Unit : Economic Growth

Lecture

EMAC

February
 Outline

Economic convergence
Absolute convergence
Conditional convergence
Club convergence
Reasons for di ferences in income and growth over time and space
Income di ferentials
Growth di ferentials
The golden rule of savings
Growth accounting
Simulation using Solow model
Endogenous growth models
Productive externalities
The AK model
Endogenous growth models v. exogenous growth models
In closing
 Objectives/outcomes for U

* Use Solow models of economic growth to analyse the determinants of di ferences in output levels and
output growth rates across time and space.
* Describe economic convergence
* Analyse the e fects of changes in the savings rate on economic welfare
* Contrast the assumptions and predictions of endogenous and exogenous growth models
* Use macroeconomic data to perform a growth accounting exercise
* Use macroeconomic data to analyse cross-country growth
 Literature

* Sorensen & Whitta-Jacobsen (SW): Ch. . . , . & . & ., . - .

/ Economic convergence
 Economic convergence

* Countries with low initial income levels will grow faster than countries with high initial income levels.
* Baumol ( ) observed that countries with low initial GDP/worker levels have high subsequent growth rates,
leading to the convergence of income/worker over time (F . )
* Based on a sample of OECD countries
 Absolute convergence

* Absolute convergence hypothesis: the LR growth path of yi is the same for all countries ... all have same
growth path ... in very LR, all countries will have same y
* Implication: Average annual growth rate of low income countries is always faster than that of high-income
countries
* To test this hypothesis econometrically: (i) Estimate gi = β0 + β1 y0i . (ii) Then, test: H1 : β1 < 0
yT −yt
* Note that gy = T−t
* Graphically: looks like F . (note: draw gy against y0 )
* Note: if true, no scope for foreign aid (only to help very worst-o f or to help speed up movement toward world
growth path)
* This hypothesis does not hold: see F . and work by De Long ( ) – sample selection bias (only OECD
countries). Once sample is extended (F . ), no negative relationship
Absolute convergence: diagram (F . - )
* Despite absolute convergence hypothesis being rejected, how does one explain F . ?
 Conditional convergence

* Conditional convergence hypothesis: yi converges to a country-specific (i) long-run growth path, which is
determined by a country's structural characteristics (in Solow model: s, n, δ, g).
* Countries with the same/similar structural characteristics have the same long-run growth paths ...
* Implication: for countries with same structures, low income countries will grow faster than high-income
countries (F . , F . )
* and the further a country lies from its growth path, the faster its subsequent growth rate will be
* It can be shown (SW, . ) that speed of convergence is λ = (1 − α)(s + g + δ)
* To test this hypothesis econometrically: (i) estimate gi = β0 + β1 y0i + β2 si + β3 ni + ..... (ii) Then, test
H1 : β1 < 0
* Note: scope for foreign aid = structural adjustment... use aid to improve s, lower n or increase g...
Conditional convergence: modified Solow diagram (CS, F . )
Conditional convergence: modified Solow diagram
Conditional convergence: modified Solow diagram
Conditional convergence: modified Solow diagram
Conditional convergence: modified Solow diagram
Conditional convergence: modified Solow diagram
 Discussion: diagram

* Let di ferences in structural characteristics be captured by di ferent savings rate (sH and sL)
* While xH and xL denote the steady-states for sH and sL, respectively
* For countries with same structural characteristics, low initial income (khat) associated with faster
subsequent growth
* A and C have same structure, A grows faster
* E, D and B have same structure: D grows faster than B, while E grows faster than D
* Two countries with the same income level (khat) can grow at di ferent rates
* A and D – same income level (khat), but A has higher savings rate, so A grows faster than D
* C and B – same khat, but B has lower savings rates, so B grows slower than C
* Country with higher income (khat) can grow faster than country with lower income (khat), if high income
country has more favourable structure (e.g. higher s)
* A has higher income than E, but has “better” structure (sH > sL, so A grows faster than E
* Under conditional convergence, only the structural characteristics of a country determines its LR growth path
...
* But what if the initial level of income + structural characteristics determines growth path??
 club convergence

* Club convergence hypothesis: yi converges to i specific LR growth path, which depends on structural
characteristics (s, n, δ, g) and whether y0 lies above or below an income-threshold
* Countries with same/similar structure, that have y0 > ythreshold : high LR growth path ...
* Countries with same/similar structure, that have y0 < ythreshold : low growth path ...
* Countries with same structure could have di ference LR growth paths, because the lie on di ferent sides of income
threshold ... re lects history (in part) ...
* Test: Split countries into groups: high and low initial income (yH0 and yL0 ). Then, estimate
gHi = β0H + β1H yH0 + ... and gLi = β0L + β1L yL0 + .... Then, test H1H : β1H < 0 and H1L : β1L < 0
* Under club convergence, much greater scope for foreign aid: improve structural characteristics and use aid to
push country over threshold ...
Club convergence: growth path diagram
Club convergence: growth path diagram
Club convergence: growth path diagram
Club convergence: modified Solow
Club convergence: modified Solow
 Club convergence: modified Solow

* Based on: multiple equilibria/poverty trap literature

* Below threshold – move to steady-state L; above threshold – move to steady-state H
* How do A and B end up at L? How do C and D end up at H?
Solow diagram on which the previous modified Solow is based
/ Reasons for di ferences in income and growth over time and space
 Reasons for income di ferentials: Solow model

( )α/1−α ( )α/1−α
s s
* In steady-state, ŷ∗ = n+g+δ , while y∗ = A n+g+δ

* So, country y in country A will be greater than y in country B if:

* Country A has higher savings rate and/or
* Country A has lower population growth and/or
* Country A has lower depreciation rate and/or
* Country A has more advanced of technology (maybe due to faster technological progress)
* Remember: g = ȦA and A = A0 egt
* So, increase in g will lead to lower ŷ∗ , but A will be much higher, so that y∗ = Aŷ∗ will be higher

Use Solow diagrams (one for savings, one for population/depreciation, and then another Solow, drawn in the y-k
space) to compare the steady-state outcomes of A with the steady-state outcomes of B ...
Why is SA income higher?
Why is SA income lower?
 Reasons for growth di ferentials: Solow model

* At a particular point in time, countries can have di ferent growth rates for y
* These di ferences can be permanent or temporary
* Of course, when studying LR growth, temporary di ference is in the eye of the beholder...
* Permanent di ference: Two countries have di ferent technological growth rates (one with higher g has higher
growth rate of income per worker)
* Because in steady-state, gy = g
Could show this using the Solow diagram drawn in the y − k space following an increase in g... see Lecture slide
deck... Countries A and B start o f with same rate of g. Then country A's g increases
 Temporary growth di ferentials: convergence

* If two countries have the same structural characteristics, the one with the lower initial income level will grow
faster; if two countries have di ferent structural characteristics, they will grow at di ferent rates, even if their
initial income levels are the same, because they are on di ferent growth paths
* If country A has more favourable structural characteristics than country B, it may grow faster, even if it has a higher
income level, because it may lie further from its growth path's steady-state than B;
* or, A and B have the same structural characteristics, but A has a lower initial income level, which means that
country A will grow faster
Can show this using a modified Solow diagram [similar to the one used to show conditional convergence. But only
need to discuss/compare A, D and C, for example]
 Temporary growth di ferentials: steady-state transition

* If s or n changes, the economy transitions from an old to new steady-state (k̂ and ŷ is higher if s is higher or n
is lower)
* In steady-state, k̂ and ŷ grow at %. But during transition to new steady-state, k̂ and ŷ grow at non-zero rate:
gŷ = αgk̂
* A country transitioning to more favourable steady-state will grow faster than country who is already close to
its steady-state (or, a country transitioning to less favourable steady-state will have lower growth rate than
country close to its steady-state)
* Also, since y = Aŷ ⇒ gy = g + gŷ = g + αgk̂ ... income per worker will grow faster
Can show this using a modified Solow diagram [similar to F . , increase in s – also see L slides. Set-up: both
countries have same steady-state and structure, so that their level of income and capital is the same in
steady-state. And also growth. But now, suppose that country A's savings rate increases. Even with same income
level, A will grow faster. And, over time, even with the higher income level, A will grow faster, until it has reached
it's new steady-state. Only then will A and B grow at same rate again].
/ The golden rule of savings
 Golden rule of savings

* In Solow model: higher s leads to higher k and y. Should policymakers aim for s = 1 (max savings rate)?
* Suppose policymakers wish to maximise the welfare of their citizens ... suppose welfare/utility =
consumption
* Recall that C = Y − S = Y − sY
* If s = 1 ⇒ C = 0, if s = 0, k → 0 over time
* Maximising welfare entails finding the savings rate that maximises consumption
* This implies s ̸= 0, and s ̸= 1
* Between 0 and 1 is sG , the level that will maximise ĉ, ĉG
 Obtaining/deriving the golden rule savings rate

ĉ∗ = (1 − s) ŷ∗
( ) 1−α
α
s
∴ ĉ∗ = (1 − s) n+g+δ
( ) 1−α
α
s
ln (ĉ∗ ) = ln (1 − s) + ln n+g+δ
( ) ( )
s
ln (ĉ∗ ) = ln (1 − s) + 1−α α
× ln n+g+δ
( )
ln (ĉ∗ ) = ln (1 − s) + 1−α α
[ln (s) − ln (n + g + δ)]
( )( )
d ln (ĉ∗ ) −1 α 1
ds = 1−s + 1−α s
∗
( )( )
∗ d ln (ĉ ) −1
Max ln (ĉ ): ds = 0 ⇒ 1−s + 1−α α 1
s =0
( )( )
∴ 1−α α 1
s = 1−s
1

∴ α 1
s(1−α) = 1−s
s(1−α)
∴ α = (1 − s) ⇒ s (1 − α) = α (1 − s)
s ∗∗
∴ 1−s = 1−α ⇒ s
α
= sGR = α
 in more general terms

* In Solow diagram: steady-state golden-rule k̂, k̂∗∗ , is the level of k̂ at which the slope of k̂α = slope of
(n + g + δ)k̂
* Because slope of production function is MPK, this means that ∂ ŷ = αk̂α−1 = (n + g + δ) at the golden
∂ k̂
rule level of capital
* So, the return to capital is MPK = r = (n + g + δ), to max ĉ
Golden rule: Solow diagram
Golden rule: Solow diagram
Golden rule: Solow diagram
Golden rule: Solow diagram
Golden rule: Solow diagram
Golden rule: Solow diagram
Golden rule: Solow diagram
Golden rule: Solow diagram
Golden rule: Solow diagram
Golden rule: Solow diagram
Golden rule: Solow diagram
Golden rule: Solow diagram
 Tips on drawing this diagram

* First, draw sk̂α and (n + g + δ)k̂

* Find k̂∗∗ (at point A)
* Now, draw line parallel to (n + g + δ)k̂, going through A′
* Draw production function (y = k̂α ) which is tangent to this parallel line at A′
* Di ference A′ − A = ĉ∗∗ , which is maximised at k̂ = k̂∗∗ , which is the case if s = s∗∗
* At any other s, lower ĉ
/ Growth accounting
 Growth accounting (SW, . )

* In a growth accounting exercise, our goal is to decompose output growth into its sources: capital growth,
labour growth and TFP growth
* Suppose Yt = Bt Kαt Lt
1−α

* Then, ln Yt = ln Bt + α ln Kt + (1 − α) ln Lt
* Subtracting ln Yt−1 = ln Bt−1 + α ln Kt−1 + ln(1 − α)Lt−1 (or, di ferentiating w.r.t. t) yields:
gY = gB + αgK + (1 − α)gL
* We mostly don't observe TFP growth ... But we can calculate it as the so-called Solow residual:
gB = gY − αgK − (1 − α)gL
1/(1−α)
* If At ≡ Bt , then Yt = Kα t (At Lt )
1−α

* log: ln Yt = α ln Kt + (1 − α) ln(At Lt )
* Subtract lagged log (ln Yt−1 ) or di ferentiate w.r.t. t, and then gY = αgK + (1 − α)(gA + gL )
* Don't observe A. So, calculate A as residual...
gB
* Also note the relationship between gA and gB : gB = (1 − α)gA , while gA = 1−α
* We can also perform growth accounting on output per worker
* If yt = Bt kα
t , then ln yt = ln Bt + α ln kt
* If we subtract lagged log, or di ferentiate w.r.t. to time: gy = gB + αgk (see T . )
/ Simulation using Solow model
 Simulation with Solow model

* Use the discrete-time general Solow model

* Eq. - , p. ; Eq - (p. - ); Eq. - (p. - )
* NB examples:
* SW, p. - (T . , F . )
* Exercises . - . (p. )
* Exercise . (p. -)
* Set-up: economy is in steady-state from period to period a
* In period b (a+ ), one of the parameters s, n, g or δ changes
* Simulate values of k̂ ŷ, k, y etc. for period to period T
* Since the economy is in steady-state for period to a, use the steady-state solution to find the values of the
variables from to a (in steady-state, actual values = steady-state values!)
* From period b to T, the economy is no longer in steady-state (steady state and actual values no longer the same)
 Running the simulation: periods b to T

* To obtain the values of the variables from period b:

* First, find values of A for b to T (if g changes, set g equal to its new value from b onward)
* Then, find values of k̂ (use Solow equation or transition equation)
* Then, find values of ŷ (use production function)
* Then, find other ”hat” variables (e.g. ĉ)
* Only try to find values of other variables (e.g. y, k , c , w) once you've found the values of A, k̂, ŷ
/ Endogenous growth models
 Productive externalities & increasing returns to scale (SW, . )

* If Y = BKα L1−α , then production function exhibits constant returns to scale and decreasing returns to
capital (and capital per worker
* Implication: Output growth due to capital accumulation (due to increased savings) can't be sustained, due to
diminishing returns
* But, if there were increasing returns to scale in Y = Kα Lβ , so that α + β > 1 ...
* Suppose that an individual firm's production is given by Y = Kα (AL)1−α
* Also suppose that A = Kϕ , ϕ ≥ 0
* Then, aggregate production is Y = Kϕ(1−α)+α L1−α
* Notice that ϕ(1 − α) + α + (1 − α) = ϕ(1 − α) + 1 ≥ 1
* Now have constant returns at firm level, and increasing returns at aggregate level (wages and return to
capital still given by marginal products; individual firms are still price takers)
 Learning-by-doing

* Lots of empirical studies suggest that ϕ > 0, so that aggregate production exhibits IRS
* The fact that ϕ > 0 implies that there is a productive externality associated with use of capital ... Firm gains
benefit of installing new capital; and there are spill-overs to other parts of the economy
* One mechanism: learning-by-doing: as firms use more capital, workers learn how to use this capital (skills
⇑). But this knowledge is not just limited to firm installing capital... Workers change firms ...
 Solow model with productive externalities (SW , p. )

Below follows the continuous time version of the discrete time model presented on p.
* Y = Kα (AL)1−α
* A = Kϕ
L̇
* n= L
* K̇ = sY − δK
 Endogenous growth and the AK model

* Suppose that ṅ = 0, while ϕ = 1

* Allows us to focus on factors other than population growth for growth in y, and allows us to examine models that
never truly converge to steady-state (see S . in SW)
* With these assumptions in place, the model becomes:
* A=K
* Y = Kα (KL)1−α = Kα × K1−α × L1−α = K(α+1−α) L1−α = KL1−α ≡ AK
* where A ≡ L1−α
* Note: This A is time-invariant (constant), because L is constant ... In the AK-model, A is not the technology variable, but is
chosen for notational simplicity ... unfortunately ...
 The AK model

Per worker production = Y/L = y = Ak

* Solow equation: k̇ = sAk − δk
k̇
* Modified Solow equation: k = sA − δ = ge
* Note that both of these equations are linear equations ... no diminishing returns to capital ... see F .
* ge is the common, endogenous growth rate of all per worker variables (and technology and wages)
* ge is endogenous, because it is explained by model parameters ...
AK model: Solow diagram
AK model: Solow diagram
AK model: modified Solow diagram
AK model: modified Solow diagram
 Structural policy I

* An increase in s leads to a permanent increase in the level and growth rate of y; lower s permanently
decreases level and growth rate of y
* Because there are constant, not diminishing returns to k
* So, policies to boost savings rate, will now boost level and growth rate of y
* Also, policies to reduce δ will also lead to permanent increase in level and growth rate of y
* Increase e ficiency of investment... tax treatment of investment ...
* But ... An increase in A will also increase ge
* And A ≡ L1−α
* Implication: increase ge by increasing level of population; increase ge explosively if n > 0
* Problem: no empirical support ... countries with higher population growth rates tend to have lower levels of per
capita income growth (see F . )
* Also, if this applies to global output ... We have seen very strong population growth in past years, but output
per worker growth has been relatively constant, has not increased explosively
* This feature of AK model is known as scale e fect.
( K )ϕ
* Can get rid of scale e fect by letting A (technology) depend on k = KL . Specifically, let A = L = kϕ
* Then, ge = s − δ
 Structural policy II

* But now, Y is insensitive to labour input, because e fect of increase in L is o fset by decrease in KL
* Can't get around this problem when setting ϕ = 1
* Therefore, develop alternative models (see Ch. : R& D-based models of growth)
An increase in the savings rate: AK model, Solow diagram (see F . )
An increase in the savings rate: AK model, Solow diagram
An increase in the savings rate: AK model, Solow diagram
An increase in the savings rate: AK model, modified Solow diagram
An increase in the savings rate: AK model, modified Solow diagram
Lower savings rate: AK model, Solow diagram
Lower savings rate: AK model, Solow diagram
Lower savings rate: AK model, Solow diagram
Lower savings rate: AK model, modified Solow diagram
Lower savings rate: AK model, modified Solow diagram
 Endogenous v. exogenous model (SW, . )

* Explaining growth: argument in favour of endogenous models.

Ȧ
* Exogenous (general Solow): g = A
* Endogenous (AK): ge = sA − δ
* Knife-edge argument: argument in favour of exogenous... but not if we instead view ϕ = 1 as an
approximation for large ϕ
* In AK model, truly endogenous growth achieved only by setting ϕ = 1 (as limiting case of large ϕ). But empirical
results show that ϕ tends to lie closer to . than to ... But works OK as approximation...
* Convergence: in simple AK model, no convergence. But allowing for gradual technological transmission
between countries (from tech frontier to others), allows for convergence in endogenous models. No clear
winner (but, if simple AK is our model of endogenous growth, then exogenous is clear winner)
* Scale argument: higher L associated with higher growth in endogenous model, while growing L associated
with explosive growth... no evidence to support this ... argument in favour of exogenous models
* Growth and investment rates: positive relationship between growth and savings/investment rates
* Favours endogenous, but could also be explained by exogenous models if transition to steady-state is low.
* But, no relationship in exogenous models between s and A ... how does one explain F . , then?... note typo in x-axis
– should be ”average investment rate in physical capital”, not ”average labour force growth”
* Therefore, evidence favours endogenous models in this case ...
NB argument in favour of endogenous models (F . , updated + expanded here)
/ In closing
 In closing

* Work through this lecture's homework questions

* Next time: Unit a: Unemployment (SW, ch. - & Marinkov & Geldenhuys, )