Balance of payment crises in emerging markets: how early were the ‘early’ warning signals?

Abstract

Although many papers have already proposed empirical models of currency crises, the timing of such crises has received relatively little attention so far. Most papers use indeed a static specification and impose the same lag structure across all explanatory variables. This, by construction, prevents from specifically timing the crisis signals sent by the leading indicators. The objective here is to fill this gap by considering a set of dynamic discrete choice models. The first contribution is to identify how early in advance each explanatory variable sends a warning signal. Some indicators are found to signal a crisis in the very short run while others signal a crisis at more distant horizons. The second contribution is to show that state dependence matters, albeit mostly in the short run. The results have important implications for crisis prevention in terms of the timeliness and usefulness of the envisaged policy response.

Keywords:

• financial crises
• currency crises
• emerging market economies
• sudden stops
• dynamic discrete choice model
• panel data
• state dependence

JEL Classification::

• F31
• F32
• C23

Acknowledgements

I am very grateful to the Editor, Mark Taylor, and to two anonymous referees for their valuable suggestions. I would also like to thank for helpful comments and discussion with Mike Artis, Anindya Banerjee, Agnès Bénassy-Quéré, Marcel Fratzscher, Eleni Kyriazidou, Søren Johansen, François Laisney, Philip Lane, Helmut Lütkepohl, Daniel McFadden, Michael Sager, Frank Vella and seminar participants at the European Central Bank, at the Banque de France and at the European University Institute. I am indebted to Noëmie Lisack for outstanding research assistance. The views expressed in this article are those of the author and do not necessarily reflect those of the European Central Bank, the Banque de France or the Eurosystem.

Notes

¹ See also Calvo et al. (2004), and Calvo and Talvi (2005). The expression refers to the saying ‘it is not speed that kills, it is the sudden stop’ and was previously used in this context by Dornbusch et al. (1995).

² Exceptions include Sachs et al. (1995) and Bussière and Mulder (1999a, b), who use a continuous dependent variable.

³ See Berg and Pattillo (1999) for a review.

⁴ Some attempt has been made at estimating the role of political variables in a cross-sectional context; see for instance Bussière and Mulder (1999a). However, although political variables can increase the goodness-of-fit of the model, substantial cross-country heterogeneity remains and calls for the use of fixed effects.

⁵ ‘Forwarding’ the crisis index in order to define a time window is relatively common in the literature on financial crises, see for instance Fuertes and Kalotychou (2004) as well as the references therein.

⁶ In fact the article used three regimes in a multinomial logit framework, which was shown to increase the fit of the model compared with a two-regime model. While this technique proved useful to obtain a good fit, it does not specifically address the issue of state dependence and does not allow for different time lags across explanatory variables, which are the two objectives of the present article.

⁷ Several papers have tested for slope homogeneity. For instance, Bussière and Fratzscher (2006) run out-of-sample forecasts for the Asian crisis, which suggest that the same model can apply to different countries/episodes.

⁸ See also Frankel and Rose (1996), Sachs et al. (1996), Eichengreen et al. (1995) or Bertoli et al. (2010) for a discussion of crisis indices.

⁹ See Bussière (2007) for further details on the methodology. Several alternative measures have been used, such as the rate of change in the preceding 12 or 24 months, with broadly similar results. Jongwanich (2008) discusses the role of exchange rate overvaluation in explaining the 1997–1998 currency crisis in Thailand.

¹⁰ Time and country indices are omitted in this section to simplify notation.

¹¹ Heckman (1981a, p. 122). The way crises have been defined implies that crises are typically separated rather than appearing in succession so the notion of accumulation in states is not directly relevant here.

¹² It also suggests that results for the effect of the lending boom variable are simply not very robust, as noticed already in a cross-section context (Bussière and Mulder, 1999).

¹³ Bussière and Fratzscher (2006), where the dependent variable is also a transformed index similar to the one used in Tables 7 and 8, also concludes that fixed effects do not have a significant impact on the results. This contrasts with the results presented in Table 2–5 (with the untransformed index).