What Seasonal Adjustment Changes in Retail Sales Data for a Buyer

Census publishes monthly retail sales both adjusted and not adjusted, and the headline of its advance release is the adjusted figure. In the advance report for September 2026, retail and food services sales were $745.0 billion, adjusted for seasonal variation and holiday and trading-day differences but not for price changes, up 0.8 percent from the previous month. That one sentence carries three conditions a planner has to keep with the number: an adjustment, a price basis and an early estimate that will be revised.

What does Census remove when it seasonally adjusts retail sales?

According to Census's adjustment-factor page, the bureau derives factors with the X-13ARIMA-SEATS program for seasonal variations and, for sales, for trading-day and holiday differences, and it publishes the combined factors it uses to adjust sales estimates for each kind of business. In a Federal Reserve Bank of Dallas explainer, the method starts by estimating the effects that come back in the same month each year at a similar size and in the same direction, and it aims to split that seasonal part from the series' longer trend-cycle and its irregular movements.

For a buyer, the gain is a fairer comparison between neighbouring months. In the methodology of the advance survey, Census states that the estimates are adjusted for seasonal variation and holiday and trading-day differences with the X-11 filter-based procedure inside X-13ARIMA-SEATS, so the expected holiday lift is taken out of a November-to-December change in the adjusted headline. The unadjusted series keeps that swing, and several years of it, or the published factors, show a category plan how much heavier one month runs than another for a kind of store. The factor panel below shows how different that swing is from one kind of store to the next.

What does an adjusted series still contain?

Adjustment takes out the calendar Census can predict, so anything unusual in a month stays in the adjusted figure. In the Dallas Fed explainer, the irregular part of a series is traced to causes such as unseasonable weather, natural disasters, strikes or sampling error, and adjustment leaves that part in place. In Census's weather FAQ for the advance survey, the bureau says it cannot isolate the effect of weather events on those estimates, because the sample is designed to measure retail and food services activity at the national level.

The factors are estimates as well. On the same adjustment-factor page, Census calls adjustment an approximation based on current and past experience, which could become less precise if competitive pressures or changes in holiday buying shift the seasonal, trading-day and holiday patterns. The survey adds its own uncertainty: according to the monthly survey's methodology, imputed data make up about 36.3 percent of the total monthly retail and food services sales estimate in any given month, and its standard errors measure sampling variability only, not systematic biases.

Why are adjusted retail dollars still not a measure of volume?

Seasonal adjustment and price adjustment are different operations, and Census performs only the first in these tables. In the retail program's definitions, all dollar values are current dollars, not adjusted to a constant dollar series, and users who compare estimates with prior years are told to consider price level changes. The advance release repeats the point in its own label: in the September 2026 report, the figures are adjusted for seasonal variation and holiday and trading-day differences but not for price changes.

In practice, a kind of store whose adjusted sales rose may have sold the same quantity at higher prices or more units at lower prices, and the retail table cannot separate the two. A buyer who needs volume has to set a price measure for comparable goods beside the sales figure and accept a loose match, or use unit sales data instead, such as sell-through data from customers or point-of-sale records in the team's own systems.

Why does an adjusted figure change after it is published?

An adjusted figure moves for three separate reasons, each on its own timetable. In the advance survey's methodology, Census says its advance estimates come out about ten business days after the month ends and are revised one month later by the larger monthly survey. On the adjustment-factor page, it explains that each month concurrent adjustment feeds every available unadjusted estimate, including the latest preliminary and advance ones, into X-13ARIMA-SEATS, and that the published projected factors are estimates of the factors it will use once unadjusted estimates arrive.

The third reason is the annual revision. According to Census's explanation of this year's annual revision, it combined historical corrections, benchmarking to the 2024 Annual Integrated Economic Survey and updates to the seasonal adjustment models, and adjusted estimates were revised from both the revised unadjusted estimates and revised seasonal factors. In the same explanation, Census adds that monthly reports can rest on incomplete or unaudited records and on estimates respondents make themselves.

Recent releases show how much a printed figure can move. In the September 2026 advance report, the July-to-August change in total sales was revised from up 1.1 percent to up 1.3 percent, and the report notes that monthly estimates had been revised on September 28, 2026, using historical corrections and the 2023 and 2024 Annual Integrated Economic Survey results. The revision panel below compares December 2025 as Census printed it in August with the figure it prints now.

Why is neither series an item demand forecast?

In Census's retail definitions, sales are counted by the store that made them, so the estimates by type of store are not meant to measure total sales of any one commodity or merchandise line. A grocery store's sales include everything it sells, and one product sold through department stores, warehouse clubs and online retailers lands in several kinds of business. In the advance survey's methodology, retail sales are one of the primary measures of consumer demand for durable and non-durable goods, which is demand in aggregate rather than for an item. If you buy several tiers upstream of retail, read these figures as end-consumer context next to indicators closer to your own suppliers.

The survey is also national. In the weather FAQ, Census says the monthly retail programs are not designed to measure activity at specific geographic levels, so a region, a store cluster or a distribution center's catchment sits outside what the table can describe. The projected factors described above estimate the calendar effect for coming months, not the sales of any item. Build the item demand forecast from the team's own sales history, cleaned of stockout periods and promotion spikes, and from open customer orders. Then set the quantity to buy against open purchase orders, supplier lead times and what suppliers can deliver, the ground covered in our guide to supplier risk assessment methods.

How should a buyer record a retail figure before using it?

Before a retail figure enters a category plan, write its adjustment state down next to it. The register below is our proposal for doing that in one line per figure, with fields taken from the conditions the sources above attach to every estimate.

The register works in both directions. When a supplier argues for a price increase because retail sales are up, ask which line, which adjustment state and which release it means, and test the request against the supplier's cost breakdown and the indices named in the contract rather than against retail sales, which also gives supplier relationship reviews a shared fact base. When your own sales dip in January, compare the dip with the unadjusted January drop for the same kind of store before treating it as weaker demand, and if your team reports on a 4-5-4 or other fiscal calendar, restate its months as calendar months first. A year-over-year comparison helps with the January check, though in the Dallas Fed explainer it relies heavily on data that is 12 months old, so the register keeps both comparisons.

What should automated planning keep from these tables?

For the next category review, pick one retail figure that already sits in a plan or a supplier letter and fill in its register line, which takes a few minutes once the Census table is open. Then ask whether the decision it supports needs the team's item history instead.