Analyze a file →
FRC Intelligence · Original Research

The FHA Lender Taxonomy

The first behavioral classification of major FHA lenders. Four institutional types, validated across 8 years of federal data. Not ranked by rate — classified by behavior.

Data: CFPB HMDA 2018–2025 · the 100 largest lenders · 51 jurisdictions · ~2.9M FHA records
Method: BIC model selection, LOO ρ=0.894, label-randomization p<0.0001, bootstrap ARI 0.58
Published: June 2026 · FRC Intelligence

Moody's classifies bonds. Fitch classifies banks. No one has formally classified FHA lenders by behavioral type — until now. Using eight years of federal HMDA data, we identified four stable institutional types that explain why identical borrowers get opposite outcomes at different lenders.

The core insight: Lenders aren't just "strict" or "lenient." They have behavioral DNA — stable institutional patterns that survive market cycles, management changes, and regulatory shifts. Knowing the type predicts behavior better than any single rate metric.

Type I — Stable Core

Stable Core Identity-driven. Predictable. Clock-like.
Lender2025 DenialLSI8-yr StabilityRouting Signal
CrossCountry6.37%6★★★★★Primary for all borderline files
Guild7.14%21★★★★★Primary for W2, strong FHA fit
Rocket29.5%16★★★★☆Speed-first; credit-history files
Mr. Cooper20.1%32★★★★☆Servicing-focused; consistent credit
loanDepot29.1%44★★★☆☆Pipeline-driven; watch LSI trend
UWM21.6%30★★★★☆⚠ 2026 watchlist: +6.8pp drift
Planet Home18.2%38★★★★☆Specialty channel; complex files

Stable Core lenders show year-over-year rank correlation averaging 0.905 across the full panel. Their behavior is explained primarily by institutional identity, not market conditions.

Type II — Regime-Sensitive

Regime-Sensitive OFI-coupled. Market-driven. Conditional.
Lender2025 DenialLSIOFI CouplingCurrent Status
Freedom24.4%50r=0.66⚠ OFI 52 — regime gate marginally active
NewRez53.27%57r=0.84⚠ OFI 52 — lowest routing priority

Regime-sensitive lenders behave normally when OFI ≤ 50. Above that threshold, denial rates decouple from identity and track market friction. Current OFI: 52 (model: 4.81 × 30yr rate + 20.17, rate 6.60). Both lenders sit just above the gate threshold — regime effects present but moderate.

Type III — Improving Trend

Improving Trend Secular drift. Historical data misleads. Use recent.
Lender2018 Denial2025 Denial8-yr ChangeRouting Signal
PennyMac52.8%23.6%−29.2ppUse 2024-25 data only; 8-yr avg overstates risk

PennyMac is the only lender showing sustained secular improvement across the full 8-year period. Lag analysis confirms this is structural (operational change), not market-driven. Do not use historical averages.

Type IV — Unpredictable

Unpredictable Idiosyncratic. Uncorrelated. Avoid for routing.
Lender2025 DenialOFI CorrelationFHA Volume TrendStatus
Wells Fargo48.63%r=0.27 (uncorrelated)−63% since 2019Possible FHA exit strategy

Wells Fargo's denial behavior is uncorrelated with market conditions (r=0.27 vs 0.66-0.84 for regime-sensitive lenders). Its FHA volume has declined 63% since 2019. The data is consistent with an institution gradually exiting FHA lending. Not suitable for routing.

Why this classification matters

Type classification predicts behavior more reliably than any single metric. A lender's denial rate tells you where it is; its type tells you where it's going under different conditions. A Type I lender at 25% denial is more predictable than a Type II at 15% — because the Type II's 15% may become 45% when markets tighten.

This taxonomy is the foundation of the Zai routing system. Every file analysis starts with type — then adjusts for borrower profile, market OFI, and lender-specific denial genome.

Taxonomy understood. Now route your file.
See which type fits your borrower profile.
Get a file analysis — free →
Related: Denial Genome →  |  71% Not Your Fault →  |  9+2 Research →
Source: CFPB HMDA 2018–2025 public data. Classification methodology: two-way variance decomposition, BIC model selection (9+2 structure, BIC 384.0), leave-one-year-out validation (ρ=0.894), label-randomization testing (p<0.0001), bootstrap ARI (0.58±0.45). Type assignments are research outputs, not financial ratings. Individual lender behavior may change.

UWM classified as Stable Core with watchlist flag: 2025 denial rate +6.8pp above 8-year average. 2026 HMDA data is the formal falsification test.

Citation: FRC Intelligence (2026). "The FHA Lender Taxonomy." financeratecalc.com/lender-taxonomy.html
⚡ Analyze a file