Methodology & data inventory

HUD's Compare Ratio is a one-number summary of relative delinquency performance, but its real power is that it's the only performance signal HUD itself uses to trigger an enforcement chain with statutory teeth. That makes it a leading indicator for several downstream outcomes that are otherwise hard to see:

• Enforcement risk.CR > 150 puts a lender on the Credit Watch monitoring list. CR > 200 and above the national rate is the legal threshold HUD uses to terminate Direct Endorsement authority in a specific HUD field office (24 CFR 203.3(d)). Termination notices land in the Federal Register; we track them and link them to lender IDs.
• Secondary-market pricing. Counterparties (correspondent investors, warehouse lenders, MSR buyers) price overlays off Compare Ratio bands. A lender drifting from 110 to 175 will see pricing widen and credit overlays added 6–9 months before any HUD action — but only if a counterparty is actually watching the NW data.
• Lender Insurance authority. The LI Compare Ratio uses a stricter 150 threshold than the general 200 line. Losing LI authority means every loan goes back to HUD pre-endorsement review — a 30–60 day cycle-time hit per file.
• Branch & geographic risk. CR is computed at six geographic levels (national, state, HUD field office, MSA, county, ZIP). A lender with a healthy national CR can still have hotspots — we expose all six levels so you can find them.
• Sponsored origination risk. Many small lenders originate FHA through a sponsor / principal- agent arrangement. HUD reports the sponsor's CR separately from the originator's CR (the "principal" report family). Both numbers matter when underwriting counterparty risk.

Our database mirrors HUD's FHA Neighborhood Watch system, plus a handful of complementary sources that NW doesn't expose through its UI. Current inventory:

What we deliberately don't have: anything behind FHA Connection login (loan-level case status, underwriter notes, indemnification rosters). That data exists, but accessing it would require user credentials we're contractually prohibited from using.

• Deal screening before correspondent / TPO onboarding. Pull a prospective seller's national CR plus their three highest-volume HUD field offices. Anything over 130 nationally or 175 in any single office is a conversation, not a green light.
• Branch expansion underwriting. Compare your own state-level CR against the state benchmark before staffing up. A CR of 140 in a state where you have 8 LOs says hire #9 won't fix the problem — file mix or QC will.
• MSR / portfolio buy diligence. Cross the servicer's TRSII tier against their NW delinquency trend. A Tier 1 servicer drifting from 4% SDQ to 6.5% SDQ across two quarters is a price negotiation; a Tier 3 doing the same is a walk-away.
• Counterparty monitoring. Watch your top 10 originator counterparties' sponsor CR monthly. A drift past 150 typically precedes either an LI termination or a sponsor exiting wholesale entirely.
• Regulatory / examination prep. Pull your own historical CR series and identify the HUD office most likely to draw an exam. Have the QC narrative ready before the auditor brings up the number.
• Market entry analysis. For a target metro, rank every active FHA lender by 24-month volume. Identifies who you're competing against, who's slipping, and which DBAs are actually the same parent.

CR = (lender SDQ + claims rate) ÷ (peer area SDQ + claims rate) × 100

• SDQ = loans 90+ days delinquent as of the last servicer report.
• Claims = FHA insurance claims paid in the cohort window.
• Cohort = loans with beginning amortization date in the trailing N months (24 for fhax / d3, 12 for 1yrx / d1).
• Peer area = whichever geography you queried at (national, state, MSA, HUD office, county, ZIP). Each level has its own benchmark.
• Streamline refinances are excludedfrom the published CR — HUD added this exclusion in 2013 so lenders weren't penalized for taking other lenders' streamlines.

• CR > 150 → HUD Credit Watch monitoring (CWT flag set, no automatic action).
• CR > 200 AND above national rate → HUD may terminate Direct Endorsement authority in the affected HUD field office, 60-day notice + informal hearing right (24 CFR 203.3(d)).
• LI CR > 150 → HUD may terminate Lender Insurance authority — every loan goes back to pre-endorsement HUD review.

Termination is geo-scoped — it applies to the field office where the CR breached, not to the lender's overall FHA approval. A terminated lender can still originate via sponsor / principal-agent arrangements or in non-terminated offices, and can apply for reinstatement after six months.

Every metric we surface ties back to one of these:

• HUD FHA Neighborhood Watch (Early Warnings, Analysis, Lender Details)
• NW — Compare Ratio definition
• NW — Streamline-excluded CR
• NW FAQ (cohort, refresh cadence)
• 24 CFR 203.3(d) (Credit Watch / DE termination)
• HUD Lender Performance overview
• FHA Single Family Handbook 4000.1
• Mortgagee Review Board overview (12 USC 1708)
• Federal Register — CWT termination notices

HUD publishes everything as fragmented Excel exports inside a 1990s ColdFusion app. We do three things HUD doesn't:

• Normalize and stitch. Hundreds of thousands of per-state, per-ZIP, per-lender exports unified into one queryable schema with stable IDs and consistent geo joins.
• Track history. HUD only shows the current snapshot. We archive each quarter-end so you can chart drift and run change-over-time analysis.
• Link enforcement to lender IDs. CWT termination notices live as PDFs in the Federal Register. We parse them and resolve every named lender back to the same lender_id you use on the rest of the API.

• Refresh latency is bounded by HUD's — daily fields can be ≤24 hours behind; monthly performance fields are typically current to the last calendar month-end and refresh in the second week of each month.
• Some loans HUD attributes to a state lack a parseable ZIP — they appear at the state level but not in any ZIP rollup. Reconciling ZIP-sum to state-total recovers ~90% on average.
• Pre-2010 CWT termination notices use prose-paragraph formats that don't parse cleanly into structured affected-roster rows. Modern notices (post-2010) parse fully.
• HUD-office-level performance is sparse for low-volume lenders; we don't fall back to ZIP-aggregated estimates because mixing methodologies would give misleading numbers.