Forum: The government needs to know the score
How credit-scoring policies affect the economy.
By Douglas A. Samuelson
The article in the May/June issue of Analytics on credit scoring by FICO’s Andrew Jennings and Carolyn Wang  illuminated a key reason for the lingering economic slump. Most of us know that business, and in particular business investment, is not rebounding as hoped. We also know that borrowing is below expectations, even at record low interest rates. Now we can see a major element of the problem.
Much of the boom of the previous decade was fueled by small-business innovation. From conversations with many people in the high-tech community, I believe that much of the credit card and home equity borrowing of that era was small business finance, not profligate consumer spending: It was easier to tap those sources than to explain a high-tech business plan to a banker. (Of course, some of the business financing was spent in ill-advised ways, but that doesn’t detract from the reasoning here.)
How credit scoring works
Now consider how credit scoring works. Essentially it is a mathematical method of identifying the handful of applicant characteristics that best distinguish between high and low risks, in the creditor’s recent experience. These quantitative models are generally better than human judgment, but assumptions, including those we don’t immediately recognize as assumptions, matter. One of these assumptions is that behavioral patterns seen in recent experience will continue without significant change.
During the boom, those who ran up balances close to their credit limits were, indeed, poorer risks than those who did not, as the difference mostly reflected money management habits. As the downturn began, however, an increasing number of fiscally conservative business managers were also using more credit, trying to keep paying their suppliers and creditors as their own revenues declined. As scoring systems indicated tightening credit on these borrowers, therefore, the effect was an accelerating, self-feeding squeeze on small businesses.
Thus Jennings and Wang’s recommendation for more aggressive credit tightening is, unfortunately, a fine example of getting the wrong answer by looking at too narrow a view of the problem. Indeed, a few credit-granting institutions that followed their advice could have cut their short-term losses substantially as the country slid into recession, by tightening credit to people who had performed well up until then. When the preponderance of credit granting institutions did this simultaneously later in the downturn, however, the result was a cascade of defaults, and now a slower recovery than economists predicted. In fact, at this point, a modest increase in available credit would be more likely to lower the losses economy-wide. That is what the mortgage assistance program was supposed to do, but ‚Äì again ‚Äì few borrowers have been able to meet the tightening qualification standards for the loans.
Ironically, credit-scoring companies such as Jennings and Wang’s employer, FICO, could lead the way out of this crisis. When I worked for Fair, Isaac and Company, as FICO was then called, in the early 1970s, the company developed scoring systems for several large retailers. Typically what best separated good from bad risks were characteristics such as length of time at current address, length of time in current job and length of time the best good credit account has been open. This meant that the system predicted risk well but would deny credit to virtually everyone under 30. Understandably, this strategy was unacceptable to the retailers, as it would likely have severely curtailed their growth, at the very least. The solution was to segment the applicant pool and develop a different scoring system for applicants under 30. This system used different characteristics, such as educational level, occupation and whether the applicant owns or rents his or her residence.
At that time Fair, Isaac also found itself, unwittingly, to be something of a champion for women and minorities, as the scoring systems often identified as good risks some members of these traditionally underrepresented groups —underrepresented because they were undervalued by the judgment of loan officers as circumstances changed and more economic value emerged in these subpopulations.
As the retailers did in the 1970s, credit grantors would do well to segment their applicant populations and re-estimate risks accordingly. In particular, risk varies with the purpose for which the money is to be used. Risk and recovery for individuals reflect good money management. However, as I learned in the late 1990s analyzing Small Business Administration disaster loans, risk and recovery for businesses reflect mostly the robustness of sources of revenue. For example, small businesses near Homestead Air Force Base mostly did badly after Hurricane Andrew in 1992, regardless of their previous soundness of management, because the Air Force base never reopened. Small businesses affected by the Northridge earthquake (Southern California) in 1994 tended to do well, regardless of soundness of management practices. These businesses were mostly specialty suppliers to the high-tech and aerospace industries. As soon as their phones were back in service, they started ringing with orders.
Characteristics of New Models
Here are some suggested characteristics these new models might include:
- Is the credit to be used solely for personal purposes or for business capital?
- If the credit is for personal use, in what industry is the applicant employed, and what are that employer’s and industry’s prospects?
- If the credit is for business financing, what industry and sector is the business in, and what are its products or services? What are its current sales, and how stable are its customers? How long has the company been in business, what is its management structure and how much experience does its management team have?
Opening up more credit to the more promising businesses, especially the smaller ones in newer areas of industry — hence unlikely to have other ready sources of capital — would most likely prove to be both beneficial to the economy and profitable to the lenders.
The federal government would do well to consider incentives, possibly guarantees, in addition. It has done so, with considerable success, in the housing market (especially FHA and VA loans). The automobile insurers’ assigned risk pool, in which less risky people cross-subsidize the riskiest, is another good example of an approach that improves the overall situation. (Less risky people benefit from lowering their risk of incurring damage inflicted by an uninsured and insolvent motorist, while the economy benefits by permitting high-risk motorists to continue to drive to work.) Some provisions of the 2010 health care law also aim to provide similar arrangements regarding access to health care by high-risk patients, again with the expectation of reducing overall risk. One option here is an expanded SBA loan program, with a fast-track option for selected industries, calling on the assessment expertise of previously successful managers and financiers in those industries. Several states have used this approach in centers to promote innovative technology.
Even without federal guarantees, however, this proposed approach to increasing available credit should appeal to wise credit scorers and credit grantors. As credit scoring in the 1970s opened access to credit for women and minorities, and for younger applicants to some retailers, so it could lead the way out of the slump now — and generate substantial rewards for those willing and able to take the right risks.
Doug Samuelson (email@example.com) is president and chief scientist of InfoLogix, Inc., a small R&D and consulting firm in Annandale, Va., and a frequent contributor to Analytics and OR/MS Today.
1. Andrew Jennings and Carolyn Wang, “The Secret to Better Risk Management: Economically Calibrated Models,” Analytics, May-June 2011, pp. 11-15.