Periods of economic crisis arouse interest in exploring the causes of firms closure, for preventive and predictive purposes. Failure prediction models are useful tools for bankers to measure the risk of lending and minimise losses, for firms wishing to evaluate their market position, and, also for investors, asset managers and rating agencies. Quantitative methods to assess the performance of firms and to predict the bankruptcyevent based on balance sheet indicators are widely used in the credit risk context. Logistic regression and survival analysis techniques based on hazard models are among the methods often employed. AlargedatasetoncapitalcompaniesinItalyfrom2008to2013,including Business Registerdata supplying a complete picture of their legal situation, was used to develop survival models. Training (n = 27286) and holdout (n = 7124) samples were constructed for developing and testing models, respectively. Fixed and time-varying covariates were taken into account and macro-economic variables were included besides the firms individual financial indicators. Furthermore, we considered one- and two-year lagged values of each time-varying covariate. ROC curves that vary as a function of time and AUC up to a given time were used to compare models and obtain global concordance measures
Bankruptcy prediction by survival models based on current and lagged values of time-varying financial data
Francesca Pierri;
2016
Abstract
Periods of economic crisis arouse interest in exploring the causes of firms closure, for preventive and predictive purposes. Failure prediction models are useful tools for bankers to measure the risk of lending and minimise losses, for firms wishing to evaluate their market position, and, also for investors, asset managers and rating agencies. Quantitative methods to assess the performance of firms and to predict the bankruptcyevent based on balance sheet indicators are widely used in the credit risk context. Logistic regression and survival analysis techniques based on hazard models are among the methods often employed. AlargedatasetoncapitalcompaniesinItalyfrom2008to2013,including Business Registerdata supplying a complete picture of their legal situation, was used to develop survival models. Training (n = 27286) and holdout (n = 7124) samples were constructed for developing and testing models, respectively. Fixed and time-varying covariates were taken into account and macro-economic variables were included besides the firms individual financial indicators. Furthermore, we considered one- and two-year lagged values of each time-varying covariate. ROC curves that vary as a function of time and AUC up to a given time were used to compare models and obtain global concordance measuresI documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
