Skip to content
FinObservatory

Banks / Backtest

Did the composite see failures coming?

Every FDIC-resolved failure since 2016 (32 banks), against the FinObservatory composite the bank carried at its last call report before failing. This is the out-of-sample record, published in full: the median pre-failure composite is 20.7 against a panel-wide median of 50.5, so the score does discriminate; and 8 of the 32 failures scored 35 or better at their last report, so it also misses, and the misses are named below.

The base rate, first

A low score is common; failure is rare. Of the 1,255 bank-quarters below composite 20 between 2016 and 2024Q1 (every one with a full two-year window inside the panel), 76 were followed by that bank's failure within two years: 6.1%. Read the other way, 93.9% of flagged bank-quarters were NOT followed by failure. 384 distinct banks have spent at least one quarter below 20 since 2016; 21 of them later failed. A screen this size cannot be read as a failure prediction, and this page never presents it as one.

The misses, named

8 failures scored 35 or better at their last report. The largest is Silicon Valley Bank (composite 63.5 at 2022Q4, ten weeks before it failed): its held-to-maturity losses and uninsured-deposit concentration are invisible in the score's public call-report fields, a limit documented on the banks methodology page since the score shipped. The others are in the table. This page reports the scores and does not assign failure causes.

All 32 failures since 2016

Composite at the bank’s last call report before failure; the gap column is the days between that report and the failure date.

BankFailedLast reportCompositeGap (days)
NORTH MILWAUKEE STATE BANK2016-03-112015-12-3112.171
TRUST COMPANY BANK2016-04-292016-03-3111.029
FIRST CORNERSTONE BANK2016-05-062016-03-3114.636
THE WOODBURY BANKING COMPANY2016-08-192016-06-3018.050
ALLIED BANK2016-09-232016-06-3011.185
HARVEST COMMUNITY BANK2017-01-132016-12-3110.213
SEAWAY BANK AND TRUST COMPANY2017-01-272016-12-318.627
PROFICIO BANK2017-03-032016-12-3113.862
FIRST NBC BANK2017-04-282017-03-3123.628
GUARANTY BANK2017-05-052017-03-3121.335
FAYETTE COUNTY BANK2017-05-262017-03-3115.156
THE FARMERS AND MERCHANTS STATE BANK OF ARGONIA2017-10-132017-09-3014.413
WASHINGTON FEDERAL BANK FOR SAVINGS2017-12-152017-09-3074.576
ENLOE STATE BANK2019-05-312019-03-3152.761
LOUISA COMMUNITY BANK2019-10-252019-09-304.925
RESOLUTE BANK2019-10-252019-09-3018.325
CITY NATIONAL BANK OF NEW JERSEY2019-11-012019-09-3023.632
ERICSON STATE BANK2020-02-142019-12-3119.545
THE FIRST STATE BANK2020-04-032020-03-3120.03
FIRST CITY BANK OF FLORIDA2020-10-162020-09-3026.816
ALMENA STATE BANK2020-10-232020-09-3012.123
SILICON VALLEY BANK2023-03-102022-12-3163.569
SIGNATURE BANK2023-03-122022-12-3142.771
FIRST REPUBLIC BANK2023-05-012023-03-3136.731
HEARTLAND TRI-STATE BANK2023-07-282023-03-3148.1119
CITIZENS BANK2023-11-032023-09-3029.634
REPUBLIC BANK2024-04-262023-12-3125.7117
FIRST NB OF LINDSAY2024-10-182024-09-3018.918
PULASKI SAVINGS BANK2025-01-172024-09-3051.3109
SANTA ANNA NATIONAL BANK2025-06-272025-03-3140.888
METROPOLITAN CAPITAL B&T2026-01-302025-12-3125.430
COMMUNITY BANK AND TRUST - WEST GEORGIA2026-05-012025-12-3127.3121

Green: below the 20 reporting threshold at the last report (15 of 32). Red: 35 or better (the misses). The threshold is a reporting convention for this page, not a model parameter; the panel's 10th percentile sits near 34 and its median near 50.5.

Method

  • Failures: FDIC Failures and Assistance Transactions, RESTYPE=FAILURE, 2016 onward. Scores: the same quarterly composite panel the bank screener serves (2015Q1 onward), so every number here is recomputable from the site’s own data.
  • “Last report” is the latest call-report quarter strictly before the failure date; every failure since 2016 has one.
  • The false-positive window ends at 2024Q1 so each flagged quarter has a full two-year horizon inside the panel; flags after that date are excluded from the base rate, not resolved optimistically.
  • Formula and disclaimers: banks methodology. The composite is a transparent proxy from public filings; it is not CAMELS, and this backtest is descriptive, not a prediction model.