a mixed historical formula to forecast volatility.pdfVIP

A Mixed Historical Formula to forecast volatility Roberto Ferulano First Draft: 20 November of 2007 1 Introduction Since the introduction of the CAPM and Black-Scholes models, volatility plays a crucial role in the definition of derivative price. However, volatility is both a random and an unobserved variable, that must be inferred from data. Several models have been proposed to estimate volatility. They typically use either his- torical data (historical volatility) or option market prices (implied volatility). A complete review of volatility model used for forecast can be found in Poon and Granger (2003). Many subsequent papers attempted to improve the forecast volatility and to introduce the practice of checking the quality of the forecast. The work of Hansen and Lunde (2005) and McMillan and Speight (2004) con- firms the superiority of the ARCH model over all of the other models as far as daily forecasts are concerned. The quality of their predictions is due to the comparison of the forecast variance with the cumulative squared returns from intra-day data (realized volatility), instead of the daily squared return. This new measure of volatility was recommended by Andersen and Bollerslev (1998). In order to reduce the high persistence typical of the GARCH model, Marcucci (2005) introduces a new regime-switching GARCH model, that gives optimal results for one day forecast. Koopman et al. (2005) compare historical and im- plied volatility on the SP100 stock index. Their results show that historical volatility allows for better results on a 1-day horizon and that the best per- forming model is an ARFIMA model. Gonzalez-Rivera et al. (2004), instead, introduce four different loss functions (two from the statistical world and two from the economic