FDIC Bank M&A, Charter & Structure-Change Monitor

A typed delta feed of every structure change in the US banking system, straight from the FDIC's own institution-history record: mergers paired acquirer-to-target and sized on both sides, de-novo charters, failures and assisted resolutions, branch openings and sales, relocations, regulator and class conversions, legal name changes.

Keyless, no login, no scraping. One mode, one date window, optionally a state list.

v1.1 adds deal-size financials — total assets, deposits, offices, holding company and RSSD for the acquirer, target, survivor and subject on every row — plus three live drift assertions that fail the run rather than hand you a plausible wrong answer.

---

Why this exists

api.fdic.gov/banks/history is public and free, and also close to unusable raw: 150 columns per record, 41 undocumented numeric change codes, acquirer/target linkage smeared across ACQ_*/OUT_*/SUR_* triples plus an underscore-packed ID, no financials at all, and an aggregation endpoint (agg_by) that returns HTTP 500.

Worse, it fails quietly. Two requests, verified live on 2026-07-28:

Request	meta.total	What FDIC did
PSTALP:TX AND CHANGECODE:223	1,457	applied both clauses
PSTALP:TX and CHANGECODE:223	52,138	echoed the filter back byte-for-byte, HTTP 200, and never applied the CHANGECODE clause

52,138 is exactly 35,239 (all TX) + 18,356 (all code 223) − 1,457 (the overlap). A filter FDIC declines to honour is not an error; it is a different, larger answer, and you would be billed for it. This actor asserts at runtime both that the echoed filter matches what was sent and that every returned row satisfies the constraints you asked for, and fails the run outright if either does not hold. Nothing is emitted and nothing is billed.

---

Who it is for

Persona	What they run
Bank M&A / corp-dev analyst	mode: "mergers", 365-day window, enrichFinancials: true — a deal book with both balance sheets, relative deal size, intrastate/interstate, and both holding companies.
Community-bank strategist	mode: "mergers" + states — who is consolidating your footprint, and how big the targets are.
Bank vendor / fintech BD team	mode: "denovo" — every new charter the moment FDIC records it. New banks buy core, compliance and card platforms.
Risk / credit analyst	mode: "failures" — failures, payoffs, conservatorships, open-bank assistance and assisted mergers, back to the 1930s if you widen the window.
Branch-network / site-selection analyst	mode: "branches" — openings, closings, purchases and sales with coordinates.
Compliance / KYB pipeline	mode: "events" — legal name changes, cert changes, charter-agency and primary-regulator moves. A bank that changed its name is not a new counterparty.
Market researcher / journalist	mode: "rollup" — consolidation counts by state, event type, quarter and acquirer.
AI agent	Any mode over MCP. Every field is described in the output schema, and every field says whether it was actually checked.

---

Example input

{
  "mode": "mergers",
  "sinceDays": 365,
  "dateField": "processed",
  "states": ["TX", "OK"],
  "maxResults": 500,
  "enrichFinancials": true
}

Every input carries a working prefill and a default. Running with the defaults returns a full year of US bank mergers, sized.

Input	Type	Notes
mode	string	events · mergers · denovo · failures · branches · rollup. Defaults to mergers.
sinceDays	integer	Delta anchor. Ignored when sinceDate is set. Default 365.
sinceDate / untilDate	YYYY-MM-DD	Explicit window. sinceDate overrides sinceDays.
dateField	processed | effective	processed = FDIC's PROCDATE, the right anchor for delta monitoring. effective = EFFDATE, the legal effective date.
states	array	Two-letter USPS codes. Empty = all states and territories.
changeCodes	array	Explicit FDIC CHANGECODE list. Overrides the set implied by mode.
maxResults	integer	Hard row cap. Billing is per result, so this is also your cost ceiling.
resolveMissingNames	boolean	Backfill a counterparty name from /banks/institutions when the history row omits it.
enrichFinancials	boolean	Deal-size lookup (below). Default on. Costs about one extra request per 250 certificates.

---

What comes back

One row per structure-change event, or one row per aggregate bucket in rollup mode. 184 declared fields; a mergers run populates 162 of them.

Identity and classification

Field	Notes
eventId	FDIC ID. On institution rows it decodes as TRANSNUM_CHANGECODE_UNINUM_ACQUNINUM_OUTUNINUM. Branch rows carry an opaque hash instead; the parser refuses to invent linkage from one.
transactionNumber	Shared by every record belonging to the same deal.
changeCode, changeCodeDescription	Raw FDIC code plus FDIC's own text.
eventType, eventCategory	Normalized snake_case type and coarse category, decoded from all 41 change codes present in the live index.
unmappedChangeCode	true if FDIC publishes a code this build does not decode. Also raises a run-level warning. It is never silently bucketed as "other".
effectiveDate, processedDate, eventQuarter, eventYear, processYear
isMerger · isFailure · isAssisted · isDeNovoCharter · isNewCharter · isRelocation · isRegulatorChange · isClassChange · isNameChange · isVoluntaryLiquidation · isBranchEvent · isParticipantMirror	Booleans, null-safe: false means FDIC published a 0; null means FDIC did not publish the flag at all.

The institution the event happened to

institutionName · cert · uninum · city · state · county · institutionClass · institutionClassDescription · charterAgency · regulator · organizationType · organizationTypeNumber · serviceType · trustPowers · insuranceFund · establishedDate · institutionEndDate · bankInsured · fdicRegion · latitude · longitude · reportType · fdicBankFindUrl

Counterparties

Flat columns for CSV and Sheets, plus the full nested acquirer / target / survivor objects for JSON consumers.

acquirerName · acquirerCert · acquirerUninum · acquirerCity · acquirerState · acquirerCounty · acquirerClass · acquirerClassDescription · acquirerCharterAgency · acquirerRegulator · acquirerAddress · acquirerZip · acquirerLatitude · acquirerLongitude · acquirerFdicRegion · acquirerBranchesAcquired — the same set for target* and survivor*, plus survivorChangeCode and survivorChangeCodeDescription (the mirror record on the surviving charter's timeline).

Field	Notes
linkageMethod	How we looked: acq_out_fields, id_uninum, institutions_lookup, none.
linkageStatus	Whether we found anything: resolved · partial · unresolved · not_applicable. A merger with no named side reports unresolved — deliberately not the same value as a de-novo charter, which genuinely has no counterparty.
survivorUninumSource	FDIC does not publish SUR_UNINUM on this endpoint. When SUR_CERT matches ACQ_CERT the survivor's UNINUM is taken from ACQ_UNINUM and this field says so; otherwise survivorUninum stays null rather than being guessed.

Deal size (enrichFinancials, new in v1.1)

/banks/history carries no financials. /banks/institutions retains the final call report of a bank that has already merged away — Stellar Bank, CERT 58629, charter ended 2026-07-01, still reports $10.89 B assets and $9.04 B deposits as of 2026-03-31 — so a completed deal can be sized on both sides.

For each of institution, acquirer, target, survivor: *AssetsUsdThousands · *DepositsUsdThousands · *Offices · *HoldingCompany · *HoldingCompanyRssd · *Specialization · *IsCommunityBank · *IsActive · *FinancialsAsOf · *CharterEndDate

Plus targetToAcquirerAssetsPct, combinedAssetsUsdThousands and dealGeography (intrastate / interstate).

Assets and deposits are US dollar THOUSANDS. *HoldingCompanyRssd is FDIC's RSSDHCR, the regulatory high-holder pointer — note that ULTCERT is not a parent pointer and is deliberately not used. A null financial figure means not retrieved, never zero; enrichmentStatus on the same row tells you which.

Before / after

former and current objects, plus changedAttributes — an explicit list of only the attributes that actually moved, e.g. [{"attribute":"regulator","from":"FED","to":"FDIC"}] — and activeFlags, the FDIC boolean event flags set to 1 on the record.

Provenance — every row says what was and was not checked

Field	Meaning
historySourceStatus	ok, or partial when a page failed after retries.
nameLookupStatus	not_needed · ok · unavailable · disabled. When unavailable, a blank acquirerName means "not resolved", not "no acquirer".
enrichmentStatus	ok · partial · unavailable · disabled · not_applicable.
resultCompleteness / partialResult	complete · capped_by_maxresults · partial_upstream_error. A partial run says so on the row, not only in the log.
filterEchoVerified	Always true on emitted rows — the run fails if it is not.
queryFilters	The exact FDIC filter expression sent, for audit against the echo.
fdicIndexName	The FDIC search index that served the run, e.g. history_20260724090007.
fdicIndexMaxProcessedDate	Newest PROCDATE anywhere in the index, from an unfiltered probe.
pagesRequested / pagesFailed / rowsAvailableUpstream	Fetch accounting. rowsAvailableUpstream is FDIC's meta.total for your query, before maxResults capping.
queryMode, windowStart, windowEnd, retrievedAt, sourceUrl

rollup mode

Buckets by state, eventType, quarter and acquirer (dimension + key), each with eventCount, distinctInstitutions, distinctAcquirers, distinctStates, firstEventDate, lastEventDate, and summed targetAssetsUsdThousands / acquirerAssetsUsdThousands when enrichment ran.

mergerCount, failureCount, deNovoCount, relocationCount, regulatorChangeCount, classChangeCount and otherCount are a single exclusive chain — each event lands in exactly one, in that order — and always sum to eventCount. assistedCount, branchCount, nameChangeCount, unmappedCodeCount and linkageUnresolvedCount are independent, non-exclusive counters that deliberately overlap the chain. Every bucket also carries the provenance fields above.

FDIC's agg_by returns HTTP 500 on this endpoint, so rollups are computed client-side from the rows actually fetched. eventsAnalyzed tells you how many that was, and rowsAvailableUpstream how many existed — if the second is larger, widen maxResults.

Note that firstEventDate / lastEventDate use the effective date while the default window filters on the processed date, so a bucket can legitimately reach back years: a relocation processed in 2026 may have taken effect in 2023.

---

Reliability: what fails the run instead of billing you

Three assertions run live, against the actual response, on every execution. An offline fixture cannot see a degraded upstream, because the fixture was shaped from it.

1. Upstream health probe. One unfiltered request. The national row count must sit inside a sane band (583,491 on 2026-07-28) and the newest PROCDATE must be within 60 days. A truncated or frozen index fails the run.
2. Echoed-filter assertion, on every page. meta.parameters.filters must equal what was sent, byte for byte — including the enrichment requests to /banks/institutions.
3. Row-level constraint check. Every returned row must carry a CHANGECODE in the requested set, a PSTALP in the requested states, and a date inside the window (1-day tolerance). This is the gate that catches the lowercase-and case above, where the echo matched and the query still had not been applied. A row with no PSTALP at all is counted as unverifiable and logged, not treated as a violation.

Beyond that: an invalid mode is rejected at run creation (HTTP 400, nothing consumed); an unreachable FDIC endpoint fails with the request URL and the last error; and if every page fails, the run fails rather than emitting an empty success. Zero rows is reported as a genuine empty result only when the health probe passed, the filter was echoed intact, and FDIC honestly returned nothing.

An entirely empty input is not an error — the schema defaults apply (mergers, 365 days), so {} returns a year of US bank mergers.

---

Use as an MCP tool

Available to AI agents through mcp.apify.com. Every input and output field carries a description, so an agent can pick a mode and read the result without extra prompting. The provenance fields matter here in particular: an agent can distinguish a verified negative (false) from an unchecked one (null), and will not report "no acquirer" when the lookup merely failed.

Useful agent phrasings: "which banks were acquired in Texas this year, and how big were they", "list new bank charters recorded since January", "has this bank changed its legal name or primary regulator", "summarise US bank consolidation by state for the last four quarters".

---

Pricing

$15 per 1,000 results, billed as the standard Apify Result event (apify-default-dataset-item) — one event per dataset item, which is one structure-change event in the event modes and one bucket in rollup. Graduated Apify plan discounts apply (Bronze −20 % through Diamond −70 %).

maxResults caps rows fetched and emitted, so it is also your cost ceiling. A failed run emits nothing and bills nothing. The health probe and the enrichment lookups are requests, not results — they are never billed.

---

FAQ

Where does the data come from? https://api.fdic.gov/banks/history for events and https://api.fdic.gov/banks/institutions for the deal-size enrichment. Both are public and keyless. Do not use banks.data.fdic.gov/api/* — it 302-redirects.

How far back does it go? To the 1930s. Set sinceDays large, or use sinceDate. failures with a wide window returns the full historical record of US bank failures.

Why does a merger row show the acquired bank as institutionName? Because that is what the FDIC record is about: the change happened to the institution that went away. The acquirer is in acquirerName, the surviving charter in survivorName.

FRM_INSTNAME / FRM_CERT are missing on merger rows — is that a bug? It is FDIC. Those columns are absent on 100 % of sampled merger-family rows, not merely null. The actor reconstructs the before-state from the OUT_* side and never zero-fills the gap.

Why are organizationType, serviceType, institutionOffices and targetOffices empty on my merger run? Because FDIC does not publish them for those rows. ORGTYPE and SERVTYPE_DESC are absent from merger-family history records — run mode: "events", or changeCodes: [360], and they populate. OFFICES is omitted for closed charters, and in a merger the target is the closed charter. Null means "FDIC does not have it here", not zero.

Why does the same deal appear twice if I pass changeCodes by hand? FDIC writes an 8xx "Participated in…" mirror on the acquirer's timeline sharing the deal's TRANSNUM. The institution-level modes exclude 810/811/812 for exactly this reason. If you request them explicitly they are returned and flagged isParticipantMirror: true.

Can I get quarterly totals without pulling every row? Not from FDIC — agg_by returns HTTP 500 on this endpoint. Use mode: "rollup", which aggregates client-side over the rows it fetched.

Is the run charged if FDIC is down, or answers with the wrong universe? No. Every failure path calls Actor.fail() before anything reaches the dataset, and the message ends with "Nothing was emitted and nothing was billed."

---

Related actors

Actor	Use it for
FDIC Branch Network Churn Rollup	Office-level entry and exit by bank and state, from a full chronological replay of state history — the branch-footprint counterpart to this actor's institution-level feed.
FDIC Deposit Market Share Rollup (SOD)	County and MSA deposit share and HHI. Run it after a merger found here to see what the combined entity's concentration looks like.
FDIC & NCUA Bank Health Rollup	Capital, asset quality and CRE-concentration screening on the banks in this feed — peer cohorts, the 2006 CRE guidance two-prong test, and credit unions.

A typical chain: this actor to find the deal → health rollup to underwrite both sides → SOD rollup to check the deposit overlap → branch churn to see which offices actually close.