Nobody should be surprised that it is hard to forecast crises with high reliability; low-risk opportunities for profits are never easy to find. Thus it is especially hard to predict the timing of a crisis. Some economists, however, are skeptical that Early Warning Indicators (EWIs) have any useful predictive ability at all. A common assessment is that EWIs have failed, in the sense that in each historical round of emerging market crises (1982, 1994-2001, 2008) those particular variables that appeared statistically significant in that round did not perform well in the subsequent round. This is not the right conclusion.
In a recent NBER working paper (featured in a ReutersTV interview today), George Saravelos and I began by examining more than eighty contributions to the pre-2008 literature on EWIs. The table below reports which variables were most often found to have performed consistently well in predicting crises in the past. Among 17 categories of indicators, the level of international reserves and exchange rate overvaluation stand out by far as the two most useful leading indicators. These results hold across different crisis episodes stretching from the 1950s to the early 2000s, even though different authors have defined “crisis” and “useful” in different ways.
The 2008-09 financial crisis
The recent global financial crisis was in a sense a perfect experiment for testing the performance of EWIs, because it originated in a shock that was exogenous to the smaller countries of the world. It hit everyone at the same time, so we don’t have to worry about the issue of timing. We can focus on what economic variables indicate vulnerability to such a shock.
To summarize the findings in our paper, the EWIs from the pre-2008 literature do relatively well in predicting which countries got hit in 2008-09, and the indicator that was found to be the top performer in past crises was also the top performer this time: foreign exchange reserve holdings, especially expressed relative to a denominator such as debt.
Other economists have not gotten such strong results. The first wave of analysis of the global crisis — by top scholars including Blanchard, Obstfeld, and Rose — found that few, if any, indicators were useful in explaining which countries got hit the most.i Why do we get stronger results? These papers necessarily defined the crisis period as the year 2008; they lacked data on 2009 at the time when they were written. We define the global crisis as beginning in earnest in the latter part of 2008. (Recall that the failure of Lehman Brothers came in September.) We extend the period under consideration through early 2009, because many aspects of global financial markets and the real economy did not begin to recover until the 2nd quarter of that year. We believe that the difference in the period that is defined to be the crisis is the reason for the difference in results.
The table below summarizes our results. Each column represents another criterion for gauging the severity of the 2008-09 crisis in a particular country. We consider a country to have suffered more from the crisis if it experienced larger output drops, bigger stock market falls, loss in demand for its currency, or a need for access to IMF funds. We regressed each of these dependent variables against the list of useful EWIs indentified in our literature review. A darker color in the figure indicates greater statistical significance for that indicator.
We controlled for GDP per capita, allowing us to mix high-income and low-income countries in the same data set. The last three years have featured a historic role reversal in which some “advanced countries” were badly hit (Iceland, Greece, and others on the periphery of Europe) while some big emerging markets” did much better (China, Brazil, Indonesia and others).
At the top of the list is central bank reserves. We find that the level of reserves in 2007 was a statistically significant predictor of crisis incidence in 2008-09. The best-performing of the reserve measures expresses them relative to short-term debt. This is consistent with earlier findings, including the Guidotti Rule that tells emerging market central banks to hold reserves equal to at least the amount of debt maturing within one year.[ii]
Next on our list is overvaluation measured by past movements in the real effective exchange rate (REER). It is statistically significant in predicting devaluation, exchange market pressure and access to IMF stand-by arrangement programs. A few of the other leading indicators stand out as well. A higher current account balance or level of national savings is associated with lower crisis intensity. Other variables with some explanatory power but less consistently useful across different measures include the level of external debt, its composition (e.g. short-term vs. long-term), and the extent of Foreign Direct Investment (FDI). Broadly speaking, many Eastern European countries suffered in the crisis for these reasons, while most Asian countries did much better.
We conclude that early warning exercises can indeed be a useful tool for assessing future vulnerabilities. The same variable that topped the list of indicators in the earlier literature, central bank reserves, also worked the best in predicting who got hit in the 2008-09 crisis.