Default Prediction for Housing and Utilities Management Firms Using Non-Financial Data

Abstract

For many years, financial ratios have been used as predictors of default. However, biases in financial statements of companies in Russia call into question the applicability of this approach. An alternative approach is to use non-financial data in such models. The purpose of this paper is to find out whether non-financial data, such as information related to court trials, unscheduled inspections and firm age, can significantly improve the accuracy of default prediction in the housing and utilities management industry. This part of the services sector is chosen as one of the riskiest industries, in which firm default affects not only conventional stakeholders such as banks, shareholders, employees, etc, but also customers. A dataset of 378 housing and utilities management firms which have faced default and 765 solvent “healthy peers” is used to create and test default prediction models. Logistic regression is used as the classification algorithm. The results suggest that addition of non-financial data can significantly improve the accuracy of default prediction, and moreover, non-financial data can be used exclusively without any financial ratios to create classification models which show acceptable accuracy. The paper contributes to the existing literature by providing new evidence on the benefits of using non-financial data in default prediction models. In addition, we were able to collect a unique dataset of unscheduled inspections and use this data for default prediction, which appears to be the first case of this kind.

Keywords

• default prediction
• credit risk assessment
• housing and utilities management firms
• non-financial data