Predicting IPO Performance from Prospectus Sentiment

Abstract

With the development of society, how to apply artificial intelligence in finance becomes a hot topic. IPO performance is the most significant part of these areas. Therefore, this paper aims to use sentiment analysis to predict the IPO performances in the Hong Kong stock market. In specific, 6 machine learning models are trained, namely Random Forests, Decision Tree, Naïve Bayes, Logistic Regression, LightGBM and a stack model to predict the direction of Hong Kong stocks IPO performances (positive, negative, or little change) in 3, 5,10, 20 and 30 days after their IPOs. This paper will compare all the models with a baseline model which generates random guesses for the direction of IPO performances to conclude about which model works better and which one is more predictable. This paper will show that instead of logistic regression, random forest performs relatively the best across all Ys, and this may be due to the different sample size.