Predicting Debt Crises
In this project, Matt DiGuiseppe and I use random forest models to derive predictions of debt crises. We use the existing political economy literature to help us build our forecasting model, allowing theory and forecasting to connect more than in previous financial forecasting efforts. In addition, our forecasting models will generate predictions for financial crises for the next 18 months (July 2023-December 2024). These predictions, as derived from the model parameters, will be time-stamped, preregistered, and publicly displayed. This approach alleviates concerns that we are attempting to overfit existing observational data. The quality of our forecasts will be evaluated by events that have yet to occur, akin to the pre-registration of experimental data. Our pre-analysis plan can be found here.