A Fundamental Tool for Predicting Bankruptcy

Abstract

ABSTRACT : This thesis is partially a replication of a study conducted by Altman (1968) on a different sample size and genre. The main purpose of this thesis is to provide investors with a simple tool that will help them make primary decisions regarding their current or prospective investments. The tool, a mathematical model, predicts companies' failures before their actual occurrence, thus offering investors key information. Knowing that the current literature is filled with models that predict bankruptcy, we attempted to utilize one of the most known models, Altman's Z-score, and investigate the prospect of expanding its applicability to embrace a wider-range of observations. We derived a new model out of the original one, in an attempt to enhance the model's accuracy. The study was conducted using publicly-available financial statements from 115 US companies across a wide range of sectors including services, healthcare, and technology of which 39 filed for Chapter 7 or Chapter 11 bankruptcy in the US courts between 2010 and 2014. Our results ascertain that Altman's model is capable of predicting bankruptcy in companies from a wide range of sectors, exhibiting an accuracy of up to 74.7% one year prior to bankruptcy, 70.2% two years prior to bankruptcy, and 66.3% three years prior to bankruptcy. The developed model, however, exhibits higher accuracy than Altman's in predicting the bankruptcy of companies from the same range of sectors, at 84.1% one year prior to bankruptcy, 79.8% two years prior to bankruptcy, and 73.5% for three years prior to bankruptcy. In a bid to discover whether the competency of the discriminant models presented can be improved upon when implemented on a homogeneous sample (a sample consisting of bankrupt and non-bankrupt companies from the same sector), the analysis was conducted once again on each sector separately. The results affirm that when using Altman's model on segregated samples, its accuracy in predicting bankruptcy does not significantly improve. In contrast, devising a discriminant model for each sector significantly improves the model's predictive competency.