|
Company financial performance prediction using statistical and artificial intelligence methods: a case study of listed firms after going public |
|---|---|
| รหัสดีโอไอ | |
| Title | Company financial performance prediction using statistical and artificial intelligence methods: a case study of listed firms after going public |
| Creator | Pornpawee Supsermpol |
| Contributor | Navee Chiadamrong, Advisor |
| Publisher | Thammasat University |
| Publication Year | 2568 |
| Keyword | Initial public offering (IPO), IPO performance prediction, Financial performance prediction, Hybrid modelling, Machine learning, Logistic regression, Thai capital market |
| Abstract | Predicting the financial performance of firms following an Initial Public Offering (IPO) is an important yet challenging task for investors and market participants. Accurate forecasts can assist stakeholders in evaluating the potential prospects of newly listed firms and making more informed investment decisions. However, predictive modelling in this context is often complicated by limited data availability, particularly in emerging markets. In addition, many Machine Learning (ML) models prioritize predictive accuracy while offering limited interpretability, making it difficult to understand the factors associated with post-IPO outcomes. This thesis investigates the prediction of post-IPO financial performance for firms listed on the Stock Exchange of Thailand (SET). Using a dataset of 134 IPO firms listed between 2003 and 2018, the study examines whether firm internal capability and IPO-specific variables observed prior to the IPO can be used to predict firm performance in the years following listing. Post-IPO financial performance is evaluated across three years after the IPO and classified using profitability-based measures derived from return on assets, return on equity, and return on sales. Several benchmark ML models are first evaluated to assess their predictive performance in this setting. To improve interpretability while maintaining predictive capability, this study proposes a hybrid modelling framework that combines threshold patterns extracted from decision trees with penalised multinomial logistic regression. The tree-based component identifies informative thresholds and interaction structures among variables, which are then incorporated as predictors in the regression model. This approach enables the model to capture non-linear relationships while preserving a transparent statistical structure. The robustness of the results is assessed through repeated experiments using multiple random seeds.The results indicate that the proposed hybrid framework achieves competitive predictive performance while providing interpretable insights into the determinants of post-IPO outcomes. Several financial indicators and threshold patterns are consistently associated with different performance classes across experimental runs. These findings provide empirical evidence on post-IPO performance prediction in the Thai capital market and illustrate how interpretable hybrid modelling approaches can be applied to relatively small financial datasets. The results may also offer practical insights for investors evaluating IPO opportunities and for firms seeking to understand financial characteristics associated with stronger post-listing performance. |
