Extreme: about 1-2.5% of NFIP-insured properties account for roughly a quarter to half of all federal flood claims dollars

Summary

Federal flood claims are extremely concentrated in a small number of properties that flood again and again. Across every published FEMA/GAO/CRS estimate spanning 2004 to March 2026, properties FEMA classifies as repetitive loss (RL) - buildings that have collected two or more paid flood claims over $1,000 within a 10-year window since 1978 - make up only about 1% to 2.5% of NFIP policies, yet those same properties have absorbed roughly 25% to 48% of all NFIP claims dollars ever paid, depending on the measurement window and definition used. The most recent figure, from a GAO report published March 26, 2026, puts unmitigated repetitive-loss properties at 2.5% of policies but 48% of claims dollars. A narrower category, severe repetitive loss (SRL) properties (4+ claims over $5,000, or 2+ claims exceeding the building's value), is a small fraction of RL properties but carries an even more extreme dollar share per property.

Separately, we computed our own concentration measure directly from the askamerica warehouse's disasters.nfip_claims table (1,009,346 claims, $57.3B paid, 2010-2026): claims dollars are extremely concentrated by county as well (Gini = 0.955; the top 10% of counties by dollars paid account for 94% of the total). This is a genuinely different, geography-level finding - it does not by itself prove any single property flooded repeatedly - but it is directionally consistent with, and adds a corroborating dimension to, the property-level repetitive-loss literature.

What "repetitive loss" means and why the concentration is so extreme

FEMA defines a Repetitive Loss (RL) property as an NFIP-insured building that has incurred two or more paid flood claims of more than $1,000 each within any rolling 10-year period since 1978. A Severe Repetitive Loss (SRL) property is a tighter subset: one that has either (a) 4+ claims of more than $5,000 each, or (b) 2+ building-only claims whose combined payments exceed the property's value. Neither category caps how many times a property can claim or how much it can collect - there is no lifetime claims limit under the standard NFIP policy, which is a structural reason the dollar concentration compounds over decades. Pew's 2016 analysis found roughly 1 in 10 repeatedly flooded properties had received cumulative payments exceeding the property's own value, and cited extreme individual cases - e.g., a $69,000 Mississippi home that flooded 34 times in 32 years and collected $663,000 in claims (nearly 10x its value).

The published numbers, by vintage

The figures are not perfectly comparable - they use different cutoff years, different claim windows, and sometimes different RL/SRL definitions - but every vintage over more than 20 years of reporting lands in the same range: a low-single-digit share of properties absorbing a quarter to nearly half of the program's total claims payments. Since 1978, nearly 100,000 structures have had two or more paid flood losses, and nearly 1,000 of those have flooded 10 or more times (Pew, 2016, citing FEMA).

What we verified directly in the askamerica warehouse, and its limits

We queried disasters.nfip_claims directly: it holds 1,009,346 individual paid claims for 2010-2026, totaling $57.3 billion in building + contents + ICC payments. This table is the right source for realized flood losses, but it has no property-level identifier - no address, parcel ID, or FEMA repetitive-loss flag - and its latitude/longitude fields are rounded to one decimal degree (roughly 6-8 miles of resolution) for FEMA privacy reasons. That means this corpus cannot, on its own, identify which specific claims landed on the same structure twice; doing so would require joining to FEMA's separate NFIP Multiple Loss Properties / Repetitive Loss dataset, which is not currently loaded here (search_catalog confirms no table or column in this corpus matches "repetitive").

This is not just a claim about our own corpus. A March 26, 2026 Congressional Research Service statement to the House Financial Services Committee (Diane Horn, "Mitigation and Multiple Loss Properties") states explicitly that FEMA's own public OpenFEMA repetitive-loss file gives the number of claims per RL/SRL structure but not the dollar amount of those claims - so "no information is available online with which to calculate total claims from multiple loss structures" independently. Every dollar-share figure above is therefore FEMA's own internal tabulation, reported secondhand in GAO/CRS/Pew documents, not something an outside analyst (including us) can currently reproduce from public microdata alone.

What we could compute directly and did: a county-level concentration measure of the claims dollars. For this calculation we excluded 3,039 claim rows (out of 1,009,346, about 0.3% of rows) that carry a null county_fips - these rows total $83.6 million, about 0.15% of the $57.3B all-claims total, so their exclusion does not materially change the concentration result. With those rows dropped, the Gini coefficient across the remaining 2,735 counties is 0.955 (0-1 scale, 1 = total concentration), and the top 10% of counties by dollars paid account for 94% of the $57.2B counted total. This confirms flood losses are geographically as well as structurally concentrated, but it is a distinct measure from property-level repetition - a county can have a high dollar total either because a few properties there flood repeatedly, or because many different properties there flooded once during a single large event (e.g., a hurricane). We are not conflating the two in the headline figures above.

Every query behind this report

3 warehouse calls ran in this session, in order. Each is reproducible against the same snapshot.

SELECT county_fips, SUM(COALESCE(amount_paid_building,0)+COALESCE(amount_paid_contents,0)+COALESCE(amount_paid_icc,0)) AS total_paid FROM disasters.nfip_claims WHERE county_fips IS NOT NULL GROUP BY county_fips

SELECT COUNT(*) AS n_claims, SUM(COALESCE(amount_paid_building,0)+COALESCE(amount_paid_contents,0)+COALESCE(amount_paid_icc,0)) AS total_paid FROM disasters.nfip_claims

SELECT COUNT(*) AS n_null_county, SUM(COALESCE(amount_paid_building,0)+COALESCE(amount_paid_contents,0)+COALESCE(amount_paid_icc,0)) AS paid_null_county FROM disasters.nfip_claims WHERE county_fips IS NULL