economics paper reinhart

8/8/2019 Economics Paper Reinhart

1/28

The Twin Crises: The Causes of Banking and

Balance-of-Payments Problems

By GRACIELA L. KAMINSKY AND CARMEN M. REINHART*

In the wake of the Mexican and Asian currency turmoil, the subject of financialcrises has come to the forefront of academic and policy discussions. This paperanalyzes the links between banking and currency crises. We find that: problems inthe banking sector typically precede a currency crisisthe currency crisis deepensthe banking crisis, activating a vicious spiral; financial liberalization often precedesbanking crises. The anatomy of these episodes suggests that crises occur as theeconomy enters a recession, following a prolonged boom in economic activity thatwas fueled by credit, capital inflows, and accompanied by an overvalued currency.

(JEL F30, F41)

Pervasive currency turmoil, particularly inLatin America in the late 1970s and early1980s, gave impetus to a flourishing literatureon balance-of-payments crises. As stressed inPaul Krugmans (1979) seminal paper, in thisliterature, crises occur because a country fi-nances its fiscal deficit by printing money to theextent that excessive credit growth leads to theeventual collapse of the fixed exchange-rate

regime. With calmer currency markets in themid- and late 1980s, interest in this literaturelanguished. The collapse of the EuropeanExchange Rate Mechanism, the Mexican pesocrisis, and the wave of currency crises sweeping

through Asia have, however, rekindled interestin the topic. Yet, the focus of this recent liter-ature has shifted. While the earlier literatureemphasized the inconsistency between fiscaland monetary policies and the exchange-ratecommitment, the new one stresses self-fulfillingexpectations and herding behavior in interna-tional capital markets.1 In this view, as Calvo(1995 p. 1) summarizes: If investors deem you

unworthy, no funds will be forthcoming and,thus, unworthy you will be.

Whatever the causes of currency crises, nei-ther the old literature nor the new models ofself-fulfilling crises have paid much attention tothe interaction between banking and currencyproblems, despite the fact that many of thecountries that have had currency crises havealso had full-fledged domestic banking crisesaround the same time. Notable exceptions are:Carlos F. Daz-Alejandro (1985), AndresVelasco (1987), Calvo (1995), Ilan Goldfajnand Rodrigo O. Valdes (1995), and VictoriaMiller (1995). As to the empirical evidence onthe potential links between what we dub thetwin crises, the literature has been entirely si-lent. The Thai, Indonesian, and Korean crisesare not the first examples of dual currency andbanking woes; they are only the recent additionsto a long list of casualties which includes Chile,Finland, Mexico, Norway, and Sweden.

* Kaminsky: Department of Economics, George Wash-ington University, Washington, DC 20052; Reinhart:School of Public Affairs and Department of Economics,University of Maryland, College Park, MD 20742, and theNational Bureau of Economic Research. We thank twoanonymous referees for very helpful suggestions. We alsothank Guillermo Calvo, Rudiger Dornbusch, Peter Montiel,Vincent Reinhart, John Rogers, Andrew Rose, and seminarparticipants at Banco de Mexico, the Board of Governors ofthe Federal Reserve System, Florida State University, Har-vard University, the International Monetary Fund, JohnsHopkins University, Massachusetts Institute of Technology,Stanford University, the State University of New York-Albany, the University of California-Berkeley, UCLA, theUniversity of California-Santa Cruz, the University ofMaryland, the University of Washington, the World Bank,and the conference on Speculative Attacks in the Era of theGlobal Economy: Theory, Evidence, and Policy Implica-tions (Washington, DC, December 1995) for very help-ful comments, and Greg Belzer, Kris Dickson, and NoahWilliams for superb research assistance.

1 See Maurice Obstfeld (1994, 1995) and Guillermo A.Calvo (1995).

473


2/28

In this paper, we aim to fill this void in theliterature and examine currency and banking cri-ses episodes for a number of industrial and devel-oping countries. The former include: Denmark,Finland, Norway, Spain, and Sweden. The latter

focus on: Argentina, Bolivia, Brazil, Chile, Co-lombia, Indonesia, Israel, Malaysia, Mexico, Peru,the Philippines, Thailand, Turkey, Uruguay, andVenezuela. The period covered spans the 1970sthrough 1995. This sample gives us the opportu-nity to study 76 currency crises and 26 bankingcrises. Out of sample, we examine the twin crisesin Asia of 1997.

Charles Kindelberger (1978 p. 14), in study-ing financial crises, observes: For historianseach event is unique. Economics, however,

maintains that forces in society and nature be-have in repetitive ways. History is particular;economics is general. Like Kindelberger, weare interested in finding the underlying commonpatterns associated with financial crises. Tostudy the nature of crises, we construct a chro-nology of events in the banking and externalsectors. From this timetable, we draw inferenceabout the possible causal patterns among bank-ing and balance-of-payments problems and fi-nancial liberalization. We also examine thebehavior of macroeconomic indicators that have

been stressed in the theoretical literature aroundcrisis periods, much along the lines of BarryEichengreen et al. (1996b). Our aim is to gaugewhether the two crises share a common macro-economic background. This methodology also al-lows us to assess the fragility of economies aroundthe time of the financial crises and sheds light onthe extent to which the crises were predictable.Our main results can be summarized as follows.

First, with regard to the linkages among thecrises, our analysis shows no apparent link be-

tween balance-of-payments and banking crisesduring the 1970s, when financial markets werehighly regulated. In the 1980s, following theliberalization of financial markets across manyparts of the world, banking and currency crisesbecome closely entwined. Most often, the be-ginning of banking-sector problems predate thebalance-of-payment crisis; indeed, knowing thata banking crisis was underway helps predict afuture currency crisis. The causal link, never-theless, is not unidirectional. Our results showthat the collapse of the currency deepens thebanking crisis, activating a vicious spiral. We

find that the peak of the banking crisis mostoften comes after the currency crash, suggestingthat existing problems were aggravated or newones created by the high interest rates requiredto defend the exchange-rate peg or the foreign-

exchange exposure of banks.Second, while banking crises often precede

balance-of-payments crises, they are not neces-sarily the immediate cause of currency crises,even in the cases where a frail banking sectorputs the nail in the coffin of what was already adefunct fixed exchange-rate system. Our resultspoint to common causes, and whether the cur-rency or banking problems surface first is amatter of circumstance. Both crises are pre-ceded by recessions or, at least, below normal

economic growth, in part attributed to a wors-ening of the terms of trade, an overvalued ex-change rate, and the rising cost of credit;exports are particularly hard hit. In both types ofcrises, a shock to financial institutions (possiblyfinancial liberalization and/or increased accessto international capital markets) fuels the boomphase of the cycle by providing access to fi-nancing. The financial vulnerability of the econ-omy increases as the unbacked liabilities of thebanking-system climb to lofty levels.

Third, our results show that crises (external

or domestic) are typically preceded by a multi-tude of weak and deteriorating economic fun-damentals. While speculative attacks can and dooccur as market sentiment shifts and, possibly,herding behavior takes over (crises tend to bebunched together), the incidence of crises wherethe economic fundamentals were sound are rare.

Fourth, when we compared the episodes inwhich currency and banking crises occurredjointly to those in which the currency or bank-ing crisis occurred in isolation, we find that for

the twin crises, economic fundamentals tendedto be worse, the economies were considerablymore frail, and the crises (both banking andcurrency) were far more severe.

The rest of the paper is organized as follows.The next section provides a chronology of thecrises and their links. Section II reviews thestylized facts around the periods surroundingthe crises, while Section III addresses the issuesof the vulnerability of economies around thetime of the crisis and the issue of predictability.The final section discusses the findings and pos-sibilities for future research.

474 THE AMERICAN ECONOMIC REVIEW JUNE 1999


3/28

I. The Links Between Banking

and Currency Crises

This section briefly discusses what the theo-retical literature offers as explanations of the

possible links between the two crises. The the-oretical models also guide our choice of thefinancial and economic indicators used in theanalysis.

A. The Links: Theory

A variety of theoretical models have been putforth to explain the linkages between currencyand banking crises. One chain of causation,stressed in James Stoker (1994), runs from

balance-of-payments problems to banking cri-sis. An initial external shock, such as an in-crease in foreign interest rates, coupled with acommitment to a fixed parity, will result in theloss of reserves. If not sterilized, this will lead toa credit crunch, increased bankruptcies, and fi-nancial crisis. Moreover, Frederic S. Mishkin(1996) argues that, if a devaluation occurs, theposition of banks could be weakened further ifa large share of their liabilities is denominatedin a foreign currency. Models, such as Velasco(1987), point to the opposite causal direction

financial-sector problems give rise to the cur-rency collapse. Such models stress that whencentral banks finance the bailout of troubledfinancial institutions by printing money, we re-turn to the classical story of a currency crashprompted by excessive money creation.

A third family of models contend that cur-rency and banking crises have common causes.An example of this may be found in the dynam-ics of an exchange-rate-based inflation stabili-zation plan, such as that of Mexico in 1987.

Theory and evidence suggest that such planshave well-defined dynamics2: Because inflationconverges to international levels only gradually,there is a marked cumulative real exchange-rateappreciation. Also, at the early stages of theplan there is a boom in imports and economicactivity, financed by borrowing abroad. As thecurrent account deficit continues to widen, fi-nancial markets become convinced that the sta-

bilization program is unsustainable, fueling anattack against the domestic currency. Since theboom is usually financed by a surge in bankcredit, as banks borrow abroad, when the capitalinflows become outflows and asset markets

crash, the banking system caves in. Ronald I.McKinnon and Huw Pill (1996) model howfinancial liberalization together with microeco-nomic distortionssuch as implicit deposit in-surancecan make these boom-bust cycleseven more pronounced by fueling the lendingboom that leads to the eventual collapse of thebanking system. Goldfajn and Valdes (1995)show how changes in international interest ratesand capital inflows are amplified by the inter-mediating role of banks and how such swings

may also produce an exaggerated business cyclethat ends in bank runs and financial and cur-rency crashes.

So, while theory does not provide an unam-biguous answer as to what the causal links be-tween currency and banking crises are, themodels are clear as to what economic indicatorsshould provide insights about the underlyingcauses of the twin crises. High on that list areinternational reserves, a measure of excessmoney balances, domestic and foreign interestrates, and other external shocks, such as the

terms of trade. The inflation stabilization-financial liberalization models also stress theboom-bust patterns in imports, output, capitalflows, bank credit, and asset prices. Some ofthese models also highlight overvaluation of thecurrency, leading to the underperformance ofexports. The possibility of bank runs suggestsbank deposits as an indicator of impending cri-ses. Finally, as in Krugman (1979), currencycrises can be the by-product of governmentbudget deficits.

B. The Links: Preliminary Evidence

To examine these links empirically, we firstneed to identify the dates of currency and bank-ing crises. In what follows, we begin by describ-ing how our indices of financial crises areconstructed.

Definitions, Dates, and Incidence of Crises.Most often, balance-of-payments crises areresolved through a devaluation of the domesticcurrency or the floatation of the exchange rate.

2 See Reinhart and Carlos A. Vegh (1996) for a review ofthis literature and the empirical regularities.

475VOL. 89 NO. 3 KAMINSKY AND REINHART: THE TWIN CRISES


4/28

But central banks can and, on occasion, doresort to contractionary monetary policy andforeign-exchange market intervention to fightthe speculative attack. In these latter cases, cur-rency market turbulence will be reflected in

steep increases in domestic interest rates andmassive losses of foreign-exchange reserves.Hence, an index of currency crises should cap-ture these different manifestations of specula-tive attacks. In the spirit of Eichengreen et al.(1996a, b), we constructed an index of currencymarket turbulence as a weighted average ofexchange-rate changes and reserve changes.3

With regard to banking crises, our analysisstresses events. The main reason for followingthis approach has to do with the lack of high-

frequency data that capture when a financialcrisis is under way. If the beginning of a bank-ing crisis is marked by bank runs and withdraw-als, then changes in bank deposits could be usedto date the crises. Often, the banking problemsdo not arise from the liability side, but from aprotracted deterioration in asset quality, be itfrom a collapse in real-estate prices or increasedbankruptcies in the nonfinancial sector. In thiscase, changes in asset prices or a large increasein bankruptcies or nonperforming loans couldbe used to mark the onset of the crisis. For some

of the earlier crises in emerging markets, how-ever, stock-market data is not available.4 Indi-cators of business failures and nonperformingloans are also usually available only at lowfrequencies, if at all; the latter are also madeless informative by banks desire to hide theirproblems for as long as possible.

Given these data limitations, we mark thebeginning of a banking crisis by two types ofevents: (1) bank runs that lead to the closure,merging, or takeover by the public sector of one

or more financial institutions (as in Venezuelain 1993); and (2) if there are no runs, the clo-sure, merging, takeover, or large-scale govern-ment assistance of an important financial

institution (or group of institutions) that marksthe start of a string of similar outcomes for otherfinancial institutions (as in Thailand in 19961997). We rely on existing studies of bankingcrises and on the financial press; according to

these studies the fragility of the banking sectorwas widespread during these periods. This ap-proach to dating the beginning of the bankingcrises is not without drawbacks. It could datethe crises too late, because the financial prob-lems usually begin well before a bank is finallyclosed or merged; it could also date the crisestoo early, because the worst of crisis may comelater. To address this issue we also date whenthe banking crisis hits its peak, defined as theperiod with the heaviest government interven-

tion and/or bank closures.Our sample consists of 20 countries for theperiod 1970mid-1995. The countries are thoselisted in the introduction and Appendix TablesA1 and A2. We selected countries on the mul-tiple criteria of being small open economies,with a fixed exchange rate, crawling peg, orband through portions of the sample; data avail-ability also guided our choices. This period en-compasses 26 banking crises and 76 currencycrises.

As to the incidence of the crises (Table 1 and

Figure 1), there are distinct patterns across de-cades. During the 1970s we observe a total of26 currency crises, yet banking crises were rareduring that period, with only three taking place.The absence of banking crises may reflect thehighly regulated nature of financial marketsduring the bulk of the 1970s. By contrast, whilethe number of currency crises per year does notincrease much during the 1980s and 1990s(from an average of 2.60 per annum to 3.13 perannum, Table 1, first row), the number of bank-

ing crises per year more than quadruples in thepost-liberalization period. Thus, as the secondrow of Table 1 highlights, the twin crisis phe-nomenon is one of the 1980s and 1990s.

Figure 1 also shows that financial crises wereheavily bunched in the early 1980s, when realinterest rates in the United States were at theirhighest level since the 1930s. This may suggestthat external factors, such as interest rates in theUnited States, matter a great deal as argued inCalvo et al. (1993). Indeed, Jeffrey Frankel andAndrew K. Rose (1996) find that foreign inter-est rates play a significant role in predicting

3 The construction of the index is described in the DataAppendix. The dates of the crises appear in Appendix TableA1, and the level of the index and key events around thecrises dates are reported in the working paper version of thispaper (Kaminsky and Reinhart, 1996).

4 Bank stocks could be an indicator, but in many of thedeveloping countries an important share of the banks are nottraded publicly.



5/28

currency crashes. A second explanation why

crises are bunched is that contagion effects maybe present, creating a domino effect amongthose countries that have anything less thanimmaculate fundamentals. Sara Calvo andReinhart (1996) present evidence of contagionin capital flows to Latin American countrieswhile Eichengreen et al. (1996a) find evidencethat knowing there is a crisis elsewhere in-creases the probability of a domestic currencycrisis.

Table 2 provides the dates of financial liber-alization, the beginning and peak of the bankingcrisis, and the date of the balance-of-payments

crisis that was nearest to the beginning of the

banking crisis.5

By selecting the nearest cur-rency crisis, whether it predates or follows thebeginning of the banking crisis, we allow thedata to reveal what the temporal patterns are.The dates for the remaining crises are given inthe Appendix tables.

The Twin Crises.We next examine howthe currency and banking crises are linked.

5 If the peak month for the banking crisis is not known,we list the midpoint of that year as the date.

FIGURE 1. NUMBER OF CRISES PER YEAR

TABLE 1FREQUENCY OF CRISES OVER TIME

Type of crisis

Number of crises

19701995 19701979 19801995

Total

Average

per year Total

Average

per year Total

Average

per year

Balance-of-payments 76 2.92 26 2.60 50 3.13Twin 19 0.73 1 0.10 18 1.13Single 57 2.19 25 2.50 32 2.00

Banking 26 1.00 3 0.30 23 1.44

Note: Episodes in which the beginning of a banking crisis is followed by a balance-of-payments crisis within 48 months areclassified as twin crises.



6/28

We begin by calculating the unconditional proba-

bility of currency crises and banking crises in oursample. For instance, the probability that a cur-rency crisis will occur in the next 24 months overthe entire sample is simply 24 times 76 (the totalnumber of currency crises in the sample) dividedby the total number of monthly observations in thesample. These calculations yield unconditionalprobabilities for currency and banking crises,which are 29 percent and 10 percent, respectively(Table 3). The difference in the probabilities of thetwo kinds of crises highlights the relatively higherfrequency of currency crises in the sample.

We next calculate a family of conditional

probabilities. For instance, if knowing that there

is a banking crisis within the past 24 monthshelps predict a currency crisis then the proba-bility of a currency crisis, conditioned on infor-mation that a banking crisis is under way,should be higher than the unconditional proba-bility of a balance-of-payments crisis. In otherwords, a banking crisis increases the probabilitythat a country will fall prey to a currency crisis.This is precisely what the results summarized inTable 3 show. The probability of a currencycrisis conditioned on the beginning of banking-sector problems is 46 percent, well above theunconditional estimate of 29 percent. Hence, it

TABLE 2THE TIMING OF THE TWIN CRISES AND FINANCIAL LIBERALIZATION

CountryFinancial

liberalization

Banking crisis Closest balance-of-payment

crisisBeginning Peak

Argentina 1977 March 1980 July 1982 February 1981May 1985 June 1989 September 1986December 1994 March 1995 February 1990

Bolivia 1985 October 1987 June 1988 September 1985Brazil 1975 November 1985 November 1985 November 1986

December 1994 March 1996 October 1991Chile 1974 September 1981 March 1983 August 1982Colombia 1980 July 1982 June 1985 March 1983Denmark Early 1980s March 1987 June 1990 August 1983Finland 1982 September 1991 June 1992 November 1991Indonesia 1983 November 1992 November 1992 September 1986Israel 1985 October 1983 June 1984 October 1983Malaysia 1978 July 1985 August 1986 July 1975Mexico 1974 September 1982 June 1984 December 1982

1991 October 1992 March 1996 December 1994Norway 1980 November 1988 October 1991 May 1986Peru 1991 March 1983 April 1983 October 1987Philippines 1980 January 1981 June 1985 October 1983Spain 1974 November 1978 January 1983 July 1977Sweden 1980 November 1991 September 1992 November 1992Thailand 1989 March 1979 March 1979 November 1978

October 1983 June 1985 November 1984Turkey 1980 January 1991 March 1991 March 1994Uruguay 19761979 March 1971 December 1971 December 1971

March 1981 June 1985 October 1982Venezuela 1981, 1989 October 1993 August 1994 May 1994

Memorandum item:

Out of sample

Indonesia November 1992 Ongoing August 1997Malaysia September 1997 Ongoing August 1997Philippines July 1997 Ongoing July 1997Thailand May 1996 Ongoing July 1997

Note: Episodes in which the beginning of a banking crisis is followed by a balance-of-payment crisis within 48 months areclassified as twin crises.Sources: American Banker, various issues; Gerald Caprio, Jr. and Daniela Klingebiel (1996); New York Times, various issues;Sundararajan et al. (1991); Wall Street Journal, various issues.



7/28

could be argued, as Daz-Alejandro (1985) andVelasco (1987) did for the Chilean crisis in theearly 1980s, that, in an important number ofcases, the bailout of the banking system mayhave contributed to the acceleration in creditcreation observed prior to the currency crises(see Herminio Blanco and Peter M. Garber,1986; Sebastian Edwards, 1989; Eichengreen etal., 1996b; and this paper). Even in the absenceof a large-scale bailout, a frail banking system is

likely to tie the hands of the central bank indefending the currencywitness Indonesia inAugust 1997.

If, instead, the peak of the banking crisis is usedas the conditioning piece of information, no valu-

able information is gained; indeed, the conditionalprobability is 22 percent and below the uncondi-tional. This result follows from the fact that amore common pattern (see Table 2) appears to bethat the peak of the banking crisis comes after thecurrency crisis. For instance, knowing that there isa currency crisis does not help predict the onset ofa banking crisis, this conditional probability is 8percent; knowing that there was a currency crisisdoes help to predict the probability that the bank-ing crisis will worsen, this conditional probability

is 16 percent.Taken together, these results seem to point tothe existence of vicious circles. Financial-sectorproblems undermine the currency. Devalua-tions, in turn, aggravate the existing banking-sector problems and create new ones. Theseadverse feedback mechanisms are in line withthose suggested by Mishkin (1996) and can beamplified, as we have seen in several of therecent Asian crises, by banks inadequate hedg-ing of foreign-exchange risk. The presence ofvicious circles would imply that, a priori, the

twin crises are more severe than currency orbanking crises that occur in isolation.

To measure the severity of a currency crisis,we focus on a composite measure that averagesreserve losses and the real exchange-rate depre-ciation.6 For reserves, we use the six-monthpercent change prior to the crisis month, asreserve losses typically occur prior to the deval-uation (if the attack is successful). For the realexchange rate, we use the six-month percentchange following the crisis month, because

large depreciations occur after, and only if, thecentral bank concedes by devaluing or floatingthe currency. This measure of severity is con-structed for each currency crisis in our sampleand the averages are reported in Table 4 sepa-rately for the 19 twin crises in our sample andfor the others. In line with our results that thebeginning of the banking crisis precedes thebalance-of-payments crisis, we define the twin

6 The real exchange rate is used, as high inflation coun-tries will typically have larger nominal devaluations.

TABLE 3PROBABILITIES OF CRISES

Probabilities of balance-of-payment crises

TypeValue

(in percent)

Unconditional 29Conditional on the beginning of a

banking crisis46

Conditional on the peak of a bankingcrisis

22

Probabilities of banking crises

TypeValue

(in percent)

Unconditional 10Beginning of a banking crisis conditional

on a balance-of-payments crisis8

Beginning of a banking crisis conditional

on financial liberalization

14

Peak of a banking crisis conditional on abalance-of-payments crisis

16

Notes: The balance-of-payment crisis windows are definedas the 24 months preceding the crisis. The banking crisiswindows are defined as the 12 months before and the 12months after the beginning (or peak) of the crises. Theunconditional probabilities of balance-of-payment andbanking crises are calculated as the total number of monthsin the respective crisis windows divided by the total numberof months in the sample. The balance-of-payment probabil-ities conditional on a banking crisis (beginning or peak) arecalculated as the number of months in the balance-of-

payment crisis windows that occur within 24 months of thebanking crises (beginning or peak) divided by the totalnumber of months in the banking crisis windows. Theprobabilities of banking crises conditional on balance-of-payment crises are calculated as the number of months inthe banking crisis windows that occur within 24 months ofa balance-of-payment crisis divided by the total number ofmonths in the balance-of-payment crisis windows. Theprobability of a banking crisis conditional on financial lib-eralization is calculated as the total number of months in thebanking crisis windows that occur during times of financialliberalization divided by the total number of months duringwhich the banking sector was in a regime of financialliberalization. All probabilities were estimated using the

data for the 20 countries in the 1970mid-1995 period.



8/28

crises as those episodes in which a currencycrisis follows the beginning of the banking cri-sis within the next 48 months. For banking

crises, we use the bailout costs, as a percent ofGDP, as the measure of severity. As Table 4highlights, bailout costs are significantly larger(more than double) in the twin crises than forbanking crises which were not accompanied bya currency crisis. As to balance-of-paymentscrises, the results are mixed. Reserve lossessustained by the central bank are significantlybigger (Table 4) but the real depreciations are ofcomparable orders of magnitude.

Our results also yield an insight as to the links

of crises with financial liberalization (Table 3). In18 of the 26 banking crises studied here, the fi-nancial sector had been liberalized during the pre-ceding five years, usually less. Only in a few casesin our sample countries, such as the early liberal-ization efforts of Brazil in 1975 and Mexico in1974, was the liberalization not followed byfinancial-sector stress. In the 1980s and 1990smost liberalization episodes have been associatedwith financial crises of varying severity. Only in ahandful of countries (for instance, Canada, whichis not in the sample) did financial-sector liberal-ization proceed smoothly. Indeed, the probability

of a banking crisis (beginning) conditional onfinancial liberalization having taken place ishigher than the unconditional probability of abanking crisis. This suggests that the twin crisesmay have common origins in the deregulation of

the financial system and the boom-bust cycles andasset bubbles that, all too often, accompany finan-cial liberalization. The stylized evidence presentedin Caprio and Klingebiel (1996) suggests that in-adequate regulation and lack of supervision at thetime of the liberalization may play a key role inexplaining why deregulation and banking crisesare so closely entwined.

II. The Macroeconomic Background

of the Crises

To shed light on whether both types of crisesmay have common roots, we analyze the evo-lution of 16 macroeconomic and financial vari-ables around the time of the crises. Thevariables used in the analysis were chosen inlight of theoretical considerations and subject todata availability. Monthly data was used to geta clearer view (than would otherwise be re-vealed by lower frequency data) of develop-ments as the crisis approaches and by the desireto evaluate to what extent these indicators were

giving an early signal of impending troubleanissue that will be taken up in the next section.

The indicators associated with financialliberalization are the M2 multiplier, the ratio ofdomestic credit to nominal GDP, the real interestrate on deposits, and the ratio of lending-to-deposit interest rates. Other financial indicatorsinclude: excess real M1 balances, real commer-cial-bank deposits, and the ratio of M2 (convertedinto U.S. dollars) divided by foreign-exchangereserves (in U.S. dollars).7 The indicators linked to

the current account include the percent deviationof the real exchange rate from trend, as a measureof misalignment, the value of exports and imports(in U.S. dollars), and the terms of trade.8 The

7 M2 to reserves captures to what extent the liabilities of thebanking system are backed by international reserves. In theevent of a currency crisis, individuals may rush to convert theirdomestic currency deposits into foreign currency, so that thisratio captures the ability of the central bank to meet thosedemands (Calvo and Enrique Mendoza, 1996).

8 An increase in the real exchange-rate index denotes adepreciation.

TABLE 4THE SEVERITY OF THE CRISES

Severity measure

Banking crises

Balance-of-payments

crises

Twin Single Twin SingleCost of bailout

(Percent of GDP) 13.3 5.1* NA NALoss of reserves

(Percent) NA NA 25.4 8.3*Real depreciation

(Percent) NA NA 25.7 26.6Composite index NA NA 25.6 17.5

Notes: Loss of reserves is the percentage change in the level ofreserves in the six months preceding the crises. Real depreci-ation is the percentage change in the real exchange rate (withrespect to the dollar for the countries that peg to the dollar andwith respect to the mark for the countries that peg to mark) inthe six months following the crises. The composite index is theunweighted average of the loss of reserves and real deprecia-tion. Episodes in which the beginning of a banking crisis isfollowed by a balance-of-payments crisis within 48 months areclassified as twin crises.

* Denotes that the measure of severity of single-crisesepisodes is statistically different from the twin-crises sever-ity at the 5-percent level. An NA denotes not applicable.



9/28

indicators associated with the capital accountare: foreign-exchange reserves (in U.S. dollars)and the domestic-foreign real interest-rate dif-ferential on deposits (monthly rates in percent-age points). The indicators of the real sectorareindustrial production and an index of equityprices (in U.S. dollars).9 Lastly, the fiscal vari-

able is the overall budget deficit as a percent ofGDP.

Of course, this is not an exhaustive list of po-tential indicators. In particular, political variables,such as the timing of an election, can also belinked to the timing of these crises. Indeed, theevidence presented in Deepak Mishra (1997), whoexamines a subset of the currency crises in thisstudy, suggests that devaluations, more often than

not, follow elections. Indeed, an election raises theprobability of a future devaluation, even after con-trolling for economic fundamentals.

Except for the interest-rate variables, thedeviations of the real exchange rate from trend,our proxy for excess real M1 balances, and thelending/deposit interest-rate ratio, which are in

levels, we focus on the 12-month percent changesof the remaining 10 variables. The pre- and post-crises behavior of all variables is compared to theaverage behavior during tranquil periods, whichare all the remaining observations in our sampleand serves as our control group.

Figures 2, 3, and 4 illustrate the behavior ofthe variables around the time of the balance-of-payments crises, banking crises, and twincrises, respectively; each panel portrays a dif-ferent variable. The horizontal axis recordsthe number of months before and after thebeginning of the crises; the vertical axis

9 Detailed definitions of all the variables and theirsources are provided in the Data Appendix.

FIGURE 2. EMPIRICAL REGULARITIES DURING BALANCE-OF-PAYMENTS CRISES

Notes: The values of the variable relative to tranquil times are reported on the vertical axes. The horizontal axes representthe number of months before (with a negative sign) and after the crisis. The solid line represents the average for all the crisesfor which data was available. The dotted lines denote plus/minus one standard error around the average. Unless otherwisenoted, all variables are reported as 12-month changes, in percent, relative to tranquil times. 1. Monthly rates, in percentagepoints, relative to tranquil times. 2. Actual less estimated money demand. Percent deviation relative to tranquil times. 3.

Deviations from trend, in percent, relative to tranquil times.



10/28

records the percent difference (percentage-point difference for interest rates) betweentranquil and crisis periods. In all the figuresthe solid line represents the average for all thecrises for which data was available.10 Hence,if no data points are missing, the solid linerepresents the average behavior of that indi-

cator during the months around 76 currencycrises and 26 banking crises. For Figures 2and 3, the dotted lines denote plus/minus onestandard error around the average. For exam-ple, the top center panel of Figure 2 showsthat, on average, the 12-month growth in thedomestic credit/GDP ratio is about 15 percenthigher than in tranquil times. In Figure 4 the

solid line shows the evolution of the indica-tors for the twin-crises episodes while thedashed line denotes the averages for the cur-rency crises that were not accompanied by abanking crisis.

For currency crises we focus on the 18-month period before and after the crisis. Un-

like balance-of-payments crises, in whichreserves are lost abruptly and currency pegsabandoned, banking crises are protracted af-fairs which tend to come in waves and, hence,the depth of the crisis is seldom reached at thefirst sign of outbreak (see Table 2). For thisreason, we widen the window and focus onthe 18 months before the onset of the crisis,an 18-month arbitrarily chosen crisis period,and the 18-month post-crisis period. At anyrate, because most of our analysis focuses onthe causes leading up to the crises, our main

results will not be affected whether the crises

10 See Appendix Tables A1 and A2 for a detailed indi-cation of any missing data around crisis dates.

FIGURE 3. EMPIRICAL REGULARITIES DURING BANKING CRISES

Notes: The values of the variable relative to tranquil times are reported on the vertical axes. The horizontal axes representthe number of months before (with a negative sign) and after the crisis. The solid line represents the average for all the crisesfor which data was available. The dotted lines denote plus/minus one standard error around the average. Unless otherwisenoted, all variables are reported as 12-month changes, in percent, relative to tranquil times. 1. Monthly rates, in percentagepoints, relative to tranquil times. 2. Actual less estimated money demand. Percent deviation relative to tranquil times. 3.

Deviations from trend, in percent, relative to tranquil times.



11/28

lasted less or more than a year. For the 19episodes of the twin crises, we focus on the 18months prior to the balance-of-payments cri-sis. Given that banking crises usually predatecurrency crises in our sample, this implies weare already looking at a period of heavyfinancial-sector stress.

A. The Financial Sector

Until the 1970s, most financial markets wereregulated with rationed credit and, often, nega-tive real interest rates. The late 1970s andbeginning of the 1980s, however, witnessedsweeping financial reforms both in developedand emerging markets, which led to, amongother things, increases in real interest rates.11

Because financial liberalization often precedes

banking crisesthe indicators associated withfinancial liberalization presented in the first fourpanels of Figures 2, 3, and 4 (from left to right)merit scrutiny. The growth in the M2 multiplierrises steadily up to nine months prior to thecurrency crisis and the onset of the bankingcrisis; indeed, for banking crises the multipliergrows at above normal rate in the entire 18

months prior to the crisis. The draconian reduc-tions in reserve requirements that often accom-pany financial liberalization play a role inexplaining the large increases in the M2 multi-plier. Yet the rise in the multiplier prior tocurrency crises is entirely accounted for by itsevolution ahead of the twin crises, as shown inFigure 4.

The growth in domestic credit/GDP re-mains above normal as the balance-of-payments crisis nears (Figure 2) but particu-larly accelerating markedly as the twin crisesapproaches; throughout this period it remains11 See Vincente Galbis (1993).

FIGURE 4. EMPIRICAL REGULARITIES DURING TWIN CRISES

Notes: The values of the variables relative to tranquil times are reported on the vertical axes. The horizontal axes representthe number of months before (with a negative sign) and after a crisis. The solid lines show the behavior during twin-crisesepisodes, and the dotted lines show the behavior during single-crises episodes. Unless otherwise noted, all variables arereported as 12-month changes, in percent, relative to tranquil times. 1. Monthly rates, in percentage points, relative to

tranquil times. 2. Actual less estimated money demand. Percent deviation relative to tranquil times. 3. Deviations fromtrend, in percent, relative to tranquil times.



12/28

well above the growth rates recorded for tran-quil periods, consistent with a credit boom(and bust) story. This ratio also rises in theearly phase of the banking crisis. It may bethat, as the crisis unfolds, the central bank

may be pumping money to the banks toalleviate their financial situation or the evolutionof the denominator has changed. While credit israpidly expanding 18 to 6 months before the crisis,the economy is still in a vigorous expansion phase(see below), with healthy GDP growth. The lever-aging of households and business becomes evi-dent as the economy slips into recession. The realinterest rate evolves very differently ahead ofbalance-of-payments and banking crises. For cur-rency crises, interest rates bounce around in the

range of 0 to 2 percentage points per month belowthe average during periods of tranquilitythismay reflect lax monetary policy ahead of the cur-rency crisis or simply the fact that 26 of thecurrency crises are in the 1970s, when interestrates were regulated and not particularly informa-tive. By contrast, prior to banking crises and,therefore, twin crises (which are almost exclu-sively in the post-liberalization part of the sam-ple), real interest rates are 1 to 2 percentage pointshigher (at a monthly rate) than in tranquil times inthe pre-crisis period. The above normal real inter-

est rates may have a variety of causes: Thesecould be the product of a recent financial liberal-ization; high real rates could also reflect increasedrisk taking by banks;12 they could be the productof a tight monetary policy stance. Real interestrates do not return to their levels in tranquil timesas the crisis deepens, perhaps reflecting that banksmay respond to deposit withdrawals by keepingdeposit interest rates high. The lending-depositrate ratio hovers around its level in tranquil timesup until about six months prior to the balance-of-

payments crises and then begins to climb; by thetime of the crisis it is about 10 percent higher thanin tranquil times, possibly reflecting a deteriora-tion in credit risk. For banking crises, the lending/deposit ratio remains close to normal levels in thepre-crisis period. Only at around the peak of thebanking crises does the lending/deposit ratio in-crease above its level in tranquil times, as banksbecome increasingly unwilling to lend.

The next three panels show the evolution of the

monetary indicators. The middle panel in the sec-ond row of Figures 2 and 3 show the excess M1balances. The periods prior to the currency andbanking crises are characterized by an excess sup-ply of real M1 balances; the excess liquidity is

particularly pronounced for the twin-crises epi-sodes, which nearly account for all the above-normal behavior ahead of currency crises.Without overinterpreting this result, given theshortcomings of money-demand estimation, thepicture that emerges is consistent with the deficitfinancing as in the Krugman (1979) framework orthe excess liquidity may be created to ease condi-tions for troubled financial institutions. In anycase, at some point the excess liquidity becomesincompatible with maintaining the exchange-rate

commitmentand a currency crisis emerges. Thiswould suggest that the high real interest rates priorto banking crises were due to factors other thanmonetary policy. The next panel shows the evo-lution of the 12-month change in M2/reserves ofcentral banks. For both currency and banking cri-ses, this ratio grows well above its norm prior tothe crises. The increases are associated with botha vigorous expansion in M2 (witness the multi-plier) and a sharp decline in foreign currencyreserves (discussed below). As Calvo and Men-doza (1996) do for Mexico 1994, we find that the

M2/reserves ratio over the 76 currency crises in-dicates an abrupt decrease in the backing ratio inthe months preceding the crisis. Indeed, thegrowth rate is 70 percent in excess of the tranquilperiod average, highlighting vulnerability of thesystem. This observation is equally descriptive ofboth single-currency and twin-crises episodes.The growth rate ofbank deposits remains close tonormal during the 18 months prior to the financialcrises, but the loss of deposits accelerates as thecrises unfold. There may be multiple reasons for

this sudden decline. Past financial-crises periodshave often been characterized by massive andpersistent capital flight. Deposits only start to re-cover a year and a half after the onset of thefinancial crises.

B. The External Sector

The next four panels of Figures 2, 3, and 4present indicators associated with the currentaccount. The middle panel of the third row ineach figure chronicles the abysmal performanceof the growth of exports in the year and a half12 See V. Sundararajan and Tomas Balino (1991).



13/28

preceding the currency and banking crisesexports consistently underperform (relative tonormal times) during this period. By the time abalance-of-payments crisis is under way, exportgrowth is about 20 percent below (annual rate)

the average growth observed in tranquil periods.Once the appreciation is reversed, export per-formance improves sharply, outdoing the per-formance observed during tranquil periodsabout nine months after the crisis began. Exportperformance is particularly poor during thetwin-crises episodes. The behavior of importgrowth is more difficult to justify on the basis ofrelative price developments (see below). Importgrowth remains close to the norm during tran-quil periods up to about nine months before a

currency crisis and then declines; for bankingcrises, we see the tail end of the import boomand the subsequent slide prior to the crisis.During this pre-crisis period, income and rela-tive price effects are moving in opposite direc-tions, and the observed decline in import growthmay well be accounted for by the slowdown ineconomic activity (see below) during that time.Import growth remains below that of normalperiods throughout the post-crisis period.

The next panel provides evidence on theterms of trade. Crises are preceded, on aver-

age, by a deterioration of the terms of trade,with an annual decline that is about 10 per-cent deeper than those observed in tranquiltimes prior to a balance-of-payments crisis.This persistent adverse performance of theterms of trade erodes purchasing power andmay also account for the weakness in importsin the months preceding the crisis. This weak-ness is equally evident in single- and twin-crises episodes. For banking crises, up toabout a year prior to the crisis, terms-of-trade

shocks appear to have been positiveperhapshelping to explain the earlier boom (see be-low); as the crisis nears we see some evidenceof adverse terms-of-trade shocks. The middlepanel in the fourth row shows the evolution ofreal exchange rates. During the year beforethe balance-of-payments and banking crises(as stressed in Rudiger Dornbusch et al.,1996), the real exchange rate shows evidenceof being overvalued, relative to its averagelevel during tranquil times. In periods preced-ing the currency crash, it is appreciating rel-ative to its trend (an overvaluation of about 20

percent relative to tranquil periods). The realexchange-rate appreciation does reverse itselfrapidly with the devaluation, suggesting thatproductivity shocks or preference changeswere unlikely to account for the initial appre-

ciation. Exchange-rate-based inflation stabili-zation plans have often given rise to largecumulative real exchange-rate appreciations,as domestic inflation fails to converge to in-ternational levels. As noted in Reinhart andVegh (1996) and Kaminsky and LeonardoLeiderman (1998), many of those plans endedin a balance-of-payments crisis. Followingthe crash, the real exchange rate depreciatessubstantially (and is about 10 percent higherthan in tranquil times). Over time, higher

domestic inflation erodes in part the improve-ment in competitiveness.In the absence of monthly data on capital

flows for most of the period and most of thecountries in our sample, we extract informationabout capital account developments by focusingon the indicators shown in the next two panels.As expected, the 12-month percentage changein foreign-exchange reserves of the centralbanks falls substantially in the months prior toboth banking and balance-of-payments crises.The loss of reserves is particularly steep and

longer lived following the crises for the 19twin-crises episodes. As early as 12 monthsprior to the balance-of-payments crisis, reservegrowth is about 20 percent below that observedduring tranquil periods; although we report 12-month changes, which introduce positive serialcorrelation in the data, reserves do not decreasecontinuously. There are modest short-lived re-versals in the path followed by reserves, whichsuggest that the central banks may have hadspells in which they fought the reserve loss with

contractionary monetary policy (note that thereare brief spells where real interest rates riseprior to the crisissee the third panel) beforefinally conceding defeat and devaluing. Follow-ing the devaluation (or flotation), foreign-ex-change reserves of central banks start toincrease again.

Finally, the first panel in the bottom rowshows the evolution of the domestic-foreignreal interest-rate differential on deposits. Inter-est differentials do not reflect increasing expec-tations of a devaluation as the currency crisisnears. Turning to banking crises, the picture that



14/28

emerges is quite distinct from its counterpart inFigure 2; while in balance-of-payments crisesinterest-rate differentials were not appreciablydifferent from tranquil periods prior to crises,differentials in the case of banking crises remain

above those observed in periods of tranquility.One explanation for this difference among thetwo crises has to do with the bunching of thebanking crises in the post-financial liberaliza-tion period.

C. The Real Sector

The last two panels in the figures show theevolution ofoutputgrowth and changes in stock

prices. The deterioration of the terms of trade,

the overvaluation of the currency, and the weak-ening export performance are reflected in amarked slowing in economic activity and a de-cline in output prior to both crises. For balance-of-payments crises, the 12-month growth inoutput bounces in a range of 2 to 6 percentbelow the comparable growth rates during tran-quil periodswith a tendency for the recessionto deepen as the crisis nears. Interestingly, andin line with the greater severity of the twincrises, the combination of currency and bankingproblems appears to take a more devastating toll

on the real economy as the recession is fardeeper and longer than the recessions associatedwith currency crashes alone. At growth rateswhich are 8 percent below those observed intranquil periods, the twin-crisis recession istwice as severe. As Kindelberger (1978) ob-serves: Financial crises are associated with thepeaks in business cycles ... the financial crisis isa culmination of a period of economic expan-sion that leads to downturn. While in the 18months prior to a balance-of-payments crisis

there is no evidence of a residual economicboom, that is not the case in the pre-banking-crisis period. As Figure 3 shows, up to about 8months before the banking crises the economywas recording growth rates above those ob-served during tranquil periods. Yet, the realexchange-rate appreciation that characterizespre-crisis periods is often cited as a key factorbehind the squeeze in profit margins that even-tually leads to increased bankruptcies, a risein nonperforming loans, a deepening in theeconomic contraction, and banking-sectorproblems.

The last panel shows the evolution of stockprices. During the 18 months prior to a balance-of-payments crisis, the equity market steadilyunderperforms (relative to tranquil times)atfirst, not by much, but as the crisis nears,

changes in stock prices (that is, stock returns [indollars]) are about 40 percent below those ob-served in noncrisis periods. The weakening inequity prices is, most likely, reflecting both thedeteriorating cyclical position of the economy,reduced foreign demand as capital inflows arereversed, and the worsening balance sheets offirms, as the overvaluation takes its toll. Thecrash is particularly severe when currency andbanking crises nearly coincide (Figure 4). Un-like the onset of a banking crisis (see below),

the equity market was already past it cyclicalpeak well before the crisis begins. On the eve ofbanking crises, the return on equity prices up toabout nine months prior to the crises suggests aboom (relative to tranquil periods) which may(or may not) be an asset-price bubble. Duringthe boom phase, returns exceed those of non-crises periods by about 40 percent on an annualbasis. The beginning of the recession is alsoreflected in the stock market, which collapsesthe year before the crisis; this collapse is alsoapparent in other asset markets, most notably

real estate.13

Finally, although not shown in the figures,the fiscal deficit/GDP ratio is higher in the twoyears prior to the currency crisis and one yearprior to the banking crisis. While the biggerdeficit could stem from higher governmentspending, the weakness in output prior to crisescould lead to a shortfall in revenues.

III. The Anatomy of Crises

In what follows, we offer an alternative ap-proach to examine the evolving nature of the

13 For example, in the boom period leading up to the1981 Argentine banking crisis, stock returns (in U.S. dol-lars) were as high as 813 percent during the 12 monthsending May 1979; by May 1981, the 12-month capital losswas 60 percent. The crash in asset values is cited in mostcase studies as an important factor contributing to the prob-lems of the banks. Also, due to either mismanagement oroutright fraud, in many of the crises in our sample a sub-stantial portion of banks and finance companies were con-siderably overexposed to real estate.



15/28

crises, pinpoint their origin, and gauge theirprobability conditioned on signals from one ormore indicators. The methodology used, whilenot previously applied to analyze currency andbanking crises, has a long history in the rich

literature that evaluates the ability of macro-economic and financial time series to predictbusiness-cycle turning points.14 The remainderof this section is divided into two parts, the firstdescribes the statistical methodology used,while the second applies that methodology tothe 102 currency and banking crises that makeup our sample.

A. Methodology

To examine the causes of crises, gauge thevulnerability of the economy on the eve ofcrisis, and assess whether the crisis itself couldbe forecasted by anomalous economic develop-ments, we need to make four sets of judgments:First, we must have a well-defined notion ofwhat is classified as a crisis. Second, we mustagree on a list of variables that are potentialleading indicators. Third, we need to decideupon a criteria that allows us to classify thebehavior of an indicator as either a signal of acrisis or normal (no signal). Last, if an indicator

is giving a signal, we have to determine if acrisis happens within a reasonable period oftime or if the signal was a false alarm. Hence,we also need to define what is considered to bea reasonable period of time. Section I deals withthe definition and dating of banking and cur-rency crises, while the previous section and theData Appendix discuss the indicators. In thissubsection, we describe the approach used todefine what is a signal and what is a reasonableperiod of time.

The Interval Between Signals and Crisis: De- fining a Reasonable Period of Time.In whatfollows, the maximum interval of time betweenthe signal and the crisis was decided a priori as24 months in the case of balance-of-paymentscrises.15 Hence, any signal given within the24-month period before the beginning of the

crisis is labeled a good signal; any other signaloutside that 24-month window is labeled a falsealarm or noise. For banking crises, any signalgiven within the 12-month period before thebeginning of the crisis or within 12 months

following the beginning of the crisis is labeled agood signal. The two different signaling win-dows for currency and banking crises have to dowith the different timing of the peaks of bothcrises, as previously discussed. In addition, theevents that mark the beginning of a bankingcrisis are often not seen as systemic at the timeand are not treated by policy makers as harbin-gers of a crisis. Since symptoms of a crisis aresometimes evident well before the crises erupt,the narrower windows (say, 12-month) were

thought to penalize indicators (such as M2/re-serves) that tend to give an early warning.

The Threshold: Defining a Signal.In Sec-tion II we noted that the crises were preceded bymarked declines in equity returns. Yet, surely,not every decline in equity returns presages acrisis. Hence, we need to select an appropriatethreshold or cutoff that separates when a declinein equity returns is considered a signal of acrisis and when it is not. As is the case ofselecting the size of the rejection region in hy-

pothesis testing, choosing the optimal thresholdinvolves a trade-off. Suppose that our null hy-pothesis is that we are in a tranquil state ofnature and (for a particular country) we areweighing whether to arbitrarily set the thresholdfor annual equity returns at minus 15 percent orat minus 40 percent. Suppose further, that forthis country 10 percent of the observationsposted annual equity returns below minus 15percent, but only 3 percent of the observationsshowed equity returns below minus 40 percent.

Our aim is to use the readings for this indicatorto test the null hypothesis that we are in atranquil state of nature. If we choose the minus15 percent threshold, the size of (the rejectionregion) is 10 percentthis is the probability ofrejecting the null hypothesis when it is true(Type I error). In this case, the threshold may betoo laxit is likely to catch all the crises but itis also likely to generate a lot of false alarms.Instead, we could adopt the minus 40 percentthreshold, which cuts the size of to 3 percent;this reduces the probability of Type I error at theexpense of increasing the probability of Type II

14 See, for instance, Francis Diebold and Glen Rude-busch (1989), James H. Stock and Mark W. Watson (1989),and Reinhart and Vincent R. Reinhart (1996).

15 An 18- and 12-month window were also used; theresults are available from the authors.



16/28

error (not rejecting the null hypothesis when itis false). With this tight threshold we may missall but the most severe of the crisesthe priceof reducing the number of false alarms is accu-rately calling a lower proportion of crises.

We select the threshold value on an indicator-by-indicator basis, by performing a fine gridsearch over a broad range of critical regions upto a maximum of 30 percent. For each thresholdvalue in our grid search we compute the noise-to-signal ratio.16 We then select the thresholdvalue that minimizes the noise-to-signal ratio.As to the location of the rejection region,whether it is the upper or lower tail of thefrequency distribution for each indicator, werely on the theory as a guide. The threshold

values for the 16 indicators, as well as thelocation of the rejection region and its theoret-ical justification, are given in Table 5. For ex-ample, for currency and banking crises largeoutput declines signal a crisis, so a sign inTable 5 denotes that the rejection region islocated at the bottom tail of the distribution.

This criterion does have drawbacks which areworth mentioning. First, if an indicator gives an earlysignal and policy makers heed the signal and pre-empt a crisis, that signal is labeled as false and theindicator is penalized with an unduly high noise-to-

signal ratio. In addition, a signal within the windowis treated the same irrespective of whether it was

given 12 months before the crisis erupts or only themonth before. Naturally, from the vantage point ofthe policy maker the earlier signal is the more valu-able one.

B. The Anatomy of Crises

The methodology just described was ap-plied to the 16 indicators and 102 crises in thesample and the four recent Asian crises out ofsample.

Appendix Tables A1 and A2 show the resultson a crisis-by-crisis and indicator-by-indicator ba-sis. An NA denotes some or all the observationswere missing during the pre-crisis 24-month win-dow; a 1 denotes at least one signal was given

during the 24-month window, and a zero indicatesno signals were issued. Hence, for example, col-umn (12) in Appendix Table A1 scores the per-formance of foreign-exchange reserves; there arefour NA entries, hence we have full data for thisindicator for 72 balance-of-payments crises. In 75percent of the crises [row (1) Summary Statistics,bottom of the table] there were one or more sig-nals during the 24 months prior to the crisis. Thelast column (17) lists what proportion (in percent)of the indicators were sending signals. Data avail-ability permitting, the tables also show the evolu-

tion of the indicators out of sample for the Asiancrises of 1997.

About the Origins of Crises.Table 6 summa-rizes the results in Appendix Tables A1 and A2.The indicators are shown individually and are alsogrouped into sectors along the lines described inthe previous section: financial liberalization, otherfinancial, current account, capital account, real-side, and fiscal. For balance-of-payments crises,we also examine subsamples before financial lib-

eralization, which encompasses the 1970s, andafter financial liberalization, as well as those cur-rency crises which occurred alongside a bankingcrises. The latter appear under the column labelledTwin. As nearly all banking crises fall in thepost-liberalization period, no subsamples for theseare reported. Table 6 presents the percentage ofcrises accurately called by each indicator. As tothe various groups we also report the simple arith-metic average of the proportion of crises accu-rately called by all the indicators in that subgroup.Capital account indicators accurately called thehighest proportion of balance-of-payments crises

16 The definition of noise-to-signal ratio used throughoutis best illustrated by considering the following two-by-twomatrix:

Crisis occursin the

following 24months

No crisis occursin the following

24 months

Indicator issues asignal A B

Indicator does notissue a signal C D

If a variable signals and a crisis occurs in the following 24months (counted in cell A) the signal is considered accurate.If a variable signals and no crisis occurs in that time frame(counted in cell B), the signal is said to be a false alarm ornoise. Hence, a perfect indicator would only have entries incells A and D. More generally, the noise-to-signal ratio forany indicator is given by the number of entries in [B/(B

D)]/[A/(A C)]. Hence, it is the ratio of false signals to allpossible bad signals divided by the ratio of good signals toall possible good signals. An extremely noisy indicatorwould have few entries in A and D, many in B and C.



17/28

(about 81 percent). Financial liberalization indica-tors were next in line, accurately signaling 74percent of the currency crises before they oc-curred; for the twin crises their performance is

even better. Among the capital account and finan-cial indicators that fared the worst are bank de-posits, the lending-deposit ratio, and excess M1balances. Current account indicators followed

TABLE 5THRESHOLD VALUES FOR SIGNALING CRISES

Indicators

Threshold values andthe location of the

critical region

Comments

Balance-of-paymentcrises

Bankingcrises

Financial sectorFinancial liberalization

M2 multiplier 0.86 0.90 Both banking and currency crises have been linked to rapid growth(boom-bust) in credit and the monetary aggregates (see McKinnonand Pill, 1996).

Domestic credit/GDP 0.90 0.95

Real interest rateLending-deposit rate

ratio

0.880.80

0.800.87

For banking crises, the choice is unambiguous since financialderegulation is associated with high interest rates (which could reflectincreased risk taking [see Galbis, 1993]). A liquidity crunch (say todefend a peg) will also hurt banks. For balance-of-payments crises itis less clear-cut; higher real interest rates could reflect a higher risk

premia and fears of devaluation. Yet, using the lower interest ratesfor signals could be justified for balance-of-payments crises on thebasis of loose monetary policy. An increase in the lending/depositratio can capture a decline in loan quality.

Other

Excess M1 balances 0.94 0.91 This is a loose monetary policy story (see Krugman, 1979).M2/reserves 0.87 0.90 For the motivation on M2/reserves, see Calvo and Mendoza (1996).Bank deposits 0.10 0.16 Capital flight and a run against the domestic banks may precede

both currency and banking crises (see Goldfajn and Valdes, 1995).External sector

Current account

ExportsTerms of tradeReal exchange rate

0.100.160.10

0.100.190.10

Real exchange-rate overvaluations and a weak external sector are apart of a currency crisis. It adds vulnerability of the bankingsector, since a loss of competitiveness and external markets could

lead to recession, business failures, and a decline in the quality ofloans. Thus, large negative shocks to exports, the terms of trade,and the real exchange rate are associated with signals (seeDornbush et al., 1995).

Imports 0.90 0.80 Theory is ambiguous as to where we should locate the rejectionregion. Rapid import growth could be the sign of a buoyanteconomy (this would argue for a negative shock to imports); itcould also be the sign of overvaluation. Hence a positive shock toimports could be a signal. Both possibilities were explored.

Capital account

ReservesReal interest-rate

differential

0.15

0.890.28

0.81See discussion under bank deposits and real interest rates.

Real sector

OutputStock prices

0.110.11

0.140.10

Recessions and the burst of asset price bubbles precede financialcrises (see Gary Gorton, 1988; Charles W. Calomiris and Gorton,1991).

Fiscal sectorDeficit/GDP 0.86 0.86 Loose fiscal policy financed by credit from the central bank (see

Krugman, 1979).

Note: The definitions and sources of the indicators are described in the Data Appendix.



18/28

next (68 percent accurately called) but this islargely owing to the weak performance of importsin accurately calling crises; exports, the terms oftrade, and the real exchange rate do much better.The fiscal variable fared the worst, accurately call-ing only slightly over a quarter of the currency

crises.One key difference between banking and cur-

rency crises, highlighted in Table 6, is the role ofthe real sector, which appears to be consider-ably more important for banking crisesgiv-ing early signals in 85 percent of the crises.17

Indeed, output and stock prices signaled in89 and 81 percent, respectively, of the bank-ing crises for which data for these indicators

was available. As much of the literature onbanking crises stresses, particularly asymmet-ric information models (see Calomiris andGorton, 1991), the evidence presented heresuggests that the bursting of asset-price bub-bles and increased bankruptcies associated

with an economic downturn appear to beclosely linked to domestic financial problems.

Yet another feature that is revealed in Table 6is that the proportion of crises accurately calledrises for 13 out of the 16 indicators when single-currency crises are compared to their twin coun-terparts. The improved performance of most of theindicators is not entirely surprising, in light of thegreater severity of the twin-crises episodes.

Fragility on the Eve of Crises.Table 7 pre-sents strong evidence that, for both banking andcurrency crises, multiple economic problems

17 For a discussion of the evolving nature of crises seethe working paper, Kaminsky and Reinhart (1996).

TABLE 6THE ONSET OF FINANCIAL CRISES: EARLY SIGNALS

Indicators

Percent of crises accurately called

Balance-of-payment crises

Total Single Twin

Before

financialliberalization

After


Bankingcrises

Financial sector 67 67 67 67 68 65Financial liberalization 74 72 78 64 77 71

M2 multiplier 76 75 78 74 77 73Domestic credit/GDP 61 59 67 56 65 50Real interest rate 89 86 94 78 91 100Lending-deposit rate ratio 71 70 73 50 73 57

Other 57 58 53 57 56 57Excess M1 balances 37 43 22 52 26 32M2/reserves 81 79 89 74 86 75Bank deposits 51 52 47 44 56 67

External sector 72 71 74 72 72 82

Current account 68 67 70 70 66 75Exports 85 83 89 78 89 88Terms of trade 75 72 83 73 77 96Real exchange rate 59 57 67 58 60 58Imports 52 57 39 73 40 60

Capital account 81 80 83 74 83 96Reserves 75 74 79 70 78 92Real interest-rate differential 86 86 88 78 89 100

Real sector 69 69 70 61 72 85Output 74 73 77 68 76 89Stock prices 64 65 63 53 68 81

Fiscal sector 28 27 29 21 31 44

Notes: Episodes in which the beginning of a banking crisis is followed by a balance-of-payments crisis within 48 months are

classified as twin crises. An indicator is said to have accurately called a crisis if it issues at least one signal in the crisis windowon the basis of the criterion shown in Table 5. For each indicator, each cell in the table represents the number of times thatindicator correctly calls a crisis as a percentage of the total number of crises. For the different sectors, each cell representsthe simple average of the percentage of crises accurately called by all the individual variables in that group.



19/28

were simultaneously building. We construct ameasure of the fragility of the economy in the24 months preceding the crisis by tallying on acrisis-by-crisis basis what proportion of the in-dicators were signaling during that period.18

Hence, if 14 of the 16 indicators are sending asignal prior to the crisis, this crisis would becounted in the first row of Table 7, labeled 80percent to 100 percent. It appears that crises arenot simply a story of an overvalued exchangerate or too rapid a monetary expansion. In about30 percent of the currency crises, 80 percent ormore of the indicators were sending signals.The economies appear to be particularly frail onthe eve of twin crises, with a higher proportionof the indicators signaling. Indeed, in about 80percent of the twin crises, at least 60 percent ofthe indicators were sending a signal. There werebasically no banking crises with less than 20percent of the indicators signaling. For furtherevidence of the diversity of the economic prob-lems on the eve of crises on a crisis-by-crisis

basis, see Appendix Tables A1 and A2. Thefinding that when the balance-of-payments cri-ses occur jointly with a banking crisis (under theheading Twin, Table 7) economies appear tohave more widespread problems perhaps is notentirely surprising, given the earlier resultswhich suggest the twin crises tend to be moresevere.

These results would appear to suggest that the

overwhelming majority of crises, external or do-mestic, have a multitude of weak economic fun-damentals at their core. While speculative attacksdo occur as market sentiment shifts and, possibly,herding behavior takes over, such self-fulfillingcrises appear to be quite rare. Indeed, in the con-text of the Exchange Rate Mechanism crises thisissue has been the subject of much debate.19 Notonly are the signals many, but their sources aremultiple, as shown in Table 7with the financialsector external (capital account) and domesticplaying a key role.

IV. Final Remarks

We have examined the empirical regularitiesand the sources and scope of problems in theonset of 76 currency crises and 26 bankingcrises. We find that banking and currency crisesare closely linked in the aftermath of financialliberalization, with banking crises, in general,beginning before the currency collapse. We also

find evidence of vicious cycles, in which thecurrency collapse further undermines an alreadyailing banking sector. When currency and bank-ing crises occur jointly, they are far more severethan when they occur in isolation. In both typesof crises, a financial shock, possibly financialliberalization or increased access to interna-tional capital markets, appears to activate aboom-bust cycle by providing easy access tofinancing. Finally, in both crises we find a

18 These are reported for each crisis in column (17) of theAppendix tables. 19 See Eichengreen et al. (1996a).

TABLE 7ECONOMIC FRAGILITY ON THE EVE OF CRISES

Number of indicators

signaling a crisis(in percent)

Number of crises (in percent)

Balance-of-payments crises

Total Single Twin

Before


After


Bankingcrises

80100 26.7 28.6 21.1 40.0 17.8 30.86079 45.3 41.1 57.9 23.3 60.0 53.84059 20.0 21.4 15.8 20.0 20.0 11.52039 6.7 8.9 0.0 13.3 2.2 3.9Less than 20 1.3 0.0 5.3 3.3 0.0 0.0

Notes: This table captures the state of distress of the economy in different crisis episodes. Each cell represents the proportionof crises with a given proportion of signals. For example, 21.1 percent of the twin balance-of-payment crises had 80100percent of indicators signaling a crisis. Episodes in which the beginning of a banking crisis is followed by a balance-of-payments crisis within 48 months are classified as twin crises.



20/28

TABLE A1ANATOMY OF BALANCE-OF-PAYMENTS CRISESa

Country Crisis

M2multiplier

(1)

Domesticcredit/GDP

(2)

Realinterest

rate(3)

Lending-deposit

rate ratio(4)

ExcessM1

balances(5)

M2/reserves

(6)

Bankdeposits

(7)

Argentina June 70 NA NA NA NA 1 NA NAJune 75 0 1 NA NA 1 1 0Feb. 81 1 0 1 NA 0 1 0July 82 1 1 1 1 0 1 0Sept. 86 1 NA 1 1 0 1 0April 89 1 NA 1 1 0 1 0Feb. 90 0 NA 1 1 1 1 1

Bolivia Nov. 82 0 1 1 NA 0 1 1Nov. 83 0 1 1 0 0 1 1Sept. 85 0 0 1 1 1 1 1

Brazil Feb. 83 0 0 NA NA 0 1 1Nov. 86 1 1 1 1 1 1 0July 89 1 1 1 1 1 1 1

Nov. 90 1 1 1 1 0 1 1Oct. 91 1 0 1 1 0 0 1Chile Dec. 71 1 1 NA NA NA 1 NA

Aug. 72 1 1 NA NA 1 1 NAOct. 73 1 1 NA NA 1 1 1Dec. 74 0 0 NA NA 1 1 1Jan. 76 1 1 NA NA 0 1 1Aug. 82 1 1 1 1 0 0 0Sept. 84 0 1 1 1 1 0 1

Colombia March 83 1 1 1 NA 0 NA 1Feb. 85 1 1 1 NA NA 1 1

Denmark May 71 1 NA NA NA 1 NA 1June 73 1 0 NA NA 1 0 0Nov. 79 1 0 NA NA 0 0 1

Aug. 93 1 0 1 1 1 1 0Finland June 73 0 0 NA NA 1 1 0

Oct. 82 1 0 NA NA 0 1 0Nov. 91 1 0 1 0 0 1 1Sept. 92 1 1 1 0 0 1 1

Indonesia Nov. 78 1 1 1 NA 0 0 1April 83 1 1 0 NA 1 1 1Sept. 86 1 1 1 1 0 0 0

Israel Nov. 74 1 1 NA NA 1 1 1Nov. 77 1 1 NA NA 0 1 0Oct. 83 0 0 NA NA 0 0 0July 84 0 0 NA NA 1 0 0

Malaysia July 75 1 0 NA NA 0 1 1Mexico Sept. 76 1 1 NA NA 0 1 0

Feb. 82 0 0 0 0 0 0 0Dec. 82 0 1 0 0 1 1 1Dec. 94 0 1 1 1 1 1 0

Norway June 73 1 0 NA NA 0 0 0Feb. 78 1 1 NA NA 1 1 0May 86 1 1 NA NA 0 1 0Dec. 92 1 0 NA 1 0 1 1

Peru June 76 1 0 NA NA 1 1 0Oct. 87 1 1 NA NA 0 1 0

Philippines Feb. 70 NA NA NA NA NA NA NAOct. 83 1 1 1 0 0 1 NAJune 84 1 1 1 1 NA 1 NA



21/28

TABLE A1Continued

Exports(8)

Imports(9)

Terms oftrade(10)

Realexchange

rate(11)

Reserves(12)

Realinterest-ratedifferential

(13)Output

(14)

Stockprices

(15)

Deficit/GDP(16)

Totalnumber of

signalsb

(17)

NA NA NA NA NA NA NA NA NA 1001 1 1 1 1 NA NA NA NA 801 1 0 1 0 1 NA 0 NA 541 1 NA 1 1 1 1 1 0 801 0 1 0 0 1 1 1 0 601 0 0 0 1 1 0 0 0 471 0 0 0 1 1 1 0 1 670 1 0 1 1 1 0 NA NA 621 1 0 1 1 1 1 NA NA 711 1 NA 1 1 1 0 NA NA 771 0 1 0 1 NA NA 0 0 421 0 1 0 1 1 NA 0 0 671 0 1 0 1 1 1 1 0 81

1 0 1 1 1 1 1 1 0 811 0 1 1 1 1 1 1 1 751 1 1 NA 1 NA NA NA 0 881 1 1 0 1 NA NA NA 0 801 1 0 1 1 NA 1 NA 0 831 1 1 1 1 NA 1 NA 0 751 1 1 0 1 NA 1 NA 0 751 1 1 1 0 1 1 1 1 751 0 1 0 1 1 1 1 0 691 1 1 1 1 NA 1 1 0 851 0 0 1 1 1 1 1 0 79NA NA NA NA NA NA NA 1 0 800 1 NA 0 0 NA 0 0 1 330 1 1 1 0 NA 1 0 0 46

1 0 NA 0 1 1 1 1 0 670 0 NA 0 1 NA 0 0 NA 271 0 NA 0 1 NA 1 0 0 421 0 1 1 1 1 1 1 1 751 0 1 1 1 1 1 1 1 811 1 1 0 0 1 0 NA 0 571 1 NA 1 1 0 1 NA 0 771 0 1 1 0 1 1 NA 1 671 1 1 1 1 NA 1 NA 0 920 0 0 1 1 NA 1 NA 0 501 0 0 0 0 NA 1 NA 0 171 0 0 0 0 NA 1 1 0 311 1 1 0 0 NA 1 NA 0 581 1 0 0 0 NA 0 NA 0 42

1 1 0 1 0 0 0 1 0 251 1 1 1 1 0 1 1 0 690 0 1 1 1 1 0 0 NA 600 1 NA 0 0 NA 1 1 0 331 1 1 1 1 NA 1 1 0 851 1 1 0 0 NA 1 0 0 541 0 1 0 1 1 1 1 1 731 1 NA 1 1 NA NA NA 1 801 1 1 0 1 NA 0 NA 0 58NA NA NA 1 NA NA NA NA NA 1001 0 1 1 1 0 NA 0 0 501 0 1 1 1 0 NA 1 0 71



22/28

multitude of weak and deteriorating economicfundamentals suggesting that it would be diffi-cult to characterize them as self-fulfilling crises.

During much of 1997 and 1998, the financialpress has frequently stressed that the crises inAsia are a new breed, as they supposedly oc-curred against a backdrop of immaculate fiscal

and economic fundamentals. Yet our analysis ofearlier episodes reveals that many of the fea-tures and antecedents of the crises in Asia werecommon to a substantial number of crisis epi-sodes in Latin America, Europe, and elsewhere.Consider an economy that had successfully sta-bilized inflation, enjoyed an economic boom,

TABLE A1Continued

Country Crisis

M2multiplier

(1)

Domesticcredit/GDP

(2)

Realinterest

rate(3)

Lending-deposit

rate ratio(4)

ExcessM1

balances(5)

M2/reserves

(6)

Bankdeposits

(7)

Spain Feb. 86 NA NA 1 1 NA NA 1Feb. 76 1 1 NA NA 1 1 1July 77 0 1 NA NA 0 1 1Dec. 82 0 1 1 NA 0 1 0Sept. 92 1 0 1 0 0 0 0May 93 1 0 1 1 0 1 0

Sweden Aug. 77 1 0 0 1 NA 1 1Sept. 81 1 1 0 0 NA 1 0Oct. 82 1 1 1 0 NA 1 0Nov. 92 1 0 1 1 NA 1 1

Thailand Nov. 78 1 0 1 NA 0 0 0July 81 1 0 1 NA 0 1 1Nov. 84 1 1 1 NA 0 1 0

Turkey Aug. 70 NA NA NA NA NA NA NAJan. 80 0 NA NA NA NA 1 1March 94 1 1 1 1 0 1 1

Uruguay Dec. 71 NA 1 NA NA 1 1 NAOct. 82 1 1 1 NA 0 1 0

Venezuela Feb. 84 1 1 NA NA 0 1 0Dec. 86 1 0 1 1 1 1 0March 89 1 0 1 0 0 1 1May 94 1 1 1 1 0 1 1Dec. 95 1 1 1 1 0 1 1

OUT-OF-SAMPLE CRISES

Indonesia Aug. 97 0 NA 1 0 0 0 0Malaysia Aug. 97 NA NA 1 NA 1 1 NA

Philippines July 97 1 1 1 1 1 0 0Thailand July 97 0 NA NA NA 0 1 1

SUMMARY STATISTICS

Percent of crises calledc 76 61 89 71 37 81 51Percent called before FLd 74 56 78 50 52 74 44Percent called after FLd 77 65 91 73 26 86 56

Noise-to-signal ratioe 0.67 0.64 0.75 1.52 0.56 0.52 0.67



23/28

and was running fiscal surpluses. However, thiseconomy had liberalized its capital account andits domestic financial sector amidst an environ-ment of weak regulation and poor banking su-pervision. Banking-sector problems emergedand intensified, eventually undermining theability of the central bank to maintain its

exchange-rate committment. While this profilefits Asia rather well, this was Daz-Alejandrosdescription of the antecedents to the fierce Chil-ean crisis of 1982. At the roots of the meltdownof the Thai baht, Korean won, and Indonesianrupiah lay systemic banking problems. Thus, itwould appear that we can only consider these

TABLE A1Continued

Exports(8)

Imports(9)

Terms oftrade(10)

Realexchange

rate(11)

Reserves(12)

Realinterest-ratedifferential

(13)Output

(14)

Stockprices

(15)

Deficit/GDP(16)

Totalnumber of

signalsb

(17)

1 0 1 1 1 1 NA 1 0 821 1 1 0 0 NA 1 1 1 851 1 1 0 1 NA 1 1 1 771 0 1 0 1 1 1 1 0 600 0 0 1 0 1 1 1 1 441 0 0 1 1 1 1 1 0 631 1 1 1 1 0 1 1 0 731 1 1 1 1 1 1 0 0 671 0 1 1 1 1 1 0 0 671 0 1 1 1 1 1 1 1 871 0 1 0 0 1 NA NA 1 461 1 1 1 1 1 NA 1 1 861 1 1 1 1 1 NA 0 0 71

NA NA NA NA NA NA NA NA NA NA1 1 NA 1 0 NA NA NA 1 780 0 0 1 1 1 0 0 0 561 NA NA NA 1 NA NA NA 0 831 1 1 1 1 1 1 NA 0 791 0 1 1 1 NA 1 0 0 621 0 1 0 1 1 0 0 1 630 1 1 0 1 1 0 1 1 631 0 1 0 1 1 0 1 0 690 1 1 1 1 1 0 1 0 75


0 0 NA NA 0 1 1 0 NA 250 0 NA 1 1 1 0 0 NA 60

0 1 NA 1 0 1 NA 0 NA 621 0 NA 1 1 NA 1 1 NA 70

SUMMARY STATISTICS

85 52 75 59 75 86 74 64 2778 73 73 58 70 78 68 54 2189 40 77 60 78 89 76 68 31

0.40 1.10 0.70 0.14 0.55 0.90 0.46 0.38 0.49

a A 1 indicates that there was at least one signal in the 24 months preceding a crisis. NA indicates that some or all of thedata were missing for the 24-month period.

b Number of variables signalling a crisis as a proportion of the number of indicators for which data are available (inpercent).

c Number of crises accurately called divided by the number of crises for which data are available for that indicator. Forexample, for the M2 multiplier, the indicator correctly identified crises 76 percent of the time (54 out of 71).

d FL: Financial liberalization.e The noise-to-signal ratio is the number of bad signals as a proportion of the number of months outside the crisis window

divided by the number of good signals as a proportion of the number of months in the crisis window.



24/28

crises as a new breed if we ignore the numerouslessons history offers. Thus, among the lessonsthat emerge from this analysis is the obviouscase for strong banking regulation and supervi-sion to allow countries to sail smoothly throughthe perilous waters of financial liberalization.Yet, the Asian episodes of 19971998, likemany of their earlier Latin American counter-

parts, also remind us that capital inflows can onoccasion be too much of a good thing.

The results presented in this paper are a firststep in evaluating the complex linkages betweencurrency and domestic financial crises. Analyzinghow the authorities deal with the banking prob-lems and how the problems affect exchange-rateexpectations will help determine whether a bank-

TABLE A2ANATOMY OF BANKING CRISESa

Country Crisis

M2multiplier

(1)

Domesticcredit/GDP

(2)

Realinterest

rate(3)

Lending-deposit

rate ratio(4)

ExcessM1

balances(5)

M2/reserves

(6)

Bankdeposits

(7)

Argentina March 80 1 0 1 0 0 1 0May 85 1 NA 1 1 0 1 0Dec. 94 NA NA 1 1 NA NA NA

Bolivia Oct. 87 1 1 1 0 NA 1 0Brazil Nov. 85 1 1 1 1 1 1 0

Dec. 94 NA NA 1 1 NA 1 1Chile Sept. 81 1 1 1 0 1 0 0Colombia July 82 1 NA 1 NA 0 NA 1Denmark March 87 NA 1 1 0 0 1 1Finland Sept. 91 1 1 1 0 0 1 1Indonesia Nov. 92 1 0 1 1 0 0 1Israel Oct. 83 0 0 1 NA 1 1 1Malaysia July 85 1 0 1 0 0 1 1

Mexico Sept. 82 0 1 1 NA 1 1 1Oct. 92 0 0 1 1 1 0 0Norway Nov. 88 1 1 1 NA 0 1 1Peru March 83 0 1 NA NA 0 0 1Philippines Jan. 81 1 0 NA NA 0 1 1Spain Nov. 78 0 0 NA NA 0 0 1Sweden Nov. 91 1 0 1 1 NA 1 1Thailand March 79 0 0 1 NA 0 1 1

Oct. 83 1 1 1 NA 0 1 0Turkey Jan. 91 1 0 1 1 1 0 1Uruguay March 71 NA 1 NA NA 1 1 NA

March 81 1 0 1 NA 0 1 0Venezuela Oct. 93 1 1 1 1 0 1 1


Malaysia Sept. 97 NA NA 1 NA 1 NA NAPhilippines July 97 NA 1 1 1 1 NA NAThailand May 96 NA NA 1 NA 1 1 1

SUMMARY STATISTICS

Percent of crises calledc 73 50 100 57 32 75 67Noise-to-signal ratiod 0.50 0.59 0.45 1.93 0.82 0.71 1.03



25/28

ing crises will lead to a balance-of-payments cri-sis. We have only considered macroeconomic datain our list of indicators, but data of the health ofbank balance sheets would be a logical comple-ment to the macro data. Future analysis couldprovide a more detailed evaluation of the univar-iate and multivariate signaling properties of vari-ous macroeconomic time series and composite

indices along the lines of Diebold and Rudebusch(1989) and Stock and Watson (1989). Indeed, thatwould appear to be a logical first step in the designof an early-warning system designed to help de-tect when a crisis is coming.

While this paper has focused on the similar-ities and common patterns across crises, itwould also be useful to investigate whether

TABLE A2Continued

Exports(8)

Imports(9)

Terms oftrade(10)

Realexchange

economics paper reinhart

Documents