(A) study on the bankruptcy prediction model using survival analysis

Abstract

In this thesis, I investigate the applicability of survival analysis technique to quantitative prediction models for corporate bankruptcy. Most of the traditional bankruptcy models suffer from the low performance of ex ante (out-of-sample) forecasting. I suggest that the consequence of low performance results mainly from ignorance a temporal concept from the model building stage. Survival analysis is a proper alternative method to reflect the panel property of financial statement and the common effect of macroeconomic variables. I provide three empirical studies in this thesis. Firstly, I compare a survival technique with a conditional probability model to verify the applicability of survival analysis which is eligible to dichotomous classification. Secondly, I assessed the violation of proportionality assumption, which is one of the core assumptions of Cox’s (1972) proportional hazard model, in the estimated PH-model by the listed companies in the Korea Stock Exchange (KSE) from 1991 to 2000. It can be shown that a non-proportional hazard model including a change point model is a proper alternative when the proportionality assumption is violated by the change of macroeconomic environment, such as the financial crisis in 1997. Finally, extending the work of Shumway (2001), I present a duration model with time varying covariates and a baseline hazard function incorporating macroeconomic dependencies. I investigate the performances of out-of-sample forecasting using the suggested models and demonstrate the improvements, which are produced by allowing temporal and macroeconomic dependencies. I conclude that the survival model is required to be modified and complemented to fully reflect the natures of financial statement because it is early stage in the finance or accounting literature, nevertheless it is an appropriate alternative to solve the limitation of traditional dichotomous classification method for bankruptcy prediction.