Reporting/Market Focus from the March 2005 Edition of the SGS Newsletter

The Census Bureau’s retail sales report is the broadest measure of U.S. consumer activity published on a monthly a basis. The monthly data are compiled using a random sampling of 5,000 retail outlets nationwide, which are estimated to account for 65% of the total volume of retail sales. The numbers are revised the next month and also annually, in a comprehensive benchmark revision.

The problems with the reporting generally are tied to the sampling and/or the seasonal adjustments. The sampling appears well designed, and the Census Bureau is upfront about sampling error. For example, the following was noted in the January "advance" report, where sales were reported as down 0.3% (+/- 0.7%): "The 90 percent confidence interval includes zero. The Census Bureau does not have sufficient statistical evidence to conclude that the actual change is different from zero."

There are some issues, though. For example, new auto sales account for roughly 15% of retail sales volume and about half the statistical margin of error for the total survey. Not only does the Census Bureau conduct a sampling of auto industry sales, but so too does the Bureau of Economic Analysis, which gives us the GDP. What is hard to understand is why neither government agency uses the 100% survey published on a timely basis by the auto industry, since that would reduce sampling error significantly.

Retail sales vary meaningfully by season, with monthly sales often affected by when Easter, Labor Day or Thanksgiving fall. Accordingly, seasonal adjustment is a necessity, but the government’s seasonal-adjustment process is far from perfect and often distorts reported month-to-month changes. This problem can be overcome by looking at retail sales on a smoothed three-month moving average basis and by tracking year-to-year change.

Unlike the related personal consumption component of the GDP, however, the retail sales report is not burdened by massive theoretical imputations, and is not reported adjusted for inflation, with real growth inflated by artificially low inflation estimates.

Nonetheless, retail sales do reflect inflation pressures, directly, and it helps to know what portion of sales growth is due to physical volume and what portion is due to changing prices. Unfortunately, the government does not publish a meaningful inflation number any more. Our "consistent methodology" estimates, using pre-Clinton era CPI methodology offers one option (see graph on the home page).

Working with the limitations of official CPI reporting, it still is possible, however, to make meaningful predictions of broad economic activity using retail sales. Deflated by the official CPI, year-to-year change in a three-month moving average retail sales number shows a high correlation with economic activity, with roughly a three-month lead-time.

Retail sales is used as one of the SGS key leading indicators of economic activity, where the historical series has been statistically analyzed against reported year-to-year GDP growth. While GDP is distorted, there is some fair correlation with the pattern of the ups and downs in annual growth of reliable economic indicators. It is the reported magnitude of change in real annual GDP that is particularly off from reality.

The adjusted retail sales growth was assessed for levels that would signal either an economic boom (annual GDP growth of 6% or higher, as last seen in 1984) or bust (annual GDP growth of 0% or below, indicating a formal recession). What was found was that smoothed, real year-to-year retail sales of 9.5% or above and 1.8% or below reliably signaled a respective boom or bust.

The January 2005 smoothed real growth in retail sales is 4.2%, down from a near-term peak of 6.2% in April 2004. Look for a meaningful move to the downside after the next reporting and benchmark revision.

At such time as an actual recession signal is generated, full detail and graphics of the involved indicator will be highlighted in the SGS.