NATURAL RESOURCE CURSE AND EXTERNALITIES FROM NATURAL RESOURCE EXPORTS
Author: FREDERICO ROCHA e-mail: [email protected]
Associate Professor at the Instituto de Economia – Universidade Federal do Rio de Janeiro Address: Av. Pasteur, 250 Rio de Janeiro – RJ – Brazil CEP: 22.290-240 Tel: +(55) 21 38725242; Fax: +(55) 21 25418148 Visiting Fellow at University of Siena (January 2010-December 2010)
Abstract Some economists have argued that natural resources provide low level of externalities and, as a consequence, that resource dependent economies tend to grow their GDP per capita at a slower rate. This paper addresses the issue of externality generation from natural resource exports. Using panel data regressions, the paper shows that natural resource exports generate positive externalities that are at least as high as those generated by manufacture exports and, thus, resource rich countries that manage to grow their exports at a fast rate do not suffer from the natural resource curse. Keywords: natural resource; growth; externalities; exports; resource curse JEL: O13; Q33; Q32
The development economics literature has found evidence that resource rich economies have been outperformed by resource poor economies in terms of GDP per capita rates of growth (Sachs and Warner 1997). The literature has followed two trajectories in dealing with this apparent paradox. On the one hand, there have been studies that challenge or reproduce Sachs and Warner’s results through the use of alternative measures and techniques. On the other hand, there have been attempts to find the transmission channels for the curse, that is, to explain how the abundance of natural resources may have a negative impact on growth. Many transmission mechanisms have been examined. One of the earliest discussed mechanisms is the lack of externalities provided by natural resources, due to feeble linkages backward and forward linkages or low demand of skills and infrastructure (Singer 1950, Hirschman 1958). This argument has been questioned recently (Maloney 2008, Wright and Czelusta 2008). Singer (1950) and Hirschman (1958) argued for the connections that manufacturing would have with the rest of the economy and that would be absent in natural resource staples. This would be responsible for the structural heterogeneity of some resource rich developing economies, where a high productivity exporting sector does not provide positive spillovers for domestic producers, increasing the productivity distance among the sectors (Furtado 1957, 1974 examining the Venezuelan case). This kind of hypothesis has also been present in Dutch disease models. Dutch disease theorists (Van Winjberger 1984, Torvik 2001) have elaborated models in which an external foreign currency shock over evaluates domestic currency and shrinks the tradable manufacturing sector that is unable to enjoy increasing returns from additional production and therefore in the next period presents a smaller size that damages domestic economy. The external shock may be a consequence of a price increase in the natural resource exporting product. However, no attention has been given to the hypothesis that the foreign currency premium is result of a production movement. 2
Therefore, extensions of the Dutch Disease paradigm take for granted that natural resource exports show low level of externalities (Sachs and Warner 1999). More recent work argues however that there is no reason for the natural resource tradable sector to provide less externality than the manufacturing sector (Maloney 2008 and Wright and Czelusta 2008). They argue that, on the contrary, successful resource rich economies have been able to grow through the externalities provided by these activities. This paper aims to contribute to this literature arguing that there are as much externalities from the natural resource exports sector as there are from the manufacture exports sector. The point also attempts to show that the real curse is on some resource rich economies that fail to expand their production. These tasks are carried out through the testing of two hypotheses. First, the paper tests the hypothesis that the great problem about resource rich economies is the lack of dynamism of its external sector. Second, using a model elaborated by Feder (1982) that measures spillovers from the tradable sector, this paper compares spillovers provided by the natural resource and the manufacture exports sector. The remaining of this paper is composed of four sections. Section 2 provides the analytical background for the paper, highlighting previous literature contributions and formulating the two hypotheses to be tested. Section 3 is dedicated to the formulation of the models to be used in the empirical section and to the presentation of the database. Section 4 presents and comments the empirical results. The final section drives the main conclusions from the paper.
Natural resource curse literature has gained large attention from scholars and policy makers since the publication of Sachs and Warner (1995, 1997, 1999, 2001) series of papers. In these papers, using cross-section OLS regressions, the authors show thorough evidence for the 19701990 period that countries that were specialized in natural resources presented lower average
rates of growth of GDP per capita. They use four distinct measures for natural resource abundance: (i) the ratio of natural resource exports to GDP, (ii) the share of natural resource exports in total exports, (iii) the ratio of mineral production to total GNP, and (iv) per capita agricultural land extent. Though the careful examination of the authors, a large literature has followed these papers trying to reproduce or challenge their initial findings. The main strategy has been to use alternative measures of resource abundance in order to show different results. Stijns (2005) measures natural resource abundance by the present value of mineral reserves and finds no correlation between this variable and economic growth. Lederman and Maloney (2008) use the Leamer index (natural resources net exports/labor force) and, in a panel data setting, find it not significant with respect to GDP per capita growth. A second stream of challenges on Sachs and Warner’s tests arises from the use of cross-section regressions. Measures of resource abundance usually maintain some endogeneity with respect to growth variables. It is difficult to address the issue of causality whenever using cross-section settings. Some other country specific characteristics may be influencing the result. Panel data settings appear to be more adequate for the treatment of these issues. Some recent work has tried to overcome this empirical shortcoming with results that do not allow for the generalization of the resource curse hypothesis (Lederman and Maloney 2008, Manzano and Rigobón 2008 and Bravo-Ortega and De Gregorio 2008). Literature has also been concerned about investigating the transmission channels for the resource curse. In Sachs and Warner (1997) they established some lines of work that should be undertaken in order to investigate the issue a bit further. First, they argued for the role played by currency over evaluation, explored later in Sachs and Warner (1999, 2001), which conduct to high wages and de-industrialization as argued in the Dutch Disease literature (Corden and Neary 1982) and by Furtado (1974) in the analysis of the Venezuelan case. Though there is evidence that natural resource rich economies suffer from high prices and, in some cases, of
factor re-alocation, what may also be transformed in a crowding out argument, the literature shows little evidence over its impact on long run growth. Second, part of the literature has been dedicated to show the interaction between natural resource curse and institutions. They argue that natural resource rich countries are more likely to present weak institutions that promote rent-seeking behavior and mismanagement that would displace efforts towards less productive activities. Mehlum et al. (2006) show that when one includes in Sachs and Warner (1997) regressions a variable composed by the interaction of a institutional quality variable and the resource abundance variable the direct effect of the resource abundance variable becomes stronger but the interaction variable has the opposite (positive) sign that offsets the effect of the resource abundance variable, that is, natural resource intensive countries that have good institutions do not suffer from the curse. Studying the isolated case of Norway, Larsen (2005, 2006) shows how strong institutions have helped Norway to overcome the natural resource curse and surpass its neighbors’ level of growth and GDP per capita. Third, Gylfasson et al. (1999) and Gylfasson (2001) argue that the lack of investment in education is a transmission mechanism for the natural resource curse. This would be a consequence of the low requirement for human capital in resource intensive activities that would drive to a reduction on long term growth. Gylfasson (2001) shows evidence that natural resource intensive nations spend less and have a lower average level of education in their population. Nonetheless, he fails to show causality. Bravo-Ortega and De Gregorio (2008) gather some evidence that interacting an education variable with the resource abundance variable, in a panel data setting, the interaction variable would offset the resource abundance variable whenever the average years of schooling for over 25 years old population would be around three. They show that in a 60 country sample, 2/3 of the countries would be free from resource curse.
Fourth, some argue that the problem may be on the structure of exports. They have emphasized the role played by concentration of exports in a few products (Lederman and Maloney 2008). Others have directed their attention to what they argue is the wrong specialization into activities that show little technological spillovers (ECLAC 2007) as the main causes of the curse. The problems associated with the concentration of exports have been treated before. Furtado (1957, 1974) deals with this issue when analyzing performance of the Venezuelan oil dependent economy. He argues that over dependence on one product exposes the country to periodical balance of payments crises due to price fluctuations. Furthermore, Furtado stresses that the productivity difference between the resource sector and the non-resource sector would cause wages and shrink the low productivity non-resource sector. This process would be even more radical when resources prices are high. This view anticipates the Dutch Disease mechanism (Medeiros 2007). Furtado argues that the generation of externalities from the natural resource exporting sector is low and, therefore, structural heterogeneity tends to persist. According to Furtado, the main reason for this structural problem was in the use of the rents from oil production that would go into the import of luxury consumption goods rather than into machinery, which also anticipates discussions on institutions. Implicitly, therefore, Furtado was arguing that there were little spillovers from production in the resource sector and little spillovers from the use of oil rents. The lack of externalities was also treated by Singer (1950:476) that stated that manufactures “provide the growing points for increased technical knowledge, urban education, the dynamism and resilience that goes with urban civilization”. Therefore, trade specialization in natural resources would provide low spillovers, compared to trade specialization in manufacturing. This position is coupled by ECLAC (2007) that argues that the striking difference between high and low growth countries is the share of what they call engineering
intensive sectors in total manufacturing. These sectors would play the role of technological diffusers to the rest of the economy. Furthermore, comparing the cases of Latin American countries with a group of successful resource rich economies – Australia, Canada, Finland, Ireland, Norway and New Zealand – they show that the latter group has a much higher share of engineering sector, a more similar industrial structure to what they classify as the leading economy, US, measured by the Krugman index, finally, they show that the share of exports in fast growing sectors in the latter group is much larger than in Latin America. However, when multivariate analysis is undertaken, they do not introduce a resource abundant variable, though the share of engineering sectors and the Krugman index show respectively positive and negative significant impact on catch-up. The issue about externalities (increasing returns) and natural resources has been dealt with by Dutch disease theorists. They have built models that consider learning by doing as a transmission mechanism, mostly associated with the tradable sector. van Winjberger (1984) builds a two-period model with a tradable and a non-tradable sector. The tradable sector is subject to learning by doing from production. Therefore, the level of production in the first period may affect the outcome of the second period. He shows that with a foreign currency premium that implies a real valuation of the country’s currency, the production of tradables in the first period will be smaller, generating negative effects in the second period, which damages welfare. He proposes therefore a subsidy or the use of the foreign currency premium to accumulate foreign assets. Sachs and Warner (1999) propose a model with similar mechanics but with learning spillovers from the tradable sector affecting the non-tradable sector as well. Torvik (2001) adds learning by doing in the non-tradable sector. All these papers depart from the premise that trade in manufacturing has spillovers through learning by doing and that there is little (less) learning in natural resource exporting sector.
Maloney (2008) argues that there is nothing special about manufactures and that technical change, education and externalities in natural resource sector are the same. Using data from Maddison database and the ratio of net natural resources exports to worker as natural resource abundance indicator, he finds that, in the 1820-1950 period, natural resource rich countries grew at faster rates than others (Auty 2000 makes the same point using descriptive statistics). For the post war period, he finds that the resource abundance variable lacks significance whenever variables indicating the degree of a country’s openness to trade and the level of the country’s technological capabilities are introduced. Therefore, he goes on to reverse the causality proposed by Singer, indicating that the lack of absorptive capacity and the reliance on import substitution industrialization were the main causes for the Latin America feeble performance in post-war years. As ECLAC (2007), he searches for empirical support in the successful resource abundant nations, as Australia, Canada, Sweden and the US. He argues that, previous to Second World War, these countries already had a well established education system that allowed for absorptive capacity, while in Latin American countries1 the education system was precarious which may have undermined absorptive capacity and learning. Maloney (2008) also puts that the establishment of property rights and free trade were distinguished factors of these countries’ cases and that those features may have allowed for the higher rates of growth showed by the former countries. David and Wright (1997) show evidence that the development of the US was based on natural resource industries that were capable to introduce technological innovations in their processes and generate spillovers for the rest of the economy. On the one side, this was a consequence of the country’s territorial expansion. On the other side, this was a result of well organized mineral research and of the introduction of technical innovations that made viable the processing and use of mineral reserves that previous technology did not allow. This success is
Some of which had the same GDP per capita as those successful nations in the late XIX century.
attributed to a well developed legal environment, infrastructure of public knowledge and education in mineral and metallurgy disciplines. Based on the same arguments, Wright and Czelusta (2008) defend that US resource abundant economy was not a gift from nature, but was a conquest of technical change. Natural resource reserves were expanded during the country’s development. In this case, they agree with Krautkraemer (1998) on the endogeneity of natural resources. One point that these economic historians stress is that technological improvements allowed for the development of the US mineral industries were able to use and supply technological knowledge and, most importantly, to expand mineral production. The correlation between exports and economic growth has been widely tested and its beneficial spillover effect has been stated in many studies, though some doubts seem to arise on the direction of causality (see Edwards 1998). Nonetheless, literature seems to have collected little evidence about the distinction made by Dutch Disease theorists that only the manufacturing tradable sector provides positive externalities through learning by doing. What Dutch Disease theorists seem to model is the occurrence of price effects that have influence on the exchange rate, they throw little attention on the expansion of production of natural resources. In fact, many resource abundant countries that have been stagnated in the last forty years seem to be characterized by the stagnation of their production of natural resource. Manzano and Rigobón (2008) show evidence that the world per capita production of fuel and mineral ores has decreased from 1978 to 1996,2 however, the per capita production of these fuel and mineral ores in resource abundant countries has shown higher rates of decrease. This may suggest that it is not low spillover from natural resource but slow growth in exports the main reason for the slow growth in resource abundant countries.
This should not hold for more recent years due to the China effect.
However, as stressed by Wright and Czelusta (2008), the highly successful resource abundant economies managed to expand their production/sales of natural resources through the introduction of technological innovations, whereas the most striking and known case of resource rich stagnant economies – the OPEC countries – have suffered from imposed production rationing to achieve price increases or prevent price decreases. These features raise the first question this paper intends to address: if it is true that the tradable sector provides positive externalities to the economy, then, do resource abundant countries perform as well as non-resource abundant economies whenever they have good export performance? The second question derives from the frequently hypothesized learning by doing effects of manufacturing. There are good reasons to think that trade provides positive externalities. Some are associated with learning by doing, as the Dutch Disease literature has stressed, others are associated with balance of payment effects such as those presented in the two-gap models (Chenery and Bruno 1962) and the export-led growth models (Thirlwall 1979, McCombie 1997). There are no reasons to suppose that the balance of payment externality effects of the growth of manufacture exports outperform those of the rate of growth of natural resource exports. However, though some have argued for positive effects of manufacturing on infrastructure, technological knowledge and the production chain, there are good reasons to suppose that these effects are also present in the natural resource sector. So, the paper attempts to address this question: do the externalities that arise from the growth of natural resource exports match those that arise from manufacture exports expansion?
EMPIRICAL STRATEGY Effects of Exports Dynamism on GDP per Capita Growth
In order to answer the first question, the paper departs from the basic cross-section equation tested by Sachs and Warner (1997), originated in the Solow growth model:
where y is per capita GDP, t is the final year, t-1, the initial year, T, the number of years from the initial to the final year, I/Y, the average investment to GDP ratio from the initial to the next to initial year, NR a natural resource abundance variable and Z is a vector of economic characteristics that determine the country’s growth. This paper will use the ratio of natural exports to GDP (sxp) as the natural resource abundance indicator. This indicator is chosen for two reasons. First, it has been the most tested and preferred measure of resource abundance in the literature. Second, and most importantly, it fits better in growth accounting modeling as will be shown. This paper will introduce an interaction variable between NR and a export dummy (EXPDUM), that assumes value one, whenever the country showed above median rates of export growth, and zero, otherwise. Sachs and Warner (1997) have used cross-section OLS estimation for equation (1). Crosssection regressions show two problems in this case. They are sensitive to omitted variables that may be associated with country specific characteristics and, furthermore, they do not account for endogeneity that may occur in the case of the sxp indicator. In order to deal with these problems, the paper uses panel data models (see Manzano and Rigobón 2008 for comments and justification). 3.2
In order to test for the spillover effects of natural resource and non-natural resource exports, this paper uses a model specification elaborated by Feder (1982). Feder uses a two-sector model that this paper extends to three sectors. Denote N for the non-exporting sector, NR for the natural resource exports sector and MX for the manufacture exports sector. (2) (3) 11
(4) And Feder departs from the same hypothesis used in most of the above surveyed literature that the exports sectors have higher productivity than the non-exports sector. Therefore, let the ratio of the marginal productivity of capital and labor in the two exporting sectors to the marginal productivity in the non-exports sector equal (1+δ):
Taking , where
(6) where I, stands for investment and ,
Suppose a linear relationship between the marginal product of labor and the average product per worker, such that:
Then, (10), (9), (8) and (7) into (6) and dividing by Y, we have:
Where the coefficients and
gather respectively the spillover effects,
and the productivity differential effects,
. The dots over the
variables mean rates of growth. In order to separate spillovers from productivity differential effects, one further hypothesis is introduced: as Feder (1982), it is necessary to suppose a constant elasticity function. Therefore, introducing a constant elasticity hypothesis of the effect of natural resource and manufacturing exports on N, from equation (2), we obtain (2’)
and substituting into (11):
where θ and μ are the externality parameters and off course,
the productive differential and xn, the share of natural resource exports in total exports. If or
there is no productivity differential.
In order to deal with the direct effect of the rate of growth of natural resource and manufacture exports, the paper addresses equations (11) and (12) in two different forms. First, equations are run against GDP growth as dependent variable. Then, the external trade sector is excluded from the dependent variable by subtracting from GDP the difference between exports and imports. 3.3
As has been stated above, this paper adopts the ratio of natural resources exports to GDP (sxp) as a measure of resource abundance. This variable has been built using the percentage of 13
natural resource exports and exports in US$ current value and GDP in US$ current value. All measures have been taken from World Development Indicators, 2009 (WDI). The catching-up variable is the initial period per capita GDP. The paper uses the chained per capita GDP in constant prices for 2005 variable from the Penn World Tables 6.3 (rgdpch). In order to cope with high variability of the data, the paper uses a five year average to obtain the variable, two years before and two years after the base year.3 Per capita GDP rates of growth have been obtained from the local currency constant prices GDP per capita of the WDI. The use of this measure avoids problems with transformation into PPP, but may affect results when compared to other papers. Anyway, the use of different versions of PWT 6.3 tends to show variability of data in similar dimensions. The investment to GDP ratio has been obtained from ten year average investment to gross GDP ratio of the PWT 6.3. Data for exports growth have been obtained from WDI in current prices (see Table 1 and Table 2). Data have been collected for four years: 1970, 1980, 1990 and 2000. The panels will have as reference three ten year growth periods: 1970-80, 1980-90, 1990-2000. Cross-section regressions will adopt 1970-2000 period. The database is composed of 74 countries that had data for all the years. The countries are listed in annex. In the case of the use of grgdpdom, only 61 countries are used in the estimation due to data shortages.
One of the first concerns of the paper was to match results with Sachs and Warner (1997). However, due to the need to find data for more recent years, new versions of the Penn World Tables had to be used. One empirical consequence of this search was the high variability of data according to the version of PWT. As has been put by Johnson et al. (2009) problems associated with the PPP measure make annual data change more than it would be desirable. The authors find that studies that use five year averages show greater robustness of their results.
Table 1 Variable gea
Variables Calculation and Data Source Method for Calculation
Rate of growth of per capita GDP, calculated by ) where, T is the number of years from t to t0. In this case, the paper uses the GDP per capita at constant local currency for 2005 for calculation of the gea lgdp Natural logarithm of five year average of the real GDP per capita in PPP at 2005 prices in the PWT chained series, the rgdpch variable Linv Natural logarithm of the average share of investments in GDP for the period considered in the gea variable, the ci variable from PWT xrn Percentage of natural resource exports in total exports. It is composed by the sum of fuel, mineral ore and metals, food and agricultural shares in total exports. grexp Export growth in current US$, expdum A dummy variable that assumes value one, whenever grexp is above its median value for the period, 0, otherwise sxp The natural resource exports to GDP ratio , where GDP is obtained in current US$ from WDI grgdp Rate of growth of GDP at local currency constant prices, 2005, calculated in the same way as gea grgdpdom Rate of growth of GDP of the domestic sector at local currency constant prices, 2005, obtained by GDP-(X-M). Exports (X) and imports (M) are measured at constant local currency, calculated in the same way as gea sxpexp Interaction variable: sxp*expdum grpop Rate of growth of the population grn Rate of growth of natural resource exports grm Rate of growth of manufacture exports special
WDI WDI WDI
WDI WDI WDI
Descriptive Statistics Variable gea lgdp linv xrn grexp expdum sxp grgdp sxpexp grpop grn grm special sxpgrn sxmgrm
Obs 222 222 222 222 222 222 222 222 222 222 222 222 222 222 222
Mean 1.817 8.812 3.031 60.527 0.097 0.505 0.136 3.511 0.065 1.694 0.079 0.122 -0.043 0.010 0.010
Std. Dev. 2.187 0.969 0.505 30.997 0.076 0.501 0.123 2.178 0.113 1.072 0.081 0.102 0.097 0.021 0.019
Min -5.983 6.430 1.209 2.600 -0.090 0.000 0.002 -2.269 0.000 -0.316 -0.099 -0.167 -0.462 -0.064 -0.021
Max 8.828 10.580 3.970 99.800 0.383 1.000 0.761 12.493 0.761 5.338 0.382 0.493 0.319 0.180 0.133
EMPIRICAL RESULTS Do Exports Matter?
Table 3 shows the results of eight regressions that have the rate of growth of GDP per capita as the dependent variable. Equations 1.1 to 1.4 present regressions that reproduce Sachs and Warner (1997) equations for the data set used in this paper. Equation (1.1) shows the results for the cross-section regression with rates of growth of GDP per capita from 1970 to 2000 as dependent variables. All variables have the expected sign and significance. However, the coefficient for the sxp variable is much smaller (in module) than the one found in Sachs and Warner (1997) equations. Three factors contribute for this difference. First, the samples are not the same. Second, the period is larger. Third, and most importantly, there are differences in the database as put in section 3.3. Equations (1.2) to (1.4) show the result for panel data settings. Variables maintain the expected sign and significance. However, the magnitudes of coefficients vary according to the
model. The coefficient for sxp assumes its lower value in the pooled data equation and its higher value in the fixed effects panel data model. According to this model, an increase of one standard deviation in the sxp variable implies a 1.1 percentage point decrease in the rate of growth of the GDP per capita. In the cross-section model, this impact is of only minus 0.28 percentage point.4 Equations (1.5) to (1.8) add an interaction variable formed by the product of sxp by the export dummy. When the interaction variable sxpexp is added, the coefficient sxp increases in absolute value and its level of significance also augments in all equations. However, the interaction variable assumes a positive and significant value. In all equations with the exception of the fixed effects model, the coefficient of the interaction variable is larger than the coefficient of the sxp variable, that is, a resource abundant country that manages to grow its exports at a fast rate does not suffer from the natural resource curse. In the fixed effects model, however, the interaction variable coefficient is in module a little lower than the coefficient of sxp, ending up in a net negative value even for natural resource economies that have grown their exports fastly. Yet, the overall negative effect of an increase of a standard deviation in sxp is reduced from 1.1 percentage point to 0.27 percentage point.5 This means that in a thirty year period, the resource rich economy would have grown its per capita income by less 8.4%, which is very far away from the disaster proposed by the results obtained by the natural resource curse literature. It could be that resource rich countries that show greater rates of growth of their exports are those that diversify towards manufacturing activities. Equations (1.9) to (1.11) attempt to address this question. It introduces a variable for the direction of specialization, formed by the difference between the rate of growth of natural resources exports and the rate of growth of
The F test for the significance of the intercepts and the Hausman test suggest the acceptance of the fixed effects model. 5 Again, the F test for ui=0 and the Hausman test suggest the acceptance of the fixed effects model.
manufacturing exports. The variable is insignificant and the coefficients and significance of the remaining variables are not affected by the introduction of this variable. Table 4 splits the sample into those economies that have shown low exports growth rates and those that show high exports growth rates for each ten year period. Equations (2.1) to (2.3) deal with the former case, while equations (2.4) to (2.6) deal with the latter. In all equations for the low exports growth case, sxp has a negative and significant effect on growth rates. The sxp coefficients in these equations are also higher than those presented in Table 3. Again, the highest value is in the fixed effects model, which has the best fit as well. In that model the coefficient for sxp is -12.57. This means that if we increase sxp by one standard deviation, the economy will grow less -1.55 percentage point per year, which would result an accumulated rate of growth of -63% in a thirty year period, which is clearly relevant. However, when we take the sample for fast exports growth economies, in equations (2.4) through (2.6), sxp does not have a statistically significant coefficient. Thus, the lack of dynamism of the tradable sector of resource abundant economies seems to affect their long run performance. Resource rich economies that are able to grow their exports at a fast rate do not suffer from the curse.
sxp lgdp linv
Regressions – Dependent Variable gea (growth in GDP per capita) Crosssection (1.1) -2.28*
Pooled (1.2) -3.35***
Fixed Effects (1.3) -8.93***
Random Effects (1.4) -4.91***
Crosssection (1.5) -5.13***
Fixed Effects (1.7) -9.77***
(-1.75) -0.37* (-1.94) 2.16*** (5.79)
(-3.15) -0.47*** (-2.83) 2.40*** (7.5)
(-3.6) -3.27*** (-6.97) 0.12*** (3.47)
(-3.82) -0.64*** (-3.35) 0.14*** (7.14)
(-3.49) -0.32* (-1.80) 1.88*** (5.25) 5.74*** (3.42)
(-6.96) -0.42*** (-2.76) 1.87*** (6.26) 9.43*** (7.09)
(-4.30) -2.98*** (-6.89) 0.08** (2.39) 7.61*** (5.44)
R-square 0.35 0.09 0.36 0.23 0.45 F 12.67 21.11 27.16 13.87 Wald Chi2 58.42 Obs 74 222 222 222 74 Groups 74 74 F test for 2.81 ui=0 Hausman 48.1 LM Breusch 7.98 and Pagan t-statistics in parenthesis. * 10% significance **5% significance ***1% significance
222 74 2.70
Random Pooled Fixed Effects (1.9) Effects (1.8) (1.10) -8.93*** -8.30*** -9.81*** (-6.81) (-6.93) (-4.19) -0.55*** -0.41*** -2.97*** (-3.09) (-2.74) (-6.83) 0.10*** 1.87*** 1.77** (5.56) (6.25) (2.49) 8.97*** 9.48*** 7.56*** (6.86) (7.09) (5.36) -0.55 -0.10 (-0.45) (-0.08) 4.93*** 0.30 23.47*** (3.43) (0.27) (5.21) 0.35 117.66 222 74
Random Effects (1.11) -8.60*** (-6.55) -0.63*** (-3.51) 2.11*** (6.03) 9.13*** (7.05) 0.04 (0.04) 1.56 (1.15) 0.36 124.63 222 74
2.66 40.9 8.87
Table 4 Sample sxp lgdp linv _cons
Regressions – Dependent Variable gea (growth in GDP per capita), samples low and high export growth Low Exports Growth Pooled Data Fixed Effects Random Effects -7.16*** -12.57*** -8.01*** (-4.78) (-2.99) (-4.88) -0.30 -4.29*** -0.48* (-1.38) (-4.28) (-1.95) 1.37*** 0.98 1.61*** (3.41) (0.88) (3.50) 0.54 37.24*** 1.52 (0.34) (3.65) (0.84)
R2 0.24 0.46 0.32 F 11.12 12.97 Wald Chi2 32.64 Obs 183 183 183 Groups 61 61 F test for ui=0 2.3 Hausman 22.4 LM Breusch and Pagan 2.33 t-statistics in parenthesis. * 10% significance **5% significance ***1% significance
High Exports Growth Pooled Data Fixed Effects Random Effects 0.32 -2.31 0.24 (0.25) (-0.67) (0.16) -0.52** -2.67*** -0.56** (-2.47) (-4.81) (-2.43) 2.46*** 2.17 2.32*** (5.19) (1.47) (4.52) -0.44 19.93 0.31 (-0.28) (3.25) (0.17) 0.2118 9.68 183
0.35 8.11 183 61 1.9
0.15 21.15 112 64 19.2 0.65
Table 5 and Table 6 tackle the problem of externalities testing equations (11) and (12) with two different dependent variables. In Table 5, the dependent variable is the rate of growth of GDP, while, in Table 6, the dependent variable is the rate of growth of the GDP with the exclusion of the trade sector, that is, subtracting from the GDP the difference between exports and imports. Equations (3.1) to (3.3) test the neoclassical growth equation and both the capital and labor variables have positive and significant signs. The F test for the fixed effects dummies is significant inducing to accept the pooled data regression as a better specification. The Hausman test also rejects the fixed effects model. Therefore, we should take the strong significance of the variables in the pooled data and random effects models. However, when we introduce the export sector variables formed by the interaction of the product of the natural resources exports to GDP ratio by the rate of growth of natural resources exports (sxpgrn) and by the product of the manufacture exports to GDP ratio by the rate of growth of manufacture exports (sxmgrm) in equations (3.4) to (3.6) both capital and labor variables lose significance. In the pooled data model, capital and labor still maintain significance, but in the fixed effects model only the labor variable is significant and in the random effects model, only the capital variable is significant.6 One possible explanation for the erratic behavior of the capital variable may be found in Bond and Malik (2010) that shows that a greater weight of fuel exports in the economy implies a higher level of investment. In all three models, sxpgrn is the most significant explicative variable and its coefficient is greater than the one for sxmgrm. This leads to the conclusion that large as
is at least as
, that is, the sum of the excess of productivity of the natural resource
It should be stated that the F test for the fixed effects dummies and the Hausman value suggest the acceptance of the fixed effects model.
trade sector over the domestic sector, trade sector over the domestic sector,
, and the externality effect of the natural resource , is at least as large as the same sum for the
manufacturing sector. Equations (3.7) to (3.9) in Table 5 attempt to separate the externality effect from the productivity effect. In these equations the coefficient for grn, θ, and grm, , represent respectively, the externality effect of the natural resource exports and the externality of manufacture exports over the domestic sector. In all three equations, the externality effect for the natural resource exports sector is significant at the 1% level, while the manufacture exports sector is significant at the 1% level only in the pooled and the random effects models. In the fixed effects, it only achieves a 10% level of significance.7 In all three equations, θ> . In the fixed effects model, an increase of one standard deviation in the rate of natural resource exports growth results is an increase of 0.48 percentage point in the rate of growth of the GDP due to the externality effect, while an increase in one standard deviation of the manufacturing sector, in the same model, provides an increase of 0.21% in the rate of growth of the GDP.
The coefficients ,
are also positive and significant in all three
equations. These coefficients are a measure of the impact of the higher relative productivity of each of the two external sectors to the domestic sector. In the fixed effects model, the coefficient for the natural resource exports sector is larger than the one for the manufacture exports sector. In the other two models the reversal is true. These results suggest that the two exports sector show greater productivity than the domestic sector. Table 6 reproduces equations (3.4) to (3.9) with the use of GDP minus trade surplus as the dependent variable. The results confirm the conclusions obtained from the equations in Table 5. Again the investment variable lacks significance in most equations. The labor variable is
Again the F and the Hausman tests suggest the acceptance of the fixed effects model.
positive and significant in all equations with exception of equation (4.5). Most importantly, the coefficients of the natural resource exports externalities are positive and significant in all equations. In equations (4.1), (4.2) and (4.3), larger than
is the most relevant variable and is
. In equations (4.4) to (4.6), the externality coefficient for natural
resource, θ, is positive and significant. The externality variable for manufacture exports is significant in equations (4.4) and (4.6), but lacks significance in equation (4.5). The results from these two tables seem to confirm the hypothesis made by the literature that the tradable sectors have higher productivity than the non-tradable sector. In the two sets of equations that analyze separately the productivity effect and the externality effect, the productivity effects of the natural resource and the manufacturing tradable sectors are positive and achieve at least 10% of significance.
Regressions – Dependent Variable grgdp Pooled (3.1) 0.09*** (5.55) 0.81*** (5.85)
Fixed (3.2) 0.13*** (2.98) 0.95*** (2.69)
Random (3.3) 0.09*** (5.26) 0.83*** (5.52)
Pooled (3.4) 0.78*** (2.81) 0.60*** (4.76) 42.03*** (7.33) 32.30*** (5.14)
Fixed (3.5) 1.05 (1.40) 0.74*** (2.65) 50.98*** (9.56) 45.06*** (3.09)
Random (3.6) 0.03** (2.06) 0.54*** (4.38) 42.60*** (7.32) 32.47*** (4.98)
sxpgrn sxmgrm grn (θ) grm (μ) _cons
R2 0.40 0.48 0.1728 0.1226 0.12 F 36.76 32.99 22.88 10.2 Wald Chi2 41.8 Obs 222 222 222 222 222 Groups 74 74 74 F test for ui=0 2.2 1.49 Hausman 1.6 LM Breusch and Pagan 3.94 t-statistics in parenthesis. * 10% significance **5% significance ***1% significance
0.48 161.01 222 74 21.53 15.8
Pooled (3.7) 0.57** (2.26) 0.53*** (4.52) 17.21*** (2.57) 27.53*** (4.73) 8.20*** (4.52) 3.68*** (3.09) -0.68 (-0.80)
Fixed (3.8) 0.65 (0.88) 0.39 (1.39) 30.33*** (4.06) 25.70* (1.74) 6.01*** (3.12) 2.10* (1.68) -0.41 (-0.19)
Random (3.9) 0.56** (2.01) 0.51*** (4.05) 21.16*** (3.28) 27.32*** (4.28) 7.52*** (4.35) 3.19*** (2.80) -0.52 (-0.56)
222 74 1.8
221.03 222 74 29.07 8.04
Linv grpop sxpgrn sxmgrm
Regressions – Dependent Variable grgdpdom Pooled Data (4.1) 0.60* (1.71) 0.49*** (2.81) 67.90*** (6.62) 38.57*** (4.13)
Fixed Effects (4.2) -0.12 (-0.11) 0.81* (1.76) 85.36*** (8.20) 44.25** (2.07)
Random Effects (4.3) 0.58 (1.57) 0.48*** (2.65) 71.32*** (7.19) 38.93*** (3.94)
grn (θ) grm (μ) _cons
R2 0.32 0.42 0.42 F 21.27 21.78 Wald Chi2 90.81 Obs 183 183 183 Groups 61 61 F test for ui=0 1.53 Hausman 27.4 LM Breusch and Pagan 3.12 t-statistics in parenthesis. * 10% significance **5% significance ***1% significance
Pooled Data (4.4) 0.35 (1.08) 0.39** (2.40) 31.05** (2.54) 25.29*** (2.81) 9.21*** (3.67) 4.23** (2.50) -0.23 (-0.21)
Fixed Effects (4.5) -0.36 (-0.34) 0.35 (0.74) 55.87*** (3.85) 21.72 (0.98) 7.22** (2.47) 1.43 (0.73) 2.29 (0.75)
Random Effects (4.6) 0.35 (1.08) 0.39** (2.40) 31.05** (2.54) 25.29*** (2.81) 9.21*** (3.67) 4.23** (2.50) -0.23 (-0.21)
183 61 1.21
129.57 183 61 27.2 0.8
The results suggest a revision of the natural resource curse assertions. It does not seem that the specialization in natural resources is negatively correlated to growth, but to the lack of growth of exports in resource dependent economies. In this case, as put by Wright and Czelusta (2008:208), “the danger of the resource-curse thesis is that countries may be discouraged from pursuing this reasonable and potentially fruitful avenue for economic progress”. The high level of externalities showed by natural resource exports in this paper recommends that countries that are able to use their natural resources should be encouraged to do so. The paper also argues for the risk of extending Dutch Disease model implications to the whole set of natural resource rich economies. The hypothesis of an exogenous foreign currency premium is associated with a price increase in existing exports (in a small open economy) and therefore is suitable when these situations are present. It should not be assumed as a plausible hypothesis in situations where natural resource production is increasing and, as shown in this paper, providing positive externalities to the economy. As a consequence of this line of argument, one should be wary about policies that aim to diminish periodical price decreases or reverse price decrease trends by the establishment of quotas or production restrictions. In most cases, these policies are more likely to generate market share losses in the long run and displace possible positive externalities that would be gained through productivity growth in the affected sectors. The arguments raised in this paper recommend caution in the use of dual economy with structural heterogeneity models. These models are based in the incapacity of the high productivity sector to generate positive spillovers to the low productivity sector, straightening the gap between them. The results from this paper do not invalidate structural heterogeneity models but suggest that if there is structural heterogeneity it is unlikely that its origin should 27
be associated with the type of trade specialization of the country. In this case, more general causes, such as the lack of absorptive capacity, should be pursued. The paper does not address some issues that may be important. First, it does not tackle the type of externalities generated by each exporting sector. Second, it does not deal with possible problems of production or exports growth that may arise from the specific specialization in natural resource. Third, the paper does not address problems associated with excessive specialization of the export basket. In the first case, Esfahani (1991) argues that exports affect balance of payment and therefore the capacity to import inputs for production and to acquire capital goods for other sectors. His results show that balance of payments type of externalities are quite important. It is unlikely that any balance of payment externality would be uneven depending on the sectors that generated the currency. Thus, in what matters for this paper, this type of externality could not be the origin of asymmetries. The other main externalities would be through capability accumulation generated by learning by doing type of effect or backward and forward linkages. This paper advances in this respect stating that if there are across sectors differences between these two types of externalities, they seem to be counterbalanced or are in favor of the natural resource sector. The second cavity of the paper questions whether the previously identified natural resource curse isn’t really the reflection of a stagnation of exports problem in an export dependent economy. Therefore, the real question to be answered would be the reasons why some natural resource rich countries grow their exports slowly and why some resource rich economies manage to grow their exports at a fast rate. Natural resource friendly literature conjectures that the main driver of expansion is technical progress (Maloney 2008, Wright and Czelusta 2008). In this case, learning and capability accumulation processes should be the most important drivers of success and failure and should affect indistinctly any sector. Problems in 28
the expansion of production front would be overcome by the introduction of technical progress, which is why they understand natural resources reserves as being intrinsically endogenous. However, some critics of natural resource specialization may argue that production or export expansion may be related to other situations. Low income elasticity and deterioration of the terms of trade may appear as one of the main obstacles, though Prebisch and Singer theses have failed to demonstrate their empirical pertinence (Cuddington et al. 2008). Other limitations may arise by production quota administration as the OPEC cartel does in the case of oil. The third cavity persists and, in fact, may be an important cause for poor performance of some of the slow export growth and GDP growth economies. Lederman and Maloney (2008) address this issue and find that the inclusion of a Herfindahl concentration index reduces the significance and the coefficient of the natural resource variable suggesting that the concentration of exports is the problem. It should be added that the proposition that slow exports growth is an important transmission mechanism does deny importance of other mechanisms. In fact, some of the mechanisms, such as institutional fragility, may help to explain why some countries have grown their exports at a slower rate.
This paper has shown that when one controls for export performance, only slow export growth resource rich economies grow significantly slower than the sample’s average. Thus, it argues that the problem with natural resource curse may be associated with slow export growth. Countries that are specialized in natural resources grow slower because their export base shows lower growth. Resource rich countries that overcome slow exports growth manage to engage in fast growth.
Literature has dedicated much attention to the role that would be played by asymmetric externalities that would arise from manufacture and natural resource industries as a transmission mechanism for the resource curse. This paper has raised evidence that this asymmetry is unlikely to occur. The results from the estimates presented suggest three important features: (i) natural resource export sector has higher productivity than the domestic sector; (ii) natural resource exports provide positive spillovers to the domestic sector; and (iii) its productivity and spillovers to the economy are at least as high as those provided by the manufacture exports sector. One can invert the order of presentation of the results and state that natural resources exports show positive externalities at least as high as those that arise from manufacture exports and, thus, resource rich countries that manage to grow their exports at a fast rate are able to grow as fast as non-resource rich economies.
AUTY, R. (2001) Introduction and Overview. In Auty, R. (Ed.) Resource Abundance and Economic Development, Oxford, Oxford University Press, 2001a.
BLOMSTRÖM, M. E KOKKO, A. (2008) From Natural Resources to High-Tech Production: The Evolution of Industrial Competitiveness in Sweden and Finland. In Lederman, and Maloney, W. F. (eds.) Natural Resources Neither Curse nor Destiny. Stanford University Press, and the World Bank.
BOND, S. and MALIK, A. (2009) Natural Resources, Export Structure, and Investment. Oxford Economic Papers 61, 675–702.
BRAVO-ORTEGA, C. and GREGORIO, J. (2008) The Relative Richness of The Poor? Natural Resources, Human Capital, And Economic Growth In Lederman, and Maloney, W. F. (eds.)
Natural Resources Neither Curse nor Destiny. Stanford University Press, and the World Bank. 5.
CHENERY, H. and BRUNO, M. (1962) Development Alternatives in an Open Economy: The Case of Israel. The Economic Journal. 72, 79-103.
CORDEN, W. AND NEARY (1982) Booming Sector and De-Industrialization in a Small Open Economy. The Economic Journal, Vol. 92, No. 368. (Dec., 1982), pp. 825-848.
CUDDINGTON, J., LUDEMA, R. and JAYASURIYA, S. (2007) The Prebisch-Singer Redux. In Lederman, and Maloney, W. F. (eds.) Natural Resources Neither Curse nor Destiny. Stanford University Press, and the World Bank.
DAVID, P.A., e WRIGHT, G. The Genesis of American Resource Abundance. Industrial and Corporate Change 6: 203–45, 1997.
ECLAC, (2007) “Progreso Técnico y Cambio Estructural en América Latina”. http://www.cepal.org.
10. EDWARDS, S. (1993) Openness, Trade Liberalization, and Growth in Developing Countries. Journal of Economic Literature. 31(3), 1358-1393. 11. ESFEHANI, H. (1991) Exports, Imports and Economic Growth in Semi-Industrialized Countries. Journal of Development Economics, 35(1), 93-116. 12. FEDER, G. (1982) On exports and economic growth. Journal of Development Economics, 12(1-2), 59-73. 13. FURTADO, C. (1957) O desenvolvimento recente da economia venezuelana. Reproduced in Arquivos Celso Furtado nº 1: subdesenvolvimento com abundância de divisas, Contraponto, Rio de Janeiro, 2007. 14. FURTADO, C. (1974) Notas sobre a economia venezuelana e suas perspectivas atuais. Reproduced in Arquivos Celso Furtado nº 1: subdesenvolvimento com abundância de divisas, Contraponto, Rio de Janeiro, 2007. 15. GYLFASON, T. (2001) “Natural Resources, Education, and Economic Development.” European Economic Review 45: 847–59. 16. GYLFASON, T., HERBERTSON, T., and ZOEGA, G. (1999), A Mixed Blessing: Natural Resources and Economic Growth, Macroeconomic Dynamics, 3, 204–25. 17. KRAUTKRAEMER, J. (1998) Nonrenewable resource scarcity. Journal of Economic Literature. Vol. XXXVI, pp. 2065-2107, December. 18. JOHNSON, S., LARSON, W., PAPAGEORGIOU, C., SUBRAMANIAN, A. (2009) Is newer better? Penn World Table revisions and their impact on growth estimates. NBER Working
Paper Series, Working Paper 15455, http://baselinescenario.files.wordpress.com/2009/10/jlps_oct-2009-final.pdf. 19. LARSEN, E. R. (2005) Are rich countries immune to the resource curse? Resources Policy, 30, 75-86. 20. LARSEN, E. R. (2006) Escaping the Resource Curse and the Dutch Disease? American Journal of Economics and Sociology, 65(3), 605-640. 21. LEDERMAN, D. e MALONEY, W. Trade Structure and Growth. In Lederman, and Maloney, W. F. (eds.) Natural Resources Neither Curse nor Destiny. Stanford University Press, and the World Bank, 2007. 22. MALONEY, W. F. (2008) Missed Opportunities: Innovation and Resource-Based Growth in Latin America. In Lederman, and Maloney, W. F. (eds.) Natural Resources Neither Curse nor Destiny. Stanford University Press, and the World Bank. 23. MANZANO, O. and RIGOBÓN, R. (2008). Resource Curse or Debt Overhang? In Lederman, and Maloney, W. F. (eds.) Natural Resources Neither Curse nor Destiny. Stanford University Press, and the World Bank. 24. MEDEIROS, C. (2007) Celso Furtado e o desenvolvimento a partir da exportação de recursos naturais não renováveis. In: Gregorio Vidal; Arturo Guillén R.. (Org.). Repensar la Teoría del Desarrollo en un contexto de Globalización. Buenos Aires: Clacso. 25. MEHLUM, H., MOENE, K. and TORVIK, R. (2006) Institutions and the Resource Curse. The Economic Journal, 116, January, 1-20. 26. PREBISCH, R. (1950) El Desarrollo Económico de América Latina e sus Principales Problemas, reprodução na Revista de Economia Política, VIII, 296-314, 1957.
27. SACHS, J, e WARNER, A. (1995) Economic Reform and the Process of Global Integration. Brookings Papers on Economic Activity, 25th Anniversary Issue, Washington, DC: Brookings Institution, 1–118. 28. SACHS, J, e WARNER, A. (1997) Natural Resource Abundance and Economic Growth. Center for International Development and Harvard Institute for International Development, Harvard University, Harvard, http://www.cid.harvard.edu/ciddata/warner_files/natresf5.pdf. 29. SACHS, J, e WARNER, A. (1999) The big push, natural resource booms and growth. Journal of Development Economics, 59, 43-76. 30. SACHS, J, e WARNER, A. (2001) “Natural Resources and Economic Development: The Curse of Natural Resources.” European Economic Review 45: 827–38. 31. SINGER, H. (1950) The Distribution of Gains between Investing and Borrowing Countries. The American Economic Review, 40(2), 473-485. 32. STIJNS, J. P. (2005) Natural resource abundance and economic growth revisited, Resources Policy 30: 107-130. 33. THIRLWALL, A. and HUSSEIN, M. (1982) The Balance of Payments Constraint, Capital Flows and Growth Rate Differences between Developing Countries. Oxford Economic Papers 34(3), 498-510. 34. TORVIK, R. (2001), Learning by Doing and the Dutch Disease. European Economic Review, 45, 285–306. 35. van Wijnbergen, S. (1984), The ‘Dutch Disease’: A Disease After All?, The Economic Journal, 94, 41–55. 36. WRIGHT, G. e CZELUSTA, J. Resource-Based Growth Past and Present. In Lederman, and Maloney, W. F. (eds.) Natural Resources Neither Curse nor Destiny. Stanford University Press, and the World Bank, 2007. 34
Country Sample Argentina Australia Austria Belgium Belize Bolivia Brazil Barbados Canada Switzerland Chile Cote d'Ivoire Cameroon Colombia Costa Rica Germany Denmark Algeria Ecuador Egypt, Arab Rep.
Spain Finland Fiji France United Kingdom Greece Guatemala Hong Kong, China Honduras Hungary Indonesia India Ireland Iceland Israel Italy Jamaica Japan Korea, Rep. Morocco
Madagascar Mexico Mali Malta Malawi Malaysia Nicaragua Netherlands Norway Nepal New Zealand Pakistan Panama Peru Philippines Papua New Guinea Portugal Paraguay Saudi Arabia Senegal
Singapore El Salvador Sweden Seychelles Syrian Arab Republic Togo Thailand Trinidad and Tobago Tunisia Turkey Uruguay United States Venezuela, RB South Africa