VaR is an important tool to control financial risks which can be expressed as capital gains rate into the upper or lower side of quantile. And its effective measurement has always been a research hotspot. In this paper, we extend the linear quantile model (Huangand Lee, 2013) based on high frequency information to the nonlinear model, and further use the Moving-Block bootstrap simulation to test the robustness of the model. Using China stock index futures continuous contract if000 for 5-minute high-frequency data to forecast VaR with combination of factors and variables. Empirical results show that, these models are better than the existing linear quantile prediction models, in particular, considered at high frequency information of the influence of model. Moreover, the models are robust.