Research On Optimization Of Performance Attribution Based On Barra Model

By virtue of its specialized operation,risk sharing and transparent information,public offering of fund has gradually become a significant role in the capital market.With the rapid expansion of the industry scale,China’s public fund industry is moving towards a period of high grade development,and its proportion in residents’ wealth management is also rising.It has become the largest institutional investor in A-shares and gradually controls the pricing power of the A-share market.However,due to the late rise of our fund industry,there is still a certain gap between Chinese and worldwide head countries,and the market lacks a complete and effective performance analysis system.Fund performance attribution model is widely used,which can be used to measure positions,help stock investors observe market changes,guide fund investors to adjust portfolio allocation,and reduce the information asymmetry between investors and fund managers.In addition,by providing portfolio style exposure and style drift,product investment strategy can be displayed and future return of the fund can be predicted,so as to enhance investors’ judgment of fund performance and reduce the difficulty of decision-making.Thus,the capital market is in urgent need of a comprehensive,objective and effective fund evaluation system suitable for the A-share market.The series of Barra models launched by MSCI greatly enriched the theoretical basis of the structured risk model.At present,the version widely used by Chinese institutional investors is Barra China Equity Model(Barra CNE5 model).This paper introduces the revenue decomposition ideas of Barra model and the sources and meanings of various Barra factors of CNE5 model.Based on Barra CNE5 model,improvements are made in every link of regression to realize model optimization.The specific approach is to establish time series regression of fund returns and factor returns on the premise of known fund daily net value and Barra factor return,estimate the fund’s factor exposure,and predict the fund’s future returns with the help of the results.Traditional models do separate regression for style factors and industry factors.Based on existing studies,this paper chooses both style and industry factors as explanatory variables.Considering the sparse nature of industry factors and the threshold constraints of factor exposure,this paper transforms multiple regression into an optimization problem with constraints,and carries out sparse reconstruction of industry factor exposure.In order to further improve the accuracy of estimation,multiple linear regression methods were optimized during the selection of independent variables,setting of constraint conditions,sample weighting method and referring to posterior information.Evaluation indicators were designed from the perspectives of whether the estimation was accurate and whether the future income could be effectively predicted,so as to evaluate the accuracy of the model in estimation and prediction.In this paper,895 common stock funds and 3,320 commingled funds are taken as the fund pool,and the funds that have been issued for 120 natural days and have not matured are studied.From January 2010 to January 2022,all semi-annual reports/annual report disclosure dates and 10 trading days before and after each date,a total of 483 cross-sectional days are selected for regression analysis,and the two evaluation indicators are compared.Test the direction and strength of different regression model settings on estimating factor exposure and predicting future returns.The model is compared with the Lasso model commonly used in the industry and the stepwise regression model recognized by the academia.Two sample funds were selected for out-of-sample testing to confirm the effectiveness and practicability of the optimization method and the rationality of the parameter settings,so as to provide reference for market investors.