A multilogit approach to predicting corporate failure—Some evidence for the UK corporate sector

In this paper we demonstrate, using conventional logit analysis, how a number of non-financial variables, and more particularly the time lag in publishing annual accounts, can discriminate between failing and non-failing quoted firms up to three years before failure. We then, in a novel way, demonstrate how this information can be combined in a multilogit model. Previous models which have incorporated data sets one or more years prior to failure have only addressed the problem of whether a company will fail and not when it will fail. The new approach adopted here is concerned not only with predicting whether a company will fail, but with simultaneously predicting when it will fail, based on data up to three accounting periods before failure. This approach, if successful, would provide users with a much more informative analytical tool than that reported in previous failure prediction studies.