Public Capital and Growth

Serkan Arslanalp, Fabian Bornhorst,

Sanjeev Gupta, and Elsa Sze

WP/10/175

2010 International Monetary Fund WP/10/175


IMF Working Paper

Fiscal Affairs Department

Public Capital and Growth

Prepared by Serkan Arslanalp, Fabian Bornhorst, Sanjeev Gupta, and Elsa Sze
1


July 2010

Abstract

This Working Paper should not be reported as representing the views of the IMF.
The views expressed in this Working Paper are those of the author(s) and do not necessarily
represent those of the IMF or IMF policy. Working Papers describe research in progress by the
author(s) and are published to elicit comments and to further debate.

This paper estimates the impact of public capital on economic growth for forty-eight OECD
and non-OECD countries during 19602001. Using the production function and its
extensions, it finds a positivebut concaveelasticity of output with respect to public
capital, which is robust to changes in time intervals and varying depreciation rates.
Furthermore, in non-OECD countries the growth impact of public capital is higher once
longer time intervals are considered.


JEL Classification Numbers: E22, H54, O40, O57

Keywords: Public investment, public capital, economic growth.

Authors E-Mail Address: [email protected], [email protected], [email protected],
[email protected]

1
The authors wish to thank Emanuele Baldacci, Andrew Berg, Giovanni Callegari, Reda Cherif, Benedict
Clements, Carlo Cottarelli, Luc Eyraud, Julio Escolano, Annalisa Fedelino, Mark Horton, Doug Hostland, Paolo
Mauro, Nkunde Mwase, Catherine Pattillo, Mauricio Villafuerte, Abdoul Aziz Wane and Felipe Zanna helpful
comments on earlier versions of the paper. The usual disclaimer applies.
2

Contents Page
I. Introduction ............................................................................................................................3
II. Theory ...................................................................................................................................5
A. Standard Model .........................................................................................................5
B. Extended Model ........................................................................................................6
III. Data ......................................................................................................................................6
A. Construction of the Dataset .......................................................................................6
B. Depreciation Rates ....................................................................................................8
C. Caveats ......................................................................................................................9
IV. Results................................................................................................................................10
A. Descriptive Statistics ...............................................................................................10
B. Regression Results ..................................................................................................14
C. Post Estimation Tests ..............................................................................................22
V. Alternative Depreciation Rates ...........................................................................................22
VI. Conclusions........................................................................................................................25

Tables
1. List of Countries in the Sample .............................................................................................7
2. Depreciation Rates .................................................................................................................9
3. GDP Growth, Public Investment and Public Capital Stock Growth, 19602000 ...............10
4. Average Real GDP Growth Before and After Public Capital Stock ...................................11
5. Panel Unit Toot Tests for Non-OECD Countries ................................................................14
6. Regression Results for All Countries ...................................................................................18
7. Regression Results for OECD Countries .............................................................................20
8. Regression Results for Non-OECD Countries .....................................................................21
9. Sensitivity Analysis: Alternative Depreciation Rates for OECD Countries ........................23
10. Sensitivity Analysis: Alternative Depreciation Rates for Non-OECD Countries .............24

Figures
1. Long-Term Real GDP Growth, Public Investment and Public Capital Growth Rates,
19602000................................................................................................................................12
2. Public Investment, Public Capital Stock and Real GDP Growth for OECD and
non-OECD Countries, 19602000 ...........................................................................................13
3. Standard Model with Varying Coefficients of Public Capital .............................................16

Appendix Table: Literature Survey .........................................................................................32

References ................................................................................................................................26

3

I. INTRODUCTION
An understanding of the impact of public investment on growth is relevant for at least three
reasons. First, it has been argued that tight budgets constrain public investment more than
current spending because it is easier to cut the former for political and other reasons (Roy,
Heuty, and Letouze, 2006). Since the late 1990s, this has led to calls to correct the bias
against public investment, most importantly in infrastructure, and create fiscal space for
funding such investment (Heller, 2005).
2
Underlying this premise is the belief that public
investment is productive.

Second, in a somewhat similar vein, it has been argued that constraints on external borrowing
have prevented governments with large infrastructure gaps from undertaking productive
investments. Although a countrys borrowing capacity depends primarily on its
macroeconomic policies, ability to collect taxes, and strength of its public financial and debt
management systems, the contribution of debt-financed public investment to growth and
exports also plays a role in external borrowing limits.

Finally, fiscal policy has a countercyclical role to play in supporting growth and recovery,
which has been recognized during the recent financial crisis. In this context, fiscal stimulus
packages in many countries have included a large share of public investment spending in the
belief that such investment is productive and better for future growth.
3


However, the empirical evidence on the impact of public investment on growth is mixed.
Previous studies on the impact of public investment on growth have not produced clear-cut
results (IMF 2004 and IMF 2005b). This has led some to conclude that public investment is
not productive. Some have also argued that total factor productivity (TFP), rather than capital
accumulation, matters in explaining growth differentials (Easterly and Levine (2001).

At the same time, a more recent study by the World Bank (2007) concluded that there are
positive growth effects of public spending in general, and that of infrastructure, education,
and health spending in particular. The report from the Commission on Growth and
Development (2008) came to an even stronger conclusion by noting that a common element
in fast-growing countries is high public investment, defined as 7 percent of GDP or more.
Other studies have argued that fiscal multipliers for investment spending are higher than
those for other types of public spending or tax cuts (Perotti, 2005; Zandi, 2008).

Why is the evidence on the relationship between public investment and growth mixed? There
could be three reasons. First, many empirical studies use the public investment rate, as

2
Fiscal space refers to room in a governments budget that allows it to provide resources for a desired purpose
without jeopardizing the sustainability of its financial position (IMF, 2005a).
3
The share of infrastructure spending in fiscal stimulus packages for 200910 is about 20 percent for advanced
G-20 countries, and more than 50 percent for emerging G-20 countries (Horton et al. 2009).
4

opposed to the rate of change in public capital in explaining differences in growth across
countries. This is because public investment data are easier to obtain, even though public
investment and public capital variables can differ substantially for a country. In particular,
these variables can grow at different rates, depending on the initial level of capital stock.

Second, there is an endogenous link between public investment and growth that can
complicate econometric identification. Public investment and growth are flow variables
determined in equilibrium and observed over the same period. For example, public
investment may fall in an economic downturn simply due to lack of resources, as it is often
among the first expenditure items to be cut in a downturn. In contrast, public capital stock
does not suffer from this drawback, given that it is measured at the beginning of the period.
A year in which growth is low would not affect the (beginning of the period) capital stock.

Third, most studies do not take into account the governments budget constraint and the costs
of financing public investment, which could lead to diminishing returns to investment. The
relationship between public investment and growth could even turn negative once public
capital is above a certain threshold. For example, maintaining and/or expanding the existing
capital stock may require high (and potentially distortionary) tax rates, which would reduce
growth, all else being equal (Aschauer 1998; Barro 1990). Thus, the productivity of public
investment can vary depending on the initial stock of public capital.

We therefore use measures of the capital stock, rather than investment rates in our study,
notwithstanding difficulties in estimating the capital stock. We find that the importance of
assumed initial capital stock diminishes significantly in long time series.
4
We address the
issue of choosing an appropriate depreciation rate by using different rates. Indeed, our
sensitivity analysis shows that under varying depreciation rates, the results of the paper hold.

An important contribution of the paper lies in the construction of public capital stock series
for 26 middle-income and low-income countries. Such estimates already exist for 22 OECD
countries (Kamps 2006). Our estimates rely on the same methodology as that used for OECD
countries.

Our results for a panel of 48 OECD and non-OECD countries show that there is a positive
elasticity of output with respect to public capital. This supports the view that changes in
public capital stock can explain growth differences across countries, even though the
evidence on the impact of public investment is mixed. The literature review in the appendix
table, drawn from Romp and de Haan (2005), further indicates that studies that used changes
in public capital stock have typically obtained positive results, while the picture is less clear
for those using public investment rates.


4
In particular, if the capital stock in 1960 is assumed to be zero in our sample, the capital stock in 2001 would
differ by only 6 percent for the average country in the sample.
5

We also find that the elasticity of output with respect to public capital depends on the income
level of countries (OECD versus non-OECD) and the initial level of public capital. The
elasticity is somewhat stronger for OECD countries, possibly suggesting the importance of
institutional factors. At the same time, we find that the positive impact of public capital on
output varies with the level of public capital, and that it can be partially offset by high levels
of public capital. Such inefficiencies may be engendered by taxes levied to maintain or
expand the existing capital stock.

The paper is organized as follows. Section II presents the theoretical models underlying the
empirical estimation. Section III explains the construction of the public capital stock dataset.
Section IV discusses the empirical results and Section V presents sensitivity analysis with
respect to depreciation rates. Section VI concludes and discusses policy implications.

II. THEORY
We use the standard production function approach in the literature and its extension to
estimate the impact of public capital on output (Aschauer 1989; Aschauer 1998).

A. Standard Model
In this model, public capital is explicitly included in the aggregate production function by
redefining the production function Y= A* F(L,K) as

Y= * F(L,K,G)

where Y is the level of output, A is the level of productivity, L is employment, K is private
capital, G is public capital and is the total factor productivity purged of the influence of
public capital.
5
A commonly used specification is the Cobb-Douglas production function:

Y = L
a
K
b
G
c


Taking natural logarithms yields the equation:

InY = In + a lnL + b lnK + c lnG

Taking first differences yields the equation:

InY = constant + a lnL + b lnK + c lnG

The elasticity of output with respect to public capital, c, is the main variable of interest in this
study. The other production elasticities, a and b, are of interest mainly to assess the shape of
the production function. Assuming perfect competition in factor markets, private capital and
labor must be paid their marginal products (i.e., a + b = 1). In this case, given that c is

5
Some studies also include human capital in the production function. We stick to the standard approach, where
human capital is assumed to be included in the TFP measure .
6

expected to be positive, the model would generate increasing returns to scale (a + b + c > 1).
On the other hand, if one assumes constant returns to scale (a + b + c = 1), then labor and
private capital must be paid more than their marginal products (i.e., a + b < 1). That would be
the case when public capital generates indirect income for labor and private capital, even
though it is not paid directly for its services. In this paper, the model is estimated without any
restrictions on the coefficients a, b, and c, but they are subsequently tested.

B. Extended Model
The extended model is the same as the standard model, except that it allows for a diminishing
elasticity of public capital as a function of the initial stock of public capital (in percent of
GDP). The motivation for this control variable is the potential non-linearities in the
productivity of public capital. For example, if public investment is financed by capital taxes,
a higher public investment would reduce after-tax profits for investors and curtail their
incentive to invest. At some point, the disincentive effect from higher taxes would exceed the
productivity gains from higher public investment, and the net impact on growth would
become negative (Barro 1990). Furthermore, the marginal productivity of public capital
could decline due to inefficiencies in the capital spending process and due to difficulties in
finding investment projects with high returns.

To capture this relationship, we extend the standard model by adding an interaction on the
right hand side of the equation. In particular:

InY = constant + a lnL + b lnK + c lnG + d lnG * (G/Y)

The interaction term implies that the production elasticity of public capital is no longer just c,
but is equal to (c + d G/Y). This implies that the productivity of public capital can vary
according to its initial level. The coefficient on the interaction term, d, is expected to be
negative, while the coefficient c is expected to remain positive.

III. DATA
A. Construction of the Dataset
The dataset includes public and private capital stock series for 48 countries from 1960 to
2001. Table 1 lists the countries in the sample. The data for 22 OECD countries is from
Kamps (2006).
6
We construct public and private capital stock data for the other 26 non-
OECD countries
7
using the same methodology for OECD countries. The estimation is based
on internationally comparable total investment series from Penn World Tables (Heston,

6
The underlying investment data used in Kamps (2006) are from the OECD Analytical Database, Version 2002
and available for 1960-2001.
7
The countries referred to as non-OECD include the countries not covered in Kamps (2006).
7


Summers, Aten 2006) and public and private investment series from the World Economic
Outlook database (IMF, April 2009).
8


Table 1. List of Countries in the Sample

OECD
Non-OECD
(Middle-income)
Non-OECD
(Lower-income)
Australia Argentina Bangladesh
Austria Brazil Bolivia
Belgium Colombia Egypt
Canada Dominican Republic Guatemala
Denmark Malaysia Honduras
Finland Mexico India
France Panama Jordan
Germany Peru Kenya
Greece South Africa Morocco
Iceland Thailand Pakistan
Ireland Tunisia Paraguay
Italy Turkey Senegal
Japan Uruguay Swaziland
Netherlands
New Zealand
Norway
Portugal
Spain
Sweden
Switzerland
United Kingdom
United States
Note: Middle income non-OECD countries have per capita income of
more than $3,700 in 2000 prices. Lower income non-OECD countries
have per capita income of less than $3,700 in 2000 prices.


The Penn World Table (PWT) is one of the most widely used sources for cross-country
growth studies. It provides data on internationally comparable output and investment series
for a broad group of countries. The data are based on national accounts and adjusted for
purchasing power parity (PPP) to make them internationally comparable.
9
PWT also assigns
grades to countries from A to D based on the quality and consistency of their data over time.
Countries with a grade of D are excluded from the sample.


8
The public and private capital stock estimates for the non-OECD countries in this paper are available for
download at http://www.imf.org/external/pubind.htm.
9
The output and investment series are in constant (2000) international prices. For investment, we use total gross
domestic investment in 2000 international prices (PWT code: KI).
8

One drawback of PWT is that it does not provide a breakdown of investment into its public
and private components. For that, we turn to the World Economic Outlook (WEO) database
(IMF 2009), which provides disaggregated data on public and private investment.
10
We use
the share of public and private investment in total investment to split the PWT investment
series into public and private components.

Based on the disaggregated PWT investment series, the public and private capital stocks are
estimated using the perpetual inventory method employed by Kamps (2006) for OECD
countries. First, the initial capital stock is set to zero for all countries in 1860.

Second, a
hypothetical investment series is constructed between 1860 and 1960 based on a 4 percent
growth rate for investment during 18601960 to calculate the capital stock in 1960. Lastly,
the investment series for 19602000 are accumulated to construct the capital stock series for
19602001.

B. Depreciation Rates
The net capital stock accounts for the wear and tear of an asset (i.e., depreciation), thus it
excludes assets that are no longer used in production. The choice of depreciation rates
present perhaps the biggest challenge in the construction of the capital stock estimates
because country specific estimates of depreciation rates are typically not available, with the
U.S. being an important exception. According to the Bureau of Economic Analysis,
estimated depreciation rates for public capital in the U.S. were about 2.5 percent in 1960 and
increased to 4 percent in 2001 (and from 4.25 to 8.5 percent for private capital, respectively).
Kamps (2006) uses these U.S. depreciation rates to construct the capital stock estimates for
other OECD countries.

Different depreciation assumptions may be more appropriate for non-OECD countries, given
differences in the types of assets they hold. The composition of underlying assets affects the
average depreciation rate of the aggregated stock because different types of assets have
different life spans. For instance, concrete structures are typically estimated to have longer
lifetime (e.g., 80100 years), while assets related to IT tend to have much shorter life span
(e.g., a few years). As countries become richer, the share of assets with shorter life span rises,
thereby raising the overall depreciation rate.

The rising depreciation rates in OECD countries reflect the growing importance of high
technology assets in their public capital. In contrast, non-OECD countries with relatively
lower share of technological assets and more traditional physical infrastructure in their
public capital are more likely to have lower depreciation rates, as the average life span of
those assets is higher. Arestoff and Hurlin (2006) find that this is in fact the case for a
number of developing countries.
11


10
We use the series gross public fixed capital formation, current prices (WEO code: NFIG) and gross private
fixed capital formation, current prices (WEO code: NFIP).
11
Arestoff and Hurlin (2006) estimate depreciation rates based on six components of public investment(roads,
railways, electricity, gas, water and telecom).
9


Table 2 shows the depreciation rates for public and private capital assumed for constructing
the capital stock for non-OECD countries. For middle-income countries, we use a time-
varying profile for public and private capital with a flatter slope than the one used for OECD
countries. For low-income countries, we hold the depreciation rate constant over time.

Table 2. Depreciation Rates
(In percent)


C. Caveats
Our capital stock dataset is novel in several ways: It includes both OECD and non-OECD
countries, differentiates between public and private capital, and applies time- and income-
level varying depreciation rates to capture the nature of the underlying public and private
assets. Previous studies, such as by Nehru and Dhareshwar (1993) and Marquetti and Foley
(2008), have estimated the capital stock of non-OECD countries but have not made a
distinction between public and private capital. On the other hand, Collier, Hoeffler and
Pattillo (2001) have estimated the public capital stock for 22 sub-Saharan African countries,
but assuming constant depreciation rates. Finally, Arestoff and Hurlin (2006) have estimated
the public capital stock for a group of non-OECD countries using infrastructure specific
depreciation rates, but not the private capital stock. Our dataset includes both public and
private capital stock data, which are necessary to estimate production functions, and covers
countries with different income groups from all five continents.

However, some caveats are in order:

First, investments in public capital may not always be productive (Pritchett 1996, Canning
1999, Easterly and Serven 2004). Reasons range from administrative inefficiencies to pork-
barrel politics to corruption. This unobservable factor could cause public capital in some
countries to be overestimated. While this is clearly an important issue, the estimated
elasticity of output with respect to public capital should reflect this spending inefficiency. All
else being equal, a country with a lower spending efficiency and overstated capital stock
would have a lower elasticity of output with respect to public capital.

Second, non-OECD countries tend to spend less on operations and maintenance, which could
cause depreciation rates to be higher in non-OECD countries compared to OECD countries.
1860 1960 2001
Public capital
OECD 2.50 2.50 4.00
Non-OECD (middle-income) 2.50 2.50 3.25
Non-OECD (low-income) 2.50 2.50 2.50
Private capital
OECD 4.25 4.25 8.50
Non-OECD (middle-income) 4.25 4.25 7.00
Non-OECD (low-income) 4.25 4.25 4.25
10

On the other hand, as mentioned above, non-OECD countries tend to have a lower share of
high-technology assets, which would make depreciation rates lower in non-OECD countries.
While the net effect is unknown, we address this issue by conducting a sensitivity analysis
using different depreciation rates (Section V).

Third, there may be differences in the way countries classify their investments as public or
private, given the presence of public-private partnerships and the treatment of quasi-public
enterprises in the national accounts. We do not address this issue in the paper, but note that
this problem is common to all studies on the subject, regardless of whether they use public
investment or public capital as the explanatory variable.

Finally, by applying the same depreciation rate to a group of countries, the approach
disregards natural disasters and other Force Majeure events that may impact the capital stock
of any particular country.

IV. RESULTS
A. Descriptive Statistics
Table 3 provides the average real GDP growth, public investment rate, and the public capital
stock growth for each income group during 19602000.
12
It shows that the average GDP
growth for the OECD countries was 3.4 percent during this period. For the average non-
OECD countries, the growth rates were higher at 4.4 percent, suggesting that some catching-
up took place during this period.

Table 3. GDP Growth, Public Investment and
Public Capital Stock Growth, 19602000

Real GDP growth 3.4 (0.7) 4.4 (1.3)
Public investment (percent of GDP) 3.6 (1.2) 3.9 (1.7)
Public capital stock growth 3.3 (1) 5.7 (2.3)
Source: Authors' calculations. Standard deviations in paranthesis.
OECD Non-OECD


The difference in public investment rates between OECD and non-OECD countries was
relatively small during 19602000 (3.6 versus 3.9 percent of GDP). This contrasts with the
difference in public capital growth, which was much higheralmost twice as muchin non-
OECD countries (5.7 versus 3.3 percent).

12
The growth rate for each country, Yi, is calculated as the geometric average growth rate between 1960 and
2000. More specifically, Y
i
= 1/40 [log (Y
i2000
- Y
i1960
)]. The average for the income group is calculated as the
average of the growth rates for each country in the group.
11


Figure 1 plots a scatter of real GDP growth, public investment, and public capital growth in
sample countries during 19602000. It shows that public capital growth is better in
explaining the variation in GDP growth than public investment rates. Public investment rates
lie in a narrow range, between 2 and 6 percent of GDP for most countries, whereas variation
in public capital growth is larger. The public investment rate explains only 12 percent of the
cross-country variation in GDP growth, whereas public capital growth explains 51 percent of
the variation.

Figure 2 plots the public investment rate and public capital stock (both as a percent of GDP)
for the average OECD and non-OECD countries. Note that public investment and public
capital follow significantly different paths for each group. For example, public investment in
the average OECD country started declining in the 1970s, whereas public capital was still
increasing until the mid-1980s. For the average non-OECD country, both public investment
and public capital stock peaked in the mid-1980 and since then has been on a declining
trend.

Figure 2 also shows that the peak level of public capital was found in non-OECD countries.
This is explained by the higher public investment rates in these countries (4 percent of GDP),
compared to OECD countries (around 3 percent of GDP), during this period. Finally,
Figure 2 shows the relationship between public capital and GDP growth. During the pre-
peak years, the average real GDP growth was consistently high for OECD and non-OECD
countries. In fact, Table 4 shows that real GDP growth was about one percentage point
higher for both OECD and non-OECD countries in the pre-peak period when public capital
stock was still increasing as a percent of GDP.

Table 4. Average Real GDP Growth Before and After Public Capital Stock

OECD 3.8 (1.8) 2.8 (1.1)
Non-OECD 4.8 (1.5) 3.7 (1.1)
Source: Authors' calculations. Standard deviations in paranthesis.
Before the peak After the peak
Note: Public capital stock (as a percent of GDP) peaked in 1983 for the average OECD
country and in 1985 for the average non-OECD country. The peak level was 60 percent
of GDP for the average OECD country, 62 percent of GDP for the average non-OECD
country.

12

Figure 1. Long-Term Real GDP Growth, Public Investment and Public Capital
Growth Rates, 19602000

Sources: PWT (Version 6.2), Kamps (2006), Authors' calculations.
AUS
AUT
BEL
CAN
DNK
FIN
FRA
GER
GRC
ISL
IRL
ITA
JPN
NLD
NZL
NOR
PRT
ESP
SWE
CHE
GBR
USA
ARG
BRA
COL
DOM
MYS
MEX
PAN
PER
ZAF
THA
TUN
TUR
URY
BGD
BOL
EGY
GTM
HND
IND
JOR
KEN
MAR
PAK
PRY
SEN
SWZ
y = 0.2738x + 2.8842
R
2
= 0.1169
0.0
1.0
2.0
3.0
4.0
5.0
6.0
7.0
8.0
0.0 2.0 4.0 6.0 8.0 10.0 12.0
Public Investment to GDP
R
a
t
e

o
f

C
h
a
n
g
e

i
n

R
e
a
l

G
D
P
AUS
AUT
BEL
CAN
DNK
FIN
FRA
GER
GRC
ISL
IRL
ITA
JPN
NLD
NZL
NOR
PRT
ESP
SWE
CHE
GBR
USA
ARG
BRA
COL
DOM
MYS
MEX
PAN
PER
ZAF
THA
TUN
TUR
URY
BGD
BOL
EGY
GTM
HND
IND
JOR
KEN
MAR
PAK
PRY
SEN
SWZ
y = 0.3862x + 2.1254
R
2
= 0.5147
0.0
1.0
2.0
3.0
4.0
5.0
6.0
7.0
8.0
0.0 2.0 4.0 6.0 8.0 10.0 12.0
Rate of Change in Public Capital
R
a
t
e

o
f

C
h
a
n
g
e

i
n

R
e
a
l

G
D
P

13

Figure 2. Public Investment, Public Capital Stock and Real GDP Growth
for OECD and non-OECD Countries, 19602000
(percent of GDP and annual percentage change)

Source: PWT (Version 6.2), Kamps (2006), Authors' calculations.
The red dotted lines indicate the average real GDP growth before and after the year in which the public
capital stock peaked for each income group. Public capital stock (as a percent of GDP) peaked in 1983
for the average OECD country and in 1985 for the average non-OECD country. The peak levels were 60
percent of GDP for the average OECD country and 61percent of GDP for non-OECD countries.
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
4.5
5.0
Public Invesment Rate
( percent of GDP)
OECD
0.0
1.0
2.0
3.0
4.0
5.0
6.0
7.0
Public Invesment Rate
(percent of GDP)
Non-OECD
30.0
35.0
40.0
45.0
50.0
55.0
60.0
65.0
Public Capital Stock
(percent of GDP)
OECD
30.0
35.0
40.0
45.0
50.0
55.0
60.0
65.0
Public Capital Stock
(percent of GDP)
Non-OECD
0.0
1.0
2.0
3.0
4.0
5.0
6.0
7.0
1960 1965 1970 1975 1980 1985 1990 1995 2000
Real GDP growth
OECD
0.0
1.0
2.0
3.0
4.0
5.0
6.0
7.0
1960 1965 1970 1975 1980 1985 1990 1995 2000
Real GDP growth
Non-OECD

14

B. Regression Results
We first examine the question whether regressions should be run in levels or differences.
Table 5 shows that the variables of interest for non-OECD countries are stationary when
expressed in first differences, extending a previous finding for OECD countries (Kamps
2006). We do not find any relationship between the variables in levels that would point to a
common cointegrating relationship in the panel, neither for the combined dataset nor for non-
OECD and OECD countries separately.
13
We thus proceed to estimate the equation in
differences to obtain consistent estimates of the coefficients.


Table 5. Panel Unit Toot Tests for Non-OECD Countries

Variable Deterministic IPS (t-value) LLC (t-value) Result
ln GDP constant 1.29 -1.07 at least I(1)
trend -0.80 -1.96 at least I(1)
ln L constant -0.08 -1.77 at least I(1)
trend -3.97 -7.14 I(0)
ln K constant 1.79 -1.36 at least I(1)
trend -0.11 -2.93 at least I(1)
ln G constant -2.81 -5.40 I(0)
trend -1.14 -6.32 inconclusive
IG/GDP constant -3.22 -4.31 I(0)
trend -1.20 -1.05 at least I(1)
G / GDP constant -1.38 -5.27 inconclusive
trend -2.91 -4.36 I(0)
ln GDP constant -10.52 -9.78 I(0)
ln L constant -0.97 -2.34 I(0)
ln K constant -3.04 -3.00 I(0)
ln G constant -2.97 -3.14 I(0)
(IG/GDP) constant -15.98 -13.49 I(0)
G / GDP constant -6.03 -6.00 I(0)
Source: Authors' calculations. IPS and LLC columns report the Im, Pesaran and Shin (2003) and
the Levin, Lin and Chu (2002) corrected t-value of the lagged variable, which is normally
distributed under the null hypothesis of nonstationarity. One lagged difference included. Panel unit
root tests for OECD countries are conducted by Kamps (2006).

Table 6, 7, and 8 provide the main results of the paper. The OLS panel regressions include
fixed effects. These are better at capturing cross-country differences in technological growth,
human capital accumulation, and any other factor affecting total factor productivity, which
are reflected in the intercept term (ln). This is confirmed by Hausman tests on alternative
random effect specifications. Table 6 shows results for the combined OECD and non-OECD

13
We test for common panel cointegration using a two step approach and residual based panel (non)stationarity
tests. We focus on a common cointegrating relation because we are interested in exploiting the cross section
information in the panel, reflecting the common set of assumptions underlying the capital stock estimate. While
country specific cointegrating relations between the variables may existpossibly including additional country
specific variables we do not identify them as we are interested in a cross country validation of the model, and
therefore use a first difference representation of the model.
15

dataset; Tables 7 and 8 provide separate results. The results are presented for three models:
naive, standard and extended. We also present three different time intervals in the analysis,
namely one year, five years, and ten years.
14

One reason for analyzing different time intervals is the existence of both long and short-term
effects of public capital accumulation on growth that may not be captured well in annual
data. For example, some public investments may take more than a year to complete and, in
addition, the payoff may accrue over a longer time horizon. The longer time horizons,
especially the five-year interval, may be better suited to capture the indivisibility of
investment and lags in its effectiveness.
Naive model
The naive model uses the public investment rate (IG/GDP) instead of public capital growth
(ln G) in the standard model.
15
As expected, we do not find any statistically significant
relationship between the public investment rate and GDP growth for the combined dataset
and for the non-OECD countries. For the OECD, the coefficient is significant but negative
only in the five-year interval specification.
Non-zero depreciation rates imply that the same gross investment can lead to very different
rates of net capital accumulation. For example, in our sample of non-OECD countries,
countries with a low public capital stock (30 percent of GDP, the 25
th
percentile) on average
lose public capital worth around one percent of GDP annually due to depreciation, whereas
those with a high stock of public capital (70 percent of GDP, the 75
th
percentile) lose more
than double that amount, close to two percent of GDP. Consequently, a gross investment rate
of 7 percent of GDP (Growth Commission report, 2008) would translate into net investment
rates between 5 and 6 percent of GDP, different paths of capital accumulation, and different
growth effects. Another way of making the same point is as follows: a 7 percent of GDP
gross investment rate implies a public capital accumulation rate of 20 percent if the initial
stock is 30 percent of GDP, while it implies an accumulation rate of only 7 percent, if the
initial stock is 70 percent of GDP. This shows that there could be large differences between
the public investment rate and the rate of change in the public capital stock depending on the
initial capital stock.
Standard model
In the combined OECD and non-OECD panel dataset (Table 6) the standard model produces
plausible and statistically significant results in the one-year specification, with a public
capital stock coefficient of 0.05. Coefficients for L and K are significant and positive in all
specifications. When varying the coefficient for G with the distribution of public capital

14
When moving to the five and ten-year intervals, we redefine Y and L as flows over a five or ten-year period,
while K, and G are defined as the stock at the beginning of these periods.
15
Arithmetically, the growth rate of public capital (G/ G) is equal to the investment rate (G /Y) divided by
the capital stock in percent of GDP (G/Y) if the depreciation rate is zero.
16

stock (by multiplying with quartile dummy variables), a concave pattern emerges with
respect to the elasticity of public capital, indicating a maximum impact of additional public
capital (growth elasticity of 0.2) when the public capital stock is between 15 and 60 percent
of GDP (Figure 3). For very high levels of public capital, the coefficient loses statistical
significance. We now proceed to estimate separate models for OECD and non-OECD
countries.
Figure 3. Standard Model with Varying Coefficients of Public Capital
-0.1
0.0
0.1
0.2
0.3
14 37 58 169
E
l
a
s
t
i
c
i
t
y

o
f

p
u
b
l
i
c

c
a
p
i
t
a
l
Initial public capital (percent of GDP)
- standard deviation
Coefficient
+ standard deviation

Source: Authors calculations.
For OECD countries alone, we also find a positive and statistically significant relationship
between public capital and output for OECD in the one-year interval specification. The
coefficients for L, K, and G have values that are consistent with those found in the literature.
Public capital is significant with a coefficient of 0.13, and the coefficients attached to labor
and private capital are also significant and in the plausible range. When moving to five and
ten-year time intervals, the coefficient on private capital is the only one to remain significant,
and the estimate increases to 0.5. At the same time, the fit of the model improves when
moving to higher time intervals, a feature that cannot be related to the literature since earlier
studies do not report measures of fit.
For non-OECD countries, the standard model has a poor fit in the one-year interval
specification, but the fit improves significantly when moving to the five and ten-year
interval. While public capital remains insignificant in most specifications, private capital and
labor are highly significant and the estimated coefficients increase with time intervals. In the
five-year interval the coefficient on labor is estimated to be 0.5, whereas that on private
capital is 0.2, in line with the one-year interval estimate obtained for OECD countries.
Extended model
The extended model produces estimates that are statistically more significant than those
produced by the standard model, while the fit of the model also improves. For non-OECD
17

countries, the coefficient on public capital, c, becomes statistically significant, while the
coefficient on the interaction term, d, is negative, as expected.
For OECD countries, the coefficient c is estimated to be 0.13 (using one-year interval),
however, when moving to higher time intervalsjust as in the standard modelthe
coefficients for public capital lose significance. For example, in the five-year interval the
estimated coefficient for public capital in OECD countries is smaller (0.08) and not
significant. The interaction variable dthat captures non-linearitiesis not significant in the
case of OECD countries.
For non-OECD countries, the coefficient c on public capital averages 0.2 and is significant in
all specifications. Its value and significance increases with the length of time interval, from
0.12 in the one-year specification to 0.26 in the ten-year specification. The coefficient d
attached to the interaction term is negative and significant in all specifications, and averages
-0.5. This suggests that in some instances, a significant scaling up of investment may not
yield positive returns, depending on the size of the existing public capital stock. Evaluated at
the 25 percentile of public capital stock for non-OECD countries (public capital is around
30 percent of GDP), the marginal effect of additional public capital is positive (0.1), the
effect eventually reduces to zero around the 50
th
percentile (public capital is 50 percent of
GDP), before turning slightly negative, reaching -0.06 at the 75
th
percentile (public capital is
70 percent of GDP).
Finally, the fit of the model improves as one moves from one-year to five and ten-year
intervals. This indicates that in non-OECD countries, capital spending impacts growth over a
period of time in reflection of indivisibility of investments.
In summary, once we enter the interaction term in the standard model, we find the following:
(i) the fit of the regressions improve; (ii) the impact of public capital on growth becomes
positive also for non-OECD countries; (iii) a negative relationship between growth and the
level of public capital is uncovered, in particular for non-OECD countries. These results are
robust to a variety of alternative specifications and to inclusion and exclusion of outliers.
16

The results suggest that some non-OECD countries may not absorb a large scaling up of
capital investment, reflecting implementation weaknesses, and that weak absorptive capacity
should be taken into account in setting borrowing limits for non-concessional loans.
Furthermore, from a policy perspective, the results also suggest that public investment can be
used to boost aggregate demand in OECD countries, while it can boost aggregate supply in
non-OECD countries.

16
We also estimated specifications in which we scaled the capital stock to the population (G/L), and modeled
the diminishing returns by including a quadratic term of the public capital variable. In each of these cases, the
concave curvature of the elasticity of public capital was confirmed, with sample extreme points very similar to
the ones presented.
18

Table 6. Regression Results for All Countries

ln L 0.213* 0.219* 0.216*
(0.128) (0.127) (0.127)
ln K 0.204*** 0.176*** 0.169***
(0.041) (0.043) (0.043)
IG/GDP 0.018
(0.035)
ln G 0.049**
(0.022)
ln G [1
st
quartile] 0.032
(1.342)
ln G [2
nd
quartile] 0.183***
(2.769)
ln G [3
rd
quartile] 0.072
(0.866)
ln G [4
th
quartile] 0.028
(0.336)
Countries 44 44 44
Observations 1782 1782 1782
R-squared (within)
0.02 0.02 0.02
R-squared (between)
0.47 0.52 0.50
Hausman test (p-val) 0.13 0.03 0.00
ln K=ln G (p-val) 1/ 0.02 0.87
ln L+ln K+ln G=1 (p-val) 1/ 0.00 0.00
ln L 0.285 0.327 0.337
(0.213) (0.214) (0.214)
ln K 0.366*** 0.273** 0.258***
(0.082) (0.108) (0.109)
IG/GDP -0.016*
(0.008)
ln G 0.076 -0.096
(0.084) (0.176)
ln G * (G/GDP) 0.868
(0.781)
Countries 22 22 22
Observations 282 282 282
R-squared (within)
0.10 0.09 0.09
R-squared (between)
0.10 0.64 0.13
Hausman test (p-val) 0.06 0.08 0.06
ln K=ln G (p-val) 0.26 0.28
ln L+ln K+ln G=1 (p-val) 0.12 0.15
Regression Results for OECD and Non OECD Countries
(3 year intervals)
Standard model (interacted)
(1 year intervals)
Dependent variable: ln GDP
Naive model Standard model

19

Table 6. Regression Results for All Countries (concluded)

ln L 0.357** 0.337** 0.334**
(0.164) (0.163) (0.165)
ln K 0.278*** 0.249*** 0.266***
(0.056) (0.058) (0.059)
IG/GDP -0.020
(0.201)
ln G 0.040
(0.028)
ln G [1
st
quartile] 0.036
(1.177)
ln G [2
nd
quartile] 0.215**
(2.301)
ln G [3
rd
quartile] -0.093
(-0.953)
ln G [4
th
quartile] -0.010
(-0.095)
Countries 44 44 44
Observations 347 347 347
R-squared (within)
0.10 0.11 0.13
R-squared (between)
0.50 0.52 0.41
Hausman test (p-val) 0.13 0.03 0.06
ln K=ln G (p-val) 1/ 0.00 0.65
ln L+ln K+ln G=1 (p-val) 1/ 0.02 0.32
ln L 0.957*** 0.897*** 0.874***
(0.185) (0.192) (0.197)
ln K 0.238*** 0.201** 0.190**
(0.080) (0.083) (0.085)
IG/GDP -0.109
(0.462)
ln G 0.042
(0.040)
ln G [1
st
quartile] 0.032
(0.744)
ln G [2
nd
quartile] 0.099
(0.854)
ln G [3
rd
quartile] 0.098
(0.703)
ln G [4
th
quartile] 0.076
(0.524)
Countries 173 173 173
Observations 44 44 44
R-squared (within)
0.26 0.27 0.27
R-squared (between)
0.40 0.41 0.44
Hausman test (p-val) 0.13 0.03 0.03
ln K=ln G (p-val) 1/ 0.13 0.57
ln L+ln K+ln G=1 (p-val) 1/ 0.46 0.44
1/ In extended model, coefficient of the second quartile.
Regression Results for OECD and Non OECD Countries
(10 year intervals)
Note: Standard errors in parenthesis, statistical significance at the 10, 5 and 1 percent levels denoted by *, ** and *** respectively.
Hausman test indicates p-value of the null hypothesis that the difference in estimated coefficients between the fixed effect and a
random effects specification (not reported) is not systematic. L: Labor; K: Private Capital Stock; IG/GDP: Public Investment to
GDP ti G P bli C it l St k
Standard model (interacted)
(5 years intervals)
Dependent variable: ln GDP
Naive model Standard model
20

Table 7. Regression Results for OECD Countries

ln L 0.392** 0.391** 0.392**
(0.164) (0.164) (0.164)
ln K 0.365*** 0.253*** 0.255***
(0.065) (0.082) (0.082)
IG/GDP -0.002
(0.002)
ln G 0.129** 0.132**
(0.064) (0.065)
ln G * (G/GDP) -0.000
(0.002)
Countries 22 22 22
Observations 892 892 892
R-squared (within) 0.05 0.06 0.06
R-squared (between) 0.03 0.63 0.61
Hausman test (p-val) 0.25 0.02 0.02
ln K=ln G (p-val) 0.34 0.35
ln L+ln K+ln G=1 (p-val) 0.16 0.17
ln L 0.140 0.192 0.173
(0.225) (0.227) (0.227)
ln K 0.499*** 0.401*** 0.422***
(0.087) (0.118) (0.118)
IG/GDP -0.038**
(0.016)
ln G 0.064 0.084
(0.089) (0.090)
ln G * (G/GDP) -0.003
(0.002)
Countries 22 22 22
Observations 174 174 174
R-squared (within) 0.22 0.20 0.21
R-squared (between) 0.09 0.68 0.10
Hausman test (p-val) 0.08 0.38 0.17
ln K=ln G (p-val) 0.08 0.08
ln L+ln K+ln G=1 (p-val) 0.11 0.14
ln L 0.248 0.317 0.265
(0.309) (0.312) (0.310)
ln K 0.573*** 0.496*** 0.531***
(0.119) (0.163) (0.162)
IG/GDP -0.082
(0.051)
ln G 0.045 0.065
(0.116) (0.116)
ln G * (G/GDP) -0.004
(0.002)
Countries 22 22 22
Observations 87 87 87
R-squared (within) 0.40 0.37 0.40
R-squared (between) 0.04 0.62 0.12
Hausman test (p-val) 0.48 0.86 0.51
ln K=ln G (p-val) 0.08 0.07
ln L+ln K+ln G=1 (p-val) 0.60 0.60
(5 years intervals)
(10 year intervals)
Note: Standard errors in parenthesis, statistical significance at the 10, 5 and 1 percent levels denoted by *, ** and ***
respectively. Hausman test indicates p-value of the null hypothesis that the difference in estimated coefficients between the
fixed effect and a random effects specification (not reported) is not systematic. L: Labor; K: Private Capital Stock; IG/GDP:
Public Investment to GDP ratio; G: Public Capital Stock.
Regression Results for OECD Countries (different time intervals)
Dependent variable: ln GDP
Naive model Standard model Extended model
(1 year intervals)

21

Table 8. Regression Results for Non-OECD Countries
ln L 0.224 0.221 0.206
(0.174) (0.174) (0.173)
ln K 0.170*** 0.150*** 0.143***
(0.052) (0.054) (0.054)
IG/GDP -0.001
(0.067)
ln G 0.034 0.123***
(0.025) (0.035)
ln G * (G/GDP) -0.342***
(0.095)
Countries 26 26 26
Observations 1043 1043 1043
R-squared (within) 0.01 0.01 0.03
R-squared (between) 0.25 0.34 0.08
Hausman test (p-val) 0.61 0.02 0.00
ln K=ln G (p-val) 0.08 0.77
ln L+ln K+ln G=1 (p-val) 0.00 0.00
ln L 0.511** 0.491** 0.435**
(0.223) (0.224) (0.214)
ln K 0.251*** 0.235*** 0.168**
(0.071) (0.073) (0.071)
IG/GDP -0.077
(0.370)
ln G 0.021 0.183***
(0.032) (0.049)
ln G * (G/GDP) -0.460***
(0.109)
Countries 26 26 26
Observations 202 202 202
R-squared (within) 0.10 0.10 0.18
R-squared (between) 0.24 0.28 0.08
Hausman test (p-val) 0.80 0.14 0.10
ln K=ln G (p-val) 0.02 0.89
ln L+ln K+ln G=1 (p-val) 0.27 0.33
ln L 1.189*** 1.123*** 0.984***
(0.227) (0.242) (0.211)
ln K 0.163 0.134 0.030
(0.104) (0.108) (0.096)
IG/GDP -0.032
(0.799)
ln G 0.036 0.256***
(0.046) (0.060)
ln G * (G/GDP) -0.644***
(0.129)
Countries 26 26 26
Observations 100 100 100
R-squared (within) 0.31 0.32 0.44
R-squared (between) 0.16 0.18 0.21
Hausman test (p-val) 0.05 0.01 0.00
ln K=ln G (p-val) 0.46 0.64
ln L+ln K+ln G=1 (p-val) 0.24 0.58
Regression Results for Non-OECD Countries (different time intervals)
(10 year intervals)
Note: Standard errors in parenthesis, statistical significance at the 10, 5 and 1 percent levels denoted by *, ** and ***
respectively. Hausman test indicates p-value of the null hypothesis that the difference in estimated coefficients between the
fixed effect and a random effects specification (not reported) is not systematic. L: Labor; K: Private Capital Stock; IG/GDP:
Public Investment to GDP ratio; G: Public Capital Stock.
Extended model
(1 year intervals)
(5 years intervals)
Dependent variable: ln GDP
Naive model Standard model

22

C. Post Estimation Tests
First, even after controlling for non-linearities, non-OECD countries have a slightly lower
coefficient for public capital than OECD countries (0.123 versus 0.132) in the one-year
specification. All else being equal, this could be attributed to lower spending efficiency in
non-OECD countries. However, the difference is not statistically significant.

Second, private capital has a higher coefficient than public capital in all models for both
OECD and non-OECD countries. This is consistent with other studies that find private
investment to be more productive than public investment (Khan and Kumar 1997). However,
the difference is not significant for OECD countries and is significant only in non-OECD
countries when using the standard model.

Third, the estimation results with one year intervals for OECD countries (standard model)
yield elasticities of output with respect to public capital of 0.13, private capital of 0.25, and
labor input of 0.39. Therefore, one cannot reject the null hypothesis that the production
function has constant returns to scale (a+b+c=1). The estimation results for non-OECD
countries (extended model, one year interval) are also reasonable with elasticities of output
with respect to public capital of 0.12, private capital of 0.14, and labor input of 0.20 (all
coefficients are significant at the 5 percent level, except for labor). However, one can reject
the null hypothesis that the production function in non-OECD countries displays constant
returns to scale in the one year interval. Results for non-OECD countries differ slightly if
five-year intervals are considered. In this case, in line with results for OECD countries, the
null hypothesis of constant returns to scale cannot be rejected, and the estimated coefficients
move somewhat closer to the findings for OECD countries.

Finally, we ran Hausman tests, finding that the random effect specification is rejected in most
cases.

V. ALTERNATIVE DEPRECIATION RATES
We vary the depreciation rates used in the construction of the public capital stock. We run
three scenarios. In each scenario, depreciation rates are the same for both OECD and non-
OECD countries. In the first scenario, they are time-varying and equal to the ones used by
Kamps (2006), which increase from 2.5 to 4.0 percent between 19602001. In the second
scenario, they are constant and equal to the average rate of 3.25 percent over the same
period. In the third scenario, they are constant and equal to the maximum rate of 4.0 percent.

The results are provided in Tables 9 and 10. Using alternative depreciation rates does not
change the results significantly. For OECD countries, the estimated coefficient c varies
between 0.129 and 0.110 in the standard model, and 0.22 and 0.25 in the extended model.
For non-OECD countries, c varies between 0.21 and 0.036 in the standard model and 0.09 to
0.2 in the extended model. More importantly, in all cases, c and d remain statistically
significant, despite the use of different depreciation rates.
23

Table 9. Sensitivity Analysis: Alternative Depreciation Rates for OECD
Countries
ln L 0.392** 0.391** 0.392**
(0.164) (0.164) (0.164)
ln K 0.365*** 0.253*** 0.255***
(0.065) (0.082) (0.082)
IG/GDP -0.002
(0.002)
ln G 0.129** 0.132**
(0.064) (0.065)
ln G * (G/GDP) -0.000
(0.002)
Countries 22 22 22
Observations 892 892 892
R-squared (within) 0.05 0.06 0.06
R-squared (between) 0.03 0.63 0.61
Hausman test (p-val) 0.25 0.02 0.02
ln K=ln G (p-val) 0.34 0.35
ln L+ln K+ln G=1 (p-val) 0.16 0.17
ln L 0.392** 0.394** 0.388**
(0.164) (0.164) (0.164)
ln K 0.365*** 0.276*** 0.285***
(0.065) (0.079) (0.079)
IG/GDP -0.002
(0.002)
ln G 0.110* 0.131**
(0.063) (0.065)
ln G * (G/GDP) -0.003
(0.002)
Countries 22 22 22
Observations 892 892 892
R-squared (within) 0.05 0.06 0.06
R-squared (between) 0.03 0.63 0.08
Hausman test (p-val) 0.25 0.02 0.01
ln K=ln G (p-val) 0.19 0.22
ln L+ln K+ln G=1 (p-val) 0.17 0.22
ln L 0.392** 0.394** 0.391**
(0.164) (0.164) (0.164)
ln K 0.365*** 0.267*** 0.274***
(0.065) (0.079) (0.079)
IG/GDP -0.002
(0.002)
ln G 0.115* 0.129**
(0.059) (0.060)
ln G * (G/GDP) -0.002
(0.002)
Countries 22 22 22
Observations 892 892 892
R-squared (within) 0.05 0.06 0.06
R-squared (between) 0.03 0.63 0.29
Hausman test (p-val) 0.25 0.02 0.01
ln K=ln G (p-val) 0.22 0.24
ln L+ln K+ln G=1 (p-val) 0.16 0.20
(Scenario 2)
(Scenario 3)
Note: Standard errors in parenthesis, statistical significance at the 10, 5 and 1 percent levels denoted by *, ** and ***
respectively. Hausman test indicates p-value of the null hypothesis that the difference in estimated coefficients between the
fixed effect and a random effects specification (not reported) is not systematic. L: Labor; K: Private Capital Stock; IG/GDP:
Public Investment to GDP ratio; G: Public Capital Stock. Scenaro 1: Time-varying Kamps depreciation rates for public capital
from 2.5 to 4.0 percent; Scenaro 2: Constant depreciation rate at 3.25 percent; Scenaro 3: Constant depreciation rate at 4.0
percent.
Regression Results for OECD Countries (different depreciation rates)
Dependent variable: ln GDP
Naive model Standard model Extended model
(Scenario 1)


24

Table 10. Sensitivity Analysis: Alternative Depreciation Rates for
Non-OECD Countries
ln L 0.224 0.221 0.206
(0.174) (0.174) (0.173)
ln K 0.170*** 0.150*** 0.143***
(0.052) (0.054) (0.054)
IG/GDP -0.001
(0.067)
ln G 0.034 0.123***
(0.025) (0.035)
ln G * (G/GDP) -0.342***
(0.095)
Countries 26 26 26
Observations 1043 1043 1043
R-squared (within) 0.01 0.01 0.03
R-squared (between) 0.25 0.34 0.08
Hausman test (p-val) 0.61 0.02 0.00
ln K=ln G (p-val) 0.08 0.77
ln L+ln K+ln G=1 (p-val) 0.00 0.00
ln L 0.224 0.222 0.201
(0.174) (0.173) (0.173)
ln K 0.170*** 0.148*** 0.143***
(0.052) (0.054) (0.054)
IG/GDP -0.001
(0.067)
ln G 0.037 0.116***
(0.025) (0.035)
ln G * (G/GDP) -0.301***
(0.096)
Countries 26 26 26
Observations 1043 1043 1043
R-squared (within) 0.01 0.01 0.02
R-squared (between) 0.25 0.35 0.11
Hausman test (p-val) 0.61 0.02 0.00
ln K=ln G (p-val) 0.09 0.70
ln L+ln K+ln G=1 (p-val) 0.00 0.00
ln L 0.224 0.224 0.204
(0.174) (0.173) (0.173)
ln K 0.170*** 0.145*** 0.143***
(0.052) (0.054) (0.054)
IG/GDP -0.001
(0.067)
ln G 0.035 0.102***
(0.021) (0.031)
ln G * (G/GDP) -0.313***
(0.102)
Countries 26 26 26
Observations 1043 1043 1043
R-squared (within) 0.01 0.01 0.02
R-squared (between) 0.25 0.35 0.12
Hausman test (p-val) 0.61 0.02 0.00
ln K=ln G (p-val) 0.08 0.55
ln L+ln K+ln G=1 (p-val) 0.00 0.00
Regression Results for Non-OECD Countries (different depreciation rates)
(Scenario 3)
Note: Standard errors in parenthesis, statistical significance at the 10, 5 and 1 percent levels denoted by *, ** and ***
respectively. Hausman test indicates p-value of the null hypothesis that the difference in estimated coefficients between the
fixed effect and a random effects specification (not reported) is not systematic. L: Labor; K: Private Capital Stock; IG/GDP:
Public Investment to GDP ratio; G: Public Capital Stock. Scenaro 1: Time-varying Kamps depreciation rates for public capital
from 2.5 to 4.0 percent; Scenaro 2: Constant depreciation rate at 3.25 percent; Scenaro 3: Constant depreciation rate at 4.0
percent.
Extended model
(Scenario 1)
(Scenario 2)
Dependent variable: ln GDP
Naive model Standard model

25

VI. CONCLUSIONS
This paper revisits the debate on the effect of public investment on growth by estimating a
production function for forty-eight OECD and non-OECD countries, using capital stock as
the explanatory variable. The results indicate that increases in public capital stock are
positively correlated with growth, after controlling for the initial level of public capital. The
effect is stronger for OECD countries in the short-term, while it is stronger for non-OECD
countries in the long-term. This is in contrast with the mixed results obtained by studies that
mainly use the investment rate as the explanatory variable.

In some countries, the positive impact of public capital on output partially or wholly offset if
the initial capital stock is high in relation to GDP. However, these considerations do not seem
to matter in countries with a relatively low public capital stock. An important by-product of
this study is the construction of a public capital stock series for 26 middle- and low-income
countries. Such estimates are already available for 22 OECD countries.

A number of policy implications could be drawn from these results. First, while debate on
fiscal space has centered on creating room in the budget for higher public investment, the
results show that certain types of constraints (financing or the ability to absorb) can limit
growth benefits of higher capital stock. Second, developing countries can avail of non-
concessional foreign borrowing to finance new investments, provided these resources are
invested in projects that have been subjected to a rigorous cost-benefit analysis and hold
strong prospects for enhancing future growth. However, unlike OECD countries, the benefits
tend to accrue over time. This would necessitate extending the timeframe of debt
sustainability frameworks so that they can take into account the positive effects of public
investments. Finally, public investment projects included in fiscal stimulus packages for
200910 should increase growth, even with lags in their implementation, provided public
capital levels are not too high to begin with and the resulting financing costs and high tax
rates do not negate the positive benefits of new public investments.




26

References

Albala-Bertrand, J. M., (2004), Can the Composition of Capital Constrain Potential Output?
A Gap Approach, Queen Mary University of London, Department of Economics,
Working Paper No. 510.

Albala-Bertrand, J. M., and Mamatzakis, E. C., (2004), The Impact of Public Infrastructure
on the Productivity of the Chilean Economy, Review of Development Economics,
(8:2), pp. 266278.

Arestoff, Florence, and Christophe Hurlin, (2006), Estimates of Government Net Capital
Stocks for 26 Developing Countries, 19702002, Policy Research Working Paper
Series 3858, The World Bank.

Aschauer, D. A., (1989), Is Public Expenditure Productive? Journal of Monetary
Economics, 23, 177200.

Aschauer, D. A., (1998), How Big Should the Public Capital Stock Be? The Jerome Levy
Economics Institute of Bard College Public Policy, No:43.

Barro, Robert, (1990), Government Spending in a Simple Model of Exogenous Growth,
Journal of Political Economy, 98, S103S125.

Batina, R. G., (1998), On the Long-run Effects of Oublic Capital and Disaggregated Public
Capital on Aggregate Output, International Tax and Public Finance, (5:3), pp. 263
281.

Bonaglia, F., La Ferrara, E., and Marcellino, M., (2001), Public capital and economic
performance: Evidence from Italy, IGIER Working Paper No. 163.

Bosc, J. E., Escriba, J., and Murgui, M. J., (2000), The Effect of Public Infrastructure on
the Private Productive Sector of Spanish Regions, Journal of Regional Science, (42),
pp. 301326.

Bose, N., Haque, M. E., and Osborne, D., (2003), Public Expenditure and Economic
Growth. A disaggregated Analysis for developing Countries, Centre for Growth and
Business Cycle Research, Economic Studies, University of Manchester; Unpublished.

Cadot, O., Rller, L. H., and Stephan, A., (1999), A Political Economy Model of
Infrastructure Allocation: An Empirical Assessment, CEPR Discussion Paper No.
2336.

Cadot, O., Rller, L. H., and Stephan, A., (2002), Contribution to Productivity or Pork
Barrel? The Two Faces of Infrastructure Investment, WZB Discussion Paper No.
0209.

Caldern, C., Easterly, W., and L. Servn, (2004), Latin Americas Infrastructure in the Era
of Macroeconomic Crises, in The Limits of Stabilization: Infrastructure, Public
Deficits and Growth in Latin America, Easterly W. and Servn L. (eds), Stanford
University Press, California.
27


Caldern, C., and Servn, L., (2002), The output cost of Latin Americas Infrastructure
Gap, Central Bank of Chile Working Paper No. 186.

Canaleta, C. G., Arzoz, P. P., and Grate, M. R., (1998), Public Capital, Regional
Productivity and Spatial Spillovers, Universidad Pblica de Navarra, Lan Gaiak
Departamento de Economa, Working Paper No. 9811.

Canning, David, 1999, Infrastructures Contribution to Aggregate Output, The World Bank
Policy Research Working Papers Series #2246, (November 30, 1999).

Canning, D., and Bennathan, E., (2000), The Social Rate of Return on Infrastructure
Investments, World Bank Working Paper No. 2390.

Canning, D., and Pedroni, P., (1999), Infrastructure and Long-run Economic Growth,
Mimeo.

Charlot, S., and Schmitt, B., (1999), Public Infrastructure and Economic Growth in Frances
Regions, Paper No.129 for ERSA 39th Congress, Dublin, Ireland.

Cohen, J.P. and Morrison Paul, C.J. (2004), Public Infrastructure Investment, Interstate
spatial Spillovers, and Manufacturing Costs, The Review of Economics and
Statistics, (86:2), pp. 551560.

Collier, Hoeffler, and Catherine Pattillo, (2001), Flight Capital as a Portfolio Choice, The
World Bank Economic Review, Vol. 15, No. 1 (2001), pp. 5580

Commission on Growth and Development, 2008, The Growth Report: Strategies for
Sustained Growth and Inclusive Development.

Crowder, W.J., and Himarios, D., (1997), Balanced Growth and Public capital: An
Empirical Analysis, Applied Economics, (29:8), pp.10451053.

Demetriades, P. O., and Mamuneas, T. P., (2000), Intertemporal Output and Employment
Effects of Public Infrastructure Capital: Evidence from 12 OECD economies,
Economic Journal, (110), pp. 687712.

Dessus, S., and Herrera, R., (2000), Public Capital and growth Revisited: a Panel Data
Assessment, Economic Development and Cultural Change, 48, 407418.

Devarajan S., Swaroop, V., and Zou, H. F., (1996), The Composition of Public Expenditure
and Economic Growth, Journal of Monetary Economics, (37), pp. 313344.

Duggal, V. G., Saltzman, C., and Klein, L. R. (1999), Infrastructure and Productivity: a
Nonlinear Approach, Journal of Econometrics, (92), pp. 4774.

Easterly, William, and R., Levine, (2001), It's Not Factor Accumulation: Stylized Facts and
Growth Models, World Bank Economic Review 15 (2001), pp. 177219.

28

Easterly, William, and Servn Luis, 2004, The Limits of Stabilization: Infrastructure, Public
Deficits and Growth in Latin America, Stanford University Press, California.

Esfahani, H., and Ramres, M. T., (2003), Institutions, Infrastructure and Economic
Growth, Journal of Development Economics, (70), pp. 443477.

Everaert, Greetje., (2003), Balanced Growth and Public Capital: an Empirical Analysis with
I(2) Trends in Capital Stock Data, Economic Modelling, (20), pp. 741763.

Everaert, Greetje., and Heylen, F., (2004),, Public Capital and Long-Term Labour Market
Performance in Belgium, Journal of Policy Modelling, (26), pp. 95112.

Fernald, J. (1999), Assessing the Link Between Public Capital and Productivity, American
Economic Review, (89:3), pp. 619638.

Ferrara, E. L. and Marcellino, M., (2000), TFP, Costs, and Public Infrastructure: An
Equivocal Relationship, IGIER Working Paper No. 176.

Flores de, Frutos, R., Garcia-Diez, M., and Perez-Amaral, T., (1998), Public Capital and
Economic Growth: An Analysis of the Spanish Economy, Applied Economics,
(30:8), pp. 985994.

Ghali, K.H. (1998), Public Investment and Private Capital Formation in a Vector Error-
Correction Model of Growth, Applied Economics, (30), pp. 837844.

Gupta, Sanjeev, Benedict Clements, Emanuele Baldacci, and Carlos Mulas-Granados,
(2005), Fiscal Policy, Expenditure Composition, and Growth in Low-Income
Countries, Journal of International Money and Finance, 24, 441463.

Gwartney, J., Holcombe, R. G., and Lawson, R., (2004), Institutions and the Impact of
Investment on Growth, Paper Presented at the Conference of the Association Private
Enterprise Education, APEE, (April), Bahamas.

Haque, M. E., (2004), The Composition of Government Expenditure and Economic Growth
in developing Countries, Global Journal of Finance and Economics, 1, 97117.

Heller, Peter S., 2005, Understanding Fiscal Space, IMF Policy Discussion Papers (May 4,
2005).

Heston, Alan, Robert Summers, and Bettina Aten, (2006), Penn World Table Version 6.2,
Center for International Comparisons of Production, Income and Prices at the
University of Pennsylvania, September 2006.

Holtz-Eakin, D., and Schwartz, A. E., (1995), Infrastructure in a Structural Model of
Economic Growth Regional Science and Urban Economics, (25), pp. 131151.

Horton Mark, Mohan Kumar, and Paolo Mauro, (2009): The State of Public Finances: A
Cross-Country Fiscal Monitor, IMF Staff Position Note 09/21 (Washington:
International Monetary Fund).

29

Im, K. S., M. H. Pesaran, and Y. Shin, (2003) Testing for Unit Roots in Heterogeneous
Panels, Journal of Econometrics (115), pp. 5374.

IMF, 2004, Public Investment and Fiscal Policy, available at
http://www.imf.org/external/np/fad/2004/pifp/eng/pifp.pdf.

IMF, 2005a, Understanding Fiscal Space, IMF Policy Discussion Paper, (March 2005)

IMF, 2005b, Public Investment and Fiscal PolicyLessons from the Pilot Country
Studies, International Monetary Fund, (April 2005).

IMF, 2009, World Economic Outlook, (April 2009).

Kamps, C., (2006), New Estimates of Government Net Capital stocks for 22 OECD
Countries 19602001, IMF Staff Papers (53:1), pp. 120150.

Kamps, C., (2004b), The Dynamic Effects of Public Capital: VAR Evidence for 22 OECD
Countries, Kiel Institute of World Economics Working Paper No. 1224.

Kemmerling, A., and Stephan A. (2002), The Contribution of Local Public Infrastructure to
Private Productivity and Its Political Economy: Evidence from a Panel of Large
German Cities, Public Choice, (113), pp. 403422.

Levin, A., C.-F. Lin, and C.-S. J. Chu. (2002), Unit Root Tests in Panel Data: Asymptotic
and Finite-Sample Properties Journal of Econometrics (108), pp. 124.

Ligthart, J. E. (2002), Public Capital and Output Growth in Portugal: An empirical
Analysis, European Review of Economics and Finance, (1:2), pp. 330.

Mamatzakis, E.C. (1999a), Public Infrastructure, Private Input Demand, and Economic
Performance of the Greek Industry, Queen Mary & Westfield College Working
Paper No. 406.

Mamatzakis, E.C., (1999b), Testing for Long run Relationship Between Infrastructure and
Private Capital Productivity: A Time Series Analysis for the Greek Industry,
Applied Economics Letters, (6:4), pp. 243246.

Marquetti, Adalmir and Foley, Duncan, 2008, Manuscript.

Milbourne, R., Otto, G., and Voss, G. (2003), Public Investment and Economic Growth,
Applied Economics, (35:5), pp. 527540.

Mittnik S., and T. Neumann, (2001), Dynamic Effects of Public Investment: Vector
Autoregressive Evidence From Six Industrialized Countries, Empirical Economics,
Springer, vol. 26(2), pages 429446

Moreno, R., Lpez-Bazo, E., and Arts, M., (2003), On the Effectiveness of Private and
Public Capital, Applied Economics, (35), pp. 727740.

Nehru, Vikram and Ashok Dhareshwar , 1993, A New database on Physical Capital Stock:
Sources, Methodology and Results, Revista de Anlisis Econmico, 8(1), pp. 3759.
30


Perotti, Roberto, 2005, Estimating the Effects of Fiscal Policy in OECD Countries, CEPR
Discussion Paper No. 4842 (London: Centre for Economic Policy Research).

Pereira, A. M., (2000), Is All Public Capital Created Equal? Review of Economics and
Statistics, (82:3), pp. 513518.

Pereira, A. M., (2001), On The Effects of Public Investment on Private Investment: What
Crowds in What? Public Finance Review, (29:1), pp. 325.

Pereira, A. M., and Andraz, J. M., (2001), On the Impact of Public Investment on the
Performance of U.S. Industries, Public Finance Review, (31:1), pp. 6690.

Pereira, A. M., and Flores de Frutos, R., (1999), Public Capital Accumulation and Private
Sector Performance, Journal of Urban Economics, (46:2), pp. 300322.

Pereira, A. M., and Roca Sagales, O., (2003), Spillover Effects of Public Capital Formation:
Evidence from the Spanish Regions, Journal of Urban Economics, (53:2), pp. 238
256.

Pritchett, Lant, 1996, Mind Your Ps and Qs: The Cost of Public Investment is Not the
Value of Public Capital, Policy Research Working Paper 1660
(Washington: World Bank).

Romp, W., and J. de Haan, 2005, Public Capital and Economic Growth: a Critical Survey,
EIB Papers 2/2005, European Investment Bank, Economic and Financial Studies.

Roy, R., A. Heuty, and E. Letouze, 2006, Fiscal Space for Public Investment: Towards a
Human Development Approach, Paper Prepared or the G-24 Technical Meeting,
Singapore, September 1314, 2006.

Sanchez-Robles, B., (1998), Infrastructure Investment and Growth: Some Empirical
Evidence, Contemporary Economic Policy, (16), pp. 98108.

Seung, C. K., and Kraybill, D. S. (2001), The Effects of Infrastructure Investment: A Two
Sector Dynamic Computable General Equilibrium Analysis for Ohio, International
Regional Science Review, (24:2), pp. 261281.

Shioji, E., (2001), Public Capital and Economic Growth: A convergence Approach,
Journal of Economic Growth, (6), pp. 205227.

Stephan, A., (2000), Regional Infrastructure Policy and Its Impact on Productivity: A
Comparison of Germany and France, Applied Economics Quarterly, (46), pp. 327
356.

Stephan, A., (2003), Assessing the Contribution of Public Capital to Private Production:
Evidence from the German Manufacturing Sector, International Review of Applied
Economics, (17), pp. 399418.

31

Sturm, J. E., Jacobs, J., and Groote, P., (1999), Output Effects of Infrastructure Investment
in the Netherlands, 18531913, Journal of Macroeconomics, (21:2), pp. 355380.

Vijverberg, W.P.M., Vijverberg, C.P.C., and Gamble, J.L., (1997), Public Capital and
Private Production Review of Economics and Statistics, (79:2), pp. 267278.

Voss, G.M., (2002), Public and Private Investment in the United States and Canada,
Economic Modelling, (19), pp. 641664.

World Bank, 2007, Fiscal Policy for Growth and Development Further Analysis and Lessons
from Country Studies, March 28, 2007.

Zandi, Mark, 2008, Financial Shock, Pearson Education, INC., New Jersey.





3
2



Appendix Table: Literature Survey

Study Countries Sample Variable / Concept Specification Conclusion
Albala-Bertrand &
Mamatzakis (2004)
Chile 1960-98 Infrastructure capital stock Translog production function Infrastructure capital growth appears to reduce
productivity slightly up to 1971. From 1972 onwards,
the reverse seems true
Albala-Bertrand (2004) Chile and Mexico,
regions
1950-2000 Infrastructure capital stock Gap approach using a Leontief production function
(with private and public capital as inputs)
In Chile potential output is mostly constrained by
shortages of normal capital, in Mexico infrastructure
is the binding factor
Bosc et al. (2000) Spain, regions 1980-93 Infrastructure capital stock Generalized Leontief Elasticity is 0.08
Cadot et al. (1999) France, regions 1985-91 Infrastructure capital stock Production function combined with policy equation
for transport
infrastructure
Elasticity is 0.10
Cadot et al. (2002) France, regions 1985-92 Infrastructure capital stock Cobb-Douglas production function combined with
policy equation for transport infrastructure
Elasticity is 0.08
Caldern & Servn (2002) 101 countries 101 countries Infrastructure capital stock Cobb-Douglas production function with different
types of infrastructure as separate factor
Elasticity is 0.16
Canaleta et al. (1998) Spain, regions 1964-91 Infrastructure capital stock Flexible cost function Public capital reduces private production costs,
public and private capital factors are complementary;
Canning & Bennathan
(2000)
62 countries 1960-90 Infrastructure capital stock Cobb-Douglas and translog production function with
different types of infrastructure as separate factor
On average, only the low- and middle-income
countries benefit from more infrastructure
Canning (1999) 57 countries 1960-90 Infrastructure capital stock Cobb-Douglas production function with different
types of infrastructure as separate factor
Electricity and transportation routes have normal
capitals rate of return, telephone above normal
Canning and Pedroni (1999) Panels of countries
with different size
1950-92 Infrastructure capital stock Tests whether infrastructure has long-run effect on
growth based
on dynamic error-correction model
Evidence of long-run effects running from
infrastructure to growth, but results differ across
countries and type of infrastructure.
Cohen and Morrison Paul
(2004)
US, states 1982-1996 Infrastructure capital stock Generalized Leontief Infrastructure investment reduces own costs and
increases cost reducing effect of adjacent states
Fernald (1999) US, 29 sectors 1953-89 Infrastructure capital stock Sectoral productivity growth taking network
approach.
Roads contribute about 1.4 percent per year to
growth before 1973 and about 0.4 percent
Flores de Frutos et al.
(1998)
Spain 1964-92 (A) Infrastructure capital stock VARMA (first differences log
levels)
Transitory increase of public capital growth implies a
permanent increase of output, private capital and
employment
Holtz-Eakin & Schwartz
(1995)
US states 1971-86 Infrastructure capital stock Neo-classical growth model that separates
adjustment effects from steady state effects
Infrastructure has a negligible effect on output
nowadays
Kemmerling & Stephan
(2002)
87 large German
cities
1980, 1986 and
1988
Infrastructure capital stock Cobb-Douglas production function combined with
policy equation for transport infrastructure and
investment function for private capital
Rate of return on infrastructure is 16%. Political color
is important determinant for receiving grants
Mamatzakis (1999a) Two digit Greek
industries (20)
1959-90 Infrastructure capital stock Translog cost function Cost elasticity of public infrastructure ranges from
0.02% in food manufacturing to 0.78% in wood and





3
3



Appendix Table: Literature Survey (continued)

Study Countries Sample Variable / Concept Specification Conclusion
Moreno et al. (2003) Spain, regions and
sectors
198091 Infrastructure capital stock Translog cost function Public and private investments increase efficiency.
Pereira & Roca Sagales
(2003)
Spain (regional and
national)
1970-95 (A) Infrastructure capital stock VAR, first differences log levels Positive and significant long-run effects on output,
employment, and private capital
Stephan (2000) West-German and
French regions
Germany: 1970-
95,
France: 1978-92
Infrastructure capital stock Cobb-Douglas production function with public capital
as separate factor and translog production function
Cobb Douglas gives elasticity of 0.11. Translog
specification runs into multicolinearity problems.
Stephan (2003) West-German
regions (11)
1970-96 Infrastructure capital stock Cobb-Douglas production function with public capital
as separate factor
Elasticity between 0.38 (first differences) and 0.65
(log levels)
Vijverberg et al. (1997) US, time series 1958-89 Infrastructure capital stock Cobb-Douglas and semi-translog Result are not reliable due to multicollinearity
Vijverberg et al. (1997) US 1958-89 Infrastructure capital stock Translog cost and profit functions Both cost and profit function estimates suffer from
multicollinearity
Batina (1998) US 1948-93 (A) Public capital stock VAR and VECM Public capital has long-lasting effects on output and
vice versa
Bonaglia et al. (2001) Italy, regions 1970-94 Public capital stock Cobb-Douglas production function with public capital
as separate
factor
Elasticity is 0.05 (insignificant) for Italy as a whole,
large variation between regions
Bonaglia et al. (2001) Italy, regions 1970-94 Public capital stock Cobb-Douglass variable cost function Inconclusive, no good measure of the social user
cost of public capital available
Charlot & Schmitt (1999) France, regions 198293 Public capital stock Cobb-Douglas and translog production function with
public capital as separate factor
Elasticity is 0.3 (Cobb-Douglas), 0.4 (translog), but
very sensitive to region and period
Crowder and Himarios
(1997)
US 1947-89 (A) Public capital stock VECM Public capital is at the margin slightly more
productive or as productive as private capital
Demetriades & Mamuneas
(2000)
12 OECD countries 1972-91 Public capital stock Quadratic cost function Output elasticity varies from 2.06 (Norway) to 0.36
(UK)
Duggall et al. (1999) USA, national 1960-89 Public capital stock Production function, technology index is non-linear
function of infrastructure and time trend
Elasticity for infrastructure is 0.27
Everaert (2003) Belgian regions 1953-96 (A) Public capital stock VECM Output elasticity of public capital is 0.14, which is
only a fraction (0.4) of output elasticity of private
capital
Ferrara & Marcellino (2000) Italy, total and
per region
1970-94 Public capital stock Cobb-Douglas production function with physical
capital stocks as separate input
Italy: Negative output elasticity in the 1970s, positive
in the 1980s and 1990s. Regions: Negative in North-
West and North-East, positive in Centre and
South.
Ferrara & Marcellino (2000) Italy, total and regions 1970-94 Public capital stock Cobb-Douglas and generalized Leontief with
physical capital stocks as separate input
Public capital is cost increasing over the whole
sample (only cost decreasing in the 90s). Suggests
over-investment in public capital
Ghali (1998) Tunesia 1963-93 Public capital stock VECM Public investment has a negative effect on growth
Kamps (2006) 22 OECD countries 1960-2001 Public capital stock Aschauer (1989) model for individual countries and
panel
Elasticity is 0.22 in panel, but much higher in time-
series models'
Kamps (2004) 22 OECD countries 1960-2001 (A) Public capital stock VECM For majority of countries there is a positive and
significant effect on growth
Ligthart (2002) Portugal 1965-95 Public capital stock Cobb-Douglas production function, with and without
CRS
Positive and significant output effects of public
capital
Ligthart (2002) Portugal 1965-95 (A) Public capital stock VAR (log levels) Positive output effects of public capital
Mamatzakis (1999b) Greece 1959-93 Public capital stock VECM Positive effect of public capital on productivity, no
reverse effect




3
4



Appendix Table: Literature Survey (concluded)

Study Countries Sample Variable / Concept Specification Conclusion
Pereira and Flores de
Frutos
(1999)
US 1956-89 (A) Public capital stock VAR, first differences log levels Public capital is productive, but substantially less
than suggested by Aschauer (1989)
Seung & Kraybill (2001) Ohio Calibrated
on 1990
Public capital stock Computable general equilibrium model with
congestion adjusted infrastructure as third factor in
Cobb-Douglas production function
Welfare effects of infrastructure are non-linear
Shioji (2001) US states and
Japanese regions
US: 1963-93, 5
year interval,
Japan: 1955-95 (5
year interval)
Public capital stock Computable general equilibrium model with public
capital in the technology term of a Cobb-Douglas
production function
Elasticity between 0.10 and 0.15
Bose et al (2003) 30 LICs, MICs, 1970-90 Public investment Significant positive effect
Dessus and Herrera (2000) 29 LICs and MICs 1981-91 Public investment Significant positive effect
Devarajan et al. (1996) 43 LDCs Public investment Significant negative effect
Everaert & Heylen (2004) Belgian regions 1965-96 Public investment Translog production function. Using a general
equilibrium model, they analyze labor market effects
Elasticity is 0.31
Gupta et al. (2005) 39 LICs 1990s Public investment Significant positive effect
Gwartney et al. (2004) 86 countries of which
66 LDCs
1980-2000 Public investment Significant positive effect, but coefficient is less than
coefficient of private investment
Haque (2004) 33 LICs, MICs 1970-99 Public investment Significant positive effect
Milbourne, Otto and Voss
(2003)
74 countries 1960-85 Public investment Not significantly different from zero in steady state
model; in transition model with IV also not
significantly different from zero
Mittnik & Neumann (2001) Canada, France,
UK, Japan, The
Netherlands and
Germany
Different periods
per
country (Q)
Public investment VECM Weak positive output effect of infrastructure, public
investment induces private investment; no reverse
causation from GDP to public capital
Pereira & Andraz (2001) US (sectoral and
national)
1956-97 (A) Public investment VAR, first differences log levels Public investment positively affects private
investment, employment and output
Pereira (2000) US 1956-97 (A) Public investment VAR, first differences log levels Positive effect through crowding in of private
investment
Pereira (2001) US 1956-97 (A) Public investment VAR, first differences log levels All types of public investment are productive, but
core infrastructure displays the highest rate of
return
Sturm et al. (1999) Netherlands 1853-1913 Public investment VAR, levels Positive and significant short-run effect; no long-run
effect
Voss (2002) US and Canada US: 1947-88 (Q);
Canada: 1947-96
(Q)
Public investment VAR (11 lags); first differences Public investment tends to crowd out private
investment
Source: Romp and de Haan (2005), IMF (2004) and World Bank (2007)