Bankruptcy prediction : static logit and discrete hazard models incorporating macoreconomic dependencies and industry effects

Abstract

In this thesis, we present firm default prediction models based on firm financial statements and macroeconomic variables. We seek to develop reliable models to forecast out-of-sample default probability, and we are particularly interested in exploring the impact of incorporating macroeconomic variables and industry effects. To the best of our knowledge, this is the first study to account for both macroeconomic dependencies and industry effects in one analysis. Additionally, we investigate the impact of the 2008 financial crisis on bankruptcies. We develop five models, one static logit model and four hazard models, and compare the out-of-sample predictive performance of these models. To explore the impact of industry effects and the financial crisis, our study includes 562 U.S. public companies across all sectors (except financial) that filed for bankruptcy between 2003 and 2013. These were matched to a control group of non-bankrupt firms. We find that the cash flow, profitability, leverage, liquidity, solvency, and firm size are all significant determinants of bankruptcy. The ratio of cash flow from operations to total liabilities, and total debt to total assets, are the most significant variables in the static logit model. In addition to these ratios, cash to total assets and net income to total assets are also among the most important covariates in the hazard models. Next, we find that the forecasting results are improved by incorporating macroeconomic variables. Finally, we find that the hazard model with macroeconomic variables and industry effects has the best outof-sample accuracy.