A New Zealand failure prediction model: Development and international implications

In this study we develop a failure prediction model for New Zealand listed companies and theorise on the usefulness and international implications of the results. In developing the model, we obtained data from 60 failing and 60 matched non failing New Zealand listed companies, providing an unusually rich database for model development. Binary logistic regression (logit) was applied on 21 potentially significant variables. The predictive strength was tested using holdout tests, with support from applications to more recent company data. Multicollinearity was minimised through a theoretically informed selection of variables, the Pearson correlation and stepwise regression. The resulting model is significant to 91.7% overall, which tested well for prediction and non-multicollinearity, and reveals a unique combination of variables. We conclude by considering the research implications of small-market model development, and the benefits of local, contextually-based models as evidenced by the results of this study.