Takeover likelihood modelling: target profile and portfolio returns

This thesis investigates four interrelated research issues in the context of takeover likelihood modelling. These include: (1) the determinants of target firms’ takeover likelihood, (2) the extent to which targets can be predicted using publicly available information, (3) whether target prediction can form the basis of a profitable investment strategy, and – if not – (4) why investing in predicted targets is a suboptimal investment strategy. The research employs a UK sample of 32,363 firm-year observations (consisting of 1,635 target and 31,737 non-target firm-year observations) between 1988 and 2010. Prior literature relies on eight (old) hypotheses for modelling takeover likelihood – determinants of takeover likelihood. Consistent with prior studies, I find that takeover likelihood increases with the availability of free cash flow (Powell (1997, 2001, 2004)), the level of tangible assets (Ambrose and Megginson (1992)) and management inefficiency (Palepu (1986)), but decreases with firm age (Brar et al. (2009)). The empirical evidence lends no support to the firm undervaluation, industry disturbance, growth-resource mismatch or firm size hypotheses (Palepu (1986)). I extend prior research by developing eleven (new) hypotheses for target prediction. Consistent with the new hypotheses, I find evidence that takeover likelihood is an inverse U-shaped function of firm size, leverage and payroll burden. Takeover likelihood also increases with share repurchase activity, market liquidity and stock market performance and decreases with industry concentration. As anticipated, the new hypotheses improve the within-sample classification and out-of-sample predictive abilities of prior takeover prediction models. This study also contributes to the literature by exploring the effects of different methodological choices on the performance of takeover prediction models. The analyses reveal that the performance of prediction models is moderated by different modelling choices. For example, I find evidence that the use of longer estimation windows (e.g., a recursive model), as well as, portfolio selection techniques which yield larger holdout samples (deciles and quintiles) generally result in more optimal model performance. Importantly, I show that some of the methodological choices of prior researchers (e.g., a one-year holdout period and a matched-sampling methodology) either directly biases research findings or results in suboptimal model performance. Additionally, there is no evidence that model parameters go stale, at least not over a ten-year out-of-sample test period. Hence, the parameters developed in this study can be employed by researchers and practitioners to ascribe takeover probabilities to UK firms. Despite the new model’s success in predicting targets, I find that, consistent with the market efficiency hypothesis, predicted target portfolios do not consistently earn significant positive abnormal returns in the long run. That is, despite the high target concentrations achieved, the portfolios generate long run abnormal returns which are not statistically different from zero. I extend prior literature by showing that these portfolios are likely to achieve lower than expected returns for five reasons. First, a substantial proportion of each predicted target portfolio constitutes type II errors (i.e., non-targets) which, on average, do not earn significant positive abnormal returns. Second, the portfolios tend to hold a high number of firms that go bankrupt leading to a substantial decline in portfolio returns. Third, the presence of poorly-performing small firms within the portfolios further dilutes its returns. Fourth, targets perform poorly prior to takeover bids and this period of poor performance coincides with the portfolio holding period. Fifth, targets that can be successfully predicted tend to earn lower-than-expected holding period returns, perhaps, due to market-wide anticipation. Overall, this study contributes to the literature by developing new hypotheses for takeover prediction, by advancing a more robust methodological framework for developing and testing prediction models and by empirically explaining why takeover prediction as an investment strategy is, perhaps, a suboptimal strategy