FEMA NFIP Redacted Claims v2 (OpenFEMA)

FEMA NFIP Redacted Claims v2 is the loss ledger of America's flood insurer of last resort, published through FEMA's OpenFEMA program. Every row is one claim transaction on an NFIP policy: when the loss happened (dateOfLoss, yearOfLoss), what zone the property was rated in (ratedFloodZone, floodZoneCurrent), how the building is occupied (occupancyType), how badly it was damaged (buildingDamageAmount, contentsDamageAmount), and what the program actually paid - gross amountPaidOnBuildingClaim sitting beside net netBuildingPaymentAmount, contents versions of both, and the Increased Cost of Compliance column that pays to elevate, demolish or floodproof after a loss.

Around the money sits everything a severity analyst asks for next. Physical readings: waterDepth, floodWaterDuration, baseFloodElevation, lowestFloorElevation, elevationDifference. Construction facts: originalConstructionDate, the post-FIRM indicator, elevated-building and floodproofed flags, floors and units. Occupant flags: primary residence, small business, house of worship, agriculture, non-profit, rental, state-owned. Community context: crsClassificationCode and nfipCommunityName. Storm attribution: causeOfDamage, the named floodEvent, and an eventDesignationNumber tying claims to federal disaster declarations. Geography arrives nested and deliberately blurred - state, reported ZIP, county FIPS, census tract, census block group, then rounded coordinates no row can be resolved to a household through. All 72 fields are documented at verified confidence.

Within Datadory's catalog of 1,744 datasets across 159 viable industries, this record anchors the realized-loss end of the Property & Casualty Insurance slice - sixteen primary records pooled there, nineteen counting secondary tags - and scores 9/10. Get a sample of this dataset to see the full dictionary mapped against your territories.

One row, one claim, one receipt. The verified record below comes from the December Storm - Nor'easter of 1992, on the New Jersey coast:

# one row = one claim transaction; the record below is verified, not synthetic

id                              : <uuid>
dateOfLoss                      : 1992-12-11
yearOfLoss                      : 1992
ratedFloodZone                  : A07
occupancyType                   : 1
amountPaidOnBuildingClaim       : 7,243.04
amountPaidOnContentsClaim       : 3,000.00
totalBuildingInsuranceCoverage  : 59,700
totalContentsInsuranceCoverage  : 8,800
state                           : NJ
reportedZipCode                 : 07732
countyCode                      : 34025
censusTract                     : 34025800100
latitude                        : 40.4
longitude                       : -74.0
floodEvent                      : December Storm - Nor'easter

# column families that ride on every row (shape shown; values ship with your sample):

# money
buildingDamageAmount            : <usd>     contentsDamageAmount : <usd>
netBuildingPaymentAmount        : <usd>     netContentsPaymentAmount : <usd>
netIccPaymentAmount             : <usd>     iccCoverage : <usd>
buildingDeductibleCode          : <code>    contentsDeductibleCode : <code>
buildingReplacementCost         : <usd>     contentsReplacementCost : <usd>
nonPaymentReasonBuilding        : <coded reason>

# physics
waterDepth                      : <depth>   floodWaterDuration : <duration>
baseFloodElevation              : <ft>      lowestFloorElevation : <ft>
elevationDifference             : <ft>      lowestAdjacentGrade : <ft>

# property and occupant flags
originalConstructionDate        : <date>    numberOfFloorsInTheInsuredBuilding : <n>
postFIRMConstructionIndicator   : <bool>    elevatedBuildingIndicator : <bool>
primaryResidenceIndicator       : <bool>    smallBusinessIndicatorBuilding : <bool>

# community and storm attribution
crsClassificationCode           : <class>   nfipCommunityName : <community>
causeOfDamage                   : <coded cause>   eventDesignationNumber : <declaration no.>

That single row already answers three questions analysts usually need a data team for. The program paid $7,243.04 on the building and $3,000 on contents against $59,700 and $8,800 of carried coverage - a building payout of roughly twelve cents on the dollar, computable from adjacent columns with zero enrichment. The geography nests from state down to census tract (34025800100), so the loss aggregates cleanly to any jurisdiction between. And the storm is named - the row attributes itself to an event rather than waiting to be clustered by date.

Multiply that anatomy by 2,721,780 rows spanning 1970 to the present and you get the dataset's real shape: every claim carries the same identifier-plus-measure-plus-context skeleton, so event studies, severity curves and ZIP-level loss runs assemble without reshaping. The placeholder families in the lower block - money, physics, property flags, community and attribution - ride on every row; their values arrive confirmed in your sample rather than invented here.

Six groups cover the 72 documented fields. Money: damage amounts, gross payments, net payments and ICC payments for building and contents, deductible codes, coverage carried, ICC coverage, replacement costs and assessed values, and the coded non-payment reasons that explain zeroes. Physics: water depth, flood duration, base flood elevation, lowest floor and lowest adjacent grade, elevation difference, obstruction and basement-enclosure types. Property: construction and map dates, the post-FIRM indicator, floors, units, and the occupancy, rental and use flags. Community: CRS classification, FICO and NFIP community numbers past and current, community name. Attribution: cause of damage, named flood event, disaster designation number. Geography: state, reported ZIP, county FIPS, census tract, census block group, rounded latitude and longitude.

The dictionary above lists the load-bearing core at verified confidence; the remaining columns fold under additional fields on request and are confirmed during sample preparation rather than asserted here.

Geography - the United States and territories wherever NFIP policies are written. Location lands nested and redacted: two-letter state, reported ZIP code, county FIPS, census tract, census block group FIPS, then latitude and longitude rounded so the point cannot resolve to a parcel. Reported city reads Currently Unavailable in recent vintages, which is one more reason the census geography is the join you want - stable, complete and built for aggregation.

Temporal - loss dates run from 1970-08-31 forward, five decades and change of American flooding: pre-map-era buildings, the 1992 nor'easters, the Katrina era, Sandy and everything since. yearOfLoss supports annual panels without touching the exact date. Two lifecycle facts belong in any model built on this record: FEMA has published a successor v3 release and retires v2 after October 15, 2026, and the v2 extract itself is preserved at its final June 2026 vintage.

Granularity - one row per claim transaction, 2,721,780 of them. A property that flooded twice appears twice, which is precisely what repeat-loss analysis requires. Because FEMA built the policies extract from the same administration system, 42 of these 72 fields also appear on the NFIP Redacted Policies v2 record - flood zones, coverage carried, deductibles, the elevation complex, census geography - so claims join to the exposed book they landed on without fuzzy matching.

API, files, or your warehouse. Daily, weekly, or hourly.

Your cadence is decoupled from any upstream publication rhythm. Take the full claim panel once as files, or keep a warehouse table current so new rows diff cleanly into yesterday's - either way the schema is the one documented above, typed and stable across deliveries.

Census geography travels with every row as the join spine, so claims attach to your own portfolio, territory or peril references on codes rather than names. Deliveries arrive with the field dictionary and validation rows attached; samples precede any commitment, and the schema you validate in the sample is the schema you ship against.

Half a century of realized federal losses earns its keep in five jobs:

• Severity and payout modelling - insurers, reinsurers and cat analysts fit empirical severity curves and payout ratios from the government's own claim receipts, separating gross from net and paid from closed-without-payment via the coded non-payment reasons.
• Portfolio flood-exposure screening - lenders and credit teams score mortgage and asset books by ZIP, tract and county loss history before exams, deals or renewals; the workflow continues on our credit risk screening page.
• Model training - data scientists train severity and outcome models on labeled rows where physical predictors (waterDepth, elevationDifference) ride beside financial targets; the pattern lives on our ml model training page.
• Catastrophe strategy backtests - quants group losses into events using named floodEvent and eventDesignationNumber instead of inferring clusters from dates; see quant backtesting.
• Journalism and academic research - writers and researchers cite claim-level receipts behind disaster narratives rather than press-release aggregates; the practice continues on citation-grade research.

Data scientists get a labeled, fixed-schema panel - outcomes, non-payment reasons and physical predictors on one spine - that drops into any severity pipeline without cleanup; see property casualty insurance data for data scientists. Investors and quants get five decades of event-attributed losses for stressing catastrophe-linked positions; see property casualty insurance data for investors quants. Market researchers and consultants get tract- and county-scale loss histories that put federal numbers under resilience and mitigation engagements; see property casualty insurance data for market researchers. Developers and builders embed loss-history lookups on stable census codes; see property casualty insurance data for developers builders. Journalists and academics get the citable receipt behind every flood-damage headline.

Within property-casualty data, this record owns the realized-loss layer: what actually flooded, when, and what the federal program paid, claim by claim, back to 1970. Its neighbours own different layers. FEMA NFIP Redacted Policies v2 owns the premium side - 73.6 million policy transactions with premiums, limits and rating detail, the exposed book these claims landed on; the pairing is close enough that we scored it head-to-head in Claims v2 vs Policies v2. The NFIP Geospatial Hub owns hazard, not history - flood zones and FIRM panels marking where flooding is expected, next to where it happened. NOAA NCEI's Billion-Dollar Disasters supplies event-scale totals, useful context above the claim-level receipts, while First Street Foundation's climate risk data looks forward with modeled risk where these rows look backward with measured outcomes. Layer hazard, book, event and claim and you have the whole flood economy. The full slate sits on our best property casualty insurance datasets shortlist.

Provenance note - published through FEMA's OpenFEMA program, the statistical arm of the agency that writes the cheques; its profile lives on our FEMA OpenFEMA source page.

Redaction note - coordinates are rounded and reported city reads Currently Unavailable in recent vintages, by design. Treat the census geography as the intended analytical grain; attempts to sharpen it past that defeat the point and the terms alike.

Lifecycle note - FEMA has released a successor v3 of the redacted claims extract and retires v2 after October 15, 2026; the v2 record described here holds at its final June 2026 vintage, which matters for anyone reconciling a series across the cutover.

Completeness note - Datadory scores this record 9/10; the 72-field dictionary verifies at verified confidence, and remaining detail folds under additional fields on request. Related records follow, and the whole vertical maps out in the property casualty insurance data guide.