Business — Banking — Management — Marketing & Sales

Scoring is widely used for consumer loans. Scoring allows us to automate the credit process because there is no real need to examine in detail the profile of individuals. There are plenty of reliable portfolio loss statistics, given the large number of individuals. The intensity of services provided to customers is a major source of high revenues, both fees and spreads. The relevant criteria combine both the potential revenues, the value added for the bank and the risks. Many fit such conditions. Income per dependent in the family, renting or owning home, marital status, occupation are all criteria for assessing both the credit standing and the potential revenues. Information on other credit cards and loans by other institutions provides information on the exposure to other competitors and the level of debt. Years spent at the same address and years spent in the same job provide information on the mobility to be expected and the length of the financial services to be expected.

The Economics of Scoring Systems

The usage of scoring has economic implications, which are the relative costs of rejecting a good credit and of accepting a bad credit.

Type I and Type II Errors

Scoring is a statistical technique that can fail to make the right predictions, as all do. Beyond some value, the likelihood of failing is negligible, below another lower value, the likelihood is close to 1. In between, we do not know, but we might infer some relationship between default probability and the value of the Z-score. Used in this way, we have both a default predictive tool and a rating device. Types of errors for cut-off scores are:

• Type I error—accept bad credits.

• Type II error—reject good credits.

Since the costs of the different errors are widely different, the economic side of errors has an influence on the cut-off value of the score.

Scoring Criteria and Revenues

Scoring does not apply only to credit risk measurement. It extends to attributes representative of the potential richness of a customer to the bank. Richness means intensity of services, such as the number of accounts and services with a customer, future transactions expected with the customer, such as new loans and credit card loans. Richness, or service intensity and profitability, are criteria that prove correlated to the personal profile of individuals.

In the end, the modelling provides information on both credit standing and potential richness. An individual can pass the credit standing test and not the richness test. Others might be risky customers, but seem to have plenty of business potential. This extension of the same type of technique can also serve to assess the risk-return profile of individuals. Embedding both revenue and risk criteria in scoring raises the wider issue of the economic implications for individual decisions and for the entire loan portfolio.

For individual decisions, type I errors result in a full loss of principal and interest should risk materialize. Type II error costs are opportunity costs of not lending to a wealthy borrower. The opportunity cost is the lost income. Hence, the cost of a type II error is much lower than that of a type I error. Based on these simple economics, we can tolerate more type II errors than type I errors, because the unit loss is lower. Since the cut-off values of scoring models drive the probabilities of type I and II errors, choosing them depends on both the quality of predictions and the cost of errors.

Rejecting too many good credits results in adverse selection, or taking too many bad credits. Extended at the scale of the portfolio of the banks, this implies that some banks will concentrate on bad credits and others on good credits. If the bank uses risk-based pricing according to credit quality, this is acceptable, since not all banks need to run for the same market segments. Nevertheless, because of competition between banks, pricing might not really differentiate the good and bad accounts. The biased score becomes a problem since pricing does not compensate any more the differential cost of errors. Therefore, the cutoff points should not depend only on the individual loss trade-off between type II errors and type I errors and their frequencies. It should also depend on the portfolio effect once scoring is in force and actually structures the loan portfolio. The portfolio view changes the picture because it relates the implementation of scoring to the portfolio structure and the pricing policy. In addition, it should also extend to any adverse effects of attracting too many bad credits, which influences the banks reputation.