An evaluation of open- and closed-form distress prediction models: The nested logit and latent class models

Abstract

Introduction As was seen in Chapter 2, the discrete choice literature has witnessed tremendous advances over the past decade. A range of sophisticated choice models have been developed and applied throughout the social sciences. Only very recently has this literature been applied to accounting and finance-related research (see Jones and Hensher 2004). Essentially, the discrete-choice literature has developed down two distinct paths: one is towards open-form (simulation based) choice models, the most prominent of which is the mixed logit model and extensions such as the error component logit model. The other approach has developed down the path of closed-form models (also called generalized extreme value or GEV models), the most prevalent of which are the multinomial nested logit and latent class MNL models. Both open- and closed-form models have a number of unique advantages as well as some limitations associated with their use, hence the issue of their comparative performance is an important empirical question in evaluating the full potential of these models in accounting research. In this chapter, we compare the explanatory and predictive performance of the open-form mixed logit model with two sophisticated and widely used closed-form models, multinomial nested logit and latent class MNL (see Train 2003). Chapter 2 provided an illustration of the performance of the open-form mixed logit model (with error components) in the context of financial distress prediction.