I’ve been talking with various economists about the inconsistency of the economic data we have been seeing since December regarding job creation and consumer spending. Retail sales were unexpectedly down 1.2% in December, but in January beat expectations by 1.3%. Then they were unexpectedly down 0.2% in February. Similarly, job growth of 20,000 was a large disappointment in February, but in March, U.S. job creation jumped to a better-than-expected 200,000.
Is our economy really this inconsistent? Was the accuracy of the economic data impacted by the government shutdown? By the Polar Vortex? Possibly, but I believe the real culprit is a practice that compromises most economic data – seasonal adjustments.
I like to use non-seasonally-adjusted data, which is the actual data. I believe it shows what really happens, and is a more meaningful way to measure year-to-year changes in activity. However, most economic data reported on the news is seasonally adjusted. In other words, the data you see reported is not the actual number – it’s an adjusted figure, changed by someone’s estimate of a 10-year seasonal trend.
I can explain the difference in the example below. (My friend Steve is an economist.)
Me: “Hey, Steve, nice June weather we have had. Only 10 days of rain.”
Steve: “Yes, Jeff, that is quite the break from May – five fewer days of rain.”
Me: “That was twice as much rain as last year, but June 2017 was an unusually dry June.”
Steve: “Well, actually, when we report this number to the media, it’s going to be only one fewer day of rain, as seasonally, June has a 70% adjustment factor.”
Me: “Isn’t that telling a different story than the five fewer days of rain we had?”
Steve: “Yes, but seasonally, June had only one fewer rain day.”
As you can see, the seasonal adjustment paints a different picture than the reality of the data. While this seems simple, let’s raise the complexity level and look at retail sales or auto sales.
The rapid growth of Amazon Prime, e-commerce, and the move toward home delivery has led to significant changes in consumer buying habits, as well as the employment of labor across many other industries related to the change in the holiday season and winter shopping in general. September/October has picked up in activity, and November/December now sees less of a late holiday surge, with more gifts exchanging in the January/February time frame because of gift cards and the like. However, seasonally, the adjustment factors typically used are 10-year averages, which do not reflect the realities of the past few years as accurately, because the trends have changed.
As such, how surprised should we be that retailers reported weaker-than-expected sales in December (formerly the big shopping rush), and stronger-than-expected results in January? What about the lack of job growth in the February data? An anomaly, perhaps, created by seasonally adjusted reported data, only to be shown materially stronger than expected in the month of March. The trick is, many of us listening to the headline reports on the news do not realize that the actual data has been adjusted by a seasonal factor.
The same applies to auto and truck order data. These figures may be shown both ways in the data, but the headline data is altered by someone’s estimate of a seasonal adjustment. With truck OEMs this year largely sold out through 2019, should we be surprised that the data is weaker than the normal seasonal trend?