Census Bureau Revisions to Retail Sales

Earlier today I posted my monthly update on Retail Sales. Those of us who routinely track this series know that the Advance Estimate will be followed by a second estimate next month and a third estimate the month after. How big are those revisions? Are they big enough to warrant skepticism about the Advance Estimate?

See for yourself. Here is a visualization of the cumulative change from the first to third estimates from January 2007 through October 2014, the most recent month for which we have all three data points.

As we see, revisions abound, and they move in either direction, although mostly downward during the last recession. For a better sense of the magnitude of the revisions, the next chart shows the percent change from the first (advance estimate) to third (second revision).

During this timeframe there were 45 upward revisions and 49 downward revisions. The absolute mean (average) revision was 0.35%, which comprises a 0.28% average for the upward adjustments and -0.41% for the downward adjustments.

The Census Bureau's latest Advance Monthly Retail Trade and Food Services reported a 0.0% month-over-month change. The CB adds a parenthetical (±0.5%)* for that MoM advance estimate. The asterisk points to the following explanation:

* The 90 percent confidence interval includes zero. The Census Bureau does not have sufficient statistical evidence to conclude that the actual change is different than zero. [PDF source]

Last year Bob Bronson of Bronson Capital Markets Research has shared some interesting observations:

Note that when revisions are as much as +0.5%, as they were in the June 2014 data, they have always been revised lower the next month, and four of the last seven times those follow-on revisions were negative, eliminating (on average) the previous upward revision. The Census Bureau's ARIMA X-13 seasonal adjustment formula has the most difficulty with highly seasonal data like retail sales.

The Ultimate Direction of Revisions

Actually there's much more to the Census Bureau's revisions than the change from the advance to second estimates. We get annual revisions of the series as well. The longer term pattern of revision is quite volatile, but if we study the percent change from the advance estimate to latest revision, it is consistently downward.

The message is clear: Don't take the advance estimate of retail sales, or the initial revisions, too seriously.