Debt overhang and economic growth the Asian and the Latin American experiences

Transcription

1 Economic Systems 31 (2007) Debt overhang and economic growth the Asian and the Latin American experiences Swapan Sen a, *, Krishna M. Kasibhatla b, David B. Stewart a a Winston-Salem State University, Winston-Salem, NC, United States b North Carolina A&T State University, Greensboro, NC, United States Received 13 March 2006; received in revised form 16 May 2006; accepted 28 May 2006 Abstract Debt overhang may impede a country s investment and growth. Accordingly, the World Bank initiated debt relief programs. However, doubts have been raised regarding the empirical validity of the debt overhang hypothesis. We employ panel data for testing the existence of debt overhang for Latin American and Asian borrowers. A variety of dynamic panel data econometric estimations as well as dynamic and system GMM estimations are conducted. Our results indicate that debt overhang impeded growth in Latin American economies severely and the impact was moderately negative in the Asian region. # 2007 Elsevier B.V. All rights reserved. JEL classification : P52; F34; G20; C33 Keywords: Debt overhang hypothesis; Asian and Latin American debt; Dynamic panel data; System and dynamic GMM 1. Introduction Debt overhang for a country exists when the country s debt service burden is so heavy that a large portion of the current output accrues to foreign lenders and consequently creates disincentive to invest (Krugman, 1988; Sachs, 1989). The hypothesis suggests that if there is some likelihood in the future that external debt will be larger than the country s repayment ability, then the expected debt-service costs would discourage further domestic and foreign investment and harm economic growth (Pattillo et al., 2002). Potential investors would be afraid that the more the country produces owing to additional investment, the more it would be taxed * Corresponding author. Tel.: ; fax: address: sens@wssu.edu (S. Sen) /$ see front matter # 2007 Elsevier B.V. All rights reserved. doi: /j.ecosys

2 4 S. Sen et al. / Economic Systems 31 (2007) 3 11 by external creditors in the form of debt service. Thus, investors would be less willing to incur costs today for the sake of increased output in the future as the additional output would be used to meet external debt service requirements. 1 High debt service burden increases expected future taxes on the private sector and lowers private investment. Resources that might have funded investments are consumed by debt servicing. In addition, debt overhang can worsen economic performance by changing the quality of investment if myopia in policy sets in when quick-yielding projects are favored over highervalued long-term investments. This and the uncertainty of debt service repayment create disincentives and difficulties to pursue economic reform (Clements et al., 2003). Debt overhang negatively impacts a country s economic growth through the adverse impact on investment and policy. Further, the negative impact of high debt on economic growth operates mainly through the negative effect on physical capital accumulation (Pattillo et al., 2003). That debt overhang adversely affects growth is generally acknowledged 2 and international programs of debt relief have been put in place to alleviate the problems of growth due to debt overhang. 3 Debt relief is expected to stimulate growth by releasing resources from debt service to investments in infrastructure and institutions. Such investments in turn enhance domestic investment as well as attract private foreign investment. The debt overhang hypothesis has, however, been questioned on both theoretical and empirical grounds. In the past, Bulow and Rogoff (1991) suggested that the borrowers underdevelopment was due more to their own economic mismanagement than to the burden of external debt and thus debt overhang was a symptom rather than a cause of low economic growth in the indebted countries. More recently, Arslanalp and Henry (2004) have claimed that the highly indebted poor countries do not suffer from debt overhang. Also, Cordella et al. (2005) are unable to rule out the possibility that debt does not matter. Whereas the negative implications of debt overhang for growth are abundantly realistic, the doubt raised regarding the empirical validity of the hypothesis is unsettling. If indeed debt overhang does not exist, debt relief is unlikely to stimulate growth. 4 It is therefore desirable to seek new evidence on the hypothesis. Thus, the primary objective of our paper is to empirically test the debt overhang hypothesis. We employ both static and dynamic panel data econometric models. The countries selected for this study are Argentina, Brazil, Colombia, Mexico, and Venezuela from the Latin American region and China, India, Indonesia, Philippines, South Korea, and Thailand from the South and East Asian region. The sample period is for the Latin American countries and for the Asian region. Section 2 presents the analytic models of estimation employed in the paper. It describes the econometric methodology as well as the variables included in the study. Section 3 presents evidence on the hypothesis. The results and concluding observations are summarized in Section 4. 1 The debt Laffer curve postulates that expected debt repayment first rises with the rise in the debt stock and then declines as debt increases. Debt overhang occurs at the peak of the debt Laffer curve. 2 Observers note that growth declined during the 1980s when debt was accumulating and growth accelerated during the 1990s when debt reduction occurred (Pattillo et al., 2002). 3 The World Bank initiated HIPC I and HIPC II, whereby highly indebted poor countries apply to obtain debt relief by fulfilling some criteria including poverty reduction targets. Recently, US$ 50 billion of African debt was forgiven. 4 Arslanalp and Henry (2004), for example, suggest that foreign aid will be more effective than external debt relief in stimulating these economies.

3 2. Dynamic panel data analytic models The debt overhang hypothesis is tested by using several different panel data models including pooled ordinary least squares (POLS), the so-called two-way fixed effects (FE) and two-way random effect (RE) models, two-stage least squares instrumental variable (IV) model, firstdifferenced GMM (DGMM) and system GMM (SGMM) models. Several reasons prompted us to employ all available models for testing the debt overhang hypothesis. First, the debate about the hypothesis appears to be an empirical one. Second, dynamic panel estimation methods differ in their assumptions regarding correlations, heterogeneity, and disturbance terms. They also differ in efficiency and bias. There are other problems, especially in growth empirics, such as endogeneity of the regressors, measurement errors, and omitted variable problems. Instead of selecting one or two of the models for conducting the hypothesis tests, we therefore considered it appropriate to employ all of the models. All of these estimators have been applied in growth research (Benhabib and Spiegel, 1997, 2000). In general terms, the equation to estimate the impact of external debt and debt service on economic growth of a panel of countries can be stated as a static linear model: DY it ¼ y it 1 þ a þ DX 0 it b þ e it; e it INð0; s 2 Þ (1) where Y is the dependent variable, a a scalar, X is k-vector of regressors, and the subscripts (i =1,..., N and t =1,..., T) identify the cross-section and the time dimensions; b is k 1 and X it is the it-th observation on k explanatory variable. Y represents the growth rate in per capita GDP and X includes commonly used debt burden indicators such as debt to GDP ratio, debt service ratio, and debt to export ratio; all in their log first differences. In addition, the regressors include gross capital formation and labor force or population also in log first differences. Eq. (1) does not make any distinction among the countries in the pool, either cross-section-wise (intercepts) or temporally. This model features constant coefficients, intercepts and slopes. Pooled OLS can be used to estimate the model under those assumptions Fixed effects model There may be characteristics that are unique to a country and these characteristics among the countries in the pool can be captured cross-sectionally by employing the least squares dummy variable (LSDV) estimator. The LSDV model (also called the fixed effects model) is given by DY it ¼ y it 1 þ DX 0 it b þ X a i d i þ e it (2) where d i is the dummy variable for the i-th country. The parameters a 1, a 2,..., a N and b can be estimated by ordinary least squares. The estimated results include a common intercept for the pool and intercepts for each country Random effects model The random effects model is given by S. Sen et al. / Economic Systems 31 (2007) DY it ¼ y it 1 þ a i þ DX 0 it b þ g t þ e it (3) where g t represents period specific effects and a i + e it is treated as an error term consisting of two components, an individual specific component that does not vary over time, and the other

4 6 S. Sen et al. / Economic Systems 31 (2007) 3 11 component that is assumed to be uncorrelated over time. Thus, the correlation of error terms over time is due to the individual effects, a i Dynamic generalized method of moments (D-GMM) The first differenced or dynamic GMM states the equation in first differences while instrumenting the right-hand side variables by using the levels of the series involved, lagged by two or more periods. Plausible results can be obtained employing GMM estimators as suggested by Arillano and Bover (1995). Consider an AR(1) process with unobserved country-specific effects model a la Blundell et al. (2000). Y it ¼ ay iðt 1Þ þ h i þ n it ; jaj < 1 (4) where i =1,..., N, and t =2,..., T, and h i + n it = u it has the standard error component structure. With E[h i ]=0,E[n it ]=0;E[n it h I ] = 0, for i =1,..., N, and t =2,..., T. Let us further assume that the transient errors are serially uncorrelated, i.e., E[n it,n is ] = 0 for i =1,..., N, and s 6¼ t, and that the initial conditions are predetermined E[y i1,n it ] = 0, for i =1,..., N and t =2,..., T. Together, the above assumptions imply 0.5(T 1)(T 2) moment restrictions. The per capita growth equation we wish to estimate has the following form: DY it ¼ g t þða 1Þy i;ðt 1Þ þ x 0 it b þ h i þ n it ; for i ¼ 1;...; N and t ¼ 2;...; T (5) where Dy it is the log first difference of per capita GDP (i.e., growth rate of per capita gdp), Y i(t 1) is the initial log gdp per capita, and x 0 it is a vector of regressors measured at the start of the period. The unobserved country-specific effects, h i, reflect differences in the initial conditions, and the period-specific intercepts, g t, capture certain changes, such as productivity changes, that are common to all countries. Country effects and time effects may also reflect country and periodspecific components of measurement errors System generalized method of moments (S-GMM) We need to make some additional assumptions to estimate Eq. (5) by system GMM. Blundell et al. (2000) consider a similar model without the time effects (g t ). Similar to the above basic AR(1) specification, in this case constant means of y it and x 0 it series through time for each country would be sufficient for the validity of the moment conditions E(h I, Dy it )=0 and E(h I, Dx 0 it )=0. This allows for the levels of x 0 it variables and Y it to be correlated with the unobserved countryspecific effects, but allows lagged first-differences of x 0 it and Y it to be used as instruments in the levels equation. Blundell and Bond (1998) consider that the S-GMM may have superior finite sample properties. Furthermore, Bond et al. (2001) add the following assumption to D-GMM: Eðh I DY i ; t 1Þ ¼0; for i ¼ 1;...; N and t ¼ 3; 4;...; T: This assumption requires stationarity restriction on the initial condition Y it. This condition will hold if the means of the Y it series, while different for each country across, are constant (or stationary) over time from periods 1, 2,..., T. Combined with the AR(1) model, presented above, this additional assumption will yield T 2 linear moment conditions which allow us to use the lagged first differences of the series as instruments for the equations stated in levels, as suggested by Arillano and Bover (1995).

5 Table 1 Panel estimation of Latin American debt and growth ( ) Variables PLS FE RE IV D-GMM S-GMM Lgdppct( 1) ( ) ** ( ) * ( ) * ( ) * ( 3.003) * ( ) * DLedtoGDP ( ) * ( ) * ( ) * ( ) * ( ) * ( 2,6053) * DLedtoexp (3.5071) * (3.6434) * (3.9152) * (3.6031) * (3.4856) * (2.1868) * DLdsr ( ) ( ) ( ) (0.1981) ( ) ( ) DLcap to GDP (5.2206) * (4.9778) * (5.9078) * (5.0393) * ) * (5.3593) * DLtlf ( ) (3.2688) * (3,1691) * (1.9081) (1.3677) (9.0807) * R DW statistic F-statistic [0.0000] [0.0000] [0.0000] [0.0000] J-statistic p-value (Sargan test) Countries: Argentina, Brazil, Colombia, Mexico, and Venezuela. Dependent variable: growth rate of per capita GDP in natural log. Regressors: ledtogdp: ln(external debt/gdp); ledtoexp: ln(external debt/exports of goods and services); ldsr: ln(debt service/exports of goods and services); lcaptogdp: ln(capital stock/gdp), ltlf: ln(total labor force). All regressors in log first differences. Dynamic panel data models estimated: PLS: panel least squares; FE: (panel least squares) fixed effects (cross-section two-way fixed effects); RE: (panel least squares) random effects (period two-way random effects); IV: (panel two-stage least squares), instrumental variables; D-GMM: first differenced generalized method of moments; S-GMM: system GMM; t-statistic in parentheses. * Significant at 5%. ** Significant at 10% [probability of F-statistic]. S. Sen et al. / Economic Systems 31 (2007)

6 8 S. Sen et al. / Economic Systems 31 (2007) Data and empirical results The dependent variable is the rate of growth of per capita GDP and the independent variables used in the estimation of Latin American countries are: LGDPPC: Log of GDP per capita (dependent variable). LEDGDP: Log of debt to GDP ratio. LDSR: Log of debt Service ratio to exports. LCAPGDP: Log of capital stock to GDP ratio. LTLF: Log of total labor force. LEDTOEXP: Log of external debt to export. For the Asian sample, data availability was different and we used population instead of labor force and gross national income instead of GDP. Data on capital stock was sketchy and the variable was omitted altogether, although we do not justify such omission. All variables are as defined and reported by the International Monetary Fund. The key question that we are interested in is whether a high level of external debt leads to slow-down in economic growth. Growth theory suggests that we should expect positive coefficients for all independent variables except the ratio of debt service to exports. In particular, if the debt to GDP ratio shows negative coefficient, we would conclude that debt overhang impeded growth. Table 1 summarizes results for the Latin American sample and the Asian experience is summarized in Table 2. Our basic result is that the coefficient of external debt to GDP is negative and statistically significant. The magnitude of the coefficient ranges from 2.10 (for D-GMM estimation) to 2.04 (for PLS estimation) indicating that the impact of high external debt on the growth rate of the economy ranges between a decrease of 2.04 and 2.10% for the five Latin American countries. Thus, on average, economic growth in these countries has been approximately 2% below what it would have been without the heavy external debt burden. This result provides strong evidence that a high level of external debt caused a significant slowdown of economic growth in the Latin American countries. The estimated results of all six specifications unanimously support the debt overhang hypothesis. Since the D-GMM estimate of DLedtogdp is slightly higher than the estimated values of PLS, FE, and RE for Latin American countries, the estimates may have been biased downwards due to weak instruments included. The debt overhang is true of the Asian countries as well, although to a much lesser extent. On average, the negative impact of debt on growth ranged from 0.06 to 0.13% across the models. The coefficient of external debt to GDP is negative and significant in all estimated equations. Thus, debt overhang reduced Asian growth, but not as severely as in the case of Latin America. In fact, the magnitude of decline of growth due to debt overhang in Asia is quite moderate. The results are, however, different when different debt burden indicators are used. The results do not support the debt overhang hypothesis as strongly when an external debt-to-exports ratio is used as the debt burden indicator. For instance, the coefficient of external debt to exports is positive and significant for Latin American countries. For the Asian countries, however, debt to export ratio is not significant. Similarly, the third debt burden indicator (debt service ratio) is not statistically significant either for Latin America or for Asia. Growth in capital formation contributed positively to economic growth in both regions. It was statistically significant in all models for Latin America, but for the Asian countries, it was significant for pooled least squares and fixed effect models only. Similarly, labor force impacted Latin American growth significantly (except for the PLS model). It was, however, not significant for the Asian countries.

7 Table 2 Panel estimation of Asian debt and growth ( ) Variables PLS FE RE IV D-GMM S-GMM Lgdppct( 1) ( ) ( ) ( ) * ( ) ( ) (0.8710) DLedtoGDP ( ) * ( ) * ( ) * ( ) * (.8135) ** ( ) ** DLdsr ( ) ( ) ( ) ( ) ( ) ( ) DLedtoexp (0.5709) (0.2562) (0.9139) (1.1254) (0.8776) ( ) DLcaptoGDP (7.4972) * (5.5600) * (1.0064) (1.1576) (1.4598) (1.2870) Dltlf ( ) (0.9769) (1.5821) (1.3167) (1.5186) (1.4399) R DW statistic F-statistic [0.0000] [0.0000] [0.0000] [0.0000] J-statistic p-value (Sargan test) Sample period: Countries: China, India, Indonesia, Philippines, South Korea, and Thailand. Dependent variable: growth rate of per capita GDP in natural log. Regressors: ledtogdp: ln(external debt/gdp); ledtoexp: ln(external debt/exports of goods and services); ldsr: ln(debt service/exports of goods and services); lcaptogdp: ln(capital stock/gdp), ltlf: ln(total labor force). All regressors in log first differences. Dynamic panel data models estimated: PLS: panel least squares; FE: (panel least squares) fixed effects (cross-section two-way fixed effects); RE: (panel least squares) random effects (period two-way random effects); IV: (panel two-stage least squares), instrumental variables; D-GMM: first differenced generalized method of moments; S-GMM: system GMM; t-statistic in parentheses. * Significant at 5%. ** Significant at 10% [probability of F-statistic]. S. Sen et al. / Economic Systems 31 (2007)

8 10 S. Sen et al. / Economic Systems 31 (2007) 3 11 Table 3 Comparative features of Asian and Latin American debt Debt stock (billion US$) Debt as % of GDP Debt stock (billion US$) Debt as % of GDP Asian countries India China Korea Malaysia Philippines Thailand Latin American countries Argentina Brazil Colombia Mexico Venezuela Total debt stock (US$ billion) and total debt as a % of GDP. Source: International Monetary Fund. The inclusion of both India and China, the two highly populous countries who have achieved significant growth in recent years, seems to have impacted the Asian results. The total debt stock (in billion US$) and total debt (as a percentage of gross national income) for 1990 and 2000 for the two regions are shown in Table 3 below. It appears that the two fastest-growing Asian economies, viz., India and China, have kept their debt level as well as their debt as a percentage of GDP moderate. Both India and China have succeeded in drastically lowering total debt as a percentage of export of goods and services, which has not occurred in the Latin American countries. Debt to export ratio for India was 3.31 in 1990 and 1.28 in 2000 and for China it was 0.91 in 1990 and 0.51 in Moderate improvement in the total debt to export ratio has been achieved by the oil-exporting Latin American countries such as Mexico and Venezuela, but not the other three Latin American economies. Our results are consistent with Bhattacharya and Clements (2004) who found that for 55 low-income countries over the period high levels of debt depressed economic growth after present value of debt reached percent of GDP. Similarly, Cordella et al. (2005) found a negative marginal relationship between debt and growth at intermediate levels of debt. 4. Summary and conclusions In view of recent works questioning the empirical validity of the debt overhang hypothesis, we undertook panel data econometrics estimations of the relationship between economic growth and the burden of external debt. By using methods of panel least squares (both fixed and random effects) and first differenced GMM and S-GMM for a sample of Latin American borrowers over , we found that debt severely impeded economic growth in these countries. Considering our various methodologies and choice of basic economic variables, the negative impact of debt stock on GDP growth seems robust. Similar estimation for the Asian borrowers over supports the debt overhang hypothesis, but the negative impact of debt on growth is much less severe. Although the two samples are not strictly comparable, the Asian

9 countries seem to be able to keep the debt level moderate while achieving substantial GDP growth. References S. Sen et al. / Economic Systems 31 (2007) Arillano, M., Bover, O., Another look at the instrumental variable estimation of error components model. J. Econom. 68, Arslanalp, S., Henry, P.B., Helping the poor to help themselves: debt relief or aid? In: Jochnik, C., Preston, F.A. (Eds.), Sovereign Debt at the Crossroads. Oxford University Press, New York, pp Benhabib, J., Spiegel, M.M., Cross-country growth regressions. C.V. Starr Center for Applied Economics research report 97 20, New York. Benhabib, J., Spiegel, M.M., The role of financial development in growth and investment. J. Econ. Growth 5, Bhattacharya, R., Clements, B., Calculating benefits of debt relief. Finance Dev. 41, Bulow, J., Rogoff, K., Sovereign debt repurchases: no cure for overhang. Q. J. Econ. 106, Blundell, R., Bond, S., Initial conditions and moment restrictions in dynamic panel data models. J. Econom. 87, Blundell, R., Bond, S., Windmeijer, F., Estimation in dynamic panel data models: improving on the performance of standard GMM estimation. IFS Working Paper W 00/12, London. Bond, S., Hoeffler, A., Temple, J., GMM estimation of empirical growth models, unpublished manuscript. Clements, B., Bhattacharya, R., Nguyen, T., External debt, public investment, and growth in low income countries. IMF working paper WP/03/249, Washington, DC. Cordella, T., Ricci, A. and Arranz R., Debt Overhang or Debt Irrelevance? Revisiting the Debt-Growth Link, WP/05/223, International Monetary Fund, Washington DC. Krugman, P., Financing versus forgiving a debt overhang. J. Dev. Econ. 29, Pattillo, C., Poirson, H., Ricci, L., External debt and growth. Finance Dev. 39, Pattillo, C., Poirson, H., Ricci, L., What are the Channels through which external debt affects growth? IMF Working Paper 04/15, Washington, DC. Sachs, J., The debt overhang of developing countries. In: Calvo, G.A., Findlay, R., Kouri, P., de Macedo, J.B. (Eds.), Debt Stabilization and Development. Blackwell, Oxford and Cambridge, MA, pp