Leading Indicators of the Business Cycle: Dynamic Logit Models for OECD Countries and Russia

In this paper, I develop the leading indicators of the business cycle turning points exploiting the quarterly panel dataset comprising OECD countries and Russia over the 1980-2013 period. Contrasting to the previous studies, I combine data on OECD countries and Russia into a single dataset and develop universal models suitable for the entire sample with a quality of predictions comparable to the analogues of single-country models. On the basis of conventional dynamic discrete dependent variable framework I estimate the business cycle leading indicator models at different forecasting horizons (from one to four quarters). The results demonstrate that there is a trade-off between forecasting accuracy and the earliness of the recession signal. Best predictions are achieved for the model with one quarter lag (approximately 94% of the observations were correctly classified with a noise-to-signal ratio of 7%). However, even the model with the four quarter lags correctly predicts more than 80% of recessions with the noise-to-signal ratio of 25% can be useful for the policy analysis. I also reveal significant gains of accounting for the credit market variables when forecasting recessions at the long horizons (four quarter lag) as their use leads to a significant reduction of the noise-to-signal ratio of the model. I propose using the “optimal” cut-off threshold of the binary models based on the minimization of regulator loss function arising from different types of wrong classification. I show that this optimal threshold improves model forecasts as compared to other exogenous thresholds.