Tis the season for economic and business data to get messed up by—the season! Things are not normal this time of year. Shopping for clothing goes up, for building materials goes down, and woe be to the executive who doesn't recognize the pattern. I pointed out the role of our December holiday in How Big Is Christmas For Retail and explained the basics in Seasonality In Business And Economics: A Primer. In this article I explain how to seasonally adjust your own data.
Understanding seasonal patterns can help you figure out whether things are going well. Let's say that your jewelry store sales are up 75 percent in December over November. Is that good, or just normal for December, or maybe sales are usually up 100 percent? A seasonal adjustment of your data helps you understand how you are doing.
The Poor Man's Seasonal Adjustment
The pre-computer method of seasonal adjustment was pretty simple and lends itself well to spreadsheets. This method also creates an understanding of the idea of seasonal adjustment. The basic concept is that for each month, we'll compute the ratio of that month's sales (or other concept, like service calls) to the entire year's sales. So for January of the first year, we'll calculate January sales divided by average annual sales. We'll do this for January of every year, average the results, and have a good idea of how January typically differs from the average month. We then repeat the exercise for every other month, February through December, and we'll know about all 12 months of the year.
Don't start creating a spreadsheet just yet, though. Many data series have an upward trend. They grow over time. If there is an underlying growth trend, then December would be higher than January just because of routine growth. So instead of dividing January sales by the calendar year average, we should divide by the centered 12-month average. That's a bit complicated, because we want the 12-month average to be centered on a single month. We'll add up the five months before January, then add January, and the five months after January. That gives us 11 months. What about July? Should we use the previous July or the next July? The best way is to use half of each, as show in this chart:
