Consumer Bankruptcy Prediction Using Balanced and Imbalanced Data

Magdalena Brygała

doi:10.3390/risks10020024

Abstract

This paper examines the usefulness of logit regression in forecasting the consumer bankruptcy of households using an imbalanced dataset. The research on consumer bankruptcy prediction is of paramount importance as it aims to build statistical models that can identify consumers in a difficult financial situation that may lead to consumer bankruptcy. In the face of the current global pandemic crisis, the future of household finances is uncertain. The change of the macroeconomic and microeconomic situation of households requires searching for better and more precise methods. The research relies on four samples of households: two learning samples (imbalanced and balanced) and two testing samples (imbalanced and balanced) from the Survey of Consumer Finances (SCF) which was conducted in the United States. The results show that the predictive performance of the logit model based on a balanced sample is more effective compared to the one based on an imbalanced sample. Furthermore, mortgage debt to assets ratio, age, being married, having credit constraints, payday loans or payments more than 60 days past due in the last year appear to be predictors of consumer bankruptcy which increase the risk of becoming bankrupt. Moreover, both the ratio of credit card debt to overall debt and owning a house decrease the risk of going bankrupt.

Highlights

Personal bankruptcy has grown from a relatively rare household event a couple of decades ago to a fairly common occurrence today (Zhu 2011)
There is a large group in society that is very close to the declaration of consumer bankruptcy, it has not yet announced bankruptcy, which may change due to, for example, a sudden increase in credit card debt or a shock-related event, for instance, loss of employment, illness, divorce or other family problems or mortgage debt (Sullivan et al 2000)
According to Akosa (2017), the results show that imbalanced data can affect the performance of a model

Summary

Introduction

Personal bankruptcy has grown from a relatively rare household event a couple of decades ago to a fairly common occurrence today (Zhu 2011). Consumer bankruptcy is one of the possibilities that natural persons can use to deal with insolvency that may result from poor financial management, inappropriate consumption habits, unexpected situations, e.g., illness, job loss (Caputo 2008). In 2019 in the United States nearly 751,000 consumers filed for personal bankruptcy (U.S Courts 2021). Consumer debtors filing for bankruptcy in 2019 reported having total assets of $82 billion and total liabilities of $112 billion White (1998) concluded that the number of bankruptcies would double if all debtors who have problems with paying their debts went bankrupt.

Objectives

Methods

Results

Conclusion