Methodology
FINASENSE research is built on publicly available NCUA 5300 Call Report data — the same quarterly financial filings that every federally insured credit union submits to the National Credit Union Administration. This page documents the data sources, analytical choices, proprietary metrics, and known limitations behind our research.
Data Source
NCUA 5300 Call Report
Quarterly (March, June, September, December)
All federally insured credit unions ( 6,080 as of latest quarter)
March 2016 – December 2025 ( 39 quarters)
The Call Report is a mandatory supervisory filing. Every federally insured credit union — regardless of size, charter type, or geography — submits the same form on the same schedule. This universality is what makes the dataset analytically valuable: there is no selection bias, no opt-in, no voluntary disclosure. The population is the sample.
FINASENSE ingests the raw Call Report archives published by the NCUA, maps account codes through temporal transitions (the NCUA periodically adds, removes, and renumbers fields), and produces a normalized analytical database spanning the full history.
Interpretation Principles
These principles govern how FINASENSE interprets data and frames analysis. They are not aspirational — they are the rules the editorial process enforces.
Data over narrative
If the numbers say one thing and the conventional wisdom says another, we go with the numbers. Every directional claim in a FINASENSE report — "delinquency rose," "margins compressed," "capital improved" — is backed by a specific computed value. Claims without numbers are not published.
Direction matters more than level
A system with 1.00% delinquency that is falling is in better shape than a system at 0.60% that is rising. FINASENSE analysis emphasizes velocity (quarter-over-quarter change) and acceleration (whether the rate of change is itself increasing or decreasing), not just the current reading. The CCPI is built entirely on this principle. Level provides context; direction provides signal.
Aggregates mask dispersion
System-wide ratios are computed from aggregated dollar totals — not by averaging institution-level ratios. This correctly weights larger institutions, but it means that a $200B credit union moving 5 bps can offset thousands of small CUs moving in the opposite direction. To address this, FINASENSE breaks out every key metric by five asset-size cohorts (see below) and flags cases where the system aggregate diverges materially from the cohort story.
Conflicting signals are the norm
A healthy credit union system does not produce uniform signals across every CAMELS dimension. Capital can strengthen while credit quality deteriorates (as it did in 2024-2025, when earnings retention rebuilt buffers even as delinquency hit dataset highs). FINASENSE reports each dimension independently and lets the reader weigh the tensions, rather than collapsing everything into a single "good" or "bad" verdict. The CUFSI provides a composite reading for those who want one, but it is presented alongside its pillar-level decomposition.
No forecasts
FINASENSE describes what the data shows and identifies the trajectory. We do not project future values, predict regulatory actions, or model hypothetical scenarios. When we write "if this trend continues," we are describing the arithmetic, not making a prediction.
Analytical Framework
CAMELS Alignment
All FINASENSE analysis is organized under the CAMELS framework — Capital adequacy, Asset quality, Management, Earnings, Liquidity, and Sensitivity to market risk — the same structure used by NCUA examiners. This is a deliberate choice: it ensures that FINASENSE output maps directly to the categories that boards, examiners, and ALM teams already think in.
Aggregation Method
System-wide ratios are always computed from aggregated dollar totals, never by averaging institution-level ratios. For example:
This correctly reflects the economic reality that a $50B credit union contributes more to system-wide capital adequacy than a $5M credit union. The tbl_ratios table in the source data contains pre-computed CU-level ratios, but FINASENSE does not average these for system-wide figures.
Annualization
Income-statement items in the Call Report are cumulative year-to-date. FINASENSE annualizes using the standard factor:
All annualized figures are labeled as such. Q1 annualization amplifies single-quarter noise by a factor of four; we flag this explicitly when a Q1 metric shows an unusual move and the adjacent quarters do not confirm the trend.
Cohort Segmentation
System-wide aggregates can mask meaningful variation by institution size. FINASENSE breaks out key metrics by five asset-size cohorts:
Cohort assignment is based on total assets as of the reporting quarter. A credit union that crosses a threshold (e.g., from $90M to $110M) moves to the new cohort in that quarter. Cohort boundaries are fixed and do not adjust for inflation.
QoQ vs. YoY
FINASENSE uses quarter-over-quarter (QoQ) change as the primary lens for trend analysis. QoQ captures turns faster than year-over-year (YoY) comparisons, which smooth out short-term movements and can mask inflection points. YoY is used as a secondary reference — particularly for income-statement items, where it eliminates the annualization artifacts that make QoQ comparisons of Q1 figures noisy.
Signal Language
FINASENSE uses three explicit signal labels for trend assessments:
Metric performing well; trend is positive or stable at healthy levels
Trend deviating from norms; may deteriorate if uncorrected
Active deterioration; metric at or beyond stress thresholds
Signals are editorial assessments informed by the data, not mechanically derived thresholds. A metric can be at a historically high level and still receive a "Favorable" signal if the trend is stable or improving. Conversely, a metric at a normal level can receive "Watch" if it is deteriorating rapidly. These are analytical assessments, not investment recommendations.
Proprietary Metrics
Credit Cycle Position Indicator™ (CCPI™)
Classifies the industry's position in the credit cycle based on velocity and acceleration of credit-sensitive metrics.
Components:
| Component | Weight | What it measures |
|---|---|---|
| Delinquency (60+ days) | 30% | Borrower distress — the leading indicator |
| Net charge-offs | 25% | Realized credit losses |
| Provision expense | 20% | Management's forward-looking loss estimate |
| Loan growth | 25% | Lending momentum (inverted: declining growth = worsening) |
How it works: For each metric, the CCPI computes the quarter-over-quarter velocity (first derivative) and acceleration (second derivative), z-scores both against the metric's own expanding history, and blends them (70% velocity / 30% acceleration) into a component score on a [-1, +1] scale. The composite CCPI is the weighted average of the four components.
Phase thresholds:
| Phase | CCPI Range | Signal |
|---|---|---|
| Expansion | Below -0.25 | Credit quality improving at an above-average pace |
| Late Cycle | -0.25 to 0.09 | Mixed signals; cycle may be turning |
| Contraction | 0.10 to 0.49 | Credit quality deteriorating at an above-average pace |
| Stress | 0.50 and above | Acute deterioration across multiple components |
The CCPI measures direction and momentum, not absolute level. A system with high delinquency but stable (non-rising) rates would score near zero — it's the change that matters.
Credit Union Financial Stress Index™ (CUFSI™)
A composite 0–100 index measuring system-wide financial stress across six pillars aligned with the CAMELS examination framework.
Pillars:
| Pillar | Weight | Variables |
|---|---|---|
| Credit Quality | 30% | Delinquency ratio, delinquency severity, NCO ratio, non-accrual ratio |
| Capital Adequacy | 20% | Net worth ratio, allowance coverage, allowance-to-loans |
| Earnings Coverage | 15% | ROAA, NIM, provision ratio, operating expense ratio |
| Liquidity | 15% | Borrowing reliance, loans-to-shares, cash buffer |
| Concentration Risk | 10% | Commercial concentration, indirect concentration, participation concentration |
| Growth Stress | 10% | Loan growth, growth-capital divergence |
How it works: Each of the 19 input variables is z-scored against its expanding history, oriented so that positive = more stress, averaged within its pillar, then combined into a weighted composite. The composite z-score is mapped to 0–100 via hyperbolic tangent:
Severity bands:
| Score | Severity | Interpretation |
|---|---|---|
| 0–30 | Low Stress | Unusually benign conditions |
| 30–45 | Below Average | Mildly favorable |
| 45–55 | Normal | Consistent with historical norms |
| 55–70 | Elevated | Multiple indicators above norms |
| 70–100 | High Stress | Acute system-wide strain |
A score of 50 represents the historical average. The CUFSI answers "how stressed is the system right now?" while the CCPI answers "which direction is the cycle moving?"
Data Limitations
Rigorous analysis requires acknowledging what the data does not capture. Users of FINASENSE research should be aware of the following:
Self-reported data
Call Report data is filed by credit unions themselves. The NCUA examines and audits these filings, but errors, restatements, and late corrections do occur. FINASENSE uses the data as published and does not attempt to adjust for suspected reporting errors. In system-wide aggregates, individual CU-level errors tend to wash out; in CU-specific analysis (where offered), they do not.
Account code transitions
The NCUA periodically adds, removes, and renumbers Call Report fields. A given financial concept (e.g., "All Other Investments") may map to different account codes depending on the reporting date. FINASENSE maintains a temporal mapping layer that resolves these transitions programmatically — the same field may be read from account code A766E before Q1 2022 and from AAS0016 afterward. This is the core complexity of the data pipeline and the primary reason raw Call Report data is difficult to use in time-series analysis.
Survivorship bias
When a credit union merges, liquidates, or loses its federal insurance, it drops from subsequent quarters. The dataset does not backfill or adjust historical figures. This means that multi-year comparisons reflect only institutions that survived to the later date. For system-wide aggregates, the effect is modest (mergers typically transfer assets and members). For cohort-level analysis — particularly the "Under $100M" tier, where merger and consolidation rates are highest — it is more significant.
Publication lag
Credit unions file Call Reports within 30 days of quarter-end. The NCUA publishes the aggregated data approximately 60–75 days after the filing deadline. FINASENSE publishes research as soon as the data is available, but the most recent quarter in any report is inherently backward-looking by roughly one quarter.
Q1 annualization noise
Income-statement items reset to zero each January. Q1 figures represent a single quarter of activity, annualized by multiplying by four. This amplifies noise: a one-time fee income spike in January becomes a 4x-inflated annualized figure. FINASENSE flags Q1 annualization effects in editorial when they produce outlier readings, and uses YoY comparisons as a cross-check.
What is not in the data
The Call Report does not capture interest rate risk modeling outputs, internal stress test results, or board governance assessments. CAMELS "M" (Management) and "S" (Sensitivity) dimensions receive less quantitative coverage than "C," "A," "E," and "L." FINASENSE acknowledges this asymmetry rather than constructing proxies with false precision.
Update and Revision Policy
Publication cadence. FINASENSE publishes the Quarterly Pulse and associated Insights articles within two weeks of the NCUA's Call Report data release, typically in March, June, September, and December.
Restatements. If the NCUA issues a material correction to a prior quarter's data, FINASENSE will update the underlying database in the next publication cycle. We do not retroactively revise published reports — instead, the correction is noted in the subsequent quarter's editorial. Historical charts and tables always reflect the latest database state.
Methodology changes. Changes to proprietary metric definitions (e.g., CCPI component weights, CUFSI pillar composition) are documented at the time of change and noted in the affected publication. Historical values are recomputed under the new methodology to maintain comparability. We do not maintain parallel series under old and new definitions.