CMStatistics 2020: Start Registration
View Submission - CMStatistics
Title: Modelling self-excitation of extreme returns in financial markets: An AR-GARCH with Hawkes Jumps approach Authors:  Steve Yang - Stevens Institute of Technology (United States) [presenting]
Anqi Liu - Cardiff University (United Kingdom)
Abstract: Extreme value theory (EVT) and Hawkes processes in the AR-GARCH framework are applied to model the tail risk clustering effect. The proposed model improves forecasts of the timing of extreme returns and is particularly useful for downside risk analysis. Due to a large parameter set, we propose a two-step calibration method to estimate the model. We apply this model on 90 stocks, including both large-caps and small-caps, in nine industry sectors. The in-sample experiments show a strong self-exciting of negative extremal of AR-GARCH residuals and it is well captured using our model. The value-at-risk forecasting experiments over the past 25 years confirm that the proposed model produces accurate downside risk estimations. More importantly, the proposed model provides more stable risk analysis results during the market crisis than other existing benchmark models.