Developing a Prediction-Based Stock Returns and Portfolio Optimization Model

The purpose of this study is to develop a prediction-based stock returns and portfolio optimization model using a combined decision tree and regression model. The empirical evidence is based on the analysis on 112 unique firms listed on the Tehran Stock Exchange from 2009 to 2019. Regression analyses, as well as six decision tree techniques including CHAID, ID3, CRIUSE, M5, CART, and M5 are used to determine the most effective variables for predicting stock returns. The results show that the six decision tree methods perform better than the regression model in selecting the optimal portfolio. Further analysis reveals that the CART model outperforms the other five decision tree models when compared using Akaike and Schwartz Bayesian. This finding is confirmed by comparing the actual returns of the selected portfolio across all six models in 2019. The findings indicate that the predicted returns on portfolio based on the CART model are not significantly different than the actual returns for 2019, suggesting that the selected model appropriately predicts the returns on the portfolio