Modeling internal credit ratings of Turkish companies listed on the ISE

An internal credit rating has wide implications for firms and the economy at large. Forming and testing the efficiency of an internal credit rating model is an essential policy tool for all economic agents due to its distributional impact. This thesis aims at developing an internal credit rating model based on two well-known techniques; ordered logit and artificial neural network.The constructed internal credit rating model was tested on a sample data that consists of financial ratios from 40 small-sized firms that are listed on the Istanbul Stock Exchange. The data is quarterly and covers 47 observations running from the first quarter of 1998 to the third quarter of 2009. The internal credit rating model contains the transformed form of the current ratio, which was used as a dependent variable and 4 financial ratios as the independent variables. The ordered logit and artificial neural network models? parameters were estimated with 41 observations. Then, their forecasting performances were measured with the remaining 6 observations.The results show that the artificial neural network model is superior to the ordered logit. As a secondary result of this thesis, manufacturing firms? rating series were at their lowest during the years of financial crisis, while the highest increase of ratings occurred in the firms of the cement sector. This thesis makes several policy suggestions in using internal credit rating models for banks and rating agencies.Keywords: internal credit rating, artificial neural network, ordered logit.