Abstract:
Over the last decades, the advanced technologies for data acquisition made it easier to collect, store and manage high-frequency data. However, the analysis of observations collected at an extremely fine time scale is still a challenge: these data are characterized by specific features, related to the trading process and the microstructure of the market, which standard time series and econometrics techniques are not able to reproduce. In particular, the behavior of the high-frequency volatility cannot be reflected by a GARCH model, hence, there is a need for more accurate ways to model it. Recently, the Heterogeneous Autoregressive model of Realized Volatility (HAR-RV) has been introduced: it allows for an easy estimation and economic interpretation of the dynamics of the Realized Volatility, a consistent estimator for daily volatility based on intraday returns. The purpose of this thesis is to model and forecast high-frequency volatility, comparing the HAR performances to those of more classical time series models. In doing so, also jump components and leverage effect have been considered.
