Methodology
How We Rate Credit unions
The Paper Trail profiles 693 credit unions and scores them on version 1.7 of a written rubric; 666 have a published score. This page is the rubric as the rating program applies it: each factor and its weight, the evidence it uses, when a factor is left out, and the gate a score must pass. Most recent data: .
Credit union scoring v1.7 (Nov. 11, 2026)
The scoring program and the second program that recomputes every score on each rebuild were written by the same research team, so the second program checks the arithmetic, not the reading of the rules. The check of the written rules is a re-implementation by a different reviewer, who built the rubric from this page alone without opening either program. When run against version 1.7 of this rubric on Nov. 11, 2026, the re-implementation matched every factor score, overall score, confidence level and publication decision for all 693 credit unions. Version 1.7, the one on this page, is the version that re-implementation checked.
What weights does the rating program use?
These weights are read from the scored profiles themselves, so they match every score on the site (666 published). A factor without enough evidence for a firm is left out and the remaining weights are scaled up to 100%.
| Factor | Weight |
|---|---|
| Capital strength | 35% |
| Asset quality | 29% |
| NCUA administrative orders | 24% |
| Earnings | 12% |
Scope: every NCUA call-report credit union headquartered in the 13 West states (693 for June 2026). A program applies the written rules below, and a second program recomputes every score.
Which factors are scored, and how much does each weigh?
| Factor | Key | Weight | Inputs (NCUA 5300 bulk file, 2026-06) | Outside record? |
|---|---|---|---|---|
| Capital strength | capital | 35% | Net worth ratio (NCUA call report account 998), NCUA net worth classification (NCUA call report account 700) | no (regulator-published call report) |
| Asset quality | asset quality | 29% | Delinquent loans (2+ months) / loans; annualized net charge-offs / loans | no |
| Earnings | earnings | 12% | Annualized return on assets | no |
| NCUA administrative orders | actions | 24% | Institution-level orders in NCUA's administrative-orders list | yes |
Consumer complaints are a displayed, unscored fact (v1.2). The CFPB supervises, and publishes complaints mainly about, institutions with more than $10 billion in assets; 686 of the 693 West credit unions are smaller, so their absence from the CFPB database is not evidence of anything, and the 7 matched credit unions cannot anchor a peer scale. Complaint counts, timeliness and relief are shown for the matched credit unions only, labeled "not scored"; a credit union with no company in the database shows no count.
Each factor 0 to 10; a factor with no usable data is N/A and the remaining weights are renormalized; the page names the factors scored. Missing data is never treated as good or bad. Each factor score is published to one decimal (rounded half-up, v1.5) and the overall is computed from those published scores: weighted mean of the scored factors x 10, rounded once, half-up. Pending matters and allegations are shown, never scored; orders against individuals never attach to the credit union.
Which thresholds are used, and where do they come from?
How is capital strength scored (35% of the score)?
- Net worth ratio r (percent): well capitalized at 7% and adequately capitalized at 6% are the statutory lines (12 U.S.C. 1790d; 12 CFR 702.102). r < 6 scores 0; 6 <= r < 7 scores 2 + 3 x (r - 6); r >= 7 scores 5 + 5 x the smaller of 1 and (r - 7)/(P90 - 7). West P90 of the ratio (Nov. 9, 2026) = 18.95%.
- NCUA's own classification caps the score (it also applies the risk-based net worth requirement to complex credit unions): Adequately Capitalized max 4.9; Undercapitalized max 2; Significantly Undercapitalized max 1; Critically Undercapitalized 0.
- In the 2026-06 data 685 credit unions are Well Capitalized, 6 Adequately, 1 Undercapitalized, 1 Significantly Undercapitalized.
How is asset quality scored (29% of the score)?
- Delinquency rate (60%) scaled P90 1.90% (0) to P10 0.04% (10); annualized net charge-offs / loans (40%), (charge-offs - recoveries) x 12/months / loans, scaled P90 1.07% (0) to P10 0.00% (10). Both are required, else N/A (2 credit unions lack a delinquency rate).
How is earnings scored (12% of the score)?
- Annualized return on assets = net income YTD x 12/months / period-end assets, scaled P10 -0.10% (0) to P90 1.54% (10).
How is nCUA administrative orders scored (24% of the score)?
- Source: NCUA's administrative-orders CSV (1991 to date; the page says it is not guaranteed comprehensive), 1,445 rows, 405 in the West. Matched to credit unions by name and state. Orders against individuals are never scored.
- Usable for federally insured credit unions (671 of 693); the 22 state-supervised, non-NCUA-insured credit unions are N/A (state regulators' actions are not in this list). As scored on Nov. 9, 2026, the factor is scored for 669 of the 692 active credit unions; the other 23 are outside the list (including the 22) or have a matched row whose respondent could not be confirmed.
- Score = 10 - 3 for each institution-level order (cease and desist; assessment of civil money penalty) in the last 5 years - 1.5 for each 5 to 10 years ago - 0.5 for each older; floor 0. A matched row that cannot be classified makes the factor N/A.
- "Complete" means every West row of NCUA's published list was read, not that no other action exists.
Why are consumer complaints shown but not scored?
- Source: CFPB Consumer Complaint Database, 36 months to Nov. 9, 2026, matched by name, with the 40% home-state test. Matched (7 credit unions): the count, the share answered on time, the relief count and the rate per 100,000 members a year are shown, not scored. None-in-database (328) and every other status: no count is shown; the page states what was searched and that the CFPB's database does not cover credit unions of $10 billion or less, so an empty search is not evidence of zero complaints.
- Until v1.1 this was a scored factor (weight fifteen of 100) that gave 10 of 10 to the 328 credit unions absent from the database and scored the 7 matched ones against a cutoff built from 7 rates. v1.2 removed it (change log).
How is confidence set?
Points: +1 for each outside-record factor scored (only the NCUA orders factor remains, so at most 1); +1 if capital, asset quality and earnings are all scored; +1 if independently verified. High 4, Medium 3, Low 2 or fewer. With one outside-record source a credit union reaches Medium at most (v1.2): High needs a second independent outside record (for example state regulators' actions), which is not collected.
How much of the rubric is scored for each firm?
Every firm file carries share of the rubric's weight that was scored (the share of the rubric's weight that was actually scored) and the factors left unscored, each with its reason (each N/A factor with its weight and the reason). When a factor of 25% weight or more is N/A, or less than 75% of the weight is scored, confidence is at most Medium and the file carries a plain note on confidence; under 60% the rating is held ("not enough of the rubric could be scored"). Measured on Nov. 11, 2026: all 666 published credit-union ratings have 100% of the weight scored; the rule holds none and changes no confidence label (credit unions are Medium at most since v1.2). Two held files carry a note on confidence.
How are firms ranked when data is missing?
Every firm file carries eligibility for ranking, ranking note, percentile within its type and rank within its type (re-derived by the audit). A firm is rank-eligible when it has a published rating, at least 85% of the rubric's weight is scored, and it does not reach the top tenth only because of a data gap. Percentiles and ranks are computed over rank-eligible firms only; other firms carry a plain ranking note and no percentile. Data-gap test: each rated firm's overall recomputed with every N/A factor at the type median of that factor instead of renormalising. Measured Nov. 11, 2026: 666 rated, 0 with a gap, rank correlation 1.0, 0 firms move 3 points or more (largest change 0); no firm reaches the top tenth only because of a data gap; 666 firms are rank-eligible.
When is a score published?
A rating is published only when ALL hold: (1) at least 3 of the 4 factors scored; (2) the outside-record factor (NCUA orders) scored; (3) confidence Medium or High; (4) the firm's facts independently verified with no open mismatch; (4a) low scores double-confirmed (CF1: a firm under 40 or classed below Adequately Capitalized needs a recorded double confirmation to pass); (5) no unresolved enforcement lead or CFPB enforcement action. The whole universe is scored and was independently verified on Nov. 9, 2026 (a separate reviewer's parse of every field plus a blind 5% manual sample); a firm publishes only when its own verification status is "verified" and the gate passes.
How sensitive are the scores to the weights?
Each weight moved by 5 points (others unchanged, scored weights renormalised per firm), overall recomputed for the 692 active credit unions:
| Factor | Shift | Spearman vs baseline | Median change | Max change | Firms moving 3+ points | Top 10% kept |
|---|---|---|---|---|---|---|
| capital | +5 | 0.998 | 0 | 3 | 1 | 96% |
| capital | -5 | 0.998 | 0 | 2 | 0 | 97% |
| asset quality | +5 | 0.997 | 1 | 3 | 19 | 97% |
| asset quality | -5 | 0.996 | 1 | 4 | 38 | 94% |
| earnings | +5 | 0.994 | 1 | 4 | 128 | 91% |
| earnings | -5 | 0.993 | 1 | 5 | 150 | 88% |
| actions | +5 | 0.999 | 1 | 3 | 12 | 96% |
| actions | -5 | 0.999 | 1 | 3 | 43 | 97% |
No shift changes any overall by more than 5 points; the ranking stays above 0.993 rank correlation. Earnings, the smallest weight, moves the most firms because its scores are the most spread.
What fairness checks apply?
- Credit unions serve defined memberships; ranks compare across fields of membership without adjustment, and the page says so.
- Period-end denominators and simple annualization make some ratios differ slightly from NCUA's published values; every credit union is computed the same way.
- Not collected: fee schedules (no common format), CRA (credit unions have none).
How complete are the fields on these profiles?
For each field, how many of the 693 credit union profiles carry a value; how many have the source read and the field blank or not reported there ("not disclosed in the source"); and how many have no such fact at all ("not collected").
Field completeness, 693 credit union profiles
| Field | Present | Not disclosed in the source | Not collected |
|---|---|---|---|
| Name | 693 (100.0%) | 0 (0.0%) | 0 (0.0%) |
| City | 693 (100.0%) | 0 (0.0%) | 0 (0.0%) |
| State | 693 (100.0%) | 0 (0.0%) | 0 (0.0%) |
| NCUA charter | 693 (100.0%) | 0 (0.0%) | 0 (0.0%) |
| Total assets | 693 (100.0%) | 0 (0.0%) | 0 (0.0%) |
| Members | 693 (100.0%) | 0 (0.0%) | 0 (0.0%) |
| Net worth ratio | 693 (100.0%) | 0 (0.0%) | 0 (0.0%) |
| Delinquency rate | 691 (99.7%) | 2 (0.3%) | 0 (0.0%) |
| Charge-off rate | 691 (99.7%) | 2 (0.3%) | 0 (0.0%) |
| NCUA order search | 693 (100.0%) | 0 (0.0%) | 0 (0.0%) |
What changed in each version?
- Data note, Nov. 11, 2026: rubric stays v1.7.
- Version 1.7, Nov. 11, 2026: peer anchors follow the written rule. Effect: 204 factor scores change (asset quality 163, earnings 32, capital 9), 56 overalls change by one point (55 down, 1 up); no publication, confidence or rank-eligibility change.
- Version 1.6, Nov. 11, 2026: identity rule. Effect: 26 overalls change; 21 held credit unions publish, none loses publication (their orders factor had been N/A because of an unread row); 5 lose points for a confirmed old civil money penalty.
- Version 1.5, Nov. 10, 2026: one rounding rule for every type. Each factor score is rounded half-up to one decimal and the overall is computed from those one-decimal scores, so a reader (and the site) recomputes the overall exactly from the factor scores shown. Effect: 76 overalls move by one point (range -1 to +1); no publication, confidence or rank-eligibility change.
- Version 1.4, Nov. 10, 2026. Ranking and data gaps; all 645 credit unions rated at the time were rank-eligible when v1.4 came in (Nov. 10, 2026); 666 of 666 are on Nov. 11, 2026.
- Version 1.3, Nov. 10, 2026. Coverage rule above; no published rating or confidence label changes.
- Version 1.2, Nov. 10, 2026. Consumer complaints are a displayed, unscored fact, mirroring banks v1.1: the CFPB publishes complaints mainly for institutions over $10 billion in assets, so absence from its database is not evidence, and 7 matched credit unions cannot anchor a scale. Effect on the June 2026 universe: 372 of 693 overall scores change (median 0, range -11 to +13; San Diego County CU 77 to 90, it had scored 0.0 on complaints at a rate 0.004 above the 7-point cutoff).
- Version 1.1, Nov. 9, 2026.
- Version 1.1 addendum, Nov. 9, 2026: institution status.
- Version 1.0, Nov. 9, 2026. First rubric.
This page is the written rubric the rating program follows, with file names and internal codes put in words. Individuals are never rated, pending matters and allegations are shown but never scored, and missing data is never counted as good or bad. Information, not investment advice.