The Ex-ante Classification of Takeover Targets Using Neural Networks

In this article we use a net with a single hidden layer and back-propagation to discriminate between targets and non-target firms. The model is estimated on a state-based sample, though the best net is selected and subsequently analysed on the basis of a cross-validation sample which is representative of the true population. Tests of model performance are constructed on the basis of performance in the cross-validation sample. In addition to the usual asymptotic assumptions commonly made we also use a bootstrap pairs sampling algorithm, and a residual based sampling algorithm to generate alternative standard errors and confidence intervals