The Application Of High-Dimensional Factor Model In Asset Pricing

With the fast development of the financial market,theories for evaluating the price of assets are becoming more and more important.Nowadays,with the increasing complexity of the domestic and international economic environment and the availability of high dimensional financial data,the traditional capital pricing model has more limitations in the interpretation of asset pricing.High-dimensional factor models can take into account more complicated situations amid large financial dataset.The capital pricing model needs to find a set of factors that have strong explanatory power to the returns,thereby improving the interpretability of the model and the prediction accuracy.The Lasso method can obtain a more refined and effective model by shrinking the high-dimensional variables.According to traditional finance theory,the unobservable potential risk information in asset pricing are contained in the covariance of factors and return,while the factors that are ignored due to high correlation are good proxies of potential risks.The standard Lasso method tends to compress this part of the high correlation factor.These potential risk information may have a large impact on the assets,resulting in inaccurate asset pricing.This paper follows the idea of a double-selection Lasso model based on the classical BJS two-step regression and the traditional Lasso(Feng et al.,2017).The double-selection Lasso model can take into account the potential risk factor information that Lasso may ignore to reducing the assets risk.At the same time,based on the characteristics of strong correlation and large number of factors in the double selection Lasso model,this paper proposes to analyze the principal component of factors after the double-selection Lasso model to reduce the number of factors and the correlation between factors.The paper compares the effect of double selection&pca and pca,and also compares the prediction abilitity of OLS model and SVR model,which proves the nonlinear relationship model improving the prediction abilitity.