Backcasting National Accounts Data Examples from United States Experience Brent Moulton Advisory Expert Group on National Accounts Washington DC 9 September 2014
2 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
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 3
Approaches ▪ Microdata approaches Detailed reclassification of micro units ▪ Macrodata approaches Concordance tables Proportional splicing Interpolation/Backward extrapolation with or without indicator 4
Examples in the US national accounts ▪ GDP-by-industry estimates 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 5
GDP by industry and NAICS ▪ U.S. statistical agencies implemented new classification system in different years Economic Census data Tax data Employment and earnings data Prices ▪ 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 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 Convert published SIC estimates to NAICS Conversion matrices for had less SIC detail For , 1977 matrix held constant V k i, t-p = V k i, t-p · (n k i, t-p / n k i, t-p+1 ) Where: 7 i = industry t = 1997 p = 1,…,10 k = VA component (output, intermediate inputs, compensation, GOS) n = conversion coefficient V = dollar value of VA component
Evaluating results 8 ▪ Reasonableness and consistency checks Growth rates compared to published SIC industries Aggregation of industry level real value added compared against expenditure-based real GDP
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 * Prior to aggregate estimates deemed more reliable than detailed industry data – proportionally scaled detail to hit aggregates 9 Time periodSource dataComments presentR&D expenditure surveys and economic census data Detailed costs by industry (business, academic, government); relatively consistent across time *R&D expenditure surveysLess consistency of surveys across time Insufficient dataGeometric interpolation Various research studies of R&D costs Selected years; straight line interpolation between data points
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 10