Corporate Financial Distress and Bankruptcy Prediction in the North American Construction Industry

This paper seeks to explore the application of Altman’s bankruptcy prediction model in the construction industry by measuring its percentage accuracy on a dataset consisting of 108 bankrupt and non-bankrupt firms selected across the timeline of 1985-2013. The main goal of this paper is to explore the predictive power of an expanded variable set tailored to the construction industry compared to the original Altman model. Specifically, this measuring process is done using machine learning algorithm based on scikit-learn library that transforms a financial statement dataset of a company into clean vectorized feature matrix. The algorithm provides various classifiers to crossvalidate the training set. Naive Bayes, Logit Regression, Support Vector Machine, K N_Neighbors, Tree, and Grid Search classifiers are used in this paper. The result shows no single dominant classifier that manages to predict bankruptcy more accurately than others, but non-linear classifiers tend to outperform their linear counterparts. Additionally, there is no clear preference in terms of the original 5-variable set versus the newly expanded construction specific 14-variable set, meaning that the Altman model stands both valid and effective in the context of bankruptcy prediction in the North American Construction Industry. JEL Classification: C5; C38; G33; G34