A0490
Title: A bias-reduced approach for dynamic estimation of extreme risk measures in financial time series
Authors: Hibiki Kaibuchi - SOKENDAI The Graduate University of Advanced Studies (Japan) [presenting]
Gilles Stupfler - ENSAI and CREST (France)
Yoshinori Kawasaki - The Institute of Statistical Mathematics (Japan)
Abstract: The question of estimating risk measures at extreme levels is important in financial applications, both from operational and regulatory perspectives. We estimate alternative risk measures to the most widely-used Value-at-Risk that are extreme expectile, both expectile- and quantile-based forms of the expected shortfall in a time dynamic setting. This is because replacing quantiles with their least square analogues, called expectiles, has recently received increasing attention. For dynamic estimations of such risk measures, we: (i) filter the financial returns using an AR(1)-GARCH(1,1) model; (ii) apply an asymptotically bias-reduced estimator of extreme quantiles to the standardized residuals after filtering; (iii) use an asymptotic relationship between quantile and expectile (or expected shortfall). The results are illustrated on a financial real dataset.
