Dynamics of the predictive power and investment performance of a generalized takeover prediction model By Julian

We here investigate the dynamical stability of both the predictions and the economic outcomes of a takeover forecasting model. The time dependence is studied by calculating multiple forecasts over the 10-years period 1999-2008. In order to account for geographical variations, the same forecasting methodology has been applied to both the UK and the US. A nominal logistic regression has been chosen to estimate the predictive models as it is considered to accurately represent the model used by the recent takeover prediction literature. Estimation and prediction sample are built using the data of one calendar year of merger and acquisition (M&A) activity. Our results show that, for high cut-off probabilities, the models exhibit high predictive performances. However, due to a few poor performances in some specific years, the predictions remain dynamically unstable. For low cut-off values, the model gains stability showing persistent and significant predictive performance. From an economical perspective, we find that, when adjusted for risk, the highest portfolio performances are achieved for low cut-off values suggesting that the maximisation of returns is correlated with the predictive stability of the model rather than with its occasional high levels of prediction. As a result, although not confirmed by both portfolio analyses, takeover models appear to be a potentially profitable basis for a dynamically stable investment.