Integrating the 2002 Benchmark Input-Output accounts and the 2002 annual industry accounts.
Rassier, Dylan G. ; Howells, Thomas F., III ; Morgan, Edward T. 等
IN SEPTEMBER, the Bureau of Economic Analysis (BEA) released the
2002 benchmark input-output (I-O) accounts. (1) Benchmark I-O accounts
are released every 5 years and provide a detailed picture of the
economy, showing relationships among hundreds of industries and
commodities. Estimates in the benchmark I-O accounts also serve as the
statistical foundation for other BEA estimates, including gross domestic
product (GDP). In addition, economists and government officials use the
benchmark I-O accounts for a wide range of research.
One improvement in the 2002 benchmark I-O accounts was the enhanced
integration with the 2002 annual industry accounts (chart 1). (2) These
complementary accounts portray, for all industries, the goods and
services purchased, the incomes earned, and the distribution of sales.
Chart 1. Basic Steps to Reconcile the 2002 Benchmark 1-0 Accounts
and Annual Industry Accounts
Step 1
Initial estimates of
intermediate inputs in the
benchmark I-O accounts
and gross operating surplus
in the annual industry
accounts are assigned
reliability indicators, tn some
cases, these indicators are
provided by the provider of
the underlying data. In other
cases, BEA assigns an
indicator based on the
strength of the underlying
data and adjustments.
Step 2
Based on reliability measures
constructed from the reliability
indicators and coefficients of
variation, the reconciliation
model is executed. The less
reliable the estimate, the more
the estimate is adjusted, The
model satisfies standard I-0
constraints (for example,
intermediate inputs plus
value-added must equal gross
output in a given industry),
Step 3
The model derives reconciled
measures of gross operating
surplus estimates and
intermediate inputs for the
benchmark I-O accounts and
annual industry accounts.
Step 4
Only the adjusted estimates
for the benchmark I-0
accounts are publicly
released. The adjusted
estimates for the annual
industry accounts will be
revised in the next
comprehensive revision, and
a new reconciliation will be
implemented.
U.S. Bureau of Economic Analysis
However, there are notable differences between the two accounts,
and the accounts are generally used for different purposes.
Because of their rich source data--mainly the every-5-year Economic
Census--the benchmark I-O accounts paint a detailed picture of the
economy at a point in time. The 2002 benchmark I-O accounts detail the
flows of 428 commodities to 426 industries and to 13 categories of final
uses. In contrast, the annual industry accounts, which are based
primarily on data from the Internal Revenue Service (IRS), Census
Bureau, and the Bureau of Labor Statistics, provide a time series of
information about the flow of goods and services at a more aggregate
level--65 industries and 65 commodities.
A long-standing goal of BEA has been to develop more consistency
among its many accounts, including its benchmark I-O accounts and annual
industry accounts, to provide a more useful view of the economy. For the
2002 benchmark I-O accounts, an improved model was used to
"reconcile" the accounts with the 2002 annual industry
accounts. The new model resulted in improved estimates of intermediate
inputs and gross operating surplus for the 2002 benchmark I-O accounts.
In theory, a reconciliation of the two accounts would adjust data
in each account to make certain aggregates equal; for example, industry
intermediate inputs in the 2002 benchmark I-O accounts would equal their
counterparts in the 2002 annual industry accounts. In reality, only the
published benchmark I-O accounts reflect adjusted data. The 2002 annual
industry accounts, which were released in 2005, will not be adjusted
until the next comprehensive revision of the annual industry accounts,
scheduled for 2010. Thus, the published data from both accounts, which
are available on BEA's Web site, will continue to differ.
However, there are several benefits from reconciling the two
accounts, notably that the estimates in both are improved because the
reconciliation takes into account the reliability of the underlying
data. In addition, the reconciliation model provides a tool for
balancing the benchmark use table, which means adjusting data so that
all I-O identities are satisfied; for example, industry output equals
commodity output.
More specifically, BEA's new reconciliation model--which is
based on a generalized least squares frame work--offers four advantages
over past models. First, the model is transparent. The technique has
been widely researched and is familiar to national economic accounting
and statistical agencies. Second, the framework provides a firm
statistical foundation. In particular, the technique uses information on
the reliabilities of initial underlying data to make adjustments to
initial estimates. Third, the framework guarantees that adjustments to
initial estimates are as small as necessary to remove discrepancies
between the estimates subject to the model's accounting
constraints. In this way, the technique yields final estimates that are
consistent with the economic concepts on which the accounts are built.
Finally, the framework yields a model that is replicable. If no changes
are made to the data that are introduced to the model, the model yields
a duplicate set of results. Alternatively, updated data can be
introduced to the model without requiring any substantial changes to the
model or efforts to run the model.
The new reconciliation model builds on a long history of scholarly
work, much of which was pioneered by Richard Stone (see the box
"History of the Reconciliation Model"). (3) The model also
builds on the work of BEA economist Baoline Chen (2006), who conducted a
pilot study to build a reconciliation model for BEA's 1997 industry
accounts.
The rest of this article includes a description of the benchmark
I-O accounts and the annual industry accounts, focusing on the source
data and adjustment methodologies that are relevant for the
reconciliation. A nontechnical explanation of the reconciliation model
follows, including a discussion of reliability measures, the technology
used to solve the model, and the results. The article summarizes a more
comprehensive paper by the authors that includes a mathematical
description of the model. The paper is available at
<www.bea.gov/papers/index.htm>.
Benchmark Input-Output Accounts
The benchmark I-O accounts are prepared every 5 years and provide a
comprehensive picture of the flows of goods and services across all
industries and the final use categories that make up gross domestic
product. The accounts are presented in a series of tables, including a
use table and a make table.
The structure of the use table is the same in the benchmark I-O
accounts and the annual industry accounts (chart 2). The difference is
that the benchmark use table includes much more industry detail than the
annual use table.
The use table for both accounts provides gross output estimates for
industries and commodities and intermediate input and value-added
estimates by industry.
The upper left part of the table shows intermediate inputs, which
are commodities purchased by industries for the production of goods and
services. Below the intermediate inputs are the value-added components,
which include compensation of employees, taxes on production and
imports, and gross operating surplus. To the right of the intermediate
inputs are final use categories. Finally, the bottom row and the far
right column show the gross output for industries and commodities,
respectively, which come from the balanced benchmark make table.
Intermediate inputs and gross operating surplus
The benchmark I-O accounts and annual industry accounts measure
intermediate inputs and gross operating surplus differently. In the
benchmark I-O accounts, gross operating surplus--a profits-like
measure--is derived as a residual: Gross output less intermediate
inputs, compensation of employees, and taxes on production and imports
less subsidies.
Intermediate inputs for specific industries are derived mainly from
Census Bureau expense data. Approximately 71 percent of the value of
intermediate inputs comes from Census Bureau data, 22 percent from other
data, and 7 percent from BEA adjustment methodologies. For the 2002
benchmark I-O accounts, Census Bureau data were more comprehensive than
data available for previous benchmark I-O accounts. Within the
manufacturing industries, 19 categories of expenses were available from
the 2002 Economic Census. Within the service industries, 19 categories
of expenses were available from the 2002 Business Expenses Survey. Given
this depth of coverage, the gross operating surplus estimates in the
benchmark I-O accounts were also improved in these industries before and
after the reconciliation because gross operating surplus is calculated
as a residual.
BEA adjusts the estimates in the benchmark I-O accounts to ensure
they conform to established benchmark I-O concepts. In particular,
adjustments are made for nonemployer expenses, misreporting and
nonfiling, and auxiliary services. (4) The reconciliation model takes
the reliabilities of these adjustments into account.
Nonemployer expenses. The Economic Census only covers
establishments with employees and payroll. To capture the inputs and
outputs of nonemployers, BEA makes an adjustment using information
derived from administrative data calculated by the Census Bureau. (5) In
part because these data are considered reliable, the nonemployer
adjustments are considered more reliable than other adjustments in the
benchmark I-O accounts but less reliable than estimates based on
Economic Census data.
Misreporting and nonfiling. Unlike nonemployer establishments,
small employers are included in estimates published for the Economic
Census. The estimates for small employers are derived from
administrative data gathered mainly by the IRS. Such data for
nonemployers typically include individual income tax returns;
administrative data for small employers typically include individual,
partnership, and corporate income tax returns. Based on these
administrative data, BEA adjusts estimates for nonemployers and small
employers for misreporting and nonfiling. (6) A misreport results when a
tax return is filed with incomplete or incorrect information. Nonfiling
results when a business or individual who earns income fails entirely to
file a return. An adjustment for misreporting is based on data from two
IRS programs: the Taxpayer Compliance Measurement Program (TCMP) and the
TCMP-Information Return Program (TCMP-IRP). (7) An adjustment for
nonfilers is based on data from an exact-match study conducted by
Census. (8) These adjustments are considered less reliable than other
adjustments because of the infrequency with which the TCMP, TCMP-IRP,
and exact-match studies are conducted and the need to approximate the
industry distributions.
[ILLUSTRATION OMITTED]
Auxiliary services. An auxiliary is an establishment that provides
services that may not be part of a company's main industry. BEA
adjusts the data to allocate the auxiliary's expenses into the
proper industry. For example, in the benchmark I-O accounts, a
management services establishment within a pharmaceutical company would
be broken out of the pharmaceutical industry and added to management
services. For a given auxiliary, the Census Bureau provides expense data
tabulated for the sector of the auxiliary. For each industry, the ratio
of industry-level payroll to sector-level payroll is assumed to be the
same as industry-level expense to sector-level expense. Additions to
intermediate inputs in a given industry served are offset by reductions
to gross operating surplus in that industry. The adjustment for
auxiliary services is considered more reliable than the misreporting and
nonfiling adjustment but less reliable than the adjustment for
nonemployer expenses because of the source data and the assumption
regarding payroll and other expenses.
Annual Industry Accounts
The annual industry accounts provide a time series of estimates for
gross output, intermediate inputs, and value added by industry. Like the
benchmark I-O accounts, they also include make and use tables. Unlike
the benchmark I-O accounts, intermediate inputs in the annual industry
accounts are derived as a residual: Gross output less compensation of
employees, taxes on production and imports less subsidies, and gross
operating surplus. (9) Gross operating surplus in the annual industry
accounts is derived from gross operating surplus estimates in the most
recent benchmark I-O accounts, extrapolated forward using annual
measures of gross operating surplus based on gross domestic income
(GDI). These GDI-based estimates of gross operating surplus are also
used as inputs for the new reconciliation model. Thus, estimates of 2002
gross operating surplus used in the reconciliation model are different
from estimates of gross operating surplus published in the 2002 annual
industry accounts.
GDI-based estimates of gross operating surplus for private, nonfarm
industries are derived using data from the IRS, other data sources, and
BEA adjustment methodologies. (10) Approximately 47 percent of private,
nonfarm industries' gross operating surplus comes from IRS data, 14
percent from misreporting and nonfiling adjustments, 22 percent from
concept and coverage adjustments, and 17 percent from other data and
related adjustments.
Misreporting and nonfiling. Similar to the benchmark I-O accounts,
BEA makes a misreporting and nonfiling adjustment for noncorporate
business income-tax based source data in the annual industry accounts
based on data from the TCMP, TCMP-IRP, and exact-match studies. Thus,
the misreporting and nonfiling adjustment is considered less reliable
than other types of adjustments. These adjustments are given the same
reliability indicator as they are in the benchmark I-O accounts.
Concepts and coverage. Concept adjustments are designed to convert
tax accounting-based concepts from IRS data to economic accounting-based
concepts consistent with national accounts. Concept adjustments include
the removal of capital gains and dividends from business income.
Coverage adjustments are designed to include the activities of entities
that contribute to gross domestic product but are not required to file a
return with the IRS. Coverage adjustments include adding income earned
by Federal Reserve banks and imputing net income for owner-occupied
housing. While concept and coverage adjustments are considered more
reliable overall than the misreporting and nonfiling adjustment,
reliability varies by specific adjustment. Adjustments based on
administrative data are considered more reliable than other adjustments,
except those based on Economic Census data. Adjustments based on survey
data are considered less reliable than those based on administrative
data.
Company-establishment. Because the tax data used to make some
estimates are classified on a company basis, BEA makes an adjustment to
convert these data to an establishment basis. A company may consist of
several establishments, each of which operates in a different industry.
This adjustment shifts gross operating surplus from one industry to
another with no impact on total gross operating surplus.
The company-establishment adjustment is limited to three of the
income components of gross operating surplus derived from corporate
income tax data: profits before tax, the capital consumption allowance,
and net interest. The adjustment is based on employment data from the
Census Bureau that relates employment by industry on a company basis and
an establishment basis. The adjustment assumes that profits before tax,
capital consumption allowance, and net interest are the same per
employee for all establishments performing the same activity, regardless
of the company to which that establishment belongs. The
company-establishment adjustment is considered less reliable than most
concept and coverage adjustments but more reliable than misreporting and
nonfiling adjustments.
Reconciling Gross Operating Surplus in a Balanced Input-Output
Framework
The objective of the reconciliation model is to adjust intermediate
inputs in the benchmark I-O accounts and gross operating surplus in the
annual industry accounts so that the industry gross operating surplus
estimates of each are equal, subject to the accounting constraints of
the I-O framework (chart 1, table 1). For the benchmark I-O accounts,
the adjusted estimates are published. For the annual industry accounts,
recall the caveat that the adjusted estimates are not published.
However, both accounts will be adjusted as part of the next
comprehensive revision.
Initial estimates and reliability measures
An essential feature of BEA's new model is that it adjusts
estimates in a way that takes the reliability of each initial estimate
into account, subject to the accounting constraints of the I-O
framework. (11) A more detailed, mathematical description of the model
is available in the BEA paper that was mentioned at the end of the
introduction to this article. In particular, the reconciliation model
makes adjustments to initial estimates based on the strengths and
weaknesses of the data that underlie those estimates. Initial estimates
that are considered relatively weak are adjusted more than initial
estimates that are considered relatively reliable.
Specific adjustments are made to the following:
* Intermediate inputs in the benchmark I-O accounts. The improved
estimates of intermediate inputs also improve the estimates of gross
operating surplus because the gross operating surplus is derived as a
residual: Gross output less intermediate inputs, compensation and taxes
on production and imports less subsidies.
* Gross operating surplus in the annual industry accounts. The
improved estimates of gross operating surplus will likewise improve the
residual estimates of intermediate inputs.
In both cases, final intermediate input estimates and gross
operating surplus estimates are informed by the reliability of the
underlying data used to produce the initial estimates. In this way, the
data from each account is used to improve estimates in the other.
To take reliability of the underlying data into account, the model
requires reliability indicators for each initial estimate. These
indicators gauge the strength of the underlying data and are used in
part to weight the adjustment. Initial estimates are assigned a
reliability indicator from two sources:
* Source data providers. Coefficients of variation, which measure
sampling errors, are available for some source data provided by the
Census Bureau and the IRS. (12) These coefficients range from 0 to 1,
with 0 denoting the highest reliability and 1 denoting the lowest
reliability. About 23 percent of the total value of benchmark
intermediate inputs and about 47 percent of the total value of annual
gross operating surplus estimates had coefficients of variation from the
source data provider.
* BEA. For estimates that do not have coefficients of variation,
BEA assigns reliability indicators (table 2). These reliability
indicators also range from 0 to 1 according to a rubric developed by BEA
economists. The rubric considers the data on which initial estimates are
based and the adjustments to those estimates. The adjustments are
designed to correct nonsampling errors in the underlying data (for
example, misreporting). However, the adjustments are themselves subject
to nonsampling errors, and the reliability of adjustments, as determined
by BEA economists, vary widely.
Chart 3 shows the distributions of BEA-assigned reliability
indicators and coefficients of variation for all industries. According
to the rubric, estimates based on Economic Census data with no
adjustments are assigned a zero, which means they are considered the
most reliable. Approximately 13 percent of the total value of
intermediate input estimates receives this indicator. No gross operating
surplus estimates are assigned a reliability indicator of zero because
none were based on Economic Census data without adjustments. Estimates
based on Economic Census data with adjustments are assigned an indicator
of 0.10, the second most reliable ranking.
Of all estimates with BEA-assigned reliability indicators, almost
45 percent of the intermediate input value and 40 percent of the gross
operating surplus value are assigned reliability indicators of 0.10 or
0.35. Reliability indicators of 1 are assigned to a higher percent of
the gross operating surplus value because the misreporting adjustment
plays a larger role in the annual industry accounts.
Constraints
The reconciliation model does not allow all estimates to adjust.
Estimates of final uses, gross output, and the compensation and tax
components of value added from the benchmark I-O accounts are fixed.
(13) The system of constraints also incorporates the accounting
identities upon which the table is built. Thus, the system of
constraints ensures that intermediate and final uses of each commodity
equals the commodity's supply, the sum of each industry's
intermediate inputs and value added equals the industry's gross
output, and that total value added equals total GDP. In this way, the
model provides a tool for balancing the benchmark use table.
[ILLUSTRATION OMITTED]
Technology
In previous years, attempts to develop robust reconciliation models
have been stymied partly because of a lack of adequate software. For the
2002 benchmark I-O accounts, the reconciliation model was executed using
the CPLEX solver in the Generalized Algebraic Modeling System (GAMS).
GAMS is a flexible optimization software package designed to handle
large mathematical programming problems. CPLEX is a GAMS solver with
solution algorithms for linear, quadratically constrained, and
mixed-integer problems. The CPLEX solver automatically chooses the
optimal combination of algorithms to efficiently solve the particular
model specified. Alternatively, GAMS also allows users to adjust tuning
parameters in order to set algorithmic options. The reconciliation model
is solved with the combination of algorithms chosen by GAMS.
The scale of the model is suggested by the number of variables
involved. At the most disaggregated industry level (987 industries and
8,910 items), the model in theory contains 8,795,157 variables to be
solved and 9,963 constraints to be satisfied. However, in practice, the
intermediate input portion of the benchmark use table is not fully
populated, resulting in fewer variables to be solved.
Results
More than 50 percent of intermediate input estimates in the
benchmark I-O accounts and gross operating surplus estimates in the
annual industry accounts were adjusted less than 5 percent from their
initial values (chart 4). As expected, the majority of these small
adjustments were made to estimates derived from Economic Census data,
survey data, and administrative data. Approximately 13 percent of
intermediate input estimates in the benchmark I-O accounts and
approximately 6 percent of gross operating surplus estimates in the
annual industry accounts were adjusted more than 50 percent from their
initial values. As expected, the majority of these adjustments were made
to estimates derived from adjustments based on analyst judgment and
misreporting and nonfiling adjustments. In addition, the majority of
these large adjustments were for small initial values.
To assess results at a more aggregate level, an aggregate industry
reliability measure was calculated for initial intermediate input
estimates in the benchmark I-O accounts and for initial gross operating
surplus estimates in the annual industry accounts. The results of the
reconciliation results in a given industry are expected to favor the
initial value with the smaller aggregate reliability measure, and
indeed, this pattern generally holds for all industries.
[GRAPHIC 4 OMITTED]
History of the Reconciliation Model
In a series of papers that began in 1942, the economist Richard
Stone advocated a framework to improve the accuracy of independent
estimates of national income and expenditures based on the reliability
of the data used to construct the statistics. The Bureau of Economic
Analysis (BEA) has drawn upon this approach to reconcile its benchmark
input-output (I-O) and annual industry accounts, culminating with the
reconciliation of the 2002 accounts, which is presented in this article.
Researchers revised the Stone method to facilitate its
implementation (Byron 1978; van der Ploeg 1982, 1984), but Federal
agencies responsible for producing national economic accounts have
generally not implemented the method. One reason for this has been a
lack of technology that is typically required to solve the complex
systems of equations faced by Federal agencies. Another reason has been
a lack of information regarding the relative reliabilities of underlying
data used to construct national accounting statistics.
While some agencies have resolved the latter challenge with
subjective measures of relative reliabilities (Mantegazza and Pisani
2000; Moyer et al. 2004a, 2004b), a lack of adequate technology has
until recently stymied implementation of the Stone method (Nicolardi
2000; Tuke and Aldin 2004). In a recent study, BEA economist Baoline
Chen (2006) addressed both challenges by building an empirical model
based on the Stone method and incorporating statistical measures of
relative reliability in the model to reconcile and balance estimates.
BEA-assigned reliability indicators were first used to reconcile
the 1997 accounts. For those accounts, value added in the benchmark I-O
accounts was reconciled with value added in the annual industry accounts
using a weighted average of the initial industry value-added estimates
in each set of accounts. The weights were based partly on the
reliability of the data from which value added was derived.
In particular, the weights in the benchmark I-O accounts were based
on the percent of industry intermediate input estimates and industry
gross output estimates that were derived from the 1997 Economic Census.
The weights in the annual industry accounts were based on the
reliability and size of the adjustments used to convert enterprise-based
income data to an establishment basis and the percent of an
industry's value added that is derived from proprietors'
income. From these criteria, industry value-added reliability measures
were calculated for each set of accounts, and these reliability measures
were used to calculate the weights to adjust initial industry
value-added estimates. Value-added estimates with smaller reliability
measures had greater weights.
The reconciliation of the 2002 accounts built on the previous
reconciliation. In particular, the new, more transparent reconciliation
model relies on a generalized least squares framework that provides a
solid statistical foundation for the adjusted estimates. The methodology
to assign reliability weights has also been improved by drawing upon
external data.
References
Brown, Robert E., and Mark J. Mazur. 2003. "IRS's
Comprehensive Approach to Compliance Measurement)' National Tax
Journal 56 (September): 689-700.
Byron, Ray P. 1978. "The Estimation of Large Social Account
Matrices." Journal of the Royal Statistical Society, Series A 141
(March): 359-367.
Chen, Baoline. 2006. "A Balanced System of Industry Accounts
for the United States and Structural Distribution of Statistical
Discrepancy." BEA working paper WP2006-8;
<www.bea.gov/papers/working_papers.htm>.
Lawson, Ann M., Kurt S. Bersani, Mahnaz Fahim-Nader, and Jiemin
Guo. 2002. "Benchmark Input-Output Accounts of the United States,
1997." SURVEY OF CURRENT BUSINESS 82 (December): 19-43.
Lawson, Ann M., Brian C. Moyer, Sumiye Okubo, and Mark A. Planting.
2006. "Integrating Industry and National Economic Accounts: First
Steps and Future Improvements?' In A New Architecture for the U.S.
National Accounts, edited by Dale W. Jorgenson, J. Steven Landefeld, and
William D. Nordhaus, 215-261. Chicago: University of Chicago Press.
Mantegazza, Susanna, and Stefano Pisani. 2000. "Present
Practices and Future Developments." Paper presented at the 13th
International Conference on Input-Output Techniques, University of
Macerata, Italy, on August 21-26.
Moyer, Brian C., Mark A. Planting, Mahnaz Fahim-Nader, and Sherlene
K. S. Lum. 2004a. "Preview of the Comprehensive Revision of the
Annual Industry Accounts: Integrating the Annual Input-Output Accounts
and Gross-Domestic-Product-by-Industry Accounts." SURVEY OF CURRENT
BUSINESS 84 (March): 38-51.
Moyer, Brian C., Mark A. Planting, Paul V. Kern, and Abigail M.
Kish. 2004b. "Improved Annual Industry Accounts for 1998-2003:
Integrated Annual Input-Output Accounts and
Gross-Domestic-Product-by-Industry Accounts." SURVEY OF CURRENT
BUSINESS 84 (June): 21-57.
Nicolardi, Vittorio. 2000. "Balancing Large Accounting
Systems: AN Application to the 1992 Italian I-O Table." Paper
presented at the 13th International Conference on Input-Output
Techniques, University of Macerata, Italy, on August 21-26.
Ploeg, Frederick van der. 1984. "Generalized Least Squares
Methods for Balancing Large Systems and Tables of National
Accounts." Review of Public Data Use 12 (January): 17-33.
Ploeg, Frederick van der. 1982. "Reliability and the
Adjustment of Sequences of Large Economic Accounting Matrices."
Journal of the Royal Statistical Society, Series A 145 (February):
169-194.
Stanley-Allen, Karla L., Nicholas R. Empey, Douglas S. Meade,
Stanislaw J. Rzeznik, Mary L. Streitwieser, and Monica S. Strople. 2005.
"Preview of the Benchmark Input-Output Accounts for 2002."
SURVEY OF CURRENT BUSINESS 85 (September): 66-77.
Stewart, Ricky L., Jessica Brede Stone, and Mary L. Streitwieser.
2007. "U.S. Benchmark Input-Output Accounts, 2002." SURVEY OF
CURRENT BUSINESS 87 (October):19-48.
Stone, Richard. 1984. "Balancing the National Accounts: The
Adjustment of Initial Estimates: A Neglected Stage in Measurement."
In Demand, Equilibrium and Trade: Essays in Honor of Ivor F. Pearce,
edited by A. Ingham and A.M. Ulph, 191-212. New York: St. Martin's Press.
Stone, Richard. 1976. "The Development of Economic Data
System." In Social Accounting for Development Planning with Special
Reference to Sri Lanka, edited by G. Pyatt et al. Cambridge: Cambridge
University Press.
Stone, Richard. 1975. "Direct and Indirect Constraints in the
Adjustment of Observations." In Nasjonalregnskap, Modeller og
Analyse: Essays in Honour of Off Aukrust. Oslo: Statistisk Sentralbyra.
Stone, Richard. 1961. Input-Output and National Accounts. Paris:
Organization for European Economic Cooperation.
Stone, Richard. 1968. "Input-Output Projections: Consistent
Prices and Quantity Structures." L'Industria 2 (February):
212-224.
Stone, Richard. 1970. Mathematical Models of the Economy and Other
Essays. London: Chapman and Hall.
Stone, Richard, James E. Meade, and David G. Champernowne. 1942.
"The Precision of National Income Estimates." The Review of
Economic Studies 9 (February): 111-125.
Tuke, Amanda, and Vanna Aldin. 2004. "Reviewing the Methods
and Approaches of the U.K. National Account." Economic Trends 602
(January): 47-57.
Yuskavage, Robert E. 2000. "Priorities for Industry Accounts
at BEA." Paper presented at the BEA Advisory Committee Meeting on
November 17; <www.bea.gov/papers/index.htm>.
(1.) For more information regarding the 2002 benchmark I-O
accounts, see Stewart et al. (2007).
(2.) For more information regarding BEEs integration initiative,
see Yuskavage (2000), Moyer et al. (2004a, 2004b), and Lawson et al.
(2006). For this article, Baoline Chen, Karen Horowitz, Douglas S.
Meade, Mark A. Planting, and George M. Smith provided early advice.
Sumiye Okubo, Erich H. Strassner, Mary L. Streitwieser, and Robert E.
Yuskavage also provided helpful comments.
(3.) In a series of papers beginning with Stone et al. (1942),
Stone (1961, 1968, 1970, 1975, 1976, and 1984) advocates a GLS framework
to improve the economic and statistical accuracy of independent
estimates of national income and expenditures based on the reliabilities
of underlying data used to construct the estimates.
(4.) While other types of adjustments are made to source data for
intermediate inputs, these three types of adjustments capture the
majority of the dollar value of adjustments made in the benchmark I-O
accounts.
(5.) The Census Bureau provides receipts for nonemployers at a more
aggregate level than receipts for employer establishments. To distribute
receipts for nonemployers to the appropriate industry level, BEA uses a
ratio of receipts for small-employer establishments by industry to total
receipts for all small employers from the Economic Census. However,
these adjustments do not affect the data for the reconciliation because
gross output is fixed in the reconciliation model.
(6.) According to IRS compliance studies, three components
contribute to a tax gap between the amount taxpayers should pay and the
amount taxpayers actually pay in a timely manner: Nonfiled returns,
underreported income, and underpaid taxes (Brown and Mazur 2003).
(7.) Discontinued in the early 1990s, the TCMP was an audit program
designed to study compliance patterns and levels of misreporting among
sole proprietors. The TCMP-IRP was designed to compare the results of
the TCMP audits to information returns fried with the IRS in order to
capture misreporting that TCMP auditors failed to find.
(8.) An exact-match study compares records from the Current
Population Survey (CPS) to records from the IRS in order to identify and
estimate nonfiled income for individuals who report income in the CPS
but do not file a return with the IRS.
(9.) Gross output in the annual industry accounts is calculated
using annual survey data to extrapolate gross output from the make table
in the most recent benchmark I-O accounts. Estimates of the compensation
and tax components of value added in the annual industry accounts are
derived from the GDI components of the NIPAs.
(10.) Gross operating surplus estimates for farm and general
government industries and the owner-occupied housing portion of gross
operating surplus estimates in the real estate industry are not
reconciled using the current model. Thus, a discussion of the data and
methodologies used to prepare estimates for these industries is outside
the scope of this article.
(11.) While Chen's (2006) model adjusts gross output,
intermediate inputs, and all components of value added in the benchmark
I-O accounts, the current model only adjusts intermediate inputs and the
gross operating surplus component of value added.
(12.) Coefficients of variation are available for all IRS
estimates. Because sampling errors apply to surveys and not censuses,
coefficients of variation are available from the Census Bureau for the
expense data from the Business Expenses Survey and purchased services
from the Annual Survey of Manufactures portion of the Census of
Manufactures. Purchased services include the following expense
categories: Accounting, auditing, and bookkeeping; advertising;
communications; computer services; legal services; management,
consulting, and administrative services; other expenses; refuse removal;
repairs and maintenance of buildings and machinery; and taxes and
license fees.
(13.) Estimates of the commodity distribution of final use
categories are reconciled through a negotiation process with the NIPAs.
Industry and commodity estimates of gross output are determined in the
make table of the benchmark I-O accounts, which was published in
September 2005 (Stanley-Allen et al. 2005). Industry estimates of the
compensation and tax components of value added are derived from Census
Bureau data and scaled to match the NIPA totals.
Table 1. Effects of the Reconciliation Model on Benchmark I-0 and
Annual Industry Accounts
Gross Intermediate
output inputs
Benchmark I-0 No change Each estimate is
accounts adjusted based on the
reliability of the
underlying data
Annual industry No change Each estimate is derived
accounts as a residual, so each is
adjusted because gross
operating surplus
estimates were adjusted
Compensation Taxes
Benchmark I-0 No change No change
accounts
Annual industry No change No change
accounts
Gross operating surplus
Benchmark I-0 Each estimate is derived
accounts as a residual, so each
estimate is adjusted
because intermediate
inputs are adjusted
Annual industry Each estimate is
accounts adjusted based on the
reliability of the
underlying data
Table 2. Rubric for Assigning Reliability Indicators
to Initial Estimates
Reliability
indicator Source of estimate
0 Economic Census data with no adjustments
0.10 Economic Census data with adjustments
Survey data with no adjustments
Concept and coverage adjustments based solely on
administrative data
Nonemployer adjustments
0.35 Survey data with adjustments
Trade association data
Concept and coverage adjustments based on survey data
Auxiliary service adjustments
0.65 Company-establishment conversion adjustment
Adjustments based on a combination of analyst
judgment and external source data
1 Misreporting adjustments
Adjustments based solely on analyst judgment
