A COMPARATIVE ANALYSIS OF FORECASTING ACCURACY BETWEEN MACHINE LEARNING MODELS AND OLS REGRESSION: EMPIRICAL EVIDENCE FROM THE VIETNAMESE STOCK MARKET

Abstract

This research investigates and compares the predictive performance of stock return forecasting between the Ordinary Least Squares (OLS) regression model and machine learning approaches in the context of the volatile Vietnamese stock market. Using a panel dataset combined with time-series data of listed firms on the Vietnamese stock exchange from 2015 to 2024, the study contrasts the OLS model with three advanced machine learning algorithms, including Artificial Neural Networks (ANN), Random Forest, and XGBoost. Predictive performance is primarily assessed using Root Mean Squared Error (RMSE), alongside Mean Absolute Error (MAE) and out-of-sample R-squared (R²OOS) as robustness measures. The empirical results demonstrate that machine learning models significantly outperform OLS in capturing complex nonlinear relationships in stock returns. Among them, the ANN model achieves the lowest RMSE, indicating the highest predictive accuracy, and generates superior long–short portfolio returns compared to the other models. Furthermore, the Diebold–Mariano test confirms that the differences in predictive accuracy between machine learning models and OLS are statistically significant. Although OLS retains advantages in terms of simplicity and interpretability, machine learning models exhibit clear superiority in predictive performance and quantitative risk management. This study provides important empirical evidence from an emerging market such as Vietnam and offers practical implications for investors and policymakers in optimizing asset allocation decisions.