Most consumer lenders allow for variances from credit score cutoffs in accepting or declining applications.
The lenders that do this acknowledge that credit bureau reports and credit scoring models, no matter how sophisticated, do not explain or predict all of the variations in credit performance and loan profitability.
Even lenders with extensive experience in developing and managing credit scoring systems typically allow or encourage underwriters to inject their judgment into the credit process.
Experienced lenders know that factors other than credit scores matter. Loan-to-value ratios, down payments, payment-to-income or debt-to-income ratios, and other factors are incorporated into the underwriting process through formal and informal guidelines.
Reduced LTV ratios or higher down payments are sometimes used to compensate for marginal or failing credit scores in underwriting individual loans. Increased debt-to-income ratios may be allowed if compensated by lower LTVs.
Increased interest rates or fees charged at origination are sometimes built into pricing schemes to compensate for higher LTVs, higher payment-to-income ratios, or marginal credit scores.
These tradeoffs are common in home equity and nonprime lending, where industry practices and competition dictate flexibility in underwriting. For most consumer lenders, whether prime or nonprime, the objective of these compensating factors is usually the same: To ensure a profit at a targeted default rate.
However, what is frequently missing is a reliable and explicit link between these tradeoffs and their effect on risk and profit. Many consumer lenders have relied on tradition, industry practices, or intuition to evaluate these tradeoffs.
Our analysis of nonprime mortgages illustrates that the relationship between these compensating factors can be statistically analyzed and modeled, much like credit scores themselves, to achieve a targeted default rate. For lending decisions where loans will be pooled into a single pricing class, lenders need a method to determine which loan applicants should be part of particular pricing pools.
Traditionally a few rules of thumb have been used to make these pool decisions. If the applicant meets these rules, the loan is considered worthy of the pool and the application is accepted; otherwise it is rejected. The disadvantage of the rule-of-thumb approach is the implicit assumption that tradeoffs generate binary yes/no decisions.
A few examples illustrate this: For some lenders, a loan-to-value ratio of 80.0% is a desirable risk exposure, but 80.1% is undesirable. A total debt payment-to-income ratio of 38.0% is desirable, but 38.1% is not. A credit score of 650 is desirable, but 649 is not. Even compensating factors applied in conjunction with credit scores often have implicit hard cutoffs.
Moreover, the appropriate degree of compensation is not typically measured or managed, but is applied intuitively. It is possible, however, to jointly estimate the effects of a number of factors, and thus map out the tradeoffs to provide a superior decision tool.
From analysis of a large number of nonprime consumer loans, University Financial Associates developed a model that predicts variations in loan behavior resulting solely from such compensating factors.
Our analysis held all credit and economic conditions constant, including the expected conditional default rate. In effect, we simulated a pool of loans identical in all respects, except for the compensating factors, performing under the same economic conditions.
From these projections we isolated combinations of equal conditional default rates and examined the tradeoffs among the compensating factors.
We have modeled numerous combinations of compensating factors at constant conditional default rates. We have also developed risk-based pricing schedules where interest rates and origination points are modeled against credit scores, LTV ratios, and debt ratios simultaneously.
Our research has generated four common threads that are not always accurately reflected in the widely used underwriting rules of thumb:
Tradeoffs are not binary yes/no decisions with distinct cutoff points, but rather continuous functions with a broad range of acceptable levels yielding the same expected conditional default rates.
Tradeoffs are not strictly linear, but may have a different effect at different levels of loan-to-value, debt ratio, etc.
Relationships between compensating factors may even reverse at different points along the curve.
Relationships between compensating factors may change at different expected default rates.
Mr Thomson is an associate professor of finance and real estate at the University of Texas at San Antonio, and a senior associate with University Financial Associates LLC, an Ann Arbor, Mich., company that develops and maintains econometric modeling systems for managing and pricing lending risks. 
