A false sense of security in applying handpicked equations for stress test purposes

The purpose of this paper is to promote the use of Bayesian model averaging for the design of satellite models that financial institutions employ for stress testing. Banks employing ’handpicked’ equations – while meeting standard economic and econometric soundness criteria – risk significantly underestimating the response of risk parameters and therefore overestimating their capital absorption capacity. We present a set of credit risk models for 18 EU countries based both on the model averaging scheme as well as a series of handpicked equations and apply them to a sample of 108 SSM banks. We thereby aim to illustrate that the handpicked equations may indeed imply significantly lower default flow estimates and therefore overoptimistic estimates for the banks’ capital absorption capacity. The model averaging scheme that we promote should mitigate that risk and also help establish a level playing field with regard to a common level of conservatism across banks.