mortgagetape^™

What it does

Lender-level Adverse Impact Ratio (AIR) audit against HMDA public data for any HMDA-reporting mortgage lender. Screens approval-rate parity by race, ethnicity, and sex against the industry-standard 4/5ths rule (AIR ≥ 0.80).

Data: HMDA LAR, activity years 2018-2024. Cohort = decisioned applications (action_taken IN ‘1’, ‘2’, ‘3’). Approved = originated + approved-not-accepted (1, 2). Reference groups: White / Not Hispanic or Latino / Male — the standard fair-lending convention. Groups with fewer than 30 applications are excluded from AIR (noisy).

Why it matters. Every regulated mortgage lender is expected to monitor disparate impact quarterly for ECOA/Reg B compliance. This tool gives a signed-in user a same-day answer to "does this lender's public HMDA book pass the 4/5ths rule this year, and if not, which protected group falls short?" — using only the lender name (fuzzy lookup) or LEI. Returns per-group approval rate, AIR vs the reference group, and pass/fail flags per protected class.

What it is NOT. A compliance-grade fair-lending analysis requires credit-score, DTI, LTV, and other underwriting-relevant controls that HMDA public data does not carry (FICO is redacted). This tool produces the unconditional AIR that regulators use as a first-pass screen — below-0.80 does not itself establish discrimination, it triggers a requirement to justify the practice on business-necessity grounds. Consult counsel for compliance-grade audits with the private-book variables added back in.

Chat invocation. Ask any of these directly in the chat: "AIR for Rocket Mortgage", "disparate impact audit for Guild", "does UWM pass 4/5ths in 2024". The intent classifier routes to this endpoint; you'll get a color-coded per-class breakdown with all violations flagged inline.

API surface. POST /api/score_di_air with {lender_name} or {lei} plus optional {activity_year}. Full request/response contract at /api/score_di_air/schema (JSON) and metadata at /api/score_di_air/info. Same auth + rate limits as the other scoring endpoints (signed-in only).

› Try it on the home page (Loan-level model scoring → Fair-Lending AIR Audit)

API connector

Programmatic access. Calibrated probability + risk band + operating recommendation in the response.

POST /api/score_di_air
Content-Type: application/json

{
  "lender_name": "Rocket Mortgage",     // fuzzy string match, OR:
  "lei":         "549300FGXN1K3HLB1R50", // 20-char LEI (preferred; unambiguous)
  "activity_year": 2024                  // optional, HMDA 2018-2024; default = 2024
}

Supply lei OR lender_name. If the fuzzy name matches multiple lenders, the response returns an ambiguous_lender: true payload with a candidates list — re-submit with a specific LEI.

Schema reference (request / response shape): GET /api/score_di_air/schema

Model metadata (training cohort, AUC, calibration): GET /api/score_di_air/info