Domestic and foreign literature mainly concerns on the perspective of investors' subjective perception for investment portfolio, but in reality, the market risk is not systemic risk which is not transferred with the will of investors and has a strong endogenous characteristic. Therefore, it believes that investors should pay attention to the impact of market risk constraints on portfolio strategy. VaR has some defects such as not meet the subadditivity, etc., therefore, this paper intends to use CVaR to build a model of market risk constraints. Firstly, this paper analyzes the optimal asset allocation strategy model under the static linear loss-aversion and constructs a dynamic market risk measure method based on TGARCH-EVT-POT-GPD method and presents the ETF optimal asset allocation strategy model based on dynamic conditional risk constraint under time-varying loss-aversion, finally solves this model based on Genetic algorithm. In order to simplify the research, this paper selects a stock index futures contract which has the characteristics of continuous trading, and choose the trading day data of the seven futures contracts as continuous sample. The conclusions of empirical research are as follows:(1)The reference return rate and the loss aversion parameter fixed, when the confidence level reached the maximum, the average investment weight of the individual asset with the low risk invested by the investors of loss averse will be higher, and the variance will be small, and the average CVaR of individual assets and its corresponding variance will also increase gradually. In addition, with the increase of the confidence level, the average return rate and the average CVaR of the portfolio will show a trend of gradually increasing. This shows that the investment strategy of the investors is more conservative, and more sensitive to the estimation of the risk under this condition.(2)When the loss aversion parameter and confidence level are fixed, if the degree of loss aversion of investor is higher, even though the high reference rate of reference return rate, the use of this model can also make the average level of suffering the excess loss by investors reduce in the future.(3)When the other parameters are fixed and the loss aversion parameter and the confidence level change respectively, the correlation value of the return rate obtained based on the dynamic CVaR constraint is more than the yield the return rate correlation value obtained based on the dynamic VaR constraint. The correlation values of CVaR based on this model are less than VaR.