Growth, Development and Natural Resources: New
Evidence Using a Heterogeneous Panel Analysis
Tiago V. de V. Cavalcanti, Kamiar Mohaddes
and Mehdi Raissi
November 2009
CWPE 0946
Growth, Development and Natural Resources:
New Evidence Using a Heterogeneous Panel Analysis
Tiago V. de V. Cavalcanti, Kamiar Mohaddes, and Mehdi Raissi
Faculty of Economics, University of Cambridge
November 4, 2009
Abstract
This paper explores whether natural resource abundance is a curse or a blessing.
In order to do so, we …rstly develop a theory consistent econometric model, in which
we show that there is a long run relationship between real income, the investment rate,
and the real value of oil p roduction. Secondly, we investigate the long-run (level) e¤ects
of natural resource abundance on domestic output as well as the short-run (growth)
e¤ects. Thirdly, we make use of a non-stationary panel approach which explicitly es-
timates the long-run relationships from annual data as opposed to the dynamic and
static panel approaches which might in fact estimate the high-frequency relationships.
Fourthly, we account for cross-country dependencies that arise potentially from oil
price shocks and other unobserved common factors, and allow countries to resp on d
di¤erently to these sho cks. Finally, we explicitly recognize that there is a substan-
tial heterogeneity in our sample, consisting of 53 oil exporting and importing countries
with annual data between 1980-2006, and adopt the methodology developed by Pesaran
(2006) for estimation. This approach considers di¤erent dynamics for each country and
is consistent under both cross-sectional dependence and cross-country heterogeneity.
We also check the robustness of these results by using the fully modi…ed OLS method
of Pedroni (2000). Our non-stationary approach also allows for country-speci…c unob-
served factors, such as social and human capital, to be captured in the …xed e¤ects
and the heterogeneous trends together with any omitted factors. Our estimation re-
sults, using the real value of oil production, rent or reserves as a proxy for resou rce
endowme nt, indicate that oil abundance is in fact a blessing and not a curse, exhibited
through both the long-run and the short-run e¤ects.
JEL Classi…cations: C23, O13, O40, Q32.
Keywords: Growth models, natural resource curse, cointegration, cross sectional
dependence, common correlated e¤ects, and oil.
We gratefully acknowledge …nancial support from the Corporación Andina de Fomento (CAF). We are
grateful to Hashem Pesaran as well as conference participants at the LACEA-LAMES 2009 for constructive
comments and suggestions. We would also like to thank Peter Pedroni and Takashi Yamagata for making
their codes available to us.
1
1 Introduction
The aim of this paper is to investigate the following questions: Is an abundance of natural
resources, in particular oil, a curse or a blessing? What are the e¤ects of natural resource
abundance on growth and economic development, as seen in the level of income per capita?
Following the in‡uential work by Sachs and Warner (1995),
1
a growing empirical literature
on and interest in the resource curse paradox was generated. According to this paradox,
resource rich countries perform poorly when compared to countries which are not endowed
with oil, natural gas, minerals and other non-renewable resources. Therefore, resource abun-
dance is believed to be an important determinant of economic failure, which implies that oil
abundance is a curse and not a blessing.
There are di¤erent explanations for why resource rich economies might be subject to this
curse. Dutch disease (see Corden and Neary (1982), Neary and van Wijnbergen (1986), and
Krugman (1987)) is one of the channels through which the resource curse makes itself felt: an
increase in natural resource revenue leads to an appreciation of the real exchange rate, which
negatively a¤ects the pro…tability of the service and manufacturing sectors. The resulting
re-allocation of resources from the high-tech and high-skill manufacturing and service sectors
to the low-tech and low-skill natural resource sector is then harmful for economic growth.
Another explanation for the resource curse paradox is based on rent-seeking theories, which
argue that natural resource abundance generates an incentive for agents to engage in non-
productive activities and for the state to provide fewer public goods than the optimum.
See for instance, Lane and Tornell (1996), Leite and Weidmann (1999), Tornell and Lane
(1999), and Collier and Hoe- er (2004). Finally, Mehlum et al. (2006) have attempted to
show that the impact of natural resources on growth and development depends primarily
on institutions, while Boschini et al. (2007) have argued that the type of natural resources
possessed is also an important factor. It is not our goal to discuss these theories in detail,
or to determine their validity. We refer to Sachs and Warner (1995), Rosser (2006), and
Caselli and Cunningham (2009) for an extensive examination of these prominent accounts
of the natural resource curse paradox, as well as van der Ploeg and Venables (2009) for a
more recent survey.
The empirical evidence on the resource curse paradox is rather mixed. Most papers in
the literature tend to follow Sachs and Warner’s cross-sectional speci…cation introducing
new explanatory variables for resource dependence/abundance, while others derive theoret-
ical models that are loosely related to their empirical speci…cation. Some of these papers
con…rm Sachs and Warner’s results (see Rodriguez and Sachs (1999), Gylfason et al. (1999),
and Bulte et al. (2005) among others). An important drawback of these studies with few
exceptions, however, is their measure of resource abundance. Sachs and Warner (1995), for
instance, use the ratio of primary-product exports to GDP in the initial p eriod as a measure
of resource abundance. This ratio, as clearly pointed out by Brunnschweiler and Bulte (2008),
measures resource dependence rather than abundance. The latter should be introduced in
the growth regressions as the stock or the ‡ow of natural resources. Moreover, a cross
sectional growth regression augmented with this regressor clearly su¤ers from endogeneity
and omitted variable problems. Brunnschweiler and Bulte (2008) argue that the so-called
1
See also Sachs and Warner (1997) and Sachs and Warner (2001).
2
resource curse does not exist, and that while resource dependence, when instrumented in
growth regressions, does not a¤ect growth, resource abundance in fact positively a¤ects eco-
nomic growth.
2
The positive e¤ect of resource abundance on development and growth is
also supp orted by Esfahani et al. (2009), who develop a long run growth model for a major
oil exporting economy and derive conditions under which oil revenues are likely to have a
lasting impact. However, this approach contrasts with the standard literature on "Dutch
disease" and the "resource curse", which primarily focuses on the short run implications of
a temporary resource discovery. On the other hand, Stijns (2005), using di¤erent measures
for resource abundance, indicates that the e¤ect of this variable on growth is ambiguous.
Another branch of the literature investigates the channels through which natural resource
abundance a¤ects economic growth negatively. Gylfason (2001), for instance, shows that
natural resource abundance appears to crowd out human capital investment with negative
e¤ects on the pace of economic activity, while Bravo-Ortega et al. (2005) show that higher
education levels can in fact o¤set the negative e¤ects of resource abundance. Therefore, it
can be seen that the empirical …ndings on the resource curse paradox are still not conclusive.
There are a number of grounds on which the econometric evidence of the e¤ects of
resource abundance on growth may be questioned. Firstly, the literature relies primarily on
a cross-sectional approach to test the resource curse hypothesis, and as such does not take
into account the time dimension of the data. As noted above, the cross-sectional approach
is also subject to endogeneity problems, and this is perhaps the most important reason for
being skeptical about the econometric studies suggesting a positive or negative association
between resource abundance and growth. Secondly, the vast majority of existing studies
focus on the e¤ects of resource abundance on the rate of economic growth, even though
most models in the Solow/Ramsey tradition suggest that the e¤ects on growth should be
transitory, though permanent for the level of per capita income.
3
In addition, even when panel data techniques are used most studies make use of homoge-
neous panel data approaches, such as the traditional …xed and random e¤ects estimators, the
instrumental variable (IV) technique proposed by Anderson and Hsiao (1981) and Anderson
and Hsiao (1982), and the generalized methods of moments (GMM) model of Arellano and
Bond (1991), Arellano and Bover (1995), among others.
4
While homogeneous panel data
models allow the intercepts to di¤er across groups all other parameters are constrained to be
the same. Therefore, a high degree of homogeneity is still imposed. As discussed in Pesaran
and Smith (1995), the problem with these dynamic panel data techniques, when applied to
testing growth e¤ects, is that they can produce inconsistent and potentially very mislead-
ing estimates of the average values of the parameters, since growth models typically exhibit
substantial cross-sectional heterogeneity. In fact Lee et al. (1997), using a panel of data on
102 countries, illustrate that there is pervasive heterogeneity in speeds of convergence and in
2
Arezki and van der Ploeg (2007) provide some support for these results as does Brunnschweiler (2009).
3
This is also consistent with the empirical evidence provided by Klenow and Rodriguez-Clare (1997). In
a di¤erent setting, Hen ry (2007) calls into question the usefulness of the cross-county approach to testing
the relationship betwee n capital account lib e ralization and growth. He argues that the capital deepening
channel of gain from …nancial integration should imply only a temporary, rather than permanent, increase
in growth, but most of the cross-sectional studies that have been conducted do not really test this.
4
For a comprehensive survey of the econometric methods employed in the growth literature, and some of
their shortcomings, see Durlauf et al. (2005) and Durlauf et al. (2009).
3
growth rates across countries and show that the conventional method of imposing homogene-
ity are subject to substantial biases. In addition, Lee et al. (1998) test the null hypothesis
of homogeneity in growth rates as well as the null of common speed of convergence and …nd
that this is rejected for 102 non-oil countries. This pattern is the same for 61 intermediate
group of countries, while for 22 OECD countries the null of common speed is not rejected.
More recently Pedroni (2007) shows that there are signi…cant di¤erences in the aggregate
production function technologies among countries. Taking into account these di¤erences he
argues that it is possible to explain the observed patterns of per capita income divergence
across countries. Finally, the current econometric evidence does not address the problem
of cross-sectional dependence arising from common factors or shocks. Thus estimations and
inference based on models that do not take into account cross-country heterogeneity and
dependence, such as the cross-sectional speci…cations widely used in the literature, can bring
about biased and misleading results.
In this paper we take a di¤erent approach in order to test the resource curse hypothesis.
We explicitly recognize that there is a substantial degree of heterogeneity in the growth
experience of di¤erent resource abundant countries. We therefore use a heterogeneous panel
data approach, which considers di¤erent dynamics for each country. We also account for
cross-country dependencies that arise potentially from multiple common factors, and we
allow the individual responses to these factors to di¤er across countries. A possible source
of cross-sectional dependency might be due to world-wide common shocks that a¤ect all
cross-sectional units. Changes in technology and in the price of oil provide examples of
such common shocks that may a¤ect real GDP per capita, but to di¤erent degrees across
countries.
5
To address the issues raised above we adopt the common correlated e¤ects
estimator of Pesaran (2006), a su¢ ciently general and ‡exible econometric approach, which
is consistent under both cross-section dependence and cross-country heterogeneity. Moreover,
we investigate the long-run (level) e¤ects of natural resource abundance on domestic output
as well as the short-run (growth) e¤ects through level and Error Correction Model (ECM)
regressions. We also check the robustness of our long-run results by using the fully modi…ed
OLS method of Pedroni (2000). An advantage of our non-stationary approach is that the
…xed e¤ects and the heterogeneous trends capture country-speci…c unobserved factors, such
as social and human capital, which are very di¢ cult to measure or observe accurately. In
addition, omitted variables that are either constant or evolve smoothly over time are absorbed
into the country speci…c deterministic trend.
Furthermore, we develop a standard growth model that requires the use of natural re-
sources as an input in the production of the consumption good. We view natural resources
as a proxy for energy and power. We assume that agents can extract natural resources at a
rate which is optimally determined, and rent them out to …rms for production. In contrast to
the literature on exhaustible natural resources and economic growth, for instance Dasgupta
and Heal (1974), we assume that costly investment will enable new reserves to be found
and old …elds to be developed. It is important to emphasize that while we do not believe
natural resources are limitless, oil production and more importantly viable reserves do seem
to increase over the horizon we are investigating empirically.
6
Finally, our theoretical model
5
Di¤erent forms of cross section dependence are discussed and formally de…ned in Pesaran and Tosetti
(2007).
6
If natural resources in our model represent power and energy used in produc tion, we can also view the
4