The federal fiscal year ends on September 30. Six days from now, agencies will finish the annual ritual of obligating whatever is left of their one-year money, because appropriations that aren't committed by the deadline go back to the Treasury. Everyone in Washington calls it use it or lose it. Two economists, Jeffrey Liebman and Neale Mahoney, put a number on it a decade ago: in 2004–2009 procurement data, the last week of the year carried 4.9 times the spending of an average week, and the IT projects bought in that week were two to six times as likely to get low quality ratings.
We wanted to see what it looks like now, from the outside, using nothing but the public API. USAspending.gov is the Treasury's open-data site for every federal award since FY2008, and its API is in the index as usaspending-gov. No key, no signup: POST a JSON body and wait. Three request bodies produced everything below.
One request, seventeen years
The workhorse is search/spending_over_time. Ask it for monthly totals of contract obligations (award types A through D: definitive contracts, purchase orders, delivery orders and BPA calls) from the start of FY2010 to today:
POST https://api.usaspending.gov/api/v2/search/spending_over_time/
{
"group": "month",
"filters": {
"time_period": [{"start_date": "2009-10-01", "end_date": "2026-09-30"}],
"award_type_codes": ["A", "B", "C", "D"]
}
}
Fifty seconds later you get 204 rows, one per fiscal month. Two things to know before you plot them. Months are numbered from the start of the fiscal year, so month 1 is October and month 12 is September. And the totals are obligations by action date: money committed on a contract that day, net of de-obligations, not cash going out the door.
Every complete year in the series puts between 15.0% and 19.1% of its contract dollars in September, 17.1% on average, which is 2.3 times an ordinary month. There are no exceptions: not the sequestration year, not the pandemic year. FY2025 set the high mark, with $148.5 billion of its $778.5 billion booked in the final month.
Swap the award types for grants (codes 02 through 05) and the pattern vanishes. Grants put 6.7% of their dollars in September and run on a quarterly clock instead, with October, January, April and July each carrying 16 to 20 percent. Formula grants tend to go out when a quarter starts; contracts go out when the money is about to expire.
Which agencies sprint hardest
The same endpoint takes an agencies filter, so a second body, sent once per agency, gives the ranking. We pooled FY2021 through FY2025 to smooth out single-year noise and kept awarding agencies with more than $5 billion in contracts over the period.
"agencies": [{"type": "awarding", "tier": "toptier", "name": "Department of State"}]
State books 35% of its contract dollars in September and Homeland Security 28%. Defense, which is most of the money, sits at 16%. Energy is the outlier at 7%: most of its contract dollars go to the management-and-operating contracts that run the national laboratories, funded steadily through the year rather than against a September deadline. The Social Security Administration tops the chart at 37%, but on a small portfolio of about $9 billion across the five years.
Down to the week, and the day
Month is the finest grouping the endpoint offers, so anything finer means one request per window. We sent 53 seven-day windows covering FY2025 and 61 single-day windows covering August and September 2025: 114 calls, three at a time, in a little over twenty minutes. The weekly and daily pulls sum to the monthly totals to the cent, which says the date filter means what it appears to mean.
| FY2025 | Contract obligations |
|---|---|
| Average week | $14.9B |
| Week of September 23–29 | $67.0B, 4.5x an average week |
| Last seven days, September 24–30 | $77.4B, 5.2x an average week |
| Average day | $2.1B |
| Monday, September 29 | $25.4B, 11.9x an average day |
| Tuesday, September 30 | $17.1B |
| First half of September / second half | $35.5B / $113.0B |
Liebman and Mahoney's 4.9x was measured on 2004–2009 data. The FY2025 figure on the same definition is 5.2x. Fifteen years, several reform efforts and one pandemic later, the sprint is a little faster than it was. The government even worked the last weekend: Saturday, September 27 saw $1.1 billion in obligations, against a typical Saturday's hundred million or so.
Notes for anyone building on it
- No key and no auth. The search endpoints are POST with a JSON body; the reference endpoints (agencies, NAICS codes, the glossary) are plain GET.
- Uncached queries that span many years take 40 to 50 seconds. An identical body comes back from a server-side cache in under a second, so save every response to disk and never re-send a request you already have.
- Under load you will see the occasional 502 or 504. Retry with backoff and keep concurrency at three or fewer; our 114-window pull needed six retries.
- Search data starts at 2007-10-01. Earlier years are bulk download only.
- The current month is always incomplete, because agencies report with a lag. Drop it before computing shares.
- The default
spending_levelis transactions, which is what you want for obligations by date. The awards level buckets whole awards instead, and the API itself warns those can spill across periods.
The listing at /api/usaspending-gov has the base URL and the docs link. It is unclaimed. If you work at the Bureau of the Fiscal Service, it's yours.
Source: USAspending.gov API, search/spending_over_time, contract obligations by action date, pulled September 16, 2026. Liebman, J. B. and Mahoney, N., “Do Expiring Budgets Lead to Wasteful Year-End Spending? Evidence from Federal Procurement,” American Economic Review 107(11), 2017; NBER Working Paper 19481.