An alternative explanation for the resource curse: the income effect channel

Abstract

This article provides an alternative explanation for the ‘resource curse’ based on the income effect resulting from high government current spending in resource rich economies. Using a simple life cycle framework, we show that private investment in the nonresource sector is adversely affected if private agents expect extra government current spending financed through resource sector revenues in the future. This income channel of the resource curse is stronger for countries with lower degrees of openness and forward altruism. We empirically validate these findings by estimating nonhydrocarbon sector growth regressions using a panel of 25 oil-exporting countries over the period 1992 to 2005.

Keywords:

• resource curse
• fiscal policy
• investment and growth

JEL Classification::

• C1
• C82
• O11
• O41
• Q30

Acknowledgements

The authors would like to thank Enrica Detragiache, Mohsin Khan, Fuad Hasanov, Rolando Ossowski, Marc Quintyn, Piroska Nagy, Alejandro Santos, Janet Stotsky and Amadou Sy for helpful comments and discussions. All remaining errors are the authors’.

Notes

¹ Arezki and van der Ploeg (2010) provide evidence that the resource curse is less severe in countries with less restrictive trade policies and good institutions. However, they also show that empirical evidence on the resource curse is not robust to correcting for the endogenous nature of some regressors.

² Other explanations based on volatility of revenues from natural resources, and on resource rich countries’ excessive borrowing have also been put forward as a possible explanation for the resource curse (e.g. Hausmann and Rigobon, 2003).

³ A related but distinct literature emphasizes the nonproductive nature of certain categories of spending (e.g. Tanzi and Schuknecht (2000)).

⁴ Karnik and Fernandes (2009) provide evidence of a heavy dependence of the nonoil economy vis-à-vis the oil sector in the UAE using a macroeconometric model. The dependence of the oil economy in UAE relies on the existence of subsidies.

⁵ The correlation coefficient associated with Fig. 1 is −0.26.

⁶ The latter claim could be verified by rewriting the transition equation as follows:

The right-hand-side of this equation is predetermined at time t + 1. Therefore, noting that β − 1 is negative, an increase in g_{ot +1} can only lead to a decrease in k_{t +1}.

⁷ We assume that the old-age transfers are anticipated by the beneficiaries when young. Otherwise there would be no income effect at play.

⁸ This is also the case for the Dutch disease channel: the smaller the size of the nontradable sector is (i.e. the more open an economy is) the less prone it is to Dutch disease.

⁹ In the case of a small open economy without any capital mobility restrictions, there is no impact of natural resources redistribution on the dynamic of capital.

¹⁰ Changes in interest rates do not qualitatively change the results of this section because they are more moderate in the open economy than the closed economy, for which we have already offered a formal proof.

¹¹ Altonji and Williams (2005) present strong evidence against forward altruism using US data at the micro level. Nevertheless, this section is presented to justify the robustness and applicability of our model to countries where altruism may be supported by data.

¹² We estimate the reduced form of this model as follows: where

¹³ First order and second order serial correlation tests and Hansen test relating to the validity of over identifying moment conditions indicate that the estimated models presented in this article are correctly specified.

¹⁴ The lagged dependent specification used in this article allows us to limit the omitted variable bias. Also, the use of unit value of hydrocarbon export allows us to exploit a plausibly exogenous source of variation that is the unit value affects the nonhydrocarbon economy only through government expenditure.

¹⁵ To account for the presence of outliers, observations with excessively high leverage have been excluded from the sample. More precisely, all observations with DFBETA _{i , j} statistics, with i indicating the country and j the time period, that have an absolute value above a cutoff point equal to , with n being the number of observations in the original sample (Besley et al., 1980; Davidson and MacKinnon, 1993, pp. 32–9) were dropped. The results presented in this article are robust to the use of different values for the cutoff point above which observations are dropped. Table A3 of Appendix A presents the countries included in the study after the removal of outliers.

¹⁶ Survey data could be used to proxy the degree of forward altruism but such data is not available for many of the countries in our sample.