Forecasting volatility

This paper studies the problem of volatility forecasting for some financial time series models. We consider several stochastic volatility models including GARCH, Power GARCH and non-stationary GARCH for illustration. In particular, a martingale representation is used to obtain the l-steps-ahead forecast error variance for the class of GARCH models. Some closed-form expressions for the variance of l-steps-ahead forecasts errors are given in terms of ψ weights and the kurtosis of the error distribution.