Searching For Asset Pricing Factors Using High

The existence of the capital market anomalies and the decline of the explanatory power of the traditional model led to more and more factors being proposed to explain the return.But most of these factors were independently proposed.We know that in econometrics,In order to minimize model deviations due to the lack of significant independent variables,it is common to choose as many independent variables as possible to improve the explanatory and predictive accuracy of the model.Therefore,these independent pricing factors may have a large model deviations,so its conclusion is not so convincing.Based on high-dimensional data and many firm level characteristics,this paper studies asset pricing from a high-dimensional perspective.Traditional seemingly unrelated regression(SUR)models are generalized in the high-dimensional case.We solve the shortcomings of covariance matrix estimation and variable selection from the high-dimensional SUR model.The empirical results show that the proposed method is superior to the single-equation Elastic Net and the Kitchen-Sink regression both in terms of explanatory power and prediction angle.In addition,the results have successfully captured the market factors such as the Fama-French five-factor,which are currently the mainstream research on asset pricing theory,as well as the firm level characteristics we are interested in.The paper also explains theoretically the main pricing factors of cpiret,share volume and special items.It shows that our method is reasonable and scientific in theory.