Adjusting loss forecasts for the impacts of government assistance and loan forbearance during the COVID-19 recession

Abstract

Since the onset of the COVID-19 recession, loss forecasting and stress testing models have dramatically overpredicted losses. As all models are pattern recognisers trained on past events, such an unprecedented event inevitably leads to model errors. Rather than, however, view the models as broken, they are useful in providing an upper bound of what could have happened if government assistance and loan forbearance had not been provided. The present work develops an approach for quantifying the short- and long-term impacts of these government and lender policies in order to create quantitative model overlays. These overlays express the problem via a set of key parameters that can be set via management judgment or simulation studies. Examples of this approach and parameter sensitivity analysis are provided using time series models of National Credit Union Administration and Federal Deposit Insurance Corporation call report data. This paper provides a framework for incorporating simulations, simple to complex, into an existing stress testing framework to better project future losses.