Forecasting Bank Capital Ratios Using the Prophet Model by Facebook

Abstract

This study investigates the efficacy of the Prophet model by Facebook with respect to forecasting bank capital ratios. Bank financial ratios and macroeconomic information are combined to forecast total risk-adjusted capital ratios for 19 large U.S. banks. Using a sample period from March 2005 to December 2020, in-sample results show that the model accurately estimates bank capital ratios over time. As validation, out-of-sample tests indicate that forecasting errors are smaller for Prophet models compared to benchmark ARIMAX models. Based on these and other results, we conclude that the Prophet model does a good job of forecasting bank capital ratios. By implication, it provides a practical forecasting tool for bank regulatory supervisors, management, and investors.

Keywords: bank capital, bank regulation, Facebook, Prophet model, stress tests

JEL Classification: C32, C53, G21