Predicting Economic Crises: An Early Warning System
- Catherine Pattillo, and Andrew Berg
- Published Date:
- September 2000
The integration of financial markets around the world over the past decade has posed new challenges for policymakers. The speed with which money can be switched in and out of currencies and countries has increased with the efficiency of global communications, considerably shortening the time policymakers have to respond to emerging crises.
Unhappily, financial markets are notoriously poor at spotting a crisis coming. Markets had little inkling of approaching crises in Mexico in 1994 or in Thailand in 1997. Currency traders did not increase their expectations of a devaluation in the months just before the currency collapsed in either Mexico or Thailand. Neither did interest rate differentials (the spread between domestic and overseas interest rates) widen significantly prior to the Mexican crisis. Spreads on Brady bonds and Eurobonds, instead of providing an early warning signal of worsening confidence, appear to have widened only after currency pressures were well under way.
As a consequence, researchers are taking a fresh look at the forces contributing to financial crises in an attempt to develop early warning systems to signal when trouble may be brewing in currency markets.
While researchers look at many of the same economic and financial variables that most financial analysts do, their strength is that they process the information in a systematic way that maximizes their ability to predict currency and balance of payments crises, based on the historical experience of a large number of countries.
Often an early warning system can translate this information into a composite measure of vulnerability. Being based on a well-defined methodology, it is less likely to be clouded by preconceptions about the expected economic performance of particular countries.
An early warning system can also be a useful tool to rank the relative vulnerability of a group of countries, which is more difficult to assess on a country-by-country basis.
What are the signs to watch for, and can an economic model be devised to predict accurately that an economy is nearing the danger point?
Research by a number of institutions, including the IMF, the U.S. Federal Reserve, the Bank for International Settlements, and others, shows that some models are useful in predicting crises, but more work needs to be done to better understand their causes.
Searching for Common Symptoms
A look at the major international crises of the past two decades illustrates an evolution in the pattern of crises.
The debt crisis of the 1980s, which started with the suspension of payments by Mexico in August 1982 and continued for almost a decade, reflected a mix of external shocks and domestic macroeconomic imbalances that built up during a period of strong capital inflows during the previous years.
Most analysts saw a combination of external factors—including a deterioration in the terms of trade, the sharp rise in U.S. dollar interest rates, and a global economic slowdown—coupled with such internal imbalances as fiscal deficits and currency overvaluation, as contributing to the crisis. Mismanagement of capital inflows, especially through the provision of implicit or explicit exchange rate guarantees to private and state borrowers, added to the problems.
In contrast, the Mexican crisis of 1994/95 suggested different explanations and different fundamentals. This time, many observers pointed to self-fulfilling prophecies by market participants as being largely responsible for the collapse of the peso. But it was also recognized that an underlying vulnerability in the economy made the speculative attacks possible. In Mexico, large current account deficits (triggered by an overvalued currency after a difficult inflation stabilization process) as well as the debt management policy followed just before the crisis had caused an accumulation of sizable short-term U.S. dollar-denominated debt; furthermore, the rapid expansion in the domestic financial sector had resulted in poor-quality loan portfolios and heavy exposure to an exchange rate devaluation.
The Asian crisis of 1997–99 put financial markets at the forefront of attention. Before the crisis erupted, traditional sources of fundamental imbalances were largely absent. The fiscal position was quite robust for all countries, and inflation had been moderate or low for a number of years. With the exception of Thailand, real exchange rates had not displayed any significant appreciation in the years leading to the crisis, and although a slowdown in export growth had been recorded in some of the economies of the region since 1996, it had come after several years of very strong expansion. The loan portfolios of financial institutions, by contrast, had deteriorated significantly, and the corporate sector was excessively indebted and financially fragile—the result of years of poor lending and investment decisions.
Weakness in the financial and corporate sectors seemed to be the only common thread among the affected countries of Asia. But what made things worse was the spread of the crisis from one country to another, as investors abruptly withdrew their money. Spillovers originating in financial markets, as well as herding behavior by investors—not only in terms of joining the stampede out of a currency, but also in their propensity to shift assets out of a whole region—appear to have played an important role in the Asian “contagion.”
Working out a common economic model for all these types of crises obviously is difficult.
But a number of common symptoms can be identified. For example, international reserves may become dangerously low, or the level of external debt commitments become too high relative to the economy as a whole. Alternatively, movements in asset prices may follow a common pattern in anticipation of currency crises.
Along with this, various fundamental data should be monitored, such as the external balance of payments of a country, and its domestic macroeconomic situation.
But there are catches in this approach. For example, the level of vulnerability of an economy implied by a given level of international reserves would vary at times when investors display a generally more or less favorable sentiment toward emerging economies, or a higher or lower propensity toward contagion, or the spread of crisis from one country to another.
A country may have a large volume of debt payments coming due in a given year for purely coincidental reasons, but this would not be a meaningful indicator of vulnerability if the country displays strong fundamentals and the mood in international financial markets is favorable. Conversely, a relatively high level of international reserves could be short-lived in the presence of large deficits and negative investor sentiment.
Another relevant set of symptoms may be found in international financial markets, where a change in prices may indicate increased risk. The expectation of a devaluation of a domestic currency, for example, would widen interest differentials between assets denominated in domestic and overseas currencies. If economic difficulties are foreseen, investors would also pull out of sovereign debt instruments, and possibly the domestic stock markets, and this would be reflected in widening spreads on instruments such as Brady bonds and declining equity prices.
Yet these different asset prices may provide apparently conflicting signals because they are affected by different risks.
This shows that it is possible to construct an early warning system that monitors different variables, but the results should be treated with caution and are open to interpretation. The range of relevant fundamentals has broadened with the changes in international financial markets. In particular, signs of stress in the banking sector should receive more attention and should be considered along with more traditional fundamentals such as those related to the external position of the economy—including the level of the real exchange rate or the current account balance—and those describing the domestic macroeconomic situation, such as fiscal balances and credit growth.
An early warning system should also consider vulnerability indicators—that is, indicators of the likelihood of a successful defense of a currency in case of an attack, as a less vulnerable currency is not likely to suffer serious attacks. In particular, the coverage offered by the level of international reserves relative to possible short-run liabilities of external and domestic origin has been identified as a measure of the vulnerability of the domestic currency to an attack. These variables could be supplemented by other data, such as the forward position of the central bank and other official or private institutions, and available lines of credit or other contingency financing, although these data may be harder to obtain.
These two types of variables—fundamentals and vulnerability indicators—play essentially complementary roles. Countries with weak fundamentals but good liquidity would not stay in a strong position for long, and conversely, countries with relatively bad liquidity but sound fundamentals, while not immune to attacks from “uninformed” investors, are less likely to be attacked and more likely to succeed in defending an attack.
Finally, indicators of market sentiment may also have a role in an early warning system, for example, indicators that can be extracted from asset prices or from developments in other countries that may trigger the risk of contagion. Market sentiment is a powerful force but is difficult to measure, and related indicators may also be relatively uninformative because they tend to provide signals only very late in the gestation of a crisis. Yet it is obvious that analysts should be watching signs of market sentiment. Thus, their incorporation in an early warning system deserves consideration.
To construct a successful early warning system, the key indicators of a looming crisis should be reasonably comparable across time and countries. Many factors that experience suggests might help predict crises are not easily measured and do not meet that standard. Perhaps the most glaring examples involve data regarding the health of financial systems, such as rates of nonperforming loans and capital adequacy. Similarly, variables may be badly mismeasured for various reasons. Accurate and comprehensive information on short-term external debt of the private sector, for example, is unavailable for most countries. These measurement or availability problems imply that it is difficult to incorporate this information into early warning systems that are calibrated using historical episodes.
How Well Do Early Warning Systems Work?
Using different criteria, economists have created a number of models that attempt to aid decision makers in predicting future crises, but, as mentioned earlier, their success rate so far is mixed.
Economists test models in two ways: on the data and time period for which they were designed (called in-sample performance), and on data or a time period for which they were not specifically designed (called out-of-sample performance). Clearly for a model to be useful it would have to provide informative signals out-of-sample, that is, beyond the time period for which the model itself was estimated.
Tests on four empirical models largely created before the Asia crisis showed that the best one was capable of predicting one half of the crises “in-sample,” and one-third “out-of-sample.” False alarms were common: over half the times all these models predicted a crisis was approaching, no crisis occurred.
Although the timing of a crisis seems quite difficult to predict, some of the models did better in predicting the relative severity of crises for different countries. This suggests the models may be more useful in identifying which countries are more vulnerable in a period of international financial turmoil than in predicting the timing of currency crises. This would still be a valuable contribution of an early warning system because it would help focus attention on the countries that need policy adjustments before a crisis develops.
The models have also highlighted the variables that are the most important determinants of crises. All approaches demonstrate that the probability of a currency crisis increases when the real exchange rate is overvalued relative to trend, and when domestic credit growth, and the ratio of M2 money supply to reserves, are high.
Large current account deficits and reserve losses increase the probability of crisis in all three of the models that include these variables. High ratios of short-term debt to reserves ratios are also found to lead to an increased probability of crisis. Some evidence is also found for the importance of other variables, such as export growth, the size of the government budget deficit, and the share of foreign direct investment in external debt. Surprisingly, output growth was not found to be a significant predictor of crises when tested. The evidence on interest rates was mixed. High domestic real interest rates provide informative signals of impending crises while the differential between foreign and domestic real interest rates does not; but in one model high foreign interest rates do increase the probability of a currency crash.
Asia: A Case Study
The crises that struck a number of Asian countries in 1997 came as a surprise to almost all observers. The economies of Korea, Thailand, Indonesia, and Malaysia had for many years been seen as “miracles” with average annual growth rates since 1970 ranging from 6.9 percent in Indonesia to 8.4 percent in Korea. Analysts disagreed largely on why the economies were so successful: Was it some special feature of “Asian capitalism,” or was it largely the application of sound economic policies, economic openness, and high rates of investment?
By 1996, a few clouds had appeared on the horizon. IMF staff had become increasingly worried about a variety of indications of over-heating in Thailand, particularly growing current account deficits, banking failures, and signs of an asset price bubble. Meanwhile, corporate profitability had been falling steadily in Korea, with 6 of the largest 30 corporations going bankrupt by June 1997. Nonetheless, before 1997, few observers expected any sort of major crisis in the region, at least outside Thailand. Dollar spreads were declining and credit ratings generally improving until at least the beginning of 1997 for the countries subsequently affected by crisis.
In July 1997, Thailand was forced to devalue its currency. Over the next few months, several other Asian countries came under severe pressure, culminating in the collapse of the Korean won and Indonesian rupiah in late 1997. Growth was sharply negative in 1998 for most countries in the region, with real gross domestic product (GDP) declines of 15.3 percent for Indonesia, 8.0 percent in Thailand, 7.5 percent in Malaysia, and 7.0 percent in Korea.
The possible causes of the crises in Asian countries can be summarized as follows:
- The crises had their origins in fundamental deficiencies in the affected countries. Some role, particularly in Thailand, was played by traditional macroeconomic problems such as current account deficits that became unsustainably large and an exchange rate that had become overvalued. More important, though, the weaknesses in the crisis countries derived from the interaction of weak domestic financial institutions with large capital flows.
- While there were indeed some structural and macroeconomic problems in the affected countries, these crises were largely avoidable financial panics—rational “bank runs” against otherwise viable economies. The most affected countries had a high ratio of short-term external debt to GDP. This implied that if foreign creditors became convinced that other creditors would not roll over their claims, there were not enough reserves to cover the maturing obligations. Panics became self-fulfilling.
Would it have been possible to predict such a crisis using an early warning model?
While some economists have developed models using up to 20 indicators, in this case the IMF researchers decided to use a simple five-variable macroeconomic model for an after-the-event analysis.
The indicators selected were the degree of real exchange rate overvaluation; the size of the current account deficit as a share of GDP; the growth rate of exports; the rate of growth of reserves; and the ratio of short-term external debt to reserves.
The first four variables are consistent with earlier models that emphasized growing inconsistencies between internal and external objectives in leading to exchange rate collapses. The fifth was suggested largely after the Asian crises as a measure of vulnerability to sudden reversals in capital flows.
A model based on these variables worked fairly well, producing results with relatively high warning signals for Korea, Thailand, Indonesia, and Malaysia, but not for the Philippines.
All crises are not the same, of course, even in the case of Asia in 1997.
- The Thai crisis was most clearly driven by fundamentals. The most important variables were the ratio of the current account deficit to GDP and the rate of decline in reserves. Excluding the short-term external debt ratio, the model yielded a relatively high crisis probability, which rose steadily until the crisis, as would be predicted by a simple first-generation model.
- Malaysia’s crisis was also driven by fundamentals. The key variables were the current account/GDP ratio, reserve losses, and the degree of real exchange rate overvaluation.
- The crises in Korea and Indonesia were predictable using this model and look more like second-generation crises. The key variable was short-term external debt/reserves. The crisis probabilities, based on this, were high and fairly stable throughout most of 1996 and early 1997, consistent with the idea that these countries were in a “zone of vulnerability,” with the Thai crisis and subsequent events tipping the scales in late 1997.
Empirical work on early warning systems is fairly new, at least for models with an explicit forecasting objective, and what triggers a crisis is relatively complex, involving the interplay of numerous economic, political, and psychological factors.
As a result, the performance of early warning systems so far has been mixed. The predictions from the most promising models contain substantial information about the risks of crises, but they often produce false alarms. Models tend to perform much more reliably for the historical period for which they were designed and estimated (“in-sample”) than for later periods (“out-of-sample”). This is because new factors come into play that had not been taken into account, or were not given sufficient weighting, in the earlier model.
Does this mean that economists are doomed to fail in the hunt for a successful early warning system that could be used by governments and financial markets to avert crises? Though not perfect, the economic models used for predicting crises are being improved by continuous study and investigation and even now are useful tools for spotting the early warning signs of distress.
While there has been a recent flurry of research on the prediction of currency crises, it is clear that more work needs to be done to better understand their causes and more successfully predict them.
An important feature of currency crises that deserves further study is the role of contagion, that is, when a crisis spreads like an infection from one country to another. It is difficult to incorporate this into a model, but it undoubtedly plays a role at least in the timing of crises.
Other issues that require further study include such institutional elements as the strength of regulatory frameworks, corporate governance, the degree of openness of the capital account, and variables accounting for political developments.
Early warning systems contain significant useful information and can produce estimates of the probability of crisis and rankings of the susceptibility to crises of different countries. At their current stage of development, they are often only partially reliable. But they remain a useful starting point for analysts and decision-makers.
To conclude, early warning systems should be used but not abused. This means that the primary use of such a system should be as only one tool among many others for analysis of external risks. Any utilization of early warning systems must be embedded in broader analysis that takes into account all the important complexities, some of which a cross-country statistical model inevitably must ignore.