US retail sales and the control group

What the survey covers 

The advance monthly retail trade survey estimates sales at retail and food service establishments in the United States. It is a business survey, not a household survey: the respondents are firms reporting their own receipts, sampled from the register of employer businesses and grouped by industry classification. Sales are recorded when the transaction happens, regardless of how it was financed, and they are recorded before any adjustment for price change.

The coverage boundary is narrower than the word consumption suggests. Retail and food services account for a minority of total household spending, because most services, including housing, healthcare, education, insurance, travel and professional services, are outside the scope of a retail trade survey entirely. The report is therefore an early and partial read on goods spending, not a measurement of consumption.

The headline is nominal 

Retail sales are published in current dollars with no deflation for price change. A rise in the headline is therefore consistent with more goods being sold at the same price, the same goods being sold at higher prices, or any combination of the two, and the report itself cannot distinguish between them.

Practitioners handle this by reading the release against a price index for the corresponding categories, which is one reason the consumer price index and the retail sales report are frequently discussed together even though they are produced by different agencies from different surveys with different definitions. Any such comparison is an analytical construction assembled by the reader, not a published statistic, and the category boundaries of the two surveys do not line up exactly.

A second consequence of the nominal presentation is that the report is not comparable across long spans without deflation. A decade of headline growth in current dollars describes the combined effect of population growth, price change and any shift in the share of household budgets going to goods rather than services, and those three influences cannot be separated inside the series itself. The Census Bureau publishes the survey as a measure of business receipts, which is what it is, and the interpretation of those receipts as a consumer indicator is the reader's construction rather than the agency's claim.

The exclusions that produce the control group 

The report is conventionally quoted at four successive levels of exclusion, each removing a category on the grounds that it is volatile, price-driven or definitionally awkward.

1. The headline: all retail and food service sales.
2. Excluding motor vehicles and parts. Vehicle sales are large, lumpy and driven by incentive campaigns and production schedules, so a single strong or weak month in the category can dominate the headline.
3. Excluding motor vehicles and gasoline. Fuel is priced globally and its dollar value moves with the oil price rather than with the quantity purchased, so including it lets a commodity price change masquerade as a change in consumer behaviour.
4. The control group: excluding motor vehicles, gasoline, building materials and food services. Building materials sit closer to construction than to consumption in the national accounts, and food services are estimated differently there.

The control group exists for a specific reason rather than an aesthetic one. Its categories are the ones that map most directly onto the goods portion of personal consumption expenditures in the quarterly national accounts, so it is the line that feeds the consumption estimate most mechanically. That is why analysts who follow the growth accounts read the control group first and the headline second, and why the two can point in different directions in the same report.

Cadence, revisions and the advance estimate 

The advance report is published monthly, in the morning Eastern Time, roughly two weeks after the reference month ends. The word advance is doing real work: it is based on a subsample of the full survey panel, and it is revised in the following month's report when the fuller sample has responded, then again in the annual revision. Monthly revisions to this series are routinely larger than the monthly change itself, which is a consequence of publishing early from a partial sample.

Seasonal adjustment carries an unusually heavy load here because retail activity has a very large regular within-year pattern. The adjustment must also cope with holidays that move between months, with the number of weekends in a month, and with the timing of promotional events that have become fixed features of the calendar. The unadjusted series is published alongside the adjusted one, and the gap between them in some months is larger than any plausible economic development.

How the release is conventionally read 

The prevailing convention reads the control group first, then examines which categories drove the difference between it and the headline, then reads the prior month's revision alongside the new month on the reasoning that a revision changes the base from which the new figure is measured. A separate convention reads the report against the same month's consumer price data to form a rough view of whether a nominal change was volume or price, with the caveat that the two surveys do not share category definitions.

One structural change worth noting is the treatment of online commerce. Sales by non-store retailers are inside the survey and inside the control group, so the long migration of spending from physical stores to online channels moves activity between categories without changing the totals. Category-level readings across long spans therefore describe a shift in the channel of purchase as much as a shift in what was purchased.