Most leading indicator models or warning systems employ economic theory to guide them in the selection of the indicators to be included. At this point, the literature is sufficiently rich in theories on the causes of currency crises to separate them into first- and second-generation models.

First-generation models focused on imbalances in macroeconomic fundamentals, often citing the role of excessive public-sector deficits. The seminal paper in the first-generation category is that of Krugman.^a In his model, under a fixed exchange rate regime, domestic credit expansion in excess of money demand growth leads to a gradual but persistent loss of international reserves and ultimately to a speculative attack on the currency. This suggests that expanding money supply to finance fiscal deficits may cause excessive credit growth, which in turn brings pressure to bear on a currency in a fixed exchange rate regime, and eventually causes its collapse. Variables like the fiscal deficit and the monetary base grow faster and beyond a certain size before a crisis. A number of papers have extended Krugman's basic model to suggest that the evolution of the real exchange rate, the trade or current account balance and domestic interest rates could be used as leading indicators of crises.

While the traditional approach stresses the role played by declining international reserves in triggering the collapse of a fixed exchange rate, some recent models have suggested that the decision to abandon the parity may stem from the authorities' concern about the evolution of other key economic variables. For example, the presence of banking problems reflected in the relative price of bank stocks, the proportion of non-performing loans, central bank credit to banks or a large decline in deposits could also indicate the high likelihood of a crisis. Leading indicators may also include political variables.

Second-generation models allow the possibility of crises even though an economy does not suffer continuous deterioration in its economic fundamentals. These models underscore the role of expectations of economic agents in a crisis. Some of these models introduce the concepts of herding behaviour and contagion effects. In herding behaviour, "information costs may lead foreign investors to take decisions based on limited information and therefore to be more sensitive to rumours".^b Further, as financial markets are now much more interrelated than they were a decade ago, individual investors now have many more ways to cover risks, and losses in one market may force liquidations in others. Thus, a loss by investors in country A may cause them to liquidate in country B (or a set of countries B, C etc.), possibly triggering a crisis, if not at least putting some stress on the financial system of country B. These models have the unpleasant implication that crises arising in financial markets are likely to be unpredictable; if a crisis can occur in a country even without poor fundamentals, then it will not be reflected in its usual macroeconomic variables since it is merely guilty by association.

There are three variants of the contagion effects hypothesis.^c The first hypothesizes that trade links may be the channel through which attacks on one currency are transmitted to another. If a country's competitor in exports suffers a devaluation of the currency, then the country itself may also suffer a currency attack owing to the perceived need to maintain export competitiveness with the other country. It has been found that the greater the correlations of export shares, the greater is the tendency of these countries' currencies to depreciate by more or less the same amount.

The second channel of contagion is related to financial market linkages. There is a demonstration effect on the part of borrowers; when domestic borrowers saw what was happening in other Asian countries where the exchange rate peg had given way they tried to cover themselves to avoid the same fate. This only served to increase the downward pressure on their own currencies. Another financial channel saw the crisis spread through the action of international lenders; heavy losses in one market by retail investors forced liquidation in several markets. It is still typical for investors to calculate historical correlations between rates of return on investments in different countries; high correlations would thus lead to immediate selling of all investments should one of them come under pressure.

A third variant relates to a common cause. For instance, policies undertaken by industrial countries may have similar effects on emerging markets. An interest rate increase in the United States may decrease the attractiveness of several other markets at the same time, leading to a sudden and unexpected outflow of short-term capital, which is highly sensitive to interest rate differentials, or a change in tax rates in a developed country may affect the savings behaviour and so the size and composition of international capital flows.

These theories of contagion provide important reasons why the empirical relationship between economic data and the occurrence of crises is likely to be imperfect.

Recent empirical studies

Most recent empirical studies on early warning systems follow one of the three models outlined in this chapter: the signals approach, probit or logit models, and regression models. A few additional studies belonging to each category are described briefly below.

(a) Signals approach

Work by Lawrence Lau and Jung-Soo Park^d (which preceded that of Salvatore) sought to identify signals of fundamental vulnerability in selected Asian countries similar to those which existed in the economy of Mexico prior to the Mexican crisis. This work was initially carried out in September 1995 and is therefore certainly a prescient one. Selected economic indicators for China; Hong Kong, China; Indonesia; Malaysia; Philippines; Republic of Korea; Singapore; Taiwan Province of China; and Thailand were compared with those of Mexico. The paper itself provides sparse details on the authors' criteria for determining whether the indicators are at problematic levels, that is, sufficiently similar to Mexico.

Most of the countries that were later hit by the Asian crisis were found to have several of the following commonalities with Mexico (though not necessarily all at the same time): falling real exchange rates; low or lowered rate of growth of real GDP; high relative rate of inflation; high interest rate differential; rising interest rate differential; high real rate of interest; large negative trade balance; large negative current account balance; and high ratio of foreign portfolio to foreign direct investment.

(b) Probit or logit models

A study by Dongchul Cho and Kiseok Hong^e examined the cause of the Asian crisis empirically. The study first identified the general causes of a currency crisis using a comprehensive data set, and then applied the result to the Asian case. The study then compared various international linkages of currency crises with each other and determined which one was most relevant.

The study used a probit equation that related crisis episodes to standard macroeconomic fundamentals along with various contagion measures to determine which variables were most significant. The dependant variable for probit estimation was a crisis index, which took the value 1 if a currency crisis occurred and 0 otherwise. A currency crisis was defined as a depreciation of the nominal exchange rate of at least 25 per cent, which was also at least a 10 per cent historical increase in the rate of depreciation for the country. Three sets of variables were used in the model: (i) macroeconomic indicators: GDP growth rate, real domestic credit growth rate, inflation rate, fiscal deficit/GDP ratio; (ii) external variables: current account/GDP ratio, changes in the terms of trade, changes in the real exchange rate, foreign reserves/short-term debt ratio, FDI/GDP ratio, total foreign debt/GDP ratio, short-term debt/total foreign debt ratio; and (iii) foreign conditions: GDP growth rate, interest rate in developed countries, crisis incidents of foreign countries. Data covered 103 developing countries, including the crisis-hit Asian and Latin American countries, for the years 1980 through 1996.

The results of the estimation suggested that in general a country might undergo a currency crisis even when its domestic fundamentals were not particularly weak. The contagion of crises was found to take place most often among countries that were in geographical proximity; economic linkages such as international trade and finance turned out to be insignificant once geographical proximity was taken into account. This pattern of contagion implied that intraregional cooperation was critical in preventing future currency crises.

A study by Daniel Hardy and Ceyla Pazarbasioglu^f used a multinomial logit model to identify the role of the macroeconomic, banking sector and real sector indicators in the emergence of banking system difficulties. The data sample covered 50 countries, 38 of which suffered a total of 43 episodes of banking system crisis and comprised 323 observations for the period 1980-1997. Three groups of explanatory variables were used: real-sector variables (rate of growth, consumption growth, investment growth and capital-output ratio); banking-sector variables (deposit liabilities, credit to private sector, and foreign gross liabilities); and potential shocks (inflation, real interest rate, real exchange rate, real growth in imports and terms of trade).

The empirical findings suggested that banking distress was associated with a fall in real GDP growth, boom-bust cycles in inflation, rapid credit expansion, rapidly increasing capital inflows, rising interest rates, a declining capital/output ratio, and a sharp reduction in real exchange rates with an adverse trade shock. The paper also found that in Asian countries, variables such as credit growth and rising foreign liabilities, which were proxies for vulnerability of the banking and corporate sector, were more important than traditional macroeconomic indicators. This suggests that the construction of early warning systems should not focus solely on macroeconomic indicators. In fact, this seems to have been the story of the Asian crisis: everyone was preoccupied with the "nice" macroeconomic picture but missed seeing the weaknesses at the micro level, for example, banking system flaws.

(c) Regression models

Kaminsky and Schmukler^g conducted a regression analysis to determine the reaction of financial markets to news - either rumours or fundamentals - in crisis episodes. The test was conducted for the Asian crisis for the period beginning in 1997 until the end of 1998. Nine economies were included: Hong Kong, China; Indonesia; Japan; Malaysia; Philippines; Republic of Korea; Singapore; Taiwan Province of China; and Thailand. The analysis concentrated on the 20 largest one-day swings in stock prices.

The results indicated that the source of the largest one-day swings could not be explained by any apparent substantial news - either economic or political - but seemed to be driven by herd instincts of the market itself. The results also indicated that rumours, or concerns unrelated to the actual information, affected foreign markets as strongly as they affected domestic financial markets, suggesting the presence of important contagion effects. There was also some evidence that investors overall reacted instantaneously and more strongly to bad news than to good news.

Dadush, Riordan and Wolfe^h used a regression model to identify the main sources of error in World Bank forecasts completed at the end of 1997. Recognizing that the evolution of a crisis was inherently impossible to anticipate with accuracy, the authors sought to draw some lessons for macroeconomic modelling and prediction which might improve the understanding of the forces at work.

Analysis of forecast errors was carried out for GDP growth, domestic demand, private consumption, total investment, exports, imports and current account balance. The substantial forecast errors for these variables were attributed to four factors: a failure to appreciate fully the interactions of foreign credit with the domestic banking sector and highly indebted domestic firms; underestimation of the extent of spillover effects within the region; inadequate forecast of the decline in regional import volumes and in prices of traded goods induced in part by the downturn in regional activity; and failure to anticipate the depth of the continued recession in Japan.

The study suggested that in the light of the foregoing, the following lessons can be drawn: (i) more careful and frequent monitoring of balance-sheet effects is called for, particularly for the accumulation of short-term foreign currency liabilities in the banking system and changes in corporate leverage; (ii) in determining the severity of a crisis, it is clear that expectations matter; the light of this, investors ought to hedge their exposures; (iii) since the advent of contagion is an important factor, analysis and forecasting of developments should be conducted in a regional or global context; (iv) forecasting models should incorporate interactions between the banking and non-financial corporate sectors; and (v) in the projection of the effects of external stocks, there is a need to account for the special nature of developing countries as marginally creditworthy borrowers.

To conclude, although each of the above models does make some contribution to understanding the causes of crises and contagion, none of them can accurately predict the timing of a crisis. Hence, these models have a useful but limited role in formulating an early warning system.

Next

Footnotes:

^a Paul Krugman, "A model of balance-of-payments crises", Journal of Money, Credit, and Banking, vol. 11, No. 3 (August 1979), pp. 311-325.