engleski

Multivariate statistical approach to credit risk assessment and valuation for loans

Bad approach to credit risk exposure in the United States and some other countries has led to financial crisis and economic depression all over the world, yet without signs of recovery. Very loose conditions for banking loans produced exaggerated financial activity and therefore, as a consequence, impossibility to get loans paid back. In Croatia, banks used much stronger criteria when allowing loans with much better performances then in the U.S., which are the main reason that we do not have financial crisis as many other countries. Still, in this paper we want to get statistical multivariate approach to get it better. Authors analyzed the sample of retail portfolio of Croatian banks classified by the loan structure and borrower selected attributes, and wanted to challenge this classification by statistical methods. Multivariate statistical approach is used to get the most significant variables which contribute to classification of the loans as risky or not (default or not). There were several groups of applicants due to loan repayments, but we have made only two groups: those who repay their loans regularly and those who get some problems (delayed or not paying at all). We used following variables as predictors: economic development of the region where subject live, age of the subjects, percentage of the loans already repaid, interest rate and the currency in which the loans are tied (Croatian kuna, Euro, US dollar and Swiss franc. We have found that probability of bad loans is higher if the subject comes from less developed region, if hi (she) is older, if interest rate is higher and if the loan is tied to Croatian kuna. Logistic regression and discriminant analysis produced similar results. Classification results were satisfactory and showed some difference regarding bank risk assessment.

credit risk ; logistic regression ; discriminant analysis

nije evidentirano