The next chapter in small business scoring

RMA Journal, The, Feb, 2002 by Marcus Bishop

More, and more recent, data from a greater number of institutions is among the improvements brought to the RMA/Fair, Isaac Small Business Scoring Service[SM]. This article provides insight into the enhancements that result in an easier-to-use and more predictive model.

Small business scoring is gaining strong momentum, according to the 2001 Small Business Risk Management Study, conducted by RMA--The Risk Management Association and BenchMark Consulting International:

"Based on the results reported by [surveyed] respondents in this year's study, credit quality remains relatively unchanged--with 71 % of survey participants indicating that they have been using credit scoring for more than four years to manage their business."

The development of SBSS 5.0 models adds to this momentum by offering the best, most predictive models Fair, Isaac has produced to date. The SBSS 5.0 suite includes the RMA/Fair, Isaac loan and line-of-credit models, the commercial card models, and the leasing models.

Improvements include the following:

* Improved risk assessment across various portfolio types. SBSS 5.0 data represents better geographic distribution, greater small business credit grantor participation, and a larger sample size than in past developments, resulting in more finely segmented and powerful models and better risk control over unique portfolios.

* Greater flexibility. Lenders can make credit decisions without necessarily obtaining financial statements, credit reports, or other time-consuming and hard-to-get information. Financial institutions can more closely align their specific credit policies and marketing strategies with the analytics, making the decision process more cost-efficient.

This article focuses on the loan and line of credit models. In addition to reviewing the development process, this article discusses performance reviews and how lenders can migrate to the new models.

Model Development

The thoroughness and objectivity of any model development process, and the underlying data, are requisite to any reliable predictor of risk. The original RMA/Fair, Isaac model data pool consisted of 17 RMA-member small business credit-granting institutions; 25 institutions have contributed data for the latest development. The additional institutions also yielded a more geographically balanced, nationwide data sample. In fact, the sample closely resembled the Small Business Association's geographic distribution of small businesses in the nation's Western, Northeastern, Southern, and Midwestern regions.

The current participating institutions also represent various asset sizes and relationships with a wide variety of small business types. Small businesses in both the previous and latest models are defined as those with $5 million or less in sales and an exposure to the credit grantor of approximately $250,000.

The SBSS 5.0 sample also generated significantly more data from all contributors, resulting in a better development sample. The sample contained more than 250,000 applications compared to approximately 30,000 in the original development. More important, the sample contains more than 6,000 "bads" compared with approximately 2,500 in the original development, allowing the developers to explore many more population segments and further pinpoint stronger predictors (characteristics) of risk. The resulting models are more robust and more broadly applicable across the industry.

Obviously, any model development benefits from a data sample that reflects recent data. In addition, however, the growth of the industry and the growth of the industry's use of credit scoring--due largely to the development of the first RMA/Fair, Isaac models--has led to a far more diverse sample this time around. Let's look at how much has changed over the years.

Before scoring became available for small business lenders back in the early 1990s, most commercial lenders evaluated and processed small business loans the same way they handled large business loans. At that time, most commercial lenders did not realize that small business lending is more closely related to consumer lending than to commercial lending. The greatest similarity between small business lending and consumer lending is that the creditworthiness of the business is tied to the financial profile of the business's principal(s).

Furthermore, in the original development and even more so in SBSS 5.0, it was found to be imperative to identify the differences in the profiles of small business owners and consumers. For example, when researching possible population segmentation splits and comparing the performance of consumers and small business owners, the Fair, Isaac team found that small business owners have:

* Comparable number of new trades opened.

* More delinquency.

* More inquiries.

* Older, thicker files.

* Higher total revolving balances.

Without the use of models incorporating this type of insight, small business credit grantors could not really take advantage of consumer information. While they would want to rely on consumer information about the business owner, they would have no method of rank-ordering risk solely among business owners. In essence, using a consumer score would be misleading. By contrast, Fair, Isaac research shows that by using the SBSS 5.0 models--which look at the applicant as a small-business owner instead of a consumer-lenders can make better, more profitable decisions.