The Census Bureau releases advance monthly retail trade at 8:30 ET on the 17th. The headline that runs on every wire is the month-over-month percent change in total retail sales, which prints to one decimal point and gets the first paragraph of the recap. That number is the noisiest one in the release. The number the Atlanta Fed and the New York Fed pull straight into their Q2 GDP nowcast is the retail sales control group, which is on a separate row halfway down the table.
The control group is total retail sales minus four categories: motor vehicles and parts dealers, gasoline stations, building materials and garden equipment, and food services and drinking places. Each exclusion has a reason that is mechanical rather than analytical.
Autos get stripped because vehicle sales are dominated by a small number of high-ticket transactions whose timing depends on incentive cycles, fleet purchases, and seasonal model-year transitions. A single month of weak auto sales can subtract 0.4 to 0.6 percentage points from the headline retail number, and a strong auto month can add the same. Neither move tells you anything about underlying consumer demand. Vehicle output is captured separately in the durable goods report and the BEA’s vehicle PCE estimate, which uses unit sales rather than dollar sales.
Gas stations get stripped because gas station revenue is a price story, not a volume story. When the retail price of gasoline rises 30 cents a gallon, gas station sales rise about 8% even if gallons sold are flat. The BEA’s PCE estimate handles gasoline through a separate volume-and-price decomposition, so leaving gas in the control group would double-count the price move.
Building materials get stripped because the BEA routes that category into residential investment in the GDP accounts rather than into goods PCE. Leaving it in the control group would assign the spending to the consumer line twice.
Food services and drinking places get stripped because that category is a service, not a good. The BEA estimates services PCE from a different set of source data, and the retail sales food services line is used as a cross-check on the BEA estimate rather than as a direct input.
What is left after the four exclusions is the control group: department stores, clothing, electronics, furniture, sporting goods, general merchandise, online retailers, miscellaneous, and a handful of smaller categories. The control group accounts for roughly 65% of total retail sales by dollar value and almost all of the read-through to non-vehicle, non-gasoline goods PCE.
The nowcast plumbing
The Atlanta Fed GDPNow model updates within 90 minutes of every retail sales release. The update path runs through the control group line. A control group print of 0.4% month-over-month, against a prior expectation of 0.3%, adds roughly 0.1 to 0.2 percentage points to the GDPNow Q2 tracking estimate, depending on where in the quarter the print lands. The New York Fed Nowcast runs a similar update with slightly different weights.
Headline retail sales can print 0.7% while the control group prints 0.2%, or the other way around. The June 2025 release was a clean example: headline came in at 0.6% on a 4.2% surge in gas station sales after a refinery outage, while the control group printed at 0.1%, the weakest print of the year. The headline lit up the wires as a strong consumer print. The nowcast tracking estimate fell 0.1 percentage point that morning. The two numbers were telling different stories about the same release, and the one that mattered for the GDP read was the smaller one.
How to read the print on the 17th
Three things, in order. First, the control group month-over-month, against the consensus from the survey distributions. Second, the prior month’s revision to the control group, which often gets buried in the table footnotes and routinely moves the nowcast more than the current month’s print. The May control group, currently logged at 0.3%, will be revised inside this release, and a revision to 0.1% or 0.5% would change the Q2 starting level. Third, the year-over-year change in the control group on a nominal basis, then mentally deflate by core goods CPI to get a rough real read.
The headline number is useful for the wires and useless for the nowcast. The control group is useful for the nowcast and never makes the headline. The two-table structure of the release is the reason. Scroll to the second table.