Title: A vine-copula extension for the HAR-RV model
Authors: Martin Magris - Tampere University of Technology (Finland) [presenting]
Abstract: The heterogeneous auto-regressive model of realized volatility (HAR-RV) is revised to account for non-linearity in the volatility variables, namely daily, weekly and monthly volatility components. The additive structure between these terms, although appealing for interpretation, estimation and inference, is linked to (i) a structural autoregressive hypothesis on the nature three components, and (ii) requires the classic set of hypotheses for the OLS estimation. With real high-frequency intraday data, the OLS hypotheses are not always met, while alternatives to the AR structure can be explored. We abandon the additive framework by modelling the joint distribution of the volatility components via Vine copulas. From the preliminary backtesting analyses, our model shows an increased forecasting accuracy (in terms of Diebold-Mariano test) and an improved description of volatility dynamics in terms of quantile exceedances (hit ratios) with respect to the standard HAR-RV model.