Title: How to predict financial stress: An assessment of Markov switching versus logit models
Authors: Thibaut Duprey - Bank of Canada (Canada) [presenting]
Benjamin Klaus - European Central Bank (Germany)
Abstract: The gap between the business cycle literature and the literature on currency, banking and financial crises is bridged by (i) comparing a Markov switching (MS) and a binary logit model to assess their ability of predicting the occurrence of high financial stress episodes, and (ii) applying a particular form of MS model to predict the transition to a high financial stress regime and back to a tranquil regime. The dependent variable used by the MS model to capture the financial cycle in the EU countries is a financial stress index. The results indicate that both models have a relatively similar ability to predict high financial stress episodes, with the MS model outperforming the logit model between six to one quarters prior to the onset of financial stress episodes. Based on cross-country estimations, debt service ratios and housing variables indicate a transition to a high financial stress regime, while equity price growth and economic sentiment indicators provide signals for a transition to a tranquil state. The MS model results based on country-specific data suggest that country-specific factors can be identified and might be valuable for the design of early warning models.