Credit rating in an agricultural credit institution: Methods and issues

This study examines models used in predicting financial performance expressed as probability of default, by researchers and credit institutions. The Adaptive Neuro-fuzzy Inference Systems, an artificial intelligence method which is known to handle linguistic variables and more commonly used for forecasting in the engineering field is compared with the logistic regression and artificial neural networks. Predictive performances of the models are examined based on their ability to accurately identify credit default. A multi-objective fuzzy simplex-genetic algorithm is also used to optimize the fuzzy rules obtained from the Adaptive fuzzy inference system.; Also, several studies have applied different methods, definitions of credit default and data types. This study examines these varieties of approaches and methods based on how and the extent to which they affect estimation results.; Using available data from the Farm Credit System over 1995-2002, empirical findings show that default probability estimates differed with default definition (data composition). However, statistical evidence does not support the hypothesis that differences exist between the prediction performance results from the three models examined. While the logistic regression showed the highest accuracy at identifying defaults in the portfolio in- and out-of-sample, the adaptive neuro-fuzzy inference system has the best ability to identify non-default cases. Also, results from the two different datasets used based on sixty and ninety default days default definitions appear to have no statistically significant effects on parameter estimates, though differences in magnitude of parameters and model accuracies occur.; Further analysis of the credit portfolio based on optimized fuzzy rules shows that low repayment capacity, low owners equity and low working capital percentages are the best indicator of default status. Low working capital percentage is also shown to be the most consistent predictor of default/non-default. The worst fuzzy rule indicator of default is low repayment capacity, high owners' equity and medium working capital.