Synthetic Recessions: An Exploration of Model Selection Techniques for Recession Nowcasting
The recession forecasting literature has invested the use of different machine learning algorithms to predict current and future economic states. Because the data itself is primary expansion observations with few recessions it is related to problem of imbalance data learning and learning from disjuncts. Here, I extend the work of (Piger 2020) to address recession nowcasting. I compare the application of resampling algorithms to the baseline approach and propose alternative model selection schemes. I apply twelve classification schemes, two wrapper feature selection methods, four resampling algorithms and four tuning specifications to the FRED macroeconomic dataset. These different results are compared and tested for their predictive performance across six metrics using two ranking schemes.
Ho-Shek, James, "Synthetic Recessions: An Exploration of Model Selection Techniques for Recession Nowcasting" (2020). ETD Collection for Fordham University. AAI27958671.