Early discovery of individual firm insolvency

This paper proposes a new methodology for the early discovery of individual firm insolvency without employing any other firm’s data. The proposed individual-level model can be applied to different firms, regardless of industry type or asset size, and thereby overcomes the sample selection problem commonly found in aggregate-level prediction models. Unlike many previous studies, which assume that the distributions of variables involved do not change over time and that the variables follow a single known distribution, the proposed model can capture each individual firm’s potential multiple data-generating processes and determine the actual distributions exhibited in its own data. Thus, it captures each individual firm’s intrinsic heterogeneity. An empirical study illustrates the greater predictive power of this model compared with the current conventional methods. Specifically, the predictive accuracy of the proposed model is 92.65% and 77.45% for 2 and 5 years prior to actual bankruptcy, respectively. Moreover, the proposed model is adaptive and simple to implement.