Eakasit Leelachutipong
This study investigates the integration of Environmental, Social, and Governance (ESG) factors alongside traditional financial data to predict stock returns in the Thai stock market. Focusing on 129 ESG-rated companies from 2020 to 2025, the research utilized machine learning models to forecast cross-sectional returns and construct monthly rebalanced trading portfolios. Feature selection analysis revealed that traditional financial metrics, such as profitability, valuation, and firm size, provided the strongest predictive signals. ESG features, on the other hand, demonstrated weak predictive power and were excluded from the optimized model. Among the evaluated algorithms, the XGBoost model achieved the best out-of-sample performance, outperforming complex deep learning architectures. By applying a 10-quantile sorting strategy, the model successfully generated long-only and long-short portfolios that outperformed the market benchmark during the 2025 testing period. The strategy's success was largely driven by a systematic preference for large-cap, liquid stocks with deep balance sheets during a broader market downturn. Ultimately, the findings suggest that while sustainable investing is growing, standalone ESG scores offer limited value for short-term return forecasting in Thailand. Instead, machine learning techniques applied to foundational financial data can reliably enhance quantitative portfolio performance.