Assessing Financial Vulnerability _Institute for International Economics

1 Introduction This study analyzes and provides empirical tests of early warning indicators of banking and currency crises in emerging economies. The aim is to identify key empirical regularities in the run-up to banking and currency crises that would enable officials and private market participants to recognize vulnerability to financial crises at an earlier stage. This, in turn, should make it easier to motivate the corrective policy actions that would prevent such crises from actually taking place. Interest in identifying early warning indicators of financial crises has soared of late, stoked primarily by two factors: the high cost to countries in the throes of crisis and an increasing awareness of the insufficiency of the most closely watched market indicators. There is increasing recognition that banking and currency crises can be extremely costly to the countries in which they originate. In addition, these crises often spill over via a variety of channels to increase the vulnerability of other countries to financial crisis. According to one recent study, there have been more than 65 developing-country episodes during 1980-95, when the banking system’s capital was completely or nearly exhausted;1 the public-sector bailout costs of resolving banking crises in developing countries during this period have 1. See Caprio and Klingebiel (1996b). Other identifications of banking crises over this period can be found in Demirguc-Kunt and Detragiache (1998), Eichengreen and Rose (1998), IMF ¨¸ (1998c), Kaminsky and Reinhart (1999), and Lindgren et al. (1996). 1 Institute for International Economics | http://www.iie.com been estimated at around $250 billion.2 In more than a dozen of these banking crises, the public-sector resolution costs amounted to 10 percent or more of the country’s GDP.3 In the latest additions to the list of severe banking crises, the cost of bank recapitalization for the countries most affected in the ongoing Asian financial crisis is expected to be huge—on the order of 58 percent of GDP for Indonesia, 30 percent for Thailand, 16 percent for South Korea, and 10 percent of GDP for Malaysia (World Bank 2000). In addition to the enormous fiscal costs, banking crises exacerbate declines in economic activity, prevent national saving from flowing to its most productive use, limit the room for maneuver in the conduct of domestic monetary policy, and increase the chances of a currency crisis as well (Lindgren, Garcia, and Saal 1996; Goldstein and Turner 1996). Illustrative of the magnitude of output losses, an International Monetary Fund study (IMF 1998c), drawing on a sample of 31 developing countries, reported that it typically takes almost three years for output growth to return to trend after the outbreak of a banking crisis and that the cumulative output loss averaged 12 percent.4 Probably the main reason Mexican authorities did not make more aggressive use of interest rate policy after the assassination of presidential candidate Luis Donaldo Colosio in March 1994 is that bad loan problems in the banking system had by then already become serious, and they were worried that recourse to higher interest rates would push Mexican banks over the edge. Yet failure to increase domestic interest rates in the face of international investors’ rising concern contributed to a rapid decline in international reserves and helped to transform a banking problem into a currency and debt crisis (Calvo and Goldstein 1996). This pattern in the timing of the banking and currency crises is not unique to the Mexican case. Drawing on a broader sample of banking and currency crises in emerging economies, there is evidence that the onset of a banking crisis typically precedes a currency crash (Kaminsky and Reinhart 1999; IMF 1998a).5 2. This figure is net of the estimated amount of loans that were eventually repaid. See Honohan (1997). 3. See Goldstein (1997) for a list of these severe banking crises. For comparison, the publicsector tab for the US saving and loan crisis is typically estimated at about 2 to 3 percent of US GDP. 4. In chapter 7, we present our own estimates of how long it takes growth rates of real output to recover after banking and/or currency crises. 5. In chapter 3, we provide further evidence that the presence of a banking crisis is one of the better leading indicators of a currency crisis in emerging economies. At the same time, the evidence also suggests that a currency crash aggravates the problems in the banking sector, as the peak of a banking crisis most often occurs following the collapse of the currency. The dating of currency and banking crises is discussed in detail in chapter 2. 2 ASSESSING FINANCIAL VULNERABILITY Institute for International Economics | http://www.iie.com Table 1.1 Emerging Asia: real GDP growth forecasts, 1996-98 Country Indonesia Thailand South Korea Malaysia The Philippines Hong Kong 1996 7.8 6.4 7.1 8.6 5.7 4.9 1997 4.9 1.3 5.0 7.8 5.1 5.3 1998 (as of May 1997) 7.5 7.0 6.3 7.9 6.4 5.5 Actual 1998 13.7 8.0 5.8 6.7 0.5 5.1 Source: International Monetary Fund, World Economic Outlook. Although the contagion of financial disturbances usually runs from large countries to smaller ones, the Asian financial crisis has shown that severe financial-sector difficulties in even a relatively small economy (namely Thailand) can have wide-ranging spillover effects if it acts as a ‘‘wake-up call’’ for investors to reassess country risk and if a set of other economies have vulnerabilities similar to those in the economy first affected.6 The costs of currency crises have likewise been shown to be significant both in terms of reserve losses and output declines (see chapter 7). During the Exchange Rate Mechanism (ERM) crises of the fall of 1992 and summer of 1993, about $150 billion to $200 billion was spent on official exchangemarket intervention in a fruitless effort to stave off the devaluation and/or floating of ERM currencies. Mexico’s peso crisis was accompanied in 1995 by a decline in real GDP of 6 percent—its deepest recession in 60 years. In emerging Asia, consensus forecasts for 1998 growth issued just before the crisis (that is, in May-June 1997) generally stood in the 6 to 8 percent range. As indicated in table 1.1, these forecasts were subject to unprecedented downward revisions in the midst of the currency, banking, and debt crises enveloping these economies. The IMF (1998c) estimates that emerging economies suffer, on average, an 8 percent cumulative loss in real output (relative to trend) during a severe currency crisis. And like banking crises, currency crises seem contagious. One recent study found that a currency crisis elsewhere in the world increases the probability of a speculative attack by an economically and statistically significant amount even after controlling for economic and political fundamentals in the country concerned (Eichengreen, Rose, and Wyplosz 1996; see also Calvo and Reinhart 1996; Kaminsky and Reinhart 2000). The more costly it is to clean up after a financial crisis, the greater the returns to designing a well-functioning early warning system. 6. See Calvo and Reinhart (1996) and Goldstein (1998a). Kaminsky and Reinhart (2000) provide an analysis of contagion in the Asian crisis that stresses the financial links among these countries—including the sudden withdrawal of funds by a common commercial bank lender or mutual fund investor. See also chapter 6. INTRODUCTION 3 Institute for International Economics | http://www.iie.com 4 Institute for International Economics | http://www.iie.com Table 1.2 Rating agencies’ performance before the Asian crisis: Moody’s and Standard & Poor’s long-term debt ratings,a 1996-97 15 January 1996 Rating Moody’s foreign currency debt Indonesia Malaysia Mexico The Philippines South Korea Thailand Standard & Poor’s Indonesia Foreign currency debt Domestic currency debt Malaysia Foreign currency debt Domestic currency debt Baa3 A1 Ba2 Ba2 A1 A2 Outlook 2 December 1996 Rating Baa3 A1 Ba2 Ba2 A1 A2 Outlook 24 June 1997 Rating Baa3 A1 Ba2 Ba2 stable A2 Outlook 12 December 1997 Rating Baa3 A1 Ba2 Ba2 Baa2 Baa1 Outlook negative negative October 1997 BBB stable BBB A A AA stable BBB A A AA stable BBB A A AA negative negative negative negative A AA stable stable positive Mexico Foreign currency debt Domestic currency debt The Philippines Foreign currency debt Domestic currency debt South Korea Foreign currency debt Domestic currency debt Thailand Foreign currency debt Domestic currency debt BB BBB BB BBB AA negative BB BBB BB BBB AA stable positive BB BBB BB A positive positive BB A stable stable Institute for International Economics | http://www.iie.com 5 positive stable stable A stable A AA stable A AA stable BBB A negative negative a. From highest to lowest, Moody’s rating system includes Aaa, Aa1, Aa2, Aa3, A1, A2, A3, Baa1, Baa2, Baa3, Ba1, Ba2, and Ba3, and Standard & Poor’s runs AAA, AA , AA, AA , A , A, A , BBB , BBB, BBB , BB , BB, and BB . Source: Radelet and Sachs (1998). The second reason for the increased interest in early warning indicators of financial crises is that there is accumulating evidence that two of the most closely watched ‘‘market indicators’’ of default and currency risks— namely, interest rate spreads and changes in credit ratings—frequently do not provide much advance warning of currency and banking crises (see chapter 4). Empirical studies of the 1992-93 ERM crisis have typically concluded that market measures of currency risk did not raise the specter of significant devaluations of the weaker ERM currencies before the fact (Rose and Svensson 1994). Another study, encompassing a larger number of crisis episodes, similarly concluded that the currency forecasts culled from ´ survey data are useless in anticipating the crises (Goldfajn and Valdes 1998). In the run-up to the Mexican crisis, market signals were again muted or inconsistent. More specifically, measures of default risk on tesobonos (dollar indexed, Mexican government securities) jumped up sharply in April 1994 (after the Colosio assassination) but stayed roughly constant between then and the outbreak of the crisis (Leiderman and Thorne 1996; Obstfeld and Rogoff 1995). From April 1994 on, market expectations of currency depreciation on the peso usually were beyond the government’s announced rate; nevertheless, this measure of currency risk fluctuated markedly. The gap between market expectations and the official rate was widest in summer of 1994, but the attack came with most ferocity only in late December (Obstfeld and Rogoff 1995; Leiderman and Thorne 1996; Rosenberg 1998). The evidence now available suggests that the performance of interest spreads and credit ratings was likewise disappointing in the run-up to the Asian financial crisis. Examining interest rate spreads on three-month offshore securities, one study found that these spreads gave no warning of impending difficulties (i.e., were either flat or declining) for Indonesia, Malaysia, and the Philippines and produced only intermittent signals for Thailand (Eschweiler 1997b). A recent analysis of spreads using local interest rates for South Korea, Thailand, and Malaysia similarly found little indication of growing crisis vulnerability (Rosenberg 1998). Sovereign credit ratings (on long-term, foreign-currency debt) issued by the two largest international ratings firms were even less prescient in the Asian crisis (see chapter 4, as well as Radelet and Sachs 1998; Goldstein 1998c).7 As shown in table 1.2, there were almost no downgrades for the 7. In a recent report, Moody’s (1998) argues that its rating record in the East Asian crisis was better than it appears at first sight from ratings changes alone. More specifically, the report argues, inter alia, that Moody’s went into the crisis with lower ratings for the crisis countries than the other major ratings agencies (i.e., Standard & Poor’s and Fitch-IBCA), that it took ratings actions before its main competitors, that its low bank financial strength ratings identified many of the banks that subsequently experienced stress in the crisis countries, that changes in sovereign credit ratings led to a widening of yield spreads in the crisis countries, and that one should examine the sovereign research reports—not just the 6 ASSESSING FINANCIAL VULNERABILITY Institute for International Economics | http://www.iie.com most severely affected countries in the 18-month run-up to the crisis. As The Economist (13 December 1997, p. 68) put it, ‘‘[I]n country after country, it has often been the case of too little, too late.’’ Looking at a larger sample of cases, a recent study by the Organization for Economic Cooperation and Development (OECD) was unable to find consistent support for the proposition that sovereign credit ratings act more like a leading than a lagging indicator of market prices (that is, of interest rate spreads; see ´ Larraın, Reisen, and von Maltzan 1997). Furthermore, international organizations such as the IMF did not do better than the rating agencies in anticipating several of the recent crises. A recent external evaluation of IMF surveillance concludes: We found that the Fund—in both bilateral and multilateral surveillance—largely failed to identify the vulnerabilities of the countries that subsequently found themselves at the center of the Asian financial crisis, except in the case of Thailand. In particular, it failed until rather late in the day to address a number of systemic issues. Moreover, to the extent that surveillance did identify these vulnerabilities, the tone of published Fund documents—notably [the World Economic Outlook]— was excessively bland prior to the December 1997 update of WEO [and the International Capital Markets Report], after the crisis had erupted. (IMF 1999, 56) There are, of course, several reasons interest rate spreads or changes in sovereign credit ratings may not anticipate financial crises well.8 For one thing, market participants may not have timely, accurate, and comprehensive information on the borrower’s creditworthiness. Several recent examples underscore the point (see also Goldstein 1998a; Corsetti, Pesenti, and Roubini 1998; BIS 1998). Thailand’s commitments in the forward exchange market and South Korea’s lending of international reserves to commercial banks meant that official figures on gross international reserves gave a misleading (i.e., overoptimistic) view of each country’s net usable reserves. Similarly, external foreign-currency denominated debt of Indonesian corporations, along with nonperforming bank loans in South Korea, Thailand, Malaysia, and Indonesia, turned out to be considerably larger than precrisis published official data suggested. Ceteris paribus, one could conjecture that if the true size of liquid assets and liabilities were known at an earlier stage, interest rate spreads would have been higher and credit ratings would have been lower than actually observed before the Asian crisis; this in turn could well have moderated the sharp ratings—in looking for early warning signals. At the same time, the report acknowledges that the firm is studying several potential enhancements to its analytical methodology to help improve the predictive power of its sovereign ratings. 8. It is sometimes also argued that even when credit rating agencies or international financial organizations (such as the IMF) conclude that crisis vulnerability is high, they will be reluctant to go ‘‘public’’ with a downgrade or a warning for fear of being accused of precipitating the crisis. INTRODUCTION 7 Institute for International Economics | http://www.iie.com change in market sentiment that was associated with the ‘‘news’’ of the lower-than-expected net worth of Asian debtors. The other reason market prices may not signal impending crises is that market participants strongly expect the official sector—be it national or international—to bail out a troubled borrower.9 In such cases, interest rate spreads will reflect the creditworthiness of the guarantor—not that of the borrower. Again, it is not difficult to find recent examples where such expectations could well have impaired market signals. In Asian emerging economies, several authors have argued that implicit and explicit guarantees of financial institutions’ liabilities were important in motivating the large net private capital inflows into the region in the 1990s. Others have emphasized that the disciplined fiscal positions of these countries may have convinced investors that, should banks and finance companies experience strains, governments would have the resources to honor their guarantees.10 In the case of the Mexican peso crisis, it has similarly been argued that, after the United States had agreed to the North American Free Trade Agreement, or NAFTA, it would have been very costly for it to stand by while Mexico either devalued the peso or defaulted on its external obligations and that expectations of a US bailout blunted the operation of early warning signals (Leiderman and Thorne 1996; Calvo and Goldstein 1996). Looking eastward, investments in Russian and Ukrainian government securities have in recent years sometimes been known on Wall Street as ‘‘the moral hazard play’’—reflecting the expectation that geopolitical factors and security concerns would lead to a bailout of troubled borrowers. Suffice it to say that the size and frequency of IMF-led international financial rescue packages—including commitments of nearly $50 billion for Mexico in 1994-95; over $120 billion for Thailand, Indonesia, and South Korea in 1997-98; over $25 billion for Russia and Ukraine in 1998; and another $42 billion for Brazil late that year—illustrate that market expectations of official bailouts cannot be dismissed lightly. If interest rate spreads and sovereign credit ratings only give advance warning of financial crises once in a while increased interest attaches to the question of whether there are other early warning indicators that would do a better job and if so, what they might be. This is a key question for this book. 9. Michael P. Dooley has stressed this point in several papers (see Dooley 1997, for instance). 10. See Krugman (1998), Dooley (1997), and Calomiris (1997) on the role of expected national and international bailouts in motivating capital flows and/or banking crises. Zhang (1999), on the other hand, tests for such ‘‘moral hazard’’ effects in private capital flows to emerging markets and finds no evidence for it. Claessens and Glaessner (1997) highlight the link between fiscal positions and the wherewithal to honor explicit and implicit guarantees in the financial sector. The Council on Foreign Relations (1999) offers a set of proposals on how the moral hazard associated with international financial rescue packages might be reduced. 8 ASSESSING FINANCIAL VULNERABILITY Institute for International Economics | http://www.iie.com Organization of the Book Chapter 2 takes up the leading methodological issues surrounding the forecasting of crisis vulnerability, including the choice of sample countries, the definition of currency and banking crises, the selection of leading indicators, the specification of the early warning window, and the signals approach to calculating optimal thresholds for indicators and the probability of a crisis. Chapter 3 presents the main empirical results for the in-sample estimation (1970-95), with a focus on the best-performing monthly and annual indicators, on a comparison of credit ratings and interest rate spreads with indicators of economic fundamentals, and on the ability of the signals approach to predict accurately previous currency and banking crises. In chapter 4, we offer some preliminary results on the track record of rating agencies in forecasting currency and banking crises. In chapter 5, we use two overlapping out-of-sample periods (namely, January 1996 through June 1997 and January 1996 through December 1997) to project which emerging economies were recently the most vulnerable to currency and banking crises. This exercise also permits us to gauge the performance of the model in anticipating the Asian financial crisis. In chapter 6, we analyze the contagion of financial crises across countries, with particular emphasis on how fundamentals-based contagion is influenced by trade and financial sector links. Chapter 7 examines data on the aftermath of crises in order to assess how long it usually takes before recovery from financial crises takes hold. Finally, chapter 8 summarizes our main results and contains some brief concluding remarks, along with suggestions for how the leading-indicator analysis of currency and banking crises in emerging economies might be improved. INTRODUCTION 9 Institute for International Economics | http://www.iie.com 2 Methodology Our approach to identifying early warning indicators of financial crises in emerging economies reflects a number of decisions about the appropriate methodology for conducting such an empirical exercise. Key elements of our thinking are summarized in the following guidelines. General Guidelines First, finding a systematic pattern in the origin of financial crises means looking beyond the last prominent crisis (or group of crises) to a larger sample. Otherwise there is a risk either that there will be too many potential explanations to discriminate between important and less important factors or that generalizations and lessons will be drawn that do not necessarily apply across a wider body of experience.1 We try to guard against these risks by looking at a sample of 87 currency crises and 29 banking crises that occurred in a sample of 25 emerging economies and smaller industrial countries over 1970-95.2 Several examples help to illustrate the point. Consider the last two major financial crises of the 1990s: the 1994-95 Mexican peso crisis and 1. One can also view ‘‘early warning indicators’’ as a way to discipline or check more ‘‘subjective’’ and ‘‘idiosyncratic’’ assessments of crisis probabilities for particular economies—just as more comprehensive, subjective assessments can act as a check on the quality of early warning indicator projections. 2. Our out-of-sample analysis spans 1996-97. Our criteria for defining a currency and a banking crisis is described later in this chapter. 11 Institute for International Economics | http://www.iie.com the 1997-99 Asian financial crisis. Was the peso crisis primarily driven by Mexico’s large current account deficit (equal to almost 8 percent of its GDP in 1994) and by the overvaluation of the peso’s real exchange rate, or by the maturity and composition of Mexico’s external borrowing (too short term and too dependent on portfolio flows), or by the uses to which that foreign borrowing was put (too much for consumption and not enough for investment), or by the already-weakened state of the banking system (the share of nonperforming loans doubled between mid-1990 and mid-1994), or by bad luck (in the form of unfortunate domestic political developments and an upward turn in US international interest rates)? Or was it driven by failure to correct fast enough earlier slippages in monetary and fiscal policies in the face of market nervousness, or by a growing imbalance between the stock of liquid foreign-currency denominated liabilities and the stock of international reserves, or by an expectation on the part of Mexico’s creditors that the US government would step in to bail out holders of tesobonos?3 Analogously, was the Asian financial crisis due to the credit boom experienced by the ASEAN-4 economies (Thailand, Indonesia, Malaysia, and the Philippines), or a concentration of credit in real estate and equities, or large maturity and currency mismatches in the composition of external borrowing, or easy global liquidity conditions, or capital account liberalization cum weak financial sector supervision? Was it the relatively large current account deficits and real exchange rate overvaluations in the run-up to the crisis, a deteriorating quality of investment, increasing competition from China, global overproduction in certain industries important to the crisis countries, or contagion from Thailand?4 There are simply too many likely suspects to draw generalizations from two episodes—even if they are important ones. To tell, for example, whether a credit boom is a better leading indicator of currency crises than are, say, current account deficits, we need to run a horse race across a larger number of currency crises.5 Equally, but operating in the opposite direction, there is a risk of ‘‘jumping the gun’’ by generalizing prematurely about the relative importance of particular indicators from a relatively small set of prominent crises. One example is credit booms—that is, expansions of bank credit that are large relative to the growth of the economy. These have been shown to 3. See Leiderman and Thorne (1996) and Calvo and Goldstein (1996) for an analysis of the Mexican crisis. 4. These alternative explanations of the Asian crisis are discussed in BIS (1998), Corsetti, Pesenti, and Roubini (1998), Goldstein (1998a), Radelet and Sachs (1998), IMF (1997), and World Bank (1998). 5. Some of these explanations, of course, are not mutually exclusive. For example, large current account deficits may be the outcome of financial liberalization and its attendant credit booms. 12 ASSESSING FINANCIAL VULNERABILITY Institute for International Economics | http://www.iie.com forerun banking crises in Japan, in several Scandinavian countries, and in Latin America (Gavin and Hausman 1996). Yet when we compare credit booms as a leading indicator of banking crises to other indicators across a larger group of emerging economies and smaller industrial countries, we find that credit booms are outperformed by a variety of other indicators. Put in other words, credit booms have been a very good leading indicator in some prominent banking crises but are not, on average, the best leading indicator in emerging economies more generally. Again, it is helpful to have recourse to a larger sample of crises (in this study nearly 30) to sort out competing hypotheses. The second guideline is to pay equal attention to banking crises and currency crises. To this point, most of the existing literature on leading indicators of financial crises relates exclusively to currency crises.6 Yet the costs of banking crises in developing countries appear to be greater than those of currency crises. Furthermore, banking crises appear to be one of the more important factors in generating currency crises, and the determinants and leading indicators of banking crises should be amenable to the same type of quantitative analysis as currency crises are.7 Some policymakers have argued that, looking forward, the emphasis in surveillance efforts should be directed to banking sector problems rather than currency crises. The underlying assumption supporting that view is that as more countries adopt regimes of managed floating, currency crises become a relic of the past. We believe this view to be overly optimistic. It is noteworthy that among all the Asian countries that had major currency crises in 1997-98 only Thailand had an ‘‘explicit pegged exchange rate’’ policy. Indonesia, Malaysia, and South Korea were all declared managed floaters, while the Philippines in principle (but not in practice) had a freely floating exchange rate. Among emerging markets, there is widespread ‘‘fear of floating,’’ and many of the countries that are classified as floaters have implicit pegs, leaving them vulnerable to the types of currency crises we study in this book.8 6. See Kaminsky, Lizondo, and Reinhart (1998) for a review of this literature. Among the relatively few studies that include or concentrate on banking crises in emerging economies, we would highlight Caprio and Klingebiel (1996a, 1996b), Demirguc-Kunt and Detragiache ¨¸ (1998), Eichengreen and Rose (1998), Furnam and Stiglitz (1998), Honohan (1997), Gavin and Hausman (1996), Goldstein (1997), Goldstein and Turner (1996), Kaminsky (1998), Kaminsky and Reinhart (1998, 2000), Rojas-Suarez (1998), Rojas-Suarez and Weisbrod (1995), and Sundararajan and Balino (1991). ˜ 7. Both Kaminsky and Reinhart (1998) and the IMF (1998c) conclude that the output costs of banking crises in emerging economies typically exceed those for currency crises and that these costs are greater still during what Kaminsky and Reinhart (1999) dubbed ‘‘twin crises’’ (that is, episodes when the country is undergoing simultaneous banking and currency crises). We provide further empirical evidence on this issue in chapter 7. 8. See Calvo and Reinhart (2000) and Reinhart (2000) for a fuller discussion of this issue. METHODOLOGY 13 Institute for International Economics | http://www.iie.com We analyze banking and currency crises separately, as well as exploring the interactions among them. As it turns out, several of the early warning indicators that show the best performance for currency crises also work well in anticipating banking crises. At the same time, there are enough differences regarding the early warning process and in the aftermath of crises to justify treating each in its own right. A third feature of our approach—and one that differentiates our work from that of many other researchers—is that we employ monthly data to analyze banking crises as well as currency crises.9 Use of monthly (as opposed to annual data) involves a trade-off. On the minus side, because monthly data on the requisite variables are available for a smaller number of countries than would be the case for annual data, the decision to go with higher frequency data may result in a smaller sample. Yet monthly data permit us to learn much more about the timing of early warning indicators, including differences among indicators in the first arrival and persistence of signals. Indeed, many of the annual indicators that have been used in other empirical studies are only publicly available with a substantial lag, which makes them plausible for a retrospective assessment of the symptoms of crises but ill-suited for the task of providing an early warning. Hence, we conclude that the advantages of monthly data seemed to outweigh the disadvantages.10 In the end, we were able to assemble monthly data for about two-thirds of our indicator variables; for the remaining third, we had to settle for annual data. A fourth element of our approach was to include a relatively wide array of potential early warning indicators. We based this decision on a review of broad, recurring themes in the theoretical literature on financial crises. These themes encompass asymmetric information and ‘‘bank run’’ stories that stress liquidity/ currency mismatches and shocks that induce borrowers to run to liquidity or quality, inherent instability and bandwagon theories that emphasize excessive credit creation and unsound finance during the expansion phase of the business cycle, ‘‘premature’’ financial liberalization stories that focus on the perils of liberalization when banking supervision is weak and when an extensive 9. For example, the studies of banking crises in emerging markets by Caprio and Klingebiel (1996a, 1996b), Goldstein and Turner (1996), Honohan (1995), and Sundararajan and Balino ˜ (1991) are primarily qualitative, while the studies by Demirguc-Kunt and Detragiache (1997), ¨¸ Eichengreen and Rose (1998), and the IMF (1998c) use annual data for their quantitative investigation of the determinants of banking crises. 10. Private-sector ‘‘early warning’’ analyses likewise seem to be moving in the direction of using monthly data. See Ades, Masih, and Tenegauzer (1998) and Kumar, Perraudin, and Zinni (1998). 14 ASSESSING FINANCIAL VULNERABILITY Institute for International Economics | http://www.iie.com network of explicit and implicit government guarantees produces an asymmetric payoff for increased risk taking, first- and second-generation models of the vulnerability of fixed exchange rates to speculative attacks,11 and interactions of various kinds between currency and banking crises. In operational terms, this eclectic view of the origins of financial crises translates into a set of 25 leading indicator variables that span the real and monetary sectors of the economy, that contain elements of both the current and capital accounts of the balance of payments, that include market variables designed to capture expectations of future events, and that attempt to proxy certain structural changes in the economy (e.g., financial liberalization) that could affect vulnerability to a crisis. Once a set of potential leading indicators or determinants of banking and currency crises has been selected, a way has to be found both to identify the better performing ones among them and to calculate the probability of a crisis. In most of the existing empirical crisis literature, this is done by estimating a multivariate logit or probit regression model in which the dependent variable (in each year or month) takes the value of one if that period is classified as a crisis and the value of zero if there is no crisis. When such a regression is fitted on a pooled set of country data (i.e., a pooled cross-section of time series), the statistical significance of the estimated regression coefficients should reveal which indicators are ‘‘significant’’ and which are not, and the predicted value of the dependent variable should identify which periods or countries carry a higher or lower probability of a crisis. A fifth characteristic of our approach is that we use a technique other than regression to evaluate individual indicators and to assess crisis vulnerability across countries and over time. Specifically, we adopt the nonparametric ‘‘signals’’ approach pioneered by Kaminsky and Reinhart (1999).12 The basic premise of this approach is that the economy behaves differently on the eve of financial crises and that this aberrant behavior has a recurrent systemic pattern. For example, currency crises are usually preceded by an overvaluation of the currency; banking crises tend to follow sharp declines in asset prices. The signals approach is given diagnostic and predictive content by specifying what is meant by an ‘‘early’’ warning, by defining an ‘‘optimal threshold’’ for each indicator, and by choosing one or more diagnostic statistics that measure the probability of experiencing a crisis. 11. First-generation models stress poor fundamentals as the cause of the currency crises, while second-generation models focus on shifts in market expectations and self-fulfilling speculative attacks. See Flood and Marion (1999) for a recent survey of this literature. 12. This approach is described in detail in Kaminsky, Lizondo, and Reinhart (1998). METHODOLOGY 15 Institute for International Economics | http://www.iie.com By requiring the specification of an explicit early warning window, the signals approach forces one to be quite specific about the timing of early warnings. This is not the case for all other approaches. For example, it has been argued that an asymmetric-information approach to financial crises implies that the spread between low- and high-quality bonds will be a good indicator of whether an economy is experiencing a true financial crisis—but there is no presumption that this interest rate spread should be a leading rather than a contemporaneous indicator (Mishkin 1996). Furthermore, the indicator methodology takes a comprehensive approach to the use of information without imposing too many a priori restrictions that are difficult to justify. Finally, we use the signals to rank the probability of crises both across countries and over time. We do so by calculating the weighted number of indicators that have reached their optimal thresholds (that is, are ‘‘flashing’’), where the weights (represented by the inverse of the individual noise-to-signal ratios) capture the relative forecasting track record of the individual indicators.13 Indicators with good track records receive greater weight in the forecast than those with poorer ones. Ceteris paribus, the greater the incidence of flashing indicators, the higher the presumed probability of a banking or currency crisis. For example, if in mid-1997 we were to find that 18 of 25 indicators were flashing for Thailand versus only 5 of 25 for Brazil, we would conclude that Thailand was more vulnerable to a crisis than Brazil. Analogously, if only 10 of 25 indicators were flashing for Thailand in mid-1993, we would conclude that Thailand was less vulnerable in mid-1993 than it was in mid-1997. Thus we can calculate the likelihood of a crisis on the basis of how many indicators are signaling. Furthermore, as will be shown in chapter 5, we can attach a greater weight to the signals of the more reliable indicators. Owing to these features, the signals approach makes it easy computationally to monitor crisis vulnerability. In contrast, the regression-based approaches require estimation of the entire model to calculate crisis probabilities. In addition, because these regression-based models are nonlinear, it becomes difficult to calculate the contribution of individual indicators to crisis probabilities in cases where the variables are far away from their means.14 13. While this is one of many potential ‘‘composite’’ indicators (i.e., ways of combining the information in the individual indicators), Kaminsky (1998) provides evidence that this weighting scheme shows better in-sample and out-of-sample performance than three alternatives. Also, see chapter 5. One can equivalently evaluate the performance of individual indicators by comparing their conditional probabilities of signaling a crisis. 14. Of course, ease of application is only one of many criteria for choosing among competing crisis-forecasting methodologies. For example, the signals approach also carries the disadvantage that is less amenable to statistical tests of significance. In addition, some of the restrictions it imposes (e.g., that indicators send a signal only when they reach a threshold) may leave out valuable information. 16 ASSESSING FINANCIAL VULNERABILITY Institute for International Economics | http://www.iie.com Guideline number six is to employ out-of-sample tests to help gauge the usefulness of leading indicators. The in-sample performance of a model may convey a misleading sense of optimism about how well it will perform out of sample. A good case in point is the experience of the 1970s with structural models of exchange rate determination for the major currencies. While these models fit well in sample, subsequent research indicated that their out-of-sample performance was no better—and often worse—than that of ‘‘naive’’ models (such as using the spot rate or the forward rate to predict the next period’s exchange rate; see Meese and Rogoff 1983). In this study, we use data from 1970-95 to calculate our optimal thresholds for the indicators, but we save data from 1996 through the end of 1997 to assess the out-of-sample performance of the signals approach, including the ability to identify the countries most affected during the Asian financial crisis. Our seventh and last guideline is to beware of the limitations of this kind of analysis. Because these exercises concentrate on the macroeconomic environment, they cannot capture political triggers and exogenous events—the Danish referendum on the European Economic and Monetary Union (EMU) in 1992, the Colosio assassination in 1994, or the debacle over Suharto in 1997-98, for instance—which often influence the timing of speculative attacks. In addition, because high-frequency data are not available on most of the institutional characteristics of national banking systems—ranging from the extent of ‘‘connected’’ and governmentdirected lending to the adequacy of bank capital and banking supervision—such exercises cannot be expected to capture some of these longerterm origins of banking crises.15 Also, because we are not dealing with structural economic models but rather with loose, reduced-form relationships, such leading-indicator exercises do not generate much information on why or how the indicators affect the probability of a crisis. For example, a finding that exchange rate overvaluation typically precedes a currency crisis does not tell us whether the exchange rate overvaluation results from an exchange rate-based inflation stabilization program or from a surge of private capital inflows. Nor is the early warning study of financial crises immune from the ‘‘Lucas critique’’: that is, if a reliable set of early warning indicators were identified empirically, it is possible that policymakers would henceforth behave differently when these indicators were flashing than they did in the past, thereby transforming these variables into early warning indicators of corrective policy action rather than of financial crisis. While this feedback effect of the indicators on crisis prevention has apparently not yet been strong enough to impair their predictive content, there is no guarantee 15. Indeed, for many countries, detailed data on the state of the banks may not even be available annually. METHODOLOGY 17 Institute for International Economics | http://www.iie.com that this feedback effect will not be stronger in the future (particularly if the empirical evidence in favor of robust early warning indicators was subsequently viewed as more persuasive). Much like the leading-indicator analysis of business cycles, we are engaging here in a mechanical exercise—albeit one that we think is interesting on a number of fronts. Moreover, this research is still in its infancy, with many of the key empirical contributions coming only in the last two to three years. In areas such as the modeling of contagion and alternative approaches to out-of-sample forecasting, too few ‘‘horse races’’ have been run to know which approaches work best. For all of these reasons, we see the leading-indicator analysis of financial crises in emerging economies as one among a number of analytical tools and not as a stand-alone, surefire system for predicting where the next crisis will take place. That being said, we also argue that this approach shows promising signs of generating real value added and that it appears particularly useful as a first screen for gauging the ordinal differences in vulnerability to crises both across countries and over time. A family of estimated conditional crisis probabilities will provide the basis of this ordinal ranking across countries at a point in time or for a given country over time. Putting the Signals Approach to Work The signals approach described above was first used to analyze the performance of macroeconomic and financial indicators around ‘‘twin crises’’ (i.e., the joint occurrences of currency and banking crises) in Kaminsky and Reinhart (1999). We focus on a sample of 25 countries over 1970 to 1995. The out-of-sample performance of the signals approach will be assessed using data for January 1996 through December 1997. These are the countries in our sample: Africa: South Africa Asia: Indonesia, Malaysia, the Philippines, South Korea, Thailand Europe and the Middle East: Czech Republic, Denmark, Egypt, Finland, Greece, Israel, Norway, Spain, Sweden, Turkey Latin America: Argentina, Bolivia, Brazil, Chile, Colombia, Mexico, Peru, Uruguay, Venezuela The basic premise of the signals approach is that the economy behaves differently on the eve of financial crises and that this aberrant behavior has a recurrent systematic pattern. This ‘‘anomalous’’ pattern, in turn, is manifested in the evolution of a broad array of economic and financial indicators. The empirical evidence provides ample support for this prem18 ASSESSING FINANCIAL VULNERABILITY Institute for International Economics | http://www.iie.com ise.16 To implement the signals approach, we need to clarify a minimum number of two key concepts which will be used throughout the analysis. Currency Crisis A currency crisis is defined as a situation in which an attack on the currency leads to substantial reserve losses, or to a sharp depreciation of the currency—if the speculative attack is ultimately successful—or to both. This definition of currency crisis has the advantage of being comprehensive enough to capture not only speculative attacks on fixed exchange rates (e.g., Thailand’s experience before 2 July 1997) but also attacks that force a large devaluation beyond the established rules of a crawling-peg regime or an exchange rate band (e.g., Indonesia’s widening of the band before its floatation of the rupiah on 14 August 1997.) Since reserve losses also count, the index also captures unsuccessful speculative attacks (e.g., Argentina’s reserve losses in the wake of the Mexican 1994 peso crisis.) We constructed an index of currency market turbulence as a weighted average of exchange rate changes and reserve changes.17 Interest rates were excluded, as many emerging markets in our sample had interest rate controls through much of the sample. The index, I, is a weighted average of the rate of change of the exchange rate, e/e, and of reserves, R/R, with weights such that the two components of the index have equal sample volatilities: I ( e/e) ( e / R) * ( R/R) (2.1) where e is the standard deviation of the rate of change of the exchange rate and R is the standard deviation of the rate of change of reserves. Since changes in the exchange rate enter with a positive weight and changes in reserves have a negative weight attached, readings of this index that were three standard deviations or more above the mean were cataloged as crises.18 For countries in the sample that had hyperinflation, the construction of the index of currency market turbulence was modified. While a 100 percent devaluation may be traumatic for a country with low to moderate inflation, a devaluation of that magnitude is commonplace during hyperinflation. A single index for the countries that had hyperinflation episodes would miss sizable devaluations and reserve losses in the moderate infla16. See Kaminsky, Lizondo, and Reinhart (1998) for a survey of this literature. 17. This index is in the spirit of that used by Eichengreen, Rose, and Wyplosz (1996), who also included interest rate increases in their measure of turbulence. 18. Of course, for a study of market turbulence as well as crisis, one may wish to consider readings in this index that are two standard deviations away from the mean. METHODOLOGY 19 Institute for International Economics | http://www.iie.com tion periods because the high-inflation episodes would distort the historic mean. To avoid this, we divided the sample according to whether inflation in the previous six months was higher than 150 percent and then constructed an index for each subsample.19 As noted in earlier studies that use the signals approach, the dates of currency crises derived from this index map well onto the dates that would be obtained if one were to define crises by relying exclusively on events, such as the closing of the exchange markets or a change in the exchange rate regime. Banking Crises Our dating of banking crises stresses events. This is because on the banking side there are no time series comparable to international reserves and the exchange rate. For instance, in the banking panics of an earlier era large withdrawals of bank deposits could be used to date the crisis. In the wake of deposit insurance, however, bank deposits ceased to be useful for dating banking crises. As Japan’s banking crisis highlights, many modern financial crises stem from the asset side of the balance sheet, not from deposit withdrawals. Hence the performance of bank stocks relative to the overall equity market could be an indicator. Yet in many of the developing countries an important share of the banks are not traded publicly. Large increases in bankruptcies or nonperforming loans could also be used to mark the onset of the crisis. Indicators of business failures and nonperforming loans are, however, usually available only at low frequencies, if at all; the latter are also made less informative by banks’ desire to hide their problems for as long as possible. Given these data limitations, we mark the beginning of a banking crisis by two types of events: bank runs that lead to the closure, merging, or takeover by the public sector of one or more financial institutions (as in Venezuela in 1993); and if there are no runs, the closure, merging, takeover, or large-scale government assistance of an important financial institution (or group of institutions) that marks the start of a string of similar outcomes for other financial institutions (as in Thailand in 1997). We rely on existing studies of banking crises and on the financial press; according to these studies the fragility of the banking sector was widespread during these periods. Our approach to dating the onset of the banking crises is not without drawbacks. It could date the crises ‘‘too late’’ because the financial problems usually begin well before a bank is finally closed or merged. It could also date crises ‘‘too early’’ because the worst of crisis may come later. 19. Similar results are obtained by looking at significant departures in inflation from a 6and 12-month moving average. 20 ASSESSING FINANCIAL VULNERABILITY Institute for International Economics | http://www.iie.com To address this issue we also indicate when the banking crisis hits its peak, defined as the period with the heaviest government intervention and/or bank closures. Identifying the end of a banking crisis is one of the more difficult unresolved problems in the empirical crisis literature—that is, there is no consensus on what the criteria ought to be for declaring the crisis to be over (e.g., resumption of normal bank lending behavior, or a marked decrease in the share of nonperforming loans, or an end to bank closures and large-scale government assistance). In our discussion of the aftermath of crises in chapter 7, however, the end of a banking crisis is understood to be its resolution (i.e., the end of heavy government financial intervention), not when bank balance sheets cease to deteriorate. Other empirical studies on banking crises have focused on annual data and provide no information on the month or quarter in which banking sector problems surface. Hence it is not possible to compare the exact dates with our own analysis. We can, however, compare the dating of the year of the crisis. In most cases, our dates for the beginning of crises correspond with those found in other studies, but there are several instances where our starting date is a year earlier than theirs. Tables 2.1 and 2.2 list the currency and banking crisis dates, respectively, for the 25 countries in our sample. The Indicators In addition to the 15 early warning indicators originally considered in Kaminsky and Reinhart (1999), we evaluate the ability of nine additional indicators that figure prominently in both the theoretical literature on banking and currency crises and in the popular discussion of these events. The indicators used in Kaminsky and Reinhart (1999) were international reserves (in US dollars), imports (in US dollars), exports (in US dollars), the terms of trade (defined as the unit value of exports over the unit value of imports), deviations of the real exchange rate from trend (in percentage terms),20 the differential between foreign (US or German) and domestic real interest rates on deposits (monthly rates, deflated using consumer prices and measured in percentage points), ‘‘excess’’ real M1 balances, the money multiplier (of M2), the ratio of domestic credit to GDP, the real interest rate on deposits (monthly rates, deflated using consumer prices and measured in percentage points), the ratio of (nominal) lending 20. The real exchange rate is defined on a bilateral basis with respect to the German mark for the European countries in the sample and with respect to the US dollar for all other countries. The real exchange rate index is defined such that an increase in the index denotes a real depreciation. METHODOLOGY 21 Institute for International Economics | http://www.iie.com Table 2.1 Currency crisis starting dates Country Argentina Currency crisis June 1975 February 1981* July 1982 September 1986* April 1989 February 1990 November 1982 November 1983 September 1985 February 1983 November 1986* July 1989 November 1990 October 1991 December 1971 August 1972 October 1973 December 1974 January 1976 August 1982* September 1984 March 1983* February 1985* May 1997 May 1971 June 1973 November 1979 August 1993 January 1979 August 1989 June 1990 June 1973 October 1982 November 1991* September 1992* May 1976 November 1980 July 1984 November 1978 April 1983 September 1986 August 1997 November 1974 November 1977 October 1983* July 1984 July 1975 August 1997* (continued next page) 22 ASSESSING FINANCIAL VULNERABILITY Bolivia Brazil Chile Colombia Czech Republic Denmark Egypt Finland Greece Indonesia Israel Malaysia Institute for International Economics | http://www.iie.com Table 2.1 (continued) Country Mexico Currency crisis September 1976 February 1982* December 1982* December 1994* June 1973 February 1978 May 1986* December 1992 June 1976 October 1987 February 1970 October 1983* June 1984 July 1997* September 1975 July 1981 July 1984 May 1996 June 1971 December 1974 January 1980 October 1997 February 1976 July 1977* December 1982 February 1986 September 1992 May 1993 August 1977 September 1981 October 1982 November 1992* November 1978* July 1981 November 1984 July 1997* August 1970 January 1980 March 1994* December 1971* October 1982* February 1984 December 1986 March 1989 May 1994* December 1995 Norway Peru The Philippines South Africa South Korea Spain Sweden Thailand Turkey Uruguay Venezuela * twin crises METHODOLOGY 23 Institute for International Economics | http://www.iie.com Table 2.2 Banking crisis starting dates Country Argentina K & R (1999) and G, K, & R (beginning) March 1980 May 1985 December 1994 Bolivia Brazil Chile September 1981 Colombia Czech Republic Denmark Egypt Finland Greece Indonesia Israel Malaysia July 1982 April 1998 1994 March 1987 January 1980 January 1990 September 1991 1991 November 1992 October 1983 July 1985 September 1997 1982 n.a. n.a. 1980 1990 1991 n.a. 1994 1977 1985 October 1987 November 1985 December 1994 C&K (1996) 1980 1985 1995 1986 1994 IMF (1996 and 1998a & b) 1980 1985 1989 1995 n.a. 1990 1994 1976 1981 1982 n.a. 1988 1981 1990 1991 n.a. 1992 1997 1983 1985 (continued next page) to deposit interest rates,21 the stock of commercial banks’ deposits (in nominal terms), the ratio of broad money (converted into foreign currency) to gross international reserves, an index of output, and an index of equity prices (in US dollars). All these series are monthly. For greater detail, see the appendix. The links between particular early warning indicators and underlying theories of exchange rate and banking crises are discussed in some detail in earlier papers (e.g., Kaminsky and Reinhart 1999). Turning to the nine ‘‘new’’ indicators introduced here, four of them are expressed as a share of GDP. These are the current account balance, short-term capital inflows, foreign direct investment, and the overall bud21. This definition of the spread between lending and deposit rates is preferable to using merely the difference between nominal lending and deposit rates because inflation affects this difference and thus the measure would be distorted in the periods of high inflation. An alternative would have been to use the difference between real lending and deposit rates. 24 ASSESSING FINANCIAL VULNERABILITY Institute for International Economics | http://www.iie.com Table 2.2 (continued) Country Mexico Norway Peru Philippines South Africa South Korea Spain Sweden Thailand Turkey January 1991 Uruguay Venezuela October 1993 n.a. not applicable K&R Kaminsky and Reinhart (1999) G, K, & R Goldstein, Kaminsky, and Reinhart C&K Caprio and Klingebiel (1996b) March 1971 March 1981 1992 1994 1981 1980 1994 K & R (1999) and G, K, & R (beginning) September 1982 October 1992 November 1988 March 1983 January 1981 July 1997 December 1977 January 1986 July 1997 November 1978 November 1991 March 1979 May 1996 C&K (1996) 1981 1995 1987 n.a. 1981 1977 n.a. 1977 1991 1983 IMF (1996 and 1998a & b) 1982 1994 1987 1983 1981 1980 1983 1997 1977 1990 1983 1997 1982 1991 1994 1981 1980 1993 get deficit. In addition, we look at the growth rates in the following variables (the first three as shares in GDP and the fourth as a share of investment): general government consumption, central bank credit to the public sector, net credit to the public sector, and the current account balance. The latter measure of the current account was motivated by the view, particularly popular in the wake of the 1994-95 Mexican peso crisis, that large current account deficits are more of a concern if they stem from low saving as opposed to high levels of investment. Recent events in Asia—a region noted for its exceptionally high levels of domestic saving and its even higher levels of investment—have led to a reassessment of that view. We also look at two measures of sovereign credit ratings. As most of the new indicators are not available at monthly or quarterly frequencies, annual data were used. Table 2.3 provides a list of the indicators we examine in this book, their periodicity, and the transformation used. In chapter 4, we examine the METHODOLOGY 25 Institute for International Economics | http://www.iie.com Table 2.3 Selected leading indicators of banking and currency crises Indicator Real output Equity prices International reserves Domestic/foreign real interest rate differential Excess real M1 balances M2/ international reserves Bank deposits M2 multiplier Domestic credit/GDP Real interest rate on deposits Ratio of lending interest rate to deposit interest rate Real exchange rate Exports Imports Terms of trade Moody’s sovereign credit ratings Institutional Investor sovereign credit ratings General government consumption/GDP Overall budget deficit/GDP Net credit to the public sector/GDP Central bank credit to public sector/GDP Short-term capital inflows/GDP Foreign direct investment/GDP Current account imbalance/GDP Current account imbalance/investment Transformation 12-month growth rate 12-month growth rate 12-month growth rate Level Level 12-month growth 12-month growth 12-month growth 12-month growth Level Level Data frequency Monthly Monthly Monthly Monthly Monthly Monthly Monthly Monthly Monthly Monthly Monthly Monthly Monthly Monthly Monthly Monthly Semiannual Annual Annual Annual Annual Annual Annual Annual Annual rate rate rate rate Deviation from trend 12-month growth rate 12-month growth rate 12-month growth rate 1-month change Semiannual change Annual growth rate Level Level Level Level Level Level Level track record of sovereign credit ratings when it comes to ‘‘predicting’’ financial crises. Specifically, we examine the performance of the Institutional Investor and Moody’s ratings. As noted, in most cases we focus on 12-month changes in the variables. This transformation has several appealing features. First, it eliminates the nonstationarity problem of the variables in levels. It also makes the indicators more comparable across countries and across time. Some of the indicators have a strong seasonal pattern, which the 12-month transformation corrects for. For some indicators, such as equity prices, one could contemplate using a measure of under- or overvaluation. However, the empirical performance of most asset pricing models is not strong enough to justify such an exercise. For the monthly variables (with the exception of the deviation of the real exchange rate from trend, the ‘‘excess’’ of real M1 balances, and the three variables based on interest rates), the indicator on a given month was defined as the percentage change in the level of the variable with respect to its level a year earlier. This filter has several attractive features: it reduces the ‘‘noisiness’’ of working with monthly data, it facilitates cross-country comparisons, and it ensures the variables are stationary with well-defined moments. 26 ASSESSING FINANCIAL VULNERABILITY Institute for International Economics | http://www.iie.com Turning to credit ratings, Institutional Investor constructs an index that rises with increasing country creditworthiness and ranges from 0 to 100; this index is published twice a year and is released in March and September.22 Hence we work with the six-month percentage change in this rating index. For Moody’s Investor services, monthly changes in the sovereign ratings are used. A downgrade takes on the value of minus one; no change in the rating takes on a value of zero, and an upgrade takes on the value of one. Since Moody’s ratings take on values from 1 to 16, we also worked with changes in the ratings that took into account the magnitude of the change. This issue will be discussed in greater detail in chapter 4. The Signaling Window Let us call a signal (yet to be precisely defined) a departure from ‘‘normal’’ behavior in an indicator.23 For example, an unusually large decline in exports or output may signal a future currency or banking crisis. If an indicator sends a signal that is followed by a crisis within a plausible time frame we call it a good signal. If the signal is not followed by a crisis within that interval, we call it a false signal, or noise. The signaling window for currency crises is set a priori at 24 months preceding the crisis. If, for instance, an unusually large decline in exports were to occur 28 months before the crisis, the signal would fall outside the signaling window and would be labeled a false alarm. Alternative signaling windows (18 months and 12 months) were considered as part of our sensitivity analysis. While the results for the 18-month window yielded similar results to those reported in this book, the 12month window proved to be too restrictive. Specifically, several of the indicators we use here, including real exchange rates and credit cycles, signaled relatively early (consistent with a protracted cycle), and the shorter 12-month window penalized those early signals by labeling them as false alarms. For banking crises, we employ a different signaling window. Namely, any signal given in the 12 months preceding the beginning of the crisis or the 12 months following the beginning of the crisis is labeled a good signal. The more protracted nature of banking crises and the high incidence of denial by both bankers and policymakers that there are problems in the banking sector motivate the more forgiving signaling window for banking crises. 22. Since there are two readings of this index per year, in a typical year, say 1995, we would have the percentage change in the rating from September 1994 to March 1995, from March 1995 to September 1995, and the change from September 1995 to March 1996. 23. Of course, normal behavior may change over time, hence, this approach, like other commonly used alternatives (such as logit or probit) is not free from Lucas-critique limitations. For further discussion of this issue, see Kaminsky and Reinhart (1999). METHODOLOGY 27 Institute for International Economics | http://www.iie.com The Threshold Suppose we wish to test the null or maintained hypothesis that the economy is in a ‘‘state of tranquility’’ versus the alternative hypothesis that a crisis will occur sometime in the next 24 months. Suppose that we wish to test this hypothesis on an indicator-by-indicator basis. As in any hypothesis test, this calls for selecting a threshold or critical value that divides the probability distribution of that indicator into a region that is considered normal or probable under the null hypothesis and a region that is considered aberrant or unlikely under the null hypothesis—the rejection region. If the observed outcome for a particular variable falls into the rejection region, that variable is said to be sending a signal. To select the optimal threshold for each indicator, we allowed the size of the rejection region to oscillate between 1 percent and 20 percent. For each choice, the noise-to-signal ratio was tabulated and the ‘‘optimal’’ set of thresholds was defined as the one that minimized the noise-to-signal ratio—that is, the ratio of false signals to good signals.24 Table 2.4 lists the thresholds for all the indicators for both currency and banking crises. For instance, the threshold for short-term capital flows as a percentage of GDP is 85 percent. This conveys two kinds of information. First, it indicates that 15 percent of all the observations in our sample (for this variable) are considered signals. Second, it highlights that the rejection region is located at the upper tail of the frequency distribution, meaning that a high ratio of short-term capital inflows to GDP will lead to a rejection of the null hypothesis of tranquility in favor of the alternative hypothesis that a crisis is brewing. While the threshold or percentile that defines the size of the rejection region is uniform across countries for each indicator, the corresponding country-specific values are allowed to differ. Consider the following illustration. There are two countries, one which has received little or no shortterm capital inflow (as a percentage of GDP) during the entire sample, while the second received substantially larger amounts (also as a share of GDP). The 85th percentile of the frequency distribution for the low capital importer may be as small as a half a percent of GDP and any increase beyond that would be considered a signal. Meanwhile, the country where the norm was a higher volume of capital inflows is likely to have a higher critical value; hence only values above, say 3 percent of GDP, would be considered signals. 24. For variables such as international reserves, exports, the terms of trade, deviations of the real exchange rate from trend, commercial bank deposits, output, and the stock market index, for which a decline in the indicator increases the probability of a crisis, the threshold is below the mean of the indicator. For the other variables, the threshold is above the mean of the indicator. 28 ASSESSING FINANCIAL VULNERABILITY Institute for International Economics | http://www.iie.com Table 2.4 Optimal thresholds (percentile) Indicator Bank deposits Central bank credit to the public sector Credit rating (Institutional Investor) Current account balance/GDP Current account balance/investment Domestic credit/GDP Interest rate differential Excess M1 balances Exports Foreign direct investment/GDP General government consumption/GDP Imports Lending-deposit interest rate ratio M2 multiplier M2/reserves Net credit to the public sector/GDP Output Overall budget deficit/GDP Real exchange ratea Real interest rate Reserves Short-term capital inflows/GDP Stock prices Terms of trade Currency crisis 15 90 11 20 15 88 89 89 10 16 90 90 88 89 90 88 10 10 10 88 10 85 15 10 Banking crisis 20 90 11 14 10 90 81 88 10 12 88 80 87 90 90 80 14 14 10 80 20 89 10 19 a. An increase in the index denotes a real depreciation. Table 2.5 illustrates the ‘‘custom tailoring’’ of the optimal threshold by showing the country-specific critical values for export growth and annual stock returns for Malaysia, Mexico, and Sweden. A 25 percent decline in stock prices would be considered a signal of a future currency crisis in Malaysia and Sweden but not in Mexico, with the latter’s far greater historical volatility.25 Figure 2.1 provides another illustration of the country-specific nature of the optimal threshold calculations. It shows for the entire sample our measure of the extent of overvaluation in the real exchange rate for Mexico. The horizontal line is the country-specific threshold, and a reading below this line (recall that a decline represents an appreciation) represents a signal. The shaded areas are the 24 months before the crisis, or the signaling window. Around 1982 the shaded area is wider due to the fact that there was a ‘‘double dip,’’ with two crises registering. If the indicator crossed the horizontal line and no crisis ensued in the following 24 months, 25. Indeed, as shown in Kaminsky and Reinhart (1998), the volatility pattern for these three countries is representative of the broader historical regional pattern. The wild gyrations in financial markets in Asia in 1997-99, however, may be unraveling those historic patterns. METHODOLOGY 29 Institute for International Economics | http://www.iie.com Table 2.5 Examples of country-specific thresholds: currency crises Country Malaysia Mexico Sweden Critical value for exports (12-month percentage change) 9.05 13.10 11.25 Critical value for stock prices (12-month percentage change) 15.20 38.30 20.78 as it did in early 1992, it is counted as a false alarm. In the remainder of this section we will define these concepts more precisely. Signals, Noise, and Crises Probabilities A concise summary of the possible outcomes is presented in the following two-by-two matrix (for a currency crisis). Crisis occurs in the following 24 months Signal No signal A C No crisis occurs in the following 24 months B D A perfect indicator would only have entries in cells A and D. Hence, with this matrix we can define several useful concepts that we will use to evaluate the performance of each indicator. If one lacked any information on the performance of the indicators, it is still possible to calculate, for a given sample, the unconditional probability of crisis, P(C) (A C)/(A B C D) (2.2) If an indicator sends a signal and that indicator has a reliable track record, then it can be expected that the probability of a crisis, conditional on a signal, P(C/S), is greater than the unconditional probability. Where P(C S) A/(A B) (2.3) Formally, P(C S) P(C) 0 (2.4) The intuition is clear: if the indicator is not ‘‘noisy’’ (prone to sending false alarms), then there are relatively few entries in cell B and P(C S) 1. This is one of the criteria that we will use to rank the indicators in the following chapters. 30 ASSESSING FINANCIAL VULNERABILITY Institute for International Economics | http://www.iie.com Figure 2.1 Mexico: real exchange rate, 1970-96 100 average of the sample Institute for International Economics | http://www.iie.com 31 We can also define the noise-to-signal ratio, N/S, as N/S [B/(B D)]/[A/(A C)] (2.5) It may be the case that an indicator has relatively few false alarms in its track record. This could be the result of the indicator issuing signals relatively rarely. In this case, there is also the danger that the indicator misses the crisis altogether (it does not signal and there is a crisis). In this case, we also wish to calculate for each indicator the proportion of crises accurately called, PC C/(A C). (2.6) In the next chapter, we employ these concepts to provide evidence on the relative merits of a broad range of indicators in anticipating crises. 32 ASSESSING FINANCIAL VULNERABILITY Institute for International Economics | http://www.iie.com 3 Empirical Results The signals approach was applied to the indicators around the dates of the 29 banking and the 87 currency crises. In what follows, we first compare our results for the 15 monthly indicators to those presented in Kaminsky and Reinhart (1999) and reproduced in table 3.1. In addition to presenting our in-sample findings, this exercise allows us to gauge robustness of the signals approach, since the results reported here are derived from a larger sample of countries (25 versus 20.)1 Moreover, in this chapter we report results for many of the indicators that have been stressed in the financial press surrounding the coverage of the Asian crisis. The Monthly Indicators: Robustness Check Tables 3.1 and 3.2 summarize the in-sample performance of the monthly indicators along the lines described in chapter 2 and presented in Kaminsky, Lizondo, and Reinhart (1998) and Kaminsky (1998). Table 3.1 covers banking crises, and table 3.2 presents the results for currency crises. The variables are shown in descending order based on their marginal predictive power. For banking crises, for instance, the real exchange rate has the greatest predictive power and imports the least. For each indicator, the first column of the tables shows the noise-to-signal ratio. An indicator with a noise-to-signal ratio of unity, such as those in the bottom of the 1. The five countries included here that were not a part of the Kaminsky and Reinhart (1999) sample are the Czech Republic, Egypt, Greece, South Africa, and South Korea. 33 Institute for International Economics | http://www.iie.com Table 3.1 Ranking the monthly indicators: banking crises Noisetosignal 0.35 0.46 0.46 0.54 0.68 0.68 0.73 0.73 0.84 0.88 0.89 0.92 1.01 1.48 1.75 Percent of crises accurately called 52 76 63 90 79 96 100 64 72 44 46 83 92 56 64 Difference Rank in in rank Kaminsky ( denotes an (1998) improvement) 1 3 4 5 7 6 8 9 10 13 11 12 14 15 16 0 0 0 0 1 1 0 0 0 2 1 1 0 0 0 Indicator Real exchange rate Stock prices M2 multiplier Output Exports Real interest rate Real interest rate differential Bank deposits M2/reserves Excess real M1 balances Domestic credit/ nominal GDP Reserves Terms of trade Lending-deposit interest rate Imports P(C S) 24.0 23.4 18.3 17.3 14.3 16.8 15.6 12.9 11.4 11.0 10.9 10.7 11.6 8.3 6.0 P(C S) P(C) 14.1 11.2 9.0 7.2 4.7 4.2 3.7 3.1 1.7 1.2 1.1 0.8 0.1 3.5 4.1 Sources: The authors and Kaminsky (1998). tables, issues as many false alarms as good signals. The second column shows the percent of crises (for which there were data for that indicator) accurately called, while the third column lists the probability of a crisis conditional on a signal from the indicator, P(C S). The fourth column shows the difference between the conditional and unconditional probabilities, P(C S) P(C), the fifth column shows the ranking that the indicator received in the previous signals approach analysis, and the last column calculates the difference between its current and previous rank. Hence, a 3 in the last column would mean that the indicator moved up three notches as the sample was enlarged, while a 2 would reflect a decline in its ranking. The indicators’ rankings based on their marginal predictive power are shown under the heading P(C S) P(C). The better the indicator, the higher the probability of crisis conditioned on its signaling—that is, the higher the P(C S)—and the bigger the gap between the conditional probability (P(C S) and the unconditional probability P(C). The unconditional probability of a banking crisis (not shown) varies slightly from indicator to 34 ASSESSING FINANCIAL VULNERABILITY Institute for International Economics | http://www.iie.com Table 3.2 Ranking the monthly indicators: currency crises Percent Noise- of crises toaccurately signal called 0.22 0.32 0.46 0.51 0.51 0.57 0.57 0.58 0.59 0.68 0.74 0.77 0.87 1.00 1.32 1.32 58 66 80 75 71 57 72 72 57 77 89 59 86 63 43 Difference in Ranking rank in K & R ( denotes an (1999) improvement) 1 2 4 3 5 6 7 8 9 10 11 12 14 12 16 15 0 0 1 1 0 0 0 0 0 0 0 0 1 1 1 1 Indicator Real exchange rate Banking crisis Stock prices Exports M2/reserves Output Excess real M1 balances Reserves M2 multiplier Domestic credit/ nominal GDP Terms of trade Real interest rate Imports Real interest rate differential Lending-deposit interest rate Bank deposits K&R P(C S) 62.1 46.0 47.6 42.4 42.3 43.0 40.1 38.9 39.2 35.6 35.4 32.0 30.1 26.1 24.4 22.3 P(C S) P(C) 35.2 17.0 18.3 15.0 14.9 12.5 12.3 12.2 11.6 8.3 6.5 5.5 2.9 0.1 4.8 5.2 Kaminsky and Reinhart (1999). Sources: The authors and Kaminsky and Reinhart (1999). indicator because of differences in data availability, since not all indicators span the entire sample.2 For some indicators the sample is such that the incidence of banking crises (i.e., their unconditional probability) is as low as 9.8 percent or as high as 12 percent. For currency crises, the unconditional probability is clustered in the 27 to 29 percent range. Several interesting features stand out from tables 3.1 and 3.2. First, the ranking of the indicators appears to be quite robust across sample selections, as shown in the last column of table 3.1. In other words, the results from the 25-country sample closely match the results of the 20-country sample. For currency crises, none of the monthly indicators changes in relative performance by more than one position as the sample is enlarged, and for 10 of the indicators, there is no change at all. For 2. As shown in Kaminsky, Lizondo, and Reinhart (1998), the bigger the gap between the conditional probability (P(C S) and the unconditional probability P(C), the lower the noiseto-signal ratio. EMPIRICAL RESULTS 35 Institute for International Economics | http://www.iie.com banking crises, the maximum ranking change is two positions and 10 of the monthly indicators show no change in their relative ranking. This is a positive factor for the expected out-of-sample usefulness of the signals approach. Specifically, it suggests that the indicators could well have a similar relative predictive ability for countries that are not included in the sample.3 Second, some of the most reliable indicators are the same for banking and currency crises. Deviations of the real exchange rate from trend and stock prices stand out in this regard. Close runners-up are output and exports. A similar statement applies to the least useful indicators; imports and the lending-deposit ratios, for example, do not have any predictive ability for either type of crisis. Several of the low-scoring indicators also carry the weakest or most ambiguous theoretical rationale.4 Third, there are some important differences in the ranking of indicators between currency and banking crises. This suggests that currency and banking sector vulnerability takes on different forms. A case in point is the ratio of M2 (in dollars) to foreign exchange reserves, a variable stressed by Calvo and Mendoza (1996) as capturing the extent of unbacked implicit government liabilities. It does quite well (ranks fifth) among the 16 indicators of currency crises, but it is far less useful when it comes to anticipating banking crises. Similarly, the money multiplier, real interest rates, and bank deposits are of little use when it comes to predicting currency crises but do much better in predicting vulnerability to banking crises. This result should not come as a surprise. Both the money multiplier and real interest rates are strongly linked to financial liberalization, which itself helps predict banking crises. As shown in Galbis (1993), real interest rates tend to increase substantially in the wake of financial liberalization. Furthermore, the steep reductions in reserve requirements that usually accompany financial liberalization propel increases in the money multi- 3. We did not included the larger industrial countries (particularly the G-7 countries) in our sample because they have characteristics (such as the ability to borrow in their own currency, a relatively good external-debt servicing history, and high access to private capital markets) that on a priori grounds would seem to make their crisis vulnerability different from that of most emerging economies. In addition, data constraints, extremely large structural shifts over time, and difficulties associated with identifying a ‘‘normal’’ period led to the decision to exclude China, Russia, and most of the transitional economies from the sample. Finally, we excluded low-income developing countries from the sample because we wanted to concentrate on emerging economies that had (in addition to the requisite data availability) significant involvement with private international capital markets. In the end, however, one can only tell whether our sample selection results in certain biases by doing further robustness checks on alternative samples of countries. 4. For instance, lending-deposit interest rate spreads could widen in advance of a crisis due to a deterioration in loan quality or a worsening in adverse selection problems. Alternatively, it could be persuasively argued that ahead of financial crises, banks may be forced to offer higher deposit rates, so as to stem capital flight. 36 ASSESSING FINANCIAL VULNERABILITY Institute for International Economics | http://www.iie.com plier. Bank runs and deposit withdrawals are at the heart of multipleequilibriums explanations of banking crises (Diamond and Dybvig 1983) yet figure less prominently in explanations of currency crises.5 Lastly, banking crises are even more of a challenge to predict than currency crises. For currency crises, the marginal predictive power of 12 of the 16 indicators (column five) is 5 percent or higher; for the real exchange rate, marginal predictive power goes as high as 35 percent. Indeed 9 of the 16 indicators have marginal predictive power in excess of 10 percent. By way of contrast, for banking crisis 11 of the 15 indicators have marginal predictive power of less than 5 percent, and even the topranked macroeconomic indicators have marginal predictive power of less than 15 percent. This relative inability of indicators to anticipate crises in sample may be due to two factors. For one thing, for the earlier part of ` the sample, banking crises were still relatively rare vis-a-vis currency crises—there is a large discrepancy between the number of currency and banking crises studied here. Detecting recurring patterns becomes more difficult in the smaller sample of banking crises. Also, pinning down the timing of a banking crisis requires a tricky judgment about when bankingsector ‘‘distress’’ turns into a full-fledged crisis. As discussed in chapter 2, the timing of currency crises is more straightforward. The empirical evidence on the ‘‘predictability’’ of banking crises is still limited to a handful of studies. Some have followed the approach pioneered by Blanco and Garber (1986) for currency crises and have attempted to model the probability of banking crises on the basis of domestic and external fundamentals. These studies have encountered some of the same problems highlighted in table 3.1—specifically, the relatively poor predictive power of the models. Moreover, the results in the studies sometimes conflict with one another. Eichengreen and Rose (1998), for example, find that external conditions, specifically international interest rates, play an important role in predicting banking crises. Real exchange rate overvaluations, growth, and budget deficits have predictive power in their regressions. The composition of external debt also seems to matter. Other variables, including credit growth, they conclude, have little or no predictive ability. In contrast, Demirguc-Kunt and Detragiache ¨¸ (1998) find no evidence in favor of budget deficits, while real interest rates, credit growth, and M2/reserves figure prominently among their significant regressors. Both studies do find, however, that slower economic growth increases the probability of a banking crisis. In any case, it appears that, to improve upon the ability to predict banking crises, we may need to look beyond macroeconomic indicators—an issue that we take up later. ´ 5. However, some recent models (Goldfajn and Valdes 1995) have highlighted the role of bank runs in precipitating currency crises. EMPIRICAL RESULTS 37 Institute for International Economics | http://www.iie.com Table 3.3 Annual indicators: banking crises Percent of crises accurately called 43 38 52 33 23 15 24 15 Indicator Short-term capital inflows/GDP Current account balance/ investment Overall budget deficit/GDP Current account balance/GDP Central bank credit to the public sector/GDP Net credit to the public sector/GDP Foreign direct investment/GDP General government consumption/ GDP Noise-tosignal 0.38 0.38 0.47 0.50 0.52 0.72 1.05 1.44 P(C S) 36.8 36.1 26.9 29.3 23.8 18.3 15.6 10.0 P(C S) P(C) 18.5 18.4 12.1 12.1 7.6 4.5 0.6 3.8 The Annual Indicators: What Works? Tables 3.3 and 3.4 present evidence on the performance of eight annual indicators that have been prominent in recent discussions of the causes of financial crises. The indicators include the fiscal variables stressed in the Krugman (1979) model of a currency crisis as well as the short-term debt exposure indicators stressed in recent theoretical and empirical explanations of the Asian crisis (Calvo 1998; Calvo and Mendoza 1996; Goldstein 1998b; Radelet and Sachs 1998). As before, the indicators are ranked according to their marginal predictive power. The first column provides information on the noise-to-signal ratio, the second column lists the percent of crises accurately called, the third column provides information on the probability of crisis conditional on signaling, while the last column provides information on the marginal predictive power of the variable. The top indicator for banking crises is the share of short-term capital inflows to GDP. This is consistent with the results in Eichengreen and Rose (1997) and supports the view that the banking sector becomes particularly vulnerable during cycles of short-term capital inflows. Such short-term inflows are more likely to be intermediated through the domestic banking sector than other types of capital flows, such as foreign direct investment (FDI) and portfolio flows. Indeed, the share of FDI/GDP does poorly as a predictor of banking crises. Two of the fiscal variables—the budget deficit and central bank credit to the public sector—do moderately well, 38 ASSESSING FINANCIAL VULNERABILITY Institute for International Economics | http://www.iie.com Table 3.4 Annual indicators: currency crises Percent of crises accurately called 56 31 22 29 15 20 13 24 Indicator Current account balance/GDP Current account balance/ investment Overall budget deficit/GDP Short-term capital inflows/ GDP General government consumption/GDP Net credit to the public sector/GDP Central bank credit to the public sector/GDP Foreign direct investment/ GDP Noise-tosignal 0.41 0.49 0.58 0.59 0.74 0.88 0.99 1.00 P(C S) 43.2 39.0 36.4 35.2 29.4 26.2 23.8 21.7 P(C S) P(C) 19.5 15.1 11.5 10.9 5.9 2.4 0.1 0.1 while the third—government consumption—does poorly. Hence the role of the public sector in fueling banking crises is somewhat mixed. Without overinterpreting the results, it is interesting that the composition of the current account matters, in the sense that the current account as a percentage of investment does better in predicting banking crises than the current account as a share of GDP. It may be that investment is more likely to be financed through the international issuance of bonds and stocks or overseas loans, while consumption is more dependent on local bank credit. Turning to currency crises, the annual indicators that perform best are those measuring current account imbalances. This finding is not representative of the broader empirical literature. As discussed in Kaminsky, Lizondo, and Reinhart (1998), most of the studies that have attempted to explain the k-period ahead probability of a currency crisis have had mixed results regarding the current account, with most studies finding it insignificant. The various fiscal indicators do moderately well in anticipating currency crises, lending some support to Krugman-type models. By contrast with banking crises, the composition of capital inflows appears to have relatively little to add to our understanding of what drives a currency crisis. This result, however, may in part be due to the fact that a large share of EMPIRICAL RESULTS 39 Institute for International Economics | http://www.iie.com Table 3.5 Short-term debt: selected countries, June 1997 (percent) Country Asia Indonesia Malaysia Philippines South Korea Thailand Latin America Argentina Brazil Chile Colombia Mexico Short-term debt/total debt 24 39 19 67 46 23 23 25 19 16 Short-term debt/reserves 160 55 66 300 107 108 69 44 57 126 Sources: Bank for International Settlements; International Financial Statistics; World Bank. the currency crises (as opposed to the banking crises) took place in the 1970s in an environment of highly regulated internal and external financial markets, where portfolio flows were negligible. While our list of indicators is comprehensive, it is by no means exhaustive. The Asian crisis in particular highlighted the importance of currency and maturity mismatches in increasing vulnerability to currency and banking crises. Table 3.5 presents an indicator of the imbalance between liquid liabilities and liquid assets: namely, the ratio of short-term debt to international reserves. All the emerging economies in this group with debt-to-reserves levels in excess of 100 percent in mid-1997 have been casualties of financial turmoil in recent years (even if not all the speculative attacks ultimately succeeded, as in the case of Argentina.) This suggests that variables such as short-term debt to reserves could be a valuable addition to our list of leading indicators of crisis vulnerability.6 Do the Indicators Flash Early Enough? The previous discussion has ranked the indicators according to their ability to anticipate crises while producing few false alarms. Such criteria, however, do not speak to the lead time of the signal. From the vantage point of a policymaker or financial market participant who wants to implement preemptive or risk-mitigating measures, it is not a matter of 6. See Calvo and Mendoza (1996) for an early discussion of this issue. We did not use the ratio of short-term debt to reserves as an indicator in our tests because its relevance was highlighted mainly by the Asian crisis and we did not want the out-of-sample tests to be biased by its inclusion. In addition, the data were not available for the early part of our sample. 40 ASSESSING FINANCIAL VULNERABILITY Institute for International Economics | http://www.iie.com Table 3.6 How leading are the signals? (average number of months from when the first signal is issued to the crisis month) Indicator Bank deposits Beginning of banking crisis Domestic credit/GDP Domestic-foreign interest rate differential Excess M1 balances Exports Imports Lending-deposit interest rate ratio M2 multiplier M2/reserves Output Real exchange rate Real interest rate Reserves Stock prices Terms of trade n.a. not applicable Currency crisis 15 19 12 14 15 15 16 13 16 13 16 17 17 15 14 15 Banking crisis 8 n.a. 7 16 6 16 11 6 12 14 13 10 16 10 12 18 indifference whether an indicator sends a signal well before the crisis occurs or if the signal is given only when the crisis is imminent. Consider for example, the Conference Board’s composite indices of business cycle activity for the United States, which are published on a monthly basis. Both financial market participants and policymakers alike find the leadingindicator composite index more valuable than the coincident and lagging indices. Market participants incorporate this information in their investment decisions, while policymakers give it weight in their policy reactions. Over the years, US monetary policy has become increasingly forwardlooking and hence preemptive rather than reactive. One could argue that this transition was facilitated by an improvement in our understanding of the business cycle and early signs of its turning points. In what follows, we tabulate for each of the monthly indicators the average number of months before the crisis when the first signal occurs. This, of course, does not preclude the indicator from giving signals through the entire period immediately preceding the crisis. Indeed, for the more reliable indicators, signals tend to become increasingly persistent ahead of crises. For the low-frequency (annual) indicators, lead time is not much of an issue since some of these are published with a considerable lag and hence tend to be of less use from an early warning standpoint. Table 3.6 presents the lead times for our monthly indicators—both for currency and banking crises. In the case of currency crises, the most striking observation is that, on average, all the indicators send the first signal anywhere between a year and 18 months before the crisis erupts, with banking-sector problems (our second-ranked indicator) offering the EMPIRICAL RESULTS 41 Institute for International Economics | http://www.iie.com Table 3.7 Microeconomic indicators: banking crises Indicator Bank lending-deposit interest rate spread Interbank debt growth Interest rate on deposits Rate of growth on loans Net profits to income Operating costs to assets Change in banks’ equity prices Risk-weighted capital-to-asset ratio Source: Rojas-Suarez (1998). Percentage of crises accurately called 73 80 80 58 60 40 7 7 Noise-to-signal 0.28 0.35 0.47 0.72 1.14 1.59 2.00 2.86 longest lead time—namely, 19 months. The average lead time for these early signals is 15 months for currency crises. All the indicators considered are therefore best regarded as leading rather than coincident, which is consistent with the spirit of an ‘‘early warning system.’’ For banking crises, there is a greater dispersion in the lead time across indicators, and the average lead time is also lower (about 11 months). Furthermore, most of the indicators signal at about the same time, thus the signaling is cumulative and all the more compelling. Thus, on the basis of these preliminary results, there does appear to be adequate lead time for preemptive policy actions to avert crises. Microeconomic Indicators: Selective Evidence If, as the previous discussion suggests, banking crises are more difficult to predict on the basis of macroeconomic indicators than currency crises, it appears that the analysis of banking crises may benefit from including a variety of microeconomic indicators of bank health. Gonzales-Hermosillo et al (1997) and Rojas-Suarez (1998) provide some insights in this direction. Rojas-Suarez uses bank-specific data from Colombia, Mexico, and Venezuela and applies the ‘‘signals’’ methodology to this data to glean which items in bank balance sheets are most useful in predicting banking distress. Her results are summarized in table 3.7. They do indeed suggest that bank-specific information could make an important contribution in assessing the vulnerability of the banking sector in emerging markets. More ‘‘traditional’’ indicators, such as liquidity ratios and bank capitalization, turn out to be less useful indicators in Rojas-Suarez’s tests, in large part because they are ‘‘noisy’’ and likely to send many false alarms while missing many of the problem spots. At the other end, bank spreads and the interest rate that banks offer on deposits appear to systematically identify the weak banks. 42 ASSESSING FINANCIAL VULNERABILITY Institute for International Economics | http://www.iie.com One possible explanation for why interest rate spreads at the micro level may be more useful indicators of banking crisis than aggregate spreads is that the latter may reflect mainly cross-country differences in the extent of banking competition. In contrast, micro spreads are more likely to be more informative about a bank’s risk taking, as all banks within a country are apt to face a more common competitive environment. Goldstein (1998b) stresses bank exposure to the property sector as an indicator in the context of banking crises. He notes that in many of the affected Asian countries, estimates of the share of bank lending to the property sector exceeded 25 percent. Banking sector external exposure, measured in terms of foreign liabilities as a percentage of foreign assets, also appears to be a worthy addition to the list of sectoral or microeconomic indicators of banking-sector problems. EMPIRICAL RESULTS 43 Institute for International Economics | http://www.iie.com 4 Rating the Rating Agencies and the Markets The discussion in the preceding chapters focused on the ability of a variety of indicators to signal distress and to pinpoint the vulnerability of an economy to banking or currency crises. In this chapter, we assess the ability of sovereign credit ratings to anticipate such crises. In addition, given the wave of sovereign credit ratings downgrades that have followed the crisis in Asia, we investigate formally the extent to which credit ratings are reactive. Along the way, we discuss a small but growing literature that examines the extent to which financial markets anticipate crises. Do Sovereign Credit Ratings Predict Crises? We attempt to evaluate the predictive ability of sovereign credit ratings using two approaches. First, we tabulate the descriptive statistics for the ratings along the lines of the ‘‘signals’’ approach and compare how these stack up to the other leading indicators we have analyzed. Second, we follow the approach taken in much of the literature on currency and, more recently, banking crises and estimate a probit model. Specifically, we estimate a series of regressions where the dependent variable is a crisis dummy that takes on the value of one if there is a crisis and zero otherwise and where the explanatory variable is the credit ratings. ´ Our exercise is very much in the spirit of Larraın, Reisen, and von Maltzan (1997), who, using Granger causality tests, assess whether credit ratings lead or follow market sentiment as reflected in interest rate differentials. These interest rate differentials reflect the ease or difficulty with 45 Institute for International Economics | http://www.iie.com Table 4.1 Comparison of Institutional Investor sovereign ratings with indicators of economic fundamentals Percent of crises accurately called Difference between conditional and unconditional probability Type of crisis and indicator Currency crisis Institutional Investor sovereign rating Average of the top five monthly indicators Average of the top three annual indicators Banking crisis Institutional Investor sovereign rating Average of the top five monthly indicators Average of the top three annual indicators Noise-tosignal 1.05 0.45 0.49 31 70 36 5.4 19.1 15.4 1.62 0.50 0.41 22 72 44 0.9 9.1 16.3 which sovereign countries can tap international financial markets. In their analysis, they focus on the sovereign ratings of Moody’s and Standard & Poor’s; in what follows, we examine the behavior around financial crises of sovereign credit ratings issues by Moody’s Investor Service and Institutional Investor (II). The II sample begins in 1979 and runs through 1995. This gives us the opportunity to study 50 currency crises and 22 banking crises. There are 20 countries in this sample, with 32 observations per country for a total of 640 observations.1 For the Moody’s ratings, we have an unbalanced panel.2 Here there are 21 currency crises and 7 banking crises. Because the II database encompasses a more comprehensive sample of crises, we will place more emphasis on these results. Table 4.1 presents the basic descriptive statistics that we used in chapter 3 to gauge an indicator’s ability to anticipate crises: namely, the noise-tosignal ratio, the percentage of crises accurately called, and the marginal predictive power (i.e., the difference between the conditional and unconditional probabilities). We compare II sovereign ratings to averages for the more reliable monthly and annual indicators of economic fundamentals. 1. The 20 countries are those in the Kaminsky and Reinhart (1999) sample: Argentina, Bolivia, Brazil, Chile, Colombia, Denmark, Finland, Indonesia, Israel, Malaysia, Mexico, Norway, Peru, the Philippines, Spain, Sweden, Thailand, Turkey, Uruguay, and Venezuela. 2. An unbalanced panel, in this case, refers to the fact that we do not have the same number of observations for all the countries. 46 ASSESSING FINANCIAL VULNERABILITY Institute for International Economics | http://www.iie.com The basic story that emerges from table 4.1 is that the credit ratings perform much worse for both currency and banking crises than do the better indicators of economic fundamentals. The noise-to-signal ratio is higher than one for both types of crises, suggesting a similar incidence of good signals and false alarms. Hence, not surprisingly, the marginal contribution to predicting a crisis is small relative to the top indicators; for banking crises the marginal contribution is nil. Furthermore, the percentage of crises called is well below those of the top indicators. Indeed, the II ratings compare unfavorably with even the worst indicators. For example, consider the performance of the terms of trade ahead of banking crises (shown in the last row of table 3.1). The terms of trade has a noiseto-signal ratio of about one, making it almost as noisy as the credit rating. Yet the terms of trade accurately called 92 percent of the crises in sample— so while it sends many false alarms, it misses few crises. The II ratings, on the other hand, score poorly on both counts as, in addition to being noisy, they miss anywhere between two-thirds and three-quarters of the crises, depending on which type of crisis we focus on. Next we assess the predictive ability of ratings via probit estimation. The dependent variable is a crisis dummy (banking and currency crises are considered separately), and the independent variable is the change in the credit rating in the preceding 12 months. The II ratings are allowed to enter with a lag. The basic premise underpinning the simple postulated model is as follows. If the credit rating agencies are using all available information on the economic fundamentals to form their rating decisions, then credit ratings should help predict crises because (as shown in the preceding chapter) macroeconomic indicators have some predictive power and the simple model should not be misspecified—that is, other indicators should not be statistically significant, since that information would already presumably be reflected in the ratings themselves. Thus the state of the macroeconomic fundamentals should be captured in a single indicator—the ratings. Recent studies that have examined the determinants of credit ratings do provide support for the basic premise that ratings are significantly linked with selected economic fundamentals (Lee 1993; Cantor and Packer 1996a). For example, Cantor and Packer (1996a) find that per capita GDP, inflation, the level of external debt, and indicators of default history and of economic development are significant determinants of sovereign ratings. The question we seek to answer is whether these are the ‘‘right’’ set of fundamentals when it comes to predicting financial crises. Table 4.2 presents the results of the probit estimation, using both the II ratings and the Moody’s ratings as regressors. The results shown in table 4.2 are based on the 12-month change in the ratings, but alternative time horizons ranging from 6-month changes to 18- and 24-month changes produced very similar results.3 The method of estimation corrected for 3. These results are not reported here but are available from the authors. RATING THE RATING AGENCIES AND THE MARKETS 47 Institute for International Economics | http://www.iie.com Table 4.2 Do ratings predict banking crises? (probit estimation with robust standard errors) Independent variable 12-month change in the Institutional Investor rating 12-month change in Moody’s rating Coefficient 0.921 Standard error 1.672 Marginal effects 0.070 Probability 0.421 Pseudo R2 0.004 0.053 0.193 0.001 0.770 0.001 Table 4.3 Do ratings predict currency crises? (probit estimation with robust standard errors) Independent variable 12-month change in the Institutional Investor rating 12-month change in Moody’s rating Coefficient 0.561 Standard error 1.250 Marginal effects 0.075 Probability 0.590 Pseudo R2 0.058 0.22* 0.101 0.009 0.013 0.021 *Denotes significance at the 5 percent level. serial correlation and for heteroscedasticity in the residuals. For banking crises, the coefficients of the credit ratings have the anticipated negative sign—that is, an upgrade reduces the probability of a crisis. However, neither of the two credit-rating variables is statistically significant, and their marginal contribution to the probability of a banking crisis is very small. These results would, on the surface, be at odds with the findings of ´ Larraın, Reisen, and von Maltzan (1997), who find evidence that ratings ‘‘cause’’ interest rate spreads. Our interpretation, however, is that, while ratings may systematically lead yield spreads (they present evidence of two-way causality)—yield spreads are poor predictors of crises, as highlighted in tables 3.1 and 3.2. Hence the inability of ratings to explain crises is not inconsistent with the ability to influence spreads. This issue will be taken up later in this section. The analogous exercise for currency crises is reported in table 4.3. Again, the estimated coefficients of the ratings have the anticipated negative sign. Only in the case of the Moody’s rating, however, is the coefficient statistically significant at standard confidence levels. Even there, its marginal contribution to the probability of a currency crisis is quite small: a one-unit downgrade in the Moody’s rating increases the probability of a currency crisis by about 1 percent. The fact that the II ratings behave differently from Moody’s in the probit regression is consistent with other 48 ASSESSING FINANCIAL VULNERABILITY Institute for International Economics | http://www.iie.com research on ratings performance. Cantor and Packer (1996b), for example, provide extensive evidence of rating agency disagreements. Why Do Credit Ratings Fail to Anticipate Crises? As discussed in chapter 1, credit ratings and interest rate spreads may fail to anticipate a crisis either because lenders do not have access to timely and comprehensive information on the creditworthiness of the borrower or because lenders expect an official bailout of a troubled sovereign borrower. We now take up two related issues that could be associated with the poor performance of credit ratings as predictors of financial crises. The first one relates to the distinction between default and financial crises. The credit rating agencies themselves often argue that sovereign credit ratings are meant to provide an assessment of the likelihood of sovereign default. Hence to the extent that a domestic banking crisis or a currency crisis is decoupled from the probability of sovereign default, credit ratings should not a priori be expected to predict currency or banking crises. For example, the three Nordic countries included in our sample had both currency and banking crises in the 1990s, yet default on sovereign debt was never a likely event. Whatever the merits of this argument for industrialized economies, it looks less persuasive for developing countries and transition economies— where many default episodes have been preceded by banking and/or currency crises.4 Latin America’s experience in the early 1980s attests to this pattern. Furthermore, had it not been for large-scale rescue packages under the auspices of the International Monetary Fund (IMF), Mexico in 1994-95 and Indonesia, South Korea, and Thailand in 1997-98 would probably have been new additions to this list. While more research on this default versus crisis distinction is warranted, one simple implication is that the rating agencies should do better in predicting currency and banking crises in developing countries (since financial crises are more closely linked to the probability of sovereign default there than in industrial countries). To examine this issue empirically, we reestimated our simple model of crises and ratings, excluding industrial countries from the sample. The results, shown in tables 4.4 and 4.5, are not appreciably different from those for the full sample. For banking crises, neither of the ratings variables are statistically significant.5 4. The IMF’s World Economic Outlook of April 2000 notes that sovereign risk and devaluation tend to move together in the case of emerging economies. 5. We do not place much weight on the Moody’s results, as the number of banking crises is very small. In future work, it would be interesting to test whether Moody’s bank financial strength ratings (which are meant to capture the health of banks independent of the likelihood of a government bailout) are better predictors of banking crises. However, as these ratings RATING THE RATING AGENCIES AND THE MARKETS 49 Institute for International Economics | http://www.iie.com Table 4.4 Do ratings predict banking crises for emerging markets? (probit estimation with robust standard errors) Independent variable 12-month change in the Institutional Investor rating 12-month change in Moody’s rating Coefficient 0.827 Standard error 2.346 Marginal effects 0.052 Probability 0.871 Pseudo R2 0.002 0.075 0.413 0.001 0.864 0.001 Table 4.5 Do ratings predict currency crises for emerging markets? (probit estimation with robust standard errors) Independent variable 12-month change in the Institutional Investor rating 12-month change in Moody’s rating Coefficient 0.753 Standard error 1.430 Marginal effects 0.093 Probability 0.690 Pseudo R2 0.048 0.34* 0.161 0.026 0.011 0.041 *Denotes significance at the 5 percent level. For currency crises, the II ratings remain insignificant, while Moody’s ratings are significant but with a quite small marginal effect (below 3 percent). The second issue challenges the basic notion that credit ratings should be expected to be leading indicators. Because rating agencies receive fees from the borrowers they rate and because downgrades can subject the agencies to charges of having precipitated a crisis, some have argued— including in the Asian crisis—that credit ratings are apt to behave as lagging indicators of crises, with downgrades coming on the heels of crises. The anecdotal evidence surrounding the events in Asia seems to point in this direction (table 4.6). Only in the case of Thailand did there appear to be any substantive anticipatory action downgrades. To examine this issue more formally, we test whether the presence of a banking or currency crisis helps to predict credit rating downgrades. In other words, our dependent and independent variables now switch roles. For the II ratings, where we have available a continuous time series, we regress the six-month change in the credit rating index on the financial crisis dummy lagged by six months.6 The method of estimation is generalwere only introduced in 1995, the time series is not yet long enough to encompass many banking crises. 6. We want to examine whether the rating changes follow immediately after the crisis, but as the index is only published twice a year this ability to discriminate is not possible. 50 ASSESSING FINANCIAL VULNERABILITY Institute for International Economics | http://www.iie.com Table 4.6 Rating agencies’ actions on the eve and aftermath of the Asian crises, June-December 1997 Country Hong Kong Date 20-23 October 30 October Malaysia 14 July 18 August 25 September South Korea 8 June 22 October 24 October 27 October 27 November 3 December 10 December 11 December Thailand September 1996 February 1997 April June 13 August 3 September October 4 November 27 November Source: Goldstein (1998a). Events and action taken Stock market plunges as speculators attack Hong Kong dollar. Moody downgrades Hong Kong banks on concerns about property exposure. Ringit falls as central bank abandons support. S&P cuts ratings from ‘‘positive’’ to ‘‘stable.’’ S&P changes outlook to negative. S&P and Moody change outlook from ‘‘stable’’ to ‘‘negative.’’ Finance minister announces small-scale bank bailout and government takeover of Kia Motors. S&P downgrades government debt. Moody downgrades government debt. Moody downgrades ratings. IMF pact. Moody downgrades ratings. S&P downgrades ratings. Moody downgrades short-term government debt. Moody puts government debt under review. Moody cuts ratings but still rates government debt ‘‘investment grade.’’ S&P reaffirms rating. S&P reaffirms rating. Government seeks IMF bailout; S&P puts ratings under review. S&P cuts rating to A . Moody and S&P downgrade government debt. Prime minister resigns. Moody lowers ratings to near junk. ized least squares, correcting for heteroskedasticity and serial correlation in the residuals. For the Moody’s ratings, the dependent variable is threemonth changes in the rating, while the explanatory variable is the financial crisis dummy lagged three months. The latter specification should allow us to glean more precisely whether downgrades follow rapidly after crises take place. For the Moody’s ratings, our dependent variable assumes the RATING THE RATING AGENCIES AND THE MARKETS 51 Institute for International Economics | http://www.iie.com Table 4.7 Do financial crises help predict credit rating downgrades? (dependent variable II 6-month changes in sovereign rating, estimated using OLS with robust standard errors) Independent variable Banking crisis dummy Currency crisis dummy Coefficient 0.009 0.04* Standard error 0.019 0.014 Probability value 0.61 0.005 Pseudo R2 0.01 0.06 Table 4.8 Do financial crises help predict credit rating downgrades? (dependent variable Moody’s 3-month changes in sovereign rating, estimated using ordered probit) Independent variable Banking crisis dummy Currency crisis dummy * Coefficient 0.11 0.27* Standard error 0.90 0.14 Probability value 0.901 0.048 Pseudo R2 0.001 0.02 Significant at the 5 percent level. value of minus one, zero, or one depending on whether there was a downgrade, no change, or an upgrade, respectively. We therefore estimate the Moody’s ratings regression with an ordered probit technique. The results of the estimation are summarized in tables 4.7 and 4.8. For banking crises, the historical experience through 1995 does not support the proposition that credit-rating agencies behaved in a ‘‘reactive’’ manner. In contrast, our results suggest that currency crises help predict credit downgrades for both the Institutional Investor and Moody’s ratings. That is, as the explanatory variable increases (from zero to one when there is a crisis), ratings fall. However, while the coefficients are significant at standard confidence levels, their marginal predictive contribution is small. For example, in the case of Moody’s ratings, a currency crisis increases the likelihood of a downgrade by only 5 percent. ´ Our results are consistent with the findings of Larraın, Reisen, and von Maltzan (1997), who find evidence of two-way causality between sovereign ratings and market spreads. That is, not only do markets react to changes in the ratings, but the ratings systematically react (with a lag) to market sentiment. Do Financial Markets Anticipate Crises? The empirical tests presented here on sovereign credit ratings and financial crises need to be supplemented with tests on the ratings and larger 52 ASSESSING FINANCIAL VULNERABILITY Institute for International Economics | http://www.iie.com samples to determine whether our findings are robust. Nevertheless, we do not find surprising the evidence suggesting that sovereign ratings fail to anticipate banking and currency crises and are instead adjusted ex post. Kaminsky and Reinhart (1999) show that domestic-foreign interest rate differentials are not good predictors of crises, particularly currency crises. This result is reenforced in tables 3.1 and 3.2. Similarly, Goldfajn ´ and Valdes (1998) use survey data on exchange rate expectations (culled from the Financial Times) to test whether market expectations are adept at foreseeing financial crises. Using a broad array of crises definitions and approaches, their answer is a negative one and much along the lines of what we have found for the sovereign ratings. We concluded tentatively that if one is looking for early market signals of crises, it would be better to focus on equity returns rather than market exchange rate expectations and sovereign ratings. RATING THE RATING AGENCIES AND THE MARKETS 53 Institute for International Economics | http://www.iie.com 5 An Assessment of Vulnerability: Out-of-Sample Results As emphasized in chapter 1, predicting the timing of currency and banking crises is likely to remain an elusive task for academics, financial market participants, and policymakers. Recent events, however, have highlighted the importance of improving upon a system of early warnings. In this chapter, we apply the signals approach to several out-of-sample exercises using data for January 1996 through June 1997. Besides providing an assessment of the model’s out-of-sample performance, this exercise may shed light on why most analysts did not foresee the Asian crisis. In the first exercise, we look at measures across countries of crisis vulnerability (e.g., total number of signals, proportion of indicators signaling, and the number of top indicators signaling). But this exercise does not weigh the signals according to the relative track record of the indicators issuing the signal, or it only does so in a very approximate way. The second exercise extends the cross-country analysis by adjusting the threshold for each indicator so as to include more borderline signals in our measure of vulnerability. A third exercise weighs the indicators by the inverse of their noise-to-signal ratio to generate a series of crosscountry vulnerability ratings for both currency and banking crises. In yet a fourth exercise, we construct a composite indicator to map the timevarying probability of crisis; we compare its in- and out-of-sample performance to that of a naive forecast and the best of the univariate indicators. Finally, our last exercise focuses on the time-series dimension by mapping out the probability of crises for four Asian countries over the January 1996-December 1997 period. Needless to say, such exercises are fraught with the traditional Type I and Type II errors. Assume that the null hypothesis is that the economy 55 Institute for International Economics | http://www.iie.com is in a state of ‘‘tranquility.’’ If a high proportion of indicators are flashing, then one could reject that hypothesis in favor of the alternative—namely, that a crisis is likely in the next 24 months. Yet even though a country may be vulnerable, in the sense that a high proportion of variables are signaling trouble, the crisis may be averted through either good luck, good policies, or credible implicit bailout guarantees. This would be an example of a Type II error (rejecting the null hypothesis when it is true). A recent example of this case is Brazil, in which multiple signals were flashing as early as 1997, but these warning signs did not culminate in a full-fledged crisis until 1999. Alternatively, the crisis may occur without much warning from the indicators; this is a Type I error (failing to reject the null hypothesis when it is false). Borrowing a phrase from Sherlock Holmes, such a situation can be regarded as ‘‘the dog that did not bark in the night’’ and could be interpreted as evidence of contagion or multiplicity of equilibriums, an issue that we take up in chapter 6 and one that is particularly relevant for understanding the Indonesian crisis. Vulnerability and Signals Table 5.1 shows how our 25 sample countries compare on vulnerability to currency crises over the June 1996-June 1997 period, using several simple measures of vulnerability. The first column shows the total number of signals from among the 15 monthly indicators listed in table 3.1 that ‘‘flashed’’ during the period. The next column indicates how many of the 15 indicators sent signals, while the third data column lists the number of ‘‘top five’’ indicators sending signals. (For banking crises, these are real exchange rates, stock prices, the money multiplier, output, and exports, and for currency crises, they are real exchange rates, stock prices, exports, M2/reserves, and output.). The next set of columns give the comparable information for the eight annual indicators. In this case, we focus on the ‘‘top three’’ indicators. (For banking crises, the share in GDP of short-term capital inflows, current account balance as a share of investment, and the overall budget deficit as a share of GDP, and for currency crises, they are the current account balance as a share of GDP, the current account balance as a share of investment, and the overall budget deficit as a share of GDP.) The last column gives the percentage of the 23 indicators that are signaling. The reason to highlight the number of top indicators signaling is that these are the indicators with the lowest noise-to-signal ratios; hence a signal from these is more meaningful than a signal from a less reliable indicator. Table 5.1 provides this information for currency crises using the thresholds reported in table 3.2. There is considerable cross-country variation, with the lowest proportion of signals coming from Egypt and the highest 56 ASSESSING FINANCIAL VULNERABILITY Institute for International Economics | http://www.iie.com Table 5.1 Signals of currency crises, June 1996-June 1997 Monthly indicators Annual indicators Total signals 2 0 0 1 3 4 1 0 2 3 1 1 0 2 1 1 2 1 3 1 1 1 3 1 1 Number of indicators signaling 2 0 0 1 3 2 1 0 2 2 1 1 0 2 1 1 2 1 3 1 1 1 3 1 1 Top indicators signaling 0 0 0 0 0 2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 Total Percentage of indicators signaling 22 26 22 13 35 52 17 13 39 43 17 22 39 17 17 13 43 39 48 30 26 30 30 26 26 Institute for International Economics | http://www.iie.com 57 Country Argentina Bolivia Brazil Chile Colombia Czech Republic Denmark Egypt Finland Greece Indonesia Israel Malaysia Mexico Norway Peru The Philippines South Africa South Korea Spain Sweden Thailand Turkey Uruguay Venezuela Total signals 35 33 37 34 27 77 21 14 74 32 6 24 36 11 9 16 59 42 32 44 55 50 22 58 18 Number of indicators signaling 3 6 5 2 5 10 3 3 7 8 3 4 9 2 3 2 8 8 8 6 5 6 4 5 5 Top indicators signaling 1 2 3 1 1 3 1 2 1 2 1 1 3 0 0 0 1 3 3 2 1 3 1 0 2 from the Czech Republic, which indeed floated following a speculative attack and substantial reserve losses in May 1997. Table 5.2 repeats the same accounting exercise, but here we include ‘‘borderline’’ signals. Specifically, we enlarged the size of the rejection region by 5 percent for all the indicators. For instance, instead of having a 10 percent threshold for stock prices, we now have a 15 percent threshold. This sensitivity analysis increases the likelihood of making a Type II error (rejecting the null hypothesis of tranquility when you should not) while reducing the probability of a Type I error (not rejecting when you should). Including borderline signals does not seem to generate large shifts in the most and least vulnerable groups. As shown in the last column in table 5.2, borderline signals do not alter the picture at all for some countries (such as Argentina), but they do markedly increase the proportion of indicators signaling, as well as the number of signals, for countries such as South Korea (from 48 to 65 percent) and South Africa (from 39 to 52 percent). Tables 5.3 and 5.4 report the results for banking crises using the original thresholds and the ‘‘borderline’’ scenario, respectively. The country profiles that emerge are similar to those for currency crises; this may reflect the fact that several of the indicators have common thresholds for currency and banking crises. While conveying useful information on vulnerability, the preceding analysis does not fully discriminate between the more and less reliable indicators. Kaminsky (1998) shows how to construct a ‘‘composite index’’ to gauge the probability of a crisis conditioned on multiple signals from various indicators; the more reliable indicators receive a higher weight in this composite index. This methodology and its out-of-sample results are described in the remainder of this chapter. In weighting individual indicators, a good argument can be made for eliminating from our list of potential leading indicators those variables that had a noise-to-signal ratio above unity; this is tantamount to stating that their marginal forecasting ability, P(C S) P(C), is zero or less. Applying this criterion to banking crises, the lending-deposit ratio, the terms of trade, government consumption growth, and FDI as a share of GDP should be dropped. For currency crises, the excluded indicators are the domestic-foreign interest rate differential, the lending-deposit ratio, bank deposits, central bank credit to the public sector, and FDI as a share of GDP. For the remaining indicators with noise-to-signal ratios below unity, we weighed the signals by the inverse of the noise-to-signal ratios reported in tables 3.1 through 3.4. For a currency crisis, suppose that both the real exchange rate and imports are issuing a signal. Because the real exchange rate has a very low noise-to-signal ratio (0.22), it would receive a weight of 4.55 (i.e., 1/0.22); in contrast, with a relatively high noise-tosignal ratio (0.87), imports would receive a weight of only 1.49 (i.e., 1/0.87). 58 ASSESSING FINANCIAL VULNERABILITY Institute for International Economics | http://www.iie.com Table 5.2 Borderline signals of currency crises, June 1996-June 1997 Monthly indicators Annual indicators Total signals 2 0 0 1 3 4 2 1 3 3 1 1 0 2 1 2 3 2 5 1 1 1 3 1 1 Number of indicators signaling 2 0 0 1 3 2 2 1 2 3 1 1 0 2 1 2 3 2 4 1 1 1 3 1 1 Top indicators signaling 0 0 0 0 0 2 0 0 1 0 0 0 0 0 0 0 0 0 1 0 0 1 0 0 0 Total Percentage of indicators signaling 22 35 32 26 43 61 30 17 43 48 22 30 39 17 35 30 48 52 65 35 30 30 35 26 26 Institute for International Economics | http://www.iie.com 59 Country Argentina Bolivia Brazil Chile Colombia Czech Republic Denmark Egypt Finland Greece Indonesia Israel Malaysia Mexico Norway Peru The Philippines South Africa South Korea Spain Sweden Thailand Turkey Uruguay Venezuela Total signals 35 33 39 40 49 85 28 22 86 41 9 37 40 24 31 26 68 63 63 55 60 54 33 71 29 Number of indicators signaling 3 8 7 5 7 10 5 3 8 8 4 6 9 2 7 5 8 10 11 7 6 6 5 5 5 Top indicators signaling 0 2 3 2 3 3 1 2 1 3 3 3 3 0 2 1 3 3 3 2 1 3 3 1 2 60 Table 5.3 Signals of banking crises, June 1996-June 1997 Monthly indicators Annual indicators Total signals 1 0 0 1 3 4 1 0 3 2 1 1 0 2 1 1 2 1 3 1 1 1 2 1 2 Number of indicators signaling 1 0 0 1 3 2 1 0 2 2 1 1 0 2 1 1 2 1 3 1 1 1 2 1 2 Top indicators signaling 1 0 0 0 1 1 0 0 1 1 1 0 0 0 0 0 0 0 1 0 0 1 0 0 0 Total Percentage of indicators signaling 22 35 26 13 35 52 22 13 39 43 17 30 39 26 39 26 43 43 57 35 26 30 30 26 26 Number of indicators signaling 4 8 6 2 5 10 4 3 7 8 3 6 9 4 8 5 8 10 10 7 5 6 5 5 4 Top indicators signaling 0 2 2 1 2 3 0 0 1 3 2 3 3 1 2 1 3 3 4 1 1 2 3 1 1 Institute for International Economics | http://www.iie.com Country Argentina Bolivia Brazil Chile Colombia Czech Republic Denmark Egypt Finland Greece Indonesia Israel Malaysia Mexico Norway Peru The Philippines South Africa South Korea Spain Sweden Thailand Turkey Uruguay Venezuela Total signals 36 42 39 34 38 81 24 18 77 39 10 32 42 16 30 19 59 55 42 51 59 53 27 74 18 Table 5.4 Borderline signals of banking crises, June 1996-June 1997 Monthly indicators Annual indicators Total signals 1 0 0 1 3 4 1 0 3 2 1 1 0 2 1 1 2 1 3 1 1 1 2 1 2 Number of indicators signaling 1 0 0 1 3 2 1 0 2 2 1 1 0 2 1 1 2 1 3 1 1 1 2 1 2 Top indicators signaling 1 0 0 0 1 2 0 0 1 1 1 0 0 0 0 0 0 0 1 0 0 1 0 0 0 Total Percentage of indicators signaling 35 35 39 35 52 52 30 13 43 48 22 39 39 30 39 30 43 48 61 43 39 43 35 30 35 Institute for International Economics | http://www.iie.com 61 Country Argentina Bolivia Brazil Chile Colombia Czech Republic Denmark Egypt Finland Greece Indonesia Israel Malaysia Mexico Norway Peru The Philippines South Africa South Korea Spain Sweden Thailand Turkey Uruguay Venezuela Total signals 46 45 44 43 70 87 29 24 88 50 14 49 49 30 37 27 73 68 74 58 66 58 35 86 34 Number of indicators signaling 7 8 9 7 9 10 6 3 8 9 4 8 9 5 8 6 8 10 13 9 8 9 6 6 6 Top indicators signaling 1 2 3 3 3 3 1 0 1 4 3 4 3 1 2 2 3 3 5 2 2 3 3 1 2 Table 5.5 Weighting the signals for currency and banking crises in emerging markets, June 1996-June 1997 Currency crises Country Argentina Bolivia Brazil Chile Colombia Czech Republic* Egypt Greece Indonesia* Israel Malaysia* Mexico Peru The Philippines* South Africa South Korea* Thailand* Turkey Uruguay Venezuela Weighted signals 5.41 6.59 7.57 5.90 10.59 15.42 6.02 14.27 7.54 6.30 12.46 2.82 2.82 14.40 16.52 14.57 14.63 8.21 4.40 5.28 Rank 16 12 10 15 8 2 14 6 11 13 7 19 19 5 1 4 3 9 18 17 Banking crises Weighted signals 7.98 7.30 6.08 5.74 11.87 17.24 8.33 14.15 8.33 10.38 7.74 2.59 5.33 11.52 12.74 14.55 12.09 7.87 4.88 6.02 Rank 10 13 14 16 6 1 9 3 9 8 12 19 17 7 4 2 5 11 18 15 Note: An asterisk (*) denotes the country had a currency crisis, a banking crisis, or both in 1997-98. Formally, we construct the following composite indicator, n It j 1 S tj / j (5.1) In equation 5.1, it is assumed that there are n indicators. Each indicator has a differentiated ability to forecast crises, and as before, this ability can be summarized by the noise-to-signal ratio, here denoted by j. S jt is a dummy variable that is equal to one if the univariate indicator, Sj crosses its critical threshold and is thus signaling a crisis and is zero otherwise. As before, the noise-to-signal ratio is calculated under the assumption that an indicator issues a correct signal if a crisis occurs within the following 24 months. All other signals are considered false alarms. If all 18 good indicators were sending signals, the maximum value that this composite vulnerability index could score is 30.05 for banking crises and 33.23 for currency crisis. This score is a simple sum of the inverse of the noise-to-signal ratios for the good indicators that are retained. However, it is seldom the case that every indicator signals. Table 5.5 presents the composite score of the indicators that are signaling for the 20 emerging 62 ASSESSING FINANCIAL VULNERABILITY Institute for International Economics | http://www.iie.com Table 5.6 Vulnerability to financial crises in emerging markets: alternative measures, June 1996-June 1997 Average proportion of indicators signaling both crises 29 35 36 31 48 57 15 48 22 35 39 30 46 24 50 63 35 35 28 31 Average proportion of top eight indicators signaling both crises 11 22 33 33 44 56 11 44 44 44 33 22 33 11 33 56 44 33 11 22 Country Argentina Bolivia Brazil Chile Colombia Czech Republic* Egypt Greece Indonesia* Israel Malaysia* Peru The Philippines* Mexico South Africa South Korea* Thailand* Turkey Uruguay Venezuela Rank 11 8 7 9 4 2 15 4 14 8 6 10 5 13 3 1 8 8 12 9 Rank 5 4 3 3 2 1 5 2 2 2 3 4 3 5 3 1 2 3 5 4 Average of ‘‘weighted’’ signals 6.69 6.94 6.82 5.74 11.23 16.33 6.42 14.21 7.93 8.34 10.10 4.08 12.96 2.71 14.63 14.56 13.36 8.04 4.88 6.02 Rank 14 12 13 17 7 1 15 4 11 9 8 19 6 20 2 3 5 10 18 16 Note: An asterisk (*) denotes the country had a currency crisis, a banking crisis, or both in 1997-98. economies in our sample; currency and banking crises are treated separately. The first data column provides the relevant value of the index for a currency crisis. The next column shows the country’s ordinal ranking for the vulnerability index relative to the remaining 19 countries. South Africa, the Czech Republic, and Thailand emerge as the most vulnerable on the basis of the signals issued and the quality of those signals during January 1996-June 1997. For banking crises, the comparable exercise ranks the Czech Republic, South Korea, and Greece as the most vulnerable. Perhaps not surprisingly, near the bottom of the list are countries such as Mexico and Venezuela, which are still recovering from their 1994-95 crises. Thus far, we have treated banking and currency crises separately in our vulnerability rankings. If one wanted to assess the ‘‘average’’ vulnerability to both banking and currency crises, one may want to combine the information contained in these two measures. Table 5.6 provides information on the average proportion of indicators signaling banking OUT-OF-SAMPLE RESULTS 63 Institute for International Economics | http://www.iie.com and currency crises, the average proportion of the top eight indicators (monthly and annual) that are signaling, and the average of the ‘‘weighted’’ indices reported in table 5.5 for currency and banking crises. The table also ranks the countries, by these three criteria, depending on the degree of ‘‘vulnerability.’’ Concentrating on the average of the ordinal rankings derived from the weighted signals (last column of table 5.6), we can see that clustered at the top of the list are several of the countries that have had or are still undergoing financial crises; these countries are denoted by an asterisk. This suggests a relatively encouraging out-of-sample performance for the signals approach. The three measures of vulnerability provide similar rankings for most of the ‘‘extreme’’ cases, such as the Czech Republic, South Korea, Malaysia, and the Philippines among the countries that have already had crises and South Africa, Colombia, and Greece among those that have not. In the case of Greece, however, there was an orderly devaluation, while in Colombia’s case there was both a devaluation (in August 1998) as well as serious banking sector difficulties. For countries such as Thailand and to a lesser degree Indonesia, taking into account the ‘‘quality’’ of the indicator that is signaling considerably changes the overall ranking. The Composite Indicator and Crises Probabilities While the foregoing exercise allows us to assess the relative propensity to crisis across countries at a point in time—like a snapshot—it does not convey information on the dynamics of the process. To assess the extent to which a country is becoming more or less vulnerable to crisis over time, one would need a continuum of such snapshots. To do so, it is convenient to link the composite index to the implied probability of crisis. Once we construct this composite indicator, we can then proceed—as we did with the individual indicators in chapters 2 and 3—to choose a critical value for the composite indicator so that when the composite indicator crosses this threshold, a crisis is deemed to be imminent.1 As before, this critical threshold could be chosen so as to minimize the noiseto-signal ratio of the composite indicator. Moreover, we could calculate the probability of a crisis conditional on the composite indicator signaling a crisis (i.e., crossing the critical threshold) as well as the odds of a crisis when the composite indicator is not signaling. However, this procedure would not give us an exhaustive reading of vulnerability as the crisis approaches because it is dichotomous—that is, it will only provide two 1. Meaning, as in the individual indicators, in the subsequent 24 months. 64 ASSESSING FINANCIAL VULNERABILITY Institute for International Economics | http://www.iie.com types of information—namely, signal or no signal. We want also to introduce shades of gray in crisis vulnerability. The idea is to analyze the empirical distribution of the composite indicator jointly with the occurrences of crises and to estimate probabilities of crises conditional on different values of the composite indicator. We would like to evaluate what the odds are of a crisis if none of the individual indicators are signaling (i.e., when the composite indicator takes on a value of zero) or when all the indicators are signaling (that is, when the composite reaches its maximum value). But we would also like to evaluate the intermediate scenarios, which depend on both how many and which of the indicators are signaling. For example, we would like to calculate the probability of a crisis conditioned on knowing that the value of the indicator is in the 9 to 14 range, which as we saw from the cross-section analysis earlier in this chapter was associated with a number of the recent crises (table 5.5). In practice, we can construct this set of probabilities using the information on the value of the composite indicator for all the countries in the sample together with the information on crises. Probabilities of crises are estimated as follows: P(C I It I) A/(A B) (5.2) where I is the lower bound of the range we are interested in (9 in our earlier example) and I is the upper bound of the range we are interested in (14 in our example). As before, we have the following two-by-two matrix, Crisis occurs in the following 24 months I It I It / [ I , I ] A C No crisis occurs in the following 24 months B D These probabilities will be estimated using all the information from all the countries in the sample. Once we estimate these probabilities and use the information on the number of signals being issued at any moment in time, we can construct time-series probabilities of crisis for every country. P tm denotes the probability of a crisis for country m in period t. Once we construct these time series of crisis probabilities, we can also evaluate the forecasting ability of the composite indicator and compare its track record with that of other indicators, such as our top-ranked univariate indicator, the real exchange rate. To conduct this horse race, we follow Diebold and Rudebusch (1989) and employ the Quadratic Probability Score (QPS) as our metric of goodness of fit. In particular, the QPS evaluates the average closeness of the predicted probabilities and OUT-OF-SAMPLE RESULTS 65 Institute for International Economics | http://www.iie.com Table 5.7 Composite indicator and conditional probabilities of financial crises Value of indicator 0-1 1-2 2-3 3-4 4-5 5-7 7-9 9-12 12-15 Over 15 Memorandum: Unconditional probability of a currency crisis 0.29 n.a. not applicable Unconditional probability of a banking crisis 0.10 Probability of a currency crisis 0.10 0.22 0.18 0.21 0.27 0.33 0.46 0.65 0.74 0.96 Probability of a banking crisis 0.03 0.05 0.06 0.09 0.12 0.13 0.16 0.27 0.37 n.a. Source: Kaminsky (1998). the observed realizations, as measured by a dummy variable that takes on a value of one when there is a crisis and zero otherwise.2 T QPS k 1/T t 1 2(P k t Rt ) 2 (5.3) where k 1,2,3 refers to the indicator Pk, refers to the probability associated with that indicator, and Rt refers to the zero-one realizations. The QPS ranges from zero to two, with a score of zero corresponding to perfect accuracy. Empirical Results Table 5.7 reports the conditional probabilities of both currency and banking crises using the composite indicator. One column reports the likelihood of currency crises. When almost none of the indicators are signaling a future crisis, the composite indicator takes on values between zero and two, and the probability of a currency crisis is only about 10 percent. The probability of a currency crisis increases sharply and nonlinearly as signs 2. This approach has also been used to assess the ability of various indicators to anticipate turning points in the business cycle (Diebold and Rudebusch 1989). 66 ASSESSING FINANCIAL VULNERABILITY Institute for International Economics | http://www.iie.com Table 5.8 Scoring the forecasts: quadratic probability scores Currency crises Indicator Naive forecast Real exchange rate Composite indicator Source: Kaminsky (1998). Tranquil times 0.173 0.115 0.110 Crisis times 1.008 0.979 0.862 Banking crises Tranquil times 0.024 0.018 0.024 Crisis times 1.620 1.589 1.309 of vulnerability in the economy increase. Specifically, the probability of a currency crisis reaches almost 100 percent when the composite indicator takes on a value of 15 or above.3 The right column reports the same evidence for banking crises. As with currency crises, the probabilities of a collapse of the banking sector increase sharply as the economy deteriorates. However, as we found with the univariate indicators, banking crises are harder to anticipate. Even when nearly all the univariate indicators are signaling, the probability of a banking crisis only climbs to about 40 percent. Table 5.8 turns to the forecasting accuracy of the composite indicator. The left side of the table looks at currency crises, while the right side examines banking crises. The performance of the composite indicator is compared with the performance of the real exchange rate—the best univariate indicator—as well as to the naive forecast based on the unconditional probability of crisis. The score statistics are reported separately for ‘‘crisis times’’ and for ‘‘tranquil times;’’ this provides information on the performance of the leading indicators across regimes. Recall that the closer the score in table 5.8 is to zero, the more accurate is the forecast. The real exchange rate does significantly better in anticipating currency crises than the unconditional forecast of currency crises. More important, the composite indicator performs better—in terms of accuracy—than the real exchange rate, but the larger improvements are obtained when forecasting in crisis times. As shown on the right side of table 5.8, all indicators score worse when predicting the onset of the banking crises—that is, the 24 months bracketing the beginning of the banking crises. Again, the real exchange rate does better than the unconditional forecast of banking crises in general. For example, the quadratic probability score declines from 0.024 and 1.620 for the naive forecast of currency crises to 0.018 and 1.589 for the real exchange rate forecast during tranquil and crisis times, respectively. The composite indicator outperforms the real exchange rate in forecasting the onset of a banking crisis but is outperformed by the real exchange rate during tranquil times. This is explained by the fact that the real exchange rate issues very few false alarms during tranquil periods. 3. Note we are not using the annual indicators in this exercise. OUT-OF-SAMPLE RESULTS 67 Institute for International Economics | http://www.iie.com An Out-of-Sample Application to Southeast Asia Using the information on the monthly value of the composite indicator and on the conditional probabilities of crises, we can construct a time series of probabilities of crises for our sample countries both in the sample period (from January 1970 to December 1995) and out of it (from January 1996 to December 1997). As an illustration, figure 5.1 reports the timeseries probabilities of currency crises for four Southeast Asian economies in the 1990s. The vertical lines in the figures represent the onset of a crisis. With the exception of Indonesia, all the Southeast Asian countries showed a severe state of distress, with about 65 percent of the indicators flashing signals during the year preceding the crisis.4 The onset of these crises occurred as the economies entered a marked slowdown in growth after a prolonged boom in economic activity fueled by rapid credit creation.5 This dramatic surge in credit is explained, in large measure, by heavy capital inflows and partly by the reform of the financial system; financial liberalization was accompanied by large reductions in reserve requirements. Overall, the explosive growth in these countries came to an end with a real appreciation of the domestic currencies (which are, in differing degrees, tied to the US dollar) and the corresponding loss of export markets. It is noteworthy that during the latter part of this period, ` there was a substantial appreciation of the dollar vis-a-vis the yen. Short-term capital inflows to Thailand amounted to 7 to 10 percent of GDP in each of the years 1994 through 1996, with the growth rate of credit to the nonfinancial private sector amounting to more than 23 percent over 1990-95. While output growth rates increased in the early 1990s to almost 9 percent, fueled in part by easy credit, this rapid growth showed signs of coming to an end with the real appreciation of the domestic currency and the corresponding loss of export markets. The annual growth rate of exports fell from a peak of 30 percent per year in 1994 to about 0 in 1996. Financial sector fragilities were also evident, with runs against major banks starting to occur as early as May 1996. Finally, the sharp increase in interest rates in 1997 to defend the baht put the nail in the coffin of the already weak banking sector.6 Overall, 75 percent of the indicators for which there are available data were exhibiting ‘‘anomalous’’ behavior. A boom-bust cycle in lending was also evident in the Philippines. As in Thailand, the boom was fueled by capital inflows but also by a dramatic 4. For a more detailed exposition of the incidence of flashing indicators in the run-up to the Asian crisis, see Kaminsky and Reinhart (1999). 5. This is at odds with the interpretation of these crises provided in Radelet and Sachs (1998), who argue these crises are the byproduct of a financial panic. 6. It is noteworthy that finance companies had been receiving substantial assistance from the central bank during this period. 68 ASSESSING FINANCIAL VULNERABILITY Institute for International Economics | http://www.iie.com Figure 5.1 Probability of currency crisis for four Southeast Asian countries, 1990-97 Indonesia 1.0 0.8 0.6 0.4 0.2 0.0 1998 1990 1991 1992 1993 1994 Malaysia 1995 1996 1997 1.0 0.8 0.6 0.4 0.2 0.0 1998 1990 1991 1992 1993 1994 1995 1996 1997 The Philippines 1.0 0.8 0.6 0.4 0.2 0.0 1998 1990 1991 1992 1993 1994 Thailand 1995 1996 1997 1.0 0.8 0.6 0.4 0.2 0.0 1998 1990 1991 1992 1993 1994 1995 1996 1997 Note: Vertical lines indicate currency crisis date. Source: Kaminsky (1998). OUT-OF-SAMPLE RESULTS 69 Institute for International Economics | http://www.iie.com reduction in reserve requirements, accompanying financial liberalization. Bank credit increased by 44 percent a year during 1995-96. As in Thailand, rapidly expanding credit was an important contributor to the rally in stock and real estate markets, with a fourfold increase in prices in both markets. Foreign currency exposure increased in the Philippines in the 1990s via foreign borrowing to finance domestic lending. Foreign borrowing was concentrated in short maturities. Consumer lending also increased and fueled a surge in consumption, leading to a deterioration of the current account. This deterioration in the external accounts was aggravated by the real exchange rate appreciation of the domestic currency. The loss of competitiveness anticipated a future decline in growth and also contributed to a substantial deterioration of the quality of banks’ assets, further reducing the odds of survival of many individual financial institutions. Overall, about 50 percent of the indicators were signaling the increased vulnerability of the economy during the two years before the collapse of the implicit peg in July 1997.7 Malaysia shared certain vulnerabilities with Thailand. It too was affected by the slowdown in the region, though to a much smaller degree. It too had large current account deficits during 1990-95, although in 1996 the outlook for the external sector improved somewhat, with the current account to GDP ratio shrinking to -5.3 percent (in Thailand, the current account to GDP ratio in 1996 was roughly -8.0 percent). Moreover, Malaysia, like Thailand, accumulated debt rapidly in the 1990s, with capital inflows fueling a stock and real estate market boom and with asset prices increasing about 300 percent in the early 1990s. Malaysia also suffered from financial sector vulnerabilities (although not to the same extent as Thailand) as a result of the high degree of leveraging in the economy. Indeed, Malaysia had one of the highest ratios of credit-to-GDP in the world, and the banks had a large exposure to the property and equity markets. For Malaysia, about 60 percent of the indicators were showing signs of distress at the onset of the crisis. Indonesia looked somewhat different. While it too exhibited banking fragilities and while short-term debt easily exceeded available foreign exchange reserves (about 1.7 times the stock of the country’s reserves),8 the current account deficit did not deteriorate as fast (reaching only 3.5 percent of GDP in 1996), the slowdown in growth was not yet evident, and the real exchange rate did not appreciate as much as in the other 7. The Philippines was classified as a managed float in the IMF’s exchange rate arrangements classification. Yet even a relatively uninformed bystander could see the large-scale extent of foreign exchange intervention before mid-1997, which kept the Philippine peso’s value virtually unchanged against the dollar. 8. The beginning of the banking crisis in Indonesia can be dated to November 1992, when a large bank (Bank Summa) collapsed and triggered runs on three smaller banks. Most state-owned banks also experienced serious difficulties. 70 ASSESSING FINANCIAL VULNERABILITY Institute for International Economics | http://www.iie.com countries in the region. Relatively few indicators (less than 20 percent) showed signs of strains in the economy in the months before the crisis. Here, over and beyond all the political uncertainty, as we explain further in chapter 6, a key factor seemed to be contagion from the flurry of financial crises elsewhere in the region—particularly the liquidity squeeze associated with the withdrawal of Japanese banks (the major lenders to the region) in the wake of losses they suffered in the Thai crisis.9 To sum up, we have seen in this chapter that the signals approach can draw coarse distinctions, both across countries and over time, in crisis vulnerability during out-of-sample periods (in this case, 1996-97). The approach does reasonably well in anticipating currency crises in most of the Asian crisis countries. At this stage, the model performs much better for currency crises than for banking crises. The evidence presented here also indicates that it is worthwhile to work with a composite index, which outperforms the best of the univariate indicators. 9. The reversal was, in fact, quite pronounced, from capital inflows in the region of $50 billion in 1996 to an outflow of $21 billion in 1997. See Kaminsky and Reinhart (2000) and the next chapter for a discussion on world and regional financial links and their effects on the probability of currency crises. OUT-OF-SAMPLE RESULTS 71 Institute for International Economics | http://www.iie.com 6 Contagion As suggested earlier, in most cases, the leading indicators signaled ahead of the 1997-98 currency and banking crises. The Indonesian case, however, is an example of an episode where ‘‘the dog did not bark.’’ Despite the fact that this country experienced a meltdown in its currency and a collapse in its banking industry, Indonesia was firmly anchored near the bottom of the list in table 5.5, as relatively few indicators gave advanced warning. In a similar vein, although Argentina was the hardest-hit country during the ‘‘tequila effects’’ that followed the Mexican financial crisis of 199495, it too would not have been judged as vulnerable on the basis of the fundamentals reviewed in the preceding chapters. Of the 89 currency crises and nearly 30 banking crises in our sample, only a handful of these occur with as few indicators flashing as was the case for Indonesia (22 percent). As shown in table 6.1, less than 15 percent of the currency and banking crises shared the Indonesian silence of signals. Still, the Indonesian crisis suggests something is missing from our previous analysis. The most obvious candidate is cross-country contagion of financial crises.1 The empirical evidence on contagion is still limited to relatively few studies, but the weight of the empirical results suggests it is important. To the extent that contagion or spillovers matter, being near the bottom of the ‘‘vulnerability’’ list based on own-country fundamentals would not preclude a country from having a crisis. In this chapter, we briefly review 1. Of course, the political turmoil at this time in Indonesia is likely to have contributed to the meltdown of the currency and the economy. 73 Institute for International Economics | http://www.iie.com Table 6.1 Crises that showed few signals, 1970-97 Number of crises that occurred with five or fewer indicators signaling 3 12 1 1 Proportion of crises that occurred with five or less indicators signaling 10.3 13.5 16.7 16.7 Type of crisis and sample Banking, 1970-95 Currency, 1970-95 Banking, 1996-97 Currency, 1996-97 Number of crises 29 87 6 6 some of the theoretical underpinnings for contagion and then construct a ‘‘contagion or spillover vulnerability index’’ that attempts to capture trade and finance links among countries. We then explore the extent to which crises probabilities increased for other emerging markets following the Mexican crisis of 1994 and the Asian crisis of 1997, owing to trade and financial links. Most of the theoretical work on contagion has attempted to provide a framework for understanding how shocks in one country are transmitted elsewhere. Our review of this literature emphasizes its empirical implications in terms of defining contagion, delineating its channels of influence, and testing for its presence. Defining Contagion Only one study that we are aware of examined the issue of contagion in the context of Latin America’s debt crisis of the 1980s. Doukas (1989) interprets contagion as the influence of ‘‘news’’ about the creditworthiness of a sovereign borrower on the spreads charged to the other sovereign borrowers, after controlling for country-specific macroeconomic funda´ mentals. Most other studies, such as Valdes (1997), define contagion as excess comovement in asset returns across countries, be it for debt or equity. This comovement is said to be excessive if it persists even after common fundamentals, as well as idiosyncratic fundamental factors, have been controlled for. A recent variant to this approach (as in Forbes and Rigobon 1998) defines ‘‘shift-contagion’’ as an increase in excess comovement of asset returns during crisis periods. Eichengreen, Rose, and Wyplosz (1996) define contagion as a case where knowing that there is a crisis elsewhere increases the probability of a crisis at home, even when fundamentals have been properly taken into account. This is the definition of contagion that we will explore in the remainder of this chapter. These fundamentals could be country-specific, along the lines analyzed in the preceding chapters, or they could be external and common to all countries or a group of countries. Changes 74 ASSESSING FINANCIAL VULNERABILITY Institute for International Economics | http://www.iie.com in international interest rates are a plausible candidate for a common shock. If international interest rates rise markedly, as they did in the early 1980s, and many countries have financial crises simultaneously, we would not attribute the common timing of the crises to contagion—we would place the blame, instead, on a common shock. In the absence of a common shock, a crisis in one country can spread to others via links in trade and finance. Some studies would not call this contagion either but rather label it a spillover (e.g., Masson 1998). These studies would reserve the term contagion for cases where a crisis spreads from one country to another despite the absence of any trade or finance link—possibly owing to shifts in sentiment and herding behavior on the part of investors. Since it is impossible to predict when such shifts in sentiment will take place and which countries will be most affected by changes in financial markets’ expectations, our focus in the empirical part of this chapter will be on assessing countries’ vulnerability to a crisis elsewhere when financial and trade links are evident. Theories of Contagion and Their Implications There are several explanations for why crises tend to be bunched or clustered. Some recent models have revived Nurkse’s story of ‘‘competitive devaluations.’’ This explanation emphasizes trade links, be they bilateral or with a third party.2 Once one country has devalued, it is costly (in terms of a loss of competitiveness and output) for other countries (with strong trade links to the first country) to maintain their parities. In this setting, subsequent devaluations reflect a policy choice, with a salutary effect on output. In any case, an empirical implication of this story of contagion is that we should either observe a high volume of trade among the ‘‘synchronized’’ devaluers or competition in a common third market.3 Calvo (1998) stresses the role of liquidity. A leveraged investor facing margin calls needs to sell his or her asset holdings to an uninformed counterpart. Because of information asymmetries, a ‘‘lemons problem’’ arises, and the asset can only be sold at a fire sale price. A variant of this story can be told about an open-end fund portfolio manager who needs to raise liquidity in anticipation of future redemptions. The strategy will be not to sell the asset whose price has already collapsed but other assets 2. See Gerlach and Smets (1995) for a model that emphasizes bilateral trade and Corsetti et al. (1998) for one in which emerging markets compete in a common third market. 3. As a story of fundamentals-based contagion, of course, this explanation does not speak to the fact that central banks often go to great lengths to avoid the devaluation in the first place. CONTAGION 75 Institute for International Economics | http://www.iie.com in the portfolio. In doing so, other asset prices are depressed, and the original disturbance spreads across markets. Yet another potentially important channel of transmission that has been largely ignored in the contagion literature but that is stressed by Kaminsky and Reinhart (2000) is the role of common lenders—in particular, commercial banks. US banks had large loan exposure to Latin America in the early 1980s, much in the way that Japanese banks did during the Asian crisis of 1997. The need to rebalance the overall risk of the bank’s asset portfolio and to recapitalize following the initial losses can lead to a marked reversal in commercial bank credit, both in the original crisis country and for others who rely heavily on the same lender. Another family of contagion models has deemphasized the role of trade in goods and services in favor of the role of trade in financial assets, particularly in the presence of information asymmetries. Calvo and Mendoza (2000) present a model where the fixed costs of gathering and processing country-specific information give rise to herding behavior, even when investors are rational. Kodres and Pritsker (1998) also present a model with rational agents and information asymmetries. However, they stress the role played by investors who engage in cross-market hedging of macroeconomic risks. In these financial contagion explanations, the channels of transmission come from the global diversification of financial portfolios. Here, the implication is that countries with more internationally traded financial assets and more liquid markets are likely to be relatively vulnerable to contagion. Small emerging economies with relatively illiquid financial markets are likely to be underrepresented in international portfolios to begin with and thus ought to be shielded from this type of contagion. In addition, cross-market hedging usually requires a moderately high correlation of asset returns. For our purposes, the key empirical implication is that countries whose asset returns exhibit a high degree of comovement with the original crisis country (for example, Argentina with Mexico in 1994-95 or Malaysia with Thailand in 1997-98) will be more vulnerable to contagion via the cross-market hedges that were in place as the crisis erupted. Empirical Studies Very few studies have attempted to run ‘‘horse races’’ among alternative models of contagion. Eichengreen, Rose, and Wyplosz (1996) tested the influence of bilateral trade links against similarities to the crisis country in macroeconomic fundamentals. Glick and Rose (1998) examined the trade issue further within a much broader country sample, while Wolf (1997) attempted to explain pairwise correlations in stock returns by bilateral trade and by common macroeconomic fundamentals. All studies 76 ASSESSING FINANCIAL VULNERABILITY Institute for International Economics | http://www.iie.com Table 6.2 Conditional probabilities and noise-to-signal ratios for financial and trade clusters Percentile of countries sharing a cluster Noise-to-signal ratio 25 to 50 50 and above Weight in vulnerability index 25 to 50 50 and above Probability of a crisis conditioned on crises elsewhere in the cluster minus unconditional probability of crisis 25 to 50 50 and above Bank 0.90 0.07 High correlation 0.58 0.39 Third-party trade 1.54 0.57 Bilateral trade 2.34 0.08 1.10 14.08 1.73 2.57 0.64 1.75 0.42 12.5 3.1 52.0 20.8 47.1 6.3 30.7 21.8 47.3 Source: Based on Kaminsky and Reinhart (2000). conclude that trade linkages play an important role in the propagation of shocks. Because trade tends to be more intra- than interregional in nature, Glick and Rose (1998) conclude that this helps explain why contagion tends to be mainly regional rather than global. Kaminsky and Reinhart (1998b) also look at trade links (both bilateral and third-party) but emphasize financial sector links. In an early paper on the subject, Frankel and Schmukler (1996) find evidence of contagion in emerging market mutual funds. Trade and Financial Clusters and a Composite Contagion Index As shown in chapter 5, one can construct a composite index to gauge the probability of a crisis conditioned on multiple signals from various indicators (i.e., economic fundamentals); the more reliable indicators receive greater weight in this composite index. This methodology can be readily applied to construct a composite ‘‘contagion vulnerability index.’’ As in Kaminsky and Reinhart (2000), we consider four channels through which shocks can be transmitted across borders: two channels deal with the interlinkages in financial markets, be they through foreign bank lending or globally diversified portfolios, and two deal with trade in goods and services. Table 6.2 reports the noise-to-signal ratios and the difference between the conditional probability of a crisis (conditioned on CONTAGION 77 Institute for International Economics | http://www.iie.com knowing there is a crisis elsewhere in that particular cluster) and the unconditional probability of crisis. Hence the four clusters—bilateral trade, third-party trade, common bank lender, and high correlations—play the same role as the indicators.4 If a country shares a common cluster with the initial crisis country, it is a signal; if a crisis occurs in the second country within the following 24 months, it is a good signal; if a crisis does not occur, it is a false alarm. Hence for these possible linkages, the number of signals could range from zero (no common clusters) to four, in which the country shares all four clusters with the initial crisis country. As when we weighted individual indicators, a good argument can be made for eliminating potential leading indicators that had a noise-tosignal ratio above unity (that is, those whose marginal forecasting ability is zero or less). Applying this criterion to our results, we would focus on the case in which more than 50 percent of the countries that share a common cluster are experiencing a crisis. As shown in table 6.2, the highest noise-to-signal ratio is 0.57, well below unity—but the track record of the signals in each of the clusters is far from uniform. Thus we weight the signals by the inverse of the noise-to-signal ratios reported in table 6.2 (see Kaminsky and Reinhart 2000 for details). Formally, as we did in chapter 5 for the macroeconomic fundamentals, we construct the following composite indicator: n It j 1 S tj / j (6.1) In equation 6.1 it is assumed that there are n indicators (i.e., clusters). Each cluster has a differentiated ability to forecast crises, and as before, this ability can be summarized by the noise-to-signal ratio, here denoted by j. S jt is a dummy variable that is equal to one if the univariate indicator, S j, crosses its critical threshold and is thus signaling a crisis and is zero otherwise. As before, the noise-to-signal ratio is calculated under the 4. The countries are classified by bank clusters according to which financial center they depend on the most (on the basis of the Bank for International Settlements data). For the high-correlation asset returns cluster, we include countries that have a correlation that is 0.35 percent or higher in their daily stock returns. For the bilateral trade cluster, we include countries for which either imports or exports to the second country are 15 percent or higher. For the third-party trade cluster, we require countries to have a common third market and similar commodity export structure. We focus on the top 10 to 15 goods that account for 40 percent or more of exports in the initial crisis country; we then see if those same goods account for a significant share (20 percent or higher) of exports of the remaining countries. For example, the top 14 Thai exports account for 46 percent of total exports; these same goods account for 44 percent of Malay exports; hence Malaysia is in the same third-party trade cluster. By contrast, those goods only account for 15 percent of Indonesia’s exports, leaving Indonesia outside the third-party trade cluster. 78 ASSESSING FINANCIAL VULNERABILITY Institute for International Economics | http://www.iie.com assumption that an indicator issues a correct signal if a crisis occurs within the following 24 months. All other signals are considered false alarms. The maximum value that this composite contagion vulnerability index could score is 30.9 if a country belonged to the same four clusters as the crisis country. This score is a simple sum of the inverse of the noise-tosignal ratio. Table 6.3 records a one if a country is in the same cluster as the original crisis country in that episode and no entry otherwise. What the Composite Contagion Vulnerability Index Reveals about Three Recent Crisis Episodes We now consider, on the basis of the trade and financial sector linkages discussed here, which countries would have been classified as vulnerable to contagion during three recent episodes of currency crises in emerging markets. The first of these episodes began with the devaluation of the Mexican peso in December 1994. On the heels of the Mexican devaluation, Argentina and Brazil were the countries to come under the greatest speculative pressure. In a matter of a few weeks in early 1995, the central bank of Argentina lost about 20 percent of its foreign exchange reserves and bank deposits fell by about 18 percent as capital fled the country. Such a severe outcome could hardly be attributed to trade linkages and competitive devaluation pressures, as Argentina does not trade with Mexico on a bilateral basis, nor does it compete with Mexican exports in a common third market. 5 In the case of Brazil, the speculative attack was brief, although the equity market sustained sharp losses. Both of these countries record high vulnerability index scores following the Mexican devaluation. While the effects on Asia of the Mexican crisis were relatively mild, the country that encountered the most turbulence in the region was the Philippines, which also registers a relatively high vulnerability score. In the case of the Thai crisis, Malaysia shares both trade and finance links with Thailand. For the other Asian countries, the potential channels of transmission are fewer. As noted earlier, the Philippines is a part of the same third-party trade cluster as Thailand, which receives a weight of 1.75 (i.e., 1/0.57) in the composite index; it is also part of the Asian high-correlation cluster, which receives a weight of 2.57 (i.e., 1/0.39) in the index. Indonesia shares the same high-correlation cluster with Thailand, and it is a part of the Japanese bank cluster, which receives a weight of 14.08 (i.e., 1/0.07). Hence, as shown in table 6.4, Indonesia and the Philippines’ contagion vulnerability index scores are 16.65 and 4.32, 5. See Kaminsky and Reinhart (2000) for details on the pattern of trade. CONTAGION 79 Institute for International Economics | http://www.iie.com 80 Table 6.3 Countries sharing financial and trade clusters with original crisis country or region Bank cluster Country Argentina Bolivia Brazil Chile Colombia Denmark Finland Indonesia Israel Malaysia Mexico Norway Peru The Philippines South Korea Spain Sweden Thailand Turkey Uruguay Venezuela Japan US 1 1 1 1 High-correlation cluster Asia Latin America 1 1 1 1 Third-party trade cluster Asia Latin America Bilateral trade cluster Latin America 1 1 1 Institute for International Economics | http://www.iie.com 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 Source: Kaminsky and Reinhart (2000). Table 6.4 Contagion vulnerability index Contagion vulnerability index Country Argentina Bolivia Brazil Chile Colombia Denmark Finland Indonesia Israel Malaysia Mexico Norway Peru The Philippines South Korea Spain Sweden Thailand Turkey Uruguay Venezuela n.a. not applicable Mexican crisis (December 1994) 16.65 0 18.4 0 12.5 0 0 0 0 0 n.a. 0 2.57 14.08 0 0 0 0 0 0 12.5 Thai crisis (July 1997) 0 0 0 0 0 0 0 16.65 0 28.33 0 0 0 4.32 26.58 0 0 n.a. 0 0 0 Brazilian crisis (January 1999) 29.15 0 n.a. 26.58 15.83 0 0 0 0 0 18.4 0 2.57 14.08 0 0 0 0 0 26.58 15.83 respectively. South Korea also borrowed heavily from Japanese banks. Accordingly its exposure to Thailand came more from having a common lender than from conventional competitive trade pressures. The most recent of these emerging market crises was Brazil’s devaluation of the real in early 1999. Not surprisingly, Argentina, which has both trade (through Mercosur) and financial linkages with Brazil, shows the highest vulnerability; other Mercosur countries come close in suit. Table 6.5 provides additional details on some of the possible channels through which the crisis may have spread during these episodes. To the extent that there is herding behavior and investors lump together all emerging markets—or perhaps only those in the infected region—that would add yet another channel of transmission to those laid out in table 6.5. As regards the potential role of bilateral and third-party trade linkages, Malaysia would be the country most closely linked with Thailand, with South Korea and the Philippines exhibiting more moderate trade exposure. Trade is certainly not the main culprit in explaining the vulnerability of Argentina and Brazil following the Mexican devaluation or of Indonesia following the Thai crisis. CONTAGION 81 Institute for International Economics | http://www.iie.com 82 Table 6.5 Characteristics of affected countries in Asian and Mexican episodes of contagion Level of liquid market/high representation in mutual funds, percentage of emerging market portfolio Level of trade with common third party in same commodities, percentage of exports competing with top exports of affected country Institute for International Economics | http://www.iie.com Affected country (onset month) Exchange rate regime at onset Nature of contagion or spillover Common bank lender High correlation of returns Level of bilateral trade, percentage of exports to affected country Tequila crisis: 1994-95 First crisis: Mexico, December 1994 Argentina Brazil Currency board Peg Turbulence Turbulence Yes Yes High, 0.56 Moderate, 0.36 Moderate, 2.98 High, 13.07 Low, 1.7 Low, 2.4 Low, 15.6 Low, 10.9 Asian flu: 1997-98 First crisis: Thailand, July 1997 Malaysia (July) The Philippines (July) Indonesia (August) Hong Kong (October) South Korea (November) Managed float Managed float Narrow band Currency board Crawling band Crisis Crisis Crisis Turbulence Crisis Yes Yes Yes No Yes Low, 0.24 High, 0.60 High, 0.68 High, 0.54 Moderate, 5.88 Low, 2.40 Moderate, 4.35 High, 15.33 Moderate, 6.16 Moderate, 4.1 Moderate, 3.8 Low, 1.8 Low, 1.0 Low, 2.0 High, 44.4 Low, 19.2 Low, 15.5 Low Moderate, 27.9 Table 6.6 Asia and Latin America: added power of Thai crisis in explaining probability of contagion in bank cluster, July 1997 Probability of a crisis conditioned on crises elsewhere in the cluster minus unconditional probability of crisis 0.60 0.35 0.02 0.02 0.02 0.02 Country Asia Indonesia Malaysia The Philippines Latin America Argentina Chile Mexico Turning to financial links stemming from a common lender, exposure to European and Japanese banks, which rapidly pulled out of the region after the outbreak of the Thai crisis, was common to all the affected countries except Hong Kong. Brazil and Argentina were in the same (US) bank cluster as Mexico in 1994-95, but US banks were not as exposed to Latin American borrowers as they were in the early 1980s, and portfolio flows had replaced bank lending as the main source of funding for these emerging economies. Most of the affected Asian countries (except South Korea) had high correlations of asset returns with Thailand, although none except Hong Kong were home to relatively liquid markets. The same is true of stock returns in Argentina, which had the highest correlation of asset returns with Mexico of any country in the region. Here, it is hard to separate cause and effect. A high correlation may reflect past contagion, but to the extent that current cross-hedging strategies use such historical correlations as a guide, it could be the vehicle for future contagion. In sum, while this is a preliminary assessment of contagion channels, it suggests that financial sector linkages, be it through banks or via international capital markets, could have been influential in determining how shocks were propagated in recent crises episodes, particularly for Argentina, Brazil, and Indonesia. In table 6.6 we take this analysis one step further. Specifically, the table compares some of the larger emerging markets in Asia and in Latin America at the onset of the Thai crisis (July 1997) based on how much added explanatory power a crisis elsewhere added to the probability of crisis at home. The numbers reported in the table are the simple difference between the probability of a crisis conditioned on our composite index of fundamentals, P(C F), and the probability of crisis conditioned on the fundamentals and a crisis elsewhere related to a common lender, P(C F, CONTAGION 83 Institute for International Economics | http://www.iie.com CE). If knowing that a crisis elsewhere in the cluster helps predict a crisis at home, then P(C F, CE) P(C F). It is noteworthy that the conditional probability of a crisis does not change much for those Latin American countries and the Philippines that are not a part of the Japanese bank cluster. For them, the contagion from the Thai crisis via this channel is minimal. By way of contrast, for countries that are in the same bank cluster as Thailand, the probability of crisis increases markedly, for Malaysia and particularly for Indonesia. Malaysia’s crisis probabilities conditioned on the fundamentals alone were well above Indonesia’s, as shown in figure 5.1. Hence, for Malaysia, the incremental explanatory power of the crisiselsewhere variable is smaller than for Indonesia. To sum up, the empirical evidence contained in this chapter suggests that the analysis of fundamentals stressed in the signals approach can be strengthened by incorporating financial sector linkages, which increase the vulnerability to contagion. While assessing the predictive ability of the individual bank clusters is a useful exercise to discriminate among competing explanations of contagion, countries that are linked in trade are also often linked in finance. This implies that multiple channels of contagion may be operating at once. 84 ASSESSING FINANCIAL VULNERABILITY Institute for International Economics | http://www.iie.com 7 The Aftermath of Crises The preceding chapters have focused on the antecedents of financial crises. The emphasis has been on the indicators’ ability to anticipate crises and to measure the extent of a country’s vulnerability. In this chapter, we begin with the premise that, whether anticipated or not, financial crises occur, and once they do, policymakers and market participants become concerned about their consequences for economic activity. In light of Asia’s recent woes, there was much speculation as to how long it would take those economies to recover from such destabilizing shocks and what the consequences for growth and inflation would be over the near and medium term. In what follows, we review the historical experience of the aftermath of currency and banking crises. The Recovery Process If we want to assess how our indicators behave following financial crises and, in particular, how many months elapse before their behavior returns to normal, we must define ‘‘normal.’’ One way to do that is to compare ‘‘tranquil’’ and ‘‘crisis’’ periods. We define periods of tranquility as the periods that exclude the 24 months before and after currency crises. In the case of banking crises, the 24 months before the banking crisis begins and the 36 months following it are excluded from tranquil periods. For each indicator, we tabulate its average behavior during tranquil periods. We then compare the postcrisis behavior of the indicator to its average in periods of tranquility. 85 Institute for International Economics | http://www.iie.com Table 7.1 Length of recovery from financial crises (average number of months for a variable to return to ‘‘normal’’ behavior)a Indicator Bank deposits Domestic credit/GDPb Exports Excess M1 balances Imports Lending-deposit rate ratio M2 multiplier M2/reserves Output Real exchange rate Real interest ratec Real interest rate differential Stock prices Terms of trade Banking crisis 30 (below) 15 (above) 20 (below) 9 (above) 29 (below) 0 7 (above) 15 (above) 18 (below) 8 (below-overvalued) 15 (above) 15 (above) 30 (below) 4 (below) Currency crisis 12 (above) 9 (above) 8 (below) 8 (below) 18 (below) 3 (above) 21 (below) 7 (above) 10 (below) 23 (above-undervalued) 7 (below) 7 (below) 13 (below) 9 (below) a. We note in parentheses whether the variable remained below or above the tranquilperiod norm. b. Domestic credit as a share of GDP remains above normal levels largely as a result of the decline in GDP following the crisis. c. The disparity between the postcrisis behavior of real interest rates lies in the fact that a large share of the currency crises occurred in the 1970s, when interest rates were controlled and not very informative about market conditions. Table 7.1 summarizes the results of that aftermath exercise for currency and banking crises. The number given after each indicator is the average number of months that it takes for that variable to reach its norm during tranquil periods. In parentheses, we note whether the level or growth rate of the variable remains above or below its norm in the postcrisis period. Several findings merit special attention. First, the deleterious effects of banking crises do linger longer than currency crises’ effects. This is evident in several of the indicators. While the 12-month change in output remains below its tranquil-period norm for (on average) 10 months following the currency crash, it takes nearly twice that amount of time to recover following the banking crisis. This more sluggish recovery pattern is also evident in imports, which take about 21⁄2 years to return to their norm. The weakness in asset prices, captured here by stock prices that are below the norm, persist for 30 months on average for banking crises—more than twice the time it takes to recover from a currency crash. There are several explanations for banking crises’ more protracted recovery periods. One concerns the special nature of the ‘‘twin’’ crises. The bulk of the banking crises in this sample were accompanied by currency crises, and twin crises ought to have more severe effects on the economy, as argued in Kaminsky and Reinhart (1999). 86 ASSESSING FINANCIAL VULNERABILITY Institute for International Economics | http://www.iie.com Table 7.2 Time from beginning of banking crises to their peaks (months) Descriptive statistic Mean Minimum Maximum Standard deviation Number of months 19 0 53 17 Source: based on Kaminsky and Reinhart (1999). A second explanation, not mutually exclusive, is that a banking crisis cuts off both external and domestic sources of funding for households and firms, whereas a currency crisis only cuts off the former. In other words, the credit crunch is more severe. A third explanation derives from the distribution of crises across the sample period. The currency crises are roughly evenly distributed between the pre- and postliberalization periods, while the banking crises are bunched in the 1980s and 1990s. To the extent that crises have become more severe following deregulation, the slower pace of recovery in banking crises may reflect that. This is an issue we take up later. A second finding highlighted in table 7.1 is that there are important sectoral differences in the pace of recovery, depending also on the type of crisis. For instance, following the devaluations that characterize the bulk of currency crises, exports recover relatively quickly and ahead of the rest of the economy at large. In contrast, following banking crises, exports continue to sink for nearly two years. This may reflect a persistent overvaluation, high real interest rates, or a ‘‘credit crunch’’ in the aftermath of banking crisis. Table 7.2 underscores the protracted nature of banking crises by showing the average number of months elapsed from the beginning of the crisis to its zenith for the 26 banking crises studied in the Kaminsky and Reinhart (1999) sample. On average, it takes a little over a year and a half for a banking crisis to ripen; in some instances it has taken over four years. Often, financial sector problems do not begin with the major banks, but rather with more risky finance companies. As the extent of leveraging rises, households and firms become more vulnerable to adverse economic or political shocks that lead to higher interest rates and lower asset values. Eventually, defaults increase and problems spread to the banks. If there are banks runs, such as in Venezuela in 1994, the spread to the larger institutions may take less time. The information in table 7.2 does not fully capture the length of time that the economy may be weighed down by banking-sector problems, since it does not cover information on the time elapsed between the crisis peak and its ultimate resolution. Rojas-Suarez and Weisbrod (1996), who examine the resolution of several banking crises in Latin America, highTHE AFTERMATH OF CRISES 87 Institute for International Economics | http://www.iie.com Table 7.3 Comparison of inflation and growth rates before and after currency crises (percent) average of t 1 and t 2 3.3 3.5 3.0 Indicator Real GDP growth All countries Moderate-inflation countriesa High-inflation countries Inflation Moderate inflation countries High-inflation countries t year of crisis t 1.0 2.1 0.6 t 1 1.8 2.4 1.0 t 2 3.1 3.3 3.1 t 3 2.9 4.0 1.7 14.0 270.9 15.7 732.8 18.0 394.8 15.7 707.4 14.8 964.7 a. Moderate-inflation countries are those with inflation rates below 100 percent in all years surrounding the crisis; high-inflation countries are those in which inflation exceeded 100 percent in at least one year. light the sluggishness of the resolution process in many episodes. The Japanese banking crisis, which has spanned most of the 1990s and is still ongoing, is a recent example of the protracted nature of the recognitionadmission-resolution process. We next focus in table 7.3 on the evolution of two of the most closely watched macroeconomic indicators—growth and inflation—in the aftermath of currency crises. Instead of comparing tranquil versus crisis periods, we compare the immediate two precrisis years with the postcrisis years.1 We distinguish between moderate-inflation and high-inflation countries; the latter encompass mostly Latin American countries. The numbers for ‘‘all countries’’ represent an average of the 89 currency crises in our sample. Perhaps the most interesting finding from table 7.3 is that it takes between two and three years in currency crisis episodes for economic growth to return to the precrisis average. Devaluations may be expansionary in industrial countries—witness the sharp recovery in the United Kingdom following its floatation of the pound during the European exchange rate mechanism (ERM) crisis and the strong growth performance of the Australian economy in 1998 coincident with a large depreciation of the Australian dollar. 1. For a comparison of the recent crises with the historic norm, see Kaminsky and Reinhart (1998). 88 ASSESSING FINANCIAL VULNERABILITY Institute for International Economics | http://www.iie.com Yet, as shown in table 7.3, devaluations in the developing world are most often associated with recessions. There are numerous theoretical explanations for this finding (see Lizondo and Montiel 1989 for a survey of the literature). Two are of particular interest. First, devaluations that occur in the context of balance of payments crises are associated with losses of confidence and increases in uncertainty that are damaging to economic activity. It is usually the case that credibility problems are more severe for developing countries than for their industrial counterparts. Second, while industrial countries do not face a higher debt-servicing costs following devaluation, as their debt is predominantly denominated in their own currencies, developing-country debt is largely denominated in US dollars or other foreign currencies. Hence a large devaluation will have staggering implications for debt servicing burdens. Furthermore, recessions following currency crises appear to be more severe among the high-inflation countries. This may be because inflation itself has adverse effects on growth (Fischer 1993) or because high-inflation countries may be especially prone to losing their access to international credit relative to their low-inflation counterparts. The evidence presented in Cantor and Packer (1996a) does indeed show that private credit ratings penalize high inflation. In any event, most existing studies find devaluation episodes in emerging economies to be contractionary, with their negative impact diminishing within two years, and table 7.3 supports these findings.2 In this connection, Morley (1992) concludes that the reason that earlier studies, which are largely focused on devaluations during the 1950s and 1960s, find milder recessions and even positive output consequences is that many of those devaluation episodes occurred in the context of trade liberalization and exchange market reform—not in the context of balance of payments crises. Table 7.3 shows that inflation picks up in the two years following the currency crisis in both moderate- and high-inflation countries. The increase is far more dramatic for high-inflation countries, where inflation remains at a substantially higher level following the crisis (usually because of recurring devaluations at an accelerating rate). For the moderate-inflation countries, inflation returns to its precrisis rate in about three years. These patterns are consistent with those found by Borensztein and de Gregorio (1998) in 19 devaluation episodes in low- and high-inflation countries. The empirical studies surveyed in table 7.4 arrive at similar conclusions. Some Caveats The preceding discussion has suggested a ‘‘representative time profile’’ for the recovery process in the wake of currency and banking crises. This 2. See also Kamin and Rogers (1997) for an interesting analysis of the case of Mexico. THE AFTERMATH OF CRISES 89 Institute for International Economics | http://www.iie.com 90 Institute for International Economics | http://www.iie.com Table 7.4 The wake of devaluations: a review of the literature Study Borensztein and de Gregorio (1998) Cooper (1971) Edwards (1986) Sample 1982-94, 19 devaluation episodes, five of which are industrial countries 1951-70, 24 large devaluations 1965-80, 12 developing countries Inflation Variable Results About one-quarter of the devaluation is offset by higher inflation after three months, about 60 percent after two years. Except for Latin American cases, inflation returned to its pre-devaluation level in three years or less. Devaluations are followed by either a recession or a reduction in the rate of growth. These output effects were small, however. Regressing income on the real exchange rate while controlling for policy fundamentals, he finds a negative and significant coefficient on the real exchange rate in the first year; this was offset by positive coefficients later on. Long-run effect is neutral. Inflation doubles, on average from about 8 to 16.7 percent one year after the crisis; net foreign assets/money fall by about 5 percent in the three years following the crisis. The trade balance does not change much the year following the devaluation; import and export growth increase. Capital inflows and reserves are about the same at t 1 as in the year of the devaluation. Inflation increases the year of the devaluation then declines. GDP growth falls the year of the devaluation then recovers the following year. GDP GDP Edwards (1989) 1962-82, 39 devaluations greater than 15 percent in 24 countries 1953-83, 50 to 90 devaluations in excess of 15 percent Inflation, GDP, current account as a share of GDP, change in net foreign assets/money Inflation, GDP, exports, imports, export prices, import prices, capital inflows, trade balance, reserves Kamin (1988) Kiguel and Ghei (1993) 1950-90, 33 devaluations in excess of 20 percent in low-inflation countries Real exchange rate, inflation, GDP growth, exports/GDP, reserves/ imports, parallel premium GDP About 60 percent of the devaluation is not eroded by increases in domestic prices. Inflation increases, on average by about 1; growth 11⁄2 percentage points, between t 3 and increases by 1 percent in that same period; exports and reserves/imports also rise between 1 and t 3; the parallel premium falls. As in Cooper (1971), devaluations are followed by either a recession or a reduction in the rate of growth. These output effects were small. After controlling for other fundamentals, the real exchange rate is found to have a negative and significant effect on capacity utilization for up to two years. He finds real devaluations are associated with sharp declines in investment. Institute for International Economics | http://www.iie.com 91 Krueger (1978) 1951-70, 22 large devaluations 1974-83, 28 devaluations in excess of 15 percent. Morley (1992) Capacity utilization Table 7.5 Comparison of severity of crises by region and period, 1970-97 Currency crises (indexa) Period 1970-94 1995-97 Latin America 48.1 25.4 East Asia 14.0 40.0 Other 9.0 n.a. Banking crises (bailout cost as share of GDP) Latin America 21.6 8.3 East Asia 2.8 15.0 Other 7.3 n.a. a. See text for description of index’s construction. Source: Kaminsky and Reinhart (1998). profile suggests that growth will return to normal within about two years of the crisis and that the inflationary consequences of the devaluation will abate within about three years. Yet this pattern would hardly describe the protracted recovery process of many Latin American economies during the 1980s, not even the relatively rapid recovery experienced by Chile.3 The speed at which the economy recovers from a financial crisis will be heavily influenced by how policymakers respond to the crisis as well as by external conditions. The high level of international real interest rates in the 1980s (the highest levels since the 1930s) were hardly conducive to speedy recovery. Moreover, as suggested in Kaminsky and Reinhart (1998a), severe currency and banking crises are apt to be associated with more delayed recoveries. This latter point is particularly relevant to the recovery from the 1997-98 crises in Asian countries, which are significantly more severe that the earlier crises in that region. To analyze this issue formally, we measure the severity of currency and banking crises, as in Kaminsky and Reinhart (1998). For banking crises, the measure of severity is the cost of the banking bailout expressed as a share of GDP. For currency crises, we construct an index that gives equal weights to reserve losses and currency depreciation. This index is centered on the month of the currency crisis, and it combines the percentage decline in foreign exchange reserves in the six months before the crisis, since reserve losses typically occur before the central bank capitulates, and the depreciation of the currency in the six months following the abandonment of the existing exchange rate arrangement. This latter component captures the magnitude of the currency meltdown. Table 7.5 presents these measures of severity for the 76 currency crises and 26 banking crises in the Kaminsky-Reinhart sample. For the 1970-94 sample, currency and banking crises were far more severe in Latin America than elsewhere. The 1970-94 crises in East Asia, by contrast, were relatively mild and not that different by these metrics from the crises in 3. Chile’s inflation rate was in single digits when it abandoned its crawling peg policy in 1982. 92 ASSESSING FINANCIAL VULNERABILITY Institute for International Economics | http://www.iie.com the European countries that largely represent the ‘‘other’’ group. This divergence may also help explain the subpar performance of the highinflation countries during recovery (table 7.3). The picture that emerges during 1995-97 is distinctly different. Both in terms of this measure of the severity of the currency crisis as well as the estimated costs of bailing out the banking sector, the severity of the recent Asian crises surpasses that of their Latin American counterparts in the late 1990s and represents a significant departure from its historic regional norm. In this sense, the V-shaped recoveries in Asia have been less protracted than the history of past severe crises would have suggested. THE AFTERMATH OF CRISES 93 Institute for International Economics | http://www.iie.com 8 Summary of Results and Concluding Remarks In this book we have introduced a set of indicators that, on the basis of both in-sample and out-of-sample tests, appear to be useful for gauging vulnerability to currency and banking crises in emerging economies. The indicators are not precise enough to make fine distinctions in crisis vulnerability across countries and over time, but they can draw some distinctions between the most and least vulnerable groups of countries and recognize large increases in the vulnerability of a given country over time. As such, they have the potential to add value as a ‘‘first screen’’ of vulnerability and as a supplementary tool to other types of analysis of crisis vulnerability. As suggested in chapter 5, we think the indicators would have been useful in anticipating the Asian currency and banking crises. In this chapter, we summarize our key results. Furthermore, in thinking about future evaluation of such leading indicators of crises, two obvious questions arise: would publication of the indicators erode their usefulness in an early warning system, and are there policy implications associated with the better performing indicators? We discuss each of these questions in turn. Summary of Findings Our main empirical findings can be summarized in 12 main points. First, banking and currency crises in emerging markets do not typically arrive without any warning. There are recurring patterns of behavior in the period leading up to banking and currency crises. Reflecting this 95 Institute for International Economics | http://www.iie.com tendency, the better-performing leading indicators anticipated between 50 and 100 percent of the banking and currency crises that occurred over the 26-year sample period. At the same time, even the best leading indicators send a significant share of false alarms (on the order of one false alarm for every two to five true signals).1 Second, using monthly data, banking crises in emerging economies are more difficult to forecast accurately than are currency crises. Within the sample, the average noise-to-signal ratio is higher for banking crises than for currency crises, and the model likewise does considerably better out-of-sample in predicting currency crises than banking crises. It is not yet clear why this is so. It may reflect difficulties in accurately dating banking crises—that is, in judging when banking sector distress turns into a crisis and when banking crises end. For example, by our criteria, banking distress in Indonesia and Mexico really began in 1992 (and not in 1997 and 1994, respectively). The absence of high-frequency (monthly or quarterly) data on the institutional characteristics of national banking systems probably also is a factor. Third, there is wide variation in performance across leading indicators, with the best-performing indicators displaying noise-to-signal ratios that are two to three times better than those for the worst-performing ones.2 In addition, the group of indicators that show the best (in-sample) explanatory power also seem, on average, to send the most persistent and earliest signals. Warnings of a crisis usually appear 10 to 18 months ahead. Fourth, for currency crises, the best of the monthly indicators were appreciation of the real exchange rate (relative to trend), a banking crisis, a decline in stock prices, a fall in exports, a high ratio of broad money (M2) to international reserves, and a recession. Among the annual indicators, the two best performers were both current account indicators—namely, a large current account deficit relative to both GDP and investment (table 8.1). Fifth, turning to banking crises, the best (in descending order) of the 15 monthly indicators were appreciation of the real exchange rate (relative to trend), a decline in stock prices, a rise in the (M2) money multiplier, a decline in real output, a fall in exports, and a rise in the real interest rate. Among the eight annual indicators tested, the best of 1. The construction of the noise-to-signal ratio is described in chapter 2. 2. When an indicator has a noise-to-signal ratio above one, crises would be more likely when the indicator was not sending a signal than when it was. Similarly, when an indicator has a conditional probability of less than zero, it means that the probability of a crisis occurring when the indicator is signaling is lower than the unconditional probability of a crisis occurring—that is, merely estimating the probability of a crisis according to its historical average. For example, if currency crises occur in a third of the months in the sample, the unconditional probability of a crisis is one-third. 96 ASSESSING FINANCIAL VULNERABILITY Institute for International Economics | http://www.iie.com Table 8.1 Currency and banking crises: best-performing indicators Currency crises High-frequency indicators Real exchange rate Banking crisis Stock prices Exports M2/reserves Output Real exchange rate Stock prices M2 multiplier Output Exports Real interest rate on bank deposits Low-frequency indicators Current account balance/GDP Current account balance/investment Short-term capital inflows/GDP Current account balance/investment Banking crises the pack were a high ratio of short-term capital inflows to GDP and a large current account deficit relative to investment (table 8.1). Sixth, while there is a good deal of overlap between the best-performing leading indicators for banking and currency crises, there is enough of a distinction to warrant treating the two separately. To highlight two noteworthy differences, the two indicators that serve as proxies for financial liberalization—namely, a rise in the real interest rate and an increase in the money multiplier—turned out to be more important for banking crises than for currency crises, whereas the opposite proved true for the two indicators designed to capture currency/maturity mismatches and excessively expansionary monetary policy—namely, a high ratio of broad (M2) money balances to international reserves and excess M1 money balances, respectively. Seventh, while our data on sovereign credit ratings cover only a subsample of crises and relate to only two of the major rating firms (Moody’s Investor Services and Institutional Investor), we find that changes in sovereign credit ratings have performed considerably worse than the better leading indicators of economic fundamentals in anticipating both currency and banking crises in emerging economies. In addition, we find no empirical support for the view that sovereign rating changes have led financial crises in our sample countries rather than reacting to these crises. In a similar vein, we have found that interest rate spreads (i.e., foreign-domestic real interest rate differentials) are not among the bestperforming group of leading indicators. More empirical work needs to be done to determine whether these results are robust to other rating agencies and other samples. Nevertheless, our findings suggest that those who are looking to ‘‘market prices’’ for early warning of crises in emerging economies would therefore be advised to focus on the behavior of real exchange rates and stock prices—not on credit ratings and interest rate spreads. SUMMARY OF RESULTS AND CONCLUDING REMARKS 97 Institute for International Economics | http://www.iie.com Eighth, in most banking and currency crises, a high proportion of the monthly leading indicators— on the order of 50 to 75 percent— reach their signaling thresholds. Indeed, both in and out of sample, we found that fewer than one-sixth of crises occurred with only five or fewer of the 15 monthly leading indicators flashing. In other words, when an emerging economy is lurching toward a financial crisis, many of the wheels come off simultaneously. Ninth, although we have just scratched the surface on testing our leading indicators out of sample, we are encouraged by the initial results— at least for currency crises. We considered two out-of-sample periods: an 18-month period running from the beginning of 1996 to the end of June 1997 (just before the outbreak of the Asian financial crisis) and a 24-month period running from January 1996 to the end of December 1997. Recall that because the indicators lead crises by anywhere from 10 to 18 months, part of the prediction period will lie outside the out-ofsample observation period. In each period, we concentrated on the ordinal ranking of countries according to their crisis vulnerability.3 In chapter 5 we also illustrate for a subset of the countries how one can calculate from this vulnerability index the probability of a crisis for a given country over time. As regards vulnerability to currency crises, the results for the two out-of-sample periods were quite similar. The five most vulnerable countries (in descending order) for the 1996 to mid-1997 period were as follows: South Africa, Czech Republic, Thailand, South Korea, and the Philippines (table 8.2). For the somewhat longer 1996 to end of December 1997 period, the list of the five most vulnerable countries is quite similar, although their ordinal ranking is slightly different: Czech Republic, South Korea, Thailand, South Africa, and Colombia. If the list were extended to the top seven, Malaysia would have been included in both periods. Perhaps the first question to ask is how many of the countries estimated to be most vulnerable to currency crises in the out-of-sample periods turned out to have undergone such crises? The answer, as shown in the upper panel of table 8.2, is almost all of them. According to our index of exchange market pressure, the Czech Republic, Thailand, South Korea, and the Philippines all experienced currency crises in 1997 (that is, depreciations or reserve losses that pushed the index of exchange market pressure to three standard deviations or more above its mean). Colombia’s currency 3. Our preferred measure of vulnerability was an index equal to the weighted average of ‘‘good’’ indicators issuing signals in the out-of-sample period. By ‘‘good’’ indicators, we mean those that had noise-to-signal ratios less than unity during the 1970-95 period. Taking the monthly and annual indicators as a group, there were 18 ‘‘good’’ indicators. We used the inverse of the noise-to-signal ratios as weights for the better indicators. We then ranked each of the 25 countries in the sample according to the computed value of this index. The index is meant to capture the probability of a crisis—not necessarily its severity. 98 ASSESSING FINANCIAL VULNERABILITY Institute for International Economics | http://www.iie.com Table 8.2 Country rankings of vulnerability to currency crises for two periodsa January 1996-June 1997 Country Most vulnerable South Africa Czech Republic Thailand South Korea Philippines Least vulnerable Chile Venezuela Uruguay Mexico Peru Rank 1 2 3 4 5 16 17 18 19 19 Experienced crisisb January 1996-December 1997 Country Czech Republic South Korea Thailand South Africa Colombia Chile Peru Venezuela Mexico Uruguay Rank 1 2 3 4 5 16 17 18 19 20 Experienced crisisb * * * * * * * * a. Weighted index is a sum of the weighted signals flashing at any time during the specified period. Monthly and annual indicators are included. Weights are equal to the inverse noiseto-signal ratios of the respective indicators. b. An asterisk (*) indicates that the country experienced a crisis during the out-of-sample period. crisis arrives later, in the summer of 1998. Moreover, while South Africa did not formally make the cut, it could reasonably be classified as a near miss since it experienced a quasi-crisis in June 1998 (a 14 percent devaluation cum a 13 percent decline in reserves that pushed the exchange market pressure index 2.7 standard deviations above its mean). Malaysia, which just makes it into the group of the seven most vulnerable, did have a currency crisis in 1997. Further information on the out-of-sample performance of the leading indicators of currency crisis can be gleaned by looking for episodes in which, to borrow from Sherlock Holmes, the ‘‘dogs were not barking’’— that is, by looking to see how often crises occurred among those countries estimated to have relatively low vulnerability. The lower panel of table 8.2 indicates the five countries that were estimated to have relatively low vulnerability to currency crises in 1996-97. As with the high vulnerability group, the ordinal rankings of countries are very similar across the two out-of-sample periods, with Venezuela, Peru, and Uruguay slightly shifting their relative positions in the least vulnerable list. Perhaps an explanation as to why the index of vulnerability is relatively low for some of these countries can be found in the fact that some of these countries were still recovering from earlier crises (Mexico and Venezuela). But what about Indonesia, which after all suffered the most severe currency crisis (beginning in the summer of 1997) among the sample SUMMARY OF RESULTS AND CONCLUDING REMARKS 99 Institute for International Economics | http://www.iie.com countries during the out-of-sample period? Why did the model miss it altogether?4 The explanation probably lies in two areas. First, most of the best-performing (higher weight) leading indicators were not flashing in Indonesia’s case. For example, in mid-1997 (just before the outbreak of the Thai crisis), the real effective exchange rate of the Indonesian rupiah was only 4 percent above its long-term average—far below its critical threshold. In a similar vein, neither the decline in stock prices, nor the decline in exports, nor the change in the ratio of M2 money balances to international reserves had hit their threshold values.5 Second, at least three of the factors important in the Indonesian crisis are not included in our list of indicators: namely, currency/liquidity mismatches on the part of the corporate sector, regional cross-country contagion effects, and political instabilities (in this case associated with the Suharto regime). In this connection, work reported in Kaminsky and Reinhart (2000) and extended in chapter 6 suggests that the withdrawal of a common bank lender (in this case, European and Japanese banks) had a lot to do with contagion in emerging Asia—and Indonesia in particular—after the outbreak of the Thai crisis. The failure of our leading indicators to anticipate the Indonesian crisis should not, however, obscure the fact that, of the five countries most adversely affected by the Asian crisis (Thailand, South Korea, Indonesia, Malaysia, and the Philippines), the indicators placed three of them (Thailand, South Korea, and the Philippines) in the top vulnerability group and another (Malaysia) in the upper third of the country vulnerability rankings. Given the well-documented failure of private credit ratings and interest rate spreads to anticipate these Asian currency crises (with the possible exception of Thailand), and given that these forecasts are based solely on own-country fundamentals (that is, with no help from contagion variables), this performance on relative-country vulnerabilities is noteworthy. By the same token, the relatively high estimated vulnerability of several of the Asian emerging economies also challenges the oft-heard view that the crisis was driven primarily by investor panic, with little basis in weak country fundamentals.6 4. It should be recognized that none of the existing early warning models—including the regression-based models—anticipated the Indonesian crisis. 5. Indonesia’s equity prices did suffer a severe decline, but it did not begin until August 1997. 6. Using a very similar approach but a slightly different set of indicators, Kaminsky (1998), who presents a time series of calculated crisis probabilities for the Asian economies, finds results that are in line with those shown in tables 1.6 and 1.7—namely, that estimated currency-crisis vulnerability increased markedly before the 1997 event in Thailand and moderately in Malaysia and the Philippines. Again, no such increase in estimated vulnerability was present for Indonesia. South Korea was not in her sample. Radelet and Sachs (1998) take the opposing view that the crisis in Asia was mainly attributable to investor panic. As discussed in chapter 6, we only find that argument to be convincing in the case of Indonesia. 100 ASSESSING FINANCIAL VULNERABILITY Institute for International Economics | http://www.iie.com Table 8.3 Country rankings of vulnerability to banking crises for two periodsa January 1996-June 1997 Country Most vulnerable Czech Republic South Korea Greece South Africa Thailand Least vulnerable Venezuela Chile Peru Uruguay Mexico Rank 1 2 3 4 5 15 16 17 18 19 Experienced crisisb January 1996-December 1997 Country Czech Republic South Korea Thailand South Africa Colombia Chile Argentina Venezuela Peru Uruguay Rank 1 2 3 4 5 16 17 18 19 20 Experienced crisisb * * * * * a. Weighted index is a sum of the weighted signals flashing at any time during the specified period. Monthly and annual indicators are included. Weights are equal to the inverse noiseto-signal ratios of the respective indicators. b. An asterisk (*) indicates that the country experienced a crisis during the out-of-sample period. Turning to banking crises, the ordinal rankings of country vulnerability again are quite similar across the two out-of-sample periods, although the correspondence is slightly lower than was the case for currency crises: four of the five countries estimated to be most vulnerable to banking crises are the same across the two periods. Specifically, for the 1996 to mid-1997 period, the five most vulnerable countries (again in descending order) were Czech Republic, South Korea, Greece, South Africa, and Thailand (table 8.3). When the out-of-sample period is extended through the end of 1997, Greece drops out of the top five and is replaced by Colombia. As with the vulnerability rankings for currency crises, it is useful to ask which of the countries estimated to be most vulnerable to banking crises actually suffered that fate during the out-of-sample periods. As suggested earlier, this is intrinsically a tougher question to answer for banking crises than for currency crises because the identification and dating of crises are subject to wider margins of error. Recall also that because our 24-month early warning window for banking crises covers both the 12-month period preceding the beginning of the crisis as well as the 12-month period following the onset, successful predictions would include some crises that began toward the end of 1995 and some that started no later than early 1998 (as well as those that began in 1996 or 1997). With these caveats in mind, the picture painted by table 8.3 can be summarized as follows. Of the five countries estimated to be most vulneraSUMMARY OF RESULTS AND CONCLUDING REMARKS 101 Institute for International Economics | http://www.iie.com ble during January 1996 to the end of June 1997, two experienced banking crises that fall in our prediction window. Specifically, we consider South Korea’s banking crisis to have begun in January 1997, with the loan losses stemming from the bankruptcy of Hanbo Steel. In a similar vein, we date Thailand’s banking crisis as starting in May 1996, when the Ministry of Finance took control of Bangkok Bank of Finance (following a run on deposits). A third member of the most vulnerable group, the Czech Republic, also experienced a banking crisis although the timing is not clear-cut. The start of the Czech crisis could be dated in August 1996, reflecting the closure of Kreditni Banka; alternatively, one could also defend a much earlier starting date, namely September 1993, when Kreditni was initially placed under supervision.7 Some researchers (e.g., Kaminsky and Reinhart 1999) also classify Malaysia and the Philippines as having registered banking crises in 1997. The results for the longer out-of-sample period, shown in the upper panel of table 8.3, are quite similar: the same three countries (South Korea, Thailand, and Czech Republic) make up the list of successful banking-crisis predictions. For the more recent sample, Colombia is also added to the list of successful predictions. In April 1998 the bailout by Banco de la Republica of several finance companies facing severe difficulties with mounting losses in their loan portfolios intensified in earnest, and banking sector problems deepened throughout 1998-99. What about the group of countries estimated to be least vulnerable to banking crises? As seen in the lower panel of table 8.3, four of the five countries in this category are common to both sample periods: Uruguay, Venezuela, Peru, and Chile. Mexico appears only in the shorter period, while Argentina makes the least vulnerable list only in the longer period. For all five of the countries estimated to be least vulnerable, no banking crisis appears to have taken place during the out-of-sample periods. As was the case with the forecasting of currency crises, Indonesia (which is ranked 11 or 12, depending on the sample chosen) emerges as a major misclassification, although timing problems somewhat cloud the issue. Many observers would regard the severity of Indonesia’s financial sector problems in 1997 as constituting a ‘‘new’’ banking crisis; others might argue that these difficulties constituted a continuation of the banking problems that began in 1992 with the collapse of Bank Summa. In any case, it is clear that the model was not picking up the increase in Indonesia’s vulnerability in 1997. Mexico presents another timing problem. Mexico remained in the throes of a banking crisis throughout the out-of-sample period and thus could be classified as highly vulnerable. At the same time, most studies (e.g., Demirguc-Kunt and Detragiache 1998; IMF 1998b) ¨¸ regard the Mexican banking crisis as having started at least as early as 7. The Czech banking crisis was not included in our in-sample test, and hence the model is not calibrated to account for this crisis. 102 ASSESSING FINANCIAL VULNERABILITY Institute for International Economics | http://www.iie.com 1994. Here, too, the model seems to have difficulty in identifying changes in vulnerability when they occur in the context of continuing banking problems. Looking at both the high and low vulnerability groups, it is clear that the early warning model is less successful out of sample in anticipating banking crises than it is in anticipating currency crises. The problem is not so much that the model misses many banking crises that do occur but rather that it generates too many false positives or ‘‘noise,’’—that is, it predicts more cases of banking crisis vulnerability than actually occur. In this connection, it is worth noting that we classify only five or six episodes as meeting our criteria for a banking crisis during the out-ofsample period (that is, the period running roughly from late 1995 to early 1998). This list comprises South Korea, Thailand, the Czech Republic, Indonesia, and Malaysia.8 Of these five crisis cases, three of the countries concerned (South Korea, Thailand, and the Czech Republic) were members of our ‘‘most vulnerable’’ group.9 This might be considered fair performance. Difficulties in forecasting Asian banking crises in 1997 seem to be common to the leading forecasting models —be they signals approach models or regression-based models. For example, Demirguc¨¸ Kunt and Detragiache (1998), using a multivariate logit model, report that the conditional probabilities for banking crises in the five most adversely affected Asian economies were actually below the unconditional crisis probabilities (Furman and Stiglitz 1998). Similarly, Kaminsky (1998) finds that estimated crisis probabilities were rising sharply in the case of the Thai banking crisis and moderately in the case of the Philippines, but not for either Malaysia or Indonesia. We conducted a number of experiments, which are described in chapter 5, to help gauge the robustness of our results on the ordinal ranking of country vulnerability to currency and banking crises. In one exercise, instead of basing the ordinal vulnerability rankings exclusively on weights derived from the noise-to-signal ratios, we looked at both the proportion of indicators signaling a crisis and the proportion of the top eight indicators signaling a crisis. In another exercise, we looked at various indicators signaling both banking and currency crises and calculated ‘‘average’’ vulnerability to banking and currency crises combined. And in yet another set of exercises, we liberalized the optimal thresholds for each of the indicators by 5 percent, thereby making it less likely that we would miss crises that were unfolding, albeit at the cost of predicting crises that never 8. The Malaysian crisis would probably best be regarded as beginning in March 1998, when the central bank announced losses at Sime Bank and elsewhere and when Malaysian President Datuk Seri Mahathir bin Mohamad pledged state funds to prop up weak institutions. 9. Malaysia was ranked fourteenth (out of 25 countries) in the shorter period and tenth in the longer one. SUMMARY OF RESULTS AND CONCLUDING REMARKS 103 Institute for International Economics | http://www.iie.com occurred. While these robustness exercises not surprisingly generated some changes in the ordinal rankings, perhaps the most important finding was that the same ‘‘core’’ set of vulnerable countries—the Czech Republic, South Korea, South Africa, Greece, Colombia, Thailand, the Philippines, and Malaysia—consistently remained in the top tier of the vulnerability list. It is also noteworthy that none of these sensitivity exercises anticipated the Indonesian crisis. All in all, we regard the out-of-sample performance of the signals approach as encouraging— particularly as regards anticipating currency crises in the Asian crisis countries.10 With the exception of Indonesia, the model did well in identifying the countries with relatively high vulnerability. In addition, the model gave strong signals for Brazil, the Czech Republic, South Africa, and Colombia, which also experienced crises (or turbulence) outside the Asian region. The results for banking crises were less impressive. While we would not place much confidence in the precise estimated ordering of vulnerability across countries, we think the signals approach looks promising for making distinctions between the vulnerability of countries near the top of the list and those near the bottom—that is, it may be useful as a ‘‘first screen,’’ which can then be followed by more in-depth country analysis. Some others are pessimistic about both the potential and actual out-ofsample performance of signals-based leading-indicator models of currency crises, including their track record in anticipating the Asian financial crisis.11 They argue that when such models do perform seemingly well, it is often because they rely on ‘‘black box’’ simple contagion variables (for example, the number of crises that have occurred in the previous period), that the methodology embedded in the signals approach is biased toward overpredicting crises in countries with good histories and this explains its successes in predicting currency crises in Asia, and that both in-sample and out-of-sample performance would be better if the good indicator variables were entered linearly (rather than sending a signal only when the indicator crossed its threshold) and if the weights on the individual indicators were estimated by a regression (rather than selected from an iterative noise-to-signal test one at a time). These critics also argue that the correlation between the severity of observed currency crises and crisis vulnerability predicted by the signals approach was low (at least in 1996) and that (also in 1996) there did not seem to be a marked distinction between the calculated currency crisis vulnerabilities of several 10. This is consistent with the results of a recent IMF study (Berg and Pattillo 1999), which found that the signals model of Kaminsky, Lizondo, and Reinhart (1998) did a better job of predicting the Asian crisis than the models of Frankel and Rose (1996) and of Sachs, Tornell, and Velasco (1996). 11. See Berg and Pattillo (1999), The Economist (1998), Furman and Stiglitz (1998), IMF (1998c), and Wyplosz (1997, 1998). 104 ASSESSING FINANCIAL VULNERABILITY Institute for International Economics | http://www.iie.com noncrisis countries (particularly Argentina and Mexico) and the Asian crisis countries (Thailand and Indonesia). Further, they find that leadingindicator models have poor in-sample performance in forecasting currency crises for developing countries, especially in emitting Type II errors (false positives) and that publication of a vulnerability index could precipitate a crisis. We do not find these criticisms to be persuasive. In the two studies (Berg and Pattillo 1999; Furman and Stiglitz 1998) that have explicitly run out-of-sample horse races between the KaminskyReinhart signals model and two other regression-based models of currency crises (Frankel and Rose 1996; Sachs, Tornell, and Velasco 1997), both concluded that the signals approach does better. Wyplosz (1998) bases his pessimistic conclusions on the in-sample performance of leading-indicator models of currency crises in developing countries using the Frankel and Rose (1996) model—not the KaminskyReinhart signals approach. Using an abbreviated search technique for the optimal threshold for various indicators, Wyplosz finds that (using a 5 percent threshold) 62 of 86 currency crises are detected, while the model signals wrongly—that is, emits false positives—in 43 percent of the crises. Our results are more favorable than Wyplosz’s for developing-country currency crises (in-sample). Out of sample, we find that the false positives problem is more serious for the banking crises than for currency crises. While we present some new results on cross-country contagion in chapter 6, the out-of-sample results—both in earlier Kaminsky-Reinhart studies and in this book—do not rely at all on cross-country contagion; instead, they reflect only own-country fundamentals. There is (at least to our knowledge) no empirical evidence to support the view that imposing a common absolute threshold for indicator variables would produce better in-sample and out-of-sample performance than our procedure of imposing a common percentile threshold and allowing the absolute threshold to differ across countries. Nor, as we have argued earlier, does it seem more reasonable on a priori grounds to impose the one-sizefits-all restriction on countries with different histories—quite the contrary. As for the alleged influence of our procedure in the context of forecasting the Asian financial crisis, one would have thought that if this bias were large, it might have led to a very successful prediction of crises in the Asian countries, yet some of these same critics find that the signals approach does very poorly in forecasting currency crises in these countries. While more work is clearly needed to assess the robustness of the results to different out-of-sample periods (since these differ and seem to generate different outcomes across studies), we do not find that there was little distinction in estimated currency-crisis vulnerabilities between most of the Asian crisis countries, on the one hand, and some other (Latin American) noncrisis countries, on the other. As indicated earlier, we found that Thailand, the Philippines, and Malaysia had higher estimated currencySUMMARY OF RESULTS AND CONCLUDING REMARKS 105 Institute for International Economics | http://www.iie.com crisis probabilities in 1996-97 than did Argentina and Mexico—not the other way around. Thailand was near the top of our vulnerability list— not near the bottom. Also, it is not obvious that out-of-sample comparisons based on the severity of crises are more meaningful than those (as above) that concentrate on the crisis/no crisis distinction. In short, just as we emphasized that it is important not to oversell the potential of early warning models to predict crises in emerging economies, we think some of the critics are too quick to dismiss the usefulness of these models because of a mixed out-of-sample performance based on runs from a single period. We should also keep in mind the apparent inability of non-model-based forecasts to foresee the Asian crisis. In our view, much more empirical work will need to be done before we can draw reliable conclusions on the out-of-sample performance of the signals approach. Examining a somewhat more limited sample (20) of small developed and emerging economies over the 1970-98 period, we looked for patterns in the cross-country contagion of currency crises. Following Eichengreen, Rose, and Wyplosz (1996), we define contagion as a case in which the presence of a crisis elsewhere increases the probability of crisis at home, even when the fundamentals have been taken into account. We considered four channels through which shocks can be transmitted across borders: two dealt with trade links (bilateral trade flows and trade competition in third-country markets) and two channels addressed financial links (correlation of asset returns in global portfolios and reliance on a common bank lender). We also demonstrated how these four contagion channels could be combined and weighted appropriately to form a ‘‘contagion vulnerability index.’’ This exercise led us to our tenth main finding: that cross-country contagion adds significantly to own-country fundamentals in furthering an understanding of emerging market vulnerability to financial crises and that (at least historically) contagion has operated more along regional than global lines. According to our contagion vulnerability index, Brazil, Argentina, and the Philippines had high vulnerability to the 1994 Mexican peso crisis; Malaysia, South Korea, and Indonesia had high vulnerability to the 1997 Thai crisis; and Argentina, Chile, and Uruguay had high vulnerability to the 1999 Brazilian crisis. Although it is difficult to separate financial contagion channels from trade channels (since countries linked in trade also are linked in finance) we concluded that withdrawal of a common bank lender (particularly Japanese banks) and high correlation of asset returns were important in the contagion in Asia in 1997-98. Eleventh, in addition to studying the antecedents of crises we also drew on our data base for information on the aftermath of crises—with particular attention to the speed with which emerging economies return to ‘‘normal’’ after a currency or banking crises. We defined normal in 106 ASSESSING FINANCIAL VULNERABILITY Institute for International Economics | http://www.iie.com two alternative ways: first, as a period of ‘‘tranquility’’ that excludes not only the crisis years but also the two- to three-year windows before and after the crisis, and second, as the average of the two years just preceding crises.12 One of our most robust findings was that the deleterious effects on economic activity are more lingering for banking crises than for currency crises.13 For example, whereas it took about two years for economic growth to return to the average of the two precrisis years after a currency crisis, that recovery was not evident even three years after a banking crisis. One possible explanation for this difference is that, whereas a currency crisis sharply reduces external sources of funding, a banking crisis curtails access to both external and domestic sources of finance for households and firms—that is, the ‘‘credit crunch’’ is more severe in the wake of banking crises. This more sluggish recovery pattern for banking crises was also evident for exports, imports, and stock prices. For instance, whereas exports recover relatively quickly (eight months) and ahead of the rest of the economy following currency crises, they continue to sink for two years following the onset of a banking crisis. Two other dimensions of the protracted nature of banking crises are that it takes about three to four years for a banking crisis to be resolved and it takes on the order of a year and half between the onset of a banking crisis and its peak. All of this highlights the challenge faced by the Asian crisis countries in sustaining their recoveries: not only did the most affected countries in emerging Asia suffer from currency crises that were accompanied by banking crises (what Kaminsky and Reinhart 1999 dub ‘‘twin crises’’), but the banking crises themselves are very severe. Our analysis of the aftermath of crises does not lend support to the notion that devaluations in emerging economies generate deflation. Instead, we find that devaluations are inflationary, that the pass-through to prices is incomplete (hence, devaluations lead to real depreciations), and that it takes between two and three years after a devaluation for inflation to return to the average of the two precrisis years. Last but not least, we offer a number of suggestions for improving early warning models of currency and banking crises. In our view, four directions for future research merit priority. As hinted above, more work needs to be done to determine the outof-sample forecasting properties of these models—be it signals approach models or regression-based logit or probit models. In particular, it would be useful to know how robust ‘‘who’s next’’ country rankings of vulnera- 12. More specifically, the ‘‘tranquil’’ period excludes the 24 months before and after currency crises and the 24 months before and 36 months after banking crises. 13. In the cases where currency and banking crises coincide, the postcrisis performance would show up in both the averages for currency and banking crises. See chapter 7 for details. SUMMARY OF RESULTS AND CONCLUDING REMARKS 107 Institute for International Economics | http://www.iie.com bility are in the face of changes in the forecasting period, different composite indicators, different definitions and transformations of the indicator variables (e.g., alternative definitions of the effective real exchange rate or of real exchange rate ‘‘misalignments,’’ and alternative ways of dating banking crises or selecting the early warning ‘‘window’’), and the restrictions imposed in the different models (e.g., imposing thresholds versus allowing indicators to enter linearly, imposing absolute thresholds versus common percentile ones). It may turn out, as suggested by Berg and Pattillo (1999), that combining certain features of the signals approach and the regression-based models would improve forecasting (e.g., using the signals approach to select the good indicators and then estimating the weights and crisis probabilities using a regression-based format). We think there is scope to bring other indicators into these horse races. For example, Kaminsky (1998) has found that the share of short-term debt in total foreign debt, as well as a proxy for capital flight (by residents of emerging economies), do quite well in anticipating currency and banking crises within the sample. Looking at the run-up to the Asian financial crisis, Furman and Stiglitz (1998) likewise make a good case for including the ratio of short-term external debt to international reserves as an indicator in future early warning exercises. If monthly data could be obtained both on real property prices and on the exposure of the banking system to property, those too could prove very helpful. A plausible extension would be to bring institutional characteristics of weak banking systems into the forecasting of banking crises. There is a strong presumption that the following all matter for vulnerability to banking crises: weak accounting, provisioning, and legal frameworks; policy-directed lending; the ownership structure of the banking system (government ownership, foreign ownership, and so on); the incidence of connected lending; the extent of diversification; the quality of banking supervision; and the incentive-compatibility of the official safety net. Yet it is only very recently that any of these factors have begun to enter the empirical literature.14 The main constraint on making use of these institutional characteristics is that one cannot get high-frequency measurements of them. Indeed, for some of these characteristics (e.g., the share of government ownership), it has proved difficult to get even annual data that is less than two or three years old. This means that such variables have to be introduced as zero-one dummy variables in a time-series context. There would be more scope to take advantage of such factors in cross-section work—that is, in explaining cross-country differences in the incidence of banking crises over long periods. 14. See, for example, Demirguc-Kunt and Detragiache (1998), who introduce law enforce¨¸ ment and deposit-insurance variables into their banking crisis model. 108 ASSESSING FINANCIAL VULNERABILITY Institute for International Economics | http://www.iie.com Lastly, we think the ongoing work on modeling the nature of crosscountry contagion of crises should be extended. One of the lessons of the last few major crises (that is, the Mexican crisis and the Asian/global financial crisis) is that the channels of cross-country contagion are more numerous and complicated than we thought earlier. Each of these factors seem to play a part in contagion: trade links (bilateral and third party), perceived similarities in macroeconomic and financial vulnerability, the dynamics of competitive devaluations, induced effects on primary commodity prices, financial links operating via withdrawal of a common bank or mutual fund lender, liquidity and margin-call effects operating via the regulatory framework, and perceived changes in the rescheduling cum capital-account convertibility regime (such as took place after the Russian unilateral rescheduling/default in August 1998 and the Malaysian imposition of wide-ranging capital controls). We need to find ways to incorporate more of these channels of contagion in our forecasting models. Would the Publication of the Indicators Erode Their Early Warning Role? It is sometimes argued that if the indices of crisis vulnerability were made publicly available on a timely basis, such publication could prompt a selffulfilling run on a country’s currency or its banks. Alternatively, it is sometimes asserted that if countries really paid heed to the message of the indicators and took preemptive action, then the indicators would lose predictive power. This latter outcome would, of course, be highly desirable. While neither of these arguments can be dismissed lightly, we would regard both as exaggerated. The conditions for generating this type of self-fulfilling runs are likely to be relatively rare. As we have stressed throughout this book, the signals approach is useful in identifying cases of high vulnerability to crises. Explaining the precise timing of the crises remains an elusive goal. To the extent that timing matters and that investment decisions are made under uncertainty, there is little reason to expect that moderate increases in the extent of vulnerability are likely to be sufficient to prompt a speculative attack. As argued earlier, negative announcements of the readings in the leading indicator index of business cycles—which have been published for many years—do not cause a recession, although investors certainly take these readings into account. By the same token, we think it unlikely that publication would cause the indicators to lose most of their predictive ability. This could certainly occur if preemptive policy was an everytime, everywhere phenomenon and if such preemptive policies were in fact successful in staving off crises. These are strong assumptions. All too often, policymakers are inclined to ignore distress signals on the grounds that, this time, the situation is really different or that there are overriding political objectives SUMMARY OF RESULTS AND CONCLUDING REMARKS 109 Institute for International Economics | http://www.iie.com against corrective action. Furthermore, even if the signals are heeded and corrective policy actions are taken, they may not be sufficient to prevent the crisis. If the feedback from the indicators to corrective policy action were strong and consistent, we would not have been able to identify useful indicators in the first place. Of course, one could always speculate that future policymakers will be wiser than their predecessors, but that remains to be seen. Do the Better Performing Indicators Carry Policy Implications? The empirical evidence presented in this book can be seen as supporting the case for including leading indicators in the analytical tool kit for diagnosing crisis vulnerability. But can one go farther and draw policy implications from the performance of the better univariate indicators? One should recall that the signals approach outlined in this book and in earlier works by Kaminsky and Reinhart looks for empirical regularities in the behavior of macroeconomic and financial variables in the run-up to currency and banking crises. We thus cannot fully identify from this exercise the channels by which policies affect economic outcomes. For some of the indicators, the results have clear implications for macroeconomic and exchange rate policies; for others there is no obvious link. For example, the performance of M2/reserves as an indicator of currency crises is suggestive of the desirability of avoiding large discrepancies between liquid liabilities and liquid assets. In this regard, the policy implication would be to encourage emerging market countries to maintain high liquidity ratios, prearranged lines of credit, and an ample stock of reserves. Much of the behavior of other indicators, notably rising real interest rates and money multipliers, are associated with financial liberalization. Indeed, the reliability of these indicators in anticipating banking crises may warn against hasty liberalizations. On the other hand, real exchange rate overvaluations are an important indicator of both currency and banking crises, but the burning policy question that remains unanswered is how emerging market countries can avoid these costly periodic overvaluations. Real exchange rate targeting has been tried by countries as diverse as Brazil, Chile, Indonesia, and Colombia, but the outcomes were quite dissimilar—particularly as regards their consequences for inflation.15 In the case of some other indicators, such as stock prices, the policy implications are even less obvious. In short, while we regard the empirical work on early warning indicators as consistent with many stories of the origins of currency and banking crises, one has to be careful not to overinterpret the results, as alternative explanations of crises often yield observationally equivalent implications. ´ 15. Calvo, Reinhart and Vegh (1995) present empirical evidence on this issue. 110 ASSESSING FINANCIAL VULNERABILITY Institute for International Economics | http://www.iie.com Appendix Data and Definitions Currency Crisis Index The index is a weighted average of exchange rate and reserve changes, with weights such that the two components of the index have equal conditional volatilities. Since changes in the exchange rate enter with a positive weight and changes in reserves have a negative weight attached, readings of this index that were three standard deviations or more above the mean were cataloged as crises. For countries in the sample that had hyperinflation, the construction of the index was modified. While a 100 percent devaluation may be traumatic for a country with low to moderate inflation, a devaluation of that magnitude is commonplace during hyperinflations. A single index for the countries that had hyperinflation episodes would miss sizable devaluations and reserve losses in the moderate inflation periods, as the historic mean is distorted by the high-inflation episode. To avoid this, we divided the sample according to whether inflation in the previous six months was higher than 150 percent and then constructed an index for each subsample. Our cataloging of crises for these countries coincides fairly tightly with our chronology of currency market disruptions. Eichengreen, Rose, and Wyplosz (1995) also include interest rates in this index; however, our data on market-determined interest rates for developing countries does not span the entire sample. The Indicators Sources include the IMF’s International Financial Statistics (IFS), the International Finance Corporation’s (IFC) Emerging Market Indicators, and the 111 Institute for International Economics | http://www.iie.com World Bank’s World Development Indicators. When data were missing from these sources, central bank bulletins and other country-specific sources were used as supplements. Unless otherwise noted, we used 12month percentage changes. M2 multiplier: the ratio of M2 to base money (IFS lines 34 plus 35) divided by IFS line 14. Domestic credit/nominal GDP: IFS line 52 divided by IFS line 99b (interpolated). Monthly nominal GDP was interpolated from annual or quarterly data. Real interest rate on deposits: IFS line 601, monthly rates, deflated using consumer prices (IFS line 64) expressed in percentage points. Ratio of lending rate to deposit rate: IFS line 60p divided by IFS line 601 was used in lieu of differentials to ameliorate the distortions caused by the large percentage point spreads observed during high inflation. In levels. Excess real M1 balance: M1 (IFS line 34) deflated by consumer prices (IFS line 64) less an estimated demand for money. The demand for real balances is determined by real GDP (interpolated IFS line 99b), domestic consumer price inflation, and a time trend. Domestic inflation was used in lieu of nominal interest rates, as market-determined interest rates were not available during the entire sample for a number of countries; the time trend (which can enter log-linearly, linearly, or exponentially) is motivated by its role as a proxy for financial innovation and/or currency substitution. Excess money supply (demand) during precrisis periods (mc) is reported as a percentage relative to excess supply (demand) during tranquil times (mt)—that is, 100 (mc mt)/mt. M2 (in US dollars)/reserves (in US dollars): IFS lines 34 plus 35 converted into dollars (using IFS line ae) divided by IFS line 1L.d. Bank deposits: IFS line 24 plus 25. Exports (in US dollars): IFS line 70. Imports (in US dollars): IFS line 71. Terms of trade: the unit value of exports (IFS line 74) over the unit value of imports (IFS line 75). For those developing countries where import unit values (or import price indices) were not available, an index of prices of manufactured exports from industrial countries to developing countries was used. Real exchange rate: based on consumer price indices (IFS line 64) and defined as the relative price of foreign goods (in domestic currency) to the price of domestic goods. If the central bank of the home country pegs the currency to the dollar (or deutsche mark), the relevant foreign price index is that of the United States (or Germany). Hence for all the European countries the foreign price index is that of Germany, while for all the other countries consumer prices in the United States were used. The trend was specified as, alternatively, log-linear, linear, and exponential; the best 112 ASSESSING FINANCIAL VULNERABILITY Institute for International Economics | http://www.iie.com fit among these was selected on a country-by-country basis. Deviations from trend during crisis periods (dc) were compared with the deviations during tranquil times (dt) as a percentage of the deviations in tranquil times (i.e., 100 (dc dt)/dt). Reserves: IFS line 1L.d. Domestic-foreign interest rate differential on deposits: monthly rates in percentage points (IFS line 601). Interest rates in the home country are compared with interest rates in the United States (or Germany) if the domestic central bank pegs the currency to the dollar (or deutsche mark). The real interest rate is given by 100 [(1 it )pt /pt 1 1 ]. Output: for most countries, industrial production (IFS line 66). However, for some countries (the commodity exporters) an index of output of primary commodities is used (IFS line 66aa). Stock prices (in dollars): IFC global indices are used for all emerging markets; for industrial countries the quotes from the main bourses are used. Overall budget balance/GDP: consolidated public-sector balance as share of nominal GDP (World Bank Debt Tables). Current account balance a share of GDP: (World Bank, World Development Report database available in CD ROM). Current account balance a share of investment: current account divided by gross investment (World Bank, World Development Report database available in CD ROM). Short-term capital inflows: Short-term capital flows as a percent of GDP, (World Bank, World Debt Tables, database available in CD ROM). Foreign direct investment (FDI): FDI as a share of GDP (World Bank, World Debt Tables, database available in CD ROM). General government consumption/GDP: General government consumption, national income accounts basis as a percent of GDP, annual growth rate (World Bank, World Development Report database available in CD ROM). Central bank credit to the public sector/GDP: Annual growth rate (World Bank, World Development Report database available in CD ROM). Net credit to the public sector/GDP: Annual growth rate (World Bank, World Development Report database available in CD ROM). DATA AND DEFINITIONS 113 Institute for International Economics | http://www.iie.com