Forecasting corporate default probabilities: a local logit approach for scenario analysis

We propose a new approach for predicting corporate default probabilities and for conducting scenario analyses by combining firm-level and macro time series data. We apply a local projection approach to a simple logit framework and bridge the gap between micro data on firms, for which no scenario is available, and macroeconomic variables, for which the forecaster instead has a scenario. We apply this model to an out-of-sample exercise, estimating it with data through the end of 2017 and forecasting corporate defaults over the following three years. We compute two sets of projections, the first based on the realized values of the macroeconomic time series (baseline), and the second conditional on a scenario that simulates a worsening in the macroeconomic environment comparable to the one observed during the European sovereign debt crisis (adverse). The baseline forecast closely matches the actual corporate debt default rate; under the adverse scenario, the default rate is similar to the one actually recorded in Italy during the sovereign debt crisis. We also run two exercises that make use of the granular forecasts of the corporate default probabilities. First, we assess which sectors are more vulnerable under each of the previous two scenarios (baseline and adverse). Second, we assume that the economy shifts from the baseline to the adverse scenario and construct transition matrices across different risk classes, showing which sectors are more exposed to the shift.