Predicting IPO Underpricing: A study on the predictability of IPO underpricing through machine learning algorithms

Abstract: This paper primarily serves to examine whether a specific subset of variables, derived from publicly available pre-IPO data, can be effectively modeled to predict and classify if an IPO will be underpriced using non-linear machine learning (ML) models. Secondly, we analyze whether the performance of ML-based models is greater compared to conventional linear models. Specifically, the focus has been on: linear, neural network (NN), random forest (RF), and gradient-boosting tree (GBT) approaches, including both regression and classification tasks. The findings indicate that given our input data, predictability is attainable solely through a classification approach, albeit with moderate support. Additionally, the evidence of this study favors machine learning models and their ability to capture complex patterns in financial data. This paper may be complemented with further analysis in order to reach a conclusion regarding the actual predictability of IPO underpricing.