Abstract: In this article, a semi-parametric alpha strategy model is proposed. The benefit of the proposed model is that the time-varying coefficient can explain the time point risk rather than the risk in a period of time. The local least square method is used to estimate time-varying coefficient and the estimate of alpha is easily derived by solving an estimating equation. Based on the semi-parametric model, the 30 low-ranking stock portfolios of alpha value are selected to observe the contrarian effect. It's found that the stock portfolio is better than market return and gain premium, showing the contrarian effect. The influence of the holding length and the stability of a stock on contrarian effect are also studied, and it's found that the selected stock portfolio performs better than CSI 300 index and its constituent stock. The empirical analysis proves that the semiparametric model can serve as a comparatively better predicting model, and our results can provide some theoretical references for managing risk and improving return.

Key words: alpha strategy, semiparametric varying-coefficient model, contrarian effect, accumulated income