THE PREDICTION OF AUSTRALIAN TAKEOVER TARGETS: A BINOMIAL AND MULTINOMIAL LOGIT ANALYSIS

This thesis provides the first attempt to predict takeover targets in the Australian context using binomial and multinomial logit models, extending the relatively small amount of work focused in the United States, Canada, and the United Kingdom. Evidence is provided concerning eight main hypothesised motivations for takeovers. Our results confirm the contention that such motivations are inconsistent both throughout time and across economies. Application of models to a true ex-ante predictive sample suggests that individual models are quite inaccurate, but that the use of certain methodological improvements can produce relatively accurate predictive classifications. Multinomial logit models are also compared to binomial logit models to examine whether theoretical benefits exist from discrimination between types of targets. Evidence is provided suggesting that the binomial model is indeed misspecified, but that it is the most appropriate model if the purpose of prediction is investment. Our main empirical finding is that a significantly positive abnormal return of 23.37 percent (68.67 percent prior to robustness adjustments) may be made from an investment in the commonly predicted targets of logit based models. This contradicts the current belief within the extant literature that such returns cannot be achieved through the use of binomial logit models for true ex-ante prediction.