A new Bayesian Backtesting framework

Abstract

We propose a new Unconditional Coverage backtest for VaR-forecasts under a Bayesian framework that significantly minimise the direct and indirect effects of p$-$hacking or other biased outcomes in decision$-$making, in general. As a consequence of the global financial crisis during 2007$-$09, regulatory demands from Basel III and Solvency II has required a more strict assessment setting for the internal financial risk models. To put our proposed backtesting technique into practice we employ linear and nonlinear Bayesianised variants of two renowned mortality models in the context of annuity pricing. In this regard, we explore whether the stressed longevity scenarios are enough to capture the experienced liability over the forecasted time horizon. Most importantly, we conclude that our Bayesian decision theoretic framework quantitatively produce a strength of evidence favouring one decision over the other.

Melvern Leung

Manager - Risk Model Validation

My research interests include distributed robotics, mobile computing and programmable matter.