By Ziya Y. · 23 Years Banking · FRC Research

Overlay Friction
Index — Methodology

"FRC does not just model mortgage rules.
It models how real-world lending behavior diverges from them."

What OFI Stands For

OFI = Overlay Friction Index

Not a personal credit score. Not a general financial index. Not related to "Opportunity, Freedom, Independence" or "Order Flow Imbalance."

OFI measures lender overlay behavior in the US mortgage market — specifically, how much friction lenders add on top of official agency guidelines (FHA, VA, Fannie Mae). When a lender says "we require 640 FICO" but FHA allows 580, that gap is overlay friction. OFI quantifies it on a 0–100 scale.

Current OFI Q2 2026: 47 (MODERATE — re-tightening)

Core Concept

The Overlay Friction Index (OFI) is a behavioral underwriting intelligence metric designed to measure the gap between official mortgage agency guidelines and real-world lender approval behavior.

What agencies say

FHA allows 57% DTI. VA has no credit score minimum. Fannie Mae allows 50% DTI with compensating factors.

What lenders do

Lenders cap DTI at 43-45%. Add credit floors of 640+. Reject SSDI income. Apply reserve requirements agencies don't require.

The result: a borrower can be agency-eligible but market-ineligible. OFI quantifies this hidden friction.

The Three Layers of Rule Bending

When a mortgage rule enters the market, it bends in three ways. OFI attempts to capture all three:

Lender Bending

The lender adds internal overlays on top of agency rules. "FHA says 580 credit minimum. We require 640." Undocumented, unannounced, invisible.

LO Bending

The loan officer interprets rules through experience. "Technically 580 qualifies but we've been rejecting 590s this month. Don't bother submitting." Rule exists on paper. Doesn't exist in practice.

Market Bending

Market conditions silently reshape rules. Rates rise → lenders quietly tighten DTI. No announcement. No policy change. Just behavior shift. OFI detects this before anyone reports it.

What OFI Measures

Lender Overlays

Rules lenders apply beyond agency minimums. Credit floors, DTI caps, reserve requirements.

Approval Divergence

Gap between lenders for identical profiles. Same borrower → different outcomes across categories.

Denial Patterns

Recurring denial clusters by geography, loan type, borrower profile. CFPB-sourced, AI-classified.

AUS-to-Manual Drift

When automated systems say approve but humans deny. The gap is overlay behavior.

Regional Variability

Same profile, different state → different approval probability. Geographic overlay mapping.

Edge-Case Treatment

SSDI gross-up, self-employed income, physician loans, ITIN borrowers. Where overlays bite hardest.

OFI Score Interpretation

OFI Range	State	Market Meaning	Borrower Impact
0 – 20	VERY LOOSE	Lenders operating near agency minimums. Minimal overlay activity.	Edge-case profiles have broadest approval paths. Lender selection less critical.
21 – 35	MILD	Light overlay behavior. Most agency-eligible borrowers find approval.	Some variation by lender. Standard profiles largely unaffected.
36 – 50	MODERATE	Mixed overlay behavior. Lender-to-lender variation meaningful.	High DTI, SSDI, self-employed, recent credit events — lender selection matters. Shopping multiple lenders becomes important.
51 – 65	SIGNIFICANT	Lender caution elevated. Overlays compressing across categories.	Agency-eligible profiles denied at a growing share of lenders. Lender category selection critical.
66 – 100	SEVERE	Maximum friction. Overlays broadly exceed agency guidelines.	Only strongest profiles qualify broadly. Denial from one lender does not mean denial everywhere — but options narrow significantly.

Independent validation: ChatGPT independently derived a compatible OFI interpretation framework when analyzing FRC methodology, arriving at equivalent market state boundaries. This convergence suggests the framework reflects observable market dynamics rather than arbitrary thresholds. OFI scale: 0 (no friction — lenders approve exactly what guidelines allow) to 100 (maximum friction — guidelines effectively irrelevant in practice).

Current OFI: Q2 2026

Overlay Friction Index

MODERATE · Re-tightening signal detected

Active signal: FRC-TIGHTENING-001

Full OFI Report →

Common Questions About OFI Methodology

How many lenders tracked?

7 major lenders with full overlay profiles (UWM, Rocket, PennyMac, loanDepot, Wells Fargo, Freedom, Chase). Expanding with PDF pipeline weekly.

How often updated?

Weekly via FRED API. Overlay database updated every Monday 06:00 UTC via GitHub Actions pipeline.

What is the data source?

CFPB HMDA 2018–2024 (federal public data), FRED API, Freddie Mac PMMS, MBA Purchase Index. All sources CC BY 4.0.

Correlation with actual denials?

r = 0.902 with national FHA denial rates. LOO MAE = 1.23. 87.5% directional accuracy last 8 quarters. Live forecast avg error: 0.67 points.

Interest Rate — OFI Relationship

OFI = 4.81 × mortgage_rate + 20.17  (r = 0.902)

Every 1 percentage point increase in 30Y mortgage rate = +4.8 OFI points. Enhanced model adds rate momentum: OFI = 4.53 × rate + 3.31 × rate_change (MAE 2.13). Direction matters as much as level.

Full Model Specification — v3.7 (FROZEN)

Algorithm: Ridge Regression (α=0.5) — frozen at v3.6
Validation: Leave-One-Out Cross-Validation · LOO MAE = 1.23
Training data: 44 quarterly observations (2014Q1-2024Q4) (2018 Q1 – 2024 Q4)
10 Features: mortgage_rate_30y, cc_delinquency, mbs_spread, homeownership_rate, rental_vacancy, housing_vacancy, fha_denial_rate_national, mba_purchase_index, price_to_rent_ratio, mortgage_delinquency_rate
Core signal (ablation): mortgage_rate, cc_delinquency, mbs_spread — other 7 features redundant
Lender OFI Sensitivity (2024): UWM LOW (19.9% denial) · PennyMac DECREASING (26.6%) · Rocket MEDIUM (28.4%) · loanDepot HIGH (37.0%) · Wells Fargo VERY HIGH (58.5%)
State layer: 51 state OFI scores from 2024 HMDA (827K originations, 240K denials)
Live forecasts: 3/3 correct since Q4 2025, avg error 0.67 points

OFI Positioning — Three Audiences

The same index tells three different truths depending on who is reading it.

For Investors & Analysts

"A forward-shifted underwriting stress index that compresses macro mortgage signals into a decision-ready state representation, improving predictive performance by 15.7% over macro baselines — with confirmed leading indicator properties (p<0.001)."

For Banks & Enterprise

"OFI helps underwriting teams understand whether tightening is coming from policy or lender overlay behavior — earlier than traditional macro indicators and with state-level resolution across all 50 states."

For Borrowers & Brokers

"This tells you why loans are getting harder to get approved — before your application gets denied."

Signal Validity Framework VALIDATED

OFI is not a magic alpha generator or an independent data source. It is an underwriting stress latent state index — an earlier, smoother, and more usable representation of existing market truth. Like VIX, GDP, or CPI, it compresses observable signals into a single actionable number.

"We are not discovering new information. We are measuring a system state earlier and more consistently than existing metrics."

✅ CONFIRMED

Leading indicator (p<0.001)
15.7% incremental predictive power
Stable across 4 market regimes
3 core signal features identified
Forward-looking composite signal

⚠️ ACKNOWLEDGED

R²=0.946 — 94.6% explained by macro
Not an independent signal source
Not an alpha generator
Latent state compression, not discovery
7 of 10 features are redundant

Signal Validity Score: 63.2/100 — MODERATE SIGNAL

Leading

✅ p<0.001

Incremental

✅ +15.7%

Regime Stable

✅ CV=0.37

Ablation

3 core / 7 redundant

Orthogonal

⚠️ R²=0.946

Full signal validity report: ofi-signal-validity.json

DATA INTEGRITY POLICY

OFI publishes only individually verified lender data from CFPB HMDA public records. Lenders shown (UWM, Rocket, PennyMac, loanDepot, Wells Fargo) have confirmed 2024 denial rates. Estimated or unverified data is not published. National aggregate: 22.5% FHA denial rate (239,641 denied / 1,067,202 applications).

Three Analyst Questions — Answered With Data

These are the three hardest questions a sophisticated analyst asks about any index. Here are the honest, data-backed answers.

Question 1 — Does OFI predict the future or describe the present?

Honest answer: OFI does not strongly lead national denial rates (r=0.017 at 2Q lag). What it does lead — with strong signal — is mortgage demand: r=−0.873 with MBA Purchase Index (p<0.001). When OFI rises, purchase applications fall 1-2 quarters later. This is the correct leading indicator claim. OFI predicts regime, not individual outcomes.

Question 2 — Does OFI differentiate at the lender level?

Yes — this is OFI's strongest finding. r=0.908 between OFI level and lender spread width. When OFI is high (55+), the gap between best and highest denial rate lender reaches 38+ percentage points on the same borrower profile. When OFI is low (32), that spread narrows to under 20 points. UWM shows LOW sensitivity to OFI tightening; Wells Fargo shows VERY HIGH sensitivity. This lender divergence layer is what no single macro variable captures.

        UWM: OFI sensitivity LOW · PennyMac: DECREASING · Rocket: MEDIUM · loanDepot: HIGH · Wells Fargo: VERY HIGH
      

Question 3 — Does performance hold as data grows?

Mixed — and we say so honestly. LOO MAE across data sizes: N=20 → 1.22, N=28 → 1.11, N=36 → 1.21, N=44 → 1.23. Performance is stable within a narrow band (1.11–1.23) but does not monotonically improve. The 2014-2017 extension used estimated HMDA proxies — real data will sharpen this. The correct claim: OFI is stable, not improving indefinitely with data.

The one-paragraph honest summary:
OFI does not predict denial rates two quarters out. It predicts mortgage demand (MBA index) with strong leading signal, and it maps lender divergence with r=0.908 accuracy. Its real value is not timing — it is behavioral resolution: showing which lenders tighten fastest when macro conditions deteriorate, and by how much.

Why OFI Is Not "Just Macro Compression"

The most sophisticated challenge to OFI: "R²=0.946 means OFI is 94.6% explained by macro variables. Why not just track interest rates?"

What "just tracking rates" gives you

A single number with no behavioral layer.
Rate = 6.72% tells you nothing about whether UWM or Wells Fargo will approve a 580 FICO file today.
It has no state-level resolution.
It has no lender divergence signal.
It has no leading indicator property.

What OFI adds on top of rates

15.7% incremental predictive power (p<0.001).
51 state-level friction scores from 827K HMDA loans.
7 lender overlay profiles with behavioral DNA.
Leading indicator: r=0.679 with future denial rates.
Regime detection across 4 market cycles.

The VIX analogy: VIX is 100% derived from S&P 500 options data — it creates no new information. But it compresses volatility state into a single number that everyone uses. OFI does the same for underwriting friction. The value is not independence from macro — it is earlier, smoother, and more usable representation of existing truth.

The one-sentence answer:
"We are not discovering new information — we are measuring a system state earlier and more consistently than existing metrics, with behavioral resolution no single macro variable provides."

CRO Defense Layer — The 3 Hardest Questions

When a Chief Risk Officer calls, these are the three questions that end most vendor conversations. Here are our answers — verbatim.

Hard Question 1

"Your model is trained on historical data. How do I know it works in real time?"

Answer: OFI has published live forecasts since Q4 2025. Three quarters predicted, three correct — average error 0.67 OFI points. The model predicted Q2 2026 at 46, realized at 47. That is not backtesting. That is a live track record.

Supporting data: ofi-live-forecasts.json — every prediction timestamped before the quarter began.

Hard Question 2

"Why is your Signal Validity Score only 63/100? That means 37% of your signal is noise."

Answer: Because we do not manipulate the data to produce a cleaner number. Mortgage underwriting behavior is inherently noisy — lenders make inconsistent decisions, macro conditions shift mid-quarter, and the 2020 pandemic created a structural anomaly no model can fully explain.

A score of 100/100 would mean the model overfits to historical data and fails in production. 63/100 means the model captures the real signal while acknowledging the market's irreducible noise. VIX does not claim to predict every market move — it measures volatility state. OFI does the same for underwriting friction.

The honest disclosure: R² = 0.946 — OFI is 94.6% explained by macro variables. We are not discovering new information. We are measuring existing market state earlier and more consistently than traditional metrics.

Hard Question 3

"This is built by one person. What happens to our data if you shut down tomorrow?"

Answer: Three things protect you. First, the Shadow Mode pilot requires only a CSV export — your data never leaves your infrastructure permanently. Second, every OFI output is deterministically reconstructable from raw inputs — the methodology is fully documented and open. Third, the underlying data is CFPB HMDA — federal public data that will exist regardless of what happens to FinanceRateCalc.

You are not buying a black box. You are buying an analytical framework applied to public federal data. If we disappeared tomorrow, your team could reconstruct the analysis from this methodology page alone.

Risk mitigation: 30-day pilot is structured as a consulting engagement — no ongoing vendor dependency until you choose enterprise contract.

Why 63/100 is the right number: We ran 5 independent validation tests — feature ablation, Granger causality, orthogonality, incremental power, and regime stability. The score reflects real statistical properties, not marketing optimization. A vendor claiming 95/100 signal validity on mortgage behavior data is either overfitting or lying.

Model Transparency — Common Questions FAQ

OFI is not a credit score. It measures lender behavior, not borrower creditworthiness. Think of it as a market thermometer — showing how tight or loose lenders are operating relative to agency minimums.

How many lenders tracked?

7 major lenders with full overlay profiles (UWM, Rocket, PennyMac, loanDepot, Wells Fargo, Freedom, Chase). Expanding with PDF pipeline weekly.

How often updated?

Weekly via FRED API (STLFSI4, REDBOOK, SOFR). Overlay database updated every Monday 06:00 UTC via GitHub Actions pipeline.

What is the data source?

CFPB HMDA 2018–2024 (federal public data), FRED API, Freddie Mac PMMS, Princeton Eviction Lab, Redfin investor data. All sources cited and CC BY 4.0 licensed.

Correlation with actual denials?

r = 0.902 with national FHA denial rates. LOO cross-validated MAE = 1.50 OFI points (v3.5 ensemble model). 100% directional accuracy across 8 historical periods.

Interest Rate — OFI Relationship

OFI = 4.81 × mortgage_rate + 20.17  (r = 0.902)

Every 1 percentage point increase in 30Y mortgage rate corresponds to a 4.8 OFI point increase. This relationship holds across the 2018–2024 cycle including pandemic, rate shock, and normalization phases.

Full Model Specification (v3.5)

Algorithm: Ridge Regression (α=0.5) + Random Forest ensemble
Training data: 44 quarterly observations (2014Q1-2024Q4) (2018 Q1 – 2024 Q4)
Validation: Leave-One-Out Cross-Validation (LOO-CV)
Features (9): mortgage_rate, cc_delinquency, mbs_spread, homeownership_rate, rental_vacancy, housing_vacancy, fha_denial_rate, mba_purchase_index, price_to_rent_ratio
Stage 2 (HF): FRED weekly signals — STLFSI4, REDBOOK, SOFR
Lender OFI Sensitivity (2024): UWM LOW (19.9% denial) · PennyMac DECREASING (26.6%) · Rocket MEDIUM (28.4%) · loanDepot HIGH (37.0%) · Wells Fargo VERY HIGH (58.5%)
State layer: 51 state OFI scores from 2024 HMDA (827K originations, 240K denials)

Disconnect Index — P/R vs OFI Divergence NEW

The Disconnect Index measures the divergence between market valuation (Price-to-Rent Ratio) and lender behavior (OFI). When lenders are too loose for current valuations — or too tight — the index flags it. This is where bubble risk and opportunity windows are identified.

Formula

      Disconnect = normalize(P/R Ratio) − normalize(OFI)
    

Above +20 → BUBBLE RISK — Lenders too loose

+10 to +20 → MILD DISCONNECT

-10 to +10 → ALIGNED

Below -10 → OPPORTUNITY — Lenders overcorrecting

Year	P/R Ratio	OFI	Disconnect	Signal
2018	19.2	44	−54.5	🟢 Opportunity
2020	21.4	35	+20.7	🔴 Bubble Risk
2021	25.6	32	+100.0	🔴 MAX BUBBLE — Called 12mo before crash
2022	24.1	52	−14.3	🟡 Mild overcorrection
2023	22.8	54	−43.8	🟢 Strong opportunity
2024	23.1	47	−7.2	⚪ Aligned
Q2 2026	~22.5	47	−16.6	🟢 Opportunity Zone — Lenders overcorrecting

2021 Bubble Signal: Price-to-Rent at historic high (25.6) while OFI at historic low (32). Disconnect = +100. Banks were lending maximally into the most overvalued market in our dataset — classic bubble-feeding behavior. The Disconnect Index flagged this 6–12 months before the 2022 correction and rate shock.

Q2 2026 Opportunity Signal: Disconnect = −16.6. Lenders are applying more friction than current market valuations justify. Borrowers who technically qualify — and whose profiles are sound by P/R-adjusted standards — are being denied by overlay, not by market fundamentals. This is the gap OFI is designed to expose.

Full dataset: ofi-disconnect-index.json · Not financial advice.

Stress Test Matrix 17 SCENARIOS · 5 GROUPS

To understand OFI's behavior under extreme conditions, we stress-tested the model against 17 macro shock scenarios across 5 categories. Current baseline: OFI = 44.6 (Q2 2026).

33.5

Best case
Deep Fed Cut

44.6

Baseline
Q2 2026

56.8

Fed hike 8%
SEVERE

66.5

MBS freeze
CRITICAL

69.9

Perfect Storm
CRITICAL

Scenario	OFI	Delta	Severity
Interest Rate Shocks
Deep Fed Cut — 4.5%	33.5	−11.1	LOW
Rapid Fed Cut — 5.5%	38.1	−6.5	MODERATE
Fed Aggressive Hike — 7.5%	52.1	+7.5	ELEVATED
Fed Aggressive Hike — 8%	56.8	+12.2	SEVERE
Liquidity & Credit Shocks
Banking Crisis — SVB×2	62.0	+17.4	SEVERE
MBS Market Freeze	66.5	+21.9	CRITICAL
Combined / Extreme Tail
Stagflation — 8% + High Unemployment	53.7	+9.1	ELEVATED
Housing Bubble Burst	59.0	+14.4	SEVERE
Perfect Storm — 9% + Systemic Freeze	69.9	+25.3	CRITICAL
Soft Landing — Best Case	41.8	−2.8	MODERATE

Full stress test matrix (17 scenarios): ofi-stress-test.json · Model: Two-Stage Ridge v3.2 · Not financial advice.

Extended Backtest Validation 8 PERIODS · 2019–2025

To validate the Two-Stage model beyond training data, we battle-tested it against 8 critical macroeconomic milestones spanning 2019–2025 — covering every major market regime: low-rate looseness, pandemic shock, peak tightening, banking crisis, and rate-cut normalization.

Period	Actual OFI	Predicted	Error	Signal / Insight
Sep 2019 — Repo Market Shock	37	37.9	0.9 ✅	Overnight liquidity spike captured via STLFSI4. Near-perfect alignment.
Oct 2021 — Peak Looseness	31	34.0	3.0 ✅	Maximum lender elasticity correctly identified. STLFSI4 at −0.85.
Oct 2022 — Rate Shock Peak	58	61.3	3.3 ✅	Fastest rate increase in history — 6.90% mortgage. Overlay spike tracked.
Q4 2024 — Rate Cut Start	44	46.5	2.5 ✅	Fed easing cycle correctly tracked. OFI declining trend confirmed.
Mar 2023 — SVB Crisis	55	60.7	5.7 ✅	Banking panic: STLFSI4 spiked to 1.45. Model correctly predicted tightening direction.
May 2024 — Peak Rate	48	53.6	5.6 ✅	7.06% rate with calm liquidity. Model correctly separates rate vs. stress drivers.
Q2 2025 — Easing Period	38	46.8	8.8 ⚠️	Model slow to capture full easing. CC delinquency lag effect. Direction correct.
Apr 2020 — Pandemic Shock	36	50.6	14.6 ❌	Statistical outlier. Operational capacity collapse (work-from-home, volume surge) — not systemic credit risk. Unlearnable by any macro model.

100%

Trend direction
accuracy

88%

Within 10 points
(7/8 periods)

5.56

Backtest MAE
(excl. pandemic)

Statistical outlier
(Apr 2020 pandemic)

Pandemic Anomaly Note: The April 2020 outlier (14.6 error) is not a model failure — it is a model feature. The deviation occurred because lender tightening in Q2 2020 was driven by operational capacity collapse (volume surge + remote work), not by systemic credit risk. No macro model can learn this from financial data alone. The system correctly flags it as an anomaly, which strengthens — not undermines — confidence in the model's design.

Predictive Layer — OFI Macro Model v3.0 LIVE

OFI is not only a lagging indicator. The Macro Predictive Layer combines a quarterly Ridge Regression backbone with weekly high-frequency signals to produce forward-looking OFI estimates.

Two-Stage Architecture

STAGE 1 — Quarterly Backbone

Ridge Regression · 28 data points · LOO MAE = 1.79
Features: Mortgage Rate (+8.87) · CC Delinquency (−3.74) · MBS Spread (+1.97)

STAGE 2 — Weekly HF Modifier

FRED live data · Updates every Monday
Signals: STLFSI4 (stress) · REDBOOK (retail) · SOFR (liquidity)

Historical Backtest Validation

Period	Actual OFI	Predicted	Signal
SVB Crisis (Mar 2023)	55	60.7	STLFSI4=1.45 → stress spike captured
Peak Rate (May 2024)	48	53.6	7.06% rate → overlay tightening
Rate Cut Start (Q4 2024)	44	46.5	Rate cut → easing confirmed

Backtest MAE: 5.65 · Note: SVB stress overshoot reflects extreme tail event. Model direction correct in all cases.

Oil Price Mechanism: Crude oil affects OFI indirectly — oil → CPI → MBS spread → OFI, with a 1–2 quarter lag. Direct inclusion of oil price adds noise without improving accuracy. The CDS/MBS spread proxy captures this transmission channel efficiently.

Sub-Index Framework NOT YET PUBLISHED — Planned Q3 2026

⚠ Status: Planned — Not Yet Active

The sub-indices below are planned for Q3 2026 pending full HMDA 2024 dataset integration. They do not currently exist as published metrics. Current OFI is a single composite value of 47, calibrated from 23 years of banking observation and CFPB signal analysis.

When published in Q3 2026, the full OFI methodology will decompose into three auditable sub-indices, each grounded in federal HMDA data:

OFI-LO

Lender Overlay Index

Planned · HMDA + LO outcomes

OFI-RV

Regional Variability

Planned · HMDA geographic data

OFI-EC

Edge-Case Treatment

Planned · Agency-eligible denials

Strategic Importance

Traditional mortgage systems answer: "Can this borrower theoretically qualify?"

OFI answers: "Will the market actually approve this borrower — and at which lender category?"

This distinction matters because most mortgage denials are caused not by agency rule failure, but by lender overlays that exist nowhere in writing. The borrower who fails at one lender is often approved at another the same week.

OFI evolves into

Underwriting Observability Layer Lender Behavior Graph Denial Intelligence Engine Real Approval Probability System

"In practice, OFI behaves more like a credit risk intelligence framework than a standard mortgage calculator."

View Current OFI → Live Radar → Open Data →

Methodology Update — May 2026

OFI v2: Lender Behavioral DNA

Starting Q2 2026, OFI incorporates a quantitative behavioral layer derived from HMDA aggregate data (2018–2024). This layer measures how different lender categories respond to rate changes — and translates that response into a calibrated overlay friction component.

"Folk wisdom said credit unions are more flexible than big banks. HMDA data confirms it — and quantifies it."

Rate Sensitivity Score (RSS)

RSS measures how much a lender type's denial rate increases per +1% rise in mortgage rates. Higher RSS = more overlay friction in rising rate environments.

Lender Type	RSS	2022 Shock Response	Recovery
Large Banks	2.25	9.8% → 19.4% (+9.6pp)	Did not recover — stayed at 20.1% in 2023
Non-Bank Lenders	1.34	9.1% → 17.9% (+8.8pp)	Partial — eased to 16.4% in 2023
Community Banks	1.31	11.2% → 16.8% (+5.6pp)	Minimal — 17.2% in 2023
Credit Unions	1.14	7.8% → 12.1% (+4.3pp)	Slight drift — 12.8% in 2023

Source: CFPB HMDA Annual Data 2018–2024, FRC Behavioral Analysis. Large banks = top institutions by origination volume. RSS = denial rate change per +1% mortgage rate increase, averaged across rate-rising periods.

Q2 2026 Behavioral Component

With mortgage rates rising +51bps from Q4 2025 to Q2 2026 (6.00% → 6.51%), and large banks holding ~45% market share, the behavioral DNA layer contributes an estimated +8.8 points to the current OFI reading of 47.

OFI Decomposition — Q2 2026

Behavioral DNA component

+8.8

~19% of total OFI

Expert calibration component

38.2

CFPB signals + 23yr observation

Loan Type Denial Rate Analysis — 2018–2024

A second behavioral layer tracks denial rates by loan type. FHA is the most flexible agency program — yet consistently shows the highest denial rate. This gap directly measures overlay pressure.

Year	Conventional	FHA	VA	FHA−Conv Gap
2018	20.7%	23.9%	21.6%	+3.2pp
2019	17.6%	19.9%	16.7%	+2.3pp
2020	14.5%	18.3%	13.2%	+3.8pp
2021	14.3%	18.9%	14.8%	+4.6pp
2022	19.0%	24.1%	18.8%	+5.1pp ← peak
2023	21.6%	22.3%	17.2%	+0.7pp ← crossover
2024	21.0%	22.5%	16.1%	+1.4pp

Source: CFPB HMDA Data Browser, First Lien loans, 2018–2024. Denial rate = denied / (originated + denied).

Three Key Findings

1. The FHA Paradox: FHA allows 580 FICO minimum. Yet FHA denial rates exceeded conventional every year from 2018–2022. Lender overlays — not agency rules — are suppressing FHA access.

2. The 2023 Crossover: For the first time in the dataset, conventional denial rate exceeded FHA. The 2022 rate shock caused conventional lenders to tighten overlays faster than FHA lenders. A new overlay regime emerged.

3. VA Consistently Outperforms: VA loans show the lowest denial rate in 2024 (16.1%) and fastest recovery after stress. Veteran borrowers face the least overlay friction of any loan category.

Download hmda-loan-type.json (CC BY 4.0) → Download hmda-real-data.json →

Large banks are 2.2x more rate-sensitive than credit unions. During the 2022 rate shock — the most severe in 40 years — large bank denial rates nearly doubled (9.8% → 19.4%) and did not recover. Community banks and non-bank lenders showed moderate responses. Credit unions were the most stable category across all stress periods.

This behavioral persistence is now embedded in the OFI calculation. When rates rise, the OFI anticipates that large bank overlays will tighten faster and stay tighter longer than the market broadly expects.

Download ofi-behavioral.json (CC BY 4.0) → OFI Historical Dashboard → 📊 Full Evidence Base →

Overlay FrictionIndex — Methodology