Bankruptcy Prediction Using Survival Analysis Technique
Currently, there is an extensive set of bankruptcy prediction models, but almost all of them are classification based, i.e., they allow to estimate the posterior probability that a particular firm will fail, given its financial characteristics. The expected time to failure is not considered explicitly. On the other hand, there is a survival analysis that deals with the time of the occurrence of the event of interest (while this event may not occur during observation). However, despite its popularity in the medical and technical sciences, survival analysis is relatively rarely used in predicting financial failure. Even when it is applied, most authors use the simplest form of a model. The goal of our work is to evaluate the applicability of survival analysis to bankruptcy prediction. We compare a few state-of-art statistical and machine learning models using a real dataset. Our findings confirm that survival analysis allows (1) to extract from given data valuable information regarding the dynamics of risks and (2) to estimate the impact of features.