III Data and Event Studies
- Richard Hemming, Axel Schimmelpfennig, and Michael Kell
- Published Date:
- May 2003
To examine the empirical relationships between fiscal variables and emerging market crises, a large fiscal dataset was constructed. This section describes the dataset and an event study analysis of fiscal variables.
The dataset consists of annual observations for 20 or so fiscal variables and three crisis variables for 29 emerging market economies, covering the period 1970 to 2000. The sample of countries and the time period is the same as for the IMF’s main EWS model.13 The 29 countries are Argentina, Bolivia, Brazil, Chile, Colombia, Cyprus, the Czech Republic, Egypt, Hungary, India, Indonesia, Israel, Jordan, Korea, Lebanon, Malaysia, Mexico, Pakistan, Peru, the Philippines, Poland, the Slovak Republic, South Africa, Sri Lanka, Thailand, Turkey, Uruguay, Venezuela, and Zimbabwe. Table A2.1 in Appendix II compares the country and area coverage of a range of empirical studies of emerging market crises.
The IMF’s main EWS model is estimated with monthly data. Most of the variables relate to financial or monetary aggregates that are widely available at a high frequency. But almost all the fiscal data available over a long time horizon in existing cross-country databases are only available on an annual basis.
While there is a consensus in the literature on how to define currency crises, the precise specification varies across studies. For purposes of comparability, this study uses the currency crisis variable constructed for the IMF’s main EWS model, based on a foreign exchange market pressure (FMP) index.14
The FMP index is calculated as the weighted average of month-on-month changes in the exchange rate and in international reserves.15 Although it would be preferable to include a measure of interest rates in the FMP index, there are insufficient data on short-term market rates for the sample under consideration.16 A crisis is said to occur if the FMP index in a given month exceeds its mean by more than three standard deviations. An annual version of the crisis variable is used, where a given year is classified as a crisis if it contains one or more crisis months.17 This definition yields 58 currency crises; Table A2.2 in Appendix II identifies these crises, and Figure 3.1 shows their distribution over time. There is some evidence of a bunching of crises in the early 1980s, around the time of the Latin American debt crisis; in the early 1990s; and in the late 1990s, reflecting the effects of the Asian crisis and its aftermath.
Figure 3.1.Number of Crises in the Sample
Source: Authors’ calculations.
Unlike currency crises, there are few recent empirical studies of debt crises, and hence less of a consensus on how to define a debt crisis. One possibility would be to use information on sovereign defaults (as in Purcell and Kaufman, 1993). But this study follows Detragiache and Spilimbergo (2001), who define a debt crisis as occurring if either or both of two conditions apply: there are arrears of principal or interest on external private and public debt to commercial creditors (banks or bondholders) of more than 5 percent of total (public and private external) debt outstanding to commercial creditors;18 and there is a rescheduling or debt restructuring agreement with commercial creditors as listed in the World Bank Global Development Finance (GDF) database.19 The first criterion is intended to rule out cases where the proportion of debt in arrears is negligible; the second criterion captures countries that are not technically in arrears because they reschedule or restructure their debt before defaulting. A crisis episode is considered to be finished when arrears fall below the 5 percent threshold, but crises beginning within four years of the end of a previous crisis are considered a continuation of the earlier event.20 This definition yields 21 debt crisis episodes;21Table A2.2 in Appendix II identifies these crises, and Figure 3.1 shows their distribution over time. Again, there is some bunching in the early 1980s and in the late 1990s. Episodes of severe external payment difficulties that do not result in arrears or rescheduling, such as the Mexican crisis of 1994–95, are not captured by this definition, even though they are often regarded to be debt crises in the policy debate.
Defining banking crises requires more judgment than in the case of currency or debt crises. This is because data on which a quantitative definition could be based are difficult to obtain. For instance, although large withdrawals of deposits may have typified banking crises in the past, the widespread existence of deposit insurance means that banking crises are no longer necessarily accompanied by large runs on banks. Another possibility would be to use the performance of bank stocks relative to the overall equity market, but often in emerging market economies many banks are not traded publicly. The approach that has been used most in the empirical literature on banking crises, following Caprio and Klingebiel (1996), is based on data on loan losses and the erosion of bank capital. A systemic banking sector crisis is said to occur when the net worth of the banking system has been almost or entirely eliminated. However, data on nonperforming loans are usually available only at low frequencies, often after considerable lags; even then, official figures typically understate the problem. So data on loan losses and bank capital are supplemented by the judgment of authors and experts familiar with the circumstances of particular countries. More specifically, this study uses the dates of systemic banking crises in Caprio and Klingebiel (1999), which updates, corrects, and expands their earlier list of crises.22 This gives 32 banking crises; Table A2.2 in Appendix II identifies these crises, and Figure 3.1 shows their distribution over time.
Box 3.1.Fiscal Variables
The fiscal variables in the dataset fall into four groups:
Variables measuring the deficit and financing: overall balance; primary balance; actuarial deficit; total financing; and the change in net claims on government (all in percent of GDP).
Variables measuring debt: total debt; public external debt (including guarantees); short-term debt; long-term debt; foreign debt (all in percent of GDP); and foreign currency debt (in percent of total debt).
Variables related to government expenditure: total expenditure (in percent of GDP); interest expenditure; defense expenditure; and social expenditure (all in percent of total expenditure).
Variables measuring government revenue: total revenue (in percent of GDP); international trade taxes; nontax revenue; and grants (all in percent of total revenue); plus two constructed variables, tax buoyancy and revenue buoyancy.1
The first two groups of variables correspond to a subset of the “fiscal position indicators” suggested in Hemming and Petrie (2002), which are intended to indicate the strength or weakness of the initial or current fiscal situation as reflected in various measures of the deficit and debt. The expenditure and revenue variables are intended to pick up some of the structural weaknesses discussed in Hemming and Petrie (2002), such as a high proportion of nondiscretionary spending, reflecting either legal or political constraints, or a revenue base dominated by more volatile inflows.1 Defined as the percentage change in tax revenue and total revenue (respectively) divided by the percentage change in nominal GDP.
A large number of fiscal vulnerability indicators are used in this paper. The approach taken was to start with the list of suggested indicators in Hemming and Petrie (2002) and to construct as many variables as possible from existing cross-country data sources, including the IMF’s International Finance Statistics (IFS) and Government Finance Statistics (GFS) databases, and the World Bank’s GDF database. Where possible, the variables refer to general government, although many variables are only available for central government. Box 3.1 lists the fiscal variables. Further details are provided in Table A2.3 in Appendix II.
The availability of fiscal data varies between countries and across variables. Table A2.4 in Appendix II shows the observations available as a percentage of the full sample period (1970 to 2000 inclusive; i.e., 31 observations per country). Coverage is reasonably full for most of the variables related to the deficit and financing, total debt, and total expenditure and revenue, although coverage for the primary balance, the variables capturing the composition of debt, social expenditure, and grants is relatively thin. Country-specific sources are not suitable for filling gaps or extending the range of fiscal vulnerability indicators, due to a lack of comparability with the base data. Overall, the fiscal variables examined are predominantly macroeconomic rather than structural.23
Event studies have become an increasingly common approach to analyzing the antecedents and effects of financial crises.24 The sample is divided between five- or seven-year crisis “windows,” and noncrisis or tranquil years. The average values of fiscal variables before, during, and after crises can then be compared to the average during tranquil periods, in graphical format. Event studies provide a simple and intuitive summary of the univariate relationship between crises and the variables of interest.25
The event study approach implicitly treats all crisis and tranquil periods as alike, regardless of country, which raises an issue of the comparability of data across countries. All fiscal variables have been scaled (either by expressing the variable as a share of GDP or as a share of total debt, expenditure, or revenue), which addresses the comparability problem to some extent. But this does not deal with the possibility that some fiscal variables are systematically more volatile in certain countries. Therefore, each variable is further standardized by subtracting the country-specific mean and dividing by the country-specific standard deviation, following the procedure in Aziz, Caramazza, and Salgado (2000). This, however, complicates the interpretation of the results, since differences between tranquil and crisis periods are in terms of country-specific standard deviations from the mean. For this reason, results are presented for both standardized and nonstandardized or “raw” data.
To give an indication of the statistical significance of the difference between crisis and tranquil periods, 95 percent confidence intervals around these differences are shown. These intervals are calculated as twice the combined standard errors for the difference between the estimated crisis and tranquil period means.26 As such, these confidence intervals reflect the sampling error for both crisis and tranquil periods. Some earlier event studies, such as Frankel and Rose (1996), show confidence intervals based on the standard errors for crisis periods only, implicitly assuming there is no sampling error for tranquil periods.
Averages are computed separately for currency, debt, and banking crises, as well as for all three types of crises pooled together. This indicates whether fiscal variables tend to behave differently before and after each type of crisis. The results are shown in Figures 3.2–3.4 and Figures A3.1–A3.8 in Appendix III (which also includes further guidance on interpreting the figures). These figures relate to slightly more than half the fiscal variables listed in Box 3.1.27
Figure 3.2.Overall Balance
Source: Authors’ calculations.
1 In the left-hand column, the solid line shows the difference between the tranquil period mean and the mean in each year of the crisis window, in units of country-specific standard deviations. The dotted lines are two standard errors around the difference in the means, corresponding to a 95 percent confidence interval.
2 In the right-hand column, the dashed horizontal line shows the tranquil period mean and the solid line shows the crisis period mean, both in percent of GDP (unless otherwise indicated). The dotted lines are two standard errors around the difference in the tranquil and crisis period means.
Figure 3.3.Public External Debt
Source: Authors’ calculations.
1 See footnotes to Figure 3.2.
Figure 3.4.Short-Term Debt
Source: Authors’ calculations.
1 See footnotes to Figure 3.2.
The Fiscal Antecedents of Crises
The overall deficit is higher in the run-up to currency crises than in tranquil periods. This is shown in Figure 3.2, and is consistent with the findings in Frankel and Rose (1996) and Aziz, Caramazza, and Salgado (2000). The same is found for debt crises. Thus the overall deficit may be a useful leading indicator of currency and debt crises. The situation is different for banking crises, where the overall deficit has on average been lower in the run-up to a crisis than in tranquil periods. The actuarial deficit rises sharply prior to currency, debt, and banking crises, although its mean value in the two years prior to a crisis is not significantly different from its tranquil period mean. Financing variables tell a similar story to the overall balance. The actuarial deficit, total financing, and the change in net claims on government are shown in Figures A3.1–A3.3 in Appendix III.
All measures of debt increase in the run-up to crises. As shown in Figure 3.3, public external debt increases from a level (two years prior to crises) close to or below the tranquil period mean. The main exception is short-term debt, which Figure 3.4 shows is significantly above the tranquil period mean two (and even three) years prior to currency or debt crises. Perhaps surprisingly, this is not the case for foreign debt, which is shown in Figure A3.4 in Appendix III. This finding is in line with several other studies (e.g., Chang and Velasco, 1999; Bussiere and Mulder, 1999), which conclude that indicators related to short-term debt are useful for predicting crises.
Expenditure and revenue variables are not markedly different in the run-up to crises as compared to tranquil periods. However, a few interesting points emerge on the expenditure side from Figures A3.5–A3.7 in Appendix III. Total expenditure increases steadily in the run-up to all three types of crises; interest expenditure is generally lower in the run-up to crises than in tranquil periods, which is unexpected; and social expenditure is above the tranquil period mean prior to currency and debt crises.
On the revenue side, it is notable that international trade taxes, shown in Figure A3.8 in Appendix III, seem to be distinctly higher two years before currency crises than in tranquil periods. This suggests that greater reliance on trade taxes has tended to be associated with higher vulnerability to currency crises, although this may reflect overvaluation of the exchange rate prior to currency crises.
The results from the event studies can also be considered in terms of type of crisis:
There are four fiscal variables that behave in a consistently abnormal manner in the run-up to currency crises. These are the overall deficit, total financing, the change in net claims on government, and short-term debt. Two other fiscal variables—the actuarial deficit and foreign debt—increase sharply just prior to currency crises, but are less informative as leading indicators.
With regard to debt crises, short-term debt is the only variable to differ significantly from the tranquil period mean in the two years prior to a crisis. But several variables—the overall deficit, the actuarial deficit, total financing, the change in net claims on government, long-term debt, and foreign currency debt—behave abnormally in the year immediately preceding a crisis.
There is less evidence of abnormal behavior of fiscal variables in the run-up to banking crises. The only exception is short-term debt. However, a few variables behave differently prior to banking crises compared to currency or debt crises: the overall deficit, for example, is lower than the tranquil period mean in the run-up to banking crises, but higher than the tranquil period mean for debt and currency crises; a similar pattern is apparent for total financing; social expenditure rises in the run-up to banking crises but is falling or stable prior to debt and currency crises; and total revenue is above the tranquil period mean for banking crises but below it for debt and currency crises.
The Fiscal Effects of Crises
The various deficit and financing measures suggest that fiscal policy becomes more restrictive in the two years following currency and debt crises. However, there is evidence that the situation is different following banking crises, with some measures suggesting an increase in the deficit in immediate postcrisis years. As might be expected, debt increases following all types of crises. The increase, relative to tranquil periods, is significant for all three types of crisis when standardized data are considered. There is some evidence of a reduction in short-term debt following currency and debt crises. Foreign debt remains fairly stable in the two years following crises.
Total expenditure generally changes little following currency and debt crises, but in the case of banking crises remains significantly above the tranquil period average. Consistent with the increase in total debt following crises, interest expenditure rises in the years following currency and debt crises, although this is less apparent for banking crises. Social expenditure appears to be somewhat lower on average following currency and debt crises than prior to them; the opposite is true of banking crises. There is very little evidence of major effects of crises on revenue indicators. The big falloff in revenues experienced by some Asian crisis countries during 1997–98 does not appear to be typical of most emerging market crises.28
Summary of Findings
The event studies provide some evidence that fiscal variables behave abnormally before and after crises. Some deficit measures have, on average, been significantly higher in the run-up to currency and debt crises than in tranquil periods. This confirms the findings of some other event studies of currency crises, though the results of this study are somewhat clearer, especially when allowance is made for systematic differences in the level and volatility of deficits between countries. By contrast, most measures of debt are not significantly above normal prior to crises. The notable exception is short-term debt, which is significantly higher than usual in the run-up to all three types of crisis, and indeed is the only fiscal variable that behaves abnormally prior to banking crises. Perhaps this is not too surprising, since the ability to borrow long term may begin to diminish at the first sign of problems, no matter what their source. But more generally, there seems to be stronger evidence of abnormal behavior of fiscal variables prior to currency and debt crises, compared with banking crises. The overall deficit tends to narrow immediately following currency and debt crises suggesting that some tightening of policy is the norm; in the case of banking crises, the deficit tends to be reduced only a year or two after the crisis begins (although banking crises are harder to date accurately). Debt typically increases following all types of crises, and there is a shift away from short-term debt.