The Impact of Classification Changes on Time Series Continuity The Case of U.S. Monthly Retail Sales

1
The Impact of Classification Changes on Time
Series ContinuityThe Case of U.S. Monthly
Retail Sales

• Presented to
• OECD Short-Term Economic Statistics Working Party
• By Thomas E. Zabelsky
• Chief, Manufacturing and Construction Division
• thomas.e.zabelsky_at_census.gov
• June 26, 2006

2
Importance of Time Series

• Historical description of occurrence or phenomena
• Analyzing and interpreting economic conditions
• Basis for forecasting

3
Time Series Continuity

• Requires
• Continuous series of observations
• Standard methods and definitions

4
Changing Industrial Classifications

• The dilemma -
• Keeps pace with evolving industrial and business
  activities, but
• Interrupts continuity of time series data

5
North American Industry Classification System
(NAICS)

• Clean slate revision to earlier system
• First NAICS-based data published from 1997
  Economic Census

6
Impact of NAICS on Industry Classifications

• 1,170 industries 15 increase over SIC
• Industries
• - 350 new
• - 390 revised
• - 422 substantially unchanged

7
Impact of NAICS on Retail Trade

• Old SIC division split into two NAIC sectors
• - retail trade
• - accommodations and food services
• Retail trade
• - 15 industries 10 new
• - eating and drinking places accounted for
  10 in retail SIC
• - retail-wholesale boundary issues

8
Impact of NAICS on Retail Trade Data

• Source 1997 Economic Census

Sales (billions)
NAICS 2,460.9
SIC 2,546.2
Change -85.3
Percent Change -3.3
9
Sources of Change on NAICS-BasedRetail Data

• Source 1997 Economic Census

Source of Change Sales (billions)
Retail and Wholesale Boundary Change (move from wholesale) 172.4
Food Services (eating and drinking) (move to Accommodation and Food Services) -251.9
Other changes (manufacturing, etc.) -5.8
Net change -85.3
10
Restructuring Retail Time Series Data

• Restated 1992 Economic Census sales on a NAICS
  basis
• - assigned NAICS code to each employer
  establishment with an SIC that directly
  converted to NAICS (74)
• - Matched employer establishments by ID and SIC
  to 1997 to obtain NAICS (6)
• - Uncoded establishments of multi-establishment
  firms based on collective characteristic of all
  establishments (0.1)
• - Random assignment (20)
• - Exceptions

11
Restructuring Retail Time Series Data (cont.)

• Restated monthly SIC-based estimates from January
  1992 March 2001
• Restated annual retail estimates from
• 1992 1998
• Distributions based on SIC to NAICS links
  developed in 1997 census
• Adjusted the restated monthly NAICS estimates
  prior to March 2001 to account for new (NAICS)
  and old (SIC) based differences

12
Computing Benchmarked Estimates

• Restated annual estimates benchmarked to 1992 and
  1997 Economic Censuses
• Minimized differences between the year-to-year
  changes of the restated annual estimates
• 1999 was computed using the published 1998
  estimate by the ratio of the 1999 to 1998
  estimates derived from the 1999 Annual Retail
  Trade Survey

13
Computing Benchmarked Estimates (cont.)

• Monthly Retail Sales
• For January 1992 through March 2001, restated
  estimates were changed in a manner that
• - constrained the sum of the 12 months to equal
  the benchmarked, restated annual estimate for
  1992 through 1999
• - minimized the difference between the
  month- to-month changes of the restated monthly
  and benchmarked series
• Constant ratio applied to monthly estimates
  following December 1999