Government Quality Improvements and Sudden Stops Incidence (Part 2)

2.1.3 Other Regressors I include a number of macro control variables that affect the probability of a country experiencing a sudden stop based on the developing empirical literature on this subject (Calvo, et al., 2004; Frankel and Cavallo, 2004). First, I include variables to control for balance sheet effects that determine whether a reversal in capital flows will result in an output contraction. Balance sheet effects, therefore, determine whether any given episode of capital flow reversal will be classified as a sudden stop. These variables are also included because anything that increases the cost associated with a reversal may also increase the probability of the reversal occurring in the first place along the lines of a self-fulfilling crisis story. The ratio of foreign liabilities of deposit money banks to M1 measures external liability dollarization of the financial system.10 In addition, countries that trade recover more quickly from the output loss following a sudden stop (Guidotti, et al., 2004). Moreover, the more open the economy, the smaller will be the required real currency depreciation following a sudden stop (Calvo, et al., 2002). The coefficient of openness, however, is ambiguous. A weakening in a country’s export markets could trigger a sudden stop, so that an open economy is more vulnerable. Frankel and Cavallo (2004) and Calvo, et al. (2004), however, find that openness to trade is in fact associated with fewer sudden stops. In an unreported regression, I control for capital account openness using the IMF’s binary indicator of capital account openness although the variable was insignificant. Using Quinn’s more refined measure of capital account restrictions that assumes values between 0 and 1, I find similar results. 10 In an unreported regression I include a measure of domestic liability dollarization, the ratio of dollar deposits to total deposits in the banking system. Although this variable significantly reduces the sample size, the results were similar. I also include the ratio of trade to GDP to proxy for the extent to firms earn revenue in dollars and therefore the ability of banks to hedge dollar deposits by making dollar loans. Results were similar when I use the ratio of exports to GDP. 11 Calvo, et al. (2004) argue that real exchange rate change necessary for current account adjustment in the presence of a sudden stop is linked to the size of the current account deficit prevailing before the sudden stop materializes. Given the effects of real exchange rate depreciations in the presence of liability dollarization on balance sheets, output and the ability to repay, the probability of a sudden stop is an increasing function of the current account deficit and the degree of liability dollarization.11 Similarly, Edwards (2004) finds that the probability of experiencing a current account reversal increases with the size of the pre-existing current account deficit. I therefore include the ratio of the current account deficit to GDP.12 The log of GDP per capita controls for the level of economic development. Its inclusion ensures that the government quality variable is not simply picking up the effects of the level of development. I also include the log of reserves in months of imports as countries with a higher stock of reserves have a lower probability of experiencing a sudden stop and are better able to finance a current account deficit.13 The ratio of foreign direct investment to GDP is included as a measure of the stability of capital flows since bank lending or portfolio investment is partially driven by market sentiment and therefore more volatile than direct investment. Calvo (2003) emphasizes the importance of public sector debt as a determinant of capital flow reversals. I therefore include the ratio of public debt to GDP. I use both data from the World Development Indicators as well as the database on central government debt used in 11 Calvo, et al., (2005) expand this analysis and study the effects of sudden stops and balance sheet effects on relative price volatility, in particular the ratio of the wholesale price index to the consumer price index. This ratio may be preferable to the real exchange rate in that it better captures the domestic behavior of the price of tradables vis-à-vis that of non-tradables following depreciations resulting from sudden stops. 12 In an unreported regression I control for the exchange rate regime using the Reinhart and Rogoff (2003) regime classification although the variable was insignificant. 13 Results were similar using the ratio of foreign reserves to the current account surplus and the ratio of net international reserves to GDP. 12 Jaimovich and Panizza (2006). The results were similar.14 I also include a measure of external debt that includes both the public and private sector. Finally, to test that hypothesis that improvements in government quality can increase the frequency of sudden stops by encouraging speculative inflows, I include capital inflows (% of GDP) as a regressor. If the hypothesis that improvements in government quality can increase the risk of crises by encouraging capital inflows, then the inclusion of this variable should eliminate the upward sloping portion of the inverted U-shaped relationship and leave only the downward sloping part. In other words, the inclusion of this variable should make the linear GovQual coefficient insignificant. As discussed in the results section, I find that this is the case. 3. Empirical Results Before turning to the estimation of equation (1), I first present graphical evidence of the hypothesized non-linear relationship. Figure 1 is a scatter plot of the incidence of sudden stops and the average level of GovQual for each country over the sample period. None of the countries with values of GovQual below two have multiple sudden stops while the majority that experience multiple episodes have higher values of GovQual between two and three. Above values of Govqual of four, very few countries experience a single sudden stop.15 16 14 The main results were also robust to the inclusion of the government’s foreign debt to GDP, total debt service as a % of exports, and the rate of growth of domestic credit (as a % of GDP) which captures the effect of rapid growth in private sector credit on sudden stops. 15 The graph is similar when broken down by region except for Eastern Europe where none of the countries experienced sudden stops. Part of the reason is that the transition economies were not in existence for half of the sample period and are therefore have less time to experience one or more sudden stops. The graph is also similar when the data are broken up by decades. 16 As can be seen from the figure, there are only three countries (Chile, Mexico and Turkey) with three or more sudden stops during the sample period, all of which have roughly average levels of institutional quality. If I omit these countries, the point estimates in the Poisson regression still imply an inverted U-shaped relationship although the coefficients are no longer significant. However, while the Poisson estimates are now insignificant, the probit estimates are still significant implying a non-linear relationship between GovQual and the probability of a sudden stop. Moreover, while I am omitting only 3 countries, these countries account for 12% of the sudden stops in the sample (10 out of 85), so I am removing a fairly large percentage of the crisis episodes. Given that sudden stops are 13 Table 2 presents results for the estimation of equation (1). The dependent variable is the number of sudden stops that country “i” experiences during the period 1982-2004. The government quality variable reduces the sample size from 154 countries to 122. All regressions contain region dummy variables although the coefficients are not shown. The standard errors are robust to heteroskedasticity. In the first column, results are presented for the regression when the government quality variables are included without the macro control variables. The coefficients of both Govqual and Govqual2 are generally significant at the 5% level and always significant at the 10% level, suggesting an inverted U-shaped relationship between government quality and the frequency of sudden stops, although the coefficients are never jointly significant at the 5% level.17 18 In fact, Govqual is insignificant when included without the quadratic term, further revealing a non-linear relationship. Further calculations reveal that improvements in government quality only begin to reduce the incidence of sudden stops at average levels of government quality, approximately 3 on a scale from 0-6. An improvement in government quality from 0 to 3 increases the number of sudden stops by approximately .70. This suggests that for a large range of countries, the increased capital inflows that improvements in government quality generate increase the probability of experiencing a sudden stop more than the corresponding improvements in prudential regulation and supervision reduce it. In fact, the relationship still holds even when omitting countries with the very worst institutional values. Specifically, excluding countries with values of government quality below 1.5 (approximately 1.5 standard deviations below the mean) still yields significant results. This rare events to begin with, it is important that the only countries with three or more sudden stops are those with average levels of institutional quality, not those with the worst institutions. 17 The standard deviation of Govqual is close to one, so a one-unit increase also represents an increase in one standard deviation. 18 The model without any of the control variables is accurate 60% of the time in predicting whether a country experiences zero or one or more sudden stops during the sample period. 14 indicates that the non-linear relationship is not just a result of the fact that countries with the worst institutions do not receive large inflows relative to GDP and therefore cannot suffer from capital flow reversals. This relationship is robust to the inclusion of a number of macro control variables. It is important to note that the importance of government quality is robust to the inclusion of GDP per capita, implying that Govqual is not just picking up the effects of the level of development.19 The lack of statistical and economic significance of the macro control variables is noteworthy. While one might be inclined to conclude that institutions matter more than policy and that once government quality is accounted for, macro policy has little residual influence on the frequency of sudden stops, this is most likely not the case. Because the macro variables, which show significant variation over time and could help to explain individual crisis episodes, are averaged over the twenty year sample period in the cross-sectional analysis, their effect is likely to underestimated. In the panel analysis presented later, many of these variables are significant.20 In the last column I include capital inflows as a % of GDP. The inclusion of this variable makes the linear GovQual coefficient insignificant at the 10% level, which implies that only the downward sloping portion of the inverted U-shaped relationship remains.21 This result is consistent with the argument that improvements in government quality can increase the frequency of sudden stops by encouraging capital inflows. Once inflows are controlled for, 19 When I include government quality and GDP per capita, as linear and quadratic terms, both show an inverted Ushaped relationship with the incidence of sudden stops although neither relationship is significant at standard confidence levels. However, the non-linear relationship between government quality and sudden stops is almost significant at the 10% level. Given the causal effect of institutions on the level of development (Acemoglu et al. 2001), the effect of government quality is more accurately portrayed when the square of GDP per capita is not included. 20 In fact, there is some time variation in institutional quality, especially in transition economies and select countries in Latin America and Asia that have ended dictatorships in the 1990’s. This suggests that the estimated coefficients may understate the true effect of government quality as well. 21 Although the inclusion of capital inflows shrinks the sample, this is not the cause of the drop in significance of GovQual (if I run the model without inflows but limit the sample to countries with data on inflows, GovQual is still significant). 15 improvements in government quality should unambiguously reduce the frequency of sudden stops. Furthermore, the drop in significance of GovQual is much larger if I control for portfolio inflows as opposed to total capital inflows (which includes foreign direct investment and portfolio inflows). A likely explanation is that portfolio flows are far more likely to reverse suddenly than FDI. Therefore, including portfolio inflows is a better control for the channel through which improvements in government quality can increase the risk of sudden stops. In order to determine whether certain policies move the “hump” to the left so that for lower values of government quality, the relationship with sudden stops is downward sloping, I interact in unreported regressions both the linear and quadratic government quality variable with variables that affect both the probability and cost of capital flow reversals. For example, greater trade openness should reduce the cost associated with a reversal. If the interaction term with GovQual is negative and the interaction with GovQual2 is positive, then greater trade openness shifts the hump to the left. I find, however, that the interaction terms are insignificant. I also include interaction terms with the IMF’s measure of capital account openness. The stimulative effect of improvements in government quality on inflows should be greater with more open capital accounts, thereby shifting the hump to the right. If true, this would imply that capital controls should accompany institutional improvements until a certain level of institutional quality is reached and it is safe to open the capital account. I find, however, that the interaction terms are insignificant. Nonetheless, there still may be merit to this policy implication since the IMF measure fails to account for the scope and effectiveness of capital controls and is therefore an imperfect measure of capital account restrictions. Interactions with measures of liberalization based on the volume of inflows were also insignificant, although these too are imperfect 16 measures since there are numerous determinants of capital flows aside from capital account restrictions. Finally, I estimate a probit model analyzing the probability that country “i” will experience a sudden stop in year “t.” The panel results are presented in Table 3. The macro variables are lagged to mitigate endogeneity problems. The standard errors are Huber–White robust and are clustered to correct for serial correlation within countries. The coefficients of the government quality variables, while still significant, are less significant than in Table 3.22 An F test of the joint significance of the linear and quadratic government quality variable yields a similar conclusion. The coefficients are jointly significant at the 5% level in two of the ten specifications. The greater difficulty of institutional quality to predict sudden stops in a given year is not surprising. Based on these coefficients, improvements in government quality (from a 0 to a 3 where 3 is the mean) increase the probability of a sudden stop by approximately 5 percentage points. Only at that point do further increases reduce the likelihood of a sudden stop. Finally, the macro variables are more significant than in the cross-sectional analysis. This is not surprising given that the macro variables of the countries that experienced sudden stops are averaged in the cross-sectional analysis and therefore their effect is downplayed. Larger CA surpluses, higher levels of development and a larger share of FDI in GDP tend to reduce the probability of a sudden stop. Greater levels of foreign liabilities and external government debt tend to increase the probability of a sudden stop.23 22 As a robustness check, I include year dummies as well. The linear government quality term, while always positive, is now insignificant in all specifications, although p-values of approximately 0.15 are at least suggestive of an initial upward sloping relationship. The quadratic term, however, is often significantly negative at the 10% level. The inverted U-shaped relationship therefore seems to hold although is less significant. A possible explanation for the reduced significance is that year dummies can account for contagion effects that may be correlated with government quality. 23 Frankel and Cavallo (2004) and Calvo, et al., (2004) estimate a similar probit model. For the most part, the results are similar. All three papers find that larger CA surpluses and increased trade reduce the probability of a sudden stop, and foreign exchange reserves and FDI/GDP are generally insignificant. Calvo, et al., (2004) also find that 17