Predicting Merger Targets and Acquirers from Text

Abstract

Mergers and acquisitions are important, infrequent events that are difficult to predict. In this study, we explore the use of a U.S. firm’s SEC filings to predict whether the firm will be an acquirer or a target of an acquisition within a year of the filing. Our approach uses text regression, in which frequencies of words and phrases in the document are used as independent variables in a logistic regression model. We find that word and phrase features can be incorporated into predictive models of being an acquirer or a target; the best performing models in each case involve a different use of text alongside standard financial variables. In particular, the best method for predicting whether a firm will be a target makes use of a novel smooth interaction model that identifies how words and phrases change meaning (with respect to the prediction) depending on a firm’s ratio of market value of firm assets to book value.