Statistics & Data Science MS Candidate
When
11 a.m. – Noon, Aug. 14, 2026
Where
Title: Extreme Pricing and Institutional Fairness: A Tail Risk Analysis of Credit Union Mortgage Lending, 2022-2024.
Abstract:
Credit unions are not-for-profit, member-owned financial institutions that operate under a dual mandate: maintaining financial stability while expanding access to credit for underserved populations. A subset of credit unions designated as Low Income Credit Unions (LICUs) by the National Credit Union Administration (NCUA) serve communities where the majority of members reside in low-income areas, receiving regulatory benefits in exchange for this mission commitment. This dual mandate means LICUs serve borrowers with different risk profiles than non-designated institutions, making direct pricing comparisons between the two groups methodologically problematic without controlling for borrower composition.
This thesis examines mortgage pricing behavior across LICU and non-LICU credit unions using Home Mortgage Disclosure Act (HMDA) data from 2022 to 2024, asking whether observed pricing disparities reflect differential institutional behavior or differences in the borrower populations each institution type serves. While LICU institutions consistently report higher average rate spreads than non-LICU institutions, extreme value theory identifies that meaningful differences are concentrated in the tail of the distribution rather than in typical lending outcomes, with an empirical tail threshold at the 83rd percentile (0.591%) that separates standard from extreme pricing outcomes. Across linear regression, logistic regression, and random forest models, LTV ratio, loan size, income, and property value are the primary drivers of extreme rate spread classification and account for approximately 65% of the random forest's feature importance. While, institutional designation ranks among the least predictive features in all three model families. Propensity score matching produces 10,942 matched pairs with near-complete covariate balance, reducing the demographic parity gap by 95% from 0.0621 to 0.0033 and the false positive rate gap by 73% from 0.0702 to 0.0193, with the four-fifths ratio improving from 0.8146 to 0.9893 and approaching regulatory parity. One asymmetry remains: the true positive rate gap widens to -0.0708 on the matched subset and persists across all three evaluation years. Diagnostic analysis establishes that this gap reflects the model's reliance on LTV ratio as its primary risk signal, which is weaker for a segment of extreme LICU loans characterized by larger loan sizes, lower leverage, and geographic dispersion where extreme rate spreads occur through pricing dynamics not captured by the available features. These findings indicate that the aggregated pricing gap between LICU and non-LICU credit unions is driven by borrower composition rather than differential institutional pricing behavior, with direct implications for how regulators evaluate fair lending compliance across institution types whose mission-driven character produces structurally different borrower populations.