The Forces of Economic Growth

A Time Series Perspective

October, 2005

Willi Semmler∗
in collaboration with
Alfred Greiner† and Gang Gong‡

∗
Center for Empirical Macroeconomics (http://www.wiwi.uni-bielefeld.de/∼ semmler/cem/), Uni-
versity of Bielefeld, and Department of Economics, New School University, New York. email:
[email protected].
†
Department of Economics, University of Bielefeld
‡
School of Economics and Management, Tsinghua Unversity, Bejing
Outline:
1. Introduction

2. Methodology: Cross-Country or Time Series Studies

3. Externalities of Investment

4. Education and Economic Growth

5. Knowledge Accumulation and Economic Growth

6. Public Infrastructure Investment

7. Evaluation and Conclusion

2
Questions:
Walt Rostow (1990), ”Theorists of Economic Growth...” and (1960) ”The Stages of Eco-
nomic Growth”

• Are we in the second, third, fourth, of fifth Industrial Revolution?

• Does the massive reallocation of resources from agriculture into manufacturing and
services enhance growth

• Will the gap between the rich and the poor nations widen?

• Will that old devil Diminishing Returns get us in the end?

1
1. Introduction
• Many generations of growth theories have studied the issue of the forces of economic
growth (Classical economists, Harrod-Domar, Solow). Kaldor (1961) has proposed
stylized fact: constant aggregate ratios, constant per capita growth rate, different
growth rates across countries, reallocation of resources. Modern modern growth
theory, building on micro foundations and dynamic optimization of agents (Uzawa,
Lucas, Romer...), has attempted to replicate some of those facts. In particular it
has pursued the question of what forces produce persistent growth rates.

• Empirically, in the spirit of the above, cross-country regressions have explored a

large number of forces of growth

– population growth, investment rates, human capital, openness and compe-

tition, well functioning financial sector, rule of law, economic and political
stability, fiscal policy, attitude toward work, climate and religion, etc...
– but there are methodological perils with this approach, see below

• We take a time series perspective and focus our attention on some major forces of
economic growth: Learning from others (externalities), human capital, knowledge
accumulation, infrastucture investment

– linear models have the problem of scale effects: There may be diminishing
returns to forces of growth
– we thus need models with nonlinearities and estimations of those
– this allows us to study stages of economic growth characterized by different
forces
– we undertake, with newly constructed times series data, empirical studies for
a few countries

• The important questions are: can a time series approach serve as a better guidance
for growth policy? Can this approach better integrate public economics and eco-
nomic growth? Does such an approach aid to identify country specific policies to
improve growth and reduce poverty for specific stages of growth?

2
2. Methodology: Cross-Country or Time Series Stud-
ies
Cross-Country Regressions

ln Yt+1 − ln Yt = α + β1 ln(Y60 ) + β2 ln(n) + β3 ln(I/Y ) + β4 ln(school).... + βi xi ....

• Disadvantages: countries may be at different stages of growth, there may be thresh-

olds in development, there may not be the same preferences and technologies across
countries, there is uncertainty and heterogeneity with respect to the underlying
model and parameters, there is doubt on robust econometric results (see Brock and
Durlauf), policy based on non-robust results

• Advantages: the studies have generated useful knowledge on factors of growth (at
least for similar countries), this approach has data advantage (averaging data)

Time-Series Perspective
• Micro behavior is important, can be allowed to change over time (though embedded
in a macro environment)

• For growth models, with preferences for households, first order conditions derived:
one obtains a system of FOC and state variables

• Nonstationarity requires the model to be transformed into a stationary model

• Scale effects require nonlinearities

• The model needs to be transformed into an estimable form (by substituting out
unobservable variables)

• Empirical estimates of the parameters and test of matching the data (using ML or
GMM estimation with Simulating Annealing).

• Data requirements: high quality data

• Our results appear to support the view that there are stages of growth

3
3. Externalities of Investment
Growth models with positive externalities of investment implying that investment not
only raises production possibilities of the investor but has also positive effects on the
aggregate economy. This type of growth model goes back to Scitovsky (1954), Myrdal
(1957), Romer (1986) and may characterize the first stage of growth.

The Estimable Model with Externalities

Z ∞
max e−(ρ−n)t U (C)dt. (1)
C 0

K̇ = Aα K 1−α − C(δK + n)K, (2)

Ȧ = ϕI − δA A . (3)
Dynamics: FOC and the state variables

Ċ ρ + δK (1 − α)K −α Aα
= − + (4)
C σ σ
µ ¶α
K̇ C A
= −(δK + n) − + (5)
K K K
µ ¶
Ȧ I
= −δA + ϕ . (6)
A A

In ratios: Presume, k = K/A and c = C/A are constant over time. This gives
k̇/k = K̇/K − Ȧ/A and ċ/c = Ċ/C − Ȧ/A or explicitly:

k̇ c
= −(δK + n) − + δA + ϕc + (1 − ϕk)k −α , (7)
k k
ċ ρ + δK 1 − α −α
= − + k + δA + cϕ − ϕk 1−α . (8)
c σ σ
which gives an estimable equation:
µ ¶α
δK + ρ 1 − α A I
D(ln(C/K)) = (δK + n) − + − , (9)
σ σ K K

4
Empirical Results

Table 1: Estimation of Equation (9) for Germany

ϕ = 0.3 ϕ = 0.4
Par. Est. Std. Err. t-Stat. Par. Est. Std. Err. t-Stat.
c1 -0.106 0.044 -2.417 c1 -0.08 0.034 -2.516
c2 0.405 0.065 6.214 c2 0.327 0.053 6.175
R2 0.38 R 2
0.38
ϕ = 0.5 ϕ = 0.6
Par. Est. Std. Err. t-Stat. Par. Est. Std. Err. t-Stat.
c1 -0.073 0.029 -2.549 c1 -0.08 0.029 -2.729
c2 0.272 0.044 6.124 c2 0.255 0.043 5.997
R2 0.38 R 2
0.38
ϕ = 0.7 ϕ = 0.8
Par. Est. Std. Err. t-Stat. Par. Est. Std. Err. t-Stat.
c1 -0.268 0.149 -1.8 c1 -0.294 0.187 -1.571
c2 0.474 0.161 2.954 c2 0.478 0.197 2.423
R2 0.4 R 2
0.4

In Sum:
For Germany, Table 3.1 gives the outcome for different values of ϕ. Overall, the model
matches the growth performance for Germany, France and Japan, but less for the US and
the UK.

5
4. Education and Economic Growth
Growth models in which individuals spend time on the formation of human capital. A
prototype example is the Uzawa-Lucas model (based primarily on Uzawa 1965 and Lucas
1988) which may characterize an important force in the second stage of growth.

The Uzawa-Lucas Model

∞
c(t)1−σ − 1 −ρt
Z
max L(t) e dt (10)
c,u 0 1−σ
subject to

K̇(t) = AK(t)1−α (u(t)h(t)L(t))α hζa − L(t)c(t) − δk K (11)

ḣ(t) = h(t)p1 κ(1 − u(t))p2 − δh h(t) (12)

In Lucas (1988:) ḣ(t) = h(t)κ(1 − u(t))

The Estimable Model

k̇ c
= Ak −α hα+ζ uα − (13)
k k
ḣ
= κ(1 − u) (14)
h
ċ A(1 − α) −α α+ζ α ρ
= k h u − (15)
c σ σ
u̇ κ(1 − α − ζ) κ(α + ζ) c
= u+ − (16)
u 1−α 1−α k
The model with the direct scale effects does not fit, therefore, we consider two variants.
First, we suppose that the equation ḣ/h is given by

ḣ(t)
= h(t)p1 −1 κ(1 − u(t))p2 − δh , (17)
h(t)
Second, it is written as

ḣ(t)
= h(t)p1 −1 κ(t)(1 − u(t))p2 − δh . (18)
h(t)
With those modifications of (12) a modified system of (13-(17) – including depreciation
rates, population growth and nonlinearity in human capital creation– using ratio formula-
tion is estimated. This is called Lucas II (Lucas I is a simpler system, with u a historical
variable)

6
Empirical Results for Lucas II

Table 2: Estimation of the Uzawa-Lucas II Model.

U.S. Germany
Parameters for Matching (kt − kt−1 )/kt−1
α 0.4560 (0.1289) 0.5192 (0.3569)
A 0.0637 (0.0062) 0.0539 (0.0010)
ζ -0.0066 (0.0681) 0.0081 (0.2850)
ρk 0.7123 (0.0582) 0.8518 (0.0516)
Parameters for Matching (ct − ct−1 )/ct−1
ρ 0.0337 (0.0042) 0.0134 (0.0008)
σ 1.5465 (1.1702) 0.4032 (0.1029)
ρc 0.7385 (0.0566) 0.5265 (0.0839)
Parameters for Matching (ht − ht−1 )/ht−1 and (ut − ut−1 )/ut−1
κ 0.0738 (0.0030) 0.0608 (0.0070)
p1 0.3295 (0.0729) 0.4300 (0.1060)
p2 0.1341 (0.0119) 0.0155 (0.0032)
ρh 0.9517 (0.4975) 0.9936 (0.0655)
ρu 0.9369 (0.0175) 0.9712 (0.0037)

In Sum:
The Uzawa-Lucas model fits the data only, if nonlinearities – as shown in (12) – are
allowed for: This means at low level of education the growth effects are stronger than for
higher level education and human capital (see also Krueger and Lindahl 2001).

7
5. Knowledge Accumulation and Economic Growth
Models in which economic agents spend resources for the creation of new knowledge
(measured by R&D, scientist and engineers, patents etc). This type is often called an
R&D model of economic growth (see Romer 1990, Grossman and Helpman 1991) and
may be characteristic for a third stage of growth.

The Model:
∞
C 1−σ − 1 −ρt
Z
e dt, (19)
0 1−σ
s.t.

K̇ = η̄Aα+β (H − HA )α Lβ K 1−α−β − C − δK K, (20)

Ȧ = µHAγ Aφ − δA A, (21)

Romer (1990):
Ȧ = µHA A

Estimated Model and Results:

K̇ C
= η̄K −α (A(H − HA )L)α − − δK (22)
K K
Ċ η̄ ρ + δK
= (1 − α)2 (A(H − HA )L)α K −α − (23)
C σ σ
Ȧ
= µHAγ Aφ−1 − δA (24)
A
˙
HA
= ..... (25)
HA

8
 Table 3: Estimation of the Modified Romer Model
U.S. Economy German Economy
Parameters for Matching (Kt − Kt−1 )/Kt−1
α 0.5564 (0.1114) 0.6171 (0.1360)
η 0.1179 (0.0065) 1.6094 (0.9742)
ρk 0.9686 (0.0039) 0.9904 (0.0002)
Parameters for Matching (Ct − Ct−1 )/Ct−1
ρ 0.0034 (0.0036) 0.0071 (0.0065)
σ 0.5358 (0.4404) 0.6953 (0.7506)
ρc 0.8234 (0.0480) 0.7494 (0.0668)
Parameters for Matching (At − At−1 )/At−1
µ 6.8065 (2.8171) 1.5806 (0.3598)
γ 0.3506 (0.049) 0.0010 (0.0681)
φ 9.43e-005 (0.0044) 0.4936 (0.0323)
ρa 0.9785 (0.0110) 0.9675 (0.0133)

In Sum:
The knowledge accumulation model fits the data better, as shown in (21), if nonlinearities
are allowed for: This means at low level of the accumulation knowledge the growth effects
may be stronger.

9
6. Public Infrastructure Investment
Public Infrastructure investment is important for all stages of growth. The above forces
of economic growth need the support by public policy and thus fiscal policy, and its
sustainability, is central. It is thus useful to connect public economics with growth theory.
We need to study the composition effect of public expenditure is important.

• We pursue an approach that originates in Barro (1990) and Futagami, Morita, and
Shibata (1993), and others. In their model the government undertakes productive
investment that raises the marginal product of capital and stimulates economic
growth (or as social service it enters households‘ welfare).

• Yet they assume a balanced budget, we allow for borrowing and debt, but we have
to define certain fiscal regimes (fiscal rules).

• We consider a broad class of public expenditure (public consumption, public in-

vestment, transfers) and consider the composition effect of public spending (social
investment could be included as well),

10
The Model with Infrastructure
Z ∞
max e−(ρ−n)t L0 u(C(t))dt, (26)
C(t) 0

subject to

K̇ + Ḃ = (w + r1 K + r2 B)(1 − τ ) − C + Tp − (δK + n)K − nB

Ġ = ϕ3 (1 − ϕ0 )T − (δG + n)G,
Ḃ = r2 B + Cp + Tp + Ip − T − nB

with per capita production function of the form

f (K, G) = K β (Ḡ/L)α , (27)

Budgetary Regimes (Blinder and Solow, Domar, Barro)

Table 4: Budgetary Regimes

Target Deficit due to

A1 C p + T p + r2 B < T public investment
A2 C p + T p + ϕ 4 r2 B < T public investment +(1 − ϕ4 ) r2 B
B1 C p + T p + Ip < T interest payment on national debt
B2 Cp + Tp + Ip > T, Cp + Tp < T Cp + Tp + Ip + interest payment

Remark:
The regimes starting from A1 and declining to B2 represent budgetary regimes with
softer and softer budget constraints: Regime A1, allows public investment to be financed
by deficit, regime A1, permits in addition some interest on debt to be paid by the deficit,
B1 allows all interest to be paid by the deficit and B2 allows to shift any component of
the public budget to be paid by deficit.

11
Solution for Fiscal Regimes
Solving the above optimization problem and taking into account the marginal productivity
rules and the budgetary regimes, we obtain:

K̇ C
= − − (δK + n) + K β−1 Gα − τ (ϕ2 + ϕ3 (1 − ϕ0 )) ×
K µK µ ¶¶
β−1 α B β−1 α δK
K G + βK G − (28)
K 1−τ
K β Gα
µ ¶
Ḃ β−1 α δK
= (ϕ0 − 1)(1 − ϕ3 )τ βK G − + −n+
B 1−τ B
µ ¶
β−1 α δK
(1 − ϕ4 ) βK G − (29)
1−τ
Ċ ρ + δK (1 − τ )βK β−1 Gα
= − + (30)
C σ σ
µ ¶
Ġ α−1 β β−1 α B B δK
= ϕ3 (1 − ϕ0 )τ G K + βK G − −δG −n (31)
G G G1−τ

Estimated Model and Results:

C B G
c≡ , b≡ , x≡ .
K K K
The new variables with respect to time gives

ċ Ċ K̇ ḃ Ḃ K̇ ẋ Ġ K̇
= − , = − , = − (32)
c C K x B K x G K
Using GMM-estimation, the set of orthogonal conditions for our GMM estimation is:
h∼ i
E c −f1 (c(ψ), x(ψ), b(ψ)) = 0 (33)
h∼ i
E b −f2 (c(ψ), x(ψ), b(ψ)) = 0 (34)
h∼ i
E x −f3 (c(ψ), x(ψ), b(ψ)) = 0 (35)

12
Table 5: Results of the GMM Estimation for U.S. Time Series, 1960.4-1992.1, with
(τ, ϕ0 , ϕ1 , ϕ2 , ϕ3 , ϕ4 ) = (0.32, 0.815, 0.4, 0.35, 1.3, 0.9) from empirical data, Regime A2.

ρ σ 1−α
Estimated Parameters 0.061 0.053 0.244
(standard errors) (75096) (614621) (0.0179)

Table 6: Results of the GMM estimation for German Time Series, 1966.1-1995.1, with
(τ, ϕ0 , ϕ1 , ϕ2 , ϕ3 , ϕ4 ) = (0.4, 0.945, 0.4, 0.42, 1.5, 1) from empirical data.

ρ σ 1−α
Estimated Parameters 0.004 0.224 0.135
(standard errors) (230566) (8425716) (0.0287)

In Sum:
• The outcome on growth rates (welfare) and growth maximizing tax rates depend on
the fiscal regimes (A1 and A2 are the superior ones)

• The empirical estimates indicate an overestimation of the contribution of public

capital (since other forces of growth are left aside)

• Sustainability tests of fiscal policy: Using the above estimated parameters one can
make inference concerning the sustainability of fiscal policy (the model allows to
assess solvency; recently we explored other tests, i.e. time varying coefficient models)

13
7. Evaluation and Challenging Issues
• through the construction of time series for externalities, human capital (cumulative
educational expenditure), knowledge capital (cumulative R&D), and public infras-
tructure, we have estimated the model for countries with high quality data

• dynamic models with nonlinearities estimated: the results suggest ”stages of eco-
nomic growth”

• we undertake an empirical study for a few countries, further modifications can be

considered (structural shifts, more complex preferences, etc )

• the study was mostly undertaken for countries with higher per capita income (with
high quality data), new studies are on the way for less developed countries, the
construction of data such as human capital, i.e. through cohorts

• our study suggests policies to improve growth prospects for specific stages (this
surely needs to be more work)

– are for each stage of economic growth particular policies needed (public goods
and services, growth enhancing policies)
– we here have explicitly modelled the effects of components of public spending
but have not much considered social expenditure improving welfare (there may
be trade-offs as well as complementarities),
– can the model be extended to include basic social services (health, desease
control) where the impact has long lags
– efficiency of governments: we are aware that publicly induced investment may
crowd out private investment, and as there are market failures there are also
government failures (accountability and transparency needed, i.e. public ac-
counting office)

14