Boon or burden? The effect of private sector debt on the risk of sovereign default in developing countries.
The past two decades witnessed a strong increase in private foreign borrowing in emerging markets and developing countries. Although in 1990 the private sector accounted for a mere 16% of all external loans disbursed to developing countries, this share has increased to 77% in 2006 (Figure 1A). Likewise, in 2006, the private sector's non-publicly guaranteed debt liabilities accounted for 44% of developing countries' total external debt--up from a mere 5% in 1990 (Figure 1B). (1) A key question in light of the ongoing global financial turmoil is what the boom in private sector external indebtedness will imply for sovereign risk in developing countries. In this paper, we empirically investigate the record of the past three decades to throw some light on this issue.
On theoretical grounds, there are arguments both in favor and against a stabilizing effect of private-sector borrowing on sovereign risk. A critical view of private-sector exposure is based on the notion that large-scale private borrowing creates vulnerabilities that may eventually lead to a sovereign default. A "sudden stop" may force the public sector to assume at least part of the private debt, and this may eventually cause debt-service difficulties for the government. Following this logic, both public and private external debt pose a threat to sovereign creditworthiness. The opposite argument that private-sector borrowing does not harm government creditworthiness can be made by invoking the idea that the private sector is exposed to greater competitive pressure, which raises the incentives to use the borrowed funds productively. More importantly, a potentially stabilizing role of private-sector borrowing can also be linked to the distributional consequences of sovereign defaults: Agents who are reliant on foreign credit are particularly vulnerable to the disruptions that come along with sovereign default. A larger share of the private sector in total external debt--a proxy of the relative size and stake of agents that would be hurt by sovereign default--would thus raise the political costs of default.
Given the competing theoretical arguments, the role of the private sector for sovereign creditworthiness is an empirical question. A first impression of how these magnitudes may be related is provided in Figure 2, which plots the Institutional Investor's measure of country creditworthiness (IICCR) against the level of external debt relative to GNI (Figure 2A) and the share of private long-term external debt in countries' total long-term external debt (Figure 2B). (2) Not surprisingly, the correlation between debt and creditworthiness is negative (-0.37). By contrast, the correlation between private-sector share and the IICCR is positive (0.54). (3) Although this picture obviously does not prove a causal relationship, it suggests that relative private-sector exposure and perceived creditworthiness might be positively related.
[FIGURE 1 OMITTED]
[FIGURE 2 OMITTED]
Further evidence that private and public debt are likely to have very different effects on the risk of sovereign default is provided in Table 1. The entries in this table are cross-country averages of various debt-related variables just before the onset of 5-yr periods in which a sovereign default did or did not take place. (4) The first column of Table 1 shows that initial debt relative to GNI is, on average, much higher before a default period than before a non-default period. Conversely, column 2 indicates that the average share of the private sector in total external debt is much higher before non-default periods than before default periods.
A first look at the data thus seems to point into the direction that a higher share of private debt in total external debt is associated with higher perceived creditworthiness--as reflected by the Institutional Investor's country credit ratings--and with a lower likelihood of sovereign default. The aim of this paper is to subject this hypothesis to closer scrutiny: Does the composition of external debt still matter if we account for other determinants of sovereign risk and the potential endogeneity of international borrowing and lending? Is this relationship driven by a particular group of countries or limited to a specific time interval? Our findings suggest that there is indeed a case to be made that a high share of the private sector in countries' external debt is more of a boon than a burden: an exogenous increase of this share reduces sovereign risk.
The rest of the paper is structured as follows: The next section offers a review of the relevant literature and highlights our own contribution. Section III introduces our empirical specification, the data we use, and comments on the results. Section IV summarizes and concludes. Detailed information on data definitions and sources are given in the data appendix.
II. REVIEW OFTHE LITERATURE
One reason why a larger share of the private sector in total external debt may improve a government's creditworthiness is that private sector debtors are particularly vulnerable to the frictions associated with sovereign default, and that the costs of default therefore increase in private-sector exposure. Although there is a rich literature on the causes and consequences of sovereign risk, (5) there are few studies that explicitly consider the distributional effects of debt crises and agents' conflicting interests with respect to sovereign default. Notable exceptions are Tomz (2002, 2004), Arteta and Hale (2008) as well as IMF (2002). (6) Tomz (2002) analyzes the shift in popular attitude that preceded the Argentine default of 2001. In particular, he documents how the sentiment of workers increasingly turned against compliance with international repayment obligations. Tomz (2004) presents the results of a survey which relates agents' attitude toward debt repayment to their professional and educational background. Not surprisingly, agents for whom access to international capital markets is important advocate debt repayment while public employees and individuals who are dependent on public welfare payments appreciate the relaxed budget constraint that comes along with sovereign default. The empirical findings of Arteta and Hale (2008) point into the same direction: A sovereign default substantially worsens firms' access to international credit markets and thus hurts those who are most reliant on foreign credit. IMF (2002) analyzes the distributional consequences of four recent default episodes. In some of these cases, default was associated with a sharp depreciation of the domestic currency. This depreciation "... eroded the balance sheets of banks, particularly those with significant open foreign exchange positions" (IMF 2002, 15). By contrast, "... others, particularly low-leveraged firms, reaped benefits from the depreciation" (IMF 2002, 16). These observations single out private agents with a large exposure to international capital markets as a group whose wealth and income is particularly affected by the government's default decision and suggests a distributional conflict between those individuals on the one hand and workers, non-leveraged firms, and public sector employees on the other hand. In Celasun and Harms (2007), we present a simple model that formalizes this notion by juxtaposing "workers" and "entrepreneurs": while workers are predominantly interested in low taxes and therefore support default, entrepreneurs who borrow abroad to finance their investments suffer a capital loss in case of default and thus advocate repayment. In this model, a larger "entrepreneurial class"--that is, a greater volume of private external borrowing--increases the political costs of default and therefore raises sovereign creditworthiness.
The alternative view that large private sector external debt is a source of financial risk for the public sector is based on the notion that private-sector overborrowing eventually harms sovereign creditworthiness through more or less explicit bailout guarantees. (7) In particular, Reinhart and Reinhart (2008) document that sovereign defaults in middle- and low-income countries in 1960-2007 were more often than not preceded by surges in overall capital inflows. They state that "on the basis of the historical track record, it is plausible to expect a higher chance of a sovereign default after a [capital flow] bonanza even in cases where government debt is not increasing. This is because the government sooner or later has usually ended up guaranteeing private sector debts." These arguments suggest that private borrowing is more of a burden than a boon to sovereign creditworthiness since private agents frequently succumb to overborrowing, and a high level of private-sector debt may threaten government solvency even in cases of healthy public finances. (8)
To determine which of the theoretical effects sketched above is dominant we estimate how the share of the private sector in total external debt affects countries' perceived creditworthiness and the likelihood of sovereign default. (9) Both the empirical studies on the determinants of sovereign defaults (Detragiache and Spilimbergo 2001; Manasse, Roubini, and Schimmelpfennig 2003; Manasse and Roubini 2005) and the literature on sovereign ratings (Cantor and Packer 1996; Haque et al. 1996; Harms and Rauber 2006; Mellios and Paget-Blanc 2006; Borio and Packer 2004; Afonso, Gomes, and Rother 2007) support the notion that high external debt is an important cause of debt crises. However, to the best of our knowledge, none of these contributions considers the potentially different impact of private versus public debt. (10)
III. EMPIRICAL ANALYSIS
A. Creditworthiness and Defaults
The goal of this paper is to investigate whether an increasing share of the private sector in external debt affects developing countries' creditworthiness and the likelihood of sovereign default. We proceed in two steps: in a first set of regressions, we estimate the impact of private sector exposure on a widely used indicator of creditworthiness, namely, the Institutional Investor's country credit rating. In a later section, we then estimate whether the share of private debt has an effect on the occurrence of actual defaults.
Our data set covers 65 developing countries and emerging markets for the years 1980-2005. The unit of time measurement we adopt is 5 yr, and the variables used in our regressions will be either 5-yr averages (1981-85, 1986-1990, ..., 2001-2005), or initial values preceding the respective 5-yr periods (1980, 1985, ..., 2000). We are interested in the following question: How does a change of the private-sector share in total external debt affect average creditworthiness and the likelihood of sovereign default in the subsequent 5 yr? Our choice of 5-yr averages is based on the notion that many of the theoretical mechanisms sketched above are likely to have a discernible effect on creditworthiness only at a low frequency. In addition, our dynamic structure has the virtues of simplicity and transparency: using annual data would require a more sophisticated dynamic specification and would possibly lead to coefficients that are difficult to interpret. Moreover, it would be much harder to address issues like unobserved heterogeneity and endogeneity.
Although focusing on actual defaults seems straightforward at first glance, it comes with a number of serious difficulties: First, there is no generally accepted definition of sovereign default. In our analysis, we rely on the definition of the rating agency Standard and Poor' s, which characterizes sovereign defaults as "... the failure to meet a principal or interest payment on the due date (or within the specified grace period) contained in the original terms of the debt issue" (Standard and Poor's 2006). Although this approach has the advantage of applying a straightforward and transparent criterion, it does not consider the size of arrears, nor does it capture those latent debt crises whose occurrence was prevented by foreign rescue operations and concessions. (11) A further problem with exclusively focusing on actual default episodes is that governments' creditworthiness frequently recovers while they are still negotiating the terms on which to repay existing arrears. During these periods, they are technically "in default," but the likelihood to deny repayment in the future may be much lower than suggested by their default status.
Therefore, as a first step, we use the IICCR which is likely to represent a more delicate and informative seismograph of investors' assessment whether current loans will be repaid in the future. The IICCR ranks countries on a scale from 0 to 100, with a lower rating reflecting a higher likelihood that borrowers in this country will default on their debt. The ratings are "... based on information provided by senior economists and sovereign risk analysts at leading global banks and money management and securities firms" (Institutional Investor 2002, 170) and have been published twice per year since 1979. (12) Although it does not exclusively refer to the likelihood of government default, we conjecture that sovereign risk makes up for a large share of "country creditworthiness." Our conjecture is confirmed by comparing the Institutional Investor's indicator to ratings which more explicitly focus on government creditworthiness, but cover a smaller number of countries and years. (13)
The performance of credit ratings in predicting financial crises has frequently been criticized in the recent past. However, Reinhart (2002) documents that ratings do a fairly good job in predicting sovereign defaults. This notion is confirmed by the numbers in Table 2, which gives the results of regressing the variable SOVDEFAULT on the Institutional Investor's measure of creditworthiness. SOVDEFAULT is a binary variable which is one if Standard and Poor's rated a government to be in default at least once during a 5-yr period (1981-85, 1986-90, etc.) and zero otherwise. [IICCR.sup.ini] is the value of IICCR in the year preceding that period (1980, 1985, etc.). The regression is based on a pro-bit model and includes both regional dummies and time dummies. The Institutional Investor credit rating has a significant relationship with the likelihood of default. Evaluated at the mean, raising [IICCR.sup.ini] by 1 point reduces the likelihood of default by about 1 percentage point. In terms of goodness of fit, the regression performs reasonably well: Approximately 81% of default episodes and 72% of the episodes without default are correctly predicted. Column (2.2) of Table 2 demonstrates that [IICCR.sup.ini] is still significant if we include the lagged value of SOVDEFAULT: Hence, it is a good predictor of future defaults even if we control for the possibility that past defaults both raise the probability of future defaults and reduce current creditworthiness.
B. Private Debt and Creditworthiness: Data and Model Specification
To investigate how a larger share of the private sector in total external debt affects perceived creditworthiness, we estimate variants of the following equation:
(1) [IICR.sup.av.sub.it] = [[beta].sub.1] [PRIVSHARE.sup.ini.sub.it] + [[beta].sub.2] [DEBT.sup.ini.sub.it]
+ [K.summation over (k=1)] [[gamma].sub.k] [x.sub.k, it] + [[xi].sub.t] + [[epsilon].sub.t]
where [IICR.sup.av.sub.it] is the Institutional Investor's average measure of country creditworthiness for country i in period t, and [PRIVSHARE.sup.ini.sub.it] is the initial percentage share of country i's long-term private external debt in its total long-term external debt. (14)
The variable [DEBT.sup.ini] is the initial level of external debt--short-term and long-term--relative to GNI. (15) Note that by using the values of PRIVSHARE and DEBT observed at the end of the previous 5-yr period we are reducing the potential for reverse causality, that is of creditworthiness affecting private and public borrowing.
Our choice of control variables [X.sub.k.it] largely follows the studies of Haque et al. (1996) as well as Harms and Rauber (2006). First, we use the lagged 5-yr average of the IICCR as a regressor ([IICCR.sup.av](-1)). A dynamic specification is suggested by Haque et al. (1996, 718) who find that "there is considerable persistence in the ratings, so that a country tends to retain its rating over time unless significant adverse or positive developments occur." Moreover, by controlling for lagged [IICCR.sup.av], we further reduce the potential endogeneity of the debt-variables: If a positive correlation between [PRIVSHARE.sup.ini] and [IICCR.sup.av] were only driven by the high persistence of credit ratings and the fact that [PRIVSHARE.sup.ini] reacts to ratings of the past, the correlation should disappear once lagged creditworthiness is explicitly taken into account.
A correlation between initial private sector debt and average creditworthiness could, of course, also reflect the expectation of more favorable economic and political conditions in the future: It is quite plausible that private sector borrowing expands more than proportionately in anticipation of a boom, and that such an upswing is also reflected by a rising measure of creditworthiness. To account for this possibility, we introduce two proxies for "economic prospects": the average growth rate of real per capita GDP in the preceding 5-yr period ([GROWTH.sup.av](-1)) and the average growth rate of the main trading partners' GDP ([TPGROWTH.sup.av]) in the current period. The advantage of using trading partners' growth is that this variable--while being significantly correlated with domestic growth--is unlikely to be endogenous with respect to [IICCR.sup.av]. (16) We also include the 5-yr average of an index of government stability ([GOVSTABILITY.sup.av]), compiled by the International Country Risk Guide, which captures the extent of political risk during a given time period, and which is likely to affect both creditworthiness and private borrowing.
To account for the possibility that the share of the private sector in total external debt merely reflects the level of economic development, we include the logarithm of real per capita income (in international dollars) at the end of the previous 5-yr period ([INCOMEPC.sup.ini]). Moreover, we use measures of financial and macroeconomic stability which are likely to affect both private borrowing and creditworthiness: the initial volume of reserves as a share of imports ([RESERVES.sup.ini]) and the log of the average inflation rate in the preceding 5-yr period ([INFLA.sup.av] (-1)). We also include the initial degree of trade openness ([OPEN.sup.ini]), measured as the ratio of exports and imports to GNI. Because more open economies are more vulnerable to the declines in foreign trade identified by Rose (2005) their willingness to default should be lower. At the same time, countries that are more open are likely to be more vulnerable to external shocks and may thus face a higher risk of default.
Finally, we use dummies for East Asia, Eastern Europe and Central Asia, South Asia, Latin America and Sub-Saharan Africa to account for regional differences, as well as time dummies [[xi].sub.t] to capture time-variant factors--changes in world interest rates or investor sentiment--that influence all countries' creditworthiness.
C. Private Debt and Creditworthiness: Results
Private Debt and Creditworthiness: Benchmark Regressions. Column (3.1) of Table 3 shows the results of estimating Equation (1) by ordinary least square (OLS). (17) All control variables have the expected sign. The coefficient on [DEBT.sup.ini] is highly significant, confirming the notion that a large level of external debt reduces creditworthiness. (18) Most importantly for our analysis, the share of private sector debt, as reflected by [PRIVSHARE.sup.ini], has a significantly positive coefficient: Although the coefficient is rather small in absolute terms--raising [PRIVSHARE.sup.ini] by one standard deviation increases [IICCR.sup.av] by 0.12 standard deviations--this result implies that, ceteris paribus, countries with a higher share of private debt in total external debt tend to have higher creditworthiness. (19)
Column (3.2) of Table 3 includes the second lag of [IICCR.sup.av] as an additional regressor: If the serial correlation of creditworthiness goes beyond one period and if private debt is slow to react to changes in credit rankings, omission of this variable could lead to biased estimates. However, this does not seem to be the case: Although the second lag of [IICCR.sup.av] has a significantly negative coefficient, the estimated coefficient of [PRIVSHARE.sup.ini] is almost unaffected.
The results so far suggest that a higher share of the private sector in total external debt is significantly associated with a higher level of creditworthiness. However, the significantly positive coefficients could just indicate that countries with a more developed financial sector have a larger share of private external debt and run a lower risk of sovereign default. To account for this possibility, we include the initial value of domestic credit to the private sector relative to GDP ([DOMCREDIT.sup.ini]) as a measure of financial depth. Column (3.3) demonstrates that, although this variable has a positive sign, it is not significant and its inclusion has almost no effect on the coefficient of [PRIVSHARE.sup.ini].
Column (3.4) of Table 3 reports the results of replacing [GOVSTABILITY.sup.av] with another measure of the investment climate. The variable [GOVERNANCE.sup.av] is also based on the International Country Risk Guide's assessments, but refers to different criteria--namely, the control of corruption, the quality of the bureaucracy, and the rule of law. Although this modification slightly lowers the size of the coefficient of [PRIVSHARE.sup.ini], the effect is not substantial, and the variable remains significant.
We also considered another disaggregation of external debt which possibly affects country creditworthiness and which might be correlated with private-sector exposure: Column (3.5) in Table 3 reports the result of including the variable [STDEBT.sup.ini], which reflects the share of short-term debt in total external debt. It turns out that this variable has no significant independent effect on perceived creditworthiness, and that its inclusion does not influence the coefficient of [PRIVSHARE.sup.ini]. (20)
Private Debt and Creditworthiness: Accounting for Unobserved Heterogeneity and Endogeneity. There is, of course, a high probability that the regressors we have included do not capture all sources of cross-country heterogeneity. The positive coefficient of [PRIVSHARE.sup.ini] may thus merely reflect the influence of other country-specific factors which affect both private borrowing and creditworthiness. Moreover, we have not yet come to terms with the possible endogeneity of [PRIVSHARE.sup.ini]: Although we have argued above that regressing the 5-yr average of [IICCR.sup.av] on the initial share of private debt in total external liabilities should reduce the potential for reverse causality--especially, when the lagged measure of creditworthiness is included--there are more sophisticated ways to deal with this issue.
In this subsection, we first follow the approach of Arellano and Bond (1991) and estimate the parameters of interest by differencing Equation (1) and by using lagged levels of the regressors as instruments. (21) The results in column (4.1) of Table 4 are based on a specification that uses up to four lags of the regressors as instruments. To avoid the overfitting that comes along with an excessive number of moment conditions and that results in biased estimates and uninformative diagnostic statistics, we impose the condition that the coefficients are uniform across the time periods in the first stage. (22) Moreover, we adopt a two-step approach that uses an optimal weighting matrix to aggregate the individual moment conditions. Standard errors are computed using the finite sample correction suggested by Windmeijer (2005). The p-values referring to Hansen's J-test of overidentifying restrictions (Hansen 1982) and to the (m2-) test of no second-order autocorrelation in the differenced residuals (Arellano and Bond 1991) are given at the bottom of the table. As the results in column (4.1) show, most coefficients change when we move from pooled OLS to the "Difference-GMM estimator" of Arellano and Bond (1991), suggesting that unobserved heterogeneity and endogeneity may indeed have influenced the results presented in Table 3. Nevertheless, our finding that the share of the private sector in total external debt raises countries' creditworthiness is strengthened rather than weakened.
Column (4.2) shows the results of applying the estimator developed by Arellano and Bover (1995) and Blundell and Bond (1998) to Equation (1). This approach simultaneously estimates the first-differenced version of the regression equation--using lagged levels of the righth-and-side variables--and the original equation in levels using lagged differences as instruments. The first advantage of this "Systems-GMM" estimator is that it exploits the information contained in the first period--a property that is of particular merit in our case where the number of periods is small. Moreover, it mitigates the weak-instruments problem that arises if the time series involved are very persistent. Column (4.2) in Table 4 demonstrates that, with this specification, the coefficient of [PRIVSHARE.sup.ini] is significantly positive.
So far, we have followed the standard approach of using lags of all regressors as instruments. Although the J-test gives no warning signs, our results may be biased if the right-hand side variables are endogenous. We are particularly concerned about [PRIVSHARE.sup.ini] and [DEBT.sup.ini] and therefore remove these variables from the list of instruments. Column (4.3) in Table 4 presents the results of following this approach when the "Difference-GMM" estimator is used, column (4.4) refers to the "Systems-GMM" estimator. In both cases the coefficient of [PRIVSHARE.sup.ini] remains significant.
The last columns of Table 4 present the results of treating the problems of unobserved heterogeneity and potential endogeneity of [PRIVSHARE.sup.ini] separately: First, we estimated our model with fixed effects. To account for the bias inherent in dynamic panel estimation (Nickell 1981), we applied the bias-correction suggested by Kiviet (1995) and Bruno (2005). (23) As indicated by column (4.5), using this "corrected LSDV (LSDVC)" estimator barely changes the coefficient of [PRIVSHARE.sup.ini] as compared to the pooled OLS results reported in Table 3. Finally, we accounted for the potential endogeneity of [PRIVSHARE.sup.ini] by instrumenting this variable with potential determinants of private-sector borrowing: the (lagged) quality of financial sector regulation ([CREDREG.sup.av](-1)), which is based on an index compiled by the Fraser Institute (Fraser Institute 2006), the distance from the equator ([LATITUDE.sup.av]) and the lagged average of the Freedom House index of civil liberties ([REPRESS.sup.av] (-1)) as proxies for the quality of governance (Freedom House 2006), the initial number of telephone main lines per 1,000 people ([TELEPHONES.sup.ini]) as a proxy for the quality of a country's infrastructure, as well as a lagged index of de jure exchange rate flexbility ([ERFLEX.sup.av] (-1)) adopted from Harms and Kretschmann (2009). The Kleibergen-Paap rk-statistic for underidentification in the presence of heteroskedastic disturbances and Hansen's J-test of overidentifying restrictions indicate that these instruments perform reasonably well in terms of relevance and exogeneity (see Baum, Schaffer, and Stillman 2007). Moreover, instrumenting for ([PRIVSHARE.sup.ini]) does not strongly affect its estimated coefficient. The notion that the coefficient of initial private sector debt is not driven by reverse causality is also supported by an explicit test for endogeneity which prevents us from rejecting the hypothesis that ([PRIVSHARE.sup.ini]) is exogenous. (24)
We conclude from the results presented in this subsection that accounting for unobserved heterogeneity and potential endogeneity has a bearing on the size of the estimated parameters--without, however, affecting our key finding that a higher share of the private sector in total external debt raises a country's creditworthiness.
Private Debt and Creditworthiness: Varying Samples and Alternative Specifications. This section reports the results of estimating Equation (1) using various subsets of the original sample and of further varying the original specification. It is apparent from Figure 2B that in a large number of countries, all external borrowing is performed by the government. We explored whether dropping the observations for which [PRIVSHARE.sup.ini] is zero changes the coefficient of this regressor. Column (5.1) of Table 5 demonstrates that it does not. We then restricted our attention to countries for whom IICCR exceeded the value of 25. Reinhart, Savastano, and Rogoff (2003) identity this value as a threshold below which countries do not really have access to international capital markets. Column (5.2) demonstrates that the size of the coefficient of [PRIVSHARE.sup.ini] is slightly reduced in this case, but the significantly positive estimated effect is not affected. Finally, we checked whether our result depended on the simultaneous decline of creditworthiness and private foreign borrowing observed during the 1980s. As column (5.3) of Table 5 reports, excluding the observations from the "lost decade" reduces the sample by almost one third, but barely affects the coefficient of [PRIVSHARE.sup.ini]. (25)
To check whether our results merely reflect the fact that a lower level of public debt is good for government creditworthiness, we replaced the regressor [DEBT.sup.ini] by the initial level of long-term public (and publicly guaranteed) debt relative to GNI ([PUBDEBT.sup.ini]). Although this variable has the expected negative effect, the coefficient of [PRIVSHARE.sup.ini] in column (5.4) indicates that the role of a larger private-sector share goes beyond just signaling the positive implications of lower government debt. In column (5.5), we explore the separate effects of long-term private, and public debt relative to GNI ([PRIVDEBT.sup.ini] and [PUBDEBT.sup.ini]). Although public debt has a significantly negative effect, the coefficient of [PRIVDEBT.sup.ini] is positive, but not significant. This could indicate that private debt per se is irrelevant for a government's creditworthiness. However, when comparing this result to our previous findings, we need to take into account that the specification in Equation (1) suggests a positive, but nonlinear influence of [PRIVDEBT.sup.ini] on [IICCR.sup.av]. Hence, imposing linearity may result in the large standard errors of column (5.5). Moreover, although explicitly accounting for endogeneity did not lead to very different results in column (4.6) of Table 4, the coefficient of [PRIVDEBT.sup.ini] increases considerably if we run a IV-regression using the set of instruments mentioned above (see column [5.6] of Table 5), and a formal test barely prevents us from rejecting the null-hypothesis that this regressor is exogenous. We conclude that a specification which considers private debt and public debt in isolation clearly rejects the notion that private-sector exposure hurts government creditworthiness. However, it does not really support the role of a stabilizing effect, either. Given the many channels through which private borrowing may influence sovereign creditworthiness this is not surprising. (26)
D. Private Debt and Sovereign Default: Results
Benchmark Results. So far, we have used the Institutional Investor's measure of creditworthiness as a dependent variable. Although we found strong evidence that the perceived likelihood of sovereign default is reduced by a larger share of the private sector in total external debt, this does not necessarily prove that governments' decisions on default versus repayment are actually affected by private sector exposure. To explore whether this is indeed the case we now use the dummy SOVDEFAULT as the dependent variable. As described in Section III.A., this variable is one if Standard and Poor's rated a government as being in default--that is failing to meet its repayment obligations--for at least 1 yr in a 5-yr interval, and zero otherwise. Except for the lagged indicator of creditworthiness, we are using the same set of covariates as in the previous subsections.
Our first regression uses the probit estimator to identify the determinants of sovereign defaults. The results are reported in column (6.1) of Table 6. With the exception of [INCOMPEC.sup.ini], the coefficients of the control variables have the expected sign--although only lagged growth and the initial debt to GNI ratio are statistically significant. [PRIVSHARE.sup.ini] clearly has a negative relationship with the likelihood of default. At the bottom of Table 6, we report the partial effect of this variable: evaluated at the sample mean, an increase of [PRIVSHARE.sup.ini] by roughly 1 percentage point ceteris paribus reduces the likelihood of default by 1 percentage point. As column (6.2) in Table 6 shows, using the logit estimator instead of probit yields a marginal effect of almost identical size. In terms of goodness of fit, both approaches do reasonably well: The pseudo-[R.sup.2] of McFadden (1974) is approximately 0.31 in both cases, and the percent correctly predicted is 75%. (27)
There might, however, be a problem with taking these results at face value: As the previous sections have indicated, sovereign creditworthihess is quite persistent--even if we focus on 5-yr averages. This is likely to apply a fortiori to actual defaults: after the initial denial of full repayment, it usually takes several years until an agreement with creditors is reached. During this period the country is rated as a defaulter, which, in turn, is likely to affect private borrowing (see Arteta and Hale ). Hence, the negative coefficient of [PRIVSHARE.sup.ini] may just capture the persistence of defaults, combined with the negative effect of past defaults on private external borrowing. Columns (6.3) and (6.4) in Table 6 indicate that this conjecture is at least partly correct: The coefficient of the first lag of SOVDEFAULT is significantly positive, and the influence of [PRIVSHARE.sup.ini] decreases once we include the first lag of the dependent variable. (28)
However, as the partial effects at the bottom of Table 6 reveal, this change is quite limited: Once we include the lagged dependent variable, it takes an increase of [PRIVSHARE.sup.ini] by approximately 1.3 percentage points to reduce the likelihood of default by 1 percentage point. The entries in the bottom rows indicate that including the lagged dependent variable improves the fit, but not by very much: the pseudo-[R.sup.2] moves from 0.31 to 0.36 while the percent correctly predicted increases from 75 to 81 and 82, respectively.
Accounting for Unobserved Heterogeneity. As in the previous subsections, we need to be concerned about unobserved heterogeneity: the likelihood of default may depend on country-specific characteristics which we have not explicitly accounted for in our regression equation. Although introducing country-specific effects is the straightforward solution to this problem in a linear regression model, things are a bit more complicated when it comes to discrete-choice models. The probit estimator, in particular, suffers from the incidental parameters problem--that is, it is not possible to consistently estimate the coefficients of the covariates using maximum likelihood without estimating the country-specific effects. This, in turn, fails if the number of time periods is finite. (29) There are several remedies to this problem: Under the assumption that the individual effects are not correlated with the covariates, the random effects probit estimator yields consistent estimates (Wooldridge 2002a). Column (7.1) in Table 7 shows the results of adopting this approach. Column (7.2) reports the coefficients we obtain when estimating the random effects model using logit: interestingly, the partial effects for [PRIVSHARE.sup.ini] (-0.014 and -0.015) do not stray too far from the values we received from the pooled regression in Table 6. (30) Column (7.3) reports the results from applying the fixed effects logit estimator. If the distribution of the underlying error term is assumed to be logistic, consistent estimation of the relevant parameters is possible even if unobserved heterogeneity is treated by means of fixed effects. However, this advantage comes at a cost: Because the fixed effects cannot be estimated, it is impossible to compute partial effects. Hence, the magnitude of the negative coefficient of [PRIVSHARE.sup.ini] in column (7.3) cannot be readily compared with our previous results. To better understand the role of unobserved heterogeneity, we finally specified the model as a simple linear equation including fixed effects. (31) Interestingly, the coefficient of [PRIVSHARE.sup.ini] in column (7.4) is very close to the partial effects we reported in Table 6.
The last two columns in Table 7 return to the issue of whether the significantly negative effect of [PRIVSHARE.sup.ini] merely reflects the persistence of sovereign defaults. To explore this issue, we include the lagged value of SOVDEFAULT as an additional regressor, adopting two alternative approaches: We first follow Wooldridge (2002b) who suggests to estimate a random effects (logit or probit) model conditioning on initial observations. The results of this strategy are presented in column (7.5) of Table 7: Although the lagged dependent variable is significant, the coefficient of [PRIVSHARE.sup.ini] is still significantly negative, and the estimated marginal effect barely differs from the one displayed at the bottom of column (7.1). However, this estimator is biased unless the other regressors are strictly exogenous. Hence, as a robustness check, we also estimated the linear probability model (LPM) with a lagged dependent variable and fixed effects. As column (7.6) demonstrates, the lagged dependent variable has no impact on the other coefficients in this case. Of course, given the caveats with respect to this estimator, we do not want to overrate this result. The main finding we take away from these estimations is that the significantly negative effect of [PRIVSHARE.sup.ini] on sovereign defaults does not seem to be an artifact of neglecting unobserved heterogeneity and the persistence of defaults.
IV. SUMMARY AND CONCLUSIONS
Although external debt figures among the usual suspects when it comes to explaining sovereign risk, little attention has been devoted to the potentially different effects of private and public external debt. The main contribution of our paper is to emphasize that this difference is substantial: in the time period covered by our sample, a higher share of the private sector in total external debt raised country creditworthiness and reduced the likelihood of default for a given stock of external debt. Our results thus offer one potential explanation for the observation of Reinhart, Savastano, and Rogoff (2003) that countries with similar levels of external debt may exhibit vast differences in their creditworthiness and their propensity to default: "debt intolerance" may more heavily afflict countries where the government accounts for most of the external borrowing.
A battery of robustness tests indicate that our results are not driven by unobserved heterogeneity--that is, creditworthiness and private sector exposure being driven by country-specific unobserved parameters--and that they do not just reflect the reverse impact of creditworthiness on private borrowing. Our findings suggest that, for a given country, an exogenous increase in the private-sector share of foreign debt improves a government's creditworthiness. An explanation as to why countries exhibit such large differences in the composition of their external debt goes beyond the scope of this paper, but offers a promising subject for further research.
Note, finally, that our analysis focused on the factors that determine the likelihood of sovereign default rather than on financial crises in general. The historical record suggests that, in the past, the stabilizing effect of private-sector debt on government creditworthiness at least compensated (if not dominated) the destabilizing effect. Whether these forces will shape governments' decisions in the crisis that started in 2007 is an open question that future research will be able to answer.
IICCR: Institutional Investor's Measure of Country Creditworthiness
LPM: Linear Probability Model
OLS: Ordinary Least Square
A. Definitions and Sources
[CREDREG.sup.av]: Five-year average of the Fraser Institute's index of credit market regulation, ranging from 0 (minimal regulation) to 10 (maximal regulation). Source: Fraser Institute (2006).
[DEBT.sup.ini]: Total external debt relative to GNI (in percent). Sources: World Bank (2007a) and World Bank (2007b).
[DOMCREDIT.sup.ini]: Domestic credit to private sector relative to GDP (in percent). Source: World Bank (2007a).
[ERFLEX.sup.av](-1): Average index of de jure exchange rate flexibility in the preceding 5-yr period. The index is based on the IMF's Annual Report on Exchange Arrangements and Exchange Restrictions and ranges from 1 to 4, where 1 represents a peg, 2 represents limited flexibility, 3 represents a managed float, and 4 represents a pure float. Source: Harms and Kretschmann (2009).
[GOVERNANCE.sup.av]: Simple average of indices measuring bureaucratic quality, corruption, and the rule of law, from the International Country Risk Guide. Source: Political Risk Services (2007).
[GOVSTABILITY.sup.av]: Index of government stability from the International Country Risk Guide. Source: Political Risk Services (2007).
[GROWTH.sup.av](-1): Average growth rate of real per capita income in the preceding 5-yr period. Source: Penn World Table 6.2 (Heston, Summers, and Aten ).
[IICCR.sup.av]: Five-year average of the country credit ratings published in the Institutional Investor magazine every March and September since 1980. Source: Institutional Investor magazine.
[INCOMEPC.sup.ini]: Log of initial value of real per capita GDP in constant PPP-adjusted dollars. Source: World Bank (2007a).
[INFLA.sup.av](-1): Log of the average growth rate of the consumer price index in the preceding 5-yr period. Source: World Bank (2007a).
[OPEN.sup.ini]: Initial value of the ratio (exports + imports)/ GNI (in percent). Source: World Bank (2007a).
LATITUDE: Squared latitude. Source: World Bank (2001).
[PRIVDEBT.sup.ini]: Initial private nonguaranteed long-term external debt relative to GNI (in percent). "Private nonguaranteed debt outstanding and disbursed (LDOD) is an external obligation of a private debtor that is not guaranteed for repayment by a public entity. Long-term debt outstanding and disbursed (LDOD) is the total outstanding long-term debt at year end. Long-term external debt is defined as debt that has an original or extended maturity of more than one year and that is owed to nonresidents and repayable in foreign currency, goods, or services." Source: World Bank (2007b).
[PRIVSHARE.sup.ini]: Initial share of private nonguaranteed long-term external debt in total long-term external debt (in percent). Source: World Bank (2007b).
[PUBDEBT.sup.ini]: Initial public and publicly guaranteed long-term external debt relative to GNI (in percent). Source: World Bank (2007b).
[REPRESS.sup.av]: Five-year average of the Freedom House index of civil liberties, ranging from 1 (maximal rights) to 7 (minimal rights). Source: Freedom House (2006).
[RESERVES.sup.ini]: Initial value of the ratio (international reserves)/(imports of goods and services) in percent. Source: World Bank (2007a).
SOVDEFAULT: Dummy variable which equals one if Standard and Poor's rates a government to be in default at least once during a 5-yr period and zero otherwise. A default is characterized by "the failure to meet a principal or interest payment on the due date (or within the specified grace period) contained in the original terms of the debt issue." Source: Standard and Poor's (2007).
[STDEBT.sup.ini]: Ratio of short-term debt to total external debt (in percent). "Short-term external debt is defined as debt that has an original maturity of one year or less. Available data permit no distinction between public and private nonguaranteed short-term debt." Source: World Bank (2007b).
[TELEPHONES.sup.ini]: Number of telephone main lines per 1,000 people. Source: World Bank (2007a).
[TPGROWTH.sup.av]: Five-year average of the growth rate of a weighted average of trading partners' GDP. Sources: World Bank (2007a) and IMF (2006).
Algeria, Argentina, Bangladesh, Benin, Bolivia, Botswana, Brazil, Bulgaria, Cameroon, Chile, China, Colombia, Congo Rep., Costa Rica, Cote d'Ivoire, Democratic Republic of Congo, Dominican Republic, Ecuador, Egypt Arab Rep., El Salvador, Estonia, Gabon, Ghana, Haiti, Honduras, Hungary, India, Indonesia, Jamaica, Jordan, Kenya, Latvia, Lithuania, Malawi, Malaysia, Mali, Mauritius, Mexico, Morocco, Nicaragua, Nigeria, Oman, Pakistan, Panama, Papua New Guinea, Paraguay, Peru, Philippines, Poland, Romania, Russian Federation, Senegal, Sierra Leone, South Africa, Sri Lanka, Syrian Arab Republic, Tanzania, Thailand, Togo, Trinidad and Tobago, Tunisia, Turkey, Uganda, Ukraine, Uruguay, Venezuela RB, Zambia, Zimbabwe.
Afonso, A., P. Gomes, and P. Rother. "What 'Hides' Behind Sovereign Debt Ratings?" ECB Working Paper No. 711, 2007.
Arellano, M., and S. Bond. "Some Tests of Specification for Panel Data: Monte Carlo Evidence and an Application to Employment Equations." Review of Economic Studies, 58, 1991, 277-97.
Arellano, M., and O. Bover. "Another Look at the Instrumental-Variable Estimation of Error-Components Models." Journal of Econometrics, 68, 1995, 29-52.
Arteta, C., and G. Hale. "Sovereign Debt Crises and Credit to the Private Sector." Journal of International Economics, 74, 2008, 53-69.
Baum, C.F., M.E. Schaffer, M.E., and S. Stillman. "Instrumental Variables and GMM: Estimation and Testing." The Stata Journal, 3, 2003, 1-31.
--. "Enhanced Routines for Instrumental Variables/GMM Estimation and Testing." Boston College Economics Working Paper No. 667, 2007.
Beim, D. O., and C. Calomiris. Emerging Financial Markets. New York: McGraw Hill, 2001.
Blundell, R., and S.R. Bond. "Initial Conditions and Moment Restrictions in Dynamic Panel Data Models." Journal of Econometrics, 87, 1998, 115-43.
Bond, S. "Dynamic Panel Data Models. A Guide to Micro Data Methods and Practice." CEMMAP Working Paper No. CWP09/02, 2002.
Borensztein, E., and U. Panizza. "The Costs of Sovereign Default." IMF Working Paper 08/238, 2007.
Borio, C., and F. Packer. "Assessing New Perspectives on Country Risk." BIS Quarterly Review, 2004, 47-65.
Bruno, G. S. F. "Approximating the Bias of the LSDV Estimator for Dynamic Unbalanced Panel Data Models." Economics Letters, 87, 2005, 361-66.
Bulow, J., and K. Rogoff. "A Constant Recontracting Model of Sovereign Debt." Journal of Political Economy, 97, 1989, 155-78.
Cantor, R., and F. Packer. "Determinants and Impact of Sovereign Credit Ratings." FRBNY Economic Policy Review, 2(2), 1996, 37-54.
Celasun, O., and P. Harms. "How Does Private Foreign Borrowing Affect the Risk of Sovereign Default in Developing Countries?" Study Center Gerzensee Working Paper 07.04, 2007.
Chang, R. "Electoral Uncertainty and the Volatility of International Capital Flows." NBER Working Paper 12448, 2006.
--. "Financial Crises and Political Crises." Journal of Monetary Economics, 54, 2007, 2409-20.
Cole, H., and P. Kehoe. "Reviving Reputation Models of International Debt." Federal Reserve Bank of Minneapolis Quarterly Review, 21(1), 1997, 21-30.
Corsetti, G., P. Pesenti, and N. Roubini. "Paper Tigers: A Model of the Asian Crisis." NBER Working Papers No. 6783, 1998.
Detragiache, E., and A. Spilimbergo. "Crises and Liquidity: Evidence and Interpretation." IMF Working Paper 01/2, 2001.
Eaton, J., and R. Fernandez. "Sovereign Debt," in Handbook of International Economics, edited by G. Grossman, and K. Rogoff, III. Amsterdam: North-Holland, 1995, 2059-77.
Eaton, J., and M. Gersovitz. "Debt with Potential Repudiation: Theory and Estimation." Review of Economic Studies, 48, 1981, 289-309.
Frankel, J., and A. Rose. "Currency Crashes in Emerging Markets: An Empirical Treatment." Journal of International Economics, 3/4, 1996, 351-66.
Fraser Institute. Economic Freedom in the World. Vancouver: Fraser Institute, 2006.
Freedom House. Freedom in the World. Washington, DC: Freedom House, 2006.
Greene, W. "Estimating Econometric Models with Fixed Effects." New York University, 2001.
Hansen, L. P. "Large Sample Properties of Generalized Method of Moments Estimators." Econometrica, 50, 1982, 1029-54.
Haque U. N., M. Kumar, N. Mark, and D. Mathieson. "The Economic Content of Indicators of Developing Country Creditworthiness." IMF-Staff Papers, 43, 1996, 688-724.
Harms, P., and M. Kretschmann. "Words, Deeds, and Outcomes: A Survey on the Growth Effects of Exchange Rate Regimes." Journal of Economic Surveys, 23, 2009, 139-64.
Harms, P., and M. Rauber. "Foreign Aid and Developing Countries' Creditworthiness." Study Center Gerzensee Working Paper No. 04.05, 2006.
Heston, A., R. Summers, and B. Aten. Penn World Table Version 6.2, Center for International Comparisons of Production, Income and Prices at the University of Pennsylvania, 2006. Accessed 8 March 2010. http://pwt.econ.upenn.edu/php_site/pwt_index.php
Institutional Investor "Credit Where It's Due?", Institutional Investor, 36(9), 2002, 167-72.
International Monetary Fund (IMF). Sovereign Debt Restructurings and the Domestic Economy Experience in Four Recent Cases. Washington DC: International Monetary Fund, 2002.
--. Direction of Trade Statistics Database. Washington DC: International Monetary Fund, 2006.
Jeske, K. "Private International Debt with Risk of Repudiation." Journal of Political Economy, 114, 2006, 576-93.
Kiviet, J. F. "On Bias, Inconsistency, and Efficiency of Various Estimators in Dynamic Panel Data Models." Journal of Econometrics, 68, 1995, 53-78.
Lancaster, T. "The Incidental Parameter Problem since 1948." Journal of Econometrics, 95, 2000, 391-413.
Larrain, F., and A. Velasco. "Can Swaps Solve the Debt Crisis? Lessons from the Chilean Experience." Princeton Studies in International Finance, 69, 1990.
Manasse, P., and N. Roubini. "Rules of Thumb for Sovereign Debt Crises." IMF Working Paper No. 05/42, 2005.
Manasse, P., N. Roubini, and A. Schimmelpfennig. "Predicting Sovereign Debt Crises." IMF Working Paper No. 03/221, 2003.
McFadden, D. "Conditional Logit Analysis of Qualitative Choice Behavior," in Frontiers of Econometrics, edited by P. Zarembka. New York: Academic Press, 1974, 105-42.
Mellios, C., and E. Paget-Blanc. "Which Factors Determine Sovereign Credit Ratings?" The European Journal of Finance, 12, 2006, 361-77.
Nickell, S. "Biases in Dynamic Models with Fixed Effects." Econometrica, 49, 1981, 1417-26.
Political Risk Services. International Country Risk Guide. New York: Political Risk Services, 2007.
Reinhart, C. M. "Credit Ratings, Default, and Financial Crises: Evidence from Emerging Markets." World Bank Economic Review, 16, 2002, 151-70.
Reinhart, C. M., and V. R. Reinhart. "Capital Flow Bonanzas: An Encompassing View of the Past and Present." CEPR Discussion Paper 6996, 2008.
Reinhart, C. M., M. Savastano, and K. Rogoff. "Debt Intolerance." Brookings Papers on Economic Activity, 2003, 1-62.
Roodman, D. "How to Do xtabond2: An Introduction to 'Difference" and 'System' GMM in Stata." Working Paper 103, Center for Global Development, Washington, 2006.
Rose, A. "One Reason Countries Pay Their Debts: Renegotiation and International Trade." Journal of Development Economics, 77, 2005, 189-206.
Standard and Poor's. Sovereign Credit Ratings--A Primer. New York: McGraw Hill, 2006.
--. Sovereign Ratings History Since 1975. McGraw Hill, 2007.
Sturzenegger, F., and J. Zettelmeyer. Debt Defaults and Lessons from a Decade of Crises. Cambridge, MA: MIT Press, 2006.
Tomz, M. "Democratic Default: Domestic Audiences and Compliance with International Agreements." Working Paper, Stanford University, 2002.
--. "Voter Sophistication and Domestic Preferences Regarding Debt Default." Working Paper, Stanford University, 2004.
Windmeijer, F. "A Finite-sample Correction for the Variance of Linear Efficient Two-Step GMM Estimators." Journal of Econometrics, 126, 2005, 25-51.
Wooldridge, J. M. Econometric Analysis of Cross Section and Panel Data. Cambridge, MA: MIT Press, 2002a.
--. "Simple Solutions to the Initial Conditions Problem in Dynamic, Nonlinear Panel Data Models with Unobserved Heterogeneity." CEMMAP Working Paper No. CWPI 8/02, 2002b.
World Bank. Global Development Network Growth Database. Washington DC: The World Bank, 2001.
--. World Development Indicators 2007 on CD-ROM. Washington DC: The World Bank, 2007a.
--. Global Development Finance 2007 on CD-ROM. Washington DC: The World Bank, 2007b.
(1.) Figure 1 and the following figures refer to long-term debt, which comprises instruments that have an original or extended maturity of more than 1 yr (World Bank 2007b). Comprehensive data on private versus public external debt and borrowing is available only for long-term debt instruments.
(2.) For the IICCR, the data points refer to 5-yr averages between 1980 and 2005, while the debt-variables are measured at the beginning of these 5-yr periods.
(3.) In both cases, extreme observations seem to play a prominent role. If we remove the top 5% of the debt variable observations--limiting our attention to countries with debt-to-GNI-ratios below 200% and a share of the private sector below 40%, respectively--the correlation in Figure 2A becomes -0.43, while the correlation in Figure 2B decreases to 0.46.
(4.) Here and in what follows we adopt the definition of the rating agency Standard and Poor's which identifies a sovereign default as the "... failure to meet a principal or interest payment on the due date (or within the specified grace period) contained in the original terms of the debt issue" (Standard and Poor's 2006).
(5.) See, for example, Eaton and Gersovitz (1981), Bulow and Rogoff (1989), Cole and Kehoe (1997), Manasse, Roubini, and Scbimmelpfennig (2003), Rose (2005), Borensztein and Panizza (2007). Eaton and Fernandez (1995) as well as Sturzenegger and Zettelmeyer (2006) offer excellent surveys of this discussion.
(6.) Chang (2006) analyzes the interdependence between capital flows and distributional conflict, but focuses on income taxation. Chang (2007) analyzes the political economy of default from a representative-agent perspective.
(7.) See, for example, Larrain and Velasco (1990) for an account of private debt nationalizations and the Chilean external debt restructuring during the 1982 crisis.
(8.) Corsetti, Pesenti, and Roubini (1998) identify excessive foreign borrowing by the private sector as one of the key causes of the Asian currency crises of 1997-1998. Indonesia, one of the countries hit hardest by the crisis, restructured its foreign currency bank debt in 1998-99 and was thus classified by Standards and Poor's as being in sovereign default status at that time.
(9.) Jeske (2006) also explores the different roles of private and public debtors; however, he does not consider potential interactions between the two components of external debt.
(10.) A notable exception is Frankel and Rose (1996) who explore inter alia how the share of the public sector in total external debt affects the occurrence of currency crises. Interestingly, a higher share of the government raises the likelihood of a currency crash in the subsequent year. By contrast, the effect of the public sector-share on currency crises in the same period is not significant.
(11.) Given these considerations, Manasse, Roubini, and Schimmelpfennig (2003) augment the Standard and Poor's data with information on concessional IMF loans, whereas Beim and Calomiris (2001) differentiate between outright repudiation and minor, pre-announced defaults. Rose (2005) identifies sovereign defaults with the onset of Paris-Club negotiations, Arteta and Hale (2008) combine information about renegotiations with those of Standard and Poor's and the Economist Intelligence Unit.
(12.) As reported by Haque et al. (1996), the individual criteria used by banks to assess default risk are not specified.
(13.) The IICCR has been widely used in empirical work on sovereign creditworthiness, given its coverage of a large number of countries and years. The rank correlation between the IICCR and the sovereign ratings published by Moody's in the 1990s is 0.92. The rank-correlation with the sovereign ratings of Fitch ratings is 0.85.
(14.) A detailed description of data definitions and sources is provided in the data appendix. The variable [PRIVSHARE.sup.ini] is based on long-term debt because the public versus private sector decomposition is not available for short-term debt. As we will show later, the focus on long-term debt does not appear to be consequential for our empirical results.
(15.) For the time being, we do not distinguish between different sources of loans. That is, [DEBT.sup.ini] comprises both loans from official sources and loans from private investors. As we will show later, this aggregation is not crucial for our results.
(16.) Including the growth rate of domestic output did not alter the qualitative results we report in subsequent sections.
(17.) The standard errors presented in square brackets are based on a covariance matrix that is robust with respect to heteroskedasticity and serial correlation of cluster-specific disturbances.
(18.) This result is because of dropping the observation for Nicaragua in 1990, which is characterized by an excessively high level of external debt (1087 percent of GNI). Including this data point substantially increases the standard error of [DEBT.sup.ini] without, however, changing the qualitative results with respect to the other regressors.
(19.) This result does not hinge on our indiscriminate treatment of private and official creditors. If we constrain our attention to debt owed to private agents and institutions-thus netting out official loans--we obtain a somewhat smaller, but significantly positive coefficient.
(20.) To further explore whether the maturity structure of external debt was important for our results we ran our benchmark regression under the two alternative assumptions that all short-term external debt was either private or public. It turned out that the modified values of [PRIVSHARE.sup.ini] still had a significantly positive impact on [IICCR.sup.av].
(21.) Bond (2002) and Wooldridge (2002a) offer excellent surveys on dynamic panel estimation. We used the xtabond2 module by Roodman (2006) to implement the difference-GMM estimator.
(22.) We do this by using the collapse option of the xtabond2 routine in Stata.
(23.) To compute these results, we used the xtlsdvc routine developed for Stata by Giovanni Bruno.
(24.) To test for endogeneity, we used both the WuDurbin-Hausman-test and the Difference-in-Sargan test implemented in Stata by Baum, Schaffer, and Stillman (2003) and Baum. Schaffer, and Stillman (2007).
(25.) Using the Difference-GMM estimator of Arellano and Bond (1991) and dropping those observations for which [PRIVSHARE.sup.ini] was zero or those in which creditworthiness did not exceed a minimum threshold did not alter our key result--nor did restricting attention to the years after 1990. These results are available upon request.
(26.) When we replaced external public debt ([PUBDEBT.sup.ini]) by total government debt (relative to GNI), the latter variable had no significant effect. The same result occured when we used both external and total public debt as regressors. These results are available upon request.
OYA CELASUN and PHILIPP HARMS *
* We are indebted to two anonymous referees for helpful comments. The views expressed in this paper are those of the authors and do not necessarily represent those of the International Monetary Fund, its Board of Executive Directors, or the governments the latter represent. Celasun: International Monetary Fund (IMF)--Research Department, 700 19th Street NW, Washington, DC 20431. Phone (202) 623 4274, Fax (202) 589 4274, E-mail OCelasun@imf.org
Harms: RWTH Aachen University, Faculty of Business and Economics, Templergraben 64/III, 52062 Aachen, Germany. Phone +49(0)241 8096203, Fax +49(0)241 8092649, E-mail email@example.com
TABLE 1 Debt Variables before Defaults Initial External Initial Prov. Ext. Debt/GNI Debt/Ext. Debt Default in t 108.51 5.02 No default in t 55.33 10.59 Source: World Bank (2007b) and Standard and Poor's (2007). TABLE 2 Institutional Investor's Country Credit Rating (IICCR) and the Likelihood of Sovereign Default (2.1) Probit (2.2) Probit [IICCR.sup.ini] -0.032 *** -0.017 ** [0.007] [0.008] East Asia and Pacific 0.623 0.483 [0.439] [0.454] South Asia -0.325 0.048 [0.678] [0.675] Europe and Central Asia 0.368 0.265 [0.388] [0.405] Sub-Saharan Africa 0.917 *** 0.874 ** [0.328] [0.344] Latin American and the 0.899 *** 0.766 ** Caribbean [0.314] [0.330] OIL 0.437 0.352 [0.270] [0.281] 1986-90 -0.086 -0.243 [0.302] [0.319] 1991-95 -0.546 * -0.903 *** [0.301] [0.324] 1996-00 -1.039 *** -1.363 *** [0.286] [0.306] 2001-05 -1.463 *** -1.684 *** [0.290] [0.306] SOVDEFR ULT(-1) 1.032 [0.2211 Constant 0.953 ** 0.245 [0.4591 [0.506] Marginal effect of IICCR -0.013 *** -0.007 ** [0.003] [0.0031 Observations 281 281 Pseudo [R.sup.2] 0.23 0.29 Percent correctly 0.77 0.76 predicted Notes: Standard errors in parentheses are based on a robust covariance matrix. ***, **, * denote significance levels of 1%, 5%, and 10%. The data sample is an unbalanced panel, comprising 5-yr averages or initial values between 1980 and 2005. The dependent variable is SOVDEFAULT, which is a binary variable indicating if Standard and Poor's rated a government as being in default at least once during a 5-yr period. TABLE 3 The Effect of PRIVSHARE on IICCR-Pooled OLS (3.1) OLS (3.2) OLS [IICCR.sup.av](-1) 0.553 *** 0.771 *** [0.040] [0.061] [PRIVSHARE.sup.ini] 0.115 *** 0.123 *** [0.030] [0.030] [DEBT.sup.ini] -0.022 *** -0.019 *** [0.007] [0.007] [TPGROWTH.sup.av] 2.695 * 1.544 [1.370] [1.356] [GROWTH.sup.av](-1) 0.713 *** 0.595 *** [0.114] [0.152] [GOVSTABILITY.sup.av] 1.253 *** 1.477 *** [0.312] [0.322] [INCOMEPC.sup.ini] 1.553 2.343 ** [1.044] [1.059] [RESERVES.sup.ini] 0.042 ** 0.057 ** [0.021] [0.024] [INFLA.sup.av] (-1) -0.365 -0.079 [0.452] [0.558] [OPEN.sup.ini] 0.027 ** 0.036 ** [0.013] [0.017] East Asia and Pacific -0.404 -0.269 [1.524] [1.777] South Asia 0.392 1.453 [1.978] [2.083] Europe and Central Asia 4.568 ** 3.655 * [1.793] [1.849] Sub-Saharan Africa -1.791 -1.07 [1.456] [1.539] Latin America and the Caribbean -2.273 ** -1.744 [1.119] [1.080] OIL -0.919 -0.553 [1.345] [1.831] [IICCR.sup.av](-2) -0.243 [0.053] [DOMCREDIT.sup.ini] [GOVERNANCE.sup.av] [STDEBT.sup.ini] Constant -17.490 ** -18.257 ** [8.585] [8.041] R-Squared (adj.) 0.88 0.89 Number of observations 257 207 (3.3) OLS (3.4) OLS [IICCR.sup.av](-1) 0.534 *** 0.526 *** [0.044] [0.047] [PRIVSHARE.sup.ini] 0.115 *** 0.093 *** [0.030] [0.029] [DEBT.sup.ini] -0.023 *** -0.026 *** [0.007] [0.007] [TPGROWTH.sup.av] 2.667 * 2.829 ** [1.398] [1.409] [GROWTH.sup.av](-1) 0.718 *** 0.705 *** [0.115] [0.119] [GOVSTABILITY.sup.av] 1.237 *** [0.310] [INCOMEPC.sup.ini] 1.372 1.391 [0.969] [1.110] [RESERVES.sup.ini] 0.041 * 0.041 * [0.021] [0.022] [INFLA.sup.av] (-1) -0.281 -0.363 [0.468] [0.452] [OPEN.sup.ini] 0.022 * 0.029 ** [0.013] [0.014] East Asia and Pacific -0.499 -0.817 [1.492] [1.704] South Asia 0.467 -1.057 [2.025] [2.067] Europe and Central Asia 5.354 *** 2.498 [1.718] [1.726] Sub-Saharan Africa -1.702 -3.187 * [1.344] [1.664] Latin America and the Caribbean -2.176 * -3.818 *** [1.128] [1.222] OIL -0.57 0.473 [1.506] [1.550] [IICCR.sup.av](-2) [DOMCREDIT.sup.ini] 0.026 [0.025] [GOVERNANCE.sup.av] 2.036 *** [0.690] [STDEBT.sup.ini] Constant -16.031 ** -8.54 [7.991] [9.418] R-Squared (adj.) 0.88 0.88 Number of observations 255 257 (3.5) OLS [IICCR.sup.av](-1) 0.548 *** [0.040] [PRIVSHARE.sup.ini] 0.114 *** [0.032] [DEBT.sup.ini] -0.023 *** [0.007] [TPGROWTH.sup.av] 2.569 * [1.417] [GROWTH.sup.av](-1) 0.706 *** [0.114] [GOVSTABILITY.sup.av] 1.259 [0.313] [INCOMEPC.sup.ini] 1.431 [1.003] [RESERVES.sup.ini] 0.043 * [0.021] [INFLA.sup.av] (-1) -0.355 [0.453] [OPEN.sup.ini] 0.027 ** [0.013] East Asia and Pacific -0.508 [1.564] South Asia 0.418 [2.010] Europe and Central Asia 4.431 ** [1.874] Sub-Saharan Africa -2.012 [1.437] Latin America and the Caribbean -2.407 ** [1.162] OIL -0.879 [1.382] [IICCR.sup.av](-2) [DOMCREDIT.sup.ini] [GOVERNANCE.sup.av] [STDEBT.sup.ini] 0.029 [0.043] Constant -16.225 * [8.529] R-Squared (adj.) 0.88 Number of observations 257 Notes: Standard errors in parentheses are based on a robust covariance matrix. ***, **, *denote significance levels of 1%, 5%, and 10%. The data sample is an unbalanced panel, comprising 5-yr averages or initial values between 1980 and 2005. The dependent variable is Institutional Investor's average country credit rating for the 5-yr period [IICCR.sup.av]. All regressions include time dummies; their coefficients are available upon request. TABLE 4 The Effect of PRIVSHARE on IICCR-GMM and Fixed Effects Estimation (4.1) (4.2) Diff-GMM Sys-GMM [IICCR.sup.av](-1) 0.392 *** 0.484 *** [0.109] [0.057] [PRIVSHARE.sup.ini] 0.213 ** 0.216 *** [0.101] [0.055] [DEBT.sup.ini] -0.013 -0.003 [0.017] [0.008] [TPGROWTH.sup.av] 4.312 ** 3.078 * [1.766] [1.750] [GROWTH.sup.av](-1) 0.705 *** 0.880 *** [0.2311 [0.143] [GOVSTABILITY.sup.av] 1.393 * 1.234 *** [0.724] [0.417] [INCOMEPC.sup.ini] 7.063 -0.061 [5.550] [3.195] [RESERVES.sup.ini] 0.066 * 0.051 * [0.033] [0.028] [INFLA.sup.av] (-1) -1.259 -0.484 [0.760] [0.757] [OPEN.sup.ini] -0.042 0.014 [0.061] [0.030] Constant -1.595 [25.857] R-Squared (adj.) Number of observations 193 257 Hansen's J-stat. (p-value) 0.26 0.40 AB m2-stat.(p-value) 0.95 0.88 Kleibergen-Paap rk-stat (p-value) (4.3) Diff-GMM (4.4) Sys-GMM (red. instr.) (red. instr.) [IICCR.sup.av](-1) 0.351 *** 0.470 *** [0.099] [0.077] [PRIVSHARE.sup.ini] 0.364 *** 0.329 *** [0.093] [0.087] [DEBT.sup.ini] -0.017 -0.014 [0.028] [0.020] [TPGROWTH.sup.av] 4.893 ** 2.422 [1.978] [2.185] [GROWTH.sup.av](-1) 0.573 *** 0.717 *** [0.200] [0.177] [GOVSTABILITY.sup.av] 1.843 *** 1.731 *** [0.590] [0.414] [INCOMEPC.sup.ini] 3.479 -0.276 [5.026] [4.448] [RESERVES.sup.ini] 0.059 * 0.053 * [0.032] [0.027] [INFLA.sup.av] (-1) -1.041 -0.706 [0.784] [0.683] [OPEN.sup.ini] -0.07 0.012 [0.062] [0.039] Constant -3.108 [35.771] R-Squared (adj.) Number of observations 193 257 Hansen's J-stat. (p-value) 0.40 0.48 AB m2-stat.(p-value) 0.72 0.56 Kleibergen-Paap rk-stat (p-value) (4.5) Corrected (4.6) Fixed Effects Pooled IV [IICCR.sup.av](-1) 0.653 *** 0.521 [0.086] [0.048] [PRIVSHARE.sup.ini] 0.099 * 0.159 ** [0.051] [0.071] [DEBT.sup.ini] -0.041 *** -0.020 ** [0.014] [0.008] [TPGROWTH.sup.av] 3.902 ** 3.102 ** [1.585] [1.409] [GROWTH.sup.av](-1) 0.717 *** 0.701 [0.208] [0.115] [GOVSTABILITY.sup.av] 1.193 *** 1.502 [0.426] [0.302] [INCOMEPC.sup.ini] -5.559 * 1.486 [3.194] [1.063] [RESERVES.sup.ini] 0.065 ** 0.042 ** [0.026] [0.019] [INFLA.sup.av] (-1) -0.657 -0.851 * [0.478] [0.461] [OPEN.sup.ini] 0.061 * 0.014 [0.033] [0.012] Constant -16.357 ** [7.698] R-Squared (adj.) 0.87 Number of observations 221 242 Hansen's J-stat. (p-value) 0.23 AB m2-stat.(p-value) Kleibergen-Paap rk-stat 0.02 (p-value) Notes: Standard errors in parentheses. ***, **, * denote significance levels of 1%, 5%, and 10%. The data sample is an unbalanced panel, comprising 5-yr averages or initial values between 1980 and 2005. The dependent variable is Institutional Investor's average country credit rating for the 5-yr period ([IICCR.sup.av]). All regressions include time dummies; their coefficients are available upon request. Estimates presented in columns (4.1)-(4.4) are based on two-step standard errors with the Windmeijer (2005) finite sample correction. For columns (4.3) and (4.4), we did not use lagged values of [PRIVSHARE.sup.ini] and [DEBT.sup.ini] as instruments. For the corrected fixed effects estimation in column (4.5), the Arellano and Bond (1991) difference-GMM estimator was used to initialize the bias-correction. Estimates presented in column (4.6) are based on robust standard errors clustered by country. TABLE 5 The Effect of PRIVSHARE on 1ICCR-Varying Samples and Specifications (5.1) OLS (5.2) OLS (5.3) OLS PRIVSHARE > 0 IICCR > 25 No 1980s [IICCR.sup.av](-1) 0.554 *** 0.484 *** 0.592 *** [0.047] [0.057] [0.054] [PRIVSHARE.sup.ini] 0.103 *** 0.100 *** 0.120 *** [0.035] [0.033] [0.027] [PUBDEBT.sup.ini] [PRIVDEBT.sup.ini] [DEBT.sup.ini] -0.028 ** -0.064 *** -0.012 * [0.013] [0.021] [0.006] [TPGROWTH.sup.av] 3.651 ** 4.249 *** 0.160 [1.420] [1.446] [1.529] [GROWTH.sup.av](-1) 0.705 *** 0.804 *** 0.729 *** [0.140] [0.160] [0.143] [GOVSTABILITY.sup.av] 1.378 *** 1.290 *** 1.188 *** [0.384] [0.463] [0.340] [INCOMEPC.sup.ini] 1.224 2.150 * 2.077 * [1.276] [1.089] [1.102] [RESERVES.sup.ini] 0.045 * 0.050 *** 0.057 * [0.023] [0.018] [0.030] [INFLA.sup.av](-1) -0.812 -2.170 *** -0.033 [0.623] [0.505] [0.595] [OPEN.sup.ini] 0.030 * 0.010 0.034 ** [0.015] [0.012] [0.017] Constant -9.766 -18.135 ** -14.301 [9.047] [8.137] [8.642] R-Squared (adj.) 0.86 0.81 0.89 Number of 208 160 176 observations Hansen's J-stat. (p-value) Kleibergen-Paap rk-stat (p-value) (5.6) (5.4) OLS (5.5) OLS Pooled IV [IICCR.sup.av](-1) 0.551 *** 0.590 *** 0.536 *** [0.040] [0.039] [0.061] [PRIVSHARE.sup.ini] 0.106 *** [0.031] [PUBDEBT.sup.ini] -0.027 *** -0.032 *** -0.031 *** [0.009] [0.009] [0.010] [PRIVDEBT.sup.ini] 0.022 0.256 [0.073] [0.247] [DEBT.sup.ini] [TPGROWTH.sup.av] 2.708 * 1.923 2.667 * [1.364] [1.325] [1.583] [GROWTH.sup.av](-1) 0.724 *** 0.736 *** 0.837 *** [0.113] [0.121] [0.111] [GOVSTABILITY.sup.av] 1.276 *** 1.173 *** 1.592 *** [0.317] [0.329] [0.398] [INCOMEPC.sup.ini] 1.522 1.856 1.577 [1.056] [1.149] [1.156] [RESERVES.sup.ini] 0.042 ** 0.044 ** 0.032 * [0.021] [0.021] [0.019] [INFLA.sup.av](-1) -0.363 -0.280 -0.424 [0.454] [0.466] [0.424] [OPEN.sup.ini] 0.026 * 0.031 ** 0.025 * [0.013] [0.014] [0.014] Constant -17.271 * -17.187 * -18.042 * [8.730] [9.305] [9.195] R-Squared (adj.) 0.88 0.87 0.86 Number of 257 257 242 observations Hansen's J-stat. 0.11 (p-value) Kleibergen-Paap 0.03 rk-stat (p-value) Notes: See notes on Table 3. TABLE 6 The Effect of PRIVSHARE on SOVDEFAULT (6.1) Probit (6.2) Logit [PRIVSHARE.sup.ini] -0.023 ** -0.040 ** [0.009] [0.016] [DEBT.sup.ini] 0.011 ** 0.020 ** [0.004] [0.008] [TPGROWTH.sup.av] -0.322 -0.774 [0.501] [0.945] [GROWTH.sup.av](-I) -0.078 * -0.152 * [0.044] [0.083] [GOVSTABILITY.sup.av] -0.173 -0.321 [0.108] [0.196] [INCOMEPC.sup.ini] 0.492 * 0.902 * [0.286] [0.482] [RESERVES.sup.ini] -0.003 -0.006 [0.004] [0.006] [INFLA.sup.av] (-1) -0.012 -0.002 [0.104] [0.184] [OPEN.sup.ini] -0.004 -0.006 [0.004] [0.008] East Asia and Pacific 1.199 *** 2.055 *** [0.398] [0.658] South Asia 0.684 1.103 [0.473] [0.853] Europe and Central Asia 0.066 -0.051 [0.520] [0.864] Sub-Saharan Africa 1.280 *** 2.078 *** [0.470] [0.789] Latin America and the Caribbean 1.172 *** 1.953 *** [0.407] [0.718] OIL 0.421 * 0.713 [0.255] [0.452] SO VDEFAULT(- I) Marginal effect of PRIVSHARE -0.009 ** -0.010 ** [0.004] [0.004] Observations 229 229 Pseudo [R.sup.2] 0.31 0.31 Percent correctly predicted 75.55 75.55 (6.3) Probit (6.4) Logit with LDV with LDV [PRIVSHARE.sup.ini] -0.018 ** -0.032 ** [0.009] [0.015] [DEBT.sup.ini] 0.009 ** 0.016 ** [0.004] [0.008] [TPGROWTH.sup.av] -0.235 -0.539 [0.500] [0.987] [GROWTH.sup.av](-I) -0.07 -0.13 [0.045] [0.096] [GOVSTABILITY.sup.av] -0.195 * -0.356 * [0.103] [0.182] [INCOMEPC.sup.ini] 0.462 * 0.867 * [0.267] [0.477] [RESERVES.sup.ini] -0.003 -0.005 [0.004] [0.006] [INFLA.sup.av] (-1) -0.014 -0.008 [0.099] [0.182] [OPEN.sup.ini] -0.003 -0.004 [0.004] [0.007] East Asia and Pacific 1.019 *** 1.715 *** [0.335] [0.578] South Asia 0.749 * 1.11 [0.433] [0.841] Europe and Central Asia -0.132 -0.351 [0.416] [0.693] Sub-Saharan Africa 1.002 ** 1.730 ** [0.391] [0.680] Latin America and the Caribbean 0.822 ** 1.369 ** [0.368] [0.663] OIL 0.419 * 0.727 * [0.218] [0.418] SO VDEFAULT(- I) 0.936 *** 1.617 [0.222] [0.403] Marginal effect of PRIVSHARE -0.0071 ** -0.0079 ** [0.0036] [0.0037] Observations 229 229 Pseudo [R.sup.2] 0.36 0.36 Percent correctly predicted 81.22 82.53 Notes: Standard errors in parentheses. ***, **, * denote significance levels of 1%, 5%, and 10%. The data sample is an unbalanced panel, comprising 5-yr averages or initial values between 1980 and 2005. The dependent variable is SOVDEFAULT, which is a binary variable indicating if Standard and Poor's rated a government as being in default at least once during a 5-yr period. All regressions include time dummies; their coefficients are available upon request. Robust standard errors clustered by country are reported for all regressions. TABLE 7 The Effect of PRIVSHARE on SOVDEFAULT--Probit and Logit with Fixed and Random Effects (7.1) (7.2) (7.3) Probit RE Logit RE Logit FE [PRIVSHARE.sup.ini] -0.035 ** -0.060 ** -0.123 ** [0.016] [0.028] [0.056] [DEBT.sup.ini] 0.012 *** 0.021 *** 0.031 ** [0.004] [0.007] [0.013] [TPGROWTH.sup.av] -0.438 -0.804 -1.064 [0.462] [0.833] [1.068] [GROWTH.sup.av](-1) -0.093 ** -0.169 ** -0.346 ** [0.041] [0.073] [0.136] [GOVSTABILITY.sup.av] -0.257 ** -0.446 ** -0.588 ** [0.122] [0.210] [0.261] [INCOMEPC.sup.ini] 0.692 * 1.201 ** 6.591 ** [0.357] [0.609] [2.580] [RESERVES.sup.ini] -0.004 -0.006 -0.006 [0.005] [0.009] [0.017] [INFLA.sup.av] (- 1) 0.047 0.074 0.2 [0.131] [0.223] [0.362] [OPEN.sup.ini] -0.004 -0.006 -0.028 [0.006] [0.010] [0.023] East Asia and Pacific 1.549 * 2.615 * [0.906] [1.536] South Asia 1.032 1.487 [1.163] [2.059] Europe and Central Asia 0.007 -0.04 [0.791] [1.336] Sub-Saharan Africa 1.621 ** 2.698 ** [0.699] [1.183] Latin America and the Caribbean 1.369 ** 2.298 ** [0.631] [1.077] OIL 0.54 0.908 [0.472] [0.802] SOVDEFAULT(-1) Marginal effect of PRIVSHARE -0.0140 ** -0.0148 ** [0.0064] [0.007] R-Squared (adj.) Percent correctly predicted 74.24 74.24 Number of observations 229 229 167 (7.4) (7.5) (7.6) LPM FE Probit RE LPM FE [PRIVSHARE.sup.ini] -0.009 ** -0.036 ** -0.009 ** [0.004] [0.016] [0.004] [DEBT.sup.ini] 0.002 * 0.006 * 0.002 * [0.001] [0.003] [0.001] [TPGROWTH.sup.av] -0.138 -0.349 -0.137 [0.159] [0.441] [0.154] [GROWTH.sup.av](-1) -0.029 *** -0.083 ** -0.028 *** [0.010] [0.039] [0.011] [GOVSTABILITY.sup.av] -0.067 * -0.17 -0.067 * [0.035] [0.127] [0.035] [INCOMEPC.sup.ini] 0.539 ** 0.524 * 0.539 ** [0.243] [0.283] [0.243] [RESERVES.sup.ini] -0.001 0.002 -0.001 [0.001] [0.006] [0.001] [INFLA.sup.av] (- 1) 0.042 -0.004 0.042 [0.034] [0.117] [0.034] [OPEN.sup.ini] -0.001 0 -0.001 [0.002] [0.005] [0.002] East Asia and Pacific 1.502 ** [0.748] South Asia 1.214 [0.849] Europe and Central Asia -0.314 [0.674] Sub-Saharan Africa 1.190 ** [0.556] Latin America and the Caribbean 0.754 [0.519] OIL 0.377 [0.368] SOVDEFAULT(-1) 1.063 *** 0.003 [0.256] [0.066] Marginal effect of PRIVSHARE -0.0142 ** [0.0065] R-Squared (adj.) 0.31 0.31 Percent correctly predicted 57.21 82.29 57.64 Number of observations 229 192 229 Notes: See notes on Table 6.
|Printer friendly Cite/link Email Feedback|
|Author:||Celasun, Oya; Harms, Philipp|
|Date:||Jan 1, 2011|
|Previous Article:||The political economy of intergenerational income mobility.|
|Next Article:||Monetary policy and endowment risk in a limited participation model.|