Sudden stops of capital flows to emerging markets: A new prediction approach

Given that financial crises not only incur huge amounts of costs but also tend to recur, we have observed a large literature focused on improving the ability to predict financial crises, and such efforts are still on going. Even though many studies have been conducted to predict financial crises, only two approaches have mainly been used for the prediction: the signal extraction (SE) approach and the statistical regression (SR) approach.

In this paper, for the purpose of predicting financial crises, we propose a new approach which combines both of the SE and the SR approaches. This new approach intends to maintain advantages of both approaches and also to alleviate their disadvantages; thereby, we hope the new approach will more accurately predict financial crises. In particular, we first categorize information variables into several sub-groups according to the information contents that the variables are expected to convey. We then construct sub-group indexes from the information variables belonging to the same sub-group. We utilize the SE approach to construct these sub-group indexes. Next, we include not only the sub-group indexes but also several relevant variables related with cross-country variations into the SR framework to predict financial crises. With this combination approach, we can consider many information variables without increasing the number of variables to be included by constructing sub-group indexes. Moreover, the SR framework in the second stage of the new approach allows for optimal fitting of a statistical parametric model as well as its associated statistical inferences and also for including variables related with cross-country variations.

We apply the new approach as well as conventional approaches into actual data and conduct prediction performance comparisons. The empirical results show that the new approach significantly improves prediction ability. The new approach has some potential merits as an alternative approach to improve prediction ability and can also be applied to various types of financial crisis events.