Estimating Value-at-Risk Conditional on Bull and Bear Predictions

This paper proposes a new risk management strategy that uses combinations of two different models to estimate VaR by conditioning the model choice on the prediction of bull and bear markets. The main goal of this research is to trigger financial risk managers to take a critical look at their internal risk model(s). They should ask themselves whether there is room for further minimization of the capital charges, by conditioning their risk model choice on the prediction of the market condition. We describe various risk models and a pragmatic strategy to predict bull and bear markets, using a binomial logit model. Using a parametric linear Student t model with EWMA volatility with an optimized l leads to the best results for VaR estimation, looking at the number of violations and the average minimum required capital. For the prediction of bull and bear markets, using the Schwarz Information Criterium for variable selection leads to an out-of- sample hitrate of more than 85%. No combinations of two different models are found, that lead to a significant decrease in the average minimum required capital.