Implications of Model Uncertainty for Bank Stress Testing

We aim to raise the awareness that model uncertainty stemming from stress test satellite equations that relate bank risk parameters to macro-financial variables can be significant. Based on a set of credit risk models derived by means of a Bayesian model averaging (BMA) methodology we conduct a stress test for 75 European banks to highlight that i) an optimistic equation choice can imply significantly overstated capital estimates, ii) model uncertainty contributes on average about 35% to overall uncertainty in our application, and iii) the impact of model uncertainty feeding through regulatory risk weights can easily turn twice as sizable as that from loan losses. Model methods that account for model uncertainty, such as the BMA, should mitigate the risks arising along these three dimensions and help establish a level playing field with regard to an equal extent of conservatism across banks.