Modelling and forecasting recessions with classical and regime-dependent logistic models: evidence from Latvia

Recession forecasting is important for economic policy, financial risk management, and business planning. This study examines recession modelling and forecasting using publicly available quarterly Latvian macroeconomic data. Three logistic specifications are developed and compared: classical logit, threshold logit, and smooth transition logistic models. The nonlinear specifications allow the effects of key recession predictors to vary across economic regimes. Predictors are selected using correlation analysis and multicollinearity diagnostics, based on which inflation, manufacturing confidence, and lagged recession dynamics are chosen. We assess model performance using in-sample fit measures, including information criteria and McFadden’s pseudo- R 2 , and out-of-sample forecast accuracy using the Brier score and mean absolute error. The models are also compared using the area under the receiver operating characteristic (ROC) curve (AUC). The results indicate that regime-dependent models provide stronger in-sample fit than the classical specification. In the out-of-sample evaluation, the smooth transition logistic model records the lowest Brier score and mean absolute error among the three models. The results provide empirical evidence that regime-dependent logistic specifications can be useful for modelling recession probabilities in small open economies like Latvia, although the limited number of observations, imbalanced data, and only 18-step-ahead forecast calls for cautious interpretation of the forecasting results.