NY Fed Household Debt Data: Validation

TREMOR tracks US household debt and credit stress from the New York Fed Quarterly Report on Household Debt and Credit, built on the Consumer Credit Panel (CCP) - a 5% random sample of all individuals with an Equifax credit report (~44 million people). This page documents how that data is validated. For the data sources and rate definitions across all three credit-stress sources (FDIC banks, NY Fed households, NCUA credit unions), see the US bank/credit data methodology page.

Validation runs in three layers. Headline verdict (as of the Q1 2026 release): Layer 1 - all 14 internal-consistency checks pass; Layer 2 - across 50 quarterly PDFs and 600 metric checks, zero mismatches; Layer 3 - no independent cross-publisher exists for CCP data (explained below). The interactive tool shows the live per-check and per-quarter results, which refresh each quarter.

Open the interactive US Bank Aggregates tool on TREMOR (NY Fed source) for the live validation tables.

Layer 1 - Pipeline internal consistency

Layer 1 is the pipeline's own internal-consistency suite. It verifies that what we extract from NY Fed's quarterly Excel reconciles to itself at every level of the data tree, that derived fields can be re-computed from raw inputs, and that the column positions we read from have not shifted. Coverage: ~93 quarters of delinquency and balance data (Q1 2003 to Q1 2026), plus age-band balances back to Q1 1999, across all loan segments (Mortgage, HELOC, Auto, Credit Card, Student, Other), all FICO bands, and all age bands.

Verdict (Q1 2026 release): all 14 checks pass. The 14 internal-consistency checks:

• Header-cell sanity - the cheap deterministic guard against the failure mode this suite was built for: a future NY Fed Excel layout change silently shifting a column. We assert each header cell contains the expected text; if any header drifts, this fails before any data is extracted.
• Page 11 row sum - proves we are correctly summing the delinquency-bucket breakdown to 100 percent. A miscounted column diverges from 100.
• Page 3 component rollup - proves segment balances reconcile to the total; any misaligned segment column shows up immediately.
• Page 12 vs Page 11 computed 90+ - the same metric (90+ DPD across all consumer debt) computed two ways (directly from Page 12, and as the sum of Page 11's 90, 120+, and Severely Derogatory buckets); they should agree closely.
• FICO band rollups (mortgage and auto) - origination volume by credit-score band sums to the published quarterly total.
• Age band vs Total Debt - reconciles Page 20's age-band breakdown to Page 3's total (NY Fed footnotes a small unknown-birthyear gap, allowed by tolerance).
• Range checks - every value sits within a plausible range.
• Sequence-gap check - no missing quarters in the time series.
• Credit-card and HELOC capacity identities - balance vs limit (utilization) identities hold for revolving products.
• Page 10 vs Page 3 (credit card and HELOC) - cross-checks the same segment balance reported on two different pages.
• Page 15 ordering - the mortgage transition-rate table rows are in the expected order.

Layer 2 - Cross-check vs the NY Fed narrative PDF

Layer 2 is independent cross-validation against NY Fed's own quarterly narrative PDF. The PDF and the Excel both come from the same underlying CCP/Equifax data but go through different editorial workflows: the Excel is mechanically generated from the panel, while the PDF narrative is hand-written by the Center for Microeconomic Data. Cross-checking the two catches column shifts, unit confusion, quarter mislabeling, and silent revisions.

Coverage: 50 quarterly PDFs from Q4 2013 (the earliest NY Fed publishes a PDF) through Q1 2026. For each quarter we extract up to 10 headline numbers from the narrative (total household debt, six segment balances, mortgage and auto originations, and the aggregate 30+ DPD rate) and compare to our Excel-derived value at the precision NY Fed used.

Verdict (Q1 2026 release): zero mismatches across 600 metric checks. Of the 343 metrics the PDF narrative actually stated, all agree with our Excel-derived values; the remaining 257 were not stated at a comparable level in the narrative.

Methodology and result categories

For each PDF we extract headline numbers using regex patterns anchored on NY Fed's specific phrasings (e.g. "Balances now stand at $X trillion", "Auto loan balances stood at $X billion"), constrained to single sentences so values from neighbouring sentences cannot leak in. Each value is rounded to the same decimals NY Fed used and compared to our Excel-derived value at the same precision. Three categories need explanation:

• Near-match - values differ by at most one unit at NY Fed's stated precision (e.g. NY Fed writes "$1.66 trillion" while our value rounds to $1.67T; the underlying difference is $0.01T, normal narrative-rounding variance). Treated as a pass.
• Methodology diff - NY Fed revised the metric definition since the PDF was published. The clearest example: 2013-2015 PDFs say "auto loan originations" (loans only) while modern PDFs say "auto loans and leases" - NY Fed expanded the definition around 2018. The Excel reflects the current definition for all history; older PDFs reflect the older one. A legitimate publisher revision, not a parsing error.
• PDF not stated - the narrative did not include a level for that metric that quarter (some quarters mention only the change, not the level). Not a failure - just nothing to compare.

Layer 3 - Why no independent cross-publisher exists

For our other consumer-credit sources (FDIC banks and NCUA credit unions) we run a cross-publisher Layer 3 check - FDIC aggregates reconciled to the FDIC QBP, and NCUA balances reconciled to the Fed Z.1 financial accounts. The NY Fed Consumer Credit Panel is different: its dollar aggregates are NY Fed's own scaled-up estimates from a 5% Equifax sample, and there is no other public statistical agency that publishes the same metric on the same definition.

• FRED's commercial-bank delinquency series (DRCCLACBS etc.) cover only loans held by commercial banks and exclude charged-off balances. NY Fed CCP covers all consumer credit (banks, credit unions, fintech, non-bank lenders) and includes charged-off balances still on credit reports. The two ask different questions and the levels differ structurally.
• The Federal Reserve Z.1 Financial Accounts publish household debt totals (CMDEBT), but from a flow-of-funds aggregation that includes debt not on credit reports (GSE-held mortgages, certain federal student loans). Levels differ by 5-10 percent; correlation is high but not exact.
• Equifax, Experian and TransUnion publish aggregate market insights, but those are derived from CCP-equivalent panels - not independent samples.

For these reasons the NY Fed CCP universe has no peer publisher with the same metric on the same definition. We rely instead on Layer 1 (internal consistency, including the column-shift defense) and Layer 2 (cross-check against NY Fed's editorially-independent narrative PDF).

Planned addition: Layer 1.5 silent-revision detection

Each new quarterly Excel includes the entire history, and NY Fed sometimes revises historical panel data quietly as the underlying Equifax sample updates. A planned "Layer 1.5" check will, going forward, compare each new Excel's pre-existing quarters to the prior Excel's same quarters and flag silent revisions. It requires retaining one prior file each quarter, so it begins producing findings from the next release onward.