An exponentially weighted quantile regression via SVM with application to estimating multiperiod VaR

The square root of time rule under RiskMetrics has been used as an important tool to estimate multiperiod value at risk (VaR). However, the conditions for the rule are too restrictive to get empirical support in practice since multiperiod VaR is a complex nonlinear function of the holding period and the one-step ahead volatility forecast. In this paper, we propose a new model by considering an exponentially weighted quantile regression via SVM to provide greater accuracy for multiperiod VaR measure. In both numerical simulations and empirical studies on three stock indices, the proposed model outperforms several traditional methods including the volatility models, filtered historical simulation, and linear quantile regression approaches in terms of the value of the number of significant entries, the mean absolute error, and the p value of prediction test in Harvey et al. (Int J Forecast 13:281-291, 1997).