Through What Channels Does External Debt Affect Growth?

Since the second half of the 1990s, policymakers and citizens around the world have been increasingly concerned that high external indebtedness in many developing countries is limiting growth and development. In Pattillo, Poirson, and Ricci (PPR), we find empirical support for a nonlinear impact of debt on growth. ¹ At low levels debt has positive effects on growth, but above particular thresholds or turning points additional debt begins to have a negative impact on growth. This type of analysis appears very relevant for current policy debates on assessing, for example, external debt sustainability in developing countries.

But there are many unanswered questions that relate to this issue. This paper's main question is what are the channels through which debt affects growth and whether nonlinear effects are also present in the relationship of debt with the different sources of growth. Specifically, we investigate whether debt affects growth mostly through an effect on factor accumulation or on total factor productivity growth, and whether these effects are nonlinear.

Both theory and policy discussions indicate that the effect of debt on growth could occur through all the main sources of growth. The capital accumulation channel in particular is supported by two arguments. First, the debt overhang concept implies that when external debt grows high investors lower their expectation of returns in anticipation of higher and progressively more distortionary taxes needed to repay debt, so that new domestic and foreign investment is discouraged. This in turn slows capital stock accumulation. Another strand of literature reaches similar conclusions by stressing that in heavily indebted countries investors hold back given the uncertainties about what portion of the debt will actually be serviced with the countries' own resources. Both arguments suggest that nonlinear effects of debt on growth are likely to occur through lower capital accumulation.

Other considerations imply that high debt levels may also constrain growth by lowering total factor productivity growth. For example, governments may be less willing to undertake difficult and costly policy reforms if it is perceived that the future benefit in terms of higher output will accrue partly to foreign creditors. The poorer policy environment in turn is likely to affect the efficiency of investment and productivity. In addition high levels of uncertainties and instabilities related to the debt overhang are likely to hinder incentives to improve technology or use resources efficiently. For example, as in other high uncertainty environments, investment may be misallocated to activities with quick returns, rather than long-term, higher risk irreversible investment, which would be more conducive to long-run productivity growth. Misallocated and less efficient investment projects could thus contribute to slower productivity growth.

Finally, debt relief advocates have argued that high debt severely constrains low-income countries' abilities to provide social services, such as education. The decision to acquire human capital is also an investment decision, which might be affected by the expectation of high marginal taxes. This would imply that high debt levels could lower growth by slowing human capital accumulation. This effect may be very difficult to detect, however, as it would only affect human capital stocks with long lags.

Our approach combines growth regressions with regressions on the sources of growth derived from a consistent growth accounting exercise. We augment a standard growth specification based on conditional convergence by adding several debt indicators. In addition we estimate the same models for the different components of growth—physical capital accumulation, human capital accumulation, and total factor productivity growth— similar to the approach of Fischer as well as Bosworth and Collins. ² We use a number of traditional estimators (ordinary least square [OLS], instrumental variables [IV], fixed effects) and more recent ones (differenced GMM, system GMM, and identification through heteroskedasticity [IH]) to control (to different extents) for biases associated with unobserved country-specific effects and the endogeneity of several regressors, particularly the debt...