Transcript PPT - United Nations Statistics Division

Backcasting National
Accounts Data
Examples from United States
Experience
Brent Moulton
Advisory Expert Group on National Accounts
Washington DC
9 September 2014
www.bea.gov
Why backcast economic data?
▪ Provide a service to data customers
▪ Maintain time-series consistency
▪ Produce longer time series to study changes
in the economy over time
▪ Understand sources of economic growth
and productivity over time
www.bea.gov
2
When is backcasting used?
▪ Changes in classification
 Industry and other classification systems
▪ Changes in concepts
 Newly recognized asset or redefined activity
▪ Expanded detail
 Sub-aggregate breakouts
▪ When data are not available to directly
measure the economic variables
www.bea.gov
3
Approaches
▪ Microdata approaches
 Detailed reclassification of micro units
▪ Macrodata approaches
 Concordance tables
 Proportional splicing
 Interpolation/Backward extrapolation with
or without indicator
www.bea.gov
4
Examples in the US national accounts
▪ GDP-by-industry estimates 1947 - 1997
 North American Industry Classification System (NAICS)
▪ Reclassifications of exports and imports
 For example, new treatment of merchandising in BPM6
▪ Recognition of R&D as fixed assets
 Newly constructed measures of R&D investment
www.bea.gov
5
GDP by industry and NAICS
▪ U.S. statistical agencies implemented new
classification system in different years




Economic Census data - 1997
Tax data - 1998
Employment and earnings data - 2001
Prices - 2004
▪ Prior to 1998, GDP by industry was based on
Standard Industrial Classification (SIC)
▪ Users urged BEA to provide NAICS time series
▪ Not feasible to convert source data to NAICS
www.bea.gov
6
Backcasting GDP by industry
▪ Designed a backcasting technique
 1997 concordance of detailed SIC to NAIC data
 Backward extrapolate concordance with SIC source data
 Create published level SIC – NAICS conversion matrices
1987-1997
 Convert published SIC estimates to NAICS
 Conversion matrices for 1977-1986 had less SIC detail
 For 1947-1976, 1977 matrix held constant
 Vki, t-p = Vki, t-p · (nki, t-p / nki, t-p+1 )
Where:
www.bea.gov
i = industry
t = 1997
p = 1,…,10
k = VA component (output, intermediate inputs, compensation, GOS)
n = conversion coefficient
V = dollar value of VA component
7
Evaluating results
▪ Reasonableness and consistency checks
 Growth rates compared to published SIC
industries
 Aggregation of industry level real value added
compared against expenditure-based real GDP
www.bea.gov
8
Recognition of R&D as fixed asset
▪ 2013 NIPA comprehensive revision
▪ New estimates of R&D output and investment
▪ Less available and reliable data further back in time
Time period
Source data
Comments
1981 - present R&D expenditure surveys Detailed costs by industry (business,
and economic census data academic, government); relatively
consistent across time
1957-1980*
R&D expenditure surveys
Less consistency of surveys across time
1953-1956
Insufficient data
Geometric interpolation
1920-1953
Various research studies
of R&D costs
Selected years; straight line interpolation
between data points
*Prior to 1981 - aggregate estimates deemed more reliable than detailed industry data –
proportionally scaled detail to hit aggregates
www.bea.gov
9
Summary
▪ Many different reasons to backcast
▪ Each instance has unique requirements
▪ Necessitates resourcefulness and
inventiveness
▪ Need to weigh the benefit of backcasting
against the resources required and the
resulting quality of the estimates
▪ Need a strong evaluation process
www.bea.gov
10