Unbalanced data, type II error, and nonlinearity in predicting M&A failure

Abstract

The traditional forecasting methods in the M&A data have three limitations: first, the outcome of M&A deal is an event with a small probability of failure, second, the consequences of misclassifying failure as success are much more severe than those of misclassifying success as failure, and third, the nonlinear and complex nature of the relationship between predictors and M&A outcome could limit the advantage of logistic regression. To overcome these limitations, we develop a forecasting model that combines two complementary approaches: a generalized logit model framework and a context-specific cost-sensitive function. Our empirical results demonstrate that the proposed approach provides excellent forecasts when compared with traditional forecasting methods.