Corruption, governance, investment and growth in emerging markets

Abstract

The article investigates the potential impact of corruption on economic growth by examining the effect that corruption may have on several significant determinants of economic growth, namely, investment in human, private and public capital, and on governance. Our theoretical approach allows for corruption to influence economic growth directly and indirectly through different investment and governance channels. All previous empirical work on this issue has been based on national income and product accounts (NIPA) data, which do not normally break down gross domestic investment into its private and public sector, and if they do, they misclassify investment by public enterprises as private investment, potentially biasing empirical findings. In this article we use a data set from the International Finance Corporation that bypasses these problems. We find that the impact of corruption on the level of public investment appears to be more ambiguous than it has been found in the previous literature. We, however, find that the impact of corruption on the accumulation of private capital is significantly more damaging than what has been previously found. We also find that the impact of corruption on governance is unambiguously negative, which further deters economic growth.

¹The views expressed in this paper are those of the authors and do not represent the official policy of the Naval Postgraduate School, the Overseas Private Investment Corporation, or the United States Government. The usual disclaimers apply.

Notes

¹The views expressed in this paper are those of the authors and do not represent the official policy of the Naval Postgraduate School, the Overseas Private Investment Corporation, or the United States Government. The usual disclaimers apply.

²A complete derivation of the theoretical model is available upon request.

³At this time, for theoretical simplicity, we assume that corruption and the set of exogenous variables are uncorrelated.

⁴The growth model specified in Equation 1 can be either a Solow-augmented neoclassical growth model with constant returns to scale for all production factors (α + β + ϕ + θ = 1), or an endogenous growth model with increasing returns to scale for all production factors (α + β + ϕ + θ ≥ 1). Also, if any combination of the capital inputs exhibits constant returns to scale (α + β = 1, β + ϕ = 1, α + ϕ = 1) then Equation 1 would similarly be characterized as an endogenous growth model. Senhadji (1999) noted that a large part of the empirical growth literature supports the assumption of decreasing returns to capital.

⁵While changes in resource endowments (the discovery of new resources or a cure for AIDS) may affect short-term capital-labour ratios, these changes would not necessarily affect the steady-state capital-labour ratio unless these changes influenced capital productivity. Gerson (1998) argues that since the convergence to the new steady state may take years to occur, fiscal policy can still lead to higher output growth rates for a significant period of time, even though the neoclassical model might imply that these policies would affect only the level of output and not its long-run growth rate.

⁶The effective unit of labour is the technology augmented unit of labour; see Islam (1995).

⁷For additional information on the International Country Risk Guide, see http://www.icrgonline.com

⁸We choose not to employ Transparency International's Corruption Perceptions Index due to the short length of the time series and the variability in the measurement methodologies over time. We also choose not to employ the World Bank's 2000 World Business Environment Survey (http://info.worldbank.org/governance/wbes) as we wished to investigate the evolution of corruption, investment and growth across time. We do note, however, that the ICRG index correlates highly with the Transparency International's Corruption Perceptions Index for the 1996–2002 periods.

⁹The precise ICRG definition of their measure is as follows: the institutional strength and quality of the bureaucracy is another shock absorber that tends to minimize revisions of policy when governments change. Therefore, high points are given to countries where the bureaucracy has the strength and expertise to govern without drastic changes in policy or interruptions in government services. In these low-risk countries, the bureaucracy tends to be somewhat autonomous from political pressure and to have an established mechanism for recruitment and training.

¹⁰For a detailed discussion of the investment data employed in this research, see Everhart and Sumlinski (2001).

¹¹See Appendix A for the countries included in the sample used for this study.

¹²Comparable data for developed countries are not available at this time. Future research could focus on developing similar measures for developed countries.

¹³See Appendix B for the variables included in the sample.

¹⁴The full set of estimation results is available upon request.

¹⁵We reject the null hypothesis of no serial correlation at the 1% significance level using a Durbin–Watson test for serial correlation. Re-specifying the model in first differences, we fail to reject the null hypothesis.

¹⁶We fail to reject the null hypothesis of exogeneity with a Hausman test statistic of 0.08 with 565 degrees of freedom.

¹⁷We employ different alternatives of conditioning variables to examine whether this result is robust and conclude that we fail to reject the null hypothesis with the given set of countries, time periods and explanatory variables. We also fail to reject the null hypothesis of exogeneity for corruption, the interactive term, and the conditioning variables, to include Current Account Balance as a percentage of GDP, Broad Moneyas a percentage of GDP and External Trade as a percentage of GDP.

¹⁸We reject the null hypothesis of no serial correlation at the 1% significance level using a Durbin–Watson test for serial correlation. Re-specifying the model in first differences, we fail to reject the null hypothesis of no serial correlation.

¹⁹We fail to reject with a Hausman test statistic of 2.07 with 678 degrees of freedom.

²⁰We employ different alternatives of conditioning variables to examine whether this result is robust and conclude that we fail to reject the null hypothesis with the given set of countries, time periods and explanatory variables. We also fail to reject the null hypothesis of exogeneity for corruption, the interactive term and the conditioning variables, to include Current Account Balance as a percentage of GDP, Broad Money as a percentage of GDP and External Trade as a percentage of GDP.

²¹A number of authors have used education as a proxy for human capital in growth regressions, and we investigated this route for our research as well. However, education data for such a broad panel of emerging market economies is not available. However, we merged the education dataset from Lee and Barro (2001) with our panel for the overlapping years and countries to investigate whether infant mortality and education were related. Using data for 35 of our countries for the years 1985, 1990 and 1995, we find a correlation of 0.71 between the two series, leading us to conclude that much of the informational content in an education proxy is also contained in our primary proxy for human capital, infant mortality.

²²We reject the null hypothesis of no serial correlation at the 1% significance level using a Durbin–Watson test for serial correlation. Re-specifying the model in first differences, we fail to reject the null hypothesis of no serial correlation.

²³We fail to reject with a test statistic of 0.2127 with 652 degrees of freedom.

²⁴We employ different alternatives of conditioning variables to examine whether this result is robust and conclude that we fail to reject the null hypothesis with the given set of countries, time periods and explanatory variables. We also fail to reject the null hypothesis of exogeneity for public health expenditures as percentage of GDP.

²⁵We reject the null hypothesis of no serial correlation at the 1% significance level using a Durbin–Watson test for serial correlation. Re-specifying the model in first differences, we fail to reject the null hypothesis of no serial correlation.

²⁶We fail to reject with a test statistic of 04875 with 684 degrees of freedom.

²⁷We employ different alternatives of conditioning variables to examine whether this result is robust and conclude that we fail to reject the null hypothesis with the given set of countries, time periods and explanatory variables.

²⁸We reject the null hypothesis of no serial correlation at the 1% significance level using a Durbin–Watson test for serial correlation. Re-specifying the model in first differences, we fail to reject the null hypothesis of no serial correlation.

²⁹We fail to reject with a test statistic of 1.541 with 680 degrees of freedom for the OLS model.

³⁰We employ different alternatives of conditioning variables to examine whether this result is robust and conclude that we fail to reject the null hypothesis with the given set of countries, time periods and explanatory variables.