Forecasting sovereign default using panel models: A comparative analysis

This paper assesses the relative merits of panel time series models in forecasting sovereign default. It explores the contentious issue of whether controlling for time-series and country heterogeneity is important in forecasting emerging market default. For this purpose, it uses conventional inference methods alongside forecasting performance statistics based on both statistical- and economic-loss functions. Since sovereign debt states are rather persistent, it is important to compare the panel model forecasts with naive competitors. For the latter we use a random walk forecast, a naive probability forecast and Pesaran-Timmermann test statistics. Diebold-Mariano tests are also deployed to assess the significance of the forecast accuracy differential across models. Our results corroborate that the choice of the best estimator depends on whether one uses economic or statistical loss functions. Interestingly, models that accommodate cross-section heterogeneity to a large extent are not favoured by either criterion. Models that allow for cross-section heterogeneity only at a regional level are superior under economic criteria, whereas the models with time heterogeneity fair slightly worse. Finally, when statistical criteria are used the homogeneous pooled estimator outperforms the other specifications