US Bank, Credit Union & Household Credit Data: Methodology

TREMOR's US Bank Aggregates tool tracks credit stress across three independent data sources: FDIC bank Call Reports, the New York Fed Consumer Credit Panel (household debt), and NCUA credit-union aggregates. This page documents the methodology for all three - data sources, indicators, how delinquency and charge-off rates are computed, and known limitations. For the validation/reconciliation detail, see the NCUA data validation page.

Open the interactive US Bank Aggregates tool on TREMOR.

FDIC banks

Data source

Data is sourced from the FDIC Statistics on Depository Institutions (SDI) API, which aggregates quarterly Call Report data filed by all FDIC-insured commercial banks and savings institutions. The API provides bank-level data; we fetch all banks per quarter and sum locally to produce industry-wide aggregates (no server-side aggregation endpoint exists).

Coverage

• Period: Q1 2001 to Q4 2025 (100 quarters, 25 years)
• Banks: ~9,900 in 2001 declining to ~4,600 in 2025 (consolidation)
• Frequency: quarterly; units: thousands of USD; update lag: ~6 weeks after quarter end

Indicators (53 across 5 metric types)

Loan balances (18): loan-segment balance fields covering Real Estate (Construction, CRE, Multifamily, Farmland, 1-4 Family Residential), C&I, Consumer (Credit Cards, Auto, Other), Agriculture, Lease Financing, and other categories. Most start Q1 2001; auto loans (LNAUTO) start Q1 2011 when FDIC introduced the separate line item.

Delinquency & charge-offs (35): for 7 segments (Total, RE, C&I, Consumer, Credit Cards, Auto, Agriculture), 5 metrics each:

• 30-89 DPD (P3*) - loans 30 to 89 days past due
• 90+ DPD (P9*) - loans 90 or more days past due
• Nonaccrual (NA*) - loans no longer accruing interest
• Noncurrent (NC*) - 90+ DPD plus nonaccrual (provided by FDIC)
• Net charge-offs (NT*) - gross charge-offs minus recoveries

Computed rates

• 30+ DPD rate (PDNA) = (P30-89 + P90+ + Nonaccrual) / Loan Balance × 100. Past Due and Non-Accrual; all loans 30+ days past due plus non-accrual.
• 90+ DPD rate (Noncurrent) = (P90+ + Nonaccrual) / Loan Balance × 100. Matches FDIC's published "noncurrent" rate.
• Net charge-off rate = (quarterly incremental NCO × 4) / Loan Balance × 100. FDIC reports YTD cumulative; we de-cumulate to quarterly (Qn YTD minus Qn-1 YTD), then annualize.
• Share of total = (Segment Balance / Total Loans) × 100.

Concept mapping: India vs US

• India 30+ DPD (everything past due) = US PDNA (30-89 + 90+ + Non-accrual)
• India 90+ DPD / NPA = US Noncurrent (90+ + Non-accrual)
• Charge-off - same concept; timing varies (typically 180 days for cards, 120 for installment)

Known limitations

• Auto loans blank before Q1 2011: a genuine regulatory gap, not an error. Before March 31, 2011, auto loans were lumped into "Other Consumer Loans". FDIC issued FIL-9-2011 introducing automobile loans as a separate line item effective Q1 2011. The numbers are consistent: Q4 2010 Other Consumer ($556B) is approximately Q1 2011 Auto ($283B) + Q1 2011 Other Consumer ($271B). No backfill is possible.
• Consumer sub-type gap: Credit Cards + Auto + Other Consumer is ~95% of total Consumer; the ~5% gap is "revolving credit plans other than credit cards" not broken out.
• RE sub-type delinquency: not available - FDIC provides RE total delinquency only.
• Agriculture delinquency: reported by only ~80-150 large banks, so the aggregate may understate.
• Gross charge-offs: not available from the FDIC API - only net charge-offs.

FDIC API: GET https://banks.data.fdic.gov/api/financials - no authentication, free public API.

NY Fed / Equifax (household debt)

Data source

The Quarterly Report on Household Debt and Credit is published by the Federal Reserve Bank of New York, based on the Consumer Credit Panel (CCP) - a nationally representative 5% random sample of all individuals with a credit report (Equifax), covering ~44 million individuals per quarter.

Coverage

• Period: Q1 2003 to Q1 2026 (93 quarters)
• Universe: ALL consumer credit - banks, credit unions, fintech, auto finance companies, mortgage companies, student loan servicers, collections agencies
• Frequency: quarterly (~6 weeks after quarter end); no API exists (Excel downloaded quarterly)

Debt categories

• Mortgage (first-lien residential), HELOC (home equity revolving), Auto Loans (incl. leases, all lenders), Credit Cards (bank/store/national), Student Loans (federal + private), Other (retail, consumer finance, installment).

Metrics

• 90+ DPD rate (stock): % of balance 90+ days late + 120+ + Severely Derogatory (includes charge-offs still on the credit report).
• 30+ DPD flow (transition rate): % of balances transitioning from current into 30+ delinquent this quarter. Leading indicator.
• 90+ DPD flow (transition rate): % transitioning into 90+ seriously delinquent this quarter (numerator: balances that moved from below 90 DPD into 90+; denominator: balances below 90 DPD last quarter). Captures both "sudden distress" and "slow roll".
• Mortgage straight-through flow (current to 90+): stricter early-warning view - mortgage balances current last quarter that skipped directly to 90+ this quarter (~3 consecutive missed payments). NY Fed publishes this transition only for mortgages ("Mortgage Cracking"); we do not approximate it for other loan types because the underlying transition-matrix data is not public.
• FICO score bands (origination quality): mortgage and auto origination volumes by Equifax Risk Score band (<620, 620-659, 660-719, 720-759, 760+). Mortgage from Q1 2003, auto from Q1 2004.
• Severely Derogatory (stock, all-debt): charge-off proxy = repossession + charge-off to bad debt + foreclosure. Total consumer debt only.

FDIC vs NY Fed - when to use which

Both sources are credible and widely cited; they answer different questions.

• Source basis: FDIC = bank-side filings (Call Reports); NY Fed = consumer-side records (Equifax credit bureau).
• Lender coverage: FDIC = FDIC-insured banks only; NY Fed = ALL lenders.
• Charge-offs: FDIC excludes them from delinquency stock (separate net charge-off metric); NY Fed includes them in 90+ DPD as "Severely Derogatory" (the loan still appears on the consumer's credit report).
• 30+ DPD definition: FDIC uses a stock measure (PDNA over balance); NY Fed uses a flow rate (% transitioning into 30+ this quarter).
• Student loans: not separately reported by FDIC; NY Fed has full coverage ($1.66T, ~9% of household debt).
• Why rates differ: NY Fed runs higher because the broader lender universe includes subprime non-bank lenders, and charge-offs remain in the numbers.
• Use FDIC for "how stressed are bank portfolios?"; use NY Fed for "how stressed are households?".

NCUA credit unions

This is the public-facing record of how the NCUA credit-union data is processed - what is published, what is deliberately hidden, and why. The companion NCUA data validation page has the full three-layer reconciliation and anomaly catalogue.

1. Data source

Data is sourced from the NCUA Aggregate Financial Performance Reports (FPR), published quarterly by the National Credit Union Administration (the federal regulator for credit unions, counterpart of the FDIC). These are aggregated 5300 Call Report numbers covering all federally-insured credit unions (FICUs). Each release contains FCU-only, FISCU-only, and FICU (combined) variants; we use the FICU file consistently.

2. Coverage

• Period: Q4 2002 to Q4 2025 (86 quarters)
• Universe: all federally-insured credit unions; 4,287 institutions in Q4 2025 (down from ~9,000 in 2002)
• Total loans: $1.72T at Q4 2025
• Tree: Total Loans -> Consumer (Auto, Credit Cards, Other Unsecured, Other Secured non-RE, Leases, Student, PAL) + Real Estate (1-4 Family 1st, Junior, Other) + Commercial (RE Secured, Not RE Secured)
• Per segment: balance, 30-59 DPD, 60+ DPD, quarterly net charge-off rate (annualized)

3. Computation methodology

3.0 Publish-as-filed policy. We publish what NCUA filed - we do not substitute values, smooth lines, or interpolate. Where our quarterly derivation produces a mathematically impossible result (a negative quarterly NCO from NCUA's own impossible-YTD-decrease), we let the math fall out and document the cause rather than overriding the figure. Adopted formally 2026-05-20; a prior catalogue of ~30 masked cells has been removed and the underlying NCUA-filed values now appear with per-quarter explanations.

3.1 Stock vs flow. Balances and delinquency dollars are quarter-end snapshots (used directly). Charge-offs and recoveries are reported year-to-date cumulative and must be converted to quarterly flows.

3.2 YTD-to-quarterly conversion. NCUA reports charge-offs/recoveries YTD, so a quarter-only figure requires subtracting the prior quarter's YTD: Q1 NCO = Q1 YTD; Q2 = Q2 YTD - Q1 YTD; Q3 = Q3 YTD - Q2 YTD; Q4 = Q4 YTD - Q3 YTD. Where the prior-quarter YTD is genuinely missing (first quarter in the extract, or pre-2008 annual-only data), the quarterly NCO is NaN (we cannot fabricate it). Where the YTD itself decreases within a calendar year (impossible for a cumulative, but observed in NCUA's own leases/MBL/student cells), the subtraction yields a negative quarterly NCO which we render on the chart and explain in the anomaly catalogue.

3.3 Net charge-off rate (annualized) = (Q-only Net CO × 4) / Average Balance × 100, where Net CO = charge-offs minus recoveries and Average Balance = (current + prior quarter balance) / 2. We use the quarterly-annualized definition (FDIC convention) rather than NCUA's YTD-annualized, for two reasons: it is more responsive to current-quarter stress, and it keeps FDIC bank rates and NCUA credit-union rates apples-to-apples on the same chart. We reconciled the YTD-annualized version against NCUA's own QCUDS time series: 14 of 14 quarters match within 0.5 basis points (see the validation page, Layer 3).

3.4 Algebraic Consumer derivation. NCUA's 3-bucket taxonomy (Consumer + Real Estate + Commercial = Total) is mathematically closed - verified at 0.0000% residual across 16 modern quarters. For pre-Q1 2022 quarters where NCUA does not publish a single Consumer-total code, we derive Consumer = Total minus RE minus Commercial using NCUA's own legacy aggregate rows. This computes an aggregate NCUA does not itself publish at that level, using NCUA-published bucket totals as inputs - it does not contradict any filed value.

3.5 Source-code fallback. For Commercial balance, NCUA switched account codes mid-series: from Q3 2017 the modern total is A400P + A718A5; from Q1 2011 to Q2 2017 it was the single legacy code A400T. We read whichever code NCUA was actively reporting each quarter (cross-verified in the overlap). Both are NCUA-filed; the only choice is which code to read.

4. Headline metrics

60+ DPD is NCUA's headline reportable delinquency rate (cited in NCUA press releases). 30-59 DPD is the early-warning flow indicator (loans newly entering distress). CU 60+ is not directly comparable to bank 90+: credit unions use a 60-day trigger (banks 90), and retain delinquent loans longer before charge-off, so a CU 60+ rate is structurally a broader, somewhat higher measure of stress than a bank 90+ rate. Read them as different definitions.

5. Display decisions

Sub-segment delinquency/charge-off charts for the five RE and Commercial sub-segments are temporarily hidden (their granular rate series only began Q1 2022 and show unexplained early spikes still being diagnosed); aggregate Real Estate and Commercial charts are published and reliable. Balance and % share charts for those sub-segments are still shown. Tiny-book segments (PAL $147M, Leases, private Student loans, Other Non-Commercial RE) produce inherently noisy rates - real NCUA data, but expect volatility; for macro signal prefer the major-book series (Total, Consumer, Auto, Credit Cards, Real Estate, Commercial).

6. NCUA anomalies catalogue

The full quarter-by-quarter catalogue of every visible blip or step-change lives on the validation page, split into Section A (anomalies - impossible/implausible/first-wave NCUA-filed values, shown as filed with explanation) and Section B (reclassifications - NCUA moving loans between buckets, same loans different presentation). Notable items include the lease NCO impossible-YTD-decrease pairs, the legacy MBL batches, the student-loan Q4 batches, the Q4 2006 credit-card 30-59 DPD revised value, and the Q1 2011 / Q3 2017 account-code restructures.

7. Coverage gaps (no data available)

• Q1-Q3 2003 and Q1-Q3 2004: only Q4 year-end files exist.
• Mid-2005 / 2006 quarters: pre-2008 layout lacks RE and MBL loss-line label sheets.
• Auto sub-segments pre-Q2 2013; modern granular RE/Commercial NCO + delinquency pre-Q1 2022.
• PAL pre-Q4 2010, private Student loans pre-Q1 2011 (product-introduction dates - NaN, not zero).
• Number of Credit Unions for some intra-year quarters (each FPR file's count cell reflects its own report quarter only).

8. Validation

Validation runs in three independent levels: Level 1 pipeline internal-consistency (subtotal math, taxonomy closure, YTD monotonicity, stock-flow, re-derivation, z-score outliers, NCUA-headline cross-check); Level 2 cross-check vs the NCUA QCUDS PDF; Level 3 cross-publisher vs Federal Reserve Z.1. Full findings are on the NCUA data validation page. A regenerated dataset is not shipped if Level 1 reports any failures.