Firm Failure Timeline Prediction: Math Programming Approaches

Young U. Ryu

DOI: https://doi.org/10.1109/HICSS.2016.149

2016-01-05

Abstract:Previous studies on firm failure prediction proposed models to predict only short-and fixed-term (normally one year) bankruptcies. Our study applies two math programming methods to predict firm failure timeline (i.e., longer-and variable-term bankruptcy): isotonic prediction and modified linear regression (a form of recurrence surface approximation). With these two methods, we also evaluate financial ratios that were previously used for fixed-term bankruptcy prediction. The experiments with actual firm data indicate that (1) both approaches generate good prediction outcomes but isotonic prediction performs better than modified linear regression and (2) the use of all twenty-three financial ratios for isotonic prediction result in the best prediction outcome after a feature selection preprocess is applied.