Factor affecting the error in individual stock’s return forecasting: Appling machine learning with Spark MLlib

Abstract

The Capital Asset Pricing Model (CAPM) measures the linear connection between risky asset return and systematic risk. CAPM is a theoretical underpinning for contemporary finance. The empirical character of the CAPM, on the other hand, is a contentious subject among scholars since the CAPM makes several assumptions that are difficult to satisfy in reality. In practice, the trend of mixing artificial intelligence with financial foundations theory has resulted in more efficient and appropriate forecasting models. The primary goals of this research are as follows: Using the CAPM and the Support Vector Regression algorithm (SVR), anticipate the return of individual stocks and identify the elements influencing the prediction inaccuracy of this combined model. The analysis makes use of data from firms listed on the Ho Chi Minh City Stock Exchange from December 2012 to September 2020, on a monthly period. The data is divided into two stages in the study: the first is used to optimize the parameters, and the second is used to assess the error of the model based on Spark MLlib. According to research, the stock return forecasting model based on the SVR algorithm is more effective than the CAPM; additionally, the study discovered that company-specific risk (VAR), overall risk (SD), CAPM error (RMSECAPM), and mean return (MEAN) are the main factors influencing the difference between the forecast error of the SVR model for each individual stock.