Factors Affecting Forecast Accuracy of Individual Stocks: SVR Algorithm Under CAPM Framework

Abstract

The research was carried out with two objectives, including applying the algorithm under the Capital Asset Pricing Model framework (CAPM) to predict individual stocks' return rates and determine the factors affecting the difference in Error for each stock. This study experimented on the Ho Chi Minh City Stock Exchange (HOSE) in the period from 12/2012 to 9/2020 with two stages; in which stage 1 is used to determine the optimal parameters in the Vector Regression algorithm (SVR), and stage 2 is used to test the predictive efficiency by rolling window method. The study pointed that the predictive model using SVR is more effective than CAPM; moreover, the study finds that the specific risk factors (VAR), the overall risk (SD), and the accuracy of CAPM (RMSECAPM) are the factors affecting the difference in the forecast error of the SVR model for individual stocks