reserves – is two or three (country-specific) standard deviations above its (country-specific) mean. 3 In this context, a relevant question is: should a currency crisis be defined as a situation when the exchange rate pressure index is two or three standard deviations above its mean? Since there is no clear consensus in the EWS literature about which crisisdefinition should be used, we attempt to fill this gap by taking an agnostic approach. We use two definitions of currency crisis. According to one definition, a currency crisis occurs when the exchange rate pressure
International Monetary Fund. Strategy, Policy, & Review Department
crisisdefinitions and assessed within a single model when effective to improve evenhandedness. They also represent a step forward in risk assessment modeling, leveraging the recent advances in modeling macroeconomic risks with ML tools.
Machine-learning tools are well-suited to the challenges of macroeconomic risk assessments . Crises are rare and almost always involve novel features (otherwise they would have been anticipated). When they materialize, they often reveal economic relationships which are not regularly observed in normal times, making it challenging to
Mr. W. R. M. Perraudin, Mr. Manmohan S. Kumar, and Ms. Uma Moorthy
. Their results are consistent with ours in that they find explanatory power for models using monthly data (especially in the case of the probit model). However, the crisisdefinition they use is based on a measure of currency pressure. They do not focus, as we do, on trading strategies and their out-of-sample evaluations are based on quite small hold-out samples (just the 1997 Asian crisis).
B. Logit-Based Crisis Forecasting
Much of the literature described above seeks to explain the origins of currency crises by relating their occurrence to contemporaneous
Mr. Richard Hemming, Mr. Axel Schimmelpfennig, and Mr. Michael Kell
F inancial crises in emerging market economies have been the subject of considerable analysis. This section draws on work to date to provide essential background for the subsequent sections.
Most studies of financial crises distinguish between currency, debt, and banking crises. 3 These three types of crisis are discussed in Box 2.1 . Empirical studies—including this one—typically make such a distinction because it is helpful to use specific, and therefore relatively narrow, crisisdefinitions. 4 But many of the
We examine the determinants of external crises, focusing on the role of foreign liabilities and their composition. Using a variety of statistical tools and comprehensive data spanning 1970-2011, we find that the ratio of net foreign liabilities (NFL) to GDP is a significant crisis predictor, and the more so when it exceeds 50 percent in absolute terms and 20 percent of the country-specific historical mean. This is primarily due to net external debt--the effect of net equity liabilities is weaker and net FDI liabilities seem if anything an offset factor. We also find that: i) breaking down net external debt into its gross asset and liability counterparts does not add significant explanatory power to crisis prediction; ii) the current account is a powerful predictor, either measured unconditionally or as deviations from conventionally estimated “norms” iii) foreign exchange reserves reduce the likelihood of crisis more than other foreign asset holdings; iv) a parsimonious probit containing those and a handful of other variables has good predictive performance in- and out-of-sample. The latter result stems largely from our focus on external crises stricto sensu.
, 2001 ) and regime switching with time-varying probabilities ( Abiad, 2003 ). Some studies depart from a regression-based setting altogether: Osband and Van Rijckeghem (2000) use classification rules to identify safety zones for fundamentals under which currency crashes are unlikely to occur, and Ghosh and Ghosh (2002) use a binary recursive tree technique to explore complex interactions among structural and economic variables. Many of the proposed approaches look promising, and most report some improvement over the benchmark models. Differing crisisdefinitions
Mr. Andrew Berg, Mr. Eduardo Borensztein, and Ms. Catherine A Pattillo
more recent model, the Deutsche Bank Alarm Clock (DBAC). Table 1 summarizes the main features of the models under consideration here. It details the crisisdefinition employed, the prediction horizon, the method used to generate predictions, and the predictor variables. Appendix I contains a more complete description of the models. Table 6 , in Appendix I , shows all the crisis dates for these models since January 1999. 4
Table 1. Specification of Early Warning Models
DCSD 1/ (Berg, Borensztein, Milesi- Ferretti, and Pattillo)