چکیده:
The aim of this research is to investigate the usefulness of variable reduction methods and the non-linear Support Vector Regression method in predicting the stock returns of companies listed on the Tehran Stock Exchange. Using correlation-based methods and factor analysis, optimal variables were selected or extracted from 52 initial variables. Subsequently, using non-linear Support Vector Regression methods, artificial neural networks, and also linear regression, the stock returns of 101 companies listed on the Tehran Stock Exchange were predicted in the time period 1383 to 1392. The empirical findings of this research indicate that Support Vector Regression performs better than the other two prediction methods, and both non-linear prediction methods perform better than linear regression. Furthermore, the research findings indicate the usefulness of variable reduction methods and the existence of a significant difference between the effectiveness of the two methods based on correlation and factor analysis, as well as the superiority of the correlation-based method.
خلاصه ماشینی:
However, despite the better performance of the non-linear Support Vector Regression 8 method compared to Artificial Neural Networks in forecasting some accounting variables (for example, Lee and Min 9, 2004 and Lee and Tao 10, 2010), no research has yet been conducted in the Tehran Stock Exchange regarding return forecasting using this method.
Furthermore, in studies conducted in the field of return forecasting, the primary goal and emphasis have been to provide suitable and accurate models for forecasting, and less attention has been paid to variable reduction 11 (variable selection 12 and extraction 1 Capital Asset Pricing Model (CAPM) 2 Factor Model 3 Arbitrage 4 Technical Analysis 5 Fundamental Analysis 6 Kothari 7 Artificial Neural Network (ANN) 8 Support Vector Regression (SVR) 9 Lee and Min 10 Lee and To 11 Variable Reduction 12 Feature (Variable) Selection variables or factors 1) and the appropriate methods for it have been addressed.
Given the importance of stock returns in the decision-making of investors and creditors, the better performance of non-linear methods, and the significant role of the optimal predictor variable selection and extraction stage in forecasting, the present research examines the performance of two non-linear methods, Support Vector Regression and Artificial Neural Networks, and compares them with linear regression in forecasting stock returns.
In this regard, the present research aims to forecast stock returns using non-linear Support Vector Regression and to examine the usefulness of correlation-based and factor analysis variable reduction methods on forecasting performance.
(
, 
, 
)
