Standard approaches rely on data that isn't there—missing credit scores, unobserved risk factors, incomplete borrower profiles. Our peer-reviewed methodology solves the omitted variable problem that undermines traditional fair lending regression.
Request a ConsultationTraditional fair lending regression compares loan pricing or approval rates between protected and non-protected groups within a single institution, controlling for observable risk factors. For internal reviews with full proprietary data, this can work. But for peer comparisons using reported HMDA data—or when regulators analyze data to flag discriminatory practices—there is a fundamental flaw:
Key underwriting variables—most notably credit scores—are unavailable in HMDA data. When these risk factors are omitted, standard tests conflate discrimination with legitimate risk-based pricing.
We developed a peer-reviewed methodology that addresses unmeasured confounding through cross-market comparisons. Rather than asking the hard-to-answer question "Does this institution discriminate?", we ask:
"How does this institution compare to appropriately matched peers?"
By comparing your institution's racial/ethnic pricing gaps to a matched benchmark of similar borrower profiles, we difference out the bias from unmeasured confounders that affects all institutions equally—isolating the excess disparity specific to your institution.
| Traditional Approach | Our Methodology |
|---|---|
| Assumes all risk factors are observed | Accounts for unmeasured confounders |
| Difficult to defend in examinations | Peer-reviewed, academically rigorous |
| Binary "discriminate or not" conclusion | Quantifies excess risk relative to peers |
| No sensitivity analysis | E-value robustness measures |
| Product | Data Availability | Our Approach |
|---|---|---|
| Mortgage (HMDA) | Race/ethnicity reported | External benchmarking |
| Home Equity | Race often missing | BISG proxy + benchmarking |
| Auto (Indirect) | Race not reported | BISG proxy + dealer analysis |
| Credit Card | Race not reported | BISG proxy + benchmarking |
| Small Business | Limited demographics | Census-based proxy |
| Service | Scope | Timeline |
|---|---|---|
| Fair Lending Risk Assessment | Single product, full analysis | 4–6 weeks |
| Comprehensive Review | All consumer products | 8–12 weeks |
| Exam Preparation | Targeted analysis + documentation | 2–4 weeks |
| Ongoing Monitoring | Quarterly analysis + dashboards | Annual engagement |
Sophisticated analytics. Defensible methodology. Actionable results.
Schedule a confidential consultation to discuss how our methodology can protect your institution.