Abstract:

Purpose - In this study, we make an empirical research and a comparison study on econometric models used with logistic link functions. We compare the predictive powers of models in credit granting process. Methodology - We collected data belonging to 87 medium sized companies. 21 of these companies are defaulted. The data set includes 15 continuous financial ratios for estimation of the models. We implement three models which are Logistic Regression, Generalized Partially Linear Models(GPLM) and Generalized Additive Models(GAM). For each model the best fitted model is selected according to AIC criteria. Findings-   GPLM have pointed out that the equity turnover ratio has a significant nonparametric effect. On the other hand GAM pointed out that (total liability)/(total assets) and Increase in Sales have significant nonparametric effects. Comparison of the models have implemented according to their accuracy ratios, Type I and Type II errors. Results show that generalized additive model with logistic link outperforms both Logistic Regression and generalized partially linear model in terms of three performance measures. Conclusion- After 1980s as a result of the financial crises the default events become a main issue of the credit agencies. For this reason, a credit agency’ objective is to determine whether a credit application should be granted or refused. Here, the problem is to learn default some time before the default event occurs. The empirical studies in this area have indicated that commonly used classification methods are good to detect signals of defaults. Especially the models which allow logistic link function are good choices for modeling default risk. In this study we mainly focused on the generalized linear models and its semi- and non-parametric extensions with logistic link function. We compare their performances in a credit granting procedure. We use a real data belonging to Turkish SMEs. Our results show that the GAM outperforms the other two models and it will be a good choice for credit granting procedure.

Keywords: