes the calculations simpler.) This firms profits are given by

t = Pt Yt Rt Kt Wt Lt (10.18)

= Pt At Kt Lt1 Rt Kt Wt Lt (10.19)

Now consider how the firm chooses how much capital and labour to use. It will maximise

profits by differentiating the profit function with respect to capital and labour and setting the

resulting derivatives equal to zero. This gives two conditions

t
= Pt At Kt1 L1
t Rt = 0 (10.20)
Kt
1
Technical Change and the Aggregate Production Function, Review of Economics and Statistics.

214
 t
= (1 ) Pt At Kt L
t Wt = 0 (10.21)
Lt

These can be simplified to read

t Pt Y t
= Rt = 0 (10.22)
Kt Kt
t Pt Y t
= (1 ) Wt = 0 (10.23)
Lt Lt

Solving these we get

Rt Kt
= (10.24)
Pt Y t
Wt Lt
1 = (10.25)
P t Yt

Take a close look these equations.

Pt Yt is total nominal GDP (the price level times real output)

Wt Lt is the total amount of income paid out as wages (the wage rate times number of

workers).

Rt Kt is the total amount of income paid to capital (the rental rate times the amount of

capital).

These equations tell us that we can calculate 1 as the fraction of income paid to workers

rather than to compensate capital. (In real-world economies, non-labour income mainly takes

the form of interest, dividends, and retained corporate earnings). National income accounts

come with various decompositions. One of them describes how different types of incomes

add up to GDP. In most countries, these statistics show that wage income accounts for most

of GDP, meaning < 0.5. A standard value that gets used in many studies, based on US

215
estimates, is = 13 . I would note, however, that some studies do this calculation assuming

firms are imperfectly competitive if this is the case (as it is in the real world) then the

shares of income earned by labour and capital depend on the degree of monopoly power. So

one needs to be cautious about growth accounting calculations as they rely on theoretical

assumptions that could potentially be misleading.

Solows 1957 paper concluded that capital deepening had not been that important for U.S.

growth for the period that he examined (1909-1949). In fact, he calculated that TFP growth

accounted for 87.5% of growth in output per worker over that period. The calculation became

very famous it was one his papers that was cited by the Nobel committee when awarding

Solow the prize for economics in 1987. TFP is sometimes called the Solow residual because

it is a backed out calculation that makes things add up: You calculate it as the part of

output growth not due to input growth in the same way as regression residuals in econometrics

are the part of the dependent variable not explained by the explanatory variables included in

the regression.

Example: The BLS Multifactor Productivity Figures

Most growth accounting calculations are done as part of academic studies. However, in some

countries the official statistical agencies produce growth accounting calculations. In the U.S.

the Bureau of Labor Statistics (BLS) produces them under the name multifactor productiv-

ity calculations, (i.e. they use the term MFP instead of the term TFP but conceptually they

are the same thing.) Many of the studies add some bells and whistles to the basic calcula-

tions just described. For example, the BLS try to account for improvements in the quality

of the labour force by accounting for improvements in the level of educational qualifications

216
and work experience of employees. In other words, they view the production function as being

of the form

Yt = At Kt (qt Lt )1 (10.26)

where qt is a measure of the quality of the labor input.

Figure 10.1 shows a summary of the BLSs calculations of the sources of growth in the

US from 1987 to 2010. They conclude that average growth of 2.2 percent in the U.S. private

nonfarm economy can be explained as follows: 0.9 percent comes from capital deepening, 0.3

percent comes from changes in labour composition and 0.9 percent comes from changes in

what they call multifactor productivity. Looking at different samples, however, we can see

large changes between different periods in the contribution of MFP.

From 1987-1995, productivity growth averaged only 1.5 percent and MFP growth was

weak, contributing only 0.5 percent per year to growth. During this period, there was

a lot of discussion about the slowdown in growth relative to previous eras, with much

of the focus on the poor performance of TFP growth. Paul Krugmans first popular

economics book was called The Age of Diminished Expectations because people seemed

to have accepted that the US economy was doomed to low productivity growth.

From 1995-2007, productivity growth averaged a very respectable 2.65 percent, with

MFP growth contributing 1.35 percent. During this period, there was a lot ofdiscussion

of the impact of new Internet-related technologies that improved efficiency. While the

peak of this enthusiasm was around the dot-com bubble of the 2000s when there was

lots of talk of a New Economy, post-tech-bubble productivity performance was also

pretty good.

217
 From 2007-2013, productivity growth has been weaker than in the previous decade,

averaging only 1.3 percent. MFP growth has been particularly weak, averaging only 0.6

percent over this period. New Economy optimism has receded, a topic that we will

return to later.

In addition to the poor performance of U.S. productivity growth, another factor that is

weighing on the potential for output growth is a slow growth rate of the labour force. After

years of increasing numbers of people available for work due to normal population growth,

immigration and increased female labour participation, the US labour force has flattened out

(see Figure 10.2). This is being driven by long-run demographic trends as the large baby

boom generation starts to retire. This trend is set to continue over the next few decades.

Figure 10.3 shows that the dependency ratio (the ratio of non-working to working people) is

projected to increase significantly as the populations grows older on average. See my blog

post Is the U.S, Set for an Era of Slow Growth?

218
Figure 10.1: Growth Accounting Calculations for the U.S.

219
Figure 10.2: The U.S. Labour Force

220
Figure 10.3: The Ratio of Non-Working to Working People in U.S.

221
Example: The Euro Area

Longer-term growth prospects in Europe appear to be worse than in the United States. I am

currently working on a paper with Kieran McQuinn that does a growth accounting analysis

for the euro area and constructs longer-term growth projections. The following discussion is

based on this work.

Table 1 shows that growth in output per worker in the countries that make up the euro

area has gradually declined over time. In particular, TFP growth has collapsed. From 2.7

percent per year over 1970-76, TFP growth has fallen to an average of 0.2 percent per year

over the period 2000-2012. Table 2 shows that weak performances for TFP growth can be

seen widely across different European countries.

Europe is also on the cusp of a significant demographic change that will reduce the potential

for GDP growth: See Figure 10.4. Population growth is slowing and total population is set to

peak in before the middle of this century. The population is also ageing significantly. Indeed

the total amount of people aged between 15 and 64 (i.e. the usual definition of work-age

population) has peaked and is set to decline substantially over the next half century.

While Europe has many short-term macroeconomics problems due to weak aggregate de-

mand and high levels of public and private debt, it is also the case that it faces severe challenges

in relation to long-term growth. Maintaining growth rates at close to those experienced histor-

ically will likely involve reforms to raise the size of the labour force (such as raising retirement

ages and immigration) and boost productivity.

222
Table 1: The Euro Areas Growth Performance

223
Table 2: Country-by-Country Growth Performance 2000-2013

224
Figure 10.4: Demographic Projections for the Euro Area from Eu-
rostat

225
Example: A Tale of Two Cities

Alwyn Youngs 1992 paper A Tale of Two Cities: Factor Accumulation and Technical Change

in Hong Kong and Singapore is an interesting example of a growth accounting study.2 He

compares the growth experiences of these two small Asian economies from the early 1970s

to 1990. Young explained his motivation for picking these two economies in terms of their

similarities and their differences:

In the prewar era, both economies were British colonies that served as entre-

pot trading ports, with little domestic manufacturing activity ... In the postwar

era, however, both economies developed large export-dependent domestic manu-

facturing sectors. Both economies have passed through a similar set of industries,

moving from textiles, to clothing, to plastics, to electronics, and then, in the 1980s,

gradually moving from manufacturing into banking and financial services ... The

postwar population of both was composed primarily of immigrant Chinese from

Southern China ... While the Hong Kong government has emphasized a policy of

laissez faire, the Singaporean government has, since the early 1960s, pursued the

accumulation of physical capital via forced national saving.

Both economies were successful: Hong Kong had total growth of 147% between the early

1970s and 1990 and Singapore had growth of 154%. But Young was interested in exploring the

extent to which TFP contributed to growth in these two economies. The results of his growth

accounting calculations are shown on the next page. He found that Singapores approach did

not produce any TFP growth while Hong Kongs more free market approach lead to strong

TFP growth with this element accounting for almost half of the growth in output per worker.
2
Available at www.nber.org/chapters/c10990.pdf

226
One can argue this was a better outcome because Hong Kong achieved the growth without

having to divert a huge part of national income towards investment rather than consumption.

As we will see in the next lecture, however, TFP-based growth has an advantage over growth

based on capital accumulation because it is more sustainable.

Table from Alwyn Youngs 1992 Paper

227
Chapter 11
The Solow Model
We have discussed how economic growth can come from either capital deepening (increased

amounts of capital per worker) or from improvements in total factor productivity (sometimes

termed technological progress). This suggests that economic growth can come about from

saving and investment (so that the economy accumulates more capital) or from improvements

in productive efficiency. In these notes, we consider a model that explains the role these two el-

ements play in generating sustained economic growth. The model is also due to Robert Solow,

whose work on growth accounting we discussed in the last lecture, and was first presented in

his 1956 paper A Contribution to the Theory of Economic Growth.

The Solow Models Assumptions

The Solow model assumes that output is produced using a production function in which output

depends upon capital and labour inputs as well as a technological efficiency parameter, A.

Yt = AF (Kt , Lt ) (11.1)

It is assumed that adding capital and labour raises output

Yt
> 0 (11.2)
Kt
Yt
> 0 (11.3)
Lt

228
However, the model also assumes there are diminishing marginal returns to capital accumula-

tion. In other words, adding extra amounts of capital gives progressively smaller and smaller

increases in output. This means the second derivative of output with respect to capital is

negative.
2 Yt
<0 (11.4)
Kt
See Figure 11.1 for an example of how output can depend on capital with diminishing returns.

Think about why diminishing marginal returns is probably sensible: If a firm acquires an extra

unit of capital, it will probably be able to increase its output. But if the firm keeps piling on

extra capital without raising the number of workers available to use this capital, the increases

in output will probably taper off. A firm with ten workers would probably like to have at

least ten computers. It might even be helpful to have a few more; perhaps a few laptops for

work from home or some spare computers in case others break down. But at some point, just

adding more computers doesnt help so much.

We will use a very stylized description of the other parts of this economy: This helps us to

focus in on the important role played by diminishing marginal returns to capital. We assume

a closed economy with no government sector or international trade. This means all output

takes the form of either consumption or investment

Yt = Ct + It (11.5)

and that savings equals investment

St = Yt Ct = It (11.6)

The economys stock of capital is assumed to change over time according to

dKt
= It Kt (11.7)
dt
229
In other words, the addition to the capital stock each period depends positively on investment

and negatively on depreciation, which is assumed to take place at rate .

The Solow model does not attempt to model the consumption-savings decision. Instead it

assumes that consumers save a constant fraction s of their income

St = sYt (11.8)

230
 Figure 11.1: Diminishing Marginal Returns to Capital

Output

Output

Capital

231
Capital Dynamics in the Solow Model

Because savings equals investment in the Solow model, equation (11.8) means that investment

is also a constant fraction of output

It = sYt (11.9)

which means we can re-state the equation for changes in the stock of capital

dKt
= sYt Kt (11.10)
dt

Whether the capital stock expands, contracts or stays the same depends on whether investment

is greater than, equal to or less than depreciation.

dKt
> 0 if Kt < sYt (11.11)
dt
dKt
= 0 if Kt = sYt (11.12)
dt
dKt
< 0 if Kt > sYt (11.13)
dt

In other words, if the ratio of capital to output is such that

Kt s
= (11.14)
Yt

the the stock of capital will stay constant. If the capital-ouput ratio is lower than this level,

then the capital stock will be increasing and if it is higher than this level, it will be decreasing.

Figure 11.2 provides a graphical illustration of this process. Depreciation is a simple

straight-line function of the stock of capital while output is a curved function of capital,

featuring diminishing marginal returns. When the level of capital is low sYt is greater than

K. As the capital stock increases, the additional investment due to the extra output tails

off but the additional depreciation does not, so at some point sYt equals K and the stock of

capital stops increasing. Figure labels the particular point at which the capital stock remains

unchanged as K . At this point, we have Kt

Yt
= s .

232
 In the same way, if we start out with a high stock of capital, then depreciation, K, will

tend to be greater than investment, sYt . This means the stock of capital will decline. When

it reaches K it will stop declining. This an example of what economists call convergent

dynamics. For any fixed set of the model parameters (s and ) and other inputs into the

production (At and Lt ) there will be a defined level of capital such that, no matter where the

capital stock starts, it will converge over time towards this level.

Figure provides an illustration of how the convergent dynamics determine the level of

output in the Solow model. It shows output, investment and depreciation as a function of the

capital stock. The gap between the green line (investment) and the orange line (output) shows

the level of consumption. The economy converges towards the level of output associated with

the capital stock K .

An Increase in the Savings Rate

Now consider what happens when the economy has settled down at an equilibrium unchanging

level of capital K1 and then there is an increase in the savings rate from s1 to s2 .

Figure 11.4 shows what happens to the dynamics of the capital stock. The line for in-

vestment shifts upwards: For each level of capital, the level of output associated with it

translates into more investment. So the investment curve shifts up from the green line to the

red line. Starting at the initial level of capital, K1 , investment now exceeds depreciation. This

means the capital stock starts to increase. This process continues until capital reaches its new

equilibrium level of K2 (where the red line for investment intersects with the black line for

depreciation.) Figure 11.4 illustrates how output increases after this increase in the savings

rate.

233
 Figure 11.2: Capital Dynamics in The Solow Model

Investment,
Depreciation Depreciation K

Investment sY

K* Capital, K

234
 Figure 11.3: Capital and Output in the Solow Model

Investment,
Depreciation, Depreciation K
Output

Output Y

Consumption
Investment sY

K*
Capital, K

235
 Figure 11.4: An Increase in the Saving Rate

Investment,
Depreciation Depreciation K

New Investment s2Y

Old Investment s1Y

K1 K2 Capital, K

236
 Figure 11.5: Effect on Output of Increased Saving

Investment, Depreciation K
Depreciation
Output
Output Y

New Investment s2Y

Old Investment s1Y

K1 K2 Capital, K

237
An Increase in the Depreciation Rate

Now consider what happens when the economy has settled down at an equilibrium unchanging

level of capital K1 and then there is an increase in the depreciation rate from 1 to 2 .

Figure 11.6 shows what happens in this case. The depreciation schedule shifts up from

the black line associated with the original depreciation rate, 1 , to the new red line associated

with the new depreciation rate, 2 . Starting at the initial level of capital, K1 , depreciation now

exceeds investment. This means the capital stock starts to decline. This process continues

until capital falls to its new equilibrium level of K2 (where the red line for depreciation

intersects with the green line for investment.) So the increase in the depreciation rate leads

to a decline in the capital stock and in the level of output.

An Increase in Technological Efficiency

Now consider what happens when technological efficiency At increases. Because investment

is given by

It = sYt = sAF (Kt , Lt ) (11.15)

a one-off increase in A thus has the same effect as a one-off increase in s. Capital and output

gradually rise to a new higher level. Figure 11.7 shows the increase in capital due to an

increase in technological efficiency.

238
 Figure 11.6: An Increase in Depreciation

Investment, New Depreciation 2K

Depreciation Old Depreciation 1K

Investment sY

K2 K1
Capital, K

239
 Figure 11.7: An Increase in Technological Efficiency

Investment, Depreciation K
Depreciation

New Technology A2F(K,L)

Old Technology A1F(K,L)

K1 K2 Capital, K

240
Solow and the Sources of Growth

In the last lecture, we described how capital deepening and technological progress were the

two sources of growth in output per worker. Specifically, we derived an equation in which

output growth was a function of growth in the capital stock, growth in the number of workers

and growth in technological efficiency.

Our previous discussion had pointed out that a one-off increase in technological efficiency,

At , had the same effects as a one-off increase in the savings rate, s. However, there are

important differences between these two types of improvements. The Solow model predicts

that economies can only achieve a temporary boost to economic growth due to a once-off

increase in the savings rate. If they want to sustain economic growth through this approach,

then they will need to keep raising the savings rate. However, there are likely to be limits in

any economy to the fraction of output that can be allocated towards saving and investment,

particularly if it is a capitalist economy in which savings decisions are made by private citizens.

Unlike the savings rate, which will tend to have an upward limit, there is no particular

reason to believe that technological efficiency At has to have an upper limit. Indeed, growth

accounting studies tend to show steady improvements over time in At in most countries. Going

back to Youngs paper on Hong Kong and Singapore discussed in the last lecture, you can

see now why it matters whether an economy has grown due to capital deepening or TFP

growth. The Solow model predicts that a policy of encouraging growth through more capital

accumulation will tend to tail off over time producing a once-off increase in output per worker.

In contrast, a policy that promotes the growth rate of TFP can lead to a sustained higher

growth rate of output per worker.

241
The Capital-Output Ratio with Steady Growth

Up to now, we have only considered once-off changes in output. Here, however, we consider

how the capital stock behaves when the economy grows at steady constant rate GY . Specif-

ically, we can show in this case that the ratio of capital to output will tend to converge to a

specific value. Recall from the last lecture that if we have something of the form

Zt = Ut Wt (11.16)

then we have the following relationship between the various growth rates

GZt = GUt + GW
t (11.17)

The capital output ratio Kt

Yt
can be written as Kt Yt1 . So the growth rate of the capital-output

ratio can be written as

K
GtY = GK Y
t Gt (11.18)

Adjusting equation 11.10, the growth rate of the capital stock can be written as

1 dKt Yt
GK
t = =s (11.19)
Kt dt Kt

so the growth rate of the capital-output ratio is

K Yt
GtY = s GY (11.20)
Kt

This gives a slightly different form of convergence dynamics from those we saw earlier. This

equation shows that the growth rate of the capital-output ratio depends negatively on the

level of this ratio. This means the capital-output ratio displays convergent dynamics. When

it is above a specific equilibrium value it tends to fall and when it is below this equilibrium

value it tends to increase. Thus, the ratio is constantly moving towards this equilibrium value.

242
We can express this formally as follows:

K Kt s
GtY > 0 if < (11.21)
Yt + GY
K Kt s
GtY = 0 if = (11.22)
Yt + GY
K Kt s
GtY < 0 if > (11.23)
Yt + GY

We can illustrate these dynamics using a slightly altered version of our earlier graph. Figure

11.8 amends the depreciation line to the amount of capital necessary not just to replace

depreciation but also to have a percentage increase in the capital stock that matches the

increase in output. The diagram shows that the economy will tend to move towards a capital
 
Kt s
stock such that sYt = + GY Kt meaning the capital-output ratio is Yt
= +GY
.

243
Figure 11.8: The Equilibrium Capital Stock in a Growing Economy

Investment,
Depreciation
Depreciation and Growth (+GY)K

Investment sY

K* Capital, K

244
Briefly, Back to Piketty

In Chapter 6, we discussed one of Thomas Pikettys explanations for why capital may tend

to growth faster than income (the r > g argument). Piketty has a different argument for why

capital may grow faster than income that relates to the result we have just derived.

In his book, Piketty describes a different assumption about savings in the economy from

the one we have just derived. Specifically, he works with a net savings rate, s, which is defined

as follows

It Kt = sYt (11.24)

In other words, defined like this, s is a savings rate that subtracts off the share of GDP taken

up by capital depreciation. In the same way, net national product is defined as GDP minus

depreciation. Given this definition, we can write the change in the capital stock as

Kt = sYt (11.25)

Repeating the calculations from above with this model, the growth rate of capital

1 dKt It Kt
GK
t = = (11.26)
Kt dt Kt

becomes
1 dKt Yt
GK
t = = s (11.27)
Kt dt Kt
So the growth rate of the capital-output ratio is

K Yt
GtY = s GY (11.28)
Kt

This gives a convergence dynamics in terms of this net savings rate.

K Kt s
GtY > 0 if < Y (11.29)
Yt G

245
 K Kt s
GtY = 0 if = Y (11.30)
Yt G
K Kt s
GtY < 0 if > Y (11.31)
Yt G

Kt s
So the capital output ratio converges to Yt
= GY
. Again showing his gift for grand terminol-

ogy, Piketty calls this result the second fundamental law of capitalism. His research has argued

that growth appears to be slowing around the world and thus, with GY in the denominator

heading towards zero, the capital-output ratio is likely to be ever-rising.

This prediction will, of course, also hold for the standard Solow model formulation in
s
which the capital-output ratio converges to +GY
. The most obvious difference, however, is

that Pikettys formulation suggests that when GY tends towards zero that we could see the

capital-output ratio head towards infinity because his steady-state ratio does not have the

in the denominator. However, this is somewhat misleading. In the standard formulation of

the model, you can show that the net savings rate along a steady growth path will be

It Kt
s = (11.32)
Yt Yt
s
= s (11.33)
GY +
sGY
= (11.34)
GY +

So when output growth goes to zero, the net savings rate also goes to zero. This means we
s
shouldnt just look at Pikettys formula of GY
for the steady-state capital-output ratio and

imagine the denominator (GY ) heading to zero while the numerator s is fixed. From this

discussion, you can take that slower output growth is likely to raise the ratio of capital to

income, but it is not likely to head towards infinity!

246
Why Growth Accounting Can Be Misleading

Of the cases just considered in which output and capital both increasean increase in the

savings rate and an increase in the level of TFPthe evidence points to increases in TFP

being more important as a generator of long-term growth. Rates of savings and investment

tend for most countries tend to stay within certain ranges while large increases in TFP over

time have been recorded for many countries. Its worth noting then that growth accounting

studies can perhaps be a bit misleading when considering the ultimate sources of growth.

Consider a country that has a constant share of GDP allocated to investment but is

experiencing steady growth in TFP. The Solow model predicts that this economy should

experience steady increases in output per worker and increases in the capital stock. A growth

accounting exercise may conclude that a certain percentage of growth stems from capital

accumulation but ultimately, in this case, all growth (including the growth in the capital

stock) actually stems from growth in TFP. The moral here is that pure accounting exercises

may miss the ultimate cause of growth.

Krugman on The Myth of Asias Miracle

I encourage you to read Paul Krugmans 1994 article The Myth of Asias Miracle.1 It

discusses a number of examples of cases where economies where growth was based on largely

on capital accumulation. In addition to the various Asian countries covered in Alwyn Youngs

research, Krugman (correctly) predicted a slowdown in growth in Japan, even though at the

time many US commentators were focused on the idea that Japan was going to overtake US

levels of GDP per capita.

1
www.foreignaffairs.com/articles/50550/paul-krugman/the-myth-of-asias-miracle

247
 Perhaps most interesting is his discussion of growth in the Soviet Union. Krugman notes

that the Soviet economy grew strongly after World War 2 and many in the West believed they

would become more prosperous than capitalist economies. The Soviet Unions achievement

in placing the first man in space provoked Kennedys acceleration in the space programme,

mainly to show the U.S. was not falling behind communist systems. However, some economists

that had examined the Soviet economy were less impressed. Heres an extended quote from

Krugmans article:

When economists began to study the growth of the Soviet economy, they did so

using the tools of growth accounting. Of course, Soviet data posed some prob-

lems. Not only was it hard to piece together usable estimates of output and input

(Raymond Powell, a Yale professor, wrote that the job in many ways resembled

an archaeological dig), but there were philosophical difficulties as well. In a so-

cialist economy one could hardly measure capital input using market returns, so

researchers were forced to impute returns based on those in market economies at

similar levels of development. Still, when the efforts began, researchers were pretty

sure about what: they would find. Just as capitalist growth had been based on

growth in both inputs and efficiency, with efficiency the main source of rising per

capita income, they expected to find that rapid Soviet growth reflected both rapid

input growth and rapid growth in efficiency.

But what they actually found was that Soviet growth was based on rapidgrowth

in inputsend of story. The rate of efficiency growth was not only unspectacular,

it was well below the rates achieved in Western economies. Indeed, by some

estimates, it was virtually nonexistent.

248
 The immense Soviet efforts to mobilize economic resources were hardly news. Stal-

inist planners had moved millions of workers from farms to cities, pushed millions

of women into the labor force and millions of men into longer hours, pursued mas-

sive programs of education, and above all plowed an ever-growing proportion of

the countrys industrial output back into the construction of new factories.

Still, the big surprise was that once one had taken the effects of these more or

less measurable inputs into account, there was nothing left to explain. The most

shocking thing about Soviet growth was its comprehensibility.

This comprehensibility implied two crucial conclusions. First, claims about the

superiority of planned over market economies turned out to be based on a mis-

apprehension. If the Soviet economy had a special strength, it was its ability to

mobilize resources, not its ability to use them efficiently. It was obvious to every-

one that the Soviet Union in 1960 was much less efficient than the United States.

The surprise was that it showed no signs of closing the gap.

Second, because input-driven growth is an inherently limited process, Soviet growth

was virtually certain to slow down. Long before the slowing of Soviet growth be-

came obvious, it was predicted on the basis of growth accounting.

The Soviet leadership did a good job for a long time of hiding from the world that their

economy had stopped growing but ultimately the economic failures of the centrally planning

model (combined with its many political and ethnic tensions) ended in a dramatic implosion

of the communist system in Russia and the rest of Eastern Europe.

249
A Formula for Steady Growth

All of the results so far apply for any production function with diminishing marginal returns

to capital. However, we can also derive some useful results by making specific assumptions

about the form of the production function. Specifically, we will consider the constant returns

to scale Cobb-Douglas production function

Yt = At Kt L1
t (11.35)

This means output growth is determined by

GYt = GA K L
t + Gt + (1 ) Gt (11.36)

Now consider the case in which the growth rate of labour input is fixed at n

GLt = n (11.37)

and the growth rate of total factor productivity is fixed at g.

GA
t = g (11.38)

The formula for output growth becomes

GYt = g + GK
t + (1 ) n (11.39)

This means all variations in the growth rate of output are due to variations in the growth

rate for capital. If output is growing at a constant rate, then capital must also be growing at

a constant rate. And we know that the capital-output ratio tends to move towards a specific

equilibrium value. So along a steady growth path, the growth rate of output equals the growth

rate of capital. Thus, the previous equation can be re-written

GYt = g + GYt + (1 ) n (11.40)

250
which can be simplified to
g
GYt = +n (11.41)
1
The growth rate of output per worker is

g
GYt n = (11.42)
1

So the economy tends to converge towards a steady growth path and the growth rate of output
g
per worker along this path is 1
. Without growth in technological efficiency, there can be

no steady growth in output per worker.

A Useful Formula for Output Per Worker

In this case of the Cobb-Douglas production function, output per worker can be written as

Yt Kt

= At (11.43)
Lt Lt

In other words, output per worker is a function of technology and of capital per worker. A

drawback of this representation is that we know that increases in At also increase capital

per worker, so this has the misleading implications about the role of capital accumulation

discussed above. It is useful, then, to derive an alternative characterisation of output per

worker, one that we will use again. First, well define the capital-output ratio as

Kt
xt = (11.44)
Yt

So, the production function can be expressed as

Yt = At (xt Yt ) L1
t (11.45)

Here, we are using the fact that

K t = xt Y t (11.46)

251
Dividing both sides of this expression by Yt , we get

Yt1 = At xt Lt1 (11.47)

1
Taking both sides of the equation to the power of 1
we arrive at
1
Yt = At1 xt1 Lt (11.48)

So, output per worker is

Yt 1
= At1 xt1 (11.49)
Lt
This equation states that all fluctuations in output per worker are due to either changes in

technological progress or changes in the capital-output ratio. When considering the relative

role of technological progress or policies to encourage accumulation, we will see that this

decomposition is more useful than equation (11.43) because the level of technology does not
Kt
affect xt in the long run while it does affect Lt
. So, this decomposition offers a cleaner picture

of the part of growth due to technology and the part that is not.

A Formal Model of Convergence Dynamics

Because At is assumed to grow at a constant rate each period, this means that all of the

interesting dynamics for output per worker in this model stem from the behaviour of the

capital-output ratio. We will now describe in more detail how this ratio behaves. Before

doing so, I want to introduce a new piece of terminology that we will use in the next few

lectures.

A useful mathematical shorthand that saves us from having to write down derivatives with

respect to time everywhere is to write

dYt
Yt = (11.50)
dt
252
What we are really interested in, though, is growth rates of series, so we need to scale this by
Yt
the level of output itself. Thus, Yt
, and this is our mathematical expression for the growth

rate of a series. For our Cobb-Douglas production function, we can use the result we derived

earlier to express the growth rate of output as

Yt At Kt Lt
= + + (1 ) (11.51)
Yt At Kt Lt

The Solow model assumes

At
=g (11.52)
At
Nt
=n (11.53)
Nt
So this can be re-written as
Yt Kt
=g+ + (1 )n (11.54)
Yt Kt
Similarly, because

xt = Kt Yt 1 (11.55)

its growth rate can be written as

xt Kt Yt
= (11.56)
xt Kt Yt

To get an expression for the growth rate of the capital stock, we re-write the capital

accumulation equation as

Kt = sYt Kt (11.57)

and divide across by Kt on both sides

Kt Yt
=s (11.58)
Kt Kt

This means we write, the growth rate of the capital stock as

Kt s
= (11.59)
Kt xt

253
 Now using equation (11.54) for output growth and equation (11.59) for capital growth, we

can derive a useful equation for the dynamics of the capital-output ratio:

xt Kt
= (1 ) g (1 )n (11.60)
xt Kt
s g
= (1 )( n ) (11.61)
xt 1

This dynamic equation has a very important property: The growth rate of xt depends nega-

tively on the value of xt . In particular, when xt is over a certain value, it will tend to decline,

and when it is under that value it will tend to increase. This provides a specific illustration

of the convergent dynamics of the capital-output ratio.

What is the long-run steady-state value of xt , which we will label x ? It is the value
x
consistent with x
= 0. This implies that

s g
n =0 (11.62)
x 1

This solves to give

s
x = g (11.63)
1
+n+
Given this equation, we can derive a more intuitive-looking expression to describe the conver-

gence properties of the capital-output ratio. The dynamics of xt are given by

xt s g
= (1 )( n ) (11.64)
xt xt 1
g
Multiplying and dividing the right-hand-side of this equation by ( 1 + n + ):
g
s/xt 1 n
!
xt g
= (1 )( + n + ) g (11.65)
xt 1 1
+n+

The last term inside the brackets can be simplified to give

!
xt g 1 s
= (1 )( + n + ) g 1 (11.66)
xt 1 xt 1
+n+

254
 g x
 
= (1 )( + n + ) 1 (11.67)
1 xt

g x xt

= (1 )( + n + ) (11.68)
1 xt

This equation states that each period the capital-output ratio closes a fraction equal to =
g
(1 )( 1 + n + ) of the gap between the current value of the ratio and its steady-state

value.

Illustrating Convergence Dynamics

Often, the best way to understand dynamic models is to load them onto the computer and

see them run. This is easily done using spreadsheet software such as Excel or econometrics-

oriented packages such as RATS. Figures 11.9 to 11.11 provide examples of the behaviour over

time of two economies, one that starts with a capital-output ratio that is half the steady-state

level, and other that starts with a capital output ratio that is 1.5 times the steady-state level.

The parameters chosen were s = 0.2, = 13 , g = 0.02, n = 0.01, = 0.06. Together these

parameters are consistent with a steady-state capital-output ratio of 2. To see, this plug these

values into (11.63):


K s 0.2

x = = g = =2 (11.69)
Y 1
+n+ 1.5 0.02 + 0.01 + 0.06

Figure 11.9 shows how the two capital-output ratios converge, somewhat slowly, over time

to their steady-state level. This slow convergence is dictated by our choice of parameters: Our

convergence speed is:

g 2
= (1 )( + n + ) = (1.5 0.02 + 0.01 + 0.06) = 0.067 (11.70)
1 3

So, the capital-output ratio converges to its steady-state level at a rate of about 7 percent

255
per period. These are fairly standard parameter values for annual data, so this should be

understood to mean 7 percent per year.

Figure 11.10 shows how output per worker evolves over time in these two economies. Both

economies exhibit growth, but the capital-poor economy grows faster during the convergence

period than the capital-rich economy. These output per worker differentials may seem a little

small on this chart, but the Figure 11.11 shows the behaviour of the growth rates, and this

chart makes it clear that the convergence dynamics can produce substantially different growth

rates depending on whether an economy is above or below its steady-state capital-output ratio.

During the initial transition periods, the capital-poor economy grows at rates over 6 percent,

while the capital-rich economy grows at under 2 percent. Over time, both economies converge

towards the steady-state growth rate of 3 percent.

256
Figure 11.9: Convergence Dynamics for the Capital-Output Ratio

257
Figure 11.10: Convergence Dynamics for Output Per Worker

258
Figure 11.11: Convergence Dynamics for the Growth Rate of Output
Per Worker

259
Illustrating Changes in Key Parameters

Figures 11.12 to 11.14 examine what happens when the economy is moving along the steady-

state path consistent with the parameters just given, and then one of the parameters is

changed. Specifically, they examine the effects of changes in s, and g.

Consider first an increase in the savings rate to s = 0.25. This has no effect on the

steady-state growth rate. But it does change the steady-state capital-output ratio from 2 to

2.5. So the economy now finds itself with too little capital relative to its new steady-state

capital-output ratio. The growth rate jumps immediately and only slowly returns to the long-

run 3 percent value. The faster pace of investment during this period gradually brings the

capital-output ratio into line with its new steady-state level.

The increase in the savings rate permanently raises the level of output per worker relative

to the path that would have occurred without the change. However, for our parameter values,

this effect is not that big. This is because the long-run effect of the savings rate on output per

worker is determined by s 1 , which in this case is s0.5 . So in our case, 25 percent increase in

the savings rate produces an 11.8 percent increase in output per worker (1.250.5 = 1.118). More

generally, a doubling of the savings rate raises output per worker by 41 percent (20.5 = 1.41).

The charts also show the effect of an increase in the depreciation rate to = 0.11. This

reduces the steady-state capital-output ratio to 4/3 and the effects of this change are basically

the opposite of the effects of the increase in the savings rate.

Finally, there is the increase in the rate of technological progress. Ive shown the effects of

a change from g = 0.02 to g = 0.03. This increases the steady-state growth rate of output per

worker to 0.045. However, as the charts show there is another effect: A faster steady-state

growth rate for output reduces the steady-state capital-output ratio. Why? The increase in

260
g raises the long-run growth rate of output; this means that each period the economy needs

to accumulate more capital than before just to keep the capital-output ratio constant. Again,

without a change in the savings rate that causes this to happen, the capital-output ratio will

decline. So, the increase in g means thatas in the depreciation rate examplethe economy

starts out in period 25 with too much capital relative to its new steady-state capital-output

ratio. For this reason, the economy doesnt jump straight to its new 4.5 percent growth rate

of output per worker. Instead, after an initial jump in the growth rate, there is a very gradual

transition the rest of the way to the 4.5 percent growth rate.

261
Figure 11.12: Capital-Output Ratios: Effect of Increases In ...

262
Figure 11.13: Growth Rates of Output Per Hour: Effect of Increases
In ...

263
Figure 11.14: Output Per Hour: Effect of Increases In ...

264
Convergence Dynamics in Practice

The Solow model predicts that no matter what the original level of capital an economy starts

out with, it will tend to revert to the equilibrium levels of output and capital indicated by

the economys underlying features. Does the evidence support this idea?

Unfortunately, history has provided a number of extreme examples of economies having

far less capital than is consistent with their fundamental features. Wars have provided the

natural experiments in which various countries have had huge amounts of their capital

destroyed. The evidence has generally supported Solows prediction that economies that

experience negative shocks should tend to recover from these setbacks and return to their

pre-shock levels of capital and output. For example, both Germany and Japan grew very

strongly after the war, recovering prosperity despite the massive damage done to their stocks

of capital by war bombing.

A more extreme example, perhaps, is study by Edward Miguel and Gerard Roland of the

long-run impact of U.S. bombing of Vietnam in the 1960s and 1970s.2 Miguel and Roland

found large variations in the extent of bombing across the various regions of Vietnam. Despite

large differences in the extent of damage inflicted on different regions, Miguel and Roland

found little evidence for lasting relative damage on the most-bombed regions by 2002. (Note

this is not the same as saying there was no damage to the economy as a whole the study

is focusing on whether those areas that lost more capital than average ended up being poorer

than average).

2
http://eml.berkeley.edu/ groland/pubs/vietnamoct09.pdf

265
Chapter 12
Endogenous Technological Change
The Solow model identified technological progress or improvements in total factor productivity

(TFP) as the key determinant of growth in the long run, but did not provide any explanation

of what determines it. In the technical language used by macroeconomists, long-run growth

in the Solow framework is determined by something that is exogenous to the model.

In these notes, we consider a particular model that makes technological progress endogeous,

meaning determined by the actions of the economic agents described in the model. The model,

due to Paul Romer (Endogenous Technological Change, Journal of Political Economy, 1990)

starts by accepting the Solow models result that technological progress is what determines

long-run growth in output per worker. But, unlike the Solow model, Romer attempts to

explain what determines technological progress.

TFP Growth as Invention of New Inputs

So what is this technology term A anyway? The Romer model takes a specific concrete view

on this issue. Romer describes the aggregate production function as

A
L1 (x1 x2 xA ) L1 xi
X
Y = Y + + .... + = Y (12.1)
i=1

where LY is the number of workers producing output and the xi s are different types of capital

266
goods. The crucial feature of this production function is that diminishing marginal returns

applies, not to capital as a whole, but separately to each of the individual capital goods

(because 0 < < 1).

If A was fixed, the pattern of diminishing returns to each of the separate capital goods

would mean that growth would eventually taper off to zero. However, in the Romer model,

A is not fixed. Instead, there are LA workers engaged in R&D and this leads to the invention

of new capital goods. This is described using a production function for the change in the

number of capital goods:

A = LA A (12.2)

The change in the number of capital goods depends positively on the number of researchers

( is an index of how slowly diminishing marginal productivity sets in for researchers) and

also on the prevailing value of A itself. This latter effect stems from the giants shoulders

effect.1 For instance, the invention of a new piece of software will have relied on the previous

invention of the relevant computer hardware, which itself relied on the previous invention of

semiconductor chips, and so on.

Romers model contains a full description of the factors that determines the fraction of

workers that work in the research section. The research sector gets rewarded with patents that

allow it to maintain a monopoly in the product it invents; wages are equated across sectors,

so the research sector hire workers up to point where their value to it is as high as it is to

producers of final output. In keeping with the spirit of the Solow model, Im going to just

treat the share of workers in the research sector as an exogenous parameter (but will discuss
1
Stemming from Isaac Newtons observation If I have seen farther than others, it is because I was standing
on the shoulders of giants.

267
later some of the factors that should determine this share). So, we have

L = LA + LY (12.3)

L A = sA L (12.4)

And again we assume that the total number of workers grows at an exogenous rate n:

L
=n (12.5)
L

Simplifying the Aggregate Production Function

We can define the aggregate capital stock as

A
X
K= xi (12.6)
i=1

Again, well treat the savings rate as exogenous and assume

K = sK Y K (12.7)

One observation that simplifies the analysis of the model is the fact that all of the capital

goods play an identical role in the production process. For this reason, we can assume that

the demand from producers for each of these capital goods is the same, implying that

xi = x i = 1, 2, ....A (12.8)

This means that the production function can be written as

Y = ALY1 x (12.9)

Note now that

K
K = Ax x = (12.10)
A
268
so output can be re-expressed as

K

Y = ALY1 = (ALY )1 K (12.11)
A

This looks just like the Solow models production function. The TFP term is written as

A1 as opposed to just A as it was in our first handout, but this makes no difference to the

substance of the model.

Steady-State Growth in The Romer Model

You can use the same arguments as before to show that this economy converges to a steady-

state growth path in which capital and output grow at the same rate. So, we can derive the

steady-state growth rate as follows. Re-write the production function as

Y = (AsY L)1 K (12.12)

where

sY = 1 sA (12.13)

Our usual procedure for taking growth rates give us

!
Y A sY L K
= (1 ) + + + (12.14)
Y A sY L K

Now use the fact that the steady-state growth rates of capital and output are the same to

derive that this steady-state growth rate is given by

! ! !
Y A sY L Y
= (1 ) + + + (12.15)
Y A sY L Y

Finally, because the share of labour allocated to the non-research sector cannot be changing

along the steady-state path (otherwise the fraction of researchers would eventually go to zero

269
or become greater than one, which would not be feasible) we have
!
Y L A
= (12.16)
Y L A

The steady-state growth rate of output per worker equals the steady-state growth rate of A.

The only difference from the Solow model is that writing the TFP term as A1 makes this
A 1 A
growth rate A
as opposed to 1 A
.

Deriving the Steady-State Growth Rate

The big difference relative to the Solow model is that the A term is determined within the

model as opposed to evolving at some fixed rate unrelated to the actions of the agents in the

model economy. To derive the steady-state growth rate in this model, note that the growth

rate of the number of capital goods is

A
= (sA L) A1 (12.17)
A

The steady-state of this economy features A growing at a constant rate. This can only be the

case if the growth rate of the right-hand-side of (12.17) is zero. Using our usual procedure for

calculating growth rates of Cobb-Douglas-style items, we get

!
sA L A
+ (1 ) = 0 (12.18)
sA L A

Again, in steady-state, the growth rate of the fraction of researchers ( ssAA ) must be zero. So,

along the models steady-state growth path, the growth rate of the number of capital goods

(and hence output per worker) is

!
A n
= (12.19)
A 1

The long-run growth rate of output per worker in this model depends on positively on

three factors:

270
 The parameter , which describes the extent to which diminishing marginal productivity

sets in as we add researchers.

The strength of the standing on shoulders effect, . The more past inventions help to

boost the rate of current inventions, the faster the growth rate will be.

The growth rate of the number of workers n. The higher this, the faster the economy

adds researchers. This may seem like a somewhat unusual prediction, but it holds well if

one takes a very long view of world economic history. Prior to the industrial revolution,

growth rates of population and GDP per capita were very low. The past 200 years have

seen both population growth and economic growth rates increases. See the figures on

the next two pages (the first comes from Greg Clarks book A Farewell to Alms which

provides a very interesting discussion of pre-Industrial-Revolution economies.)

271
Figure 12.1: World Economic History

272
Figure 12.2: Global Population

273
The Steady-State Level of Output Per Worker

Just as with our discussion of the Solow model, we can decompose output per worker into a

capital-output ratio component and a TFP component. In other words, one can re-arrange

equation (12.11) to get

Y K
  1
= A (12.20)
LY Y
and use the fact that LY = (1 sA )L to get

Y K
  1
= (1 sA ) A (12.21)
L Y

Note that the sA term reflects the reduction in the production of goods and services due

to a fraction of the labour force being employed as researchers. One can also use the same

arguments to show that, along the steady-state growth path the capital-output ratio is

K sK

= n (12.22)
Y n+ 1
+
n g
(The 1
here takes the place of the 1
in the first handouts expression for the steady-state

capital-output ratio because this is the new formula for the growth rate of output per worker).

Finally, we can also figure out the level of A along the steady-state growth path as follows.

Along the steady-state path, we have

A n
= (sA L) A1 = (12.23)
A 1

This latter equality can be re-arranged as

! 1

(1 ) 1
A = (sA L) 1 (12.24)
n

So, along the steady-state growth path, output per worker is

1
 1 !
Y sK (1 ) 1


= (1 sA ) n
(sA L) 1 (12.25)
L n+ 1
+ n

274
Convergence Dynamics for A

We noted already that the arguments showing that the capital-output ratio tends to converge

towards its steady-state are the same here as in the Solow model. What about the A term?

How do we know, for instance, that A always reverts back eventually to the path given by

equation (12.24)? To see that this is the case, let

A
gA = = (sA L) A1 (12.26)
A

The growth rate of theh right-hand-side of this equation is

gA sA
 
= + n (1 ) gA (12.27)
gA sA

One can use this equation to show that gA will be falling whenever

n sA
gA > + (12.28)
1 1 sA

So, apart from periods when the share of researchers is changing, the growth rate of A will be
n
declining whenever it is greater than its steady-state value of 1
. The same argument works in

reverse when gA is below its steady-state value. Thus, the growth rate of A displays convergent

dynamics, always tending back towards its steady-state value. And equation (12.24) tells us

exactly what the level of A has to be if the growth rate of A is at its steady-state value.

275
Optimal R&D?

We havent discussed the various factors that may determine the share of the labour force

allocated to the research sectors, sA . However, in equation (12.25) we have diagnosed two

separate offsetting effects that sA has on output: A negative one caused by the fact the

researchers dont actually produce output, and a positive one due to the positive effect of the

share of researchers on the level of technology.

Equation (12.25) looks very complicated but it looks simpler if we just take all the terms

that dont involve sA and bundle them together calling them X and also write Z = 1
. In

this case, the equation becomes


Y

= X (1 sA ) (sA )Z (12.29)
L

Written like this, it is a relatively simple calculus problem to figure out the level of sA that

maximises the level of output per worker along the steady-state growth path. In other words,

one can can differentiate equation (12.25) with respect to sA , set equal to zero, and solve to

obtain that this optimizing share of researchers is

Z
s
A = = 1 = (12.30)
1+Z 1 + 1 1+

When one fills in the model to determine sA endogenously, does the economy generally

arrive at this optimal level? No. The reason for this is that research activity generates

externalities that affect the level of output per worker, but which are not taken into account by

private individuals or firms when they make the choice of whether or not to conduct research.

Looking at the ideas production function, equation (12.2), one can see both positive and

negative externalities:

A positive externality due to the giants shoulders effect. Researchers dont take into

276
 account the effect their inventions have in boosting the future productivity of other

researchers. The higher is , the more likely it is that the R&D share will be too low.

A negative externality due to the fact that < 1, so diminishing marginal productivity

applies to the number of researchers.

Whether there is too little or too much research in the economy relative to the optimal level

depends on the strength of these various externalities. However, using empirical estimates of

the parameters of equation (12.2), Charles Jones and John Williams have calculated that

it is far more likely that the private sector will do too little research relative to the social

optimum.2

To give some insight into this result, note that the steady-state growth rate in this model is
n
1
, so 1
is the ratio of the growth rate of output per worker to the growth rate of population.

Suppose this equals one, so growth in output per worker equals growth in populationperhaps

a reasonable ballpark assumption. In this case 1
= 1 and the optimal share of researchers is

one-half. Indeed, for any reasonable steady-state growth rate, the optimal share of researchers

is very high, so it is hardly surprising that the economy does not automatically generate this

share.

This result points to the potential for policy interventions to boost the rate of economic

growth by raising the number of researchers. For instance, laws to strengthen patent pro-

tection may raise the incentives to conduct R&D. This points to a potential conflict between

macroeconomic policies aimed at raising growth and microeconomic policies aimed at reducing

the inefficiencies due to monopoly power: Some amount of monopoly power for patent-holders
2
Charles I. Jones and John C. Williams, Too Much of a Good Thing? The Economics of Investment in
R&D, Journal of Economic Growth, March 2000, Vol. 5, No. 1, pp. 65-85.

277
may be necessary if we want to induce a high level of R&D and thus a high level of output.

Robert Gordon on The Past and Future of New Technologies

Many of the facts about economic history back up Romers vision of economic growth. Robert

Gordons paper Is US economic growth over? Faltering innovation confronts the six head-

winds provides an excellent description of the various phases of technological invention and

also provides an interesting perspective on the potential for future technological progress.3

Gordon highlights how economic history can be broken into different periods based on how

the invention of technologies have impacted the economy.

The First Industrial Revolution: centered in 1750-1830 from the inventions of the steam

engine and cotton gin through the early railroads and steamships, but much of the impact

of railroads on the American economy came later between 1850 and 1900. At a minimum it

took 150 years for IR1 to have its full range of effects.

The Second Industrial Revolution: within the years 1870-1900 created within just a few years

the inventions that made the biggest difference to date in the standard of living. Electric light

and a workable internal combustion engine were invented in a three-month period in late

1879. The number of municipal waterworks providing fresh running water to urban homes

multiplied tenfold between 1870 and 1900. The telephone, phonograph, and motion pictures

were all invented in the 1880s. The benefits of IR2 included subsidiary and complementary

inventions, from elevators, electric machinery and consumer appliances; to the motorcar, truck,

and airplane; to highways, suburbs, and supermarkets; to sewers to carry the wastewater away.

All this had been accomplished by 1929, at least in urban America, although it took longer
3
CEPR Policy Insight, Number 63

278
to bring the modern household conveniences to small towns and farms. Additional follow-up

inventions continued and had their main effects by 1970, including television, air conditioning,

and the interstate highway system. The inventions of IR2 were so important and far-reaching

that they took a full 100 years to have their main effect.

The Third Industrial Revolution: is often associated with the invention of the web and

internet around 1995. But in fact electronic mainframe computers began to replace routine

and repetitive clerical work as early as 1960.

Gordons paper is very worth reading for understanding how the innovations associated

with the second industrial revolution completely altered peoples lives. He describes life in

1870 as follows

most aspects of life in 1870 (except for the rich) were dark, dangerous, and involved

backbreaking work. There was no electricity in 1870. The insides of dwelling units

were not only dark but also smoky, due to residue and air pollution from candles

and oil lamps. The enclosed iron stove had only recently been invented and much

cooking was still done on the open hearth. Only the proximity of the hearth or

stove was warm; bedrooms were unheated and family members carried warm bricks

with them to bed.

But the biggest inconvenience was the lack of running water. Every drop of water

for laundry, cooking, and indoor chamber pots had to be hauled in by the house-

wife, and wastewater hauled out. The average North Carolina housewife in 1885

had to walk 148 miles per year while carrying 35 tonnes of water.

Gordon believes that the technological innovations associated with computer technologies

279
are far less important than those associated with the second industrial revolution and that

growth may sputter out over time. Figure 1 repeats a chart from Gordons paper showing

the growth rate of per capita GDP for the worlds leading economies (first the UK, then

the US). It shows growth accelerating until 1950 and declining thereafter. Figure 2 shows a

hypothetical chart in which Gordon projects a continuing fall-off in growth.

To illustrate why he believes modern inventions dont match up with past improvements,

Gordon offers the following thought experiment.

You are required to make a choice between option A and option B. With option

A you are allowed to keep 2002 electronic technology, including your Windows 98

laptop accessing Amazon, and you can keep running water and indoor toilets; but

you cant use anything invented since 2002.

Option B is that you get everything invented in the past decade right up to Face-

book, Twitter, and the iPad, but you have to give up running water and indoor

toilets. You have to haul the water into your dwelling and carry out the waste.

Even at 3am on a rainy night, your only toilet option is a wet and perhaps muddy

walk to the outhouse. Which option do you choose?

You probably wont be surprised to find out that most people pick option B.

Gordon also discusses other factors likely to hold back growth in leading countries such

as the leveling off of a long-run pattern of educational achievement, an aging population

and energy-related constraints. Its worth noting, though, that while Gordons paper is very

well researched and well argued, economists are not very good at forecasting the invention

of new technologies or their impact on the economy. For all we know, the next industrial

280
revolution could be around the corner to spark a new era of rapid growth. Joel Mokyrs

article Is technological progress a thing of the past? is a good counterpart to Gordons

scepticism.4

Figure 12.3: Gordon on the Growth Rate of Leading Economies

4
Available at www.voxeu.org/article/technological-progress-thing-past

281
Figure 12.4: Gordons Hypothetical Path for Growth

282
Chapter 13
Cross-Country Technology Diffusion
So far, weve been discussing how the invention of new technologies promotes economic growth

by pushing out the technological frontier and allowing capital to be allocated across new

and old technologies with diminishing returns setting in. This is clearly an important aspect

of economic growth. However, we should remember that only a very few countries in the

world are on the technological frontiermost places are not relying on Apple to invent a

new gadget to promote efficiency. One way to illustrate this point is to estimate the level of

total factor productivity for different countries in the world.

An important paper that did these calculations and used them to shed light on cross-

country income differences is Hall and Jones (1999).1 The basis of the study is a levels

accounting exercise that starts from the following production function

Yi = Ki (hi Ai Li )1 (13.1)

Like the BLS multifactor productivity calculations that we discussed a few lectures ago,

Hall and Jones account for the effect of education on the productivity of the labour force.

Specifically, they construct measures of human capital based on estimates of the return to

educationthis is the hi in the above equation.

1
Robert E. Hall and Charles I. Jones. Why Do Some Countries Produce So Much More Output per
Worker than Others?, Quarterly Journal of Economics, February 1999.

283
 Hall and Jones show that their production function can be re-formulated as

Yi Ki
  1
= hi Ai (13.2)
Li Yi

Hall and Jones then constructed a measure hi using evidence on levels of educational attain-

ment and they also set = 1/3. This allowed them to use (13.2) to express all cross-country

differences in output per worker in terms of three multiplicative terms: capital intensity, hu-

man capital per worker, and technology or total factor productivity. They found that output

per worker in the richest five countries was 31.7 times that in the poorest five countries. This

was explained as follows:

Differences in capital intensity contributed a factor of 1.8.

Differences in human capital contributed a factor of 2.2

The remaining differencea factor of 8.3was due to differences in TFP.

The results from this paper show that differences in total factor productivity, rather than dif-

ferences in factor accumulation, are the key explanation of cross-country variations in income

levels. A more detailed table of Hall and Joness calculations is reproduced on the next page.

These calculations show that most countries are very far from the technological frontier, so

their growth is not likely to be reliant on the invention of new technologies.

284
Table from Hall-Jones Paper

285
Leaders and Followers

The Romer model probably should not be thought of as a model of growth in any one partic-

ular country. No country uses only technologies that were invented in that country; rather,

products invented in one country end up being used all around the world. Thus, the model is

best thought of as a model of the leading countries in the world economy. How then should

long-run growth rates be determined for individual countries? By itself, the Romer model

has no clear answer, but it suggests a model in which ability to learn about the usage of new

technologies should plays a key role in determining output per worker.

We will now describe such a model. The mathematics of the model are also formally

equivalent to a well-known model of Nelson and Phelps (AER, 1966), though the application

there is different, their subject being the diffusion of technological knowledge over time within

an individual country.

The Model

We will assume that there is a lead country in the world economy that has technology level,

At at time t which grows at rate g every period, so that

At
=g (13.3)
At

All other countries in the world, indexed by j, have technology levels given by Ajt < At . The

growth rate of technology in country j is determined by

Ajt (At Ajt )

= j + j (13.4)
Ajt Ajt

where j < g and j > 0. This tells us that technology growth in all countries apart from the

lead country is determined by two factors

286
 Learning: The second term says that their technology level will grow faster the bigger

is the percentage gap between its level of technology, Ajt and the level of the leader,

At . The larger is the parameter j , the better the country is at learning about the

technologies being applied in the lead country.

The first term, j indicates the countrys capacity for increasing its level of technology

without learning from the leader. We impose the condition j < g. This means that

country j cant grow faster than the lead country without the learning that comes from

having lower technology than the frontier.

Exponential Growth

Youve probably heard about exponential functions before but, even if you have, its worth a

quick reminder. The number e 2.71828 is a very special number such that the function

dex
= ex (13.5)
dx

One way to see why the number is 2.718 is to use something called the Taylor series approxi-

mation for a function, which states that you can approximate a function f (x) as

1 1 1
f (x) = f (a)+f 0 (x)(xa)+ f 00 (x))(xa)2 + f 000 (x))(xa)3 +... f n (x))(xa)n +... (13.6)
2 3! n!

where n! = (1)(2)(3)...(n 1)(n). If there is a number, e that has the property that ex =

f (x) = f 0 (x), then that means that all derivatives also equal ex . In this case, we have

1 1
ex = ea + ea (x a) + ea (x a)2 + ea (x a)3 + ... (13.7)
2 3!

Setting x = 1, a = 0, this becomes

1 1 1
e=1+ + + + ..... (13.8)
2 3! 4!
287
This converges to 2.71828. Ok, thats not on the test but worth knowing. Now note that
degt degt dg
= = gegt (13.9)
dt d(gt) dt
Now lets relate this back to our model. The fact that the lead country has growth such that
dAt
= At = gAt (13.10)
dt
means that this country is characterised by what is known as exponential growth, i.e.

At = A0 egt (13.11)

We write the first term as A0 because e(g)(0) = 1 so whatever term multiplies egt that is the

value that At takes in the first period.

Dynamics of Technology

Now we are going to try to figure out how the technology variable behaves in the follower

country. First, lets take equation (13.4) and multiply across by Ajt to get

Ajt = j Ajt + j (At Ajt ) (13.12)

This is what is known as a first-order linear differential equation (differential equation because

it involves a derivative; first-order because it only involves a first derivative; linear because it

doesnt involve any terms taken to powers than are not one.) These equations can be solved

to illustrate how Aj changes over time. To do this, we first draw some terms together to

re-write it as

Ajt + (j j ) Ajt = j At (13.13)

Recalling equation (13.11) for the technology level of the leader country, this differential

equation can be re-written as

Ajt + (j j ) Ajt = j A0 egt (13.14)

288
Now well move on to illustrating how people figure out how an Ajt that satisfies this equation

needs to behave.

One Possible Solution

Lets think about what we learned about exponential functions to help us see what form a

potential solution might take. The derivative of Ajt with respect to time plus (j j ) times

Ajt can be written as a multiple of the exponential function.

Looked at this way, we might guess that one possible solution for an Ajt process that will

satisfy this equation is something of the form Bj egt where Bj is some unknown coefficient.

Indeed, it turns out that this is the case. Lets figure out what Bj must be. It must satisfy

gBj egt + (j j ) Bj egt = j A0 egt (13.15)

Canceling the egt terms, we see that

j A0
Bj = (13.16)
j + g j

So, this solution takes the form

! !
j j
Apjt gt
= Bj e = gt
A0 e = At (13.17)
j + g j j + g j

A General Solution

Is that it or could we add on an additional term and still get a solution? Suppose we look for

a solution of the form

Ajt = Bj egt + Djt (13.18)

289
Then the solution would have to obey

 
gBegt + Djt + (j j ) Begt + Djt = j A0 egt (13.19)

All the terms in egt cancel out because, by construction of Bj , they satisfy equation (13.15).

This means the additional term Djt must satisfy

Djt + (j j ) Djt = 0 (13.20)

Again using the properties of the exponential function, this equation is satisfied by anything

of the form

Djt = Dj0 e(j j )t (13.21)

where Dj0 is a parameter that can take on any value. So, given the differential equation

(13.12), all possible solutions for technology in country j must take the form
!
j
Ajt = At + Dj0 e(j j )t (13.22)
j + g j

where Dj0 is an arbitrary parameter than can take any value.

Properties of the Solution

Now we like to examine the properties of this solution. Does technology in the follower country

catch up and, if not, where does it end up and why? To answer these questions, it is useful

to express Ajt as a ratio of the frontier level of technology. This can be written as

Ajt j Dj0 (j j )t
= + e (13.23)
At j + g j At

Now using the fact that At = A0 egt , this becomes

Ajt j Dj0 (j +gj )t

= + e (13.24)
At j + g j A0

290
To understand the properties of this solution, recall that we assumed j < g, which means

that on its own (without catch-up growth) the follower countrys level of technology grows

slower than the leader country and also that j > 0 (some learning takes place). Putting

these two assumptions together, we can say

j + g j > 0 (13.25)

That means that

e(j +gj )t 0 as t (13.26)

This means that the second term in (13.24) tends towards zero. So, over time, as this term
j
disappears, the country converges towards a level of technology that is a constant ratio, j +gj

of the frontier level, and its growth rate tends towards g.

Note that g j > 0 also means that

j
0< <1 (13.27)
j + g j

so each country never actually catches up to the leader but instead converges to some fraction

of the lead countrys technology level. This makes sense if you think about it. Because of their

inferiority at developing their own technologies (j < g) the follower countries will always be

falling further behind the leader unless there is a gap between their level of technology and

the leader. So, to have a steady-state in which everyones technology is growing at the same

rate, the followers must all have technology levels below that of the leader.

In addition, g j > 0 means that

!
d j
>0 (13.28)
dj j + g j

The equilibrium ratio of the countrys technology to the leaders depends positively on the

learning parameter j . The higher this parameterthe more fo the gap to the leader that

291
it closes each periodthen the close the ratio gets to one and the higher up the pecking

order the country gets. Its also true that

!
d j
>0 (13.29)
dj j + g j

In other words, the more growth the country can generate each period independent of learning

from the leader, the higher will be its equilibrium ratio of technology relative to the leader.

Illustrating the Model

Going back to the equation for the ratio of technology in country j to the leader, equation

(13.24), we noted already that the second term tends to disappear to zero over time. That

doesnt mean its unimportant. How a country behaves along its transition path depends

on the value of the initial parameter Dj0 .

If Dj0 < 0, then the term that is disappearing over time is a negative term that is

a drag on the level of technology. This means that the country starts out below its

equilbrium technology ratio, grows faster than the leader for some period of time with

growth eventually tailing off to the growth rate of the leader.

If Dj0 > 0, then the term that is disappearing over time is a positive term that is

boosting the level of technology. This means that the country starts out above its

equilbrium technology ratio, grows slower than the leader for some period of time with

growth eventually moves up towards the growth rate of the leader.

We have illustrated how these dynamics would work with the Figures 13.1 to 13.3. These

charts show model simulations for a leader economy with g = 0.02 and a follower economy

292
with j = 0.01 and j = 0.04. These values mean

j 0.04
= = 0.8 (13.30)
j + g j 0.04 + 0.02 0.01

so the follower economy converges to a level of technology that is 20 percent below that of

the leader. The first collection of charts show what happens when this economy has a value

of Dj0 = 0.5, so that it starts out with a technology level only 30 percent that of the leader.

They grow faster than the leader country for a number of years before they approach the

0.8 equilibrium ratio and then their growth rate settles down to the same rate as that of the

leader.

The second collection of charts show what happens when this economy has a value of

Dj0 = 0.5, so that it starts out with a technology level 30 percent above that of the leader,

even though the equilibrium value is 20 percent below. Technology levels in this follower

country never actually decline but they do go through a long-period of slow growth rates

before eventually heading towards the same growth rate as the leader as they approach the

0.8 equilibrium ratio.

Finally, we show how the model may also be able to account for the sort of growth

miracles that are occasionally observed when countries suddenly start experiencing rapid

growth: If a country can increase its value of j via education or science-related policies, its

position in the steady-state distribution of income may move upwards substantially, with the

economy then going through a phase of rapid growth. The third collection of charts show

what happens when, in period 21, an economy changes from having j = 0.005 to j = 0.04.

The equilbrium technology ratio changes from one-third to 0.8 and the economy experiences

a long transitional period of rapid growth.

An important message from this model is that for most countries, it is not their ability to

293
invent new capital goods that is key to high living standards, but rather their ability to learn

from those countries that are more technologically advanced.

Figure 13.1: Catching Up

A Follower Starts Out Below Their Equilibrium Technology Ratio

g=0.02, Lambda(j)=0.01, Sigma(j)=0.04

Technology Levels Over Time Ratio of Follower to Leader Technology Growth Rates of Technology
9 0.8 0.10

8 0.09

7 0.7 0.08

6 0.07

0.6
5 0.06

4 0.05

0.5
3 0.04

2 0.03

0.4
1 0.02

0 0.01
20 40 60 80 100 25 50 75 100
0.3
Leader Follower Leader Follower
20 40 60 80 100

294
 Figure 13.2: Falling Back

A Follower Starts Out Above Their Equilibrium Technology Ratio

g=0.02, Lambda(j)=0.01, Sigma(j)=0.04

Technology Levels Over Time Ratio of Follower to Leader Technology Growth Rates of Technology
9 1.3 0.0225

0.0200
8

1.2 0.0175
7

0.0150
6

1.1
0.0125

0.0100

4 1.0
0.0075

3
0.0050

0.9
2
0.0025

1 0.0000
20 40 60 80 100 25 50 75 100
0.8
Leader Follower Leader Follower
20 40 60 80 100

295
 Figure 13.3: A Growth Miracle

An Increase in the Rate of Learning

Sigma(j) Increases from 0.005 to 0.04 in Period 21

Technology Levels Over Time Ratio of Follower to Leader Technology Growth Rates of Technology
9 0.8 0.09

8
0.08

7 0.7
0.07

6
0.06

0.6
5

0.05

0.5 0.04
3

0.03
2

0.4
0.02
1

0 0.01
20 40 60 80 100 25 50 75 100
0.3
Leader Follower Leader Follower
20 40 60 80 100

296
Chapter 14
Institutions and Efficiency
We have documented huge differences in total factor productivity across countries. What

determines these differences? One answer is provided by the combination of the Romer model

and the leader-follower model. According to these models, large differences in TFP reflect

variations in the extent to which countries have adopted the latest technologies.

However, this is perhaps too mechanistic a view of what generates cross-country differences

in efficiency. TFP doesnt just reflect the technologies a countrys people use. It is a measure

of the efficiency with which an economy makes use of its resources and there are a whole range

of other factors that can affect this. For example:

Bureaucratic Inefficiency and Corruption: Satisfaction of bureaucratic requirements and

bribing of officials can be important diversions of resources in poor economies.

Crime: Time spent on crime does not produce output. Neither do resources devoted to

protecting inviduals and firms from crime.

Restrictions on Market Mechanisms: Protectionism, price controls, and central planning

can all lead to resources being allocated in an inefficient manner.

In addition, while technology adoption certainly has an impact on differences in TFP,

this still leaves open the question of what drives the pace of technology adoption in poorer

297
countries. Ultimately, the models so far dont answer the question of the deeper determinants

of economic success. We will now discuss on the idea that the ultimate explanation for patterns

of economic efficiency relates to differences in institutions.

Douglass North and Institutions

There is now a large literature that focuses on the idea that differences in institutions provides

the key to understanding TFP differences across countries. Economic activity does not take

place in a vacuum. Firms need to take account of the legal and regulatory environment,

the tax system, and the services provided by government as well as the political setting that

determines these institutions.

The work of economic historian Douglass North, winner of the 1993 Nobel prize for eco-

nomics, was particularly influential in stressing the key importance of good institutions for

economic growth. The introduction to Norths paper Institutional Change: A Framework of

Analysis gives a flavour of his arguments:

A theory of institutional change is essential for further progress in the social sci-

ences in general and economics in particular. Essential because neo-classical theory

(and other theories in the social scientists toolbag) at present cannot satisfacto-

rily account for the very diverse performance of societies and economies both at a

moment of time and over time. The explanations derived from neo-classical theory

are not satisfactory because, while the models may account for most of the differ-

ences in performance between economies on the basis of differential investment in

education, savings rates, etc., they do not account for why economies would fail to

undertake the appropriate activities if they had a high payoff. Institutions deter-

298
 mine the payoffs. While the fundamental neo-classical assumption of scarcity and

hence competition has been robust (and is basic to this analysis), the assumption

of a frictionless exchange process has led economic theory astray. Institutions are

the structure that humans impose on human interaction and therefore define the

incentives that (together with the other constraints (budget, technology, etc.) de-

termine the choices that individuals make that shape the performance of societies

and economies over time.

He goes to discuss the link between institutions and the profit-maximising decisions that

people will take:

Institutions consist of formal rules, informal constraints (norms of behavior, con-

ventions, and self imposed codes of conduct) and the enforcement characteristics

of both ... If institutions are the rules of the game, organizations are the players.

They are groups of individuals engaged in purposive activity. The constraints im-

posed by the institutional framework (together with the other constraints) define

the opportunity set and therefore the kind of organizations that will come into

existence ... If the highest rates of return in a society are to be made from piracy,

then organizations will invest in knowledge and skills that will make them better

pirates; if organizations realize the highest payoffs by increasing productivity then

they will invest in skills and knowledge to achieve that objective.

This paper contains a discussion of some aspects of the USs institutional history that have

been positive for economic growth. Much of Norths other work focuses on the development

of institutions that made some countries such as the UK successful early developers through

the industrial revolution while others lagged.

299
An Example of the Importance of Institutions

Korea provides an extreme example of the importance of institutions in determining the

success of an economy. After World War II, Korea was split into a northern zone that became

the Democratic Peoples Republic of Korea, a Soviet-style socialist republic, while South Korea

became a capitalist economy.

North Korea received external support from the USSR for many years but no longer

receives external aid. It remains a centrally planned economy with only one political party.

The economy has failed to prosper and there are reliable reports of large amounts of death

from famine in the 1990s. In contrast, South Korea has been a huge economic success and is

home to many globally successful corporations such as Samsung and Hyundai.

The figure on the next page illustrates the gap between North and South Korea. While

the two areas began with few substantive differences, sharing a common culture and identity,

their different economic institutions mean that they are now completely different. Viewed

from the sky, you can see development all over South Korea while North Korea is almost fully

dark because of a lack of electricity.

300
Figure 14.1: The Korean Peninsula at Night

301
An Econometric Approach

The historical approach adopted by North and isolated examples of extreme events (such as the

Korean split) been very valuable in highlighting cases where good institutions have facilitated

economic growth and where bad institutions have prevented it. More recently, there has

been an attempt to assess the role of institutions in economic development using more formal

econometric techniques. An early paper in this literature was the 1999 Quarterly Journal of

Economics paper by Robert Hall and Charles I. Jones (Recall that we previously discussed

this papers calculations of the sources of differences in output per worker). They used the

term social infrastructure to describe the institutions that affect incentives to produce and

invest. Their approach was to collect data on a large number of countries and then estimate

regressions of the form

Yi
= + Si + i (14.1)
Li
Y
where L
is output per worker in country i and Si is a variable that aims to measure the extent

to which institutions in country i facilitate economic activity. Hall and Jones constructed their

Si variable as an average of two different variables:

1. An index of government antidiversion policies. This is an average of five different

variables relating to (i) law and order (ii) bureaucratic quality (iii) corruption (iv) risk

of expropriation, and (v) government repudiation of contracts.

2. An index that focuses on the openness of a country to trade with other countries

There are two potentially serious econometric problems when assessing the linkage between

productivity and institutions. The first is endogeneity. Do countries get rich because they

have good institutions or do countries have good institutions because they are rich? The

302
latter linkage certainly exists. Citizens in richer countries have substantial incentives to keep

good institutions that promote productive efficiency because they would have lot to lose if

their markets ceased to work well; these incentives may be substantial weaker in the worlds

poorer countries. Hall and Jones thus describe their social infrastructure variable as being

determined by
Yi
Si = + + Xi + i (14.2)
Li
Yi
In this case, a simple OLS regression of Li
on Si will produce a positive estimate of the

effect of institutions on output per workereven if the true value of was zero.

The second econometric problem is measurement error. The variables used as measures of

institutional quality can only ever be proxies, and possibly poor proxies, for the true measure

of institutional quality that actually affects economic output. The use of proxies like this is the

same as using variables that are affected by measurement error. One of the standard results

from econometrics is that measurement error can result in downward bias in coefficients. In

other words, the OLS coefficient might be less than the true coefficient.

So the presence of these econometric problems means OLS estimation will produce biased

estimates, though whether the bias is upwards or downwards depends on the source of the bias.

The usual solution to these econometric problems is estimation via instrumental variables.

This means estimating from

Yi
= + Si + i (14.3)
Li
where Si is the fitted value from a regression of S on a set of instruments (exogenous variables

that that may be correlated with the institutions variable but that are not affected by the

countrys level of output per worker). By focusing on variations in institutions related to

exogenous factors that are not determined by output per worker, the researcher can try to

303
identify the true causal effect of institutions.

Hall and Joness Findings

Finding good instruments for this problem can be tricky. Many of the papers in this literature

have focused on either geography or history as their inspiration for truly exogenous sources of

variations in institutions.

A countrys geography is certainly exogenousit is not influenced by a countrys level of

prosperity. But certain types of geographical features may be correlated with whether

a country has good institutions or not. Hall and Jones used the countrys distance

from the equator as an instrument. Other papers have also used coastal access, average

temperature, rainfall and soil quality.

In relation to history, many countries around the world were colonised by various Eu-

ropean countries and their current institutions (e.g. whether a country uses a French or

English legal systems) are often determined, in a somewhat random fashion, by which

countries colonised them. Hall and Jones used instruments measuring the fraction of

people speaking English as a native language and a variable measuring the fraction of

people speaking other Western European languages.

Using their selected instrument set, Hall and Jones found a positive and significant effect of

their social infrastructure variable when estimating using IV methods, with the coefficient

being higher than the OLS estimate. They concluded from this that there is a large causal

effect from institutions to productivity and that the measurement error is a more important

source of bias in their OLS regressions than is endogeneity.

304
Some Other Papers

There is now a large empirical literature on this topic. Some examples:

Acemoglu, Johnson, and Robinson (AER, 2001) assess the effect on GDP per capita of

institutions, proxied by a measure of protection against expropriation risk. They use a

new instrument measuring settler mortality in different European colonies. They argue

that countries where mortality for initial settlers was low were places where Europeans

were more likely to settle and set up good institutions, with the reverse working when

settler mortality was high. With this variable as an instrument, they find a very strong

effect of certain measures of institutions on output per capita.

Rodrik, Subramanian and Trebbi (Journal of Economic Growth, 2004). These authors

assess the role of institutions (as proxied by a variable measuring the strength of the rule

of law), openness to trade and geography (as measured by distance from the equator).

To be able to assess whether geography has a direct effect on income per capita, they use

other variables such as the AJR settler mortality variable and language-related variables

as instruments. They conclude that institutions, in the form of their rule of law variable,

are the key determinant of economic success and do not find a significant role for trade

or geography.

Gillanders and Whelan (2014) compare the effect of the Rule of Law variable preferred

by Rodrik, Subramanian and Trebbi with a new variable that measures the ease of

doing business. Both are institutional variables but they measure different types of

institutions. This paper also applies IV methods using geographical variables as instru-

ments and concludes that it is the ease of doing business that is the key determinant of

output per capita rather than Rule of Law variable.

305
 Part IV
Growth and Resources

306
Chapter 15
The Malthusian Model
We have devoted the last few weeks to studying economies that grow steadily over time. For

many countries around the world, that has been a reasonable description of their behaviour

since the start of the Industrial Revolution. However, prior to around the year 1800, there is

very little evidence of steady growth in income levels. The chart on the next page repeats the

chart shown earlier from the book A Farewell to Alms by economic historian Greg Clark. It

summarises world economic history as a long period in which living standards fluctuated over

time showing no growth trend before the Industrial Revolution lead to steady growth over

time (though Clark notes that this take-off did not occur in all countries and some remain

exceptionally poor).

Measurement of living standards is an imprecise business even in modern economies with

well-resourced statistical agencies. So its hardly surprising that there is a lot of controversy

over Clarks particular interpretation of the evidence as implying no trend growth at all

in living standards prior to 1800. Other studies show slow but gradual increases in living

standards prior to the Industrial Revolution but all agree that the average rate of economic

growth was very low before 1800. In addition, Figure 15.2 shows that global population growth

was extremely slow until 1800 and then increased to much higher rates.

What explains these patterns? Our previous models would suggest the pace of technolog-

307
ical progress must have been slower before the Industrial Revolution and this is true. But

cumulatively, there was a lot of technological progress in the years prior to 1800 with many ad-

vances made in science and in the organisation of economic life. One might have expected this

to translate into growth in average living standards over time but the evidence suggests such

progress was limited. In these notes, we will present the Malthusian model, which explains

how the world works very differently when rates of technological progress are slow.

308
Figure 15.1: World Economic History (from Greg Clarks book)

309
Figure 15.2: Global Population

310
Life Expectancy and Income Levels

The Malthusian model has two key elements: A negative relationship between income levels

and the size of population and a positive relationship between income levels and population

growth. Lets start with the second relationship

By definition, population growth increases with birth rates and falls with death rates.

Death rates, in turn, are the key determinant of life expectancy. Throughout history, there

has been a strong relationship between a countrys average level of income per capita and its

average life expectancy. This relationship still holds strongly today. Figure 3 shows a chart

taken from a wonderful website called Gapminder which allows you to make animated charts

showing developments over time and around the world in income levels, health outcomes and

lots of other areas.

Figure 15.3 shows the relationship between average life expectancy and real income per

person. Each dot corresponds to a country, with the size of the dot corresponding to its

population (the big red dot is China, the big blue dot is India, the yellow dot at the top right

is the US). The chart shows that in the poorest countries in the world in 2012, average life

expectancy was as low as under 50 years of age while the richest countries tend to have average

life expectancy of over 80 years. Figure 15.4 shows a relationship of this kind holding inside

a large country: U.S. counties with higher income per capita have longer life expectancy.

Internationally, this pattern is partly related to the availability of medicines in advanced

countries that allow people to live much longer. But it is more influenced by very high

rates of child mortality. Figure 15.5 shows another Gapminder chart. This one shows that

mortality among children under 5 is still very common in the worlds poorest countries due

to malnuitrition and poor public health systems.

311
 This relationship between income levels and the rate of death among the population will

be a key element of the version of the Malthusian model that we will cover.

Figure 15.3: Life Expectancy and Real GDP Per Capita Around the
World in 2012

312
Figure 15.4: Life Expectancy and Income Levels: U.S. Counties

313
Figure 15.5: Child Mortality and Real GDP Per Capita Around the
World in 2012

314
Population and Income Levels

The second element of the Malthusian model is a negative relationship between income levels

and the level of population. Before discussing Malthuss thoughts on this issue, its worth

using the language of modern economics to describe this relationship.

Consider an economy in which aggregate output is determined by a Cobb-Douglas pro-

duction function

Yt = AK L1
t (15.1)

Here, Ive assumed that both capital and technology are fixed (and so have no time subscript),

so that labour input is the only factor that produces changes in output. We can figure out

the demand for labour by assuming that the firms in the economy maximise profits in a

competitive manner. Thus, firms are maximising

= pAK L1
t wL rK (15.2)

where p is the price of output, w is the wage rate and r is an implicit rental rate for capital.

The first-order condition for labour is

pAK L w = 0 (15.3)

This can be re-arranged as


w K

=A (15.4)
p L
Assuming that a constant fraction of the population is working

L = N (15.5)

we get

w K

=A (15.6)
p N

315
The higher the population, the lower will be the real wage. This is because of diminishing

marginal returns to labour and the fact that workers are being paid their marginal wage

product.

Now note that the direct link between higher population and lower wage rates (and thus

lower living standards) works here because technology and capital are held constant. In the

Solow growth model, there is both rising population and increasing wages because technology

improvements and capital accumulation offset the negative effects on wages of rising pop-

ulation. In this example, we have assumed something quite different, i.e. no technological

progress. We will return, however, to the question of what happens when there is a slow but

steady rate of technological improvement.

Malthus (1798)

Thomas Malthuss 1798 book An Essay on the Principle of Population put together the two

ideas that we have just discussed. He noted that rising living standards can lead to higher

population growth but the famously-gloomy Malthus believed that this increase in population

would ultimately undo the original increase in living standards.

Malthus placed a somewhat different emphasis on the various links than in our discussion.

In relation to the link between demographics and living standards, Malthus focused on two

mechanisms (checks on living standards) that would cause population growth to increase

as living standards rose and thus ultimately see the increase in living standards reversed.

The first mechanism, which Malthus labelled the preventative check was the tendency to

see more births when real wages are high. In pre-Industrial Revolution Britain, the tradition

was for people to marry relatively late as they waited to accumulate the wealth to be able to

316
support a family. This tended to keep fertility rates relatively low. In practice, as discussed in

Greg Clarks book on the Malthusian model, the evidence for a link between living standards

and birth rates prior to the Industrial Revolution is fairly weak and I will assume a constant

birth rate in the model treatment below (though the logic of the model is unchanged if you

assume a positive relationship between birth rates and living standards.)

The second mechanism, which Malthus labelled the positive check, was the negative

effect of living standards on death rates. Evidence for this mechanism is stronger and still

exists today. Malthus describes it as follows:

the actual distresses of some of the lower classes, by which they are disabled from

giving the proper food and attention to their children, act as a positive check to the

natural increase of population.

This is the mechanism that we will focus on in our description of the model.

In relation to the negative effect of population on living standards, Ive used a production

function approach and emphasised the role played by the assumption of technology increases

failing to offset the effect of increased population. Malthus focused more the idea of increased

numbers of people putting a strain on food resources:

An increase of population without a proportional increase of food will evidently

have the same effect in lowering the value of each man patent. The food must

necessarily be distributed in smaller quantities, and consequently a day labour will

purchase a smaller quantity of provisions. An increase in the price of provisions

would arise either from an increase of population faster than the means of subsis-

tence, or from a different distribution of the money of the society.

317
The Model and its Convergent Dynamics

We will now describe a Malthusian model in somewhat more formal terms than Malthus did.

Basically, Im following Greg Clarks version of the model as described in Chapter 2 of his

book, though Im using a constant birth rate rather than one that depends on income levels.

The model has four equations. First, there is the definition of the change in the population,

which just states that population equals last periods population plus last periods level of

births minus deaths. (There are lots of different possible timing conventions here. I have

in mind that the population level is measured at the start of each period, while births and

deaths occur over the course of the period, but the particular timing convention adopted isnt

important):

Nt = Nt1 + Bt1 Dt1 (15.7)

Births are a constant fraction of the population

Bt
=b (15.8)
Nt

While deaths are a decreasing function of real income per person

Dt
= d0 d1 Yt (15.9)
Nt

Finally, real income per person is a negative function of the population size:

Yt = a0 a1 Nt (15.10)

Figure 15.6 shows how the death and birth rate equations combine together to make population

dynamics a function of income per person. The death rate depends negatively on income per

person, so at sufficiently high income levelsin this case, levels above Y births are greater

than deaths and population is growing, while population is falling at income levels below Y .

318
 Figure 15.7 then shows that the economy tends to return to this equilibrium level of

income. When income is above Y , population is growing. But Figure 7 shows that growing

population means income levels are falling. So income levels tend to fall when income is

above Y and increase when it is below Y . Similarly population tends to fall when it is

above the level of population associated with Y , call this N , and rise when it is below this

level. This means that both income and population display what we have called convergent

dynamics in our discussion of the Solow model: Wherever the economy starts out, it tends to

converge towards these specific levels of income and population. Because the economy tends

to revert back to the same levels of income and population, this phenomenon is often called

The Malthusian Trap.

Figure 15.8 shows how the birth and death schedules, on the one hand, and the income-

population schedule on the other, combine to determine the models properties. Perhaps

surprisingly, it is the birth and death schedules and not the income-population schedule that

determines the long-run level of real income per person in the model. The income-population

schedule then determines how many people are alive, given that level of income.

319
 Figure 15.6: Birth and Death Rate Schedules

BIRTH
AND
DEATH
RATE
BIRTH RATE

DEATH RATE

INCOME PER PERSON

Y*

320
 Figure 15.7: The Income-Population Schedule

POPULATION

N*

INCOME PER PERSON

Y*

321
 Figure 15.8: The Full Model

BIRTH
AND
DEATH
RATE
BIRTH RATE

DEATH RATE

Y* INCOME PER PERSON

POPULATION

N*

Y0 Y* Y1 INCOME PER PERSON

322
Calculating the Long-Run Equilibrium

We can figure N and Y out algebraically as follows. Combining the birth and death schedules

with the equation for population change, we get

Nt Nt1
= b d0 + d1 Yt1 (15.11)
Nt1

Inserting the dependence of income levels on wages, we get

Nt Nt1
= b d0 + d1 a0 d1 a1 Nt1 (15.12)
Nt1

This shows that the growth rate of population depends negatively on last periods level of

population: This is what determines the convergent dynamics. The level of N such that

population stays unchanged, shown in the Figure 7 as N , is given by

b d0 + d1 a0 d1 a1 N = 0 (15.13)

which solves to give

b d 0 + d 1 a0
N = (15.14)
d 1 a1
The long-run equilibrium level of population depends positively on the birth rate, b, and on a0 ,

which effectively measures the level of technology in the model (if this increases it can offset

the negative effect of higher population on income levels). The equilibrium level of population

depends negatively on the exogenous element of the death rate (d0 ), on the sensitivity of the

death rate to income levels (d1 ), and on the sensitivity of income levels to population (a1 ).

The long-run equilibrium level of real income per person can be derived as the income

level that gives a growth rate of population of zero

Nt Nt1 d0 b
= b d0 + d1 Y = 0 Y = (15.15)
Nt1 d1

323
This level of income, as we noted above from the graphical illustration of the model, depends

only on the parameters of the birth and death schedule and not at all on the parameters of

the income-population schedule. So, for example, even if there was an increase in a0 so that

people could be paid more wages for each level of population, this would result, over time,

only in higher population rather than higher income levels. Income levels depend negatively

on birth rates, positively on death rates and negatively on the sensitivity of death rates to

income levels.

A final way of illustrating the convergent dynamics of the model is to note that equation

(15.12) for population growth can be re-written as

!
Nt Nt1 b d 0 + d 1 a0
= (d1 a1 ) Nt1 (15.16)
Nt1 d 1 a1

Using the formula for N in equation (15.14), this becomes

Nt Nt1
= (d1 a1 ) (N Nt1 ) (15.17)
Nt1

In other words, the growth rate of population is determined by how far population is from

its equilibrium level, with the speed of adjustment to this equilibrium, d1 a1 , determined by

the sensitivity of income levels to population and the sensitivity of the death rate to income

levels.

How the Malthusian Economy Responds to Shocks

Finally, we consider three kinds of shocks to the Malthusian economy. In each case, we

assume the economy starts at an equilbrium with population of N0 and income levels Y0 .

First, consider an increase in d0 which shifts the death rate schedule up. Figure illustrates

what happens: At the starting level of income, Y0 , death rates now start to exceed birth rates.

324
Population falls and income rises until we reach the new higher equilibrium level of income

Y1 with its corresponding lower level of population N1 .

Figure illustrates the consequences of an increase in the birth rate, b. This shock works in

the opposite fashion to the death rate shock.

Finally, Figure illustrates the consequences of a once-off increase in technology, i.e. an

increase in a0 so that people are able to earn more money at each level of population. The

initial response to this shock is higher income levels. However, these higher income levels

reduce the death rate and, over time, income levels return to their original equilibrium level.

While income levels return to their original level, population is permanently higher because

the new level of productivity permits a higher level of population than the old level.

There is an interesting contrast here between what happens when there is technological

progress in the Solow model and when technology improves in the Malthusian model. The

difference relates to the assumption in the Solow model that there is a consistent and non-

trivial pace of technology increase. In the Malthusian model, the instantaneous effect of an

increase in efficiency is an improvement of living standards. But this is offset over time by

population increases if there arent any further increases in technology.

In the Solow model, technology keeps increasing and keeps pushing up incomes every

period, so the population can steadily increase without pushing income levels down. Greg

Clark argues that while, cumulatively, there was a large increase in technology from ancient

times to 1800, the pace of this increase was never fast enough to prevent population growth

eroding its effects on living standards, so that prior to the Industrial Revolution, improvements

in productive efficiency only translated into higher population.

325
 Figure 15.9: A Shift in the Death Rate Schedule

BIRTH
AND
DEATH
RATE
BIRTH RATE

DEATH RATE NEW

DEATH RATE OLD

Y0 Y1 INCOME PER PERSON

POPULATION

N0

N1

Y0 Y1 INCOME PER PERSON

326
 Figure 15.10: A Shift in the Birth Rate Schedule

BIRTH
AND
DEATH BIRTH RATE NEW
RATE
BIRTH RATE OLD

DEATH RATE

Y1 Y0 INCOME PER PERSON

POPULATION

N1

N0

Y1 Y0 INCOME PER PERSON

327
 Figure 15.11: An Increase in Technological Efficiency

BIRTH
AND
DEATH
RATE
BIRTH RATE

DEATH RATE

Y0 INCOME PER PERSON

POPULATION

N1

N0

Y0 INCOME PER PERSON

328
Malthus on the Poor Laws

The Malthusian model is one in which our usual understanding of what is good and what is

bad is turned on its head. Things that we think are good, such as people living londer, turn

out to be bad for average living standards, and things that we think are bad, like plagues and

diseases, have a positive effect on those who survive. This non-intuitive worldview translated

into Malthuss own policy recommendations. For example, he argued strongly against poor

laws that provided assistance to the poor:

The poor laws of England tend to depress the general condition of the poor in these

two ways. Their first obvious tendency is to increase population without increasing

the food for its support. A poor man may marry with little or no prospect of

being able to support a family in independence. They may be said therefore in

some measure to create the poor which they maintain, and as the provisions of the

country must, in consequence of the increased population, be distributed to every

man in smaller proportions, it is evident that the labour of those who are not

supported by parish assistance will purchase a smaller quantity of provisions than

before and consequently more of them must be driven to ask for support.

Secondly, the quantity of provisions consumed in workhouses upon a part of the

society that cannot in general be considered as the most valuable part diminishes the

shares that would otherwise belong to more industrious and more worthy members,

and thus in the same manner forces more to become dependent. If the poor in the

workhouses were to live better than they now do, this new distribution of the money

of the society would tend more conspicuously to depress the condition of those out

of the workhouses by occasioning a rise in the price of provisions.

329
Over the years, Malthus has often been criticised for being overly-pessimistic about the fate of

mankind and for opposing socially-progressive policies. However, it is worth noting the date

that he wrote his famous essay1798. Up until the time that he wrote his essay, his version

of how the world worked actually described the economy remarkably well. It was only after

his book was written that technological progress became fast enough to render his analysis

less relevant.

330
Chapter 16
Malthus and the Environment
The Malthusian model may seem of interest today only for the light that it sheds on how

the world worked before the Industrial Revolution ushered in an era of growth and increasing

prosperity. Recall, however, that Malthuss views on how rising population reduced living

standards focused on how increasing numbers of people placed pressures on the allocation

of scarce resources, particularly food. In a world in which global population has just passed

7 billion, up from 4 billion in 1960 and 2 billion in 1927, it is reasonable to ask whether

important global resources, such as energy sources, agricultural land and the global resource

of a stable climate, can continue to withstand the strain of increasing population.

In these notes, we will study a model that combines a Malthusian approach to population

dynamics with an approach to modelling changes in a renewable resource base, which can

expand or contract. The model was first presented by James A. Brander and M. Scott Taylor

in their 1998 American Economic Review paper The Simple Economics of Easter Island: A

Ricardo-Malthus Model of Renewable Resource Use.

Easter Island

On Easter Sunday 1722, a Dutch explorer called Jacob Roggeveen came across a Pacific

island that is believed to be the most remote inhabitable place in the world. Situated over

331
two thousand miles west of Chile (see Figure 1) it is about 1300 miles east of its nearest

inhabited neighbour, Pitcairn Island. Known as Easter Island since Roggeveens brief visit,

the island its inhabitants called Rapa Nui has had a long and fascinating history.

There is no written history of events at Easter Island prior to Roggeveens visit so we are

relying on the interpretation of archeological evidence to reconstruct what happened a long

time ago. The interpretation Im passing on in these brief notes comes from my reading of

a chapter in Jared Diamonds book, Collapse, but there are archeologists and scientists who

disagree with some aspects of this story.

Easter Island was probably first populated sometime around 900 AD. That it was ever

populated, given its remoteness, is somewhat extraordinary. It seems likely that, once pop-

ulated, it had little (and possibly no) contact with the outside world. The most remarkable

feature of the island is its collection of hundreds of carved ceremonial statues featuring torsos

and heads (see Figures 2 and 3) which were mainly built between 1100 and 1500. The natives

most likely erected the statues as a form of religious worship. Evidence suggests that the

island was divided into twelve tribes and they competed with each other (perhaps for local

pride, perhaps for favour with the gods) by building larger and larger statues over time.

The statues were enormous. On average, they were 12 feet high and weighed 14 tons,

while the largest weighs 82 tons. There is plenty of evidence to show that the statues required

huge resources and that at least some of these resources were organised on a shared basis by a

centralised leadership. Large teams of carvers were needed to create the statues and as many

as 250 people were required to spend days transporting the statues around the island. When

first populated, the island had large amounts of palm trees which supplied the resources for

canoes, for tools for hunting and for materials used to transport the statues (sleds, rope, levers

332
etc.) Estimates of peak population vary but it appears that the population peaked at about

15,000 in the early 1600s.

By the time Europeans began to visit the island one hundred years later, however, the

island was largely deforested and population seemed to be as low as 3,000. Without palm

trees, the islanders no longer had materials with which to build good canoes and this limited

their abilities to catch fish. Without forests, the island lost most of its land birds, which had

been an important source of meat. By the 1700s, the population survived mainly on farming,

with chickens the main source of protein, but deforestation had also reduced water retention in

the soil and lead to soil erosion (the island is quite windy) so agricultural yields also declined.

Statue building had ceased by the early 1600s: Many of the statues remain today in various

states of completion at the quarry at Rano Raraku where they were carved. Archeological

evidence shows increasing numbers of spears and daggers appearing around this time, as well

as evidence of people starting to live in caves and fortified dwellings. By the time Europeans

arrived in the following century, tensions over food shortages had spilled over into intra-tribal

rivalries with tribes knocking over the statues of their rivals. By the mid-1800s, all the statues

had been toppled, so todays standing statues have been put in place in modern times.

There are many gaps in our understanding of what happened at Easter Island but the

basic story appears to be that the population expanded to the point where the islands re-

sources began to diminish and once population started to decline, the island went into a

downward spiral. By the time Europeans visited in the seventeenth century, both population

and resources had been greatly diminished from their peak levels.

The model laid out over the next few pages provides a description of how this can happen.

We conclude with some thoughts about why it was allowed to happen and the potential

333
implications for current global environmental problems.

Figure 16.1: The Worlds Most Remote Place

334
Figure 16.2: Easter Island Statues

335
Figure 16.3: Some Standing, Some Toppled

336
The Model

The model economy consists of population of Nt people at time t, who sustain themselves

by collecting a harvest, Ht from a renewable resource stock denoted by St . Think of St as

equivalent to a forest, or a herd of animals, or a stock of fishes; more realistically, think of it

as the combination of a set of different resources of this type.

The model consists of three elements. The first element describes the change in population:

This depends positively on the size of the harvest (a bigger harvest means less deaths and

perhaps more births) and on an exogenous factor d > 0 such that without a harvest, there is

a certain percentage reduction in population.

dNt
= dNt + Ht (16.1)
dt

The next element describes the harvest. We assume that the harvest reaped per person is a

positive function of the size of the resource stock.

Ht
= St (16.2)
Nt

The final element, describing the change in the resource stock, is perhaps the most important.

We are describing a resource stock that is renewable. It doesnt simply decline when harvested

until it is all gone. Instead, it has its own capacity to increase. For example, stocks of fish

can be depleted but will increase naturally again if fishing is cut back. So, our equation for

the change in resources is

dSt
= G (St ) Ht (16.3)
dt
The second term on the right-hand-side captures that the resource stock is reduced by the

amount that is harvested. The first element is more interesting. It describes the ability of

the resource to grow. Brander and Taylor use a logistic function to describe how the resource

337
stock renews itself

G (St ) = rSt (1 St ) (16.4)

This equation can be interpreted as follows. The maximum level of resources is St = 1: At this

level, there can no further increase in St . Also, if St = 0 so the resource base has disappeared,

then it cannot be regenerated. For all levels in between zero and one, we can note that

G (St )
= r (1 St ) (16.5)
St

So the amount of natural renewal as a fraction of the stock decreases steadily as the stock

reaches its maximum value of one. This means that if the stock gets very low, it can grow at

a fast rate if there is limited harvesting. However, if the stock is starting from a low base, the

absolute size of this increase may still be small.

Dynamics of Population

We are going to describe the dynamics of this model using what is known as a phase diagram,

which is a diagram that shows the direction in which variables are moving depending upon

the values that they take. In our case, we are going to describe the joint dynamics of Nt and

St .

Inserting the equation for the harvest, equation (16.2), into equation (16.1) for population

growth, we get
dNt
= dNt + St Nt (16.6)
dt
This equations shows us that the change in population is a positive function of the resource

stock. This means there is a particular value of the resource stock, S , for which population

growth is zero. When resources are higher than S population increases and when it is lower

338
than S population declines. The value of S can be calculated as the value for which the

change in population is zero meaning

d
dNt + S Nt = 0 S = (16.7)

The resource stock consistent with an unchanged population depends positively on the exoge-

nous death rate of the population, d, and negatively on the sensitivity of the population to the

size of the harvest, , and on which describes the productivity of the harvesting technology.

Figure 16.4 shows how we illustrate the dynamics with a phase diagram. We put population

on the x-axis and the stock of resources on the y-axis. Unchanged population corresponds to

a straight line at S . For all values of resources above S population is increasing: Thus in

the area above the line, we show an arrow pointing right, meaning population is increasing.

In the area below this line, there is an arrow pointing left, meaning population is falling.

Dynamics of Resources

The dynamics of resources are derived by substituting in the logistic resource renewal function,

equation (16.4), and the equation for the harvest, equation (16.2), into equation (16.3) to get

dSt
= rSt (1 St ) Nt St (16.8)
dt

The stock of resources will be unchanged for all combinations of St and Nt that satisfy

r (1 St )
rSt (1 St ) Nt St = 0 Nt = (16.9)

This means that there is downward sloping line in N S space along which each point is a

point such that the change in resources is zero. This line is shown on Figure 16.5. The upper

point crossing the S axis corresponds to no change because S = 1 and there are no people; as

339
we move down the line we get points that correspond to no change in the stock of resources

because while there are progressively larger numbers of people, the stock gets smaller and so

can renew itself at a faster pace.

Remembering that equation (16.8) tells us that the change in the stock resources depends

negatively on the size of the population, note now that every point that lies to the right of
dS
the downward-sloping St = dt
= 0 line has a higher level of population than the points on

line. That means that the stock of resources is declining for every point to the right the line

and increasing for every point to the left of it. Thus, in the area above the downward-sloping

line on Figure 5, we show an arrow pointing down, meaning the stock of resources is falling.

In the area below this line, there is an arrow pointing up, meaning the stock of resources is

increasing.

The Joint Dynamics of Population and Resources

In Figure 16.6, we put together the four arrows drawn in Figures 16.4 and 16.5. This phase

diagram shows that the joint dynamics of population and resources can be divided up into

four different quadrants.

We can also see that there is one point at which both population and resources are un-

changed, and thus the model stays at this point if it is reached. We know already from

equation (16.7) that the level of the resource stock at this point is S = d

. We can calcu-

late the level of population associated with this point by inserting this formula into equation

(16.9):  
d
r 1 r ( d)
N = = (16.10)
2
This level of population depends positively on r (so faster resource renewal raises population)

340
and on (the sensitivity of population growth to the harvest) and negatively on d (the

exogenous death rate coefficient).

This point is clearly some kind of equilibrium in the sense that once the economy reaches

this point, it tends to stay there. But is the economy actually likely to end up at this point?

The answer is yes: From any interior point (i.e. a point in which there is a non-zero population

and resource stock) the economy eventually ends up at (N , S ). Its beyond the scope of this

class to prove formally that this is the case (the Brander-Taylor paper goes through all the

gory details) but I can note that, after messing around with the equations, one can show that

1 dNt
= (St S ) (16.11)
Nt dt
1 dSt
= (Nt N ) + r (St S ) (16.12)
St dt

so the dynamics of both population and the resource stock are both driven by how far the

economy is from this equilibrium point.

Harvesting and Long-Run Population

What does changing the parameter (which determines the fraction of the resources that

is harvested) do to the equilibrium level of population? There are two different effects. On

the one hand, a higher means a smaller amount of people consume the natural growth

in resources that occurs in steady-state this would tend to reduce the sustainable level

of population. On the other hand, the smaller stock of resouces associated with the higher

value of implies a higher harvest which could sustain more people. We can calculated the

derivative of the equilibrium level of population with respect to as follows

dN r 2rd
= 2+ 3 (16.13)
d

341
 !
1 2d
= 2 1 (16.14)

r
= 2 (2S 1) (16.15)

This shows that whether an increase in raises or reduces the equilibrium population depends

on the size of the equilibrium level of resoures. If the equilibrium level of resources is over half

the original maximum amount (which we have set equal to one) then we have 2S 1 > 0

and a more intensive rate of harvesting raises the population even though it reduces the total

amount of resources. On the other hand, if the equilibrium level of resources is less than half

the original maximum amount (which we have set equal to one) then we have 2S 1 > 0

and a more intensive rate of harvesting reduces the population.

An economy like Easter Island, where the economy ended up with a hugely diminished

amount of resources, likely corresponds to the latter case, so it was an example of an economy

that would have had a higher long-run population if they had harvested less.

Back to Easter Island

Lets go back to Easter Island and imagine the island in its early days with a full stock of

resources and very few residents. What happens next? Figure 16.7 provides an illustration.

The economy starts out in what we can call the happy quadrant with resources above

the long-run equilibrium and an expanding population. How do we know the dynamics take

the curved form displayed in Figure 16.7? Well, when the economy crosses into the bottom

right quadrant, in which population is now falling, the economy doesnt suddenly jump off

in a different direction; the models equations dont allow for any sudden jumps. Thus, the

turnaround from increasing population to falling population must occur gradually over time.

342
 So what happens to our theoretical Easter Island?

For many years, the population expands and resources decline.

Then, when it moves into the bottom right quadrant, population falls and resources

keep declining.

Then the economy moves into the bottom left quadrant where population keeps falling

but resources finally start to recover.

Then the economy moves into the quadrant in the triangle under the two curves and

population starts to recover and resources increase.

Finally, the economy moves back into the quadrant where it started but with less pop-

ulation and lower resources. The process is repeated with smaller fluctuations until it

ends up at equilibrium with S = S and N = N

Our theoretical Easter Island sees its population far overshoot its long-run equilibrium

level before collapsing below this level and then oscillating around the long-run level and then

finally settling down.

343
Figure 16.4: Population Dynamics

344
Figure 16.5: Resource Stock Dynamics

345
Figure 16.6: Dynamics Differ In Four Quadrants

346
Figure 16.7: Illustrative Dynamics Starting from Low Population
and High Resources

347
Numerical Example: A Lower Harvesting Rate

One of the ways to explore the properties of models like this one is to use software like Excel

or more programming-oriented econometric software like RATS to simulate discrete-time

versions of the model. Figures 16.8 and 16.9 show time series for resources and population

generated from a RATS programme that simulates a discrete-time adaptation of the model.

The programme is shown at the back in an appendix. It implements a version of the model with

parameter values d = 0.075, r = 0.075, = 2, = 0.1 and an initial population of N1 = 0.0001.

The parameter values are set so that the equilibrium level of resources is S = d

= 0.075
0.2
=
r(1S ) (0.075)(10.375)
0.375 while the equilibrium level of population is N =
= 2
= 0.023475.

Figure 16.8 shows that, for these parameter values, the stock of resources falls to about

half of its long-run equilibrium value, then rises and overshoots this value and then oscillates

before settling down at this equilibrium level. Figure 16.9 shows the associated movements

in population. We see population surge to levels that are over twice the long-run sustainable

level, then dramatically drop to undershoot this level before eventually settling down.

Because we have chosen a base case in which S < 0.5, this is a case where there would be

higher resources and population in the long-run if we had a somewhat lower rate of harvesting.

Indeed, you can pick a rate of harvesting that avoids a collapse scenario altogether. Figures

16.10 and 16.11 compare the base case we have just looked at with a case in which the rate

of harvesting was 40 percent lower, so = 1.2. In this case, the resource stock only slightly

undershoots its long-run level and the population only slightly-overshoots. The economy ends

up a similar level of population but arrives there in a less dramatic fashion.

348
 Figure 16.8: Resources in a Simulated Easter Island Economy

1.0

0.9

0.8

0.7

0.6

0.5

0.4

0.3

0.2
50 100 150 200 250 300 350 400 450 500

349
 Figure 16.9: Population in a Simulated Easter Island Economy

0.05

0.04

0.03

0.02

0.01

0.00
50 100 150 200 250 300 350 400 450 500

350
 Figure 16.10: Resource Stock with Less Harvesting

1.0

0.9

0.8

0.7

0.6

0.5

0.4

0.3

0.2
50 100 150 200 250 300 350 400 450 500
Base Case Less Harvesting

351
 Figure 16.11: Population with Less Harvesting

0.05

0.04

0.03

0.02

0.01

0.00
50 100 150 200 250 300 350 400 450 500
Base Case Less Harvesting

352
Why Doesnt Someone Shout Stop?

The pattern demonstrated in the modelin which the economy far overshoots its long-run

level before collapsing to an equilibrium with lower population and depleted resourcesmay

seem to fit what happened at Easter Island. But it raises plenty of questions: Why did the

residents of the island allow this to happen? Why didnt they establish better governance

rules to prevent the deforestation that proved so devestating? And could this model possibly

be a warning that todays global economy could represent an overshooting with a significant

collapse awaiting us all?

In his book, Collapse, Jared Diamond discusses Easter Island and a number of other cases

in which societies saw dramatic collapses, many triggered by long-term environmental damage.

Diamond points to a number of potential explanations for why societies can let environmental

damage occur up to the point where they trigger disasters.

The Tragedy of the Commons: It may simply never be in anyones interests at any

point in time to prevent environmental degradation. A fisherman may acknowledge that

excess fishing will eventually put him out of business but there may be little he can do

to prevent others fishing and today he needs to earn an income. Some societies can put

in place centralised political institutions to prevent environmental disasters and some

cannot. At present, the society called The Earth is not known for its efficient centralised

political decision making.

Failure to Anticipate: Societies may not realise exactly how much damage they are

doing to their environment or what its long-term consequences will be. Up until the

point at which Easter Islands environment failed to support a growing population,

there was probably a limited realisation among the population of the damage being

353
 done. Once the population began to shrink and the tribes turned against each other

(theres some evidence of cannibalism during this period) the likelihood of a common

negotiated solution to cut down less trees to preserve the environment was unlikely.

Similarly today, the future effects of climate change are unpredictable and the costs

(and even potential benefits) may be unevenly distributed.

Failure to Perceive, Until Too Late: Diamond notes that environmental change

often occurs at such a slow pace that people fail to notice it and plan to deal with it.

The Easter Islanders of 1500 probably couldnt remember (and certainly had no written

record of) their island being covered in palm trees. The islander who eventually cut

down the last tree probably had little idea that these trees had once been the mainstay

of the local economy. Similarly, global climate change has occurred at such a slow pace

that, despite the mountain of scientific evidence that it is real, many simply choose to

deny it.

354
Appendix: Programme For Easter Island Simulation

Figures 8 and 9 were produced using the programme below. The programme is written for the

econometric package RATS but a programme of this sort could be written for lots of different

types of software including Excel.

allocate 10000
set d = 0.075
set r = 0.075
set gamma = 2
set theta = 0.1
set s = 1
set n = 0.0001
set h = 0
do k = 2,10000
comp s(k) = s(k-1) + r(k)*s(k-1)*(1-s(k-1) ) - h(k-1)
comp h(k) = gamma(k)*s(k-1)*n(k-1)
comp n(k) = (1-d(k))*n(k-1) + theta(k)*h(k)
end do k
graph 1
# s 1 500
graph 1
# n 1 500

355