Daily VAR Forecasts with Realized Volatility and GARCH Models

In this paper we evaluate alternative volatility forecasting methods under Value at Risk (VaR) modelling. We calculate one-step-ahead forecasts of daily VaR for the WIG20 index quoted on the Warsaw Stock Exchange within the period from 2007 to 2011. Our analysis extends the existing research by broadening the class of the models, including both the GARCH class models based on daily data and models for realized volatility based on intraday returns (HAR-RV, HAR-RV-J and ARFIMA). We find that the VaR estimates obtained from the models for daily returns and realized volatility give comparable results. Both long memory features and asymmetry are found to improve the VaR forecasts. However, when loss functions are considered, the models based on daily data allow minimizing regulatory loss function, whereas the models based on realized volatility allow minimizing the opportunity cost of capital