Outliers in Garch models and the estimation of risk measures

摘要: In this paper we focus on the impact of additive level outliers calculation risk measures, such as minimum capital requirements, and compare four alternatives reducing these measures' estimation biases. The first three proposals proceed by detecting correcting before estimating measures with GARCH(1,1) model, while fourth procedure fits a Student’s t-distributed model directly to data. former group includes proposal Grane Veiga (2010), detection based wavelets hard- or soft-thresholding filtering, well known method Franses Ghijsels (1999). results, Monte Carlo experiments, reveal that presence can bias severely requirement estimates calculated using model. message driven from second both empirical simulations, is outlier filtering generate more accurate requirements than alternative. Moreover, hard-thresholding gathers very good performance in attenuating effects generating out-of-sample, even pretty volatile periods