Considering the asymmetry of return distributions and asymmetric impact of positive and negative returns on the quantiles, the paper puts forth a new conditional autoregressive value-at-risk by regression quantiles model with asymmetric absolute values and slops (AAVS-CAViaR) quantile specification for heavy-tail data applications. An empirical study on the evolution patterns of market risk of wheat futures in Zhengzhou Commodity Exchange is performed. The dynamic quantile test, the regression quantile criteria and back testing results support our new model works well. It is found that the asymmetric impacts of price news on the quantiles of returns exist in Chinese wheat futures market. A rule to select proper VaR predicting model is suggested, and we also find AAVS model performs better than indirect GARCH model for our selected sample.