How Often FEMA Public Assistance Money Is Pulled Back, and Where It Concentrates
The gap & the data
Recovery research and reporting track what FEMA obligates — not the money later deobligated. Oversight (DHS OIG, GAO) covers deobligation qualitatively or for single disasters, from audit samples. Realized deobligation had never been measured across the national project population.
Each OpenFEMA record is one project-worksheet version carrying a signed change to the federal share; a negative version is a deobligation event. Verified directly: PW 421 (disaster 1412) runs +$2,498.25 → +$2,641.50 → +$2,887.50 → −$2,887.50. Every reported number is dual-implemented (two independent code paths, exact match) and vintage-stamped; the source dataset is restated by FEMA over time, so re-derivation against a fresh pull is part of the method.
Findings & interpretation
Deobligation concentrates where documentation and procurement burden is heaviest (water control 18.2%, buildings 12.5%, utilities 12.2% vs roads 7.7%) and among the least-resourced applicants (nonprofits 11.8% vs governments 9.8%). Management costs (19.4%) are excluded from that reading: FEMA policy FP 104-11-2 obligates category Z incrementally and prescribes a closeout true-up, and 97.8% of standalone modern-era Z deobligations are the terminal action on their worksheet (Study 9, dual-verified) — a designed cadence, not compliance exposure.
What the gross $44.3B is made of (v1.3): a penny-exact same-day screen certifies at least $3.90B (8.8%) as intra-award transfers (86.6% of it Puerto Rico sec. 428 drawdowns — budget movements, not takebacks) and $1.80B (4.1%) as category Z, the policy-designed true-up channel (event-level certification; Z's own dollar decomposition is majority-residual); 87.1% remains project-level revisions, closeouts, and recoveries, undecomposed. The transfer figure is a floor — the screen is deliberately strict.
Of 107,708 ever-deobligated projects, ~28% reverse to exactly $0 (churn or withdrawal) and ~71% end as partial reductions (median $76,790 remaining) — bounding how much of the gross $44.3B is churn versus real reduction. (An earlier snapshot's net-negative anomaly resolves to 0.8% on the completed data: it was the snapshot, not the program.) The pattern is consistent with a compliance-burden mechanism, though the data cannot isolate it from routine cost reconciliation.
Limitations (load-bearing)
- Reason is unobserved. A downward revision may be a clawback, scope change, cost underrun, insurance offset, or correction — all identical in the data. We report net deobligation, not “clawback.”
- Gross figures include obligate–reverse churn; net ($283.6B) is the conservative number.
- Vintage sensitivity. FEMA restates this dataset; an earlier snapshot gave materially different figures. Every number here is stamped to the 2026-08-26 pull; the reconciliation memo documents the differences.
- OpenFEMA is raw administrative data; recent disasters are right-censored (handled via cumulative incidence).