A feature engineering based model architecture for modeling initial public offerings

This study proposes a model architecture for modeling Initial Public Offerings (IPOs) by incorporating a diverse range of data sources, encompassing both textual and numerical inputs. Language models, machine learning models, and deep learning architectures are combined to make the final ensemble predictions. Several rich features are engineered and interpreted while providing scope for debugging using the game theory-based Shapley Additive exPlanations (SHAP) values. The study results indicate that the feature-engineering is highly eloquent in IPO performance modelling. The study findings have high economic implications range from detecting the market trends to overall market stability. • A novel feature engineered based predictive model is developed to model Initial Public Offerings. • SHAP values are used for model debugging. • The study findings have high economic implications.