This month's sales are down on last month. Is that the business, or the calendar?
Updated October 2026
A month-to-month change mixes the calendar with the business. Taking the usual yearly pattern out shows how much of a change is that pattern.
For US retail as a whole, much of what looks like a monthly rise or fall is the calendar: the yearly pattern is strong, and in 46 of 119 month-to-month changes, about two in five, the direction flips once the usual pattern is taken out.
The data are US retail and food services sales, not seasonally adjusted, month by month from January 2010 to December 2019. We separated each month's figure into a slow trend, a yearly pattern that repeats, and what is left. The tool's seasonal strength measure on this series is 0.950. It is a score from 0 to 1 for how much of the ups and downs is the repeating yearly pattern. The tool's own bands call 0.6 or more strong; they are the tool's choice, not a published rule.
Run the free seasonality tool on your own sales.
Also on this topic: the same idea for analysts, shown on carbon dioxide readings.
Our position: compare a month with the same month a year earlier, or take the usual pattern out before you compare with the last month. A month-to-month change mixes the calendar with the business.
The answer first: a real retail series
| Calendar month | Usual pattern, percent of the average trend level |
|---|---|
| January | -9.6% |
| February | -10.4% |
| March | +1.9% |
| April | -1.0% |
| May | +4.6% |
| June | +0.7% |
| July | +1.0% |
| August | +3.1% |
| September | -3.9% |
| October | -1.0% |
| November | +0.9% |
| December | +13.7% |
US retail and food services sales, total, not seasonally adjusted, 120 months, January 2010 to December 2019. December is the high, at +13.7% of the trend level, and February is the low, at -10.4%. From low to high that is a swing of 24.0 percentage points of trend, worked out from the unrounded values, so it is not exactly 13.7 plus 10.4. It is not 24 percent of sales.
Source: U.S. Census Bureau, Monthly Retail Trade Survey. No endorsement implied; no Census logos. Changes: not seasonally adjusted series, January 2010 to December 2019, millions of dollars as published; seasonal decomposition by MCP Analytics.
What this means for your decisions
- Name the calendar before you blame the business. The biggest raw fall in this window is what we call the December-to-January calendar drop: sales fell 21.9% from December 2015 into January 2016. That is the usual shape of the year, not a collapse in sales.
- Look at the same changes with the pattern removed. Adjusted for the usual yearly pattern, the biggest fall in the window is 4.2% (into May 2010) and the biggest rise is 5.1% (into February 2012). The raw series ranges from a fall of 21.9% to a rise of 16.8% (into March 2018).
- Expect the raw direction to mislead some of the time. In 46 of the 119 month-to-month changes, about two in five, the raw series moved one way and the adjusted series moved the other.
- These figures describe one series. This is an industry total from a government survey. It says nothing about the cause of any month's level, it is not a forecast, and it says nothing about any one retailer.
What a seasonal pattern is
A seasonal pattern is the part of a monthly figure that repeats every year, such as the December rush and the quiet of January and February. A series of monthly sales has three parts stacked together: a slow trend, the yearly pattern, and what is left over. Taking the pattern out lets you see the other two.
This month vs last month: not the same as this month vs a year ago
Last month sits at a different point in the yearly pattern, so a month-to-month change mixes the calendar with the business. The same month a year earlier sits at the same point in the pattern, so the pattern drops out of the comparison.
Why the window stops at 2019
The window ends at December 2019. The method assumes one yearly pattern for the whole window.
"Adjusted" here means adjusted for the usual yearly pattern. We used a method called STL, which splits a monthly series into three parts: the slow trend, the yearly pattern, and the remainder that is left over. It is our own adjusted series, the published series minus its yearly pattern. It is not the Census Bureau's own seasonally adjusted series, which the same workbook also publishes.
How to run it on your own file
You need one date column and one sales column, monthly, covering more than two full years. The free tool needs at least 25 months.
- Take one number per month.
- Estimate the usual yearly pattern, one value per calendar month.
- Subtract it from each month.
- Compare the adjusted months with each other.
The free seasonality tool does the separation on your file. Run the free seasonality tool on your own sales.
What you need and what you get back
What your file needs: one date column and one sales number per month, covering more than two full years (at least 25 months).
What you get back (from the free seasonality tool): how strong the yearly pattern is, the high and the low month, and the trend.
Questions readers ask
Is a drop from December to January a problem?
In US retail as a whole it is the usual pattern, not a sign of trouble in itself. The December-to-January calendar drop, 21.9% into January 2016, is the largest raw fall in the window: the usual pattern puts December at the high and January near the low. Whether it is normal for your business depends on your own pattern, which the free tool shows on your file.
How do I know my business has a yearly pattern?
Run your own monthly series through the free tool and read the strength. For US retail sales in this window the tool's measure is 0.950, which is strong; your series will have its own value.
Why compare with the same month a year earlier?
Because the same calendar month sits at the same point in the yearly pattern. The pattern drops out of the comparison.
Does this tell me why sales moved in a given month?
No. It separates the usual yearly pattern from what is left. It does not say what caused any month's level.
How we checked it
Three things to know about this reading.
One yearly pattern for the whole window
The method fits one fixed yearly pattern to all the months. The window ends at December 2019.
Monthly, ten full years
The tool read the data as monthly, found 120 months and ten full yearly cycles, removed no rows and filled in no months.
Our own adjusted series
The month-over-month comparison uses our own adjusted series, the published series minus its yearly pattern. It is not the Census Bureau's own seasonally adjusted series, and it is our calculation, not a tool output.