You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Replication code for: Persistent Racial Disparities in U.S. Mortgage Approval (42M HMDA applications, 2020-2024). 14.95 pp raw Black-White gap; 68% unexplained by observables. Under review, Journal of Housing Economics.
Statistical lending-disparity screening for public HMDA data — a transparent, open-source screening tool for community advocates and journalists. Not an examination methodology
Explainable and fair mortgage-denial models on 2025 Arizona HMDA data: Optuna-tuned LightGBM/XGBoost/CatBoost, SHAP vs LIME checked against lenders' stated denial reasons, and a fairlearn audit by race, ethnicity, sex and age with mitigation trade-offs.
Interactive credit-decisioning fairness tool: translates DI/EOD into per-applicant dollar units. Surfaces the Fairness Paradox on 95K HMDA mortgage applications.
Orchestrated HMDA mortgage data pipeline using Apache Airflow — automated ingestion, cleaning, quality validation, and weekly model performance monitoring on 279K NC loan applications.
An interactive app that tells the story of the 2008 mortgage crisis and its decade-long aftermath through five borrowers, built on 64 million HMDA mortgage records (2007–2017) from the CFPB.
Can anyone still afford a home: a data story in eight chapters over 393 US metros, with a pre registered plan, a simulator with known truth, a dbt warehouse and a claim gate on every number
PySpark + Delta Lake lakehouse over 222.7M US mortgage records (HMDA 2018-2025): affordability, the 2022 rate hike, denials, fairness, and how much the government's own data changes between versions. Every number checked against official totals.