Are seasoned equity offerings predictable? Predicting future SEOs with machine learning algorithms

This master thesis explores the predictability of seasoned equity issuance in the United States using the machine learning methods based on logistic regression, decision-trees, random forest and XGBoost. In addition, we investigate the practical value of predicting seasoned equity offerings. Our results show a benefit from employing machine learning for this purpose, with the best performing model (XGBoost) achieving an AUC of 0.72. The random forest model demonstrated similar capabilities with an AUC of 0.71, indicating that sophisticated nonlinear models are suited for this type of prediction problem. Further, the impact of seasoned equity offerings on stock returns is analyzed to identify the possible benefits our models provide. Our efforts included two linear regressions using separate data samples, and one difference-in-differences estimation. These tests failed to provide conclusive evidence; however existing literature implies a negative effect on stock returns from seasoned equity offerings. This thesis contributes to the extensive research conducted on the topic of seasoned equity offerings. While there are no directly comparable publications, we utilize existing literature to improve our thesis and to reflect on our findings. With this thesis we facilitate and encourage further research on this relatively unexplored area of seasoned equity offerings.