Predicting Acquisitions in India

Abstract

Statistical models for predicting takeover targets by using publicly available information, specifically historical accounting information, has attracted considerable academic endeavour. These empirical studies draw from the stylized fact that has unequivocally emerged from literature on performance of mergers and acquisitions: that target firms gain abnormal returns when a takeover announcement is made. Hence, it has been hypothesized that early prediction of takeover targets can stimulate strategic trading that can consistently ‘beat the market’, and make abnormal returns. While it has now been generally proven that such a strategy cannot succeed within semi-strong efficient markets, attempts continue to construct such prediction models to identify potentially valuable firms that can at least provide higher returns under a new management with synergistic propositions. Besides, the characteristics identified by a robust model are also used for preliminary exploration for a potentially good target by acquirers. Following this strand of literature, this paper builds a prediction model for acquisition targets in India using logistic regression. For the estimation of the logistic regression, 122 target firms of acquisitions during the three year-period from 2002 to 2005 were considered, and matched with non-acquiring, non-target firms that had similar promoters' holdings and belonged to the same industry as the target. Results from logistic regression indicate that, a typical target is inherently strong with high growth and large free cash flow, in spite of high debt levels, but encumbered by an inefficient management, who are probably disciplined by takeover market. Traditional determinants of US and UK studies, viz., size and growth-resource imbalance are not significant in the Indian context. Methodological care was taken at various steps to avoid known biases. Estimation period was taken for a modest three year period rather than a longer period to ensure minimal changes in the macro-economic landscape that might have a bearing on the target characteristics. Further, both raw accounting ratios, and industry adjusted ratios were used to account for non-normality of such data. To build the prediction model, cut-off values were calculated using two methods, one that minimized statistical errors and another that maximized returns; again, the latter was found to be superior. Finally, the prediction model was tested on an out-of-sample database of acquisitions that took place during 2005-2006 and was found to yield prediction accuracies up to 91 per cent.