WIXLIB

-5studies use somewhat different methodologies to identify currency crises (Appendix II). Forthe purposes of our analysis, we use a simple majority rule to identify the crises dates in the108 countries. Specifically, for a given country, we select a particular year as the crisis yearonly if the majority of these papers identify it as a crisis year. Using this criterion, wedropped 11 countries from our sample as crises in those countries failed to pass the majoritytest.4 The remaining sample of 97 countries had a total of 229 crises. We further dropped34 crisis episodes and 6 countries because data on key variables around the time of the crisiswere not available.5 Thus, we conduct our analysis on 195 crisis episodes in 91 countries(42 African countries, 17 Asian countries, 20 Latin American countries, and 12 countriesfrom Central and Eastern Europe and Middle East), of which, 24 crises occurred in the1970s, 83 in the 1980s, and 88 in the 1990s.6 To check for the robustness of our results,where appropriate, we also report four sets of results based on crises identified in the fourstudies.Data used in this paper were obtained from the IMF’s International FinancialStatistics, and World Economic Outlook, the World Bank's World Development Indicatorsand Global Development Finance, the Economist Intelligence Unit, and JP Morgan’swebsite. Details on the sources and construction of variables are provided in Appendix III.A. Measuring Contractions and ExpansionsTo measure the effect of a currency crisis on growth, we define (gpost n - gpre m) as thedifference in the average growth rate between m precrisis years and n postcrisis years, whereg is the growth rate of real GDP.7 If (gpost n - gpre m) is positive, we call it an expansionaryepisode and if it is negative, it is termed a contractionary episode. We also define (gpost n–gpre m, tranq) as the difference in the average growth rate between the nearest m “tranquil”precrisis years and n postcrisis years, where “tranquil” years exclude any crisis years. Wethen experimented with several different values for m and n and found that these measureswere highly correlated (Table 1). Of the measures listed in Table 1, we found (gpost 2 –gpre 3,tranq) to be the most intuitively appealing. An average of the three precrisis years4Countries excluded from our sample are Barbados, Belize, Djibouti, Grenada, Haiti, Oman,Panama, Seychelles, St. Vincent and Grenadines, Tunisia, and Servia and Montenegro(formerly Yugoslavia).5These countries are Guinea, Israel, Liberia, Samoa, Taiwan Province of China, andVanuatu.6Some of these studies do not cover the period until 1998. Therefore, where possible, weupdate the crises dates until 1998 by using the respective methodologies of the authors.7While we experimented with measures that capture the deviation of growth rate from alinear or nonlinear trend (such as the HP filter), we do not report the results as these measuresseem inappropriate for analyzing the short-run effects of currency crises, in general, or formany developing countries undergoing structural changes frequently.

-6smoothes out large fluctuations that may occur in a particular year (due to, say, a bumperharvest or a natural calamity). Similarly, an average of two postcrisis years, rather thansimply using the first postcrisis year, would take care of the measurement problem that mightarise if a crisis occurred very early or late in a given year.8 In any event, we conductedsensitivity analyses using other measures and found the results to be robust to alternativedefinitions of growth slowdown.Other studies that look at the short-run growth effects of crises have used measuressimilar to ours (for example, see Bordo and Eichengreen, 2001). Studies examining longerrun effects of crises have also used similar precrisis windows, but longer postcrisis windows(Aziz, Caramazza, and Salgado, 2000).III. STYLIZED FACTS ON GROWTH DURING CURRENCY CRISESA. Contractions and Expansions During CrisesWe begin by looking at the frequency distribution of the postcrisis growth rate ofoutput, gpost 2, in Figure 2. It shows that there is a wide variation in postcrisis growth rates,and the distribution approximates a normal distribution.9 Only 28 percent of the crises areassociated with a decline in growth rate and only in 3 percent of the episodes—Nicaragua in1979 and Uruguay in 1982—does the growth rate decline by more than 10 percent.The frequency distribution of gpost 2–gpre 3,tranq , though more skewed toward acontraction, also shows a large variation in the postcrisis growth rates. Figure 3 shows that43 percent of the crises are expansionary and 57 percent of the crises are contractionary inthe sample. Average expansion during the expansionary episodes is 3.5 percent, and averagecontraction during the contractionary episodes is 4.8 percent. Only 6 percent of the crises inthe sample experienced a contraction exceeding 10 percent. This pattern is robust toalternative ways of identifying crises (Figure 4) for the 14 countries that were commonacross the studies.B. Severity of Output Crises Over TimeThe pattern of growth rates has been quite similar in the 1970s, 1980s and 1990s(Figure 5, first panel). The growth rate significantly slows down between the precrisis andcrisis year—by 3 percentage points in the 1970s, 2.5 percentage points in the 1980s, andnearly 2 percentage points in the 1990s. The crisis year is generally the trough of the growthcurve, with the growth rate picking up the year after the crisis and reverting to the precrisis8For example, the real effects of the Mexican crisis of December 1994 would not becaptured, if n 1.9The Jarque-Bera test statistic indicates that the null hypothesis of a normal distributioncannot be rejected.

-7level by the second year after the crisis. A graph of the frequency distribution of (gpost 2–gpre 3,tranq) also seems similar across different decades (Figure 5, second panel), although thenumber of expansions and contractions vary by decades. Given that the recent literature hasfocused more on the contractionary aspects of the crises in the 1990s, it is surprising to seethat 42 percent of the crises were expansionary in the 1990s as compare with 28 percent inthe 1970s and 50 percent in the 1980s.To confirm the pattern described above, we regress (gpost 2 - gpre 3,tranq) on dummiesfor each of the three decades in a simple OLS framework (with heteroskedastically consistentstandard errors). Coefficients of the dummies measure the average contraction/expansionduring crises in the respective decades. As indicated in Table 2, on average, crises have beencontractionary in all three decades. It was somewhat of a surprise to see that the magnitudeof contraction has steadily fallen, from -3.0 in the 1970s to -1.1 in the 1980s to -0.8 in the1990s. However, the null hypothesis that the coefficients are identical across the threedecades is rejected at the 10 percent significance level for only the 1970s compared withthe 1990s. Finally, if we limit the sample to only those crises that were associated with acontraction, the average contraction was not significantly more severe during the 1990sthan in the 1970s or the 1980s. Thus, there does not appear to be any evidence to supportthe popular view that the severity of crises has systematically worsened in the 1990s.We further compare the output response during crisis episodes during which a largenumber of countries were simultaneously affected—namely, the East Asian crisis in 1997,the debt crisis in Latin America (LA) in the 1980s, and the Mexican crisis in 1994.10Interestingly, we find that, though the coefficient of the East Asian crisis is the largest, theaverage contraction during the East Asian crisis is not significantly different from that duringeither the LA debt crisis or the Mexican crisis (Table 2).C. Severity of Output Crises Across CountriesA significant number of large expansions in the sample occurred in low-income,small open economies, several of them in Africa—some examples are Central AfricanRepublic in 1994, Chad in 1981, Ethiopia in1993, Gabon in1981, Ghana in1978, Republicof Congo in 1994, and Senegal in 1981. We categorized countries by the volume of foreigncapital they received during the sample period. To see whether a pattern could be establishedamong countries that were exposed to varying levels of foreign capital, we defined largeemerging markets (LEMs) as countries that received, on average, US 100 million or moreof private external capital flows per year between 1970-98, and small emerging markets(SEMs) as those countries that received less than US 100 million (Appendix I). Indeed,only 30 percent of crises in LEMs, are associated with an expansion as compared with about50 percent in the SEMs.10The LA debt crisis episodes include 1982-83 in the following countries: Argentina,Bolivia, Brazil, Chile, Ecuador, Mexico, and Uruguay. The Mexican crisis event includesArgentina and Mexico in 1994/95. The East Asian crisis includes crises in Indonesia, Korea,Malaysia, Philippines, and Thailand in 1997.

-8-Formally, we regressed (gpost 2 - gpre 3,tranq) on dummies for LEMs and SEMs. Theestimated coefficient for the LEMs dummy variable was –3.07, while that for the SEMsdummy was 0.24 and the two coefficients were statistically significantly different from eachother. In other words, while crises in LEMs have, on average, been contractionary, crises inSEMs were expansionary but insignificant. Finally, by limiting the sample to only LEMs, weconfirm our earlier finding for the whole sample—the average growth contraction duringcrises within LEMs did not change significantly across the three decades (Table 2). Theseresults remain unchanged qualitatively if we redefine LEMs and SEMs on the basis of thesize of the economy (GDP in US ).In summary, while the majority of currency crises in the sample have beencontractionary, a large proportion, more than 40 percent, has been expansionary. Second,there is no significant difference in the ratio of expansionary to contractionary crises in thelast three decades. Contrary to the common view, the crises during the 1990s were not morecontractionary, on average, than the previous two decades. Third, the magnitude ofcontraction in contractionary crises and expansion in expansionary crises has also notchanged significantly in the last three decades. Finally, consistent with popular perception,large emerging markets experienced more contractionary crises than small emerging markets.IV. WHAT EXPLAINS THE BEHAVIOR OF OUTPUT DURING CRISES?How output responds to a currency crisis is likely to depend on a multitude offactors—the conditions prevailing in the real, external, and financial sectors at the timeof the crisis, fiscal and monetary policies implemented during the crisis, and the structuralcharacteristics of the economy. Our empirical analysis relies heavily on the explanationsprovided in the crisis literature, outlined below, that points to several channels of influence.A. Liability Dollarization and External Debt BurdenWhen the liabilities of domestic firms are denominated in foreign currency, adevaluation raises their debt burden. As a consequence, firms find it difficult to serviceexisting debt or raise new loans, which, in the aggregate, is reflected in a decline ininvestment and economic activity.11 Since it is not unusual for developing countries toborrow externally, they are particularly vulnerable to this effect. We attempt to capture thiseffect in our empirical work by including the change in external long-term debt burden asone of the explanatory variables. This variable is measured in two ways—in nominal terms(NOMINAL DEBT) and in real terms (REAL DEBT).11See Bruno (1979), van Wijnbergen (1986), Calvo (1998), and Mishkin (1999).

-9B. Sudden Stop or Reversal of External Capital InflowsCalvo and Reinhart (1999) show that, if a currency crisis is accompanied by a suddenstop or reversal of external capital inflows, and the associated loans to domestic projects isof shorter maturity than the projects themselves, it increases the incidence of nonperformingloans and reduces productive activity. Thus, countries that have been recipients of largecapital inflows are more likely to experience a contraction following a currency crisis.Accordingly, we include the three-year cumulative flow of external private capital prior tothe year of the crisis (CAPFLOW) expressed in percent of GDP as one of the explanatoryvariables.C. External LiberalizationOpen capital accounts undermine the authorities’ ability to prevent capital fromflowing out or to undertake countercyclical policies during currency crises. The severityof the crisis is worse if the capital account was opened without adequately strengtheningdomestic regulations and supervision of financial institutions. To adequately capture thiseffect, we include a variable that proxies controls on foreign exchange and capital accounttransactions (CAPCON), which is used as an interaction term with the private capital flowsvariable in our analysis.12 We want to see whether output crises were more severe incountries with more open capital accounts and experiencing capital flow reversals.D. Banking CrisisIt is not uncommon for the banking sector to come under stress at the time of thecurrency crisis. With devaluation adversely affecting the balance sheets of their clients andincreasing their nonperforming loans, banks may roll back their lending activities, giving riseto a “credit crunch” (Mishkin, 1999). Accordingly, we include a banking crisis dummy(BANK) to see whether its simultaneous occurrence worsened output performance.E. Short-Term Debt and Liquidity CrisisRodrik and Velasco (1999) show that difficulties in rolling over short-term debtduring currency crises could squeeze liquidity in the economy, and shrink the level ofeconomic activity. They find the ratio of short-term external debt to foreign exchangereserves (DEBT RESERVES) to be a robust predictor of financial crises and their severity.13To see whether output is also affectedly adversely, we include DEBT RESERVES in ourregressions.12Capital account restrictions refer to the lack of convertibility of the domestic currency forcapital account transactions. Exchange rate restrictions refer to the regulation of nominalexchange rate or when the country maintains dual or triple exchange rates.13Rodrik and Velasco (1999) define a currency crisis as a significant reversal of externalcapital flows.

- 10 -F. Devaluation and External TradeA devaluation of the currency could restore competitiveness of the economy andprovide a boost to the production of tradables (provided the Marshall-Lerner conditionholds). However, competitiveness increases only if nominal devaluation translates into a realdevaluation of the exchange rate, there is no concurrent devaluation by other competitivecountries, and the effect of devaluation is not offset by a negative terms-of-trade shock.Additionally, the trade regime needs to be sufficiently open to realize the beneficial effectsof the devaluation. 14To test whether devaluation affects output positively via the growth of exports, weuse measures of overvaluation of the exchange rate (against the U.S. dollar, OVERVAL US,and against major trading partners, OVERVAL MULTI), average export growth rate(X GROWTH) during the crisis year and the first postcrisis year, share of trade (exports plusimports) in GDP (OPEN), and a measure of competitive devaluation by other countries(COMP DEVAL) in the regressions.G. Monetary and Fiscal PoliciesMonetary policy is often tightened to stem the extent of speculative attack on thecurrency and to prevent foreign exchange reserves from falling rapidly. Similarly, fiscalpolicy may be tightened during a crisis to signal a strong policy resolve on the part of theauthorities and to compensate for the fiscal burden that inevitably arises following a crisis.While a strong case in favor of tight macroeconomic policies can reasonably be made to stemthe slide of the domestic currency or prevent capital outflows, their effect on output growth isnot obvious.To proxy for monetary and fiscal policies we use percentage change in broad realmoney supply during the crisis ( M2); change in the real interest rate between the crisis yearand the precrisis year

Authors' E-Mail Addresses: pgupta@imf.org; rsahay@imf.org; dmishra@worldbank.org 1 Poonam Gupta and Ratna Sahay are with the IMF and Deepak Mishra is with the World Bank. The views expressed in this paper are those of the authors and not necessarily of the organizations with which they are affiliated. They would like to thank Michael Adler,