Post-IPO underpricing prediction: A case study of Thai capital market

This study develops a machine learning-based model to predict Initial Public Offerings (IPO) underpricing in the Thai capital market by integrating firm-specific variables, market indicators, macroeconomic factors, and aspect-based sentiment factors derived from IPO filings. Using Light Gradient Boosting Machine (LightGBM) as the primary model, the research leverages both structured quantitative data and unstructured qualitative text to capture a wide array of underpricing determinants. Aspect-based sentiment features are extracted from issuer disclosures using a pre-trained domain-specific Thai-language model, and their validity is further evaluated against GPT-4.1. The results demonstrate that incorporating sentiment-based predictors improves model performance, with LightGBM outperforming benchmark models across all evaluation metrics. The findings reveal that market momentum, firm size, book value, and thematic sentiment—particularly in human capital and product narratives—are significant contributors to underpricing. The presence of certain narrative cues may also signal strategic orientation, including openness to innovation, which could shape investor expectations. These findings offer both methodological and empirical contributions, highlighting the utility of combining machine learning with financial text analysis for improved predictive accuracy and interpretability in emerging markets.