Predicting Corporate Takeover Outcomes Using Machine Learning

Abstract: The aim of this thesis is to investigate if the machine learning based classification procedure, Random Forest, provides superior prediction performance compared to a logistic regression model fitted using the LASSO framework, when predicting outcomes in corporate takeover situations. This is done in the context of merger arbitrage, an event-driven investment strategy. The classification models are fitted using a training data set consisting of 5 922 OECD-domiciled corporate takeover transactions and evaluated on a testing data set consisting of 1 481 observations. Variable selection is based on the extensive research done within the field of takeover prediction. The results suggest that the random forest model outperforms the logistic regression model on all relevant validation measures, such as overall prediction accuracy, sensitivity, and specificity. Given that a vast majority of previous research has been done using logistic regression, this thesis provides cause for considering alternative and complementary classification procedures when attempting takeover prediction.