Different statistical techniques are employed to represent the decision process of commercial credit officers. These techniques are illustrated by an analysis of real data from a major Canadian bank.
The process of modelling the variables important in the extension of credit is referred to as "credit scoring." This modelling process, often using statistical methodology such as discriminant analysis, has been carried out by banks, financial institutions, and corporations. A decision making model which will aid in the prediction of the outcome of small business loan applications for one division of a major Canadian bank has been developed.
Based on statistical analyses of historical data, certain financial variables are determined to be important in evaluating the customer's financial stability and strength. This information is summarized and a point system is developed where the different variables have different weights. Then, information on these variables is obtained for new customers. An overall score is produced by adding these weighted scores from answers given on an application for credit. If this overall score is above a predetermined cut-off point, the applicant receives the credit. If not, credit is denied.
Applying statistical methods to the credit decision is not a new idea. Credit scoring for consumer credit has developed into a multi-million dollar industry, and is well documented. Moreover, there has been much research regarding the creation of predictive models for corporate bankruptcy. However, little work has been done attempting to model the commercial credit officer's decision process based on actual accept/reject decisions. Due to a lack of large homogeneous groups and the absence of a general consensus on which variables are predictive, there has been a long history of reluctance to actually implement scoring models in this area. In fact, discussions with commercial credit officers reveal that many believe that it is not possible to model commercial lending in the same way that it is done for consumer credit. A number of statistical methods can be used to model the commercial credit officer's decision process.
At the bank used in this study, the need for a credit scoring model was obvious because there were only a few experienced lenders in the bank. Consequently, the bank's senior management wanted to spread the limited expertise around in the most efficient manner. Therefore, the experts' behavior was analyzed and a model produced--one which could then be incorporated by less experienced credit officers in the future. Until the introduction of credit scoring, there was no formalized training or "expertise sharing."
Variables Used in Credit Evaluation
The credit evaluation process currently in practice at one regional division of the bank was investigated. The files pertaining to small-business loans for the period of one calendar year were examined. Due to confidentiality, the investigation was limited to new customers. The outcome of the decision process (credit approved/credit declined) of the credit (loan) officer was modelled. Information was taken from 339 loan applications previously reviewed by the division managers; 252 loan applications were approved and 87 declined. The data consisted of the following variables:
* company name (NAME)
* company number (ACC)
* outcome of application-approved or declined (OUT)
* amount of the loan application (AMT) total credit position of the company (CRE)
* new vs. existing company (NEW)
* branch of bank (where the application was filed-BRN)
* type of loan (operating/term/small business loan-TY)
* statement type (past records/pro-forma/none-ST)
* sales (SA)
* gross profit (GP)
* operating profit (OP)
* bad debts (BD)
* depreciation (DEP)
* income surplus (loss) transferred to equity (INCO)
* total net worth (TNW)
* net working capital (WC)
* private/other company (TF)
* number of owners of firm (NO) and
* owners guaranteed security in the company (SEC). …