Abstract
M&A announcements can result in substantial positive or negative abnormal acquiring-firm stock returns and sizeable associated dollar value gains or losses. Unfortunately for decision makers tasked with evaluating potential deals, the existing M&A literature focuses on the in-sample analysis of cross-sectional determinants of acquirer stock price reactions, thereby providing little guidance as to whether a certain deal will generate or reduce shareholder wealth. This paper instead focuses on the out-of-sample forecasting of acquirer share price reactions to M&A announcements. We employ acquirer-, target- and deal-specific features commonly used in the literature and test the accuracy of linear and nonlinear models using state-of-the-art Machine Learning methodology. Random Forest and k-Nearest Neighbor models perform best in terms of forecasting accuracy, but are closely followed by Ridge and OLS approaches. We further document the forecasting models’ ability to disentangle value-creating from value-destroying
deals, and illustrate the substantial incremental monetary gains associated with following the suggested heuristic.