Tail Model Risk And Stress Testing
This paper incorporates uncertainties of model risk in stress testing. To do so, our approach consists of mapping the Gaussian (or other alternative) distribution quantiles to the quantiles of the empirical distribution using a statistical criterion: the mapping is implemented if the transformation factors are both statistically significant and increase a penalized measure of fit. Our home price application indicates the relevance of our mapping in terms of capturing the home price declines observed in the Great Recession. Importantly, the derivation of confidence intervals provides buffers for model risk. For example, while regulatory capital may use the point estimate of the quantile of the distribution, risk management organizations can use these confidence intervals to bolster resilience of financial institutions to home price declines.
Forecasting, Adverse Scenarios, Model Risk, Home Prices.