Based　on　realized　GARCH　model　with　skew　t　distribution,　we　proposed　a　tail　risk　estimation　method　which　incorporates　both　high　frequency　and　low　frequency　information.　Realized　GARCH　models　with　RV,　RRV　and　BPV　constitute　the　competing　models.　The　empirical　analysis　using　high-frequency　data　of　Shanghai　composite　index　shows　that　the　realized　GARCH　model　estimates　VaR　and　ES　more　accurately　than　the　EGARCH　model,　and　that　incorporating　realized　measures　robust　to　microstructure　noise　and　jump　contributes　to　improve　the　estimation　accuracy　of　VaR　and　ES.　Moreover,　the　performance　of　the　realized　GARCH　model　is　robust　during　the　pre-and　post-crisis　period.　©,　2015,　Systems　Engineering　Society　of　China.　All　right　reserved.