SECTION 01The swap nobody reads
October 2025 started on a Wednesday. October 2026 starts on a Thursday. That one day decides which three weekdays show up five times instead of four.
Last October that was Wednesday, Thursday and Friday. This October it is Thursday, Friday and Saturday. So when an owner puts this October next to last October, which is the comparison almost every owner makes, one ordinary Wednesday has quietly turned into a Saturday.
It happens every month of every year. A 30 or 31 day month always has two or three days that appear five times, and each year that set moves along by a day, two after a leap day. November 2025 had five Saturdays and Sundays. November 2026 has five Sundays and Mondays. This November hands the Saturday back.
Nobody writes it on the report. The question is how much it is worth, and to whom.
SECTION 02What we measured
The Census Bureau surveys retail and food service businesses every month and publishes their sales by kind of business, back to 1992.[1] We read the whole workbook: 41,299 published figures, 416 months, and the 23 kinds of business with an unbroken monthly record, from restaurants and liquor stores to building supply dealers and online sellers.
For each trade, every month was compared with the same month a year earlier, the way an owner does it. Then the growth of the eleven months around it was taken off, which removes inflation, the economy and the trade's own trend, and leaves only what was particular to that month. 77 of the 392 comparisons were set aside before anything was read: 34 that touch March 2020 to December 2021, 15 Februaries where a leap day changes the length, and 28 Marches and Aprils where Easter moved between them. That leaves 304 comparisons per trade.
Finally, the calendar. Every one of those comparisons swaps some weekdays for others, and the swap is known to the day.
SECTION 03One Saturday, two signs
Here is what the swap in front of us right now, a Wednesday for a Saturday, does to each trade's October.
For a liquor store the calendar alone adds +1.47% to this October. For a clothing store +1.39%, for restaurants +0.94%, for warehouse clubs and supercenters +1.07%.
For a building supply and garden dealer the same Saturday is worth -2.07%. For an online seller -2.30%, for a pharmacy -1.67%, for an auto parts and tire store -0.98%.
Same Saturday. Opposite signs.
A building supply dealer can be behind by two percent this October and have done nothing wrong. A liquor store can be ahead by one and a half and owe all of it to a Saturday.
SECTION 04Random calendars
A month moves for a hundred reasons, so the first job is to find out what ordinary looks like.
The plainest test there is: split each trade's months into the ones that gained a Saturday on the year before and the ones that lost one, and compare their averages. Across 304 comparisons, building supply dealers came in -3.27% in months that gained a Saturday and +2.74% in months that lost one, a gap of -6.00%.
Then the control. Keep every month's sales exactly as they were, and deal the calendars out to the months at random, so a month that really gained a Saturday might be told it lost one. Do that 2,000 times per trade. If the gap were just months being months, the real calendar would land somewhere in the pile.
It does not. For building supply dealers, the random calendars put the gap between -1.45% and +1.45% nineteen times in twenty. The real one is -6.00%. For liquor stores the random range is -0.98% to +1.01% and the real gap is +2.98%.
Here is every trade, sorted by the size of the gap. The gap is months that gained a Saturday against months that lost one. The random range is where nineteen in twenty of the 2,000 shuffled calendars landed.
| Trade | Gap | Random, 19 in 20 | Verdict |
|---|---|---|---|
| Online and mail order | -1.33% to +1.40% | Real | |
| Fuel dealers | -3.87% to +3.63% | Real | |
| Building materials and garden | -1.45% to +1.45% | Real | |
| Pharmacies | -0.84% to +0.85% | Real | |
| Auto parts and tire stores | -1.08% to +1.12% | Real | |
| Hobby, toy and game stores | -2.29% to +2.24% | Real | |
| Liquor stores | -0.98% to +1.01% | Real | |
| Shoe stores | -1.36% to +1.38% | Real | |
| Warehouse clubs and supercenters | -0.77% to +0.82% | Real | |
| Clothing stores | -1.23% to +1.29% | Real | |
| Restaurants | -0.81% to +0.82% | Real | |
| Gas stations | -2.53% to +2.85% | Noise | |
| Book stores | -2.08% to +2.13% | Noise | |
| Grocery stores | -0.57% to +0.53% | Real | |
| Sporting goods stores | -1.21% to +1.33% | Real | |
| Car dealers | -1.89% to +1.86% | Noise | |
| Fast food and counter service | -0.67% to +0.65% | Real | |
| Furniture stores | -1.30% to +1.29% | Noise | |
| Department stores | -1.38% to +1.46% | Noise | |
| Used merchandise stores | -2.82% to +2.72% | Noise | |
| Gift and souvenir stores | -2.22% to +2.38% | Noise | |
| Hardware stores | -1.40% to +1.32% | Noise | |
| Electronics and appliance stores | -1.09% to +1.11% | Noise |
14 of the 23 trades sit outside what random calendars produce. For the other 9, including gas stations, car dealers and department stores, the gap is inside the pile, and the calendar effect for them is not counted here.
The finding also holds in both halves of the record. Building supply dealers gapped by -7.05% from 1993 to 2009 and -4.77% from 2010 to 2026. Liquor stores by +3.14% and +2.79%. Online sellers by -6.98% and -5.86%.
SECTION 05The week, read off a calendar
The Census Bureau publishes nothing by day. But every month in the record holds a known extra Monday, or Thursday, or Saturday, so with enough months the weight of each day can be read back out of the monthly totals.
Liquor peaks on Friday and clothing on Saturday. Restaurants peak on Friday, +1.21% against the average day, well ahead of Saturday at +0.35%, and their worst day is Monday.
The other three are weekday businesses, with the shape you would expect from contractors, orders placed from a desk, and prescriptions written in office hours. A building supply dealer's Saturday sits at -1.81%. An online seller's Saturday is -2.21%, its worst day. A pharmacy loses 1.00% on a Saturday and 1.40% on a Sunday.
Which is why the same calendar flatters one owner and punishes the one across the road.
SECTION 06The Census Bureau already knew
None of this is a secret to the people who publish the data. The Bureau's adjusted retail figures, the ones its monthly release leads with, are adjusted for "holiday and trading-day differences".[1] Its seasonal adjustment estimates "trading day and moving holiday effects" and removes them from the series.[2]
That gave us a second, independent check. Dividing the not-adjusted figures by the adjusted ones leaves the Bureau's own correction, and that correction can be put through the same test. For the 13 trades where both are published:
| Trade | Our measure | From the Census adjustment |
|---|---|---|
| Liquor stores | +1.47% | +1.40% |
| Grocery stores | +0.54% | +0.52% |
| Gas stations | -1.46% | +0.17% |
| Pharmacies | -1.67% | -1.39% |
| Clothing stores | +1.39% | +2.19% |
| Shoe stores | +1.57% | +1.93% |
| Electronics and appliance stores | +0.19% | +0.11% |
| Building materials and garden | -2.07% | -1.62% |
| Department stores | -0.38% | +0.14% |
| Warehouse clubs and supercenters | +1.07% | +1.74% |
| Auto parts and tire stores | -0.98% | -1.23% |
| Online and mail order | -2.30% | -2.00% |
| Fuel dealers | -3.77% | -2.45% |
The two measures correlate at 0.95 and agree on the direction in 11 of 13. The two that disagree, gas stations and department stores, are two the random calendars had already marked as noise.
And here is the part that matters for a business. The correction exists, and it is applied to the national number. Added up across every trade, the October swap moves total retail and food service sales by only -0.31%, because the trades that gain and the trades that lose nearly cancel. The headline barely notices. A single business is never the headline: it is one trade, and it gets the whole effect, uncorrected, on its own report.
SECTION 07October and November, back to back
October 2026 gains a Saturday on last year. November 2026 loses one. So the two months that lead into the busiest stretch of the year will be wrong in opposite directions.
Take a building supply and garden dealer doing $6 million a year. In its trade, October is 8.76% of the year, which makes its October $525,418. This October's calendar costs it about $10,890 against last October, and November's hands back about $10,883. Against last year, October looks worse and November looks better than the business underneath them, by the same amount.
A liquor store doing $3 million a year gets the mirror image: about $3,679 of free October and about $4,546 gone from November.
Two months, two stories, and the calendar wrote both.
SECTION 08What it is worth
Three things a percent or two can decide in a month: whether a price change looks like it worked, whether a new hire looks like a good one, and whether the owner spends November worried. Each of those reads a month against last year.
There are a few ways to take the calendar out, and which fits depends on the business. Report in four-week blocks against the same four weeks a year earlier, Saturdays against Saturdays, which removes the swap entirely. Keep calendar months and correct them with the business's own weekday averages, which a year of daily sales is enough to build. Or keep everything as it is, and know before the month closes which way the calendar is leaning. Your call.
We wrote separately about how long it takes to tell whether a change worked. The calendar is one of the reasons the honest answer is longer than anyone would like.
SECTION 09The point
The calendar is public, printed years ahead, and identical for every business in the country. It still moves a month by two or three percent, in a direction set by the trade, and the correction lives in a national statistics agency's headline, not on a business's own report.
That is the kind of thing a data team is for. Not because the arithmetic is hard, but because it takes 41,299 figures, 2,000 random calendars per trade and a second method to check the first before the answer is worth saying out loud, and nobody running a business has that week to spare.
Everyone has the calendar. Almost nobody reads it.
If you want yours read
We are a data team that small businesses hire like staff. We work with owner-operated businesses doing $2 million to $50 million a year.
Before anyone pays us anything, we will put together a short brief on your business from what is public, so you can see how we think first. Write to data@kixik.tech, or read what we actually do. Another clock that moves a business on a schedule is written up here.