On Linear Discrimination With Accounting Ratios

Much research in Accounting and Finance is concerned with using the linear discriminant function (LDF) to model accounting‐based ratios to predict financial events and other variables. Little attention has been given to the conditions under which the model is optimal, and to any resultant biases in model output associated with accounting ratios that do not meet optimality. This study lists conditions for LDF optimality, and discusses the potential problems when accounting numbers do not meet such conditions. This knowledge is extended by reported results of an empirical study which show that the cross‐sectional properties of some ratios arr not temporally stable. Finally, suggestions are offered to improve modeling efforts.