Title: Forecasting risk measures using intraday and overnight information
Authors: Douglas Gomes dos Santos - Catholic University of Brasilia (Brazil)
Osvaldo da Silva Filho - Catholic University of Brasilia (Brazil)
Paula Tofoli - Catholic University of Brasilia (Brazil) [presenting]
Abstract: Risk measures such as value-at-risk (VaR) and expected shortfall (ES) are computed from forecasts of return volatility for the full day. When dealing with high-frequency data from markets which operate during a reduced time (e.g., six to seven hours a day), an approach to take into account the overnight return volatility is needed. In this context, we use heterogeneous autoregressions (HAR) to model the variation associated with the intraday activity, with, e.g., realized variance, bipower variation, realized semivariances and signed jump variation as regressors, and to model the overnight return variance, we use augmented GARCH type models. Then, we combine the forecasts from the two types of models to obtain forecasts for the total daily return volatility. We examine the aforementioned procedure in an extensive empirical study using high-frequency data sets (S\&P 500 index and five individual stocks), where out-of-sample VaR and ES forecasts are compared with forecasts from traditional approaches. We benefit from recent results regarding the joint elicitability of VaR and ES, being able to assess the relative forecasting performance of the models in a simple manner. The overall results indicate that the combinations of HAR models with augmented GARCH type models generally produce the most accurate forecasts.