Backtesting market risk models in a standard normality framework

ABSTRACT This paper looks at various ways in which the problem of evaluating a market risk forecasting model can be reduced to the simple problem of testing for iid standard normality. The transformations involved allow us to extend the onestep ahead Berkowitz backtesting framework and its resulting prediction of iid standard normality to encompass complications associated with truncated distributions, multi-step ahead density forecasting models, multivariate density forecasting models, risk aggregation and parameter uncertainty.