The authors show that quasimaximum likelihood (QML) estimators for conditional dispersion models can be severely affected by a small number of outliers such as market crashes and rallies, and they propose new estimation strategies (the two-stage Hampel estimators and two-stage S-estimators)...