WPS4766
Policy ReseaRch WoRking PaPeR 4766
In Search of the Missing Resource Curse
Daniel Lederman
William F. Maloney
The World Bank
Development Research Group
Trade Team
&
Latin America and the Caribbean Region
Office of the Chief Economist
November 2008
Policy ReseaRch WoRking PaPeR 4766
Abstract
The debate over the curse of natural resources has inability of econometric analysis based on international
haunted developing countries for decades if not centuries. data to capture historical processes. This paper defends
A review of existing empirical evidence suggests that an empirical proxy for relative abundance of natural
the curse remains elusive. The fragile negative effect of resources, which is based on standard growth theory.
natural resources on economic growth might be due In turn, various econometric estimations are hopelessly
to international heterogeneity in the effects of natural deployed in the search for the missing resource curse.
resources on economic growth, to the use of weak Some evidence suggests that natural resources might
indicators of natural resources that might be unrelated have large positive effects whose true magnitude remains
to relative natural-resource endowments, or to the unknown due to unresolved econometric issues.
This paper--a product of theTradeTeam, Development Research Group, and the Office of the Chief Economist for Latin
America and the Caribbean Region of The World Bank--is part of a larger effort in both departments to understand the
potential effects of the structure and quality of trade on development. Policy Research Working Papers are also posted
on the Web at http://econ.worldbank.org. The authors may be contacted at dlederman@worldbank.org and wmaloney@
worldbank.org.
The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development
issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the
names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those
of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and
its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent.
Produced by the Research Support Team
In Search of the Missing Resource Curse
Daniel Lederman
Development Economics Research Group
The World Bank
William F. Maloney
Office of the Chief Economist for Latin America
The World Bank
This paper is partly financed by the Regional Studies Program of the Latin American and Caribbean Region
of the World Bank. We would like to thank Joana Naritomi for inspired research assistance. Cameron Shelton,
Thad Dunning, and Associate Editor Francisco Rodríguez provided invaluable feedback and criticisms during
and after the 17th Economía Panel Meeting, held May 2-3, 2008 at Yale University. We also acknowledge the
comments from other participants at this event, and the comments from an anonymous reviewer.
I. Introduction
Like Dracula, the notion of a natural resource curse remerges periodically, haunting
the development debate, striking fear into the hearts of policy makers, and causing
quantities of ink to be spilled on the various ways in which being blessed with mineral,
agricultural or other natural wealth, will lead to anemic growth performance. Adam Smith
(1776) was perhaps first to articulate a concern that mining was a bad use of labor and
capital and should be discouraged.1 The idea reappeared in the mid 1950s in Latin America
when Raúl Prebisch (1959), observing slowing regional growth, argued that natural
resource industries had fewer possibilities for technological progress and were condemned
to decreasing relative prices on their exports. These stylized facts helped justify the Import-
Substitution experiment in modifying national productive structures. Subsequently,
disenchantment with the inefficiencies of protectionism and the consequences of populist
macroeconomic policies led to more open trade regimes and less intrusive micro economic
policies, partly with the example of East Asia's rapid export-led growth in mind.
However, over the interim, two stylized facts have emerged to convert a new
generation of analysts to believers in the curse. First, the liberalizing economies, with some
notable exceptions, did not become manufacturing dynamos, or major participants in what
is loosely called the new "knowledge economy." Further, growth results were not
impressive and, in the case of Africa, dramatic falls in commodity prices contributed to
negative growth rates. With the increased popularity of cross country growth regressions in
the 1990s, numerous authors offered proof that, in fact, natural resources appeared to curse
countries with slower growth: Auty (1993), Davis (1995), Gylfason et al (1999), Neumayer
(2004), Mehlum Moene and Torvik (2006), and arguably most influentially, with several
authors drawing on their data and approach, Sachs and Warner (1995, 2001), have argued
empirically that since the 1960s the resource-rich developing countries across the world
have grown slower than other developing countries. Last year Macartan Humphey, Jeffrey
Sachs, and Nobel Prize winner Joseph Stiglitz published Escaping the Resource Curse,
which adds further credence to the myth. Consequently, we find ourselves again in a time
when the conventional wisdom postulates that natural resources are a drag on development,
which contradicts the common-sense view that natural riches are riches nonetheless.
Yet there has always been a countervailing current that suggests that common sense
was not, in this case, misleading. Most recently, evidence supportive of a more positive
view was brought together in Lederman and Maloney (2007) in Natural Resources, Neither
Curse nor Destiny, although, far earlier notable observers such as Douglass North and
Jacob Viner had dissented on the inherent inferiority of, for instance, agriculture relative to
manufacturing. Even as Adam Smith was writing the Wealth of Nations, the American
colonies were declaring their independence on their way to being one of the richest nations
in the history of humanity, based for a long period, largely on natural resources (see, for
example, Findlay and Lundahl 1994). Other success stories Australia, Canada, Finland,
1"Projects of mining, instead of replacing capital employed in them, together with ordinary profits of stock,
commonly absorb both capital and stock. They are the projects, therefore, to which of all others a prudent
law-giver, who desired to increase the capital of his nation, would least choose to give any extraordinary
encouragement ..."
1
and Sweden remain, to date, net exporters of natural resources.2 Latin America's and
Africa's disappointing experiences clearly offer a counterbalance to these success stories,
but they do not negate them.
The acknowledgment of the important heterogeneity of experiences has led
tentatively to a greater circumspection about the impact of resources, although not
necessarily less enchantment with the term "curse." Humphreys, Sachs and Stiglitz begin
their book noting that resource countries often perform worse than their resource poor
comparators, and Dunning (2005) talks of "conditional resource curse" that is, under
certain conditions, there is a negative growth impact. This is without a doubt a more careful
way to frame the issue that moves explaining the heterogeneity to center stage. Yet, the
notion of a resource "curse" suggests more than the existence of a negative tail in the
distribution of impact. Dracula's sinister reputation arises not from the occasional
involuntary transfusion when in Transylvania, but rather that a bloody parasitism is the
central tendency of character. (Disclaimer: the authors have not reviewed carefully any of
the relevant empirical literature on this topic.)3
This article builds on our earlier work to argue that such a negative central tendency
does not characterize natural resource abundance, and we would do well to exorcise the
curse from the economic discourse. The next section reviews the various channels through
which the curse is thought to operate, and we argue that, in many cases, the channel is
either not convincingly present, or in fact, applies to many other factors of production. We
then review some of the existing literature, arguing that the existing stylized fact of a curse
is tentative at best and certainly not robust enough to impugn an entire category of
production. We then examine with more care the appropriate proxy for measuring resource
curse effects and, in the process, suggest what may be driving some other findings of a
negative impact. Finally, using our preferred proxy, we deploy various estimations methods
in search of the missing curse, including an aspect of the resource curse we left relatively
unexplored in Neither Curse nor Destiny, the voracious political economy channel.
Though our results in this dimension are, to some degree, rudimentary, they are suggestive
2 Across time, notable thinkers have also been less credulous about the existence of a curse. In Prebisch's
era, future Nobel Prize winner Douglass North (1955 p.252) argued that "the contention that regions must
industrialize in order to continue to grow ... [is] based on some fundamental misconceptions;" and the pioneer
trade economist Jacob Viner argued that "There are no inherent advantages of manufacturing over
agriculture" (Viner, 1952 p. 72). The literature is clear that these development successes based their growth
on natural resources and several still do (see figure 1). See Irwin (2000) for the US; Innis (1933) and Watkins
(1963) for Canada; Wright (2001), Czelusta (2001) for Australia; Blomström and Kokko (2001) and
Blomström and Meller (1987) for Scandinavia. Latin America also offers its success stories: Monterrey in
Mexico, Medellin in Colombia, and São Paolo in Brazil all grew to become dynamic industrial centers based
on mining and in the latter two cases, coffee. Copper-rich Chile has been the region's model economy since
the late 1980s.
3 We don't for example, talk about a "venture capital" curse because 19 out of 20 VC financed firms go
bankrupt. If the central tendency is that NR have a positive effect, then they remain a blessing, albeit
conditional, and we need to understand the complementary factors necessary to maximize it. This is not
different than understanding why Taiwan did better with its electronics industry than Mexico; or that Italy did
better with its fashion industry than Korea did with "Project Milan."
2
that this element of the curse, too, merits closer scrutiny and might be a figment of our
statistical imagination.
II. The Mechanics of the Curse
Numerous channels through which the curse might operate have been offered. Here
we discuss only a few.
First, Prebisch (1959), among others, popularized the idea that terms of trade of
natural resource exporters would experience a secular decline over time relative to those of
exporters of manufactures. However, Cuddington, Ludema and Jayasuriya (2007) find that
they cannot reject that relative commodity prices follow a random walk across the 20th
century with a single break in 1929. That is, there is no intrinsic force driving the observed
decline and prices could as easily rise tomorrow as fall further. While commodity by
commodity important mean reverting components are evident and are, in fact, necessary for
stabilization funds to be viable, the notion that long run prices have a strong unpredictable
and permanent component appears more relevant today than at any time across the last half
century. Paul Krugman (2008), taking exactly the opposite position from Prebisch, argues
that continued growth by China and India, combined with simply "running out of planet"
will lead to continued strong excess demand such that "rich countries will face steady
pressure on their economies from rising resource prices, making it harder to raise their
standard of living."4
Second, beginning with Smith, observers have argued that natural resources are
associated with lower human and physical capital accumulation, productivity growth, and
spillovers. This case is far from proven, however. Consistent with Viner's (1952) early
assertion, Martin and Mitra (2001) find total factor productivity growth to be higher in
agriculture than in manufactures in a large sample of advanced and developing countries.
Wright and Czelusta (2006) and Irwin (2000) have argued that, contrary to Smith's
prejudice, mining is a dynamic and knowledge intensive industry in many countries and
was critical to US development. Blomstrom and Kokko (2006) have argued the same for
forestry in Scandinavia.
Several authors stress the complementarity of essential factors, particularly human
capital (see Gylfason and Bravo Ortega and de Gregorio 2007). Relatedly, Maloney (2002,
2007) argues that Latin America missed opportunities for rapid resource-based growth due
to deficient technological adoption driven by two factors. First, deficient national
"learning" or "innovative" capacity, arising from low investment in human capital and
scientific infrastructure, led to weak capacity to innovate or even take advantage of
technological advances abroad. Second, the period of inward-looking industrialization
discouraged innovation and created a sector whose growth depended on artificial monopoly
rents rather than the quasi-rents arising from technological adoption, and at the same time
undermined natural-resource-intensive sectors that had the potential for dynamic growth.
Røed Larsen (2004) argues that Norway's surge from Scandanavian laggard in the 1960s
to regional leader in per capita income was based largely on the opposite strategy and
4Krugman, Paul, "Running Out of Planet to Exploit," The New York Times, March 21, 2008.
3
concludes that "Norwegian oil is a high technology sector which we may assume has much
the same positive spillover effects as manufacturing is supposed to have" (p 17).
These arguments are central to the discussion surrounding the "Dutch Disease"
aspect of the curse emphasized by, among others, Gylfason et al (1999), Sachs and Warner
(2001) where perhaps through an appreciated exchange rate or classic Rybczinski effects,
resource booms depress manufacturing activity. If the natural resource sector is not inferior
in terms of its growth potential, then this sectoral shift would be of similar import to the
canonical displacement of agriculture by manufacturing, or the finance-driven exchange
rate effects arising from the City of London on manufacturing in the United Kingdom.
Third, either for reasons of history or Dutch disease, natural resource abundance
may result in high levels of export concentration, which may lead to higher export price
volatility and hence greater macro volatility.5 However this is a more general concern.
Dependence on any one export, be it copper in Chile or potentially micro-chips in Costa
Rica, can leave a country vulnerable to sharp and sudden declines in terms of trade with
attendant channels of influence through volatility.
Fourth, another important literature suggests that natural riches produce institutional
weaknesses (see, among others, Auty 2001, 2005, Ross 1999, Gelb 1988 Easterly, and
Levine 2002). Tornell and Lane (1999) described the phenomenon where various social
groups attempt to capture the economic rents derived from the exploitation of natural
resources as the "voracity effect." Subsequent refinements have focused on how "point
source" natural resources -those extracted from a narrow geographic or economic base,
such as oil or minerals- and plantation crops have more detrimental effects than those that
are diffuse such as livestock or agricultural produce from small family farms (Murshed
2004, Isham et al 2006). But again, this concern is not specific to natural resources, but to
any source of rents. Autey, for instance, points to a similar impact of foreign aid.
"Natural" monopolies, such as telecom have given rise to precisely the same effects in
Mexico, and the rent-seeking literature generated by Krueger (1974)often focused on the
adverse political economy effects arising from trade restrictions. Rajan and Zingales (2004)
in Saving Capitalism from the Capitalists examine rentier attitudes among the corporate,
including manufacturing and financial elite, and the need for financial markets to ensure the
pressure of new entry.
That said, there is clearly an important agenda to understand the interaction between
political institutions and the emergence of resource sectors. Mehlum, Moene and Torvik
(2006) have argued the importance of strong institutions to minimize rent seeking activity,
and Rodríguez and Gomolin (2008) stress the pre-existing centralized state and
professionalized military as essential to Venezuela's stellar growth performance during
1920-1970 after the oil exploitation began in 1920. Dunning (2005) offers a model of how
differences in the world structure of resources, the degree of societal opposition to elites,
and the prior development of the non-resource private sector help predict the incentives to
diversification and political stability.
5Sachs and Warner (1995b) argue that Dutch disease leads to concentration in resource exports, which they
assume to have fewer possibilities for productivity growth.
4
III. The Elusive Curse
Without question, many of the channels discussed above may have important
implications for growth. However, the question is whether taking all these impacts together,
resource abundance has, as a central tendency, curse-like qualities. The literature has used
a variety of proxies for resource abundance but has not been able to demonstrate this.
By far, the best known empirical tests for the resource curse are found in the work
of Sachs and Warner (1995a, 1997a, 1997b, 1999, 2001a,b) who employ natural resource
exports as a share of GDP as their proxy. Using cross sectional data employed previously
by Barro (1991); Mankiw, Romer and Weil (1992); and DeLong and Summers (1991)
across the period 1970-1990, they persistently find a negative correlation with growth,
much to the alarm of many resource abundant developing countries.6 Yet, this proxy leads
to some counterintuitive results as a measure of resource abundance. Singapore, for
example, due to its substantial re-exports of raw materials, appears very resource abundant
and given its high growth rates, even seems to impart a positive relationship between
resource abundance and growth. Because this gross measure is clearly not capturing the
country's true factor endowments, Sachs and Warner replaced the values of Singapore and
Trinidad and Tobago with net resource exports as a share of GDP (see data appendix in
Sachs and Warner 1997a). It is not clear why net values should only be used for these two
cases. Numerous countries in Asia and Latin America have a large presence of export
processing zones that would, using the gross measure, overstate their true abundance in
manufacturing related factors.7 The issue turns out to be central to the findings of a curse.
When Lederman and Maloney (2007) replicate the Sachs-Warner results using either a net
measure of resource exports, or using the gross export measure without the adjustments for
the two countries, they find that the negative impact of natural-resource abundance on
growth disappears.
In fact, the interpretation of the Sachs Warner results is not clear even using their
modified data. Sala-i-Martin et al (2004) in their search for robust regressors across
millions of growth regressions find a persistent negative sign when the proxy enters, but it
is not robust enough to be considered a core explanatory variable for growth as other
variables appear to absorb its influence. In a similar vein, Lederman and Maloney (2007)
show that, controlling for fixed effects in a panel context, the negative impact of resources
also goes away suggesting that it is not their particular proxy, but its correlation with
unobserved country characteristics that are driving the result. Manzano and Rigobón
(2007) concur, and argue that the cross sectional result arises from the accumulation of
foreign debt during periods when commodity prices were high, especially during the 1970s,
that lead to a stifling debt overhang when prices fell. These results, and the analogy to
6The other papers by Sachs and Warner (1995b, 1997b, 1999, 2001a, 2001b) contain the basic results of
1997a, at times using a slightly longer time span (1965-1990 instead of 1970-1989), and often including
additional time-invariant explanatory variables such as dummies identifying tropical and landlocked
countries, plus some additional social variables.
7The variable also shows substantial volatility over time reflecting terms of trade movements and hence the
average for the period is probably a better measure than the initial period value that was used by Sachs and
Warner in several of their papers.
5
other bubbles, are important, not only because they further dispel the alleged curse of
natural resources, but especially because the policy implication is that the right levers to
deal with the lackluster performance of resource-rich developing countries in recent
decades lies in the realm of macroeconomic policy instead of trade or industrial policies.
Bravo-Ortega and de Gregorio (2007) using the same proxy (as well as natural
resource exports over total merchandise exports), also find a negative cross sectional
impact but trace its origin to a Dutch disease effect working through human capital.
Adding an interactive human capital term suggests that as the stock of human capital rises,
the marginal effect of the stock of natural resources on income growth rises and becomes
positive. This is broadly consistent with Gylfason et al's argument that a national effort in
education is especially necessary in resource-rich countries, although without their
hypothesis that resource-rich sectors intrinsically require, and hence induce, less education.
However, Bravo-Ortega and Gregorio find that the point at which resources begin to
contribute positively to growth occurs at around 3 years of education, a level achieved by
all but the poorest countries.
Sachs and Vial (2002), Sachs and Warner (1995) confirm a negative and robust
relationship using a second, related proxy- the share of natural resources exports in total
exports, and this proved somewhat more robust. However, it, again, does not make Sala-i-
Martin et al's core of robust regressors. Further, when Lederman and Maloney include a
generic measure of concentration, the Herfindahl index, using export data disaggregated at
4-digit SITC, the resource curse disappears. The curse is one of concentration, not
resources. This finding is consistent with Auty's (2000) concern about a resource drag on
growth arising from the limited possibilities of diversification within commodities.
However, Lederman and Xu (2007) argue that diversification into non-resource sectors
from a strong resource base is feasible.
Leamer (1984) argues that standard Heckscher-Ohlin trade theory dictates that the
appropriate measure is net exports of resources per worker. This measure has been the
basis for extensive research on the determinants of trade patterns (e.g., Trefler 1995,
Antweiler and Trefler 2002, Estevadeordal and Taylor 2002).8 This was Lederman and
Maloney's (2007) preferred measure because it obviated the Singapore issue by netting out
resource exports from the beginning.9 In cross section and in panel context across the
Sachs-Warner period, it yielded either insignificant or positive results. Using Maddison's
(1994) growth data from 1820-1989, Maloney (2002, 2007) finds suggestive evidence of a
positive growth impact of resources from 1820-1950, but then a negative impact thereafter,
driven by Latin America's underperformance.
8Assuming identical preferences, a country will show positive net exports of resource intensive goods if its
share of productivity-adjusted world endowments exceeds its share of world consumption. Usually, the net
exports are then measured with respect to the quantity of other factors of production, such as the labor force.
9It is worth mentioning that the cited references show that the HO model of factor endowments performs
relatively well for natural resources net exports, but it performs less well for manufactures. The current debate
in the trade literature revolves around the question of how the HO model might be amended (by considering,
for example, technological differences across countries, or economies of scale) to help predict better the
observed patterns of net exports across countries. But there is not debate about the use of net exports as a
proxy for revealed comparative advantage in this literature.
6
Finally, a set of papers explore the impact of more direct measures of mining
production or reserves. Stijns (2005) finds no correlation of fuel and mineral reserves on
growth during 1970-1989. This confirms earlier work by Davis (1995) that mineral
dependent economies, defined by high share of minerals in exports and GDP, did well
relative to other countries across the 1970s and 1980s. Across their several million
regressions, Sala-i-Martin et al (2004) find the mining share in GDP to be consistently
positive and in the core of explanatory variables. Nunn (2008) finds a positive partial
correlation between the per capita production of gold, oil, and diamonds and GDP per
capita, in an analysis of long-term fundamental determinants of development, with a special
focus on the role of the slave trade and its concomitant economic consequences for African
economies. Most recently, Brunnschweiler (2008) finds that per capita mineral and fuel
production in 1970 have direct positive effects on economic growth during 1970-2000.
Clearly the resource curse remains elusive. The cross-country econometric evidence
remains weak, with results changing depending on the empirical proxies used to represent
relative endowments. Moreover, there have been surprisingly few efforts to understand the
theoretical content of trade-data proxies. The following section puts the literature in
theoretical context, which in turn helps motivate our empirical strategy.
IV. Clarifying the Curse
Some simple algebra helps clarify some dimensions of the curse and possible
approaches and pitfalls to measuring it. Start with a two-factor and two-goods economy,
where labor is initially immobile across sectors, and endowments of natural resources can
only be used to produce natural-resource-intensive commodities. Denoting the natural-
resource sector by subscript "nr" and rest of the economy by subscript "1," domestic
equilibrium in the labor market entails all labor, L, is fully employed:
(3) L=L1 + Lnr
National income is simply the sum of the income produced by each sector:
(4) Y=Y1 + Ynr.
Let K denote the endowment (stock) of natural resources, which are used only in the
production of related commodities.10 Each sector has a production function determined by
sector-specific technologies with constant returns to scale:
(5) Y1=a1LL1, and
-b
(6) Ynr = anrK bL1nr , and therefore, national income is:
10The modeling approach with a stock of productive resources is standard in the growth literature, and it also
appears in recent studies of natural resource production functions (Peretto 2008).
7
(7) Y = a1 (L - Lnr ) + anrK bL1nr
-b
L .
The a's are technologically determined productivity parameters. In the case of sector 1, a1L
is output per worker, which is always positive. In the case of the natural-resource sector, anr
is the output per complementary units of K and L. Parameter "b" is the natural-resource
share in output of that sector, and it is bounded by zero and one.
By fully differentiating (7), it is easy to show that the marginal effect of natural
resource endowments on national income has three components under the assumptions that
the marginal effect on the total labor force is zero (i.e., LK = 0) and the marginal effect
on K's share in natural-resource output is also unaffected:
(8)
-1
Y K
K = a1L K (L - Lnr)+ anr -b Lnr K b K b
K K bL1nr + ((1-b)anr - a1 ) Lnr +banr Lnr
L
The first two elements in (8) are the effects of marginal changes or differences in K
on factor productivities. The literature on the voracity effect, for example, can be
interpreted as negative first derivatives of productivity with respect to K. As pointed out by
Rodríguez (2007), the empirical endogenous growth literature has revolved around the
issue of the multiplicity of variables that could affect an economy's efficiency of factor use,
and institutions have been part of that debate. The third element is the marginal effect of K
on Y through the reallocation of labor, the Dutch disease effect. If labor in the natural-
resource sector increases, then income from the rest of the economy falls as it loses labor,
but output rises in the natural-resource sector. The net effect of the reallocation effect will
depend on the difference in the effective labor productivities across the two sectors. This
issue is usually framed as the alternative sector, perhaps manufactures, exhibiting some
externality or that it implies that the private optimization that leads to the reallocation of
labor implies a social loss somehow. The fourth term is the marginal effect on the output of
the natural-resource sector, which is equal to the marginal productivity of K's share in
natural-resource output.
In a nutshell, various strands of the curse literature argue that, with greater or lesser
weight put on particular arguments:
a1 K b
K
LK < 0 K < 0
, anr , and/or Lnr ((1-b)anr - a1L) Lnr < 0 , such that
-1
a1L K (L - Lnr) + anr K b K b
K
K K bL1nr +
-b Lnr ((1-b)anr - a1L ) Lnr < -banr Lnr in
the long-run.
8
Unfortunately for empirical work, K is unobserved. Even data on known mining
reserves is inadequate as endogenous investments in exploration lead to new discoveries.
As discussed, some studies use output of oil or mining as a share of GDP (e.g., Nunn 2008,
Sala-i-Martin et al. 2004), which appear positively correlated with GDP per capita or
subsequent economic growth. Other studies use gross export receipts as a share of total
merchandise exports or as a share of GDP (e.g., Sachs and Warner 1997, 2001; Bravo-
Ortega and De Gregorio 2007, Gylfason 2004). These proxies of natural resource
dependence are found to be positively correlated with GDP per capita (Bravo-Ortega and
De Gregorio 2007) but are often found to be negatively correlated with subsequent
growth.11
V. Empirical Strategy
The discussion above suggests that empirically the resource curse, like Dracula, is
hard to nail down. More specifically, getting the right proxy for resource abundance, and
understanding its properties, is critical to establishing the credibility of any empirical
finding. The mining reserve measures are closest to measuring true abundance, but capture
only a narrow range of products and hence do not obviate the need for a trade based proxy.
In taking up the search for the curse again, we follow the vast literature on growth
empirics but with a couple of innovations that help us identify a floor of the effect natural
resource endowments. A key innovation is the inclusion of the trade-data indicator of
relative endowments, which has the desirable property of an expected positive correlation
between natural-resource endowments per worker, which is not the case for the preferred
proxies used by resource-curse believers. We also choose an institutional variable that is
commonplace in the institutions and growth literature concerning the powers of the
executive branch of government, which has been absent in the voracity effects empirical
literature.
V. A. Empirical strategy: static and dynamic growth models
If relative endowments were observed, an empirical income function consistent with
the economic growth literature can be written as:
(9) yi = a + h(ln(K/L)i) + f'Xi + ei ,
where the subscripts represent countries, a is the intercept, y is the natural logarithm of
GDP per capita adjusted for purchasing power parity, X is matrix of other determinants of
cross-country income differences, and e is assumed to be white noise error. The
corresponding literature includes Frankel and Romer (1999), Acemoglu, Johnson and
11Isham et al. (2005) used dummy variables to identify countries according to export structures by looking
only at the top two merchandise exports (according to the three-digit SITC trade classification), which further
confounds the notion of natural resource dependence. It is also noteworthy, as mentioned earlier in this paper,
that papers that have relied on the Sachs and Warner data are actually using a observations that use the gross
exports of natural resources as a share of merchandise exports where a couple of observations, Singapore and
Trinidad & Tobago, were actually net exports of natural resources.
9
Robinson (2001), Glaeser et al. (2004), Acemoglu and Robinson (2005), and Nunn (2008).
We call this the "static" growth model.
The general form of the dynamic empirical growth model found in the literature
since Barro (1991) is:
(10) yi,t - yi,t-1 = a + g(yi,t-1 ) + h(ln(K/L)i,t) + f'Xi,t + ei,t .
The obvious difference between equations (9) and (10) is the inclusion of the lagged
dependent variable as a regressor. Parameter g is the so-called convergence coefficient,
which is interpreted as the conditional rate of convergence between poor and rich countries
when g<0. This model is the dynamic version of (9), and g-1 is equal to the autoregressive
coefficient of y, because the lagged value of y also appears with a negative sign in the left-
hand side. We present estimates of the static and dynamic growth models, but with K/L
replaced with a proxy discussed below.
V.B. A proxy of relative endowments with desirable properties: Net exports per worker
A good proxy has to be positively correlated with the relevant endowments, so that we can
interpret a growth-regression coefficient as truly capturing the effect of these endowments.
Our preferred indicator is net exports per worker. Net exports are simply the difference
between what is produced and what is consumed in the domestic market:
NX
(11) L = (Ynr -Cnr )/ L = (1-cnr )anr(K / Lnr)b -cnra1 (1-lnr),
L
where 0 lnr=(Lnr/L) 1 and 0 cnr 1. The selling point of this indicator is that it strictly
rises with K/L, because the consumption of natural resources is a fraction of total income,
0< cnr <1, and, following Rybcynski, labor is attracted to the resource sector as the
endowment expands:
(12) (NX L)= b(1 lnr + cnr1L lnr 0.
(K L) -cnr )anr(Lnr )-b Kb-1L (K) (K)
L L
It is, however, not without flaws. When used as a proxy for K/L in the estimation of
income or growth models, it results in two distinct problems, which are both related to the
consumption of natural resources, which rises as income grows and as the endowments
grow.
The fact that income growth also increases consumption introduces a biased
estimate of the partial correlation between NX/L and y.. This issue strikes us as non-trivial.
Figures 1 and 2 show the relationship between the log of the absolute value of average net
exports per worker during 1980-2005 and the log of GDP per capita for exporters and
importers of natural resources. There is a strong positive correlation for both sub-samples,
10
and the bi-variate elasticity is close to one in both cases. While the positive correlation of
the net exporters with income suggests that there is no resource curse in levels, the strong
negative correlation of net importers of natural resources strikes us as most plausibly
evidence of increased consumption of NR with development, with potentially symmetrical
effects on imports and exports. This issue is even more extreme where a rise in income
from non-resource related sectors leads to an increase in consumption of resources, giving
rise to a negative correlation of growth and NX/L. If this is the case, and the technical
annex suggests it is algebraically likely, then the negative bias of the coefficient of net
exports is substantial.
The second problem created by the use of our preferred proxy, which is related to a
consumption effect of increases in K/L, affects the inferences that can be drawn even from
unbiased estimates of the (causal) partial correlation between Y and NX/L. It is clear from
equation (12) that a rise in endowments also gives rise to increased imports and,
symmetrically, decreased exports of NR that mutes the degree to which movements in
NX/L reflect movements in K/L.
Operationally, to moderate the bias in the estimate of the coefficient of NX/L in
income and growth functions, we include imports of natural resources per worker as an
additional regressor in the empirical models. Under the assumption that there is also a
symmetrical effect on exports, then the sum of the coefficients on NX/L and M/L is a good
approximation to the effect of NX/L on y. After this adjustment, we still need to be careful
about drawing inferences about the effect of K/L on y based on this unbiased estimate,
because it is still potentially a magnified effect of K/L on y see the technical annex.
Still, we think this proxy is preferable to other popular indicators of natural-resource
dependence used in the curse literature simply because these are not strictly positively
correlated with K/L. The ratio of gross natural resource exports over GDP, in addition to
the entrepot (Singapore) problem discussed earlier, potentially can be negatively correlated
with K since a rise in K leads to a rise in exports which may be less than any induced rise in
Y.12 Gross exports (which is an unknown fraction of national production) of natural
resources as a share of total merchandise exports can be negatively correlated with K/L
when the share of consumption of natural resources in national income, such as agricultural
products and food, is larger than the consumption share of other goods. Though it may be
interesting as a measure of NR exports per se, and perhaps more generally as the
concentration of exports, it is not an especially good if the goal is to show that resource
endowments themselves are pernicious.
V. D. Executive constraints as key institutional variable and other controls
As Levine and Renelt (1992) first pointed out in the growth context, cross-country
growth regressions are sensitive to the control variables included in the specification. A
substantial portion of the empirical growth literature focuses on the neoclassical growth
model, where the basic conditioning variables are related to capital accumulation and
12It is straightforward calculus to derive these conditions. They are available from the authors upon request.
11
growth of the labor force per unit of capital. In this regard, we follow the classic paper by
Mankiw, Romer, and Weil (1992) in our most fully specified models.
Regarding institutions, the literature on the resource curse recently put emphasis on
the role of so-called "point-source" resources. Isham et al. (2005, 143) define point source
natural resources as "... those extracted from a narrow geographic or economic base, such
as oil, minerals (such as copper and diamonds), and plantation crops (such as sugar and
bananas)...." They argue that "Where extractive institutions were initially laid down, they
soon consolidated themselves in ways that reduced the likelihood that over time they would
have an interest in generating or in being subjected to countervailing pressures to generate
either more diverse revenue (export) streams or more open political structures" (Isham et
al. 2005, 145-6). These statements consequently imply that natural resources should be
positively correlated with political characteristics that entail weak checks and balances,
which would otherwise limit the capacity of the rentier elites to control these resources and
public policies that might hamper economic diversification. In turn, the literature suggests
that such institutional characteristics are negatively correlated with long-term GDP per
capita.
It is possible, however, that under certain institutional arrangements, the resource
curse might come alive. That is, natural resources might not produce poor institutions as in
Isham et al. (2005), but when they interact with certain types of political institutions,
natural resources might hamper development. For example, as the voracity effect is a
problem of governing the commons, political institutions that yield fragmented and
multiple coalitions might be associated with a natural resource curse. This is the argument
by Sala-i-Martin and Subramanian (2003). But it should be noted that this is a different
argument about heterogeneity of the effects of natural resources, similar to the concerns
raised by Dunning (2005). Hence it is entirely consistent with the idea that natural
resources might not necessarily produce institutional outcomes that in turn interact with
natural resources to produce a curse. Our empirical exercises focus on testing this latter
hypothesis, and we leave it for future research to test for interactions between institutions
and natural resources as the source of the curse.
V.C.1. Estimation
In light of the discussion above, we include four interactive terms in the estimation
of 9 above: dummy variables for net exporters and net importers are multiplied with the
natural logarithm of the absolute value of net exports of natural resources per worker. In
addition, the (log of) imports of natural resources per worker are also interacted with the
relevant dummies.
To assess the possibility of heterogenous effects of resources due to interactions
with other non-observables, such as institutional quality, As mentioned, coefficient
heterogeneity is a concern in growth regressions we estimate Quantile Regressions of our
basic model with only two conditioning variables (the Sachs-Warner reform index and the
growth of terms of trade) followed by a more fully specified model that includes
macroeconomic volatility, and the Mankiw, Romer, and Weil (1992, MRW henceforth)
neo-classical growth variables. While recent literature has taken other approaches to
12
dealing with heterogeneity in growth regressions (see, also, Durlauf, Kourtellos, and
Minkin 2001), Quantile Regressions provide estimates of the coefficients across the
conditional distribution of the sample, for example, by allowing the bottom 25% of
observations to be below the predicted values, while at the same time estimating the
coefficients conditional on having 50% percent of the sample above and below the
predictions, as well as the top 75% of the sample. If the coefficients are sufficiently
different, then we can conclude that resources have distinct effects in different country
contexts. The regression errors from each quantile are allowed to be correlated and they are
bootstrapped (with 100 interactions of random sub-samples), because they are unknown a
priori.
The discussion of the Quantile Regressions is followed by the presentation of 3SLS
estimations of the basic and the fuller specifications of (9), but adding the Acemoglu-
Robinson institutional variable. We explore the role of institutions, namely constraints on
the executive branch, which could be affected by the relative abundance of natural
resources according to the resource curse hypothesis. We follow the relevant literature on
institutions and growth by complementing the static GDP per capita function with the
population densities of countries circa 1500 as the identifying instrumental variable of the
3SLS estimator, which is the approach of Acemoglu and Robinson (2002, 2005) and
Glaeser et al. (2004).13 3SLS allows for the simultaneous estimation of the growth and
institutional equation, with different explanatory variables in each structural equation. For
example, imports of natural resources per worker appear in the growth equation, but not as
a determinant of executive constraints.
VI. Data and Bi-variate Correlations
Standard publicly available sources offer all the necessary data to implement our
empirical strategy. The relevant variables are averaged over the 1980-2005 period. Table
A1 in the Appendix lists the variable definitions and sources.
VI. A. Variable definitions and data sources
The dependent variables are the real GDP per capita, its average annual growth rate
between 1980 and 2005, and the average annual growth rate in each five year period. The
data come from the latest version of the Penn World Tables, which has data until 2004. The
year 2005 is derived by applying the real growth rate of GDP per capita in local currency
and constant domestic prices between 2004 and 2005.
13There are numerous potential pitfalls in the estimation of the GDP per capita in levels model with 2SLS
even when we can conclude that the IV is correlated with the endogenous explanatory variable but not with
the dependent variable. One is that the chosen instrument could be weak in the sense of explaining a very low
share of the variance of the endogenous variable (executive constraints in this application); see Dollar and
Kraay (2003) for a similar application. Another related pitfall is that the variance of the endogenous variable
that explains the variance of the dependent variable should be related to the variance of the instrumental
variable see Dunning (2008). As discussed in the results section, these issues became unimportant since we
did not find much evidence that executive constraint affects the underlying relationship between our proxy for
relative natural-resource endowments and development.
13
The main explanatory variable of interest, as mentioned, is the natural logarithm of
the absolute value of net exports of natural resources per worker. The trade data come from
UNCOMTRADE, as downloaded from the World Bank's data server. The commodity
groups of products classified as being intensive in the use of natural resources are Leamer's
(1984, 1995) non-manufacturing commodities see Table A1.
Regarding the conditioning variables, we follow the strategy in Lederman and
Maloney (2007) by examining how changes in the specification of the model affect the
coefficient on the natural-resource variables. The initial multi-variate regression model
specification includes two variables in addition to the natural-resource proxies: (a) the share
of years during 1980-1999 when the dummy variable for policy reform was observed, as
per the data in Wacziarg and Welch (2002) and Sachs and Warner (1995);14 (b) and the
average annual growth of the terms of trade during 1980-2005, which comes from the
World Bank's World Development Indicators data, as downloaded in early 2008.
The augmented model includes the executive constraints variable, which is the
institutional variable used in Acemoglu and Robinson (2002, 2005) and Glaeser et al.
(2004). The data come from the Polity IV database, and has a value that ranges between 1
and 7, with higher values representing less executive-branch discretion. The econometric
models include the natural logarithm of this variable, which makes it easier to interpret
coefficient estimates as elasticities. The literature argues that constraints on the Executive
should be negatively partially correlated, in the first stage of 2SLS estimations, with
historical population density.15 Due to the limited coverage of the historical population
density variable, we present two sets of 3SLS estimations of the static and dynamic models;
one with the limited sample and another with a sample that incorporates observations with
imputed values for population density circa 1500. This allows for a discussion about how
the limited sample affects the estimated coefficients.16
14The policy reform index is one if import tariffs are below 40%, if there are not significant non-tariff
barriers, if governments have privatized a significant share of public enterprises, and if the foreign-exchange
black market premium is below a certain threshold see the Data Appendix for details.
15High population densities circa 1500 are presumably associated with the localization of extractive activities
and slave trade rather than with permanent settlement. Acemoglu and Robinson (2005, 960) state: "The
second determinant of European colonization strategy was initial indigenous population density. Where this
was high, Europeans were more likely to "capture" the local population and put it to work in some form of
forced labor system. Where initial population density was low, Europeans were more likely to settle
themselves and less likely to develop extractive institutions even when they did not settle. Acemoglu et al
(2002) provide evidence that for countries colonized by European powers there is a strong negative
relationship between population density in 1500 and income per capita today. This relationship is driven by
the fact that former colonies with greater population density in 1500 had, and still have, worse property rights
institutions. The density of indigenous population per square kilometer in 1500 is therefore an appealing
alternative instrument. Because settler mortality and population density in 1500 correspond to different
sources of variation in practice (the correlation between the two measures is 0.4), but should have similar
effects on property rights, using these two instruments separately is a good check on our results." If the curse-
via-politics hypothesis is correct, then we also expect a negative partial correlation in the first-stage equation
between our proxy for relative natural-resource abundance and the executive constraints index. It is
noteworthy that Glaeser and coauthors (2004) argue that education trumps institutions. This issue will be
important in the interpretation of our results, since we also control for capital accumulation and human
capital.
16The imputed values are estimated using all the other explanatory variables in the equation of the
determinants of executive constraints, namely (log of the absolute value of) net exports of natural resources,
14
Macroeconomic volatility can also affect economic performance, especially private
investment. To the extent that natural resources have higher price volatilities than other
goods, it is possible that this might be a channel through which they affect growth
performance. Following Servén (2003), our measure of macroeconomic uncertainty is the
standard deviation of the monthly variation of the log of the Real Effective Exchange Rate
from the IMF's International Financial Statistics.
The empirical neoclassical growth model of MRW includes, in addition to the level
of education of the population and the investment rate, the growth of "effective units of
labor," which is the observed growth of labor for each country minus a global capital-
depreciation rate plus an estimate of the global rate of technological progress. MRW
assume that the rate of technological progress minus the rate of capital depreciation is equal
to 0.05 per year. If the neoclassical growth model is correct, the expected coefficient on the
log of the growth of the effective labor force should be around -0.5. MRW found this to be
within the confidence intervals of the relevant coefficient only for growth regressions with
a sample of OECD high-income countries, but not for samples with developing countries.
For the sake of consistency, we include the MRW variable in the levels as wells as the
growth regressions.
Lastly, in light of evidence from the Quantile Regressions (see below) suggesting
that there is substantial heterogeneity in the intercepts, we also include regional dummy
variables in the 3SLS cross-sectional estimations of the static and dynamic models. The
regions are classified according to the World Bank's regional groups into seven groups as
described in the Appendix table.
VI. B. Descriptive statistics of net exports of natural resources and performance
Table 1 shows the average annual variation in the terms of trade, the average annual
growth of GDP per capita, the average value of the executive constraints index, and the
average growth of exports and imports of natural resources in current U.S. dollars. The data
show the average of each indicator for countries that were net exporters in all five-year
periods during 1980-2005, net importers in all periods, and two groups of switchers. The
lower panel of the table contains the corresponding information for the sample of Latin
American and Caribbean (LAC) economies with available data. It is notable that the vast
majority of countries in the data were either net exporters or importers, and the number of
switchers is small. In the case of LAC, the switchers were four small economies, including
El Salvador (became a net exporter on average only during 1985-1989), Guatemala,
Nicaragua, and Paraguay (net importer only in 1995-1999).
Interestingly, the net exporters of natural resources experienced terms of trade
deteriorations on average, while net importers experienced slight improvements. In the
sample period, the Prebisch-Singer hypothesis on the deterioration of the terms of trade
could be present. But variations in these relative prices would need to affect either the level
growth of the terms of trade, and regional dummies.
15
of GDP per capita or economic growth in the long-run for the curse to operate through this
channel. Relevant econometric evidence is discussed in the next section.
With respect to executive constraints, table 1 shows that for the global sample, net
exporters of natural resources do tend to have lower scores than net exporters. But this is
not true for the sample of LAC economies, where net exporters actually have higher
constraints on the executive branch, and in fact both net exporters and importers of natural
resources in LAC had higher average scores than the rest of the global averages during this
time period.
The growth performance of the net importers was superior to the average of the net
exporters. Surprisingly, however, the net importers of natural resources experienced
significantly higher growth rates of the value of natural-resource exports than the net
exporters, in the global sample. Hence even in this cursory look at the data, it is not clear
that growth of natural resource exports per se is in any systematic way related to low
growth.
VII. Results
The results are reported in tables 2-5. Table 2 shows the results from the Quantile
Regressions applied to the static model with both sets of control variables (but without
regional dummies). Table 3 reports the corresponding estimations of the dynamic model.
Tables 4 and 5 contain the 3SLS estimations of both models, including the results with the
larger samples that use the imputed values of historical population densities. All tables
report the probability values of the null hypothesis that the sum of the coefficient on net
exports and imports per worker are zero.17 For ease of exposition, the coefficients on most
of the other explanatory variables and the tests for equivalence of coefficients across
quantiles are not reported in the tables, but they are briefly discussed in the text below.
VII. A. Quantile regressions
In Table 2, the basic model is reported under column 1. Although the only the
coefficients on imports per worker are statistically significant across all quantiles, the F-
tests suggest that the effect of NX/L on GDP per capita is statistically significant for all
quantiles. The implied elasticities are about 0.75 for the first quantile, 0.66 for net exporters
and 0.46 for net importers in the second quantile, and 0.58 and 0.51 respectively in the third
quantile. This evidence implies that natural resources are a blessing for growth, but there is
notable heterogeneity across quantiles in this basic specification of the static model. It is
noteworthy that even if the average share of consumption of natural resources in national
income were 50%, the effect of K/L on GDP per capita would be quite large, ranging
between 0.23 and 0.38. Interestingly, the significantly highest elasticities are those of the
lower quantile, which also have the lower intercepts and are thus the low-income countries.
17For the sub-samples of net importers, the sum is the negative of the coefficient on the log of the absolute
value of net exports per worker plus the corresponding coefficient on imports per workers.
16
The fully specified static model reported under column 2 provides more mixed
evidence of a blessing effect. First, the implied effect is significant only among net
exporters in the three quantiles see the F-tests reported at the bottom of the table. These
blessing effects have implied elasticities in the range of 0.30-0.36, and there is no evidence
of cross-quantile heterogeneity. Contrary to the curse hypothesis, it seems that the
heterogenous blessing effect is associated with the macro volatility and/or the MRW factor
accumulation.
Among the other explanatory variables, only educational attainment is significant in
the second specification. The range of its coefficient is 0.75 in the lowest quantile to 0.49 in
the highest quantile. But they are statistically different across quantiles.
The results in Table 3 correspond to the dynamic growth model. The sums of the
relevant coefficients in the basic model are statistically significant at the 5% level only for
net importers in the second quantile and for net exporters and importers in the third
quantile. Although all point estimates are positive, those for the highest quantile are the
largest in magnitude, implying that a one percent increase in NX/L is associated with 1.1-
1.9 percentage points rise in the average annual growth rate of GDP per capita during 1980-
2005. Hence it seems that the inter-quantile heterogeneity in growth rates is different from
the heterogeneity observed in the static model results reported in Table 2. As in the static
model, however, the results under column 2 of Table 3 suggest that these blessing effects
disappear after controlling for macro volatility and factor accumulation.
Regarding other control variables in the dynamic growth models, we found
significant cross-quantile heterogeneity in the convergence rate and in the coefficient of
M/L in both specifications. Convergence is stronger among the second and third quantiles,
but not in the first. Although not statistically different across quantiles, the Sachs-Warner
reform index is positive and significant in the first two quantiles. Furthermore, the
investment rate is positive and significant in the upper two quantiles.
Overall the Quantile-Regressions evidence suggests that it is important to control
for heterogeneity in intercepts regardless of the model or sample. Also, there is significant
heterogeneity in the in the effect of natural resources, although it appears strongly only in
the static model. But the latter are not particularly strong when compared to the intercept
heterogeneity. Also, the existing studies that argue that there might be heterogeneity in the
effects of proxies of natural resource abundance, such as Bravo-Ortega and De Gregorio
(2007) might be confounding heterogeneity in other potential determinants of growth with
heterogeneity of the effects of natural resources. We leave this issue for future research.
More importantly, any estimation of these models without some heterogeneity of
intercepts is likely to confound the effects of the explanatory variables with differences in
the intercepts of the levels and growth rates of GDP per capita. The 3SLS estimations of the
static and dynamic models discussed below therefore include regional dummies. Future
research, however, could use panel-data estimators to assess the effect of natural resources
on growth while also controlling for heterogeneity.
VII. B. 3SLS estimates of the static and dynamic models with endogenous institutions
17
Tables 4 and 5 contain the relevant results. The first two columns in both tables
present the results of the basic model, columns 3 and 4 the fully specified model, and
columns 5-8 contain the corresponding results with the augmented samples utilizing the
imputed values of historical population densities.
In Table 4, the blessing effects appear strongly, although admittedly magnified, as
discussed earlier. Indeed, the elasticity of one for the sub-sample of net importers is
impossibly large. Moreover, there is no evidence that there is an indirect curse effect via
institutional constraints on the Executive branch of government, after controlling for
regional dummies. It is worth noting that historical population density remains a good
instrument, although only significant at the 10% in the basic model under column 1.
In the fully specified static model presented in columns 3 and 4, the blessing effect
remains significant only for the sample of net exporters of natural resources. This elasticity
is close to 0.25. The elasticity for the sub-sample of net importers declined from a bit over
one in the basic specification to about 0.19, and it is not statistically significant. Again,
there is no evidence of a statistically significant average effect of NX/L on executive
constraints. Furthermore, among the control variables, only the Sachs-Warner reform index
is statistically significant and positive, and the MRW variable capturing the accumulation
of effective labor units is negative and significant with coefficient estimate of
approximately -1.3.
The comparable results with the expanded samples are strikingly similar. In the
basic model, there are significant blessing effects among net exporters and importers, with
the respective elasticities of 0.49 and 0.38. Indeed, the latter estimate for net importers is
much more reasonable than the estimate of close to one that appeared in the restricted
sample. In the fully specified model under columns 7 and 8, the blessing effect remains
significant, but only at the 10% level. Interestingly, there might be a blessing effect via
institutions, because net exports per worker appear with a positive and significant
coefficient in column 8. But the point estimates are not significantly different from those
reported in the columns 1-4, thus suggesting that the results might not be driven by the
small sample.
Table 5 also suggests that natural resources could be a blessing, even for growth
during 1980-2005. In the restricted samples and in the basic specification, this blessing
effect is significant only among net importers. In contrast, in the fully specified model, but
with the smaller corresponding sample, the blessing effect is significant only among net
exporters. In both specifications, there is no evidence of an indirect curse effect via
institutions.
With the expanded samples, the blessing effect appears significant for both net
exporters and net importers in the basic specification. In the full specification, neither is
significant, again suggesting that if there is a blessing, it might operate through factor
accumulation and/or macro volatility. Perhaps more importantly, the changes in the point
estimates and their levels of significance due to the change of samples also suggest that
there is probably significant heterogeneity of effects across sub-samples of countries.
18
Among the control variables, the Sachs and Warner reform index appears positive
and significant in the estimation of models 1 and 3, but not in 5 or 7. The MRW variable of
the growth of effective units of labor is significant at approximately -0.04 in both the
restricted and expanded samples in models 3 and 7. This is about one tenth of the
magnitude predicted by neoclassical growth theory without controlling for the effects of the
other determinants of growth included in our specifications.
VIII. Summary and Conclusions
The review of the literature on the resource curse indicated that the evidence in
support of the curse is weak at best. There are important issues of measurement of relative
endowments, of potential heterogeneity in the effects of such endowments on development
and growth, and some of the international econometric evidence that appears to support the
curse hypothesis has been based on the use of weak proxies, and even on nonstandard
manipulations of influential data points.
This mixed evidence in favor of the resource curse also needs to face up to some
obvious historical facts, ranging from the successful development of now rich countries, to
the success of numerous developing economies, ranging from Rwanda to Chile. Moreover,
the idea that natural resources inevitably worsen political or other institutions that might be
important for development also seems to have ignored historical facts related to the
evolution of such institutions, and there are examples where "good" institutional
characteristics had emerged prior to the discovery of natural resources, for example, in the
case of Venezuela, which may explain the strong economic growth of this economy for
almost 50 years.
In the process of reviewing the existing literature and linking it to current data, we
have highlighted an important weakness in the empirical literature that uses trade-based
proxies for relative endowments. That is, the variables of choice of the propagators of the
curse are weak proxies for relative endowments. Our own previous work had ignored the
analytical contents of these empirical proxies, which might explain why we had been able
to so easily drive a stake of doubt into the heart of the resource demons.
Our preferred proxy, net exports per worker, while flawed, can help push the
literature forward, because in certain circumstances, the simultaneous estimation of
coefficients for the samples of net exporters and net importers can provide a magnified
effect of natural resources on development.
With new data, new econometric analyses provided definitive evidence that there is
no curse, not even indirectly by affecting the political institutions that would be most likely
to be affected by the curse-via-politics effects, which has been central in the literature on
the "point-source" nature of natural resources. In fact, the direct positive effect of natural
resources can be substantial, although we cannot be sure about its exact magnitude due to
the imperfect correlation between relative endowments and our trade-data proxy variable.
Furthermore, we found heterogeneity in the potential blessing effects of natural resource
endowments. More specifically, in the static model, the poorest countries benefit the most,
19
whereas in the dynamic model, the richest seem to have benefitted the most. But in both
cases, these blessing effects tend to disappear after controlling for macro volatility and
factor accumulation.
In the process, we are convinced that we know less than what we thought we knew,
especially from reading the existing literature. It remains a topic for future research to study
potential interactions between natural resources and human capital or innovation. But we
do know that there might be substantial international heterogeneity in the effects of other
determinants of growth, and there is certainly cross-country heterogeneity intercept. Hence
any cross-sectional estimation with cross-country data needs to control for mean shifters,
such as regional dummies. Preferably, panel-data estimators should be applied, because
they allow for controlling for fixed effects.
Similarly, much remains to be learned from historical case studies and perhaps from
cross-country statistical analysis of the interaction between natural resources and
institutions, in spite of the unreliable existing evidence concerning the curse-through-
politics hypothesis. From a policy viewpoint, institutional arrangements to smooth out the
economic consequences of natural-resource windfalls makes as much sense as more general
discussions about counter-cyclical fiscal policies. That is, in spite of the elusive resource
curse, we do not argue against common sense policies that in some countries might be
inextricable from natural riches.
20
References
Acemoglu, Daron, and James Robinson. 2005. "Unbundling Institutions." Journal of
Political Economy 113(5): 949-95.
Acemoglu, Daron, Simon Johnson, and James Robinson. 2001. "The Colonial Origins of
Comparative Development: An Empirical Investigation." American Economic
Review 91: 1369-1401.
Arellano, Manuel, and Olympia Bover. 1995. "Another Look at the Instrumental-Variable
Estimation of Error Component Models." Journal of Econometrics 68(1): 29-52.
Arellano, Manuel, and Stephen Bond. 1991. "Some Tests of Specification for Panel Data:
Montecarlo Evidence and an Application to Employment Equations." Review of
Economic Studies 58(2): 277-297.
Auty, Richard M. (2000) "How Natural Resources Affect Economic Development."
Development Policy Review 18: 347-364.
Auty, Richard M. "The Political Economy of Resource-Driven Growth." European
Economic Review 45 839-846.
Auty, Richard, editor. 2001. Resource Abundance and Economic Development. Oxford
University Press: New York.
Auty, Richard. 2005. "Patterns of Rent-Extraction and Deployment in Developing
Countries: Implications for Governance, Economic Policy and Performance."
Mimeographed. Lancaster University, Lancaster, United Kingdom.
Barro, Robert. 1991. "Economic Growth in a Cross-Section of Countries." Quarterly
Journal of Economics 106: 407-443.
Bond, Stephen R., Hoeffler, Anke. And Temple, Jonathan: 2001. "GMM Estimation of
Empirical Growth Models." CEPR Discussion Paper 3048, London, England.
Bravo-Ortega, Claudio, and José de Gregorio. 2007. "The Relative Richness of the Poor?
Natural Resources, Human Capital, and Economic Growth." Chapter 4 in Natural
Resources, Neither Curse nor Destiny, edited by D. Lederman and W.F. Maloney.
Washington, DC: Stanford University Press.
Brunnschweiler, Christina N. 2008. "Cursing the Blessings? Natural Resource Abundance,
Institutions, and Economic Growth." World Development 36(3): 399-419.
Cuddington, John T., Rodney Ludema, and Shamila A. Jayasuriya. 2007. "Prebisch-Singer
Redux." Chapter 5 in Natural Resources, Neither Curse nor Destiny, edited by D.
Lederman and W.F. Maloney. Washington, DC: Stanford University Press.
Dollar, David, and Aart Kraay. 2003. "Institutions, Trade, and Growth." Journal of
Monetary Economics 50(1): 133-62.
Dunning, Thad. 2008. "Model Specification in Instrumental-Variables Regression."
Political Analysis, forthcoming.
Durlauf, Steven N. & Kourtellos, Andros & Minkin, Artur, 2001. "The local Solow Growth
model," European Economic Review, Elsevier, vol. 45(4-6), pages 928-940, May.
21
Frankel, Jeffrey A., and David Romer. 1999. "Does Trade Cause Growth?" American
Economic Review 89(3): 379-399.
Findlay, Ronald and Mats Lundahl (1994) "Natural Resources, `Vent-for-Surplus' and the
Staples Theory." In Gerald M. Meir, editor, From Classical Economics to
Development Economics. New York: St. Martin's Press, 68-93.
Gelb, Alan. 1988. Oil Windfalls: Blessing or Curse? New York: Oxford University Press.
Glaeser, Edward L., Rafael La Porta, Florencio Lopez-de-Silanes, and Andrei Shleifer.
2004. "Do Institutions Cause Growth?" Journal of Economic Growth 9: 271-303.
Griliches, Zvi. and Hausman, Jerry A. 1986. "Errors in Variables in Panel Data". Journal of
Econometrics 3:93-118.
Gylfason (2001) "Natural Resources, Education and Economic Development" European
Economic Review 45:847-859.
Gylfason, Thorvaldur, Tryggvi Thor Herbertsson and Gylfi Zoega. 1999. "A Mixed
Blessing, Natural Resources and Economic Growth." Macroeconomic Dynamics
(3):204-225.
Isham, Jonathan, Michael Woolcock, Lant Pritchet, and Gwen Busby. 2005. "The Varieties
of Resource Experience: Natural Resource Export Structures and the Political
Economy of Economic Growth." The World Bank Economic Review 19(2): 141-74.
Knight, Malcolm, Norman Loayza, and Delano Villanueva. 1993. "Testing the Neoclassical
Theory of Economic Growth: A Panel Data Approach." IMF Staff Papers 40: 512-
541.
Leamer, Edward E. 1984. Sources of International Comparative Advantage: Theory and
Evidence. Cambridge, Massachusetts: The MIT Press.
Leamer, Edward E. 1995. The Heckscher-Ohlin Model in Theory and Practice. Princeton
Studies in International Finance 77. Princeton, New Jersey: Princeton University.
Lederman, Daniel, and L. Colin Xu. 2007. "Comparative Advantage and Trade Intensity:
Are Traditional Endowments Destiny?" Chapter 10 in Natural Resources, Neither
Curse nor Destiny, edited by D. Lederman and W.F. Maloney. Washington, DC:
Stanford University Press.
Levine, Ross, and David Renelt. 1992. "A Sensitivity Analysis of Cross-Country Growth
Regressions." American Economic Review 82(4): 942-963.
Levine, Ross, Norman Loayza, and Thorsten Beck. 2000. "Financial Intermediation and
Growth: Causality and Causes." Journal of Monetary Economics 46: 31-77.
Manzano, Osmel and Roberto Rigobón. 2007. "Resource Curse or Debt Overhang?"
Chapter 3 in Natural Resources, Neither Curse nor Destiny, edited by D. Lederman
and W.F. Maloney. Washington, DC: Stanford University Press.
Mankiw, Gregory, David Romer, and David Weil. 1992. "A Contribution to the Empirics
of Economic Growth." Quarterly Journal of Economics 107(2): 407-437.
Mehlum, Halvor, Karl Moene and Ragnar Torvik. 2006. "Institututions and the Resource
Curse." The Economic Journal 116:1-20.
22
Murshed, S. Mansoob (2004) "When Doe Natural Resource Abundance Lead to a Resource
Curse? Environmental Economics Programme Discussion Paper 04-11,
International Institute for Environment and Development, The Hague.
Neumayer, Eric. 2004. "Does the `Resource Curse' Hold for Growth in Genuine Income as
Well?" World Development 32(10): 1627-1640.
Nunn, Nathan. 2008. "The Long-Term Effects of Africa's Slave Trades." The Quarterly
Journal of Economics 123(1): 139-76.
Peretto, Pietro F. 2008. "Is the `Curse of Natural Resources' Really a Curse?"
Mimeographed, Department of Economics, Duke University, Durham, NC,
http://econ.duke.edu/~peretto/.
Prebisch, Raúl (1950), "The Economic Development of Latin America and its Principal
Problems, reprinted in Economic Bulletin for Latin America, Vol. 7, No. 1, 1962, 1-
22.
Pritchett, Lant. 2000. "Understanding Patterns of Economic Growth: Searching for Hills
among Plateaus, Mountains, and Plains." The World Bank Economic Review 14(2):
221-250.
Rodríguez, Francisco. 2007. "Cleaning Up the Kitchen Sink: Growth Empirics When the
World is Not Simple." Department of Economics, Wesleyan University,
http://frrodriguez.web.wesleyan.edu/.
Rodríguez, Francisco, and Adam J. Gomolin. 2008. "Anarchy, State, and Dystopia:
Venezuelan Economic Institutions before the Advent of Oil." Bulletin of Latin
American Research 28(1): ???.
Roodman, David. 2006. "How to Do xtabond2: An Introduction to "Difference" and
"System" GMM in Stata." Center for Global Development Working Paper 103,
Washington, DC.
Ross, M.L. 1999. "The Political Economy of the Resource Curse"
Sachs, Jeffrey D., and Andrew Warner. 1995 (revised in 1997). "Natural Resource
Abundance and Economic Growth." National Bureau of Economic Research
Working Paper 5398, Cambridge, Massachusetts.
Sachs, Jeffrey D., and Andrew Warner. 2001. "The Curse of Natural Resources." European
Economic Review 45: 827-838.
Sachs, Jeffrey, and Andrew Warner. 1995a. "Economic Reform and the Process of Global
Integration." Brookings Papers on Economic Activity 1: 1-95.
Sala-i-Martin, Xavier, Gernot Doppelhofer, and Ronald I. Miller. 2004. "Determinants of
Long-Term Growth: A Bayesian Averaging of Classical Estimates (BACE)
Approach." American Economic Review 94: 813-35.
Sala-i-Martin, Xavier, and Arvind Subramanian. 2003. "Addressing the Resource Curse:
An Illustration from Nigeria." NBER Working Paper 9804, Cambridge,
Massachusetts.
23
Servén, Luis. 2003. "Real Exchange Rate Uncertainty and Private Investment in LDCs."
Review of Economics and Statistics 85: 212-218.
Sevestre, Patrick, and Alain Trognon. 1996. "Dynamic Linear Models." In Matyas, L. and
P. Sevestre, The Econometrics of Panel Data. Boston: Kluwer Academic
Publishers,100-118.
Singer, Hans W. 1950. "U.S. Foreign Investment in Underdeveloped Areas: The
Distribution of Gains Between Investing and Borrowing Countries." American
Economic Review, Papers and Proceedings 40: 473-485.
Stjins, Jean-Philippe. 2005. "Natural Resource Abundance and Economic Growth",
Reources Policy, 30(2): 107-130.
Tornell, Aaron and Philip R. Lane (1999) "The Voracity Effect." American Economic
Review, 89 (1): 22-46
Torvik Ragnar (2002) "Natural Resources, Rent Seeking and Welfare." Journal of
Development Economics 67(2): 455-470.
Wacziarg, Romain, and Karen Horn Welch. 2002. "Integration and Growth: An Update."
Mimeographed. Stanford University.
Windmeijer, Frank. 2005. "A finite sample correction for the variance of linear efficient
two-step GMM estimators." Journal of Econometrics 126 (1): 25-51.
24
Figure 1. Net Exporters of Natural Resources during 1980-2005 and Real GDP per Capita
in 2005
4
SAU NOR
ISL
2 GAB KWT
e NZL CAN
AUS
TTO FIN
forc
COG VEN ARE
MYS DNK
DZA SUR CHL NLD
labor 0 ECU IRL
ARG
KAZ
LBR PNG IRN RUS SWE
CIV NAM CRI
ports/xEt NGA BOL URY
CMR IDN COL ZAF y = 1.2584x - 12.102
05 PER MEX
ZMB MRT MNG
YEM SYRHND
SLBTJKVNMBTN R2 = 0.5593
ZWE AZEPRY BRA
Nes 20-0 -2
MWI GTM LVA
198 GIN
TGO THA
ourceseR FJI
NERMDGUGAKENGHA
MOZ
-4
GNB MLI
Natural CAFTZA
log
-6 MDA
ETH
BDI
-8
5 6 7 8 9 10 11
log GDP pc 2005
25
Figure 2. Net Importers of Natural Resources during 1980-2005 and Real GDP per Capita
in 2005
2
LUX
JPN CHESGPHKG
1 MLT MAC
KOR
ISRITABEL
e PRT AUT
JOR LCA GRC ESP
0 GBR
FRA
forc MDV
DJI LBN USA
VCT SVK
BHR SVN
CPV LTU
boral HRV BLR
-1 TONWSM MKDROMBWAPAN CZE
LSO BIHJAMALB TUR TUN y = 1.0885x - 11.227
ports/ GMB VUT HUN
ARM
UKR R2 = 0.4602
mIt 50 -2 SEN MAR
PHL EGY DOM MUS
BLZ
Nes 20- ERI COM
PAK SLV BGR
80 GEO
19-3 BGD NIC LKA POL
SWZ
source RWA KHM IND
Re NPL
-4 BFA
ural SLE SDN
CHN
BEN
Natglo -5
-6 EST
-7
6 6.5 7 7.5 8 8.5 9 9.5 10 10.5 11
log GDP pc 2005
26
Table 1. Net Exporters, Net Importers, and Switchers: Selected Indicators
1980 -2005 Terms of GDP pc Executive NR NR
Trade growth Constraint imports exports
growth growth growth
Sample: All Countries, observations with data
Net Natural Resource
Exporters in all periods -0.41% 0.61% 3.97 2.22% 1.41%
(53) (59) (58) (54) (54)
Net Natural Resource
Importers in all periods 0.11% 2.18% 4.88 3.31% 3.55%
(63) (72) (57) (69) (69)
Net Exporters that became
Net Importers in any period
after 1985 0.83% -0.33% 3.60 -2.47% -0.64%
(7) (11) (10) (10) (10)
Net Importers that became
Net Exporters in any period
after 1985 0.21% 1.29% 4.05 3.35% 7.40%
(12) (14) (14) (13) (13)
Sample: LAC Countries
Net Natural Resource
Exporters in all periods -0.42% 0.34% 5.48 2.56% 2.64%
(15) (14) (13) (15) (15)
Net Natural Resource
Importers in all periods 0.01% 1.56% 4.99 -0.68% -2.07%
(6) (9) (4) (8) (9)
Net Exporters that became
Net Importers in any period
after 1985 0.42% -0.63% 4.34 3.94% 4.53%
(3) (3) (3) (3) (3)
Net Importers that became
Net Exporters in any period
after 1985 -0.49% 0.88% 5.00 6.00% 2.66%
(1) (1) (1) (1) (1)
Notes: The numbers inside parentheses are the countries in each group.
27
Table 2. Quantile Regression Results of the Static Model without Regional Dummies
Dependent Variable: Log GDP per capita in 2005 (PPP$)
q25
Positive_ln(NX/L) 0.205 0.120
(1.93) (1.17)
Negative_ln_abs(NX/L) -0.015 0.028
(0.20) (0.25)
Positive_ln(M/L) 0.541 0.184
(4.56)** (1.61)
Negative_ln_(M/L) 0.730 0.304
(5.69)** (0.92)
q50
Positive_ln(NX/L) 0.087 0.098
(1.21) (1.96)
Negative_ln_abs(NX/L) 0.068 0.084
(1.28) (0.98)
Positive_ln(M/L) 0.569 0.261
(6.60)** (2.08)*
Negative_ln_(M/L) 0.528 0.164
(8.07)** (1.73)
q75
Positive_ln(NX/L) 0.122 0.088
(2.57)* (1.32)
Negative_ln_abs(NX/L) 0.026 0.086
(0.52) (0.82)
Positive_ln(M/L) 0.460 0.240
(6.83)** (1.31)
Negative_ln_(M/L) 0.533 0.311
(9.20)** (1.84)
Constant q25 9.088 5.215
(40.37)** (3.94)**
Constant q50 9.486 6.294
(52.07)** (3.54)**
Constant q75 9.537 6.732
(65.12)** (2.74)**
Sum of Coefficients (F-Tests, P-values)
q25: Postive_ln(NX/L)+Positive_ln(M/L)=0 0.746 [0.00] 0.304 [0.00]
q25: - Negative_ln_abs(NX/L)+Negative_ln(M/L)=0 0.745 [0.00] 0.276 [0.32]
q50: Postive_ln(NX/L)+Positive_ln(M/L)=0 0.656 [0.00] 0.359 [0.00]
q50: - Negative_ln_abs(NX/L)+Negative_ln(M/L)=0 0.460 [0.00] 0.080 [0.44]
q75: Postive_ln(NX/L)+Positive_ln(M/L)=0 0.582 [0.00] 0.328 [0.04]
q75: - Negative_ln_abs(NX/L)+Negative_ln(M/L)=0 0.507 [0.00] 0.225 [0.47]
Observations 103 74
Absolute value of t statistics in parentheses
* significant at 5%; ** significant at 1%
Other control variables included (see text) but not reported.
Model 1 includes: Sachs-Warner reform index (average during 1980-2005), average annual growth of terms
of trade.
28
Model 2 includes: Variables in model (1) plus real-exchange rate volatility, log of average investment rate,
log average years of education, and MRW.
29
Table 3. Quantile Regression Results of the Dynamic Model without Regional Dummies
(1) (2)
Dependent variable: Average annual GDP per capita growth, 1980-2005
q25
Positive_ln(NX/L) 0.000 0.001
(0.18) (0.07)
Negative_ln_abs(NX/L) -0.002 -0.003
(0.81) (0.33)
Positive_ln(M/L) 0.004 0.000
(1.21) (1.03)
Negative_ln_(M/L) 0.003 0.000
(0.74) (1.52)
q50
Positive_ln(NX/L) 0.002 0.000
(0.85) (0.33)
Negative_ln_abs(NX/L) -0.002 -0.007
(0.68) (1.53)
Positive_ln(M/L) 0.004 -0.001
(1.01) (0.16)
Negative_ln_(M/L) 0.008 0.009
(1.80) (0.01)
q75
Positive_ln(NX/L) 0.003 0.000
(1.25) (0.17)
Negative_ln_abs(NX/L) -0.004 -0.001
(1.05) (0.77)
Positive_ln(M/L) 0.008 0.004
(2.11)* (0.11)
Negative_ln_(M/L) 0.015 0.000
(2.24)* (0.05)
q25: Constant 0.003 -0.037
(0.05) (0.89)
q50: Constant 0.078 0.148
(1.61) (0.55)
q75: Constant 0.19 0.057
(3.61)** (2.46)*
F-tests, P-values
q25: Postive_ln(NX/L)+Positive_ln(M/L)=0 0.004 [0.30] 0.001 [0.91]
q25: - Negative_ln_abs(NX/L)+Negative_ln(M/L)=0 0.005 [0.14] 0.003 [0.96]
q50: Postive_ln(NX/L)+Positive_ln(M/L)=0 0.006 [0.06] -0.001 [0.75]
q50: - Negative_ln_abs(NX/L)+Negative_ln(M/L)=0 0.010 [0.16] 0.016 [0.61]
q75: Postive_ln(NX/L)+Positive_ln(M/L)=0 0.011 [0.00] 0.004 [0.28]
q75: - Negative_ln_abs(NX/L)+Negative_ln(M/L)=0 0.019 [0.02] 0.001 [0.12]
Observations 84 74
Absolute value of t statistics in parentheses
* significant at 5%; ** significant at 1%
Other control variables included (see text) but not reported.
Model 1 includes: Log GDP per capita in 1980 (PPP$), Sachs-Warner reform index (average during 1980-
2005), average annual growth of terms of trade.
30
Model 2 includes: Variables in model (1) plus real-exchange rate volatility, log of average investment rate,
log average years of education, and MRW.
31
Table 4. 3SLS Estimates of the Static Model with Endogenous Institutions
(1) (2) (3) (4) (5) (6) (7) (8)
Log
GDP Log
Log per GDP
GDP Log Log Log capita Log per Log
per Execu GDP per Executi in Executi capita Executi
capita tive capita in ve 2005 ve in ve
in 2005 Const 2005 Constr (PPP Constr 2005 Constr
Dependent Variables: (PPP$) raints (PPP$) aints $) aints (PPP$) aints
Positive_ln(NX/L) 0.156 -0.02 0.015 0.045 0.14 -0.013 0.055 0.059
(2.29)* (0.65) (0.20) (1.28) (1.96) (0.54) (0.95) (2.09)*
Negative_ln_abs(NX/L) -0.105 -0.027 0.149 -0.061 0.069 -0.046 0.045 -0.031
(0.91) (0.69) (1.24) (1.51) (1.03) (1.64) (0.72) (0.95)
Positive_ln(M/L) 0.351 0.233 0.35 0.131
(3.47)* (2.77)
* (2.16)* ** (1.38)
Negative_ln_abs(M/L) 0.932 0.334 0.448 0.178
(4.50)* (5.01)
* (1.65) ** (1.57)
Log Executive Constraints 1.416 1.427 1.972 1.148
(2.72)*
* (2.28)* (1.60) (1.85)
Log population density
1500 -0.079 -0.105
(3.12)*
(1.96) *
Imputed Log Population
Density 1500 -0.084 -0.102
(3.09)*
(2.14)* *
Sum of Coefficients [F-Tests, P-
Sum of Coefficients [F-Tests, P-values] values]
Postive_ln(NX/L)+Positiv 0.507 0.248 0.490 0.19
e_ln(M/L)=0 [0.000] n.a. [0.0369] n.a. [0.00] n.a. [0.09] n.a.
- 1.040
Negative_ln_abs(NX/L)+ [0.0012 0.0190 0.38 0.13
Negative_ln(M/L)=0 ] n.a. [0.5459] n.a. [0.00] n.a. [0.39] n.a.
Observations 51 51 43 43 98 98 71 71
R-squared 0.77 0.63 0.89 0.73 0.65 0.62 0.92 0.75
Absolute value of t statistics in
parentheses
* significant at 5%; ** significant at
1%
Other control variables included (see text) but
not reported.
Models 1 and 5 include Log GDP per capita in 1980, Sachs-Warner reform index (average during 1980-
2005), average annual growth of terms of trade, regional dummies.
Models 2 and 6 include average annual growth of the terms of trade
(1980-2005), regional dummies.
Models 3 and 7 include variables in model (1) plus real-exchange rate volatility, log of average investment
rate, log average years of education, and MRW.
Models 4 and 8 include the same variables as in model (2).
32
Table 5. 3SLS Estimates of the Dynamic Model with Endogenous Institutions
(1) (2) (3) (4) (5) (6) (7) (8)
Avg. Avg.
GDP per Log GDP per Log Avg. GDP Log Avg. GDP Log
Capita Executi Capita Executi per Capita Executi per Capita Executi
Growth, ve Growth, ve Growth, ve Growth, ve
1980- Constra 1980- Constra 1980- Constra 1980- Constra
Dependent Variables: 2005 ints 2005 ints 2005 ints 2005 ints
Positive_ln(NX/L) -0.0021 -0.0205 -0.0012 0.0450 -0.0003 0.0001 -0.0007 0.0590
(1.18) (0.65) (0.76) (1.28) (0.17) (0.00) (0.37) (2.09)*
Negative_ln_abs(NX/L) -0.0028 -0.0270 0.0007 -0.0610 -0.0047 -0.0181 -0.0029 -0.0310
(1.10) (0.69) (0.26) (1.51) (2.32)* (0.56) (1.17) (0.95)
Positive_ln(M/L) 0.0060 0.0026 0.0056 0.0030
(2.18)* (1.12) (2.15)* (1.01)
Negative_ln_abs(M/L) 0.0126 0.0062 0.0104 0.0044
(2.40)* (1.37) (2.64)** (1.21)
Log Executive Constraints 0.0212 0.0020 0.2732 0.0211
(1.99) (1.58) (1.29) (0.86)
Log population density 1500 -0.0790 -0.1050
(3.12)*
(1.96) *
Imputed log population density
1500 -0.1022 -0.102
(3.09)* (3.09)*
* *
F-Tests, P-values F-Tests, P-values
Postive_ln(NX/L)+Positive_ln( 0.004 0.001 0.005 0.002
M/L)=0 [0.144] n.a. [0.037] n.a. [0.0478] n.a. [0.4679] n.a.
:-
Negative_ln_abs(NX/L)+Negativ 0.02 0.01 0.02 0.01
e_ln(M/L)=0 [0.0368] n.a. [0.5459] n.a. [0.0043] n.a. [0.1668] n.a.
Observations 51 51 43 43 79 79 71 71
R-squared 0.77 0.63 0.82 0.73 0.51 0.68 0.92 0.75
Absolute value of t statistics in parentheses
* significant at 5%; ** significant at 1%
Other control variables included (see text) but not
reported.
Models 1 and 5 include Log GDP per capita in 1980, Sachs-Warner reform index (average during 1980-
2005), average annual growth of terms of trade, regional dummies.
Models 2 and 6 include average annual growth of the terms of trade
(1980-2005), regional dummies.
Models 3 and 7 include variables in model (1) plus real-exchange rate volatility, log of average investment
rate, log average years of education, and MRW.
Models 4 and 8 include the same variables as in
model (2).
33
Data Appendix
Table A1. Data, Variable Definitions and Sources
Variable Definition Source
Natural Natural resources net exports divided by the labor force. Natural WDI and UN
Resources/labor resources net exports is defined as exports minus imports of natural COMTRADE
force resources related goods based on Leamer (1995)a commodity
clusters. Labor force is the population between 15-64 years.
GDP per capita Average yearly growth of real GDP per capita (Constant Prices: Authors'
growth 1980-2005 Chain series) in the period 1980-2005. construction
using Penn
World Tables
log GDP per capita Real GDP per capita (Constant Prices: Chain series). Ratio of total Penn World
GDP to total population in 2005. Tables
Openness Percentage of years with open economic regime. A country has a Sachs and
closed trade policy if it has at least one of the following Warner (1995a)
characteristics: 1. Nontariff barriers (NTBs) covering 40 percent or updated by
more of trade. 2. Average tariff rates of 40 percent or more. 3. A Wacziarg and
black market exchange rate that is depreciated by 20 percent or Welch (2003)
more relative to the official exchange rate, on average, during the
1970s or 1980s. 4. A socialist economic system 5. A state
monopoly on major exports.
Terms of Trade Growth of the external terms of trade, defined as the ratio of an WDI
growth export price index to an import price index of goods and services
Std. Dev. Real Standard deviation of monthly interannual changes in real effective Authors'
Effective exchange rates construction
Exchange Rate using IMF data
Executive Institutionalized constraints on the decision making powers of chief Polity IV 2004
Constraints executives, weather individuals or collectivities. It is a 1-7 category
scale, higher score means more constraint on the executive. Equals
one if country is not independent.
log Investment Natural log of the share Penn World
of investment over GDP Tables
log School Natural log of years of schooling of the adult population Barro and Lee
Attainment (2000)
MRW Based on Mankiw, Romer and Weil (1992), the variable is the WDI
natural log of the average growth of the labor force plus 0.05. The
constant 0.05 is assumed to be the sum of depreciation rate and
technological growth.
Population density Total population divided by total arable land in 1500 A.D. McEvedy and
in 1500 Jones (1978) as
cited in
Acemoglu, et al.
(2002).
Regions dummies 8 Dummies for world regions as defined by the World Bank: East The World Bank
Asia & Pacific, Europe & Central Asia, High income: OECD, High
income: non OECD, Latin America & Caribbean, Middle East &
North Africa, South Asia, Sub-Saharian Africa
a. SITC sections 0-9, 11, 12, 21-29, 32-35, 41-43, 63, 64, 68, 94
34
Table A2. Summary Statistics
Cross-Section; All Variables in 2005
Std.
Variable Obs Mean Dev.
GDP per capita growth 1980-2005 138 0.01359 0.02048
log GDP pc in 1980 152 8.30638 1.10774
Executive Constraints 155 1.23111 0.61816
Natural Resources/labor force > 0 133 0.67116 2.48421
Natural Resources/labor force < 0 133 0.15405 0.57328
Openness 141 0.55319 0.41906
Std. Dev. Real Effective Exchange
Rate 175 0.06932 0.0793
log Investment 184 2.47441 0.57929
log School Attainment 97 1.58728 0.62581
MRW 185 -2.6602 0.20589
Terms of Trade growth 145 0.00302 0.04458
log Population density in 1500 97 0.48645 1.5247
Periods Data
Std.
Variable Obs Mean Dev.
GDP per capita growth 1980-2005 1180 1.65291 3.38209
log GDP pc 1180 8.25806 1.11805
Natural Resources/labor force > 0 1133 7.10709 6.80923
Natural Resources/labor force < 0 1133 6.0296 6.76471
Executive Constraints 1238 1.12357 0.71623
log Investment 1430 2.5931 0.68728
log School Attainment 822 1.42065 0.75464
Terms of Trade growth 914 -0.0013 0.04278
Openness 904 0.44757 0.48416
log Population density in 1500 882 0.50312 1.51883
35
Appendix A3 Details on Sources of Bias in NX/L Proxy
In the context of our two-sector growth model, an exogenous labor-productivity
shock associated with sector 1 and, for simplicity, assuming that the shock does not affect
the allocation of labor (i.e., Lnr = 0) or the stock of K and L or the consumption shares,
a1 L
the marginal change in net exports per worker associated with a marginal change in
productivity:
(12) (NX / L)a1L = -cnr ·(1-lnr ).
This consumption effect of a productivity improvement is strictly negative.18 Hence the
reverse effect of productivity growth on observed net exports of natural resources would be
reflected in a decline in net exports of natural resources per worker, and any estimation of
an income or growth model with NX/L as an explanatory variable will yield downward
biased estimates of the effect of NX/L on Y. Allowing for a re-allocation of labor into
sector 1 as a consequence of the improvement in productivity in that sector would further
reduce net exports of natural resources per worker, thus further biasing estimates of the
effect of NX/L on Y.
A productivity change in the natural-resource sector would affect NX/L as follows:
(13) (NX / L)anr = (1-cnr )· . K b
L
If there is a curse, anr < 0 , then NX/L would decline as exports earnings fall more than
imports because cnr>0. The fall in gross imports of natural resources would be proportional
to the share of income spent in the consumption of natural resources: anr · cnr · < 0 K b
L
when anr < 0 . Under a curse, the negative effect of NX/L would be attenuated by the fall
of imports. If natural resources are a blessing in terms of improvements in labor
productivity in either sector, then imports would rise with income (which is the reason why
the coefficient on NX/L is biased downward in income functions). Hence reverse causality
under a blessing effect would bias the estimated coefficient of NX/L downwards due to the
rise of imports of natural resources. However tempting it is to estimate (8) and (9) with
NX/L as a proxy for unobserved K/L, this discussion highlights biases due to the
consumption effect that will contaminate estimates of the effect of NX/L on Y.
To address the endogeneity biases, we estimate augmented versions of the static and
dynamic models, which include the natural logarithm of imports of natural resources per
18Please note that a productivity shock affecting the natural-resource sector would correspond to a positive (a
"blessing") shock on output. The earlier version of this paper had erroneously confounded this positive effect
with an endogenous decline in NX/L due to a positive productivity shock in sector 1.
36
worker (M/L) as an additional regressor. Table 1 summarizes the expected coefficients
under three scenarios. If there is a resource curse, then the coefficients on NX/L and M/L
would be negative see discussion on equation (11) above. If natural resources are a
blessing, then the coefficient on NX/L and M/L should be positive. The attenuation effect
of the reverse causality effect would be captured by the coefficient on imports.
More formally, the estimated model can be re-written to account explicitly for the
reverse causality effect:
(14) yi = a + d - dx - dM (ln(NX/L)i) + f'Xi + ei ,
( )
where coefficient b in (9) equals d - d X - dM , which is the sum of the direct effect minus
the reverse-causality effect. In the context of our two-sector growth model and assuming
that the relevant effects operate only through changes in productivities in either sector, the
direct effect can be derived from equation (8):
d =
a1 L
(NX /L) (L - Lnr)+(NX anr
/ L) KbL1 (NXy/L).
-b · The reverse causality effect
can be derived from equations (12) and (13):
d X + dM = aiL · cnr(1- lnr )+ anr · cnr ·
K b (NXy/L), which is a composite
L
consumption effect coming from increases in factor productivity in each sector and will
equal the effect of y on gross imports (exports) of natural resources per worker. By
including M/L in the estimation equation we can get an estimate of dM .19 Thus the
empirical income function becomes:
(15) yi = a + d - dx (ln(NX/L)i) + dM (ln(M/L)i) + c'Xi + ei .
We assume that consumption effect on exports is equal to the consumption effect on
imports. Thus the sum of d - dx plus dM approximates an unbiased estimate of the
elasticity of y with respect to NX/L.
A remaining weakness of (15) is that we are interested in estimates of
(K /L)· y
y (K /L) whereas d = y · (NX /L). The function linking K/L with NX/L
NX y
L
was presented above in the discussion of equation (11):
19Alternatively, we could include X/L, but the resulting estimate could be contaminated by a supply-side
effect on exports.
37
(NX L)= b(1
(K L) -cnr )anr(Lnr )-b Kb 0. The supply and consumption components in our
-1
b
proxy are separable. Let z = (K) banr K
Lnr is the
(NX L)· (NX ) = z
(K L) S- zC , where zS = NX
b
un-attenuated supply-driven elasticity, and zC = cnrbanr K
Lnr
NX is the consumption-
driven attenuation of the corresponding elasticity.
It is trivial to show that the marginal effect of K/L on y is the product of the
marginal of effect of NX/L on y times the marginal effect of K/L on NX/L. After stripping
out d with the inclusion of M/L as a regressor in (15), the estimate of the effect of K/L on
y, based on our proxy, is still biased:
(16) (K /L)· y
y (K /L) = (d -dX + dM zS - zC)= (1-cnr)d -dX + dM .
)( ( )
That is, the downward bias in z produces a multiplicatively positive bias for the inference
about the effect of K/L on y based on estimates of the effect of NX/L on y. In other words,
because the supply elasticity of NX/L with respect to K/L is under-estimated due to the
consumption effect, any blessing or curse effect apparent in the approximated elasticity of y
with respect to NX/L will be an exaggerated measure of the underlying effect of K/L on y.
Multiplying (d -dX + dM by one plus the share of income spent in the consumption of
)
natural resources would yield an estimate closer to an unbiased estimate of the elasticity of
y with respect to K/L.
38