Predicting stock returns using machine learning combined with data envelopment analysis and automatic feature...

Area of Science:

• Financial Markets
• Machine Learning
• Econometrics

Background:

• Predicting stock returns is crucial for investors.
• Existing methods often overlook business efficiency metrics.
• The Vietnamese stock market presents unique challenges for return prediction.

Purpose of the Study:

• To evaluate the efficacy of business efficiency scores in predicting stock returns.
• To compare the performance of various machine learning models for stock return prediction.
• To investigate the impact of automatic feature engineering on prediction accuracy.

Main Methods:

• Data Envelopment Analysis (DEA) to calculate business efficiency scores.
• Collection of financial data for 26 real estate firms (2019-2024).
• Application and comparison of machine learning models (e.g., Deep Neural Network, Gradient Boosted Tree) with technical, fundamental, and efficiency indicators.

Main Results:

• Business efficiency scores significantly improve stock return prediction accuracy.
• Deep Neural Network model showed reduced error metrics (RMSE, MAE, MAPE) with efficiency scores.
• Gradient Boosted Tree with efficiency scores and automated feature engineering achieved the best performance (MAE: 0.122, MAPE: 103.19).

Conclusions:

• Business efficiency scores are valuable predictors of stock returns.
• Automated feature engineering further enhances predictive power when combined with efficiency metrics.
• The findings offer a novel approach for improving investment strategies in emerging markets like Vietnam.