Confidence Intervals and Seasonal Adjustment in Retail Sales

posted in: Uncategorized | 0

Retail sales were reported to have risen 0.5% for July over June, up 8.9% year over year.  Many cheered the number ex-autos which beat consensus of a 0.3% increase.  Economic data is noisy though, despite the fact that even mild “beats”can move markets.  Two factors adding to the noise are discussed here: confidence intervals and seasonal adjustments.

Confidence intervals are rarely reported with economic data, even though they are easily understood when reporting a public opinion poll.  Retail sales, like other economic data are only estimates (based on a sample of a larger group) and therefore each report has a confidence interval.  Retail sales’ confidence interval happens to be +/- 0.5%, meaning that there is at least a 30% chance that retail sales actually missed the consensus estimate of 0.3% in July.  

Another factor that significantly changes the interpretation of reported economic data is the seasonal adjustment, which is particularly true for retail sales.  Take a look at a comparison of seasonally and not seasonally adjusted charts since ’00.

Seasonally adjusted, retail sales were up in July 0.5%.  Not seasonally adjusted, they were down 1.04%. Adding to the noise, the seasonal adjustment isn’t constant and varies from year to year.  For example, the seasonal adjustment factor for January is below:
YEAR JAN
1992 0.898
1993 0.88
1994 0.876
1995 0.873
1996 0.885
1997 0.895
1998 0.897
1999 0.883
2000 0.882
2001 0.893
2002 0.901
2003 0.907
2004 0.909
2005 0.889
2006 0.889
2007 0.904
2008 0.914
2009 0.922
2010 0.906
2011 0.903
2012 0.899
2013 0.918
Seasonal adjustment is an added estimation on top of the estimated retail sales figure.  This creates the opportunity for further misstatement.  
Ignoring confidence intervals and seasonal adjustments can have a significant effect on the lens with which one views economic data.  In addition to these factors, “real” data adjustments (particularly to GDP) and historical revisions effect the interpretation of individual data-points.  Because of these inadequacies, beats and misses for any individual economic datapoint should be taken with a grain of salt.