LOGISTIC REGRESSION AND NEURAL NETWORK APPLICATION IN CREDIT SCORING: THE CASE OF LATVIA.

Interest on loans is the primary source of business income for financial institutions. Borrower solvency assessment models can increase financial institutions’ profitability and reduce high interest rates in Latvia. Borrower solvency could be assessed with different methods. The relative performance of different methods is likely to be country-specific (or case-specific), so developing a credit scoring model applicable to Latvian nonbank borrowers is required. The most popular methods to develop the models are logistic regression and artificial neural networks. In order to ensure even more stable model results, a combination of the two methods (ensemble method) has also been developed. The results show that an ensemble method is a powerful tool to significantly improve the model’s classification indicator and performance in all selected performance evaluation indicators. By introducing the best model in this work, a nonbank lending company would ensure that the number of borrowers with poor solvency would be reduced by about 35% than without implementing a credit scoring model at all. [ABSTRACT FROM AUTHOR]