External Crisis: Early Warning System Analysis for the ECCU
A recent approach to assessing external sustainability is to develop a systematic empirical framework for predicting currency and balance of payments crises (a so-called “early warning system”) using economic and financial indicators that provide a timely indication of the potential vulnerability of a country’s balance of payments position. Early warning system models can be a useful adjunct to the IMF’s traditional surveillance process, as such an objective approach avoids country-specific biases in the evaluation of the potential for crises. The predictability of currency and balance of payments crises has been examined in a number of recent papers (see Berg and Pattillo, 1999), and in this appendix an extension of the IMF’s Developing Country Studies (DCSD) model is applied to the ECCU.
The modeling approach used is as follows. A multivariate probit model is estimated on monthly data for a panel of 35 developing economies over the period 1970:1-2000:7. 1 The dependent variable in the model takes a value of one if there is a balance of payments crisis within the next 24 months, and zero otherwise. A crisis is defined to have occurred when an “exchange market pressure” index (calculated as a weighted average of monthly real exchange rate depreciations and monthly percentage declines in reserves) exceeds its country-specific mean by more than three standard deviations. The independent variables in the early warning system model include: real exchange rate overvaluation relative to trend; current account deficit as a percentage of GDP; foreign exchange reserve losses; export growth; and the ratio of external debt to foreign exchange reserves.2 The probability of a crisis is found to increase when the real exchange rate is overvalued relative to trend, reserve growth and export growth are low, and the ratios of the current account deficit to GDP and external debt to reserves are high. The estimated coefficients from the model can then be used to generate predictions in the form of the probability of a crisis occurring in any one country during the next 24 months, given the current values of the explanatory variables.3 Predicted probabilities above a certain threshold (typically taken in the literature as either 25 or 50 percent) indicate that the model is signaling the likelihood of a crisis (assuming unchanged policies) within the next 24 months.4 In effect, the signaling of an imminent crisis is tantamount to the model indicating that under unchanged policies, the path of external imbalances is unsustainable. Of course, a crisis may not eventuate if appropriate policy actions are taken to address the underlying economic problems.
The estimated crisis probability for the six ECCU countries (as a whole) peaked in late-1997 and again in late-1999 (see Appendix Figure V.1). However, since mid-2000 the ECCU crisis probability has remained relatively low at around 10 percent. It should be noted, however, that the figure for the Union as a whole masks significant differences in crisis probability across individual countries.
Berg, A., and C. Pattillo, 1999, “Predicting Currency Crises: The Indicators Approach and an Alternative,” Journal of International Money and Finance, Vol. 18, pp. 561–86.
Helbling, T., A. Mody, and R. Sahay, 2003, “Debt Accumulation in the CIS-7 Countries: Bad Luck, Bad Policies, or Bad Advice,” (unpublished manuscript, International Monetary Fund, Washington, DC).
International Monetary Fund, 2003, “Public Debt in Emerging Markets: Is It Too High?,” in World Economic Outlook (September 2003), International Monetary Fund, Washington, DC.
International Monetary Fund, 2004, “Natural Disasters and the Macroeconomic Implications,” in Eastern Caribbean Currency Union: Selected Issues, Chapter II.
Prepared by Pedro Rodriguez and Paul Cashin.
The accumulation of public debt during the second half of the 1990s has also affected the probability of an external crisis in ECCU countries. See the Appendix for an assessment of the vulnerability of the ECCU.
Four events have been identified: Antigua and Barbuda had a reduction in the value of its arrears in 1997 equivalent to 13.1 percent of GDP; the government of Grenada borrowed an amount equivalent to 11.4 percent of GDP to extinguish lease arrangements that had not been previously included as debt; public enterprises in St. Kitts and Nevis increased their debt by 8.8 percent of GDP in 1997 (not included in our fiscal accounts since only central government data is available for this country); and the government in St. Vincent and the Grenadines took over private debt in 1999 (for an amount equivalent to 17.5 percent of GDP).
Data on public sector debt, government fiscal balance, interest payments, public debt, and real GDP growth were obtained from ECCU country authorities, while those on natural disasters are from the EM-DAT database compiled by the Centre for Research on the Epidemiology of Disasters (see also IMF, 2004). Data on public debt corresponds to the public sector for all countries, but the coverage of the fiscal data varies across countries, since consolidated public sector data was only available for Antigua and Barbuda, St. Lucia, and St. Vincent and the Grenadines.
We include 1990 because natural disasters in the region usually occur in the second part of the year, and, as a result, they may have affected the public debt accumulation process of 1991.
Natural disasters are defined here as events due to natural causes that caused 10 or more fatalities, affected 100 or more people, or resulted in a call for international assistance or the declaration of a state of emergency.
The countries include the six IMF members of the ECCU (Antigua and Barbuda, Dominica, Grenada, St. Kitts and Nevis, St. Lucia, and St. Vincent and the Grenadines), the ECCU as a whole, and four non-ECCU Caribbean countries (Barbados, Guyana, Jamaica, and Trinidad and Tobago).
Data are taken from: real bilateral exchange rate, external reserves, current account, gross domestic product, and exports (IMF, IFS); external debt (Bank for International Settlements, Eastern Caribbean Central Bank, and IMF staff). For each of the ECCU countries and for the Union as a whole, the total external reserves of the Union are used in the calculations of the model.
The coefficients from the probit model and the updated independent variables are used to generate out-of-sample predicted probabilities of crisis for the period 2000:8–2003:10.
The threshold probability for an alarm that minimizes a loss function equal to the weighted sum of false alarms (as a share of total tranquil periods) and missed crises (as a share of total pre-crisis periods) is 18 percent for the 35 country sample used in this study.