This paper investigates the use of high frequency data in the estimation of daily and intraday volatility, in order to compute value at risk (VaR) forecasts for the IBOVESPA. GARCH models and deterministic methods for the filtering of seasonal patterns are used in the computation of intraday volatility and VaR forecasts. A comparison with a non-parametric method based on the empirical quantile is done. For daily VaR two simple methods seek to extract the volatility information conveyed by the high frequency data. The first method is based on the sample standard deviation with a moving window, while the second is based on exponencially weighted moving average. Some traditional methods applied to daily data are used for benchmarking. Both methods tested present good performance. For intraday VaR the results indicate that the filtering of the seasonal pattern is a fundamental step in obtaining useful forecasts of volatility and VaR.
