Joint Prediction of Turning Points in Credit and Business Cycles: Cross-Country Analysis

Abstract

This paper provides a joint analysis of business and credit cycles with a focus on unobservable factors affecting both cycles, at the cross-country level. Using quarterly data for 19 developed countries and Russia for the period from 1994 to 2018, we build a system of two dynamic probit models, which includes a cross-correlation between the errors of the equations governing the probability of a recession and the probability of credit crisis. The results show that, first, our system allows us to correctly predict 91% of episodes of joint realization of macroeconomic and credit crises and 89% of non-crisis periods in the training sample, and 92% and 95% respectively in the testing sample. Second, switching from two independent regression models to a system of correlated equations significantly (by 16 percent-age points) increases the share of correctly predicted crisis episodes while only slightly (by 7 percentage points) reducing the proportion of correctly predicted non-crisis episodes. Third, our system can predict an approaching crisis earlier, by 1–4 quarters, in comparison with similar single models. Our results complement the literature on forecasting recessions and credit crises. Fourth, it is revealed that the models which have been constructed on developed countries allow one to predict crisis events for Russia. The model we have constructed correctly predicts 100% of joint crisis episodes and 92% of joint non-crisis episodes in the training sample as well as 86% of joint crisis and 90% of joint non-crisis episodes in the testing sample for Russia.