KYKLOS, Vol. 58 – 2005 – No.

3, 307–330

Interactions Between Public and Private Investment:

Evidence from Developing Countries

Erdal Atukeren

I. INTRODUCTION

Whether public and private sector investments are substitutes, or comple-

ments, has been the subject of strong controversy in economic theory and
policy. Free markets advocates argue that government intervention in the
economy should be minimised. Among many reasons for this argument is the
view that state sector activity competes with private sector for scarce resources
and drives their prices up. For example, if public sector investments are
financed by borrowing, this leads to an increase in market interest rates and
thus raises the cost of capital for the private sector. Hence, some private sector
projects become unprofitable; in other words, they are crowded out. In the case
of tax financing of public sector investments, the taxes may distort the resource
allocation decisions of private actors in the economy by changing relative
prices.
On the other side of the coin, it is argued that public investments may indeed
be beneficial for the development of the private sector. The government sector,
for example, can afford to invest in infrastructure projects with large sunk costs,
which require long lead times to become profitable. The private sector may
benefit from the spillovers from such public sector investments during and after
the completion of the project. For instance, a better developed infrastructure
for roads and railways reduces transportation costs, and hence facilitates a
better business environment. Furthermore, public infrastructure investments
in education and health care facilities help improve the level and the quality of
human capital in an economy. In addition, as an aggregate demand manage-
ment tool, government investments might be used as a counter-cyclical eco-
nomic policy measure to smooth the business cycle and revitalise the private
 Swiss Institute for Business Cycle Research (KOF / ETH Zurich), Weinbergstrasse 35, WEH – ETH
Zentrum, CH-8092 Zurich, Switzerland. E-mail: [email protected]. I would like to thank,
without implication, the participants of the ECOMOD 2004 Conference in Paris, France, Jan Egbert-
Sturm, and an Editor of this Journal for their valuable comments and suggestions on earlier versions of
this paper.

r 2005 Blackwell Publishing Ltd., 9600 Garsington Road, Oxford OX4 2DQ, UK 307
and 350 Main Street, Malden, MA 02148, USA
 ERDAL ATUKEREN

sector activity – at least in the short run. Last but not least, the crowding-out
arguments explained in the paragraph above are based on the assumption that
the economy operates from a point on its production possibilities frontier and
that it has well-developed and efficiently functioning financial markets. These
conditions are not always fulfilled – especially in developing countries. Some
private sector investments might as well not be financed if financial markets are
shallow. In such situations, public sector investments could play the role of
catalyst in providing the economy with much needed and otherwise hard to
undertake investments. As a result, the private sector and the economy in
general may benefit from public sector investments1.
A large body of theoretical and empirical literature exists, examining the
connection between the accumulation of physical capital and economic
growth, derived from the traditional neoclassical or the more recent endogen-
ous growth theory frameworks (e.g. Lucas 1988). A comprehensive account of
the literature on economic growth is provided in Barro and Sala-i-Martin
(2003). Physical capital accumulation (investment) is a factor of production,
and higher investment shares in GDP are usually associated with higher per
capita income growth rates (e.g. Barro 1991, Mankiw et al. 1992, among
others)2. Nevertheless, the investment variable entering into cross-country
economic growth regressions generally combines both private and government
investments. Then, the question is whether public and private investments have
the same marginal productivities. As Shioji (2001) discusses, the evidence is
mixed3.
The policy implications of the possible differences in the marginal produc-
tivities of private and public capital are important. If private sector investments

1. In the discussion about the possible beneficial effects of public sector involvement in the functioning of
the economy, we restrict ourselves to productive investments only. This excludes other categories of
public spending, such as wages and salaries, subsidies and unproductive government consumption
items. Some of these expenditure items may be used as counter-cyclical policy measures and thus help
smooth business cycles, but the effect of such expenditures on private sector investments is another
topic to investigate.
2. Levine and Renelt (1992), however, find that most of the explanatory variables suggested by the
theory prove to be ‘fragile’, i.e. they do not pass a robustness test in cross-country growth regression.
They conclude that the only robust variable is trade openness. Levine and Renelt (1992) used extreme
bounds analysis (EBA) to test whether a variable is fragile or not. The EBA methodology has also
come under scrutiny since then, and some of Levine and Renelt’s results were reversed. See Temple
(2000) for a review of the issues, and Bleaney and Nishiyama (2002) for a comparison of the sensitivity
of estimates for different types of growth models.
3. Khan and Reinhart (1990) and Khan and Kumar (1997), for example, find that private capital has
higher marginal productivity. The productivity differential, however, is reported to vary over time and
across countries. Shioji (2001) examines the effects of public capital investments in a dynamic open
economy framework. It is shown that public infrastructure investments positively contribute to the
convergence of regional incomes in the USA and Japan. Gramlich (1994) and Poot (2000) provide
comprehensive surveys of the impact of different categories of government expenditures on economic
growth.

308 r 2005 Blackwell Publishing Ltd.

 INTERACTIONS BETWEEN PUBLIC AND PRIVATE INVESTMENT

have higher marginal productivity, an increase in the size of the public sector at
the expense of the private sector might hinder economic growth and well-being,
even when the share of total investments in GDP remains unchanged. Or, even
when public infrastructure investments have lower (but positive) marginal
productivity, if they crowd in private investments, they help enhance the
economy’s productive capacity. Then, the question of whether public sector
investments crowd in or crowd out private sector investments stands as an
important empirical question to examine4.
Despite the presence of a large number of papers investigating the relative
importance of private and public investments, and their sub-components in
explaining economic performance, the number of studies on the interactions
between public and private investments in developing countries is relatively
small. Even so, the method of analysis is usually restricted to single- or cross-
country contemporaneous regressions. Given that the public sector is more
likely to undertake large projects with long completion and lead times, such
studies may underestimate the effects of public investments on private sector
activity. Furthermore, as Easterly and Levine (2001) argue, the positive
association between investment and economic growth may also result from a
reverse causal relationship, where it is rather economic growth which causes
capital accumulation.
This paper attempts to fill a gap in this literature by analysing the interactions
between private and public investments by using long-term cointegration
analysis and Granger-causality tests on a sample of developing countries,
using consistently compiled private and public investment data. Furthermore,
based on the results of cointegration and Granger-causality tests, we run a
probit regression, where the dependent variable takes the value ‘1’ for the
‘crowding-out’ cases, and ‘0’ otherwise, and the explanatory variables are
various components of Gwartney and Lawson’s (2004) economic freedom of
the world index. By focussing on some of the proxies for the socio-economic
and political environment of business, this approach allows us to estimate
under which conditions public investments might be more likely to crowd out
private investments.
The rest of the paper is organised as follows. Section II discusses the data and
empirical methodology. Section III presents the results from cointegration
and Granger-causality tests, and discusses some of the results in view of the

4. The following are some of the studies that directly or indirectly provide insights on different aspects of
the crowding-in and crowding-out effects of public investments: Ahmed (1986), Ahmed and Miller
(2000), Aschauer (1989), Barro (1974, 1981, 1990, 1991), Bende-Nabende and Slater (2003), Blejer and
Khan (1984), Buiter (1977), Cruz and Teixieira (1999), Edwards (1992), Erenburg and Wohar (1995),
Fisher (1993), Greene and Villanueva (1991), Khan and Kumar (1997), Khan and Reinhart (1990),
Lynde and Richmond (1993), Onliner et al. (1995), Otto and Voss (1996, 1998), Sundararajan and
Thakur (1980), Voss (2002), among others. See also Everhart and Sumlinski (2001, Table 2.2).

r 2005 Blackwell Publishing Ltd. 309

 ERDAL ATUKEREN

previous literature. Section IV puts results from Section III in a new perspective
by means of a probit analysis, as explained above. Section V concludes.

II. DATA AND METHODOLOGY

The data on private and public investment are taken from the International
Finance Corporation’s (IFC) ‘Trends in Private Investment (TPI)’ publication,
authored by Everhart and Sumlinski in 2001. The TPI is being published
for more than a decade now, providing cross-country comparable data on
investment trends for a large number of developing countries. The capital
expenditures of state-owned enterprises are classified as private investment by
many sources. The TPI data improve on this measure. As Everhart and
Sumlinski (2001, Foreword ) explain, their methodology ‘y accounts all
investment undertaken by the public sector – including through state enter-
prises – as public sector investments’. Due to the long-term nature of our
analysis, we restricted our sample to only those countries with availability of
longer spans of data (from the early 1970s on). This left us with 25 countries. Of
these countries, 12 are Latin American, eight are Asian, and five are African
countries. The investment data are represented as a percentage of GDP.
Expressing the private and public investment data in terms of their shares of
GDP has the advantage of controlling for GDP growth. In a growing economy,
both public and private investment might be increasing, and one may
spuriously detect a positive relationship between them if GDP growth is not
controlled for.
It should be noted that the contemporaneous relationship between private
and public investment has not been stable over the last three decades. In El
Salvador, for example, the correlation coefficient is about zero for the
1970–2000 period; but it changes from 10.88 in the 1970s to
0.89 in the
1980–1989 period, and becomes a strong plus again (10.65) in the 1990s. The
sample period used in this paper includes many external shocks, debt crises,
structural adjustment programmes, further liberalisation episodes, and a drive
to cut the size of the public sector through rationalisation and privatisation in
many developing countries. It is, therefore, hard to detect robust relationships,
especially by employing contemporaneous correlation-based analysis. In
addition, correlation does not necessarily mean causation. Even if one uses a
lead-lag correlation analysis between the variables of interest, a statistically
significant lag of variable X on another variable Y does not indicate that X
causes Y.
In this study, we use Granger’s (1969) definition of causality and its
extensions. Before doing so, however, we test whether a long-run equilibrium
relation exists between public and private investments. The first step for the

310 r 2005 Blackwell Publishing Ltd.

 INTERACTIONS BETWEEN PUBLIC AND PRIVATE INVESTMENT

cointegration analysis is testing whether the variables in question are station-

ary, i.e. I(0), or they are integrated, i.e. I(1) or higher. We use Phillips and
Perron’s (1988) unit root test (PP) since it has less strict assumptions about the
behaviour of the test equation’s error term. In the PP test, serial correlation and
heteroscedasticity are also taken into account, which may be relevant in our
case. If both the private and public investment variables are found to be
I(1) processes, then it is possible that they are also cointegrated. Normally,
the combination of two I(1) series are I(1). However, there a special para-
meter configuration may exist which leads to an error term with I(0)
properties. If this is the case, then the series in question are said to be co-
integrated, and they have the property that, despite some divergences from
time to time, they move together and never diverge in the long run. To test
whether private and public investment series are cointegrated, we use the
battery of tests discussed in detail by Johansen (1995). For the countries where
a cointegrating relationship is detected, we present the normalised coefficients
for the equation and also calculate the resulting error correction term. This
term is then used as a factor driving the variables towards a long-term
equilibrium, or ‘correcting’ short-term divergences from a long-run equili-
brium, in the subsequent Granger-causality tests.
The notion of causality received much coverage in statistics, economics,
history, and philosophy. Here, we use the term causality in a pragmatic sense.
According to Granger (1969), if the history (lags) of a weakly stationary
variable X can be used to better predict another weakly stationary variable Y,
given all other relevant information, then X is a prima facie Granger-cause of
Y. In other words, for a variable X to Granger-cause another variable Y, the
information contained in the history of X should reduce the forecast error
variance of Y, when all other relevant information has already been accounted
for. Of course, how the lag structures are determined is another practical
problem to deal with. This has, indeed, important implications on the final
results. We deal with this problem by using Schwarz’s (1978) ‘Bayesian
Information Criterion (SBIC)’ in combination with a subset autoregression
and transfer function approach, as outlined by Penm and Terrell (1984) and
Kang (1989) in choosing the optimal lag orders.
The Granger-causality tests are carried out on the first differences of the ratio
of private and public investment to GDP. This means that we are rather
concerned with how a change in the ratio of public (private) investment to GDP
affects private (public) investment. If there is causal evidence that public
investment crowds in private investment, this will be confirmed as a finding that
an increase in public investment to GDP ratio will also increase the ratio of
private investment to GDP. That is, private investment will also grow above the
GDP growth rate. This is, of course, a stricter approach than just testing
whether an increase in public investments leads to an increase or decrease in

r 2005 Blackwell Publishing Ltd. 311

 ERDAL ATUKEREN

private investments. Nevertheless, since GDP growth may be a factor linking

everything together, it should be either explicitly controlled for (a trivariate
relationship), or it should be a common denominator to the variables under
investigation. We choose the latter option to keep the analysis simpler. Another
point supporting the bivariate framework is Hsiao’s (1982) result, that for a
variable X to be a direct Granger-cause of another variable Y, X should
Granger-cause Y in all systems: bivariate and multivariate. Thus, for public
investment to be a direct cause of crowding-out of private investment, public
investment should crowd out private investment in a bivariate system in the
first place.
After testing for the time series properties of the variables, testing for
cointegration, and running the Granger-causality tests, we are in a position to
evaluate the findings in an overall framework. The literature on the effects of
public investments on private investments contains mixed findings. Some
results show crowding-in for certain countries, while some others show
crowding-out in other countries. Even for the same country, there are
conflicting results coming from different studies, which use different methods
and analyse different time periods. One way to make sense out of such diverse
results is to run a meta-analysis that controls for the distinct features of each
study. Another way is to make use of the results obtained in the same study. We
follow the latter path, since the data are collected on the same basis, and a
common methodology and a similar estimation sample period are used. The
main question we address is the identification of the conditions under which
crowding-out is more likely to occur. We do this by employing a probit analysis
framework. The dependent variable is formed by denoting the cases where
crowding-out is detected in cointegration or Granger-causality tests as ‘1’ and
denoting all other cases as ‘0’. A priori, one may expect that economies with a
large public sector and those with rather closed regimes are likely to be
inefficient. Hence, the private sector may not flourish under those conditions.
In such economically less free constellations, public investments might be more
likely to crowd out private investments. The studies by Everhart and Sumlinski
(2001), Lambsdorf (2003), and Pellegrini and Gerlagh (2004), among others,
investigate the effects of corruption on investments and economic activity. One
of their main results is that corruption negatively affects investments and
economic growth by lowering the quality of public investments and by
introducing uncertainty or additional costs for private investors.
Given the importance of ‘economic freedom’, especially for creating a
suitable environment for the development of the private sector, we use the
economic freedom of the world index calculated by Gwartney and Lawson
(2004) on behalf of the Fraser Institute in Canada. There are several other
indices of economic freedom, general liberties and socio-political conditions.
We made use of the Fraser Institute’s index, since it was more available for our

312 r 2005 Blackwell Publishing Ltd.

 INTERACTIONS BETWEEN PUBLIC AND PRIVATE INVESTMENT

sample period (1970–2000) for all the countries under consideration, and it is
also available in sub-categories. Any such index, nevertheless, requires some
subjective judgement, but it should at least be applied consistently over time
and across countries. Then, by running probit regressions with the right-hand-
side variables representing different dimensions of economic liberties and other
socio-economic/political conditions, we obtain estimates showing under which
conditions public investments are likely to crowd out (or increase the marginal
probability of crowding-out) private investments. Of course, this approach is
subject to both the Type I error from the Granger-causality tests in addition to
any shortcomings, and also the Type I error from the probit regressions and
any bias in estimates due to possible missing variables. Nevertheless, this ap-
proach represents a new perspective to examine different country-specific
results in a common framework.

III. COINTEGRATION AND GRANGER-CAUSALITY

BETWEEN PUBLIC AND PRIVATE INVESTMENTS

1. Results from Unit Root and Cointegration tests

In this section, we first present the time-series properties of the GDP shares of
private and public investments in the sample of countries and time periods
as shown in Table 1. A few methodological notes about the selection of the
appropriate model in Phillips and Perron’s (1988) unit root test are in order. We
ran the tests with and without a trend variable. If the trend variable was not
significant, it was dropped. Furthermore, we selected the bandwidth (Newey-
West) by optimising Schwarz’s (1978) BIC. Employing the Phillips-Perron test
in this way, it has been found that the null hypothesis of a unit root in the
private investment / GDP process could be rejected at the 5% significance level
in Chile, Costa Rica, Ecuador, and Kenya. By the same methodology, the pub-
lic investment to GDP ratio has been found to be stationary in the Dominican
Republic and Bangladesh. Thus, these countries were excluded from the
cointegration analysis, since the combination of I(0) and I(1) series is I(1).
The last column of Table 1 reports the results from Johansen’s (1995)
cointegration tests. Both the trace and the maximum eigenvalue statistics are
reported. The form of the test employed has been selected individually for each
country, based on the minimum Schwarz BIC. In general, the number of lags to
capture autocorrelation was set to two in the test equations, since we have
annual data.
The findings indicate that private and public investments are cointegrated in
Brazil, India, Pakistan, Morocco, and South Africa. A long-tem relationship
between two series implies the existence of causality, at least in one direction.

r 2005 Blackwell Publishing Ltd. 313

 Table 1

314
Time Series Properties of the Private and Public Investment Variables
Private Investment / GDP (%) Public Investment / GDP (%) Cointegration
Sample Phillips-Perron test Sample Phillips-Perron test Johansen Test
st st
Level 1 diff. Level 1 diff. Trace Max EV
Argentina 1970–2000
2.21
8.45 1970–2000
2.27
4.27 9.61 7.23
Brazil 1970–1999
3.00
10.45 1970–1999
2.81
7.44 19.70 18.90
Chile 1970–2000
3.84
8.78 1970–2000
2.30
7.02 –
Colombia 1970–2000
2.68
6.28 1970–2000
2.40
7.33 16.82 11.83
Costa Rica 1970–1998
2.96
5.54 1970–1998
2.54
4.28 –
Dominican Rep. 1970–2000
2.83
6.14 1970–2000
3.08
7.96 18.15 11.72
Ecuador 1970–2000
5.15
9.93 1970–2000
1.57
4.19 –
El Salvador 1970–2000
2.07
4.53 1970–2000
2.67
5.81 9.03 6.16
Guatemala 1970–2000
1.67
5.07 1970–2000
2.25
5.01 16.60 11.32
Mexico 1970–2000
1.36
4.77 1970–2000
1.82
4.90 10.75 9.63
Paraguay 1970–2000
1.91
4.04 1970–2000
2.71
8.96 19.90 13.70
Uruguay 1970–2000
2.12
4.53 1970–2000
2.01
4.49 14.26 12.40
Bangladesh 1973–2000
2.40
5.47 1973–2000
3.83
4.66 –
ERDAL ATUKEREN

India 1970–1999
2.98
7.07 1970–1999
0.85
11.02 28.54 24.00

South Korea 1971–1999
1.99
3.46 1971–1999
2.34
3.90 12.32 10.39
Malaysia 1970–2000
2.00
3.93 1970–2000
2.11
3.79 14.07 8.51
Pakistan 1970–2000
2.67
5.64 1970–2000
1.39
3.51 32.54 28.36
Philippines 1975–2000
2.58
5.99 1975–2000
3.11
8.49 11.89 8.53
Thailand 1970–2000
1.47
3.42 1970–2000
1.87
3.47 13.80 10.48
Turkey 1970–2000
2.27
5.73 1970–2000
2.10
5.61 13.94 12.75
Kenya 1970–1999
3.75
7.15 1970
1999
3.37
6.42 –
Malawi 1973–2000
2.67
7.03 1973–2000
2.82
6.21 9.91 6.71
Morocco 1975–2000
2.68
5.47 1975–2000
3.10
6.41 38.36 27.91
South Africa 1970–1999
1.61
4.07 1970–1999
2.83
4.19 22.74 19.87
Tunisia 1970–1999
2.05
5.27 1970–1999
1.63
3.65 16.94 13.38

r 2005 Blackwell Publishing Ltd.

Note: (), (), and () denote significance levels of 1%, 5% and 10%, respectively. Max EV stands for Johansen’s maximum eigenvalue test.
 INTERACTIONS BETWEEN PUBLIC AND PRIVATE INVESTMENT

The estimates of the cointegrating relationships are given below for the in-
dividual countries. The variables PRI and PBI denote the shares of private and
public investments in GDP respectively, and TREND is a linear time trend.
The standard errors are given in parentheses. Note that there is no TREND
variable in the equations for Brazil and South Africa since the selection of
the optimal model specification in Johansen’s (1995) cointegration tests, via
minimising the SBIC, did not select a trend term. The long-run normalised
estimates of the cointegrating relationships are shown below.
Brazil:

PRI ¼ 17:67146
0:348509  PBI

ð1:2224Þ ð0:1975Þ

India:

PRI ¼ 0:255585  Trend
0:335567  PBI

ð0:1781Þ ð0:0913Þ

Pakistan:

PRI ¼ 0:335294  Trend þ 1:42130  PBI

ð0:4865Þ ð0:2747Þ

Morocco:

PRI ¼ 0:472261  Trend þ 1:829888  PBI

ð0:1391Þ ð0:3727Þ

South Africa:

PRI ¼ 9:383584 þ 0:361036  PBI

ð0:4293Þ ð0:04135Þ

The long-run parameter estimates indicate that public investments crowd

out private investment in Brazil and India, while a crowding in effect is found
for Pakistan, Morocco, and South Africa. These findings will be further dis-
cussed in combination with the results obtained from Granger-causality tests.

2. Results from Granger-Causality Tests

In testing for Granger-causality, we used the first differences of the PRI and
PBI variables. First differencing is necessary in the case of I(1) variables, but it
may not be required for the I(0) processes. Nevertheless, upon investigating the

r 2005 Blackwell Publishing Ltd. 315

 ERDAL ATUKEREN

correlogram of the I(0) cases, we found that the first-order autoregressive,

AR(1), terms are indeed high. Therefore, using first differencing should not
present a serious case of over-differencing. Treating all variables on the same
basis provides an ease of interpretation of the results. For the five countries
where cointegration was detected, however, we included the error correction
term in the specification.
The exact methodology used in testing for Granger-causality is as follows.
First, the best univariate specifications for PRI and PBI were selected by using
the Schwarz BIC. In doing so, we used a subset autoregression approach. This
approach has the advantage of suppressing any insignificant lags, which
leads to a gain in estimation efficiency. The maximum lag allowed, M, was
set to five. This represents a medium-term time horizon for the effects of
public or private investments to realise. As the second step, we introduced
the lags of PBI (PRI) on the best univariate specification for PRI (PBI).
We used a subset transfer function approach here as well, and suppressed
any insignificant lags. We inferred Granger (non-)causality from the resulting
best specifications on the basis of two approaches: 1) using minimum SBIC,
and 2) testing the joint significance of the coefficients. Table 2 presents the
best transfer functions of PRI (PBI) on PBI (PRI) including the error
correction term obtained from the cointegrating equations, where appropriate.
The specifications and details of the univariate models that represent the
strongest case for non-causality based on the Schwarz BIC are included
in the Appendix (Table A1). Where PBI (PRI) was found to Granger-cause
PRI (PBI), the respective signs of the coefficients were used to translate
the empirical estimates into qualitative conclusions on whether there are
crowding-out or crowding-in effects. As Sturm and De Haan (1998) discuss,
in a general VAR framework, summing up the coefficients on a variable ‘X ’ and
testing for their joint significance to decide if it Granger-causes another variable
‘Y ’ might indeed amount to a ‘neutrality test’. For instance, all the coefficients
on lagged ‘X ’ values may be individually statistically significant, but their sum
could still be zero if they cancel each other out due to opposite signs.
Furthermore, in the presence of opposing signs on the coefficients of ‘X ’, even
if the coefficients values do not sum to zero, the variable ‘Y ’ might still respond
to shocks in the variable ‘X ’ in a non-linear way, making the interpretation of
the linear sum of the coefficients difficult. These situations often arise in a
general VAR framework, where all lags of a variable up to the chosen optimal
lag are included in the specification. Then, the dynamic multipliers or impulse
response functions could provide a better view on the response of a variable to
shocks in another over time.
In this study, we use a subset autoregression and subset transfer function
modelling strategy, as discussed earlier. That is, we include only the significant
lags (in a statistical cost function sense) of a variable in the final specifications.

316 r 2005 Blackwell Publishing Ltd.

 INTERACTIONS BETWEEN PUBLIC AND PRIVATE INVESTMENT

Table 2
Best Transfer Function Specifications between the Shares of Private (PRI) and Public (PBI)
Investments in GDP
Ho: PRI does not Granger Ho: PBI does not Granger
cause PBI cause PRI
TF Spec., (Sign) SBIC TF Spec., (Sign) SBIC
Argentina t-1, (1) 2.9000z t-4, (1) 4.4460
Brazil t-5 (
) 3.4357 t-1, t-2, t-4, (1) 3.8905z
Chile t-3 (1) 3.4511 t-5, (1) 5.0613
Colombia t-2, t-5 (1) 3.2158 t-3, (1) 4.4990
Costa Rica t-4 (
) 2.5638 t-1, (
) 3.8986z
Dominican Rep. t-2 (
) 4.0769 t-2, (
) 4.6969
Ecuador t-1, t-3 (1) 3.8290z t-1, (1) 4.7381z
El Salvador t-1, t-2 (1) 2.3103z t-3, t-4, t-5, (
) 3.5156z
Guatemala t-3 (1) 2.9807 t-4, (
) 3.8682
Mexico t-1 (1) 3.2334 t-4, (
) 3.0526
Paraguay t-2 (1) 4.6246 t-2, (1) 4.9960
Uruguay t-2 (1) 2.7847 t-5, (1) 3.2340
Bangladesh t-1 (1) 2.7915z t-4, (1) 3.8885z
India t-3 (
) 2.0654z t-3 (
) 2.9906
South Korea t-1 (1) 1.2472z t-1, (
) 4.6126
Malaysia t-5 (
) 4.0651 t-2, (1) 5.5780z
Pakistan t-5 (
) 2.6564 t-2, (
) 1.5732z
Philippines t-5 (
) 3.2002 t-2, (1) 4.5461z
Thailand t-1 (1) 2.6917 t-1, (
) 4.7726z
Turkey t-1 (
) 2.9200 t-1, (1) 3.8761
Kenya t-1 (1) 2.8043 t-4, (
) 3.4558z
Malawi t-3 (1) 5.0435 t-1 (
) 5.0626
Morocco t-5 (1) 3.6026z t-2, t-3 (
) 4.0167z
South Africa t-1, t-4 (1) 2.0456z t-1, t-4, (
) 2.3954z
Tunisia t-3 (1) 4.3159 t-3 (
) 4.4744z
Notes: SBIC is the Schwarz’s (1978) Bayesian information criterion. SBIC 5 (ESS/T)T (k/T), where
ESS is the error sum of squares from estimation of the model in question, T is the sample size, and k is
the number of estimated parameters in the model.
() and () indicate statistical significance at 5% and 1%, respectively.
(z) indicates that the log(SBIC) obtained under the subset transfer function specification is less than
the SBIC value from the subset autoregression. That is, there is evidence for Granger-causality based
on the Schwarz criterion.

In many cases, this strategy led to the selection of one lag: hence, there is no
ambiguity on the coefficient. Also, in a few other cases, where more than one lag
was included in the specifications, no ambiguous coefficient behaviour that
would overturn the initial findings was found. The Granger-causal implica-
tions of the results presented in Table 2 are summarised in Table 3. Table 3
shows that, in 11 out of 25 countries, there is some evidence for crowding-out
effects from public investments to private investments. On the other hand, we
find some evidence for crowding-in effects in eight countries. In some cases,
indicated by () in Table 3, the results obtained from cointegration analysis and

r 2005 Blackwell Publishing Ltd. 317

 ERDAL ATUKEREN

Table 3
Qualitative Summary Results from Cointegration and Granger-Causality Tests
Direction of Granger-causality Countries
Public Investments crowd in Brazil (), Ecuador, Bangladesh, Malaysia, Paki-
Private Investments stan (), Philippines, Morocco, South Africa ()
Public Investments crowd out Brazil (), Costa Rica, El Salvador, Uruguay,
Private Investments India, Pakistan (), Thailand, Kenya, Morocco,
South Africa (), Tunisia
Private Investments crowd in Argentina, Colombia, Ecuador, El Salvador,
Public Investments Guatemala, Uruguay, Bangladesh, South Korea,
Thailand, Morocco (), South Africa
Private Investments crowd out India, Malaysia
Public Investments
No Causality in any direction Chile, Dominican Republic, Mexico, Paraguay,
Turkey, Malawi
Note: () indicates that the results from cointegration and Granger-causality tests are different.
Nevertheless, cointegration tests detect long-run relationships, while Granger-causality tests capture
rather short-term dynamics. Hence, different qualitative results from these tests do not imply a
conflict.

Granger-causality tests are different. Nevertheless, since Granger-causality

tests in an error correction model framework relate rather to short-term
dynamics, while the existence of cointegration indicates a long-term relation-
ship in levels, they should not be interpreted as conflicting results. For example,
there may be crowding-out in the short-run, say due to an increase in interest
rates, but as productive public investments start bearing fruit, the overall effect
on the private sector might be positive (crowding-in) in the longer term.
Another interesting aspect, and perhaps one that has not received much
attention in the literature, is the effect of private sector investments on the
public sector. According to the cointegration and Granger-causality test
results, private sector investments crowd in public investments in 11 countries.
Since the public and private sector investments are expressed as their shares in
the GDP, an increase in their first differences reflects a higher growth rate of
these investments than the GDP growth rate. This would result when an
increase in private investments also necessitates an increase in investments in
public works and infrastructure. A new factory, for example, might increase the
need for public infrastructure and services near that facility. In this case, public
and private investments have a complementary relationship in enhancing the
productive capacity of an economy.
On the other side of the coin, private investments crowd out public
investments in India and Malaysia. Examining the data from Everhart and
Sumlinski (2001), it is seen in the case of Malaysia that the public sector
investments amounted to about 10% in 1975, reached a high of 18% in the

318 r 2005 Blackwell Publishing Ltd.

 INTERACTIONS BETWEEN PUBLIC AND PRIVATE INVESTMENT

mid-1980s, and declined to about 11% in 2000 – with a high of 15% in the
mid-1990s. The private sector investments, on the other hand, were about 20%
in 1975 and, after some ups and downs, climbed up to 32% in the mid-1990s.
Nevertheless, the share of private sector investments in GDP came down to the
same level as public investments (11%) in 2000 after the Asian crisis. The
detected negative causality from private to public investments in this case must
be capturing the large increase in private investments between 1987 and 1997,
and a slight decrease in the public investments in the same period. In India, one
sees that public and private investment grew together from 1970 to the mid-
1980s, but since then the size of the public sector has decreased, while the
private sector continued to increase its share in GDP.
Coming back to the effects of public investment on private investment,
we detect more cases of crowding-out than crowding-in. One noteworthy
aspect of our findings is that, while mixed results have been obtained for
Latin America and Asia, crowding-out is found almost uniformly for the
African countries in the sample. On a country-specific basis, our results are the
opposite of Cruz and Teixeira (1999) for Brazil. Cruz and Teixeira (1999) found
that public investments crowd out private investment in the short-run, while
there is a crowding-in effect in the long-run cointegrating relationship. Their
study, however, did not use the IFC’s definition of public and private in-
vestments, and it was conducted for the 1947–1990 period. Public investment
to GDP ratio showed a steady decline from about 7% in the mid-1980s to about
3% in 2000 in Brazil, while the share of private investments to GDP fluctuated
around 16% between 1990 and 2000. This is detected as a non-diverging nega-
tive relationship, or crowding-out, in the long-run cointegration relationship.
Table 2.2 of Everhart and Sumlinski’s (2001) study further summarises the
main results of some of the studies on crowding-out and crowding-in effects. As
seen above, conflicting results might be obtained, even for the same country
in the literature. For example, both crowding-in and crowding-out were found
for Mexico, while we have not found any such effects in our study. Also for
Pakistan, where we find a long-run crowding-in but a short-run crowding-out
effect, Looney and Frederiken (1997) find crowding-in, while Sakr (1993) finds
that government expenditure on non-infrastructure components crowd out
private investment.
Overall, our findings are in line with the mixed evidence in the literature on
the interactions between public and private investments. That means that there
is no clear general verdict on whether public investments crowd out private
investments or not. The effects vary from country to country.
While previous studies conclude their analyses at this point, we take it a step
further by asking whether the countries where public investments crowd out
private investments share some common characteristics. Or, is it possible to say
that the likelihood that public investments will crowd out private investments

r 2005 Blackwell Publishing Ltd. 319

 ERDAL ATUKEREN

would be higher under certain circumstances? This analysis is the topic of the
next section.

IV. THE ROLE OF ECONOMIC FREEDOM IN EXPLAINING THE

CROWDING-OUT EFFECTS OF PUBLIC INVESTMENTS

1. Introduction

In this section, we analyse whether the crowding-out effects (or lack thereof) of
public investments, detected by cointegration or Granger-causality tests, can
be related to the general environment of economic freedom. The idea is to
employ a probit-based analysis, where the dependent variable is whether public
investments crowd out private investments (taking the value ‘1’) or not (taking
the value ‘0’)5. As candidate explanatory variables, we use Gwartney and
Lawson’s (2004) economic freedom of the world index and its sub-components.
We also include the urbanisation rate as a proxy for the level of infrastructure
development.
Before presenting the probit estimates, a few words about the economic
freedom of the world index (EFI) would be in order. The EFI is calculated for a
large number of countries, which includes all of the 25 countries in our study.
The overall EFI is a composite index of seven sub-categories and ranges from
zero (the worst or least free) to 10 (the best or most free). The index is calculated
by combining objective (e.g. inflation rate and variability, highest marginal tax
rate, share of government consumption in GDP, etc.) and subjective measures
(e.g. legal security, rule of law, etc.). The transformation of these measures on a
0–10 scale, and the weighing mechanism to obtain the overall index, might be
open to discussion, but we use it anyway since it is done on a consistent basis
for all countries and over time as far as possible6. As such, it represents
a consistent ranking. Further details on the calculation of the index can be
obtained from Gwartney and Lawson (2004). Table A2 in the Appendix lists
the components of the EFI.

5. The coding of the dependent variable in this way represents an overall average conclusion that has
tended to hold during the sample period rather than laying a strong claim that the stated causal
relationship between private and public investments in a given country always holds true. This is
similar to representing a given country’s real GDP growth rate by the sample period average in cross-
country growth regressions.
6. See Heckelman and Stroup (2000, 2002) and Sturm et al. (2002) for a discussion of the statistical issues
in the calculation of an economic freedom index.

320 r 2005 Blackwell Publishing Ltd.

 INTERACTIONS BETWEEN PUBLIC AND PRIVATE INVESTMENT

2. Cross-section Probit Estimates of the Likelihood of Crowding-out

In this section, we present the standard probit estimates of the impact of

economic freedom on the likelihood of crowding-out effects of public invest-
ments. The overall EFI is available from 1970 to 2003 in most cases. However,
some sub-components of EFI are not available in 1970 and 1975 for some of the
countries in our 25-country sample. Hence, we estimate the probit model using
the EFI index and its sub-indices starting at different time periods. The
estimation results are shown in the Appendix in Table A3.
The estimates presented in Table A3 indicate that neither the overall
economic freedom index nor the urbanisation rates alone explain the phenom-
enon of crowding-out in our sample. However, the ‘government size’, ‘mone-
tary policy and price stability’, and ‘international exchange or the trade
openness’ are found to have some predictive power in explaining the incidence
of crowding-out in bivariate probit regressions. A small government size is
represented by a high number on the 0–10 scale. The estimated coefficients on
this variable are negative and statistically significant. This indicates that the
higher the government intervention in an economy, the higher is the likelihood
of crowding-out effects on private sector investments. This result is, in principle,
in line with Guseh (1997) who found that a large government not only has an
adverse influence on economic growth in general, but also the negative impact
is higher in countries with lower levels of freedom. Better monetary conditions
and a more favourable general investment environment are indicated by a high
number on the 0–10 scale. For this variable, the estimated coefficient carries a
positive sign, with a tendency towards being slightly significant in 1990. This is
an interesting finding, since a positive coefficient implies that the better or the
more developed the macro-monetary environment is, the higher is the like-
lihood of crowding-out. This is in line with the textbook explanation of
crowding-out effects in developed countries, where increases in public invest-
ments or expenditures might crowd out private investments through the
interest rate channel. With respect to the openness to trade, the estimated
coefficients are negative and significant, especially in the 1990s. Since a high
number indicates a more liberal trade regime and higher share of exports and
imports in GDP, our findings imply that crowding-out is more likely to be
observed in economies with rather limited trade openness. This is interesting
since it indicates that, as international linkages increase, the effects of govern-
ment intervention on the economy decrease (e.g. fiscal policy might become
ineffective in an open economy, depending on the exchange rate regime, and so
crowding-out is less likely to occur). In addition, looking at the estimates in the
opposite way, the level of bureaucracy is likely to be high, and the state sector
might be more inefficient in countries with a more inward trade orientation.
This environment deters private investors and leads to a crowding-out effect.

r 2005 Blackwell Publishing Ltd. 321

 ERDAL ATUKEREN

Table 4
Multivariate Probit Model Estimates
Dependent Variable: CROWD-OUT
Method: ML – Binary Probit (BHHH)
No. of Observations: 25
Variable Coefficient Std. Error z-Statistic Prob.
Constant 70.93426 31.69344 2.238137 0.0252
Government Size in 1985
7.766842 3.841864
2.021634 0.0432
Government Size in 1990
2.799816 0.570567
4.907074 0.0000
Government Size in 1995 2.844692 1.063996 2.673592 0.0075
Trade Openness in 1980
2.604959 1.293408
2.014028 0.0440
Forex Regime in 1985
0.890040 0.461191
1.929871 0.0536
Monetary Conditions in 1985 0.987670 0.484417 2.038885 0.0415
Mean dependent var. 0.440000 Std. Dev. dependent var. 0.506623
Log likelihood
4.305824 Schwarz criterion 1.245751
Probability(LR stat.) 0.000255 McFadden R-squared 0.748906
Obs. with CROWD-OUT 5 0 14 Obs. with CROWD-OUT 5 1 11
Notes: The countries where the Crowd-out 5 1 are: Brazil, Costa Rica, El Salvador, Uruguay, India,
Pakistan, Thailand, Kenya, Morocco, South Africa, and Tunisia. The countries where the Crowd-out
variable 5 0 are: Argentina, Chile, Colombia, Dominican Republic, Ecuador, Guatemala, Mexico,
Paraguay, Bangladesh, South Korea, Malaysia, Philippines, Turkey, and Malawi. The following
observations should also be made: 1) In Uruguay, no crowding out hypothesis could be rejected only
at 10% significance level; 2) In Brazil, crowding out was detected in the long-run cointegrating
regression, while Granger-causality tests indicate crowding in effects in the short-run; and 3) In
Pakistan and South Africa, crowding-out was detected in Granger causality tests, while there is
evidence for crowding-in in the long-run cointegrating relationship.

Nevertheless, since these are preliminary results from bivariate probit

regressions, they may be biased due to missing variables that are not
orthogonal to the error term. Thus, we continue our analyses in a multivariate
framework. In constructing the multivariate model, we tried to integrate as
many different sub-categories of the EFI as possible, and selected the best
model based on the SBIC criterion – an approach consistent with the model
selection strategy employed in the earlier part of the paper. Nevertheless, the
search procedure has been subject to the limitations imposed by the sample size
and by computational considerations to estimate the model efficiently. Table 4
presents the estimation results.
In the final multivariate probit model, the following variables were found to
be significant in explaining the crowding-out effects of public investments:
‘Government size between 1985–1995’, ‘Monetary conditions in 1985’, ‘Forex
regime in 1985’, and ‘Financial Markets in 1980’. We have discussed the
interpretation of ‘government size’, ‘monetary conditions’ and ‘trade openness’
variables in the context of bivariate probit regressions above. In the multi-
variate analysis, the ‘forex regime’ variable, which represents the presence
and/or the degree of exchange rate restrictions and black markets in a country,
has a negative sign. The more liberal the exchange rate regime is, the higher the

322 r 2005 Blackwell Publishing Ltd.

 INTERACTIONS BETWEEN PUBLIC AND PRIVATE INVESTMENT

values of the ‘forex regime’ variable; thus, the negative coefficient represents
that there is more likelihood of crowding-out in a closed or controlled eco-
nomy. This may again represent bureaucratic inefficiency, corruption and black
market premia; that is, a rather uncertain environment for private investors.
In terms of the in-sample fit, the multivariate model has done strikingly well
with a prediction success of 13 out of 14 cases of no-crowding-out and 10 out
of 11 cases of crowding-out, using a cut-off probability of 50% to determine
success. Of the two unsuccessful model predictions, the cointegration and
Granger-causality tests did not indicate a crowding-out effect for Turkey, but
the probit analysis predicted a crowding-out probability of 54%. In the other
case, Thailand, no-crowding-out was falsely predicted (crowding-out prob-
ability of only 16%). In the case of Turkey, the mismatch is not so severe, as it
lies close to the 50% cut-off line. In the case of Thailand, Granger-causality
tests indicate a crowding-out effect from public investments, while there is
at the same time a crowding-in effect from private investments to public
investments. Therefore, the dynamics of the interactions between public and
private investments in Thailand turn out to be rather complex, and this may
have led to the false prediction.

V. CONCLUSIONS

In this paper, we analysed the crowding-out effects of public investments on

private investments by using time series (Granger-causality and cointegration

Table 5
Actual versus Predicted Crowding-out Probability
Country Actual Predicted
Argentina 0.0 0.01825
Brazil 1.0 0.78512
Chile 0.0 0.21478
Uruguay 1.0 0.69300
Bangladesh 0.0 0.27287
Malaysia 0.0 0.40892
Thailand 1.0 0.16979
Turkey 0.0 0.54599
Malawi 0.0 0.00662
South Africa 1.0 0.97722
Note: The actual values are derived from the results of cointegration and Granger-causality tests
and they are subject to the caveats discussed in the text. The predicted values have been obtained
from the regression estimates reported in Table 4. The predicted crowding-out probability was near
‘0’ (in line with the Actual) for Colombia, Dominican Republic, Ecuador, Guatemala, Mexico,
Paraguay, South Korea, and the Philippines. Similarly, for Costa Rica, El Salvador, India, Pakistan,
Kenya, Morocco, and Tunisia, the estimated crowding-out probability was near ‘1’, in line with the
Actual.

r 2005 Blackwell Publishing Ltd. 323

 ERDAL ATUKEREN

tests) and probit analysis in a sample of 25 developing countries. Our findings

indicate that both crowding-in and crowding-out effects of public investments
occur in developing countries. The effects vary from one country to another.
This is consistent with the mixed evidence found in the literature. Taking the
analysis one step further, we have attempted to find out some of the common
characteristics shared by the countries where public investments have been
found to crowd out private investment. This was done by means of a probit
analysis. As explanatory variables, we used an index of economic freedom
and its sub-categories. The model correctly predicted 10 out of 11 cases of
crowding-out, and 13 out of 14 cases of no-crowding-out. It was found that the
higher the share of government involvement in an economy, the lower the trade
openness, the more restrictions on the use of foreign currencies; and the more
stable and developed the macro and monetary environment is, the higher the
likelihood that public investments may crowd out private investments. Note
that the first three factors rather characterise an economy with more state
controls and inward orientation. In practice, this environment is likely to be
associated with an inefficiently functioning economy, with few incentives and
high uncertainties for private investors. Even in an economy which is liberal-
ised, with little state sector involvement and openness to foreign trade,
crowding-out may still occur, due to the effects of the financing requirements
of government investments through the financial markets. This is captured
by the macro-monetary environment factor we discussed above. This form
of crowding-out is one of the textbook cases illustrated rather for developed
economies. Furthermore, the macroeconomic and business environment
between 1980 and 1995 was found to be influential in explaining crowding-
out. Especially in the 1980s, many developing countries experienced debt crises
and undertook structural adjustment programmes. Since the structural adjust-
ment programmes were aimed at opening up the distressed economies, re-
ducing the state involvement in the economy, liberalising foreign exchange
and trade orientation, and deepening financial markets, despite their initial
negative effects on the economy, they might have also helped to strengthen the
private sector by reducing the crowding-out likelihood. After the mid-1990s,
most economies have much more favourable values (i.e. higher economic
freedom) for the variables we have in our model, which suggests that crowding-
out, when it occurs, is now more likely to be a through-the-financial markets
phenomenon. Given their still fragile economic and financial structures and
still to be improved institutions, however, capital account liberalisation, higher
trade openness, and a more important role for financial markets in emerging
market economies might also increase the risk of transmitting international
adverse shocks, or contagion, into their economies.

324 r 2005 Blackwell Publishing Ltd.

 INTERACTIONS BETWEEN PUBLIC AND PRIVATE INVESTMENT

APPENDIX
Table A1
Best Univariate Specifications for Shares of Private (PRI) and Public (PBI) Investments in GDP
d(PRI) d(PBI)
Specification SBIC Specification SBIC
Argentina t-3 4.4294 t-3, t-4 2.9956
Brazil ECT, t-4 4.1029 ECT, t-1 3.3405
Chile t-2 4.9862 t-2 3.3490
Colombia t-5 4.4522 t-3 3.1756
Costa Rica t-4 3.9387 t-4, t-5 2.4665
Dominican Rep. t-1 4.6572 t-3 4.0529
Ecuador t-1 4.7927 t-4 4.0732
El Salvador t-5 3.8261 t-1 2.4818
Guatemala t-4 3.7710 t-5 3.0696
Mexico t-3 3.0029 t-2 3.1398
Paraguay t-3 4.9079 t-2 4.5477
Uruguay t-3 3.2223 t-3 2.9485
Bangladesh t-2 3.9802 t-2 2.9669
India ECT, t-4 2.9051 ECT, t-4 2.1765
South Korea t-1, t-2 4.5361 t-3 1.6355
Malaysia t-1, t-3 5.5789 t-1, t-4 4.1041
Pakistan ECT, t-2 1.5968 ECT, t-1, t-2 2.5645
Philippines t-2 4.7165 t-5 3.1691
Thailand t-1 4.9858 t-1 2.8997
Turkey t-5 3.8474 t-3 2.8474
Kenya t-1, t-4 3.6697 t-3 2.8973
Malawi t-1 5.0188 t-3 4.9751
Morocco ECT, t-1 4.1601 ECT, t-3 3.6106
South Africa ECT, t-1, t-3 2.4605 ECT, t-4 2.2843
Tunisia t-3 4.4956 t-1 4.2914
Notes: ECT stands for the ‘error correction term’ from the cointegrating equation. t-i represents
the lag order. SBIC is the Schwarz’s (1978) Bayesian information criterion. SBIC 5 (ESS/T)
T (k/T), where ESS is the error sum of squares from estimation of the model in question, T is the
sample size, and k is the number of estimated parameters in the model.

r 2005 Blackwell Publishing Ltd. 325

 ERDAL ATUKEREN

Table A2
Components of The Fraser Institute’s Economic Freedom Index
I. Size of Government (Government Size)
a) Government Consumption
b) Transfers and Subsidies
II. Structure of the Economy and Use of Markets (Economic Structure)
a) Government Enterprises
b) Price Controls
c) Top Marginal Tax Rate
d) Conscription
III. Monetary Policy and Price Stability (Monetary Conditions)
a) Annual Money Growth
b) Inflation Variability
c) Recent Inflation Rate
IV. Freedom to Use Alternative Currencies (Forex Regime)
a) Ownership of Foreign Currency
b) Black Market Exchange Rate
V. Legal Structure and Property Rights (Legal )
a) Legal Security
b) Rule of Law
VI. International Exchange (Trade Openness)
a) Taxes on International Trade
i) Taxes as a Percentage of Exports and Imports
ii) Mean Tariff Rate
iii) Standard Deviation of Tariff Rates
b) Size of Trade Sector
VII. Freedom of Exchange in Financial Markets (Financial Markets)
a) Private Ownership of Banks
b) Extension of Credit to Private Sector
c) Avoidance of Negative Interest Rates
d) Capital Transactions with Foreigners
Source: Gwartney and Lawson (2004), Fraser Institute, Canada. The variables take a value between 0
(worst / least free) and 10 (best / most free). The short names in parentheses (in italics) are given by the
author, and used instead in the text and the tables.

326 r 2005 Blackwell Publishing Ltd.

 INTERACTIONS BETWEEN PUBLIC AND PRIVATE INVESTMENT

Table A3
Bivariate Probit Model Estimates
1975 1980 1985 1990 1995 2000
1
Overall Economic Freedom 0.1029
0.2634
0.1787
0.2239
0.2093
0.1563
Government Size
0.2939
0.4348()
0.6319
0.8909
0.3793()
0.3815
Economic Structure2
0.0713
0.0895
0.2116
0.1372
0.1897 0.0014
Monetary Conditions3 0.0863 0.0723 0.0550 0.1359() 0.0719 0.0962
Forex Regime
0.0275
0.1032
0.0961
0.0654
0.0643
0.1044
Legal4 – 0.0001 0.0714
0.0462 0.1175 0.0267
Trade Openness 0.1260
0.1021
0.1192
0.2034()
0.2226()
0.2942()
Financial Markets5 – 0.0091 0.1118 0.0494
0.1758
0.0831
Urbanisation 0.0071
0.0083
0.0010
0.0112
0.0110
0.0103
Notes: 1) El Salvador and Paraguay have missing data in 1975. 2) El Salvador, Paraguay, and
Bangladesh have missing data in 1975. 3) Bangladesh has missing data in 1980. 4) 1975 is excluded
from estimation due to missing data on eight countries. 5) 1975 is excluded from estimation due to
missing data on seven countries. Data on El Salvador is missing for 1980, while the data on Paraguay
is missing until 1990. Significant at: () 20%, 10%, 5%, 1%. Estimates for 1970 are not
available due to missing data for a large number of countries.

Table A4
Categorical Variable Statistics
Mean Crowd-out 5 0 Crowd-out 5 1 All
Government Size in 1985 8.428571 7.536364 8.036000
Government Size in 1990 8.578571 7.418182 8.068000
Government Size in 1995 8.114286 7.436364 7.816000
Trade Openness in 1980 4.257143 3.572727 3.956000
Forex Regime in 1985 5.114286 4.181818 4.704000
Monetary Conditions in 1985 5.771429 6.563636 6.120000
Standard Deviation Crowd-out 5 0 Crowd-out 5 1 All
Government Size in 1985 1.070175 0.768470 1.034762
Government Size in 1990 0.795350 0.984701 1.045116
Government Size in 1995 1.056014 1.131612 1.120521
Trade Openness in 1980 2.113588 2.138734 2.108530
Forex Regime in 1985 2.161857 2.895451 2.499580
Monetary Conditions in 1985 2.884555 3.363115 3.062815
No. of Observations 14 11 25

REFERENCES

Ahmed, Habib and Stephen M. Miller (2000). Crowding-Out and Crowding-In Effects of the
Components of Government Expenditure, Contemporary Economic Policy. 18(1): 124–133.
Ahmed, Shaghil (1986). Temporary and Permanent Government Spending in an Open Economy,
Some Evidence for the United Kingdom, Journal of Monetary Economics. 17(2): 197–224.
Aschauer, David A. (1989). Is Public Expenditure Productive?, Journal of Monetary Economics. 23:
177–200.
Barro, Robert J. (1974). Are Government Bonds New Wealth?, Journal of Political Economy. 81:
1095–1117.

r 2005 Blackwell Publishing Ltd. 327

 ERDAL ATUKEREN

Barro, Robert J. (1981). Output Effects of Government Purchases, Journal of Political Economy. 89:
1086–1121.
Barro, Robert J. (1990). Government Spending in a Model of Simple Endogenous Growth, Journal
of Political Economy. 98: 103–126.
Barro, Robert J. (1991). Economic Growth in a Cross-section of Countries, Quarterly Journal of
Economics. 106: 407–443.
Barro, Robert J. and Xavier Sala-i-Martin (2003). Economic Growth, 2nd Edition. MIT Press.
Bende-Nabende, Antony and Jim Slater (2003). Private Capital Formation: Short- and Long-run
Crowding-in (out) Effects in ASEAS, 1971–99, Economics Bulletin. 3(27): 1–16.
Bleaney, Michael and Akira Nishiyama (2002). Explaining Growth: A Contest between Models,
Journal of Economic Growth. 7: 43–56.
Blejer, Mario I. and Mohsin S. Khan (1984). Government Policy and Private Investment in
Developing Countries, IMF Staff Papers. 31: 379–403.
Buiter, Willem H. (1977). Crowding-out and the Effectiveness of Fiscal Policy, Journal of Public
Economics. 7: 109–128.
Cruz, Bruno de Oliveira and Joanı́lio R. Teixeira (1999). The Impact of Public Investment in Brazil,
1947–1990, CEPAL Review. 67: 75–84.
Easterly, William and Ross Levine (2001). It’s Not Factor Accumulation: Stylized Facts and Growth
Models, World Bank Economic Review. 15(2): 177–219.
Edwards, Sebastian (1992). Trade Orientation, Distortions and Growth in Developing Countries,
Journal of Development Economics. 39: 31–57.
Erenburg, Sharon J. and Mark E. Wohar (1995). Public and Private Investment: Are There Causal
Linkages?, Journal of Macroeconomics. 17: 1–30.
Everhart, Stephen S. and Mariusz A. Sumlinski (2001). Trends in Private Investment in Developing
Countries: Statistics for 1970–2000 and the Impact on Private Investment of Corruption
and the Quality of Public Investment, International Finance Corporation, Discussion Paper
No. 44. The World Bank, Washington, D.C. (www2.ifc.org/test/economics/pubs/dp44/dp44.
pdf)
Fisher, Stanley (1993). The Role of Macroeconomic Factors in Growth, Journal of Monetary
Economics. 32: 485–512.
Gramlich, Edward M. (1994). Infrastructure Investment: A Review Essay, Journal of Economic
Literature. 32: 1176–1196.
Granger, Clive W.J. (1969). Investigating Causal Relationships by Econometric Models and Cross-
Spectral Methods, Econometrica. 36: 424–438.
Greene, Joshua and Delano Villanueva (1991). Private Investment in Developing Countries: An
Empirical Analysis, IMF Staff Papers. 38(1): 33–58.
Guseh, James S. (1997). Government Size and Economic Growth in Developing Countries: A
Political-Economy Framework, Journal of Macroeconomics. 19: 175–192.
Gwartney, James and Robert Lawson (2004). Economic Freedom of the World: 2004 Annual Report.
Vancouver: The Fraser Institute. Data retrieved from www.freetheworld.com.
Heckelman, Jac C. and Michael D. Stroup (2000). Which Economic Freedoms Contribute to
Growth?, Kyklos. 53(4): 527–544.
Heckelman, Jac C. and Michael D. Stroup (2002). Which Economic Freedoms Contribute to
Growth? Reply, Kyklos. 55(3): 417–420.
Hsiao, Cheng (1982). Autoregressive Modelling and Causal Ordering of Economic Variables,
Journal of Economic Dynamics and Control. 4: 243–259.
Johansen, Søren (1995). Likelihood-Based Inference in Cointegrated Vector Autoregressive Models.
Oxford: Oxford University Press.
Kang, Heejoon (1989). The Optimal Lag Selection and Transfer Function Analysis in Granger-
Causality Tests, Journal of Economic Dynamics and Control. 13: 151–169.

328 r 2005 Blackwell Publishing Ltd.

 INTERACTIONS BETWEEN PUBLIC AND PRIVATE INVESTMENT

Khan, Mohsin S. and Manmohan S. Kumar (1997). Public and Private Investment and the Growth
Process in Developing Countries, Oxford Bulletin of Economics and Statistics. 59: 69–88.
Khan, Mohsin S. and Carmen Reinhart (1990). Private Investment and Economic Growth in
Developing Countries, World Development. 18: 19–27.
Lambsdorf, Johann Graf (2003). How Corruption Affects Productivity, Kyklos. 56: 457–474.
Levine, Ross and David Renelt (1992). A Sensitivity Analysis of Cross-Country Growth. Regres-
sions, American Economic Review. 82(4): 942–963.
Looney, Robert E. and Peter C. Frederiken (1997). Government Investment and Follow-on Private
Sector Investment in Pakistan 1972–1995, Journal of Economic Development. 22: 91–100.
Lucas, Robert E. Jr. (1988). On the Mechanics of Economic Development, Journal of Monetary
Economics. 22: 3–42.
Lynde, Catherine and James Richmond (1993). Public Capital and Total Factor Productivity,
International Economic Review. 34: 401–414.
Mankiw, Gregory N., David Romer and David N. Weil (1992). A Contribution to the Empirics of
Economic Growth, Quarterly Journal of Economics. 107(2): 407–437.
Onliner, Stephen, Glenn Rudebusch and Daniel Sichel (1995). New and Old Models of Investment:
A Comparison of Forecasting Performance, Journal of Money, Credit, and Banking. 27:
806–826.
Otto, Glenn and Graham M. Voss (1996). Public Capital and Private Production in Australia,
Southern Economic Journal. 62: 723–739.
Otto, Glenn and Graham M. Voss (1998). Is Public Capital Provision Efficient?, Journal of Monetary
Economics. 42: 47–66.
Pellegrini, Lorenzo and Reyer Gerlagh (2004). Corruption’s Effect on Growth and its Transmission
Channels, Kyklos. 57: 429–456.
Penm, Jack H.W. and R. Deane Terrell (1984). Multivariate Subset Autoregressive Modelling with
Zero Constraints for Detecting ‘Overall Causality’, Journal of Econometrics. 24: 311–330.
Phillips, Peter C.B. and Pierre Perron (1988). Testing for a Unit Root in Time Series Regression,
Biometrika. 75: 335–346.
Poot, Jacques (2000). A Synthesis of Empirical Research on the Impact of Government on Long-Run
Growth, Growth and Change. 31: 516–546.
Sakr, Khaled (1993). Determinants of Private Investment in Pakistan, IMF Working Paper 30.
Washington, D.C.
Schwarz, Gideon (1978). Estimating the Dimension of a Model, Annal of Statistics. 6: 461–464.
Shioji, Etsuro (2001). Public Capital and Economic Growth: A Convergence Approach, Journal of
Economic Growth. 6: 205–227.
Sturm, Jan-Egbert and Jakob De Haan (1998). Public Capital Spending in the Netherlands:
Developments and Explanations, Applied Economics Letters. 5(1): 5–10.
Sturm, Jan-Egbert, Erik Leertouwer and Jakob de Haan (2002). Which Economic Freedoms
Contribute to Growth? A Comment, Kyklos. 55: 403–416.
Sundararajan, V. and Subhash Thakur (1980). Public Investment, Crowding-out and Growth: A
Dynamic Model Applied to India and Korea, IMF Staff Papers. 27(4): 814–855.
Temple, Jonathan (2000). Cross-Country Growth Regressions and What the Textbooks Don’t Tell
You, Bulletin of Economic Research. 52: 181–205.
Voss, Graham M. (2002). Public and Private Investment in the United States and Canada, Economic
Modelling. 19: 641–664.

r 2005 Blackwell Publishing Ltd. 329

 ERDAL ATUKEREN

SUMMARY

This study examines the linkages between public investments and private investments by using Granger-
causality and cointegration tests and probit analysis in a sample of 25 developing countries. The results
from the cointegration and Granger-causality tests are further analysed in a probit framework by
assigning the dependent variable the value ‘1’ for the ‘crowding-out’ cases and ‘0’ otherwise, and the
explanatory variables are various components of Gwartney and Lawson’s (2004) economic freedom of the
world index. Using this approach, we find that the higher the share of government involvement in an
economy, the lower the trade openness; the more restrictions there are on the use of foreign currencies, and
the more stable and developed the macro and monetary environment is, the higher the likelihood that
public investments may crowd out private investments. The model correctly predicts 10 out of 11 cases of
crowding-out and 13 out of 14 cases of no-crowding-out.

330 r 2005 Blackwell Publishing Ltd.