Full text of Bureau of Economic Analysis (BEA) News Releases : Preliminary Study Shows Research and Development Contribution to Economic Growth, BEA 06-44 : Full Report

View original document

The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.

Bureau of Economic Analysis/National Science Foundation
R&D SATELLITE ACCOUNT:
PRELIMINARY ESTIMATES
Sumiye Okubo, Carol A. Robbins, Carol E. Moylan, Brian K. Sliker, Laura I. Schultz, and Lisa
S. Mataloni
Bureau of Economic Analysis (BEA)

9/28/2006

This paper presents preliminary estimates of a satellite account on research and development
(R&D) for the years 1959–2002 and of the impact that the treatment of R&D as investment
would have on Gross Domestic Product (GDP) and Gross Domestic Income (GDI). The satellite
account allows BEA to evaluate the effects of changing the asset boundary to include R&D as an
intangible asset. It provides estimates of R&D in the economy on a funder basis. Final estimates
are scheduled to be released in the fall of 2007. This work lays the ground work for
incorporating R&D as an asset into the national income and product accounts, a proposal
currently under consideration for the internationally accepted standard for national accounts,
the System of National Accounts.

Table of Contents
I. Introduction

II. Conceptual issues in the treatment of R&D as investment

III. Estimation Methodology

IV. Assumptions for the Satellite Account and Alternative Scenarios

V. Looking forward to the 2007 R&D Satellite Account

VI. Summary and Conclusion

References

Tables and Figures
Figure 1. Nominal R&D Investment as a Percent of GDP

Figure 2. Nominal R&D Investment as a Percent of GDP, funded by Business and Government
7
Table 1. Selected Summary Measures of R&D

Table 2. How the National Accounts Change when R&D is treated as Investment
13
Table 3. Summary of Estimated Rates of Return to Private R&D

Table 4. Comparison of Key Assumptions of R&D Satellite Accounts

Table 5. Assumptions for the Scenarios in the R&D Satellite Account

Table 6 Overall Impact of Capitalized R&D on GDP Level

Table 7. Impact on Gross Private Domestic Investment and the Saving Rate

Table 8. Contribution of R&D investment to growth in adjusted GDP

Appendix Tables
Definitions and General Table Notes

A-1

Table 1.1 Real Gross Domestic Product with and without R&D Adjustments 1959–2002 A-3
Table 1.2 Gross Domestic Product with and without R&D Adjustments 1959–2002

A-5

Table 1.3 Gross Domestic Income with and without R&D Adjustments 1959–2002

A-7

Table 1.4 National Savings Rate (% of GNP) 1959–2002

A-9

Table 1.5 Aggregate Current Cost Returns to R&D Assets 1959–2002

A-10

Table 2

Steps to Transform Survey Data to R&D Investment in Current Dollars

A-11

Table 3.1 Historical Cost Investment Flows in R&D Assets by Funder 1959–2002

A-12

Table 3.2 Real Investment Flows in R&D Assets by Funder 1959–2002

A-14

Table 3.3 Historical Cost Net Stock of R&D Assets by Funder 1959–2002

A-16

Table 3.4 Current Cost Net Stock of R&D Assets by Funder 1959–2002

A-18

Table 3.5 Real Net Stock of R&D Assets by Funder 1959–2002

A-20

Table 3.6 Historical Cost Depreciation of R&D Assets by Funder 1959–2002

A-22

Table 3.7 Current Cost Depreciation of R&D Assets by Funder 1959–2002
A-24
Table 3.8 Real Depreciation of R&D Assets by Funder 1959–2002

A-26

Table 4.1 Historical Cost Investment in R&D Assets by Performer 1959–2002

A-29

Table 4.2 Real Investment in R&D Assets by Performer 1959–2002
A-30
Table 5.1 Aggregate Cost-Based Price Indexes for R&D Investment 1959–2002

A-32

Table 5.2 Alternative Scenario Price Indexes for R&D Investment 1959–2002

A-35

Table 6

Deflation Summary

A-36

Table 7

Seven Account Summary of Impact on GDP

A-40

Table 8

Dimensions of International R&D

Acknowledgements:
Sumiye Okubo, Associate Director for Industry Accounts, and Carol A. Robbins, directed the
preliminary R&D satellite account project. Carol E. Moylan, Chief of the National Income and
Wealth Division, supervised the preparation of the estimates. Brian K. Sliker prepared the
macroeconomic estimates.
John W. Drewry, Lisa S. Mataloni, Jose Rodriguez-Solis, Andrew G. Schmidt, and Laura I.
Schultz prepared data analysis and other review materials. David B. Wasshausen, Erich H.
Strassner, and Gabriel W. Medeiros provided additional technical assistance with the estimates.
Claudia M. Molinar provided research assistance.
J. Steven Landefeld, Rosemary D. Marcuss, Dennis J. Fixler, Ralph H. Kozlow, Brent R.
Moulton, Bruce T. Grimm, Charles I. Mead, and William J. Zeile provided valuable comments
throughout the project. Ann M. Lawson provided comments on the tables and drafts. Lawrence
P. McNeil and Yvon Pho contributed to the early stages of the project.
John E. Jankowski, Director, R&D Statistics Program, and Francisco A. Moris of the National
Science Foundation contributed guidance and support throughout the project, prepared
supplementary data and analysis, and provided a detailed methodological and data review of the
satellite account. R&D data and discussion and consultation were provided by the members of
the Division of Science Resources Statistics (SRS) staff at the National Science Foundation
(NSF). Funding for this project was provided by NSF/SRS.

A-42

Section 1. Introduction
Economists have long sought a better understanding of research and development
as a source of innovation and growth and therefore economic well-being. This interest
was sparked in part by Robert Solow’s path-breaking productivity work in the late 1950s,
which showed that much of economic growth cannot be attributed to increases in capital
and labor.1 Since then, researchers have suggested various ways to account for the
unexplained portion of economic growth. These efforts include developing improved
theoretical underpinnings to growth models as well as better measures of technologydriven economic activity, intangible assets, real output of industries, and the so-called
knowledge economy.
BEA’s efforts have focused on improved measurement of economic output,
prices, and growth. This paper provides a set of preliminary estimates of treating R&D as
an investment, and details the potential impact of this treatment on the economy, notably
on such measures as gross domestic product, investment, and saving. These estimates are
presented as a satellite account—a set of economic estimates presented in a framework
that provides detail about R&D activity that is not reflected in BEA’s core economic
accounts.
This satellite account adjusts the accounting conventions of BEA’s core accounts
to test the impact of changing the scope of investment—in this case, recognizing R&D as
an asset. While this satellite account does not affect BEA’s official measures of GDP, it
provides an opportunity to work out new methodologies that may be incorporated into the
1

Robert Solow (1957).

accounts in the future. Using R&D expenditure data from the National Science
Foundation (NSF), BEA has developed estimates of R&D investment, R&D capital
stock, and the resulting macroeconomic impacts for 1959 to 2002. A series of revised
estimates and a final report will be released in September of 2007.
These estimates measure solely the direct impact of R&D investment. They do
not include the effect of R&D beyond those industries that conducted the R&D. For
example, the increase in output and productivity of the computer industry associated with
a new R&D-based innovation are included in the estimates, but the increase in output and
productivity of the banking industry associated with using more efficient computers are
not. The banking-industry effect is included in the GDP, but it is not attributed to R&D
investment in these estimates.
Treating R&D as a business investment, rather than an expense, has a substantial
impact on the measures of GDP and its components. This paper provides preliminary
estimates.2 Highlights include the following:
•

Recognizing R&D as investment increases the level of GDP by an average of
2.6 percent between 1959 and 2002.

•

R&D investment and the income flows arising from accumulated R&D capital
accounted for 4.6 percent of growth in real (adjusted for price change) GDP
between 1959 and 2002, and between 1995 and 2002, its contribution to
growth rose to 6.7 percent.

•

Estimates of gross private domestic investment (in current dollars) in 2002
would be more than 11 percent, or $178 billion, higher after recognizing R&D

The results reported in the conclusions of this report are based on estimates that value real (inflation
adjusted) R&D at output prices of products produced by R&D-intensive industries.

as investment; and the 2002 national saving rate would be 16 percent, instead
of 14 percent.
This paper marks another step in BEA’s ongoing commitment to adapt its
measures of economic activity to the changes in the structure of the economy. One of the
first advances made was the result of work with IBM to develop hedonic, or qualityadjusted, prices for measuring real investment in computer products and measuring the
impact on real output. The Bureau of Labor Statistics (BLS) has extended this work to a
wide range of other goods and services. Currently, approximately twenty percent of real
GDP is deflated using quality-adjusted price indexes that rely at least in part on hedonic
methods.3 This expanded coverage has been used by BEA to improve its estimates of
real GDP and the value of real output of services industries that use information
technology (IT).
A second advance was to begin to include one of the most important types of
intangible investments into the national accounts; in 1999 BEA introduced into its
estimate of GDP the investment flows of computer software. Inclusion of computer
software helped explain a significant share of the resurgence in economic growth in the
last decade. Between 1995 and 2002, software’s contribution to the growth in real GDP
was more than 5.1 percent. Between 1973 and 1994, its contribution was 3.0 percent.
These efforts have enabled researchers to decrease their estimates of the amount
of economic growth that is not explained by the contributions of labor, capital, and
intermediate inputs. In 1967, the unexplained portion of growth was approximated at 50
percent.4 The work by BEA, BLS, and others has provided the basis for better measures

3
4

Moulton, (2001).
Griliches and Jorgensen (1967).

of IT-related improvements and their contributions to growth. Indeed, researchers such as
Triplett and Bosworth have used improved BEA data on real industry output (GDP by
industry) to show that services-producing industries, “have emerged as engines of
economic growth,” during the resurgence in economic growth and productivity over the
past decade.5 Improved measures of IT have also been used by analysts of the sources of
growth to reduce their estimates of multifactor productivity—the residual portion of
growth after all the contributions of labor, capital, and intermediate inputs have been
explained.6
Because intangible assets are increasingly important components of the
knowledge economy, BEA is interested in expanding the available data that will allow
analysts to understand their role in production and economic growth. BEA is currently
engaging in research that might allow it to develop prototype accounts for health care,
human capital, and education.
A logical next step for BEA is to consider recognizing R&D in addition to
computer software in its measures of intangible fixed investment. Although R&D
activities provide private and social benefits that can be long lasting, R&D expenditures
are not currently recognized as investment in the national accounts of the United States or
other nations. Spending for R&D is more like investment than like current consumption
because it uses resources to create a future stream of benefits that extends beyond the
current period. To the extent that the benefits of R&D can be appropriated by its owner,
R&D expenditures have qualities of an economic asset.

5
6

Triplett and Bosworth (2004).
Jorgenson, Mun, and Stiroh (2005).

As part of its commitment to improving measurement of intangibles and the
knowledge economy, BEA plans to capitalize R&D in its core accounts in the future, if
resources are available. Current plans call for BEA to incorporate R&D into the inputoutput (I-O) accounts in 2012 and into the national income and product accounts (NIPAs)
around 2013.
This effort builds on the earlier work at BEA on developing R&D satellite
accounts by Carson, Moylan, and Grimm (1994) and Fraumeni and Okubo (2005).
However, an important innovation in BEA’s current satellite account is the presentation
of the investment flows and resulting stocks of R&D assets based on the funder of the
R&D. This funder-based presentation differs from the earlier versions of BEA’s R&D
satellite accounts, which were based on the performer of the R&D. These earlier versions
were created on the performer basis because the most detailed source data are available
on this basis. However, in order to assign income flows to the appropriate sector in
BEA’s national accounts, BEA needs to identify the owners of these productive assets.
Because existing R&D survey data are insufficient to completely identify ownership,
BEA has two immediate choices, the funder and performer perspectives. In this satellite
account, BEA adopts the funder perspective to represent ownership. This choice reflects
the view that the funder of R&D gains the direct economic benefits.

A Brief Overview of R&D in the U.S. Economy
Domestic R&D expenditures by business, currently treated as an expense, are
only partially identifiable in BEA’s accounts. In the I-O accounts, the identifiable portion
is based on data from the Census Bureau on establishments classified in the Scientific
Research and Development Services industry (NAICS 5417). In BEA’s GDP-by-industry

accounts, estimates for the value-added of this industry are included in a broader sector,
miscellaneous professional, scientific, and technical services. While Federal government
purchases of R&D are included in the output of colleges and universities in the I-O table,
they are not separately identified. Similarly, only a portion of R&D expenditures is
identifiable within the NIPAs; these are Federal purchases of R&D and personal
consumption expenditures (of foundations and nonprofit research). BEA’s estimates of
international trade in services provide measures of exports and imports of R&D services;
these data are described more fully in Section V.
The R&D satellite account presented in this paper enhances BEA’s measures of
R&D activity with more detailed information. The satellite account estimates show
nominal investment in R&D totaled $276.5 billion dollars in 2002 and the ratio of this
investment to GDP, adjusted for the effect of R&D, was 2.6 percent. Figure 1 shows this

Figure 1. Nominal R&D Investment as a Percent of
GDP
3.5%
3.0%
2.5%
2.0%
1.5%
1.0%
0.5%
0.0%
1959

1969

1979

1989

1999

ratio between 1959 and 2002. The ratio of R&D to GDP rose in the 1960s, as the U.S.
invested in space-related technologies, and fell in the 1970s. This ratio picks up again in
the early 1980s, followed by another drop at the end of the century.

Figure 2. Nominal R&D Investment as a Percent of
GDP, funded by Business and Government
2.5%
2.0%
1.5%

Business

1.0%

Government

0.5%
0.0%
1959

1969

1979

1989

1999

Figure 2 shows the ratio of R&D investment to GDP for R&D funded by business
and by government. (The small share funded by private nonprofit institutions is not
shown here.) The declining role of the government in the funding of R&D and the
increasing importance of business funding of R&D are evident. During the early era of
space exploration in the mid-1960s, the Federal government invested a value equivalent
to more than 2 percent of GDP in R&D. The government’s contribution to total R&D
investment was at its highest in the middle of the 1960s, when it funded almost threequarters of all R&D investment. Government investment has declined steadily since the
1960s, reaching its low in 2000, when government investment totaled about 0.8 percent
of GDP. Business investment has been steadily increasing as a percentage of GDP; by

1982, business funded more investment in R&D than government. In 2000, the business
sector investment in R&D equaled 1.8 percent of GDP.
The recent growth in business R&D expenditures has been driven by four of the
most R&D intensive industries: Chemicals (including drugs and medicine), machinery
(including computers), electrical equipment (including communication equipment) and
transportation (including aircraft). The NSF report of Science and Engineering Indicators,
2006, provides an in-depth analysis of R&D trends by performers in these industries and
many other dimensions of R&D activity.7

Table 1. Selected Summary Measures of R&D
(Percent based on current-dollar measures)
1960

1965

1970

1975

1980

1985

1990

1995

1998

1999

2000

2001

2002

R&D net fixed capital stock as a % of total
R&D adjusted net fixed capital stock

3.3

4.1

4.2

3.4

2.8

3.3

3.7

3.3

3.0

2.9

Funder-based R&D investment as a % of
R&D adjusted GDP
Business
Government
Nonprofit Institutions

1.1
1.5
0.0

0.7
2.0
0.0

1.0
1.5
0.0

0.9
1.2
0.0

1.1
1.1
0.0

1.5
1.2
0.0

1.4
1.2
0.1

1.4
1.0
0.1

1.6
0.9
0.1

1.7
0.9
0.1

1.8
0.8
0.1

1.7
0.9
0.1

1.6
0.9
0.1

Funder-based R&D investment as a % of
total R&D
Business
Government
Nonprofit Institutions

42.0
57.0
1.0

25.4
73.3
1.3

38.9
59.6
1.5

42.4
56.0
1.6

48.4
49.9
1.7

55.1
43.2
1.7

52.5
45.2
2.3

56.3
41.1
2.5

61.8
35.7
2.5

64.2
33.3
2.5

66.7
30.8
2.4

65.1
32.4
2.5

61.8
35.5
2.7

71.4
18.6
10.0

74.3
16.6
9.1

74.8
16.0
9.2

75.2
15.3
9.5

73.3
16.5
10.2

70.9
18.2
10.9

Performer-based R&D investment as a %
of total R&D
Business
76.8
70.0
68.9
67.6
69.9
73.6
72.5
Government
17.2
21.4
22.3
23.8
21.3
18.4
19.1
Nonprofit Institutions
6.0
8.6
8.8
8.6
9.4
8.0
8.4
Calculations based on:
Appendix tables 1.2, 3.1, 3.4, and 4.1
BEA fixed asset table 1.1 Current- Cost Net Stock of Fixed Assets and Consumer Durable Goods

Table 1 provides a summary of investment trends shown by BEA’s estimates for
funders and performers of R&D. In this table, public universities and colleges are

These data are available based on expenditures for R&D: National Science Board (2006).

included as part of government, and private universities and colleges are included as part
of nonprofit institutions. The share of total R&D investment by business and government,
as estimated on a performer basis, has not changed as much as the funder-based view of
investment indicates—the funder-based view shows the decline in government funded
R&D relative to business and nonprofit funded R&D over the entire range of years.

Measuring R&D as Investment
Measuring the output of R&D activity presents well-known challenges. For its
core accounts, BEA’s economic measures are primarily based on observable prices and
quantities for goods sold in the market. With data on business revenues based on prices
and quantities of goods sold, BEA develops measures of economic gross output. GDP, or
value added by business, is estimated by subtracting measured input costs from these
revenues.
For R&D and other intangibles, both prices and quantities are particularly
challenging to estimate. While there are also difficulties associated with quality
adjustments for cars, computers, or radios, these products at least have a quantifiable
output. A unit of R&D, on the other hand, does not have any commonly agreed upon
output measure. To further complicate the measurement issue, R&D investment, like
many other types of intangibles, is created primarily by firms and institutions for internal
use instead of sold on the open market. Thus, for much of the R&D conducted in the
U.S., there is neither an observable market price nor a quantity of output.8

Census data for the R&D services industries provides estimates of market R&D, but this is a relatively
small share of total domestic R&D activity.

The approach BEA has taken in this report is to apply the estimating methods
used in BEA’s accounts for nonmarket output. This method, used in the accounts both for
investment goods created internally for a firm’s own use and for the output of
government and nonprofit entities, is to measure output by the sum of costs expended to
create it. The approach described in this paper as an input-cost method is readily applied
to the detailed R&D expenditure data collected by the NSF for the United States.
How then would BEA adjust this output measure for changing prices? One of the
methods conventionally used for nonmarket output is to apply input-cost price indexes to
these aggregated costs, thereby producing a measure of real, or price-adjusted output. In
working through the methodology for this R&D satellite account, detailed input-cost
price indexes have been developed to deflate the output of R&D investment.
A well-known limitation of this input-cost approach to measuring nonmarket
output is that it is unable to capture productivity gains. In order to develop a truer
estimate of this hard-to-measure real R&D output in this paper, BEA develops three
alternative output measures for R&D and compares the results with the more
conventional input-cost based approach (see section IV for a detailed discussion).
Thus this paper provides estimates of R&D output for each of four scenarios:
•

The input-cost based approach (scenario A),

•

An output measure that takes the input-cost based approach and adjusts it to
reflect a rapid rate of productivity improvement (scenario B).

•

An output measure based on a group of high-productivity service industries
(scenario C),

•

An output measure based on four industries that perform the majority of
domestic R&D (scenario D).

While the first of these methods is, as described earlier, consistent with national
accounts methods for nonmarket output, the inability of this approach to measure
productivity gains renders it of limited value for capturing the real impact of R&D
investment in the economy. Estimates presented on that basis would underestimate R&D
impact on the economy.
The other three measures produce a set of estimates with quite similar impacts on
nominal GDP and contributions to the growth in real GDP. The preliminary estimates
described in the introduction to this paper and presented in the appendix tables (3 and 4)
are based on estimates that value real (price-adjusted) R&D at output prices of products
produced by R&D intensive industries (scenario D).

Steps in Treating R&D as Investment
The estimates presented in this paper are prepared by 1) aggregating NSF
expenditures on R&D by funder, 2) adjusting these expenditures to prevent doublecounting with other forms of investment and to include the cost of additional inputs such
as the use of structures, equipment, and software, 3) deflating the costs to develop real
(price-adjusted) R&D investment, 4) cumulating this investment into a capital stock with
an annual depreciation rate, and 5) estimating the resulting changes on GDP, investment,
and contributions to the growth in real GDP.

Limitations of the Preliminary Estimates
To calculate the preliminary estimates for R&D, BEA had to address several
substantial data gaps and unresolved conceptual issues in the treatment of R&D as
investment. This required several assumptions about R&D that BEA plans to refine over
the next several years. A partial list of these unresolved issues includes: 1) the absence of
output measures for R&D, 2) incomplete information on the ownership of R&D, 3)
limited information about market prices for R&D, 4) limited information about the rate at
which R&D depreciates, 5) incomplete information about the full scope of R&D in the
U.S, and 6) limited information about the rate of return earned on R&D investment.
These unresolved issues are addressed in two different ways—either with
simplifying assumptions or with a range of alternatives that are designed to bracket the
range of likely results. Developing improved methods for these issues forms the basis of
a research agenda for BEA’s future work on R&D and other intangibles.

Effect of R&D on Gross Domestic Product (GDP)
Recognizing R&D as investment changes the measure of the gross domestic
product (GDP). The specific impact of this treatment depends on the sector of the
economy. The largest impact on GDP occurs in the business sector. Other sectors are also
affected, including nonprofit institutions serving households, general government, and
government enterprises. These changes are shown in Table 2; Appendix Table 7 provides
a detailed view of these impacts on the NIPAs for the year 2000.
Business sector. Reclassifying business R&D expenditures from intermediate
consumption to investment leads to an increase in GDP equal to the value of the R&D

expenditures. The recognition of R&D as investment also affects the business incomes
and private consumption of fixed capital (CFC) components of gross domestic income
(GDI). Because R&D is no longer considered an expense, gross business incomes
(proprietors’ income and corporate profits) increase by the elimination of the deductions
for R&D expenditures.

Table 2. How the National Accounts Change when R&D is treated as Investment
R&D Imputations
R&D Funded by:
Business

Treatment in
Current
Measure GDP
Intermediate
consumption

Gross Domestic Product
Adjusted GDP

Reclassify to investment

Change in
Current Measure
GDP
Increase

Government
enterprises9

Deduction
from current
surplus of
government
enterprises

Reclassify to investment

Increase

Nonprofit
institutions serving
households
General
government

Consumption
(PCE)

1) Reclassify to investment
2) R&D CFC added

Increase

Government
consumption

1) Reclassify to investment
2) R&D CFC added

Increase

Gross Domestic Income
Adjusted GDI
Change in Current
Measure GDI
1) Increase in
business income
equal to R&D
investment less
CFC
2) Increase in CFC
1) Increase in
surplus of
government
enterprises equal to
R&D investment
less CFC
2) Increase in CFC
Increase in returns
to R&D capital

Increase

Increase in returns
to R&D capital

Increase

Nonprofit institutions serving households and general government. In these
two sectors, R&D expenditures are reclassified from consumption expenditures to
investment; since consumption expenditures are already part of GDP, this shift alone
does not change its measure. However, recognizing these expenditures as investment also
increases the measure of consumption by nonprofit institutions and general government
in an amount equal to the value of the CFC of the R&D, and thus GDP and GDI increase

While this row describes the way that R&D in government enterprises would be treated, the current data
and methodology does not identify any R&D in these enterprises, therefore the relevant line in Table 8
shows no impact.

correspondingly. This treatment is consistent with current treatment of government and
nonprofit investment in the NIPAs; government and nonprofit investment receives only
consumption of fixed capital as a return. The featured estimates for this account also
include a net return to government and nonprofit R&D capital in addition to consumption
of fixed capital. Therefore GDP will rise by an amount equal to the value of CFC and the
net return for government and nonprofit R&D investment.
Government enterprises. R&D expenditures by government enterprises are
currently treated as costs of production and thus are deducted in the calculation of the
current surplus of government enterprises. In recognizing R&D as investment, these
expenditures are added to gross government investment and, as a result, to GDP. The
effect on the current surplus of government enterprises is similar to that on business
income described above; that is, the surplus will increase by the elimination of the
deduction for the expenditures of R&D but will be offset partially by the deduction of the
CFC on R&D.

The BEA-NSF Collaboration
BEA is producing this satellite account in collaboration with the NSF, the agency
of the Federal government that has responsibility for producing R&D-related statistics for
the United States. NSF provided funding for the R&D satellite account project, allowing
BEA to assemble a team of researchers and analysts to work on the project. In addition to
providing the expenditure data that BEA is using to develop estimates of R&D output,

NSF staff has engaged in an analytical review of the data and methodology and has
provided guidance on issues such as the scope of R&D.10
The remainder of this paper is organized as follows:
•

Section II describes the conceptual issues involved in treating R&D as
investment. A summary of the choices made in recent experimental satellite
accounts by other countries concludes the section.

•

Section III presents an overview of the methodology used to develop currentcost and constant-cost estimates of R&D investment.

•

Section IV describes the assumptions used for the satellite account. Results
are discussed for the range of estimates based on the alternative scenarios
previously introduced.

•

Section V discusses the additional issues that will be reviewed in preparation
for the final version of the R&D satellite account to be released in the fall of
2007.

•

Section VI concludes this paper.

•

An appendix provides the data tables. The detailed methodology used to
estimate the accounts and the data improvements needed for full
implementation of capitalized R&D in the NIPAs will be released separately.
Tables 1.1-1.4 presents the macroeconomic effects of the R&D capitalization
on real GDP, nominal GDP, GDI and the national savings rate, by scenario.
Table 1.5 presents the returns to R&D assets for each scenario. Table 2

NSF staff has also undertaken several investigative and research tasks to support the satellite account
project. This support includes conducting informal surveys of R&D performers to help classify them
correctly into performing sectors, analyzing micro-data to answer questions about industry coverage and
types of R&D included in the survey data, and developing a conceptual framework for improving the
survey data on international R&D transactions (Moris (2006)).

summarizes the process required to transform the R&D survey data to the
R&D investment used in the satellite account. Tables 3.1 – 3.8 construct a net
stock of R&D assets. The investment, net stock and depreciation of the R&D
asset are presented in historical-cost, current-cost and real terms. The tables
are prepared on a funder basis using the assumptions defined in scenario D.
Tables 4.1-4.2 show the R&D investments on a performer basis in historicalcost and real dollars.
•

Table 5 show the price indices used each of the scenarios outlined in the
satellite account. Table 5.1 constructs the cost-based price index used in
scenario A. The indices used on the funder and performer basis are equal at
the total R&D investment level. Table 5.2 shows the price indices for each
scenario. Table 6 is a summary of the data and methods used for the deflation
of R&D investment. Table 7 provides a detailed view of the impacts of the
R&D treatment of R&D as an investment on the NIPAs for the year 2000.
Table 8 presents information on the trade in R&D services, international
royalties and licensing fees and the international funding of R&D.

II.

Conceptual issues in the treatment of R&D as investment

In order to prepare this satellite account, several key analytical and conceptual
issues had to be addressed. These issues include the following:
•

Defining the scope of R&D investment—that is, the boundary used to
distinguish activities within the scope of the R&D satellite account;

•

Assigning ownership of R&D investment to economic sectors;

•

Developing a measure of investment based on the available survey data on
R&D expenditures;

•

Choosing appropriate price indexes to create the real measures of R&D;

•

Identifying patterns and rates of depreciation to use when building R&D
capital stocks;

•

Defining gestation lag periods; and

•

Identifying rates of return to be used for deriving income flows from R&D
for scenarios B, C, and D that reflect a higher return to R&D compared to
other assets.

This section concludes with a look at the experiences of other countries that have worked
through these issues as they created R&D accounts.

Scope of R&D Activity
A fundamental issue BEA had to address was what activities to include as R&D
investment and what activities to exclude. Should only scientific R&D be included, or
should research in social sciences and the humanities be treated as investment as well?

What about market research and other activities that help firms determine what types of
new products to develop? How would BEA separate innovative expenditures from R&D
expenditures?
All the estimates presented in the satellite account use the same definition of
R&D; this is the definition used by NSF in its surveys. BEA evaluated two international
standards, the System of National Accounts (SNA)11 and the Frascati Manual,12 the
internationally accepted standard for R&D surveys. BEA selected the Frascati definition
of R&D as currently implemented by NSF in its R&D surveys as the measure of R&D for
the account because consistent data are available on this basis, and the resulting estimates
will allow for international comparability.
The Frascati Manual defines research and experimental development as
“…creative work undertaken on a systematic basis in order to increase the stock of
knowledge, including knowledge of man, culture, and society, and the use of this stock of
knowledge to devise new applications.”13 The inclusion of knowledge about man,
culture, and society in the definition makes it clear that, in addition to R&D in the natural
sciences, it also covers R&D in the social sciences and the humanities. The quality that
distinguishes Frascati-based R&D from related activity is “an appreciable element of
novelty and the resolution of scientific and/or technical uncertainty.”14
In contrast to the Frascati definition, the description of R&D according to the
SNA can be interpreted somewhat differently.15 The SNA emphasizes activities related

SNA (1993)
OECD (2002)
13
OECD (2002) par. 63.
14
Ibid paragraph 84.
15
In the SNA, the purpose of R&D is identified as follows: “Research and development by a market
producer is an activity undertaken for the purpose of discovering or developing new products, including
12

to innovation that bring products and processes to market and can be interpreted more
broadly than the Frascati definition. In other dimensions, the SNA definition can be
interpreted more narrowly because it does not specifically include activities that increase
knowledge without a resulting impact on economic activity. The latter type of activities
would be within the scope of R&D according to the Frascati Manual. Innovative activity
that does not involve novelty or technological uncertainty is not considered R&D in this
definition.
The current implementation of the R&D satellite account includes all
domestically performed, Frascati-defined R&D reported on NSF surveys. In practice the
survey data used to develop the estimates in the R&D satellite account cover a somewhat
narrower scope than the Frascati definition.. R&D in the social sciences and humanities
are excluded from NSF’s Survey of Industrial R&D, the primary source of time-series
related data for business sector R&D. Consequently, reported business sector R&D is
limited in practice to R&D in natural sciences and engineering.16
The estimates presented in this account are limited to those that affect GDP; a
related concept, gross national product (GNP), would include receipts of factor income
from the rest of the world, and subtract payments of factor income to the rest of the world
(for additional discussion on international transactions in R&D, see section V).
A useful recent implementation for the U.S. economy of a broader definition of
R&D is found in the work of Corrado, Hulten and Sichel (2006). Corrado, Hulten, and
Sichel (CHS) characterize investment in innovative property as composed of two parts,
improved versions or qualities of existing products, or discovering or developing new or more efficient
processes of production.” (SNA 1993 par. 6.142)
16
While for universities and colleges, R&D in the humanities is excluded from the expenditures used here,
NSF has been collecting non-scientific R&D (including humanities) since 2003, allowing these
expenditures to be included in future estimates.

scientific R&D and nonscientific R&D. The main difference between their definition of
R&D and the one implemented in this paper is the inclusion of a broad set of activities
characterized as non-scientific R&D. These include the cost of developing motion
pictures and other forms of entertainment, licensing payments and royalties, and an
estimate of spending for new product development by financial and insurance firms.

Ownership of R&D Assets
Recognizing R&D as investment implies that the satellite account must treat the
R&D stock as a fixed intangible asset. Flows from the stocks of fixed assets located in
the United States are presented in the NIPAs based on the sector that owns the assets.
Although recognition of R&D as an asset requires a presentation of domestic R&D stocks
on an ownership basis, the Frascati-based survey data used to estimate R&D investment
do not provide all the information necessary to identify ownership. In large part because
the most detailed survey data for R&D are available on a performer basis, as mentioned
earlier, the two prior versions of BEA-produced R&D satellite accounts, Carson, Moylan
and Grimm (1994) and Fraumeni and Okubo (2005), developed measures of R&D capital
stocks based on the performer of the R&D.
The limitations of the data still require BEA to make ownership assumptions; in
contrast with earlier satellite accounts, all of the capital stock and macroeconomic
estimates in this satellite account assign ownership of R&D assets to the funder. Without
complete information on transactions, BEA assumes that R&D funded by business is also
owned by business. Although, for nonprofit- and government-funded R&D, the
ownership is more ambiguous, BEA assumes that the R&D funded by them is owned by

them. The Federal government currently funds about twice as much R&D as it performs.
This amount funded by the Federal government is divided primarily between business
and academic institutions. Based on existing data, one cannot tell with certainty whether
the government bought this R&D, or whether ownership passes to the performer. Because
the government pays for this R&D and therefore can be expected to gain some economic
benefit from it, government funded R&D is currently considered an asset of the
government in this satellite account.
The impacts on GDP from recognizing R&D as investment presented in this
paper in Appendix Tables 1.1–1.4 are funder-based. Appendix Tables 3.1 and 3.2 provide
estimates of real and historical cost R&D investment and price indexes by funder.
Because many users will be interested in the performer-based data, BEA investment
measures on both a funder and performer basis are presented in Appendix Tables 4.1 and
4.2. These tables provide estimates of real and historical cost investment and price
indexes by performer.

Data for R&D Investment
As noted earlier, NSF data on R&D expenditures are used for the measurement of
R&D investment. BEA has decided to build on data published by the NSF because NSF
data provide the most comprehensive source for national level R&D expenditures for
business, government, nonprofit institutions, and academic institutions.17

BEA in fact surveys multinational companies on their R&D expenditures, collecting data on R&D
expenditures from U.S. parent companies as well as foreign-owned U.S. affiliates. However, the reported
expenditures by the U.S. units of these multinational companies account for approximately 80 percent of
industrial R&D performed in the U.S when compared to industrial expenditures reported by NSF. The
reason for this narrower coverage is that BEA data exclude the domestic R&D expenditures of firms which
are not classified as multinational companies.

The NSF’s Division of Science Resources Statistics coordinates the collection and
reporting of survey data on R&D expenditures and consolidates these data in the annual
publication, National Patterns of Research and Development Resources (NSF, various
years). These data are based primarily on two annual surveys published by the NSF, the
Survey of Industrial R&D (SIRD or RD-1) and the Survey of Research and Development
Expenditures at Universities and Colleges. An abbreviated version of the latter survey is
collected annually for Federally funded research and development centers. Two
additional annual surveys provide information on outlays and obligations by the Federal
government for R&D. These are the Survey of Federal Funds for R&D and the Survey of
Federal Science and Engineering Support to Universities, Colleges, and Nonprofit
Institutions. The Scientific and Engineering Research Facilities Survey provides
information on construction plans and capital spending and is conducted biennially. An
NSF sponsored Gallup survey, Research and Development Funding and Performance by
Nonprofit Organizations was last conducted in fiscal 1996 and 1997 and has been
discontinued. A survey of state research and development expenditures was last collected
for fiscal year 1995; an update of the survey will be fielded in the Fall of 2006.

Measurement of R&D Investment
Annual R&D investment is measured as the sum of the costs of R&D activities by
all domestic performers. This investment measure implicitly treats all reported R&D
expenditures equally, resulting in expenditures for unsuccessful R&D being considered
as investment along with expenditures for highly valuable innovations. These R&D
expenditures are then segmented by funder.

This method is used for two related reasons. First, there is currently no standard
measure of R&D output, and second, most R&D conducted in the business sector is not
sold on the market directly but is used instead within the firm that created it.18 Appendix
Table 2 provides a summary of the adjustments required to transform survey data on
R&D expenditures into an annual investment measure in current dollars. These steps
effectively link Frascati-based expenditures to the value of investment in current
dollars.19 The table provides a summary of the adjustments BEA has made to the data for
the 2006 R&D satellite account as well as an indication of what remains to be completed
for the 2007 R&D satellite account and beyond. An “x” indicates an adjustment has been
made when necessary, an “NAN” indicates that no adjustment is necessary, and a blank
indicates that the adjustment should be made in a future set of estimates as improved
methodologies and data are developed. A review of Appendix Table 2 also provides a
partial indication of BEA’s agenda for future improvements to the methodology as well
as the additional data requirements.
The first section of the table describes adjustments to the performer-based survey
data to match the scope of capitalized R&D. Since the Frascati-based definition is used
here, it includes R&D in the social sciences and humanities but excludes expenditures
solely for commercialization and marketing. Because the sum of current-period R&D
production costs are used to develop current-period investment measures, expenditures
for materials and supplies not used in the current period should be subtracted from the
output measure. Also, because R&D is being treated as an intangible asset, R&D that is

This treatment is similar to BEA’s treatment of expenditures for mineral exploration; the successful
discoveries pay for the unsuccessful ones. Moreover, knowledge gained by unsuccessful attempts is
additional output.
19
This procedure is laid out in detail in Robbins (2006).

purchased for use in the production of further R&D is the purchase of an asset and not a
current cost. However, payments for the use of R&D in the form of technology or patent
licensing fees are a cost of production.
The next section of the table makes adjustments to the data to make it consistent
with accounting conventions of the NIPAs. Following this, the next adjustments account
for expenditures in the survey data that reflect capital expenditures rather than current
costs. These include expenditures for structures, equipment, and software. Such
expenditures are subtracted and accounted for separately in the stock of fixed capital.
Accounting for the full cost of production includes the costs associated with the
use of that stock of fixed capital used to create R&D. This cost includes the economic
depreciation on the structures, equipment, and software used in production. This cost is
measured by the consumption of fixed capital (CFC) estimate. The full cost of production
includes an adjustment for taxes and subsidies on production. Finally, investment must be
adjusted to include imports of R&D and exclude exports.

Real Measures of R&D
Real (inflation-adjusted) R&D is difficult to measure largely because it is not
bought and sold in markets. Normally, the companies that conduct the R&D are also the
companies that use the R&D to produce new and better goods and services.
Conceptually, the value of R&D is equal to the discounted present value of the future
benefits that the company derives from R&D. However, the value of the R&D is
embedded in the value of all the goods and services the company sells, and there is no
direct measure of either the contribution of R&D to those sales or the market price
underlying R&D assets. Companies can normally report what they spent on wages

salaries, contactors, and other costs of conducting R&D, but not the market price of
R&D. For computers, communication equipment and other assets that are bought and
sold in final goods markets, companies know the market price of the asset, its share of
sales, as well as the share of profits that came from the difference between the sales price
and cost of producing that asset. For these assets, it is relatively simple to estimate their
real value and contribution to GDP by simply dividing the nominal value of these assets
by a price index based on their sales. For R&D, the value of the assets and their
contribution to sales are indistinguishably bundled with the companies overall assets.
Thus, the only available nominal value is the cost of their production.
The conventional approach for estimating real goods and services that are not sold
in markets is to deflate the cost of their production by a cost index based on the prices of
the components of the costs. In the U.S. national accounts, this approach is used for
computing the real value of the output of governments and nonprofits institutions serving
households (NPISHs). The cost-based output and deflation approach is also used for
estimating the real value of assets that companies build for their own use (these
expenditures are called “own-account construction” in the language of national
accountants).
While this approach produces a “real value” for the output of governments and
NPISHs output and for the value of own account construction, it necessarily implies that
real inputs grow at the same rate as real output and thus produces zero productivity
growth (e.g. real output cannot grow faster than real inputs). Unfortunately, this “default”
convention seems particularly inappropriate for measuring a dynamic sector like R&D,
especially within the context of satellite accounts.

The dynamism of the R&D sector is clear. Firm level and indirect estimates of the
rates of return to R&D are much higher than the rates of return to other investments.
While rates of return are uncertain and variable, the returns from available studies suggest
very high rates. The average private return from studies shown in Table 3 is 26 percent.
Social returns—including the returns that “spillover” to industries beyond the direct
returns to the companies conducting the R&D—are much higher, averaging 66 percent.
Although comparing rates of returns is difficult, the high average returns to R&D shown
in Table 3 probably reflect not only the reward-to-risk ratio but also the nature of
technological innovation and the high-productivity of R&D investments.
For example, products that typically embody a high level of R&D, such as
computers and communication equipment, have a short—and increasingly shorter—
lifecycle as new R&D and new technologies are introduced. This technological
obsolescence not only means that the time period over which the costs on R&D must be
recovered is short, but the value of that R&D is being reduced each year as prices fall (the
price of older models that embody previous R&D must be reduced to compete with
newer models embodying more advanced R&D). This means that R&D must
significantly raise the productivity of new products by lowering costs and/or increasing
sales by enough to earn the very high rate of returns found in R&D. Companies must also
try and maintain that productivity advantage over their competitors as long as possible by
carefully protecting their intellectual property through trade secrets, patents, licensing and
other methods.

Table 3. Summary of Estimated Rates of Return to Private R&D
Estimated Industry Rates of Return to Private R&D (%)
Source

Private

Minasian (1969)

Griliches (1980)

Mansfield (1980)

Nadiri and Bitros (1980)

Schankerman (1981)

24-73

Griliches and Mairesse (1983)

Link (1983)

Clark and Griliches (1984)

18-20

Griliches and Mairesse (1984)

Griliches (1986)

33-39

Griliches and Mairesse (1986)

25-41

Jaffe (1986)

Schankerman and Nadiri (1986)

10-15

Griliches and Mairesse (1990)

27-41

Lichtenberg and Siegel (1991)
Source: Nadiri, (1993).

Estimated Aggregate Rates of Return to Private R&D (%)
Source

Private

Social

Sveikauskas (1981)

7-25

Bernstein and Nadiri (1988)

10-27

11-111

Bernstein and Nadiri (1991)

15-28

20-110

Nadiri (1993)

20-30

(1)

Mansfield et al. (1977)

56(1)

Goto and Suzuki (1989)

Terleckyj (1974)

48-78

Scherer (1982,1984)
29-43
64-147
Source: Table 8.1, 12. Fraumeni, Barbara M., and Okubo, Sumiye. “R&D in the National
Accounts: A First Look at Its Effect on GDP.” National Bureau of Economic Research,
Studies in Income and Wealth, Volume 65, Measuring Capital in the New Economy,
Edited by Carol Corrado, John Haltiwanger, and Daniel Sichel, 2005.
(1) These rates are median rates.
Average Rates of Return to Private R&D from All Studies (%)
Source
Average of Above

Private

Social

In constructing these initial R&D satellite account real estimates, the goal is to
implement the simplest case, one consistent with existing national accounting
conventions that would be used if R&D were to be capitalized and treated like
government output; this is scenario A.
Scenario A. In this scenario, BEA uses input costs as a proxy for R&D output
and then deflates this output measure with the price index created with information on the
cost components for R&D. While this procedure is consistent with national accounts
methodologies for other nonmarket output, by measuring output with inputs and then
deflating this output by the costs of inputs, the procedure is unable to capture an increase
in output per unit of input. Therefore, if there is productivity improvement in the conduct
of R&D over time, this method will under-estimate real R&D output.
Beyond this scenario, three estimates based on proxies for the market price of
the R&D sector are introduced in order to obtain an estimate of the order of magnitude of
the effect of capitalizing R&D on real GDP. The challenge is to find proxy prices for the
market value of R&D consistent with the very high returns earned by R&D and the
characteristics of R&D without a direct link between the returns to R&D and the value of
R&D.
Scenario B. This scenario, a variant of the cost-based deflation approach,
assumes that the real value of R&D output is higher than the real value of R&D inputs by
the amount of productivity growth recorded in higher-performing industries in the U.S.
economy. This adjustment is implemented by subtracting average multifactor
productivity (MFP) growth, estimated for a group of manufacturing industries with the
highest MFP growth, from the increase in the cost-based price index from scenario A.

The result is a cost-based index that incorporates a proxy for R&D productivity growth.
The resulting R&D price index implies MFP growth in R&D conduct that averages 3.8
percent a year, but with the well-known downturn in 1973 and acceleration since the mid1990s.
Scenario C. This scenario assumes that R&D, which is most similar to a service
industry, is valued at the output prices of the most productive service industries. While
service industries have traditionally had lower productivity growth and higher inflation
than the industries in the goods sector, key industries have a good record in producing
high-productivity, declining relative prices, and ever increasing real output per unit of
input. In this scenario, real R&D output is estimated using a weighted average of BEA’s
GDP by industry value-added price indexes of these high-productivity service industries.
These industries are air transportation, broadcasting and telecommunications, securities
and commodity brokers, and information and data processing services.20
Scenario D. This scenario, the most closely linked to available R&D data,
assumes that R&D is valued at the output prices of products produced by R&D-intensive
industries. The prices of these products that have a high R&D content may be the best
proxy for the value of the R&D embodied in these products. This R&D price index is
developed using BEA’s GDP by industry value-added price indexes for the four
industries that perform the most R&D. Based on NSF industry performer data, these
industries are: radio and tv receiving equipment; drugs and medicines; office, computing,
and accounting machines; and aircraft and missiles. The weights used to combine these
20

These indexes were used instead of BLS indexes because in most cases, the time span for industry
coverage by BLS is not long enough to enable using the BLS producers price indexes as deflators. For
example, the BLS PPI for broadcast and telecom equipment – an industry that appears in the top five
Productive Services index – is only available for the period 1991 forward. The R&D work requires having
an index that covers the period 1959 forward.

price indexes vary each year based on each industry’s share of total performer-based
R&D expenditures.

Depreciation Rates
The satellite account uses two alternative measures for depreciation of R&D.
Depreciation is “the decline in the value of the stock of assets due to wear and tear,
obsolescence, accidental damage and aging.”21 While R&D capital does not wear out in
the traditional sense, it does lose value due to obsolescence. It loses value over time as
new innovations appear and earlier R&D becomes relatively less effective in the
production process and therefore less valuable. An additional loss in value comes from
the gradual leakage of information to competitors and the expiration of intellectual
property protection that render the R&D asset less valuable to its owner.
The geometric depreciation patterns used by BEA for other assets are based, to
the extent possible, on empirical studies of markets for used assets. Because most R&D is
not sold on the market, this type of information is not directly available.

Using

econometric methods to develop depreciation rates for business R&D, economists have
found the range of average annual depreciation rates to be between twelve and twentyfive percent.22 A lower rate of depreciation is likely to be appropriate for R&D
conducted by government and universities and colleges. More basic research is done by

BEA (2003)
Pakes and Schankerman (1984) found the average annual decay rate of R&D to be 25 percent; Nadiri and
Prucha (1996) estimated the annual depreciation rate of industrial R&D capital stock to be 12 percent. In
1996, Lev and Sougiannis estimated decay rates of R&D in six industries, finding a range of 12 percent to
20 percent and an average depreciation rate of 15 percent. Most recently Bernstein and Mamuneaus (2004)
calculated a 25 percent depreciation rate for the manufacturing sector.
22

these performers, and this basic research is likely to obsolesce more slowly than the
private sector R&D represented by the econometric studies noted above.
Scenario A’s estimates of depreciation are based on a 15 percent geometric rate of
depreciation; this rate was chosen as compromise between the results of the depreciation
studies and the presumed slower depreciation of non-business R&D. A 15 percent rate is
somewhat higher than the 11 percent depreciation rate that was used by Fraumeni and
Okubo (2005) and Carson, Moylan and Grimm (1994).
As an alternative to the 15 percent depreciation rate, scenarios B, C, and D
incorporate an alternative method that proxies the effect of a more rapid pace of
technological change in recent years and a resulting increase in the rate of obsolescence.
This faster rate of obsolescence is consistent with the work of Caballero and Jaffe (1993),
whose work with patents found an accelerating rate of obsolescence in the 1990s
compared with earlier decades. Scenarios B, C, and D incorporate this faster depreciation
by using the depreciation pattern from the NIPAs information processing equipment and
software between 1987 and 2002. Prior to 1987 the depreciation pattern is estimated with
the rate of change in the depreciation pattern from the NIPA series for private fixed
investment in nonresidential equipment and software. The resulting depreciation series
accelerates in recent years based on the shorter service life of new types of equipment
and software; it starts at about 16 percent in 1959 and ends up at about 23 percent in
2002.

Gestation Lags
These lags represent the time between beginning and completing an R&D project,
while application lags are the period between its completion and commercial use. For
R&D, a gestation lag implies that an expenditure made in one year becomes an
investment in a subsequent year. BEA’s earlier efforts to estimate the impact of R&D on
the economy included a lag to account for the timing of the R&D expenditure and its
impact on economic output.
Despite unarguable real-life lags between research activities and commercialized
products or implemented processes, in the current set of estimates, a zero-year gestation
lag is used, implying that the R&D investment will be added to the R&D capital stock at
the time that the expenditure is made. Real-life lags are likely to vary by type of R&D,
and the one-year lag implemented in earlier estimates is at best, an approximation of a
heterogeneous process. The zero lag implemented here is a methodological simplification
that consistent with both the current practice with tangible investment in the NIPAs and
recent recommendations of the international group of experts responsible for developing
revised SNA standards.

Rates of Return to R&D Assets
Recognizing business R&D as investment results in an increase in GDP equal to
the value of the new investment. On the income side (GDI) this treatment results in
matching additional profits and depreciation. These additional profits and depreciation
are reflected in the NIPAs as gross operating surplus. This profit-like measure can be
used to estimate before and after tax rates of return as a ratio of this surplus, less CFC,

divided by the stock of produced assets. From 1960 to 2005, the before-tax rate of return
(based on net operating surplus) for domestic nonfinancial corporations has had a median
value of 9 percent.23
Scenarios B, C, and D make some preliminary proxy adjustments for the impact
of higher net rates of return for R&D compared with other types of capital. These
scenarios assume that R&D investment earns a higher rate of return than other types of
investment because of the uncertainty or risk surrounding its future benefit. This is done
by assuming that R&D capital earns an average net rate of return of 15 percent over the
period 1959 to 2002. This is approximately four percentage points higher than the
average rate of return to all private assets.
In contrast with business sector gross operating surplus, which implicitly includes
depreciation and a net return, the current NIPA and SNA treatment of government and
nonprofit capital is to recognize only CFC; implicitly assuming a zero net return to this
investment. While scenario A of the R&D satellite account adopts this assumption,
scenarios B, C, and D explore the impact on GDP of including a net return to government
and nonprofit R&D capital. These scenarios include a return to government and nonprofit
R&D that is estimated with a long-run average rate from ten-year government securities,
boosted up to match the new ratio of the return from business R&D assets to other
business assets.24

Lally (2006) pages 7
BEA’s efforts to approximate ex ante real net rates of return—as the nominal rate on constant-maturity
10-year U.S. Treasury bonds, less a simple moving average of the rate of change of the R&D investment
deflator—produced very bumpy series, owing to a spike in interest rates in the early 1980s unmatched by
R&D inflation. The 1959–2002 average rate by this method is noticeably lower than the 15 percent average
assigned to private-business R&D assets; the ratio of these two averages, times the year-to-year privatebusiness R&D net rate, gives the smoother net rate that BEA actually used in scenarios B, C, and D.

In their estimates of the impact of R&D on the economy, Fraumeni and Okubo
(2005) included a set of estimates of the account with return to government and nonprofit
R&D capital in excess of CFC (net returns). For consistency, they extended this treatment
to all government and nonprofit fixed capital.
In this version of the satellite account, BEA chose to limit the addition of the
return to government and nonprofit investment capital to only include a return to R&D
capital (scenarios B, C, and D). The reason for this is that the additional return to nonR&D capital has a significant impact on the macroeconomic results, and would tend to
blur the interpretation of the treatment of R&D as investment with the net returns to
government and nonprofit investment capital.
To illustrate the impact, including a net return to all government and nonprofit
capital would increase the level of nominal GDP by an average of 5.4 percent from 1959
to 2002 as compared with the average annual 2.6 percent increase in scenarios B, C, and
D. Excluding the net return to government and nonprofit capital out of the estimates all
together would have resulted in an average annual increase in the level of nominal GDP
of 2.3 percent.
Including the net rate of return to government and nonprofit capital presented an
additional challenge: What price index should be used to deflate these additional returns?
These returns were deflated with the price index created for scenario B, the highproductivity service-sector industries. It is used here instead of a broader deflator in a
first attempt at capturing the public benefits of government and nonprofit R&D
investment.

Experiences of Other Countries with R&D Satellite Accounts
The national statistical agencies of several other countries have linked their R&D
survey data to their national accounting system through bridge tables and have begun to
develop satellite accounts to estimate the impact of R&D on GDP. These efforts required
an exploration of conceptual issues described in earlier parts of this section, including
specifying the scope of R&D, lags and depreciation rates for R&D assets, and the choice
of input deflators for R&D investment. This section reviews some of the results of these
efforts. Countries with relevant experimental work with R&D satellite accounts include
Australia, Canada, France, Israel, the Netherlands, Sweden, and the United Kingdom.
Table 4 compares some of the assumptions used.
In specifying the scope of R&D that should be considered as an asset, Israel,
Australia, the Netherlands, and Canada consider somewhat different frameworks from
each other. National accountants from Statistics Netherlands and Statistics Canada argue
that the SNA definition of an economic asset should be understood to exclude some
Frascati-defined R&D. De Haan and van Rooijen-Horsten of Statistics Netherlands
propose that nonmarket R&D should be excluded unless it is patented or explicitly used
in government production, as in the case of defense R&D.25 Siddiqi and Salem of
Statistics Canada include all business sector R&D but exclude R&D outside of the
business sector that is not either patented or patentable.26 Peleg of the Central Bureau of
Statistics of Israel argues for the inclusion of nonmarket R&D, noting that its use is
similar to that of roads and other public infrastructure.27 The Australian Bureau of

de Haan, et al (2004). page 19
Siddiqi and Salem (2005).
27
Peleg (2003).
26

Table 4. Comparison of Key Assumptions of R&D Satellite Accounts

Country

Average
service life
in years

Australia
(Note 1)

Canada
(Note 2)

5 to 10

Israel
(Note 3, 5)

Netherlands
(Note 4)
UK
(Note 6)

R&D input price
index
componentbased input price
index

Depreciation rate
of R&D capital

10 percent

Source of net
operating
surplus
estimate
based on a
normal return
to capital

GDP deflator
componentbased input price
index
componentbased input price
index

10 and 25 percent

R&D Services
Industry

15 percent

11 to 25 percent

R&D Services
Industry
Other business
services
industry

Gestation
lag

Impact on
Level of
Current GDP

1.5 percent

1.2 percent

3 percent

1.1–1.2
percent

GDP deflator
10 to 25 percent
NA
NA
NA
input-cost based
deflator and
15 percent,
three highinformationproductivity
processing
proxy price
equipment and
2.3 to 2.6
indexes
software
percent
U.S.
13.3
none
0
Note 1. Australia. Australian Bureau of Statistics. National Accounts Research Section. Capitalising
Research and Development in the National Accounts. March 2004.
Note 2. Siddiqi, Yusuf and M. Salem: “Treating Research and Development as Capital Expenditure in the
Canadian SNA.” System of National Accounts Statistics Canada. March 2005.
Note 3. Brenner, Nava; Peleg, Soli; and Galit Zalewsky: Updated version of the exercise to examine the
impact of capitalization of R&D in the national accounts.” Prepared for the Canberra II Group: On the
Measurement of Non-Financial Assets. August 2005.
Note 4. de Haan, Mark and Myriam van Rooijen –Horsten Measuring R&D Output and Knowledge
Capital Formation in Open Economies. Conference paper, 28th General Conference of the International
Association for Research in Income and Wealth, Cork, Ireland, August 22-24, 2004.
Note 5 Peleg, Soli: “Harmonization between R&D Statistics and the National Accounts.” Central Bureau
of Statistics, Israel. Paper presented at the NESTI/Canberra II meeting in Berlin, Germany, May 2006.
Note 6 Clayton, Tony and Prabhat Vaze. Capitalising Research and Development in the UK National
Accounts, Undated Manuscript.

Statistics’ satellite account handles this issue by providing estimates both with and
without pure basic research.
Because of the absence of output measures, output is obtained via deflation of
input costs in most of these satellite accounts. Australia, Israel, and the Netherlands

deflate cost components separately, using wage and salary indexes for R&D activity
when available. By contrast, Canada and the UK use the GDP deflator, which has been
found in many comparisons to perform similarly to component based input deflators for
R&D. All the satellite account reports describe the need to improve their deflator
measures in future work.
Based on these preliminary exercises, the impacts on the level of currentdollar GDP of these countries range between 1 and 3 percent. Of these international
examples, the largest impact of 3 percent is found in Israel, where market R&D is a major
industry and the satellite account includes all nonmarket R&D. Australia reports a 1.4
percent increase in GDP, excluding basic R&D, and a 1.6 percent increase, including
basic R&D. The Netherlands total GDP is adjusted upwards by 1.1 percent to 1.2 percent.
For Canada, total GDP increases by 1.2 percent.
Each satellite account estimates the value of R&D investment based on input
costs, and R&D survey data are adjusted to include an estimate of the cost of capital used
to produce the R&D. Similar to BEA’s construction of its satellite account, these
countries include an estimate of the value of CFC for the fixed capital used in the creation
of R&D.
Since the fixed capital used to create R&D could have been used for another
economic purpose and would have earned an average return that exceeds the CFC alone,
some satellite accounts make an additional cost adjustment to the sum of costs to
recognize this return. This adjustment is made with an estimate of the net operating
surplus (a profit-like measure net of CFC). Israel and Canada use net operating surplus
from the R&D services industry; the Netherlands uses the surplus from Other Business

Services, and the Australians use the normal return on their aggregated capital stock.
BEA does not use this practice in its national accounts methodology for own-account
output, and therefore, this adjustment is not made in the satellite account.28

Since the industry survey data used for the U.S. estimates include an overhead measure, this overhead
may also include profits for performers who sell R&D.

III. Estimation Methodology
This section provides a more detailed discussion of the transformation of NSF
expenditure data into measures of real investment and capital stock. Readers who are
interested primarily in the results of the alternative scenarios may wish to skip to Section
IV.
In the new treatment of R&D as investment, BEA measures the value of R&D
output by summing the costs of R&D activity. The methodology is similar to BEA’s
methods for estimating other types of own-account investment, for example software and
construction. The aggregated investment measures for R&D are presented in the
Appendix Tables, where they are shown by major performer and by major source of
funding. The stocks of R&D, the accumulation of investment after adjusting for their loss
in value over time, are presented by major source of funding.
The data are disaggregated into two major institutional categories: Private and
government (“public”) organizations. Several subcategories are also included. Private
organizations consist of businesses; private universities and colleges; private hospitals,
charitable foundations, and other nonprofit institutions serving households; and most
Federally Funded Research and Development Centers (FFRDCs). 29 Government
organizations consist of the Federal Government, state and local governments (excluding

FFRDCs, which are R&D organizations financed almost entirely by the Federal Government, are shown
separately and grouped with the entities that administer them in the performer-based presentation of
investment (Appendix Table 4.1). Grouping FFRDCs in the performing sector that administers them is
consistent with the NIPAs. However, NSF reports that all FFRDC activities are more similar to Federal
Government laboratories and classifies them as such. Since these institutions are by definition Federally
funded, they are included with the government-funded investments and stocks of R&D.

universities and colleges), public universities and colleges, and FFRDCs administered by
state and local governments, primarily public universities and colleges.
All estimates of current-dollar R&D investment are prepared by first compiling
data available from the various NSF surveys and then by adjusting these data to be
statistically and conceptually consistent with BEA definitions in the NIPAs.
Performer-based estimates of real R&D expenditures are derived by deflating the
most detailed current-dollar expenditures by appropriate price indexes. BEA develops
real R&D capital stocks by treating the R&D expenditures as investment and cumulating
them based on methodologies which BEA uses for other types of fixed assets.

Changes from Previous BEA R&D Satellite Accounts
The methodologies used for this satellite account are largely similar to those used
in the first version of the R&D satellite account presented in the November 1994 Survey
of Current Business (Carson, Moylan, and Grimm (1994)). However, several important
changes have been made to the account’s current-dollar estimates, real estimates and the
capital stock measures.
Important changes in current-dollar estimates include the following:
•

Industry-administered FFRDCs are now shown separately.

•

For the state and local government and nonprofit institutions sectors, indicator

series have been improved for years when there are no NSF survey data.

Important changes to real estimates include the following:

•

Improved chain-type price measures of real output and prices, eliminating the
overstatement of real R&D growth for periods after the base year and the
understatement of real R&D growth for periods before the base year.

•

Within the input-cost price index (Scenario A), a new methodology for
deflating business R&D that uses price measures based on unpublished BEA
Industry Accounts data from the Scientific Research and Development
Services industry (NAICS 5417) instead of price measures for each industry;

•

Within the input-cost price index (Scenario A), a new methodology for
deflating academic R&D that uses an academic R&D price index developed
by the National Center for Education Statistics.30

•

Real R&D investment by source of funding.

Important changes to the capital stock measures include the following:
•

A new measure of R&D capital stocks based on the funder of the R&D that is

performed; to better approximate the ownership assumed for R&D capital.
•

A geometric rate of depreciation that replaces the depreciation pattern based a
straight-line perpetual inventory method.

The key methodological issues faced in developing the R&D satellite account are
outlined in the following sections.

This series ends in 1995, BEA extrapolates with NIPA PCE for education and research between 1996 and
2002.

Historical Cost (current-dollar) R&D investment
As noted earlier, although BEA’s estimates of R&D stocks are presented on a
funder basis, the investment estimates are built up from performer-based survey data.
Current-dollar investment is based on performer-reported R&D expenditures from
several NSF surveys. These expenditures include the cost of R&D personnel, materials
and supplies, overhead (e.g., utilities, insurance, and taxes) and the depreciation charges
associated with the capital used in performing R&D. Expenditures for each performer are
then disaggregated by any available detail on the sources of funding.
Constructing funder-based R&D stocks performer-based expenditures, rather than
stocks from the source of R&D funding, attempts to avoid at least two problems. First,
the data reported by source of funding would have to be adjusted in order to be consistent
with the timing with which purchases of goods and services are generally recorded in the
NIPAs.31 Second, the data reported by source of funds would have to be adjusted to
avoid double-counting for funds passed between sectors.
Where there are gaps in the coverage of the NSF survey data, BEA uses
interpolation and extrapolation methods to fill in the missing data using time series of
related data. R&D activities for state and local governments and for nonprofit
organizations excluding universities and colleges are two categories where this method is
used extensively. Some R&D expenditure estimates prior to 1987 use data previously
estimated in Carson, Moylan, and Grimm (1994).
BEA performs a number of adjustments to the NSF survey-based expenditure data
to make them statistically and conceptually consistent with the NIPAs. The statistical

This is an adjustment from the time of payment from the source of funding to the time of payment to the
performer.

adjustments are for timing (to reflect calendar year expenditures), geographic coverage
(to reflect only domestic R&D performed), and to estimate missing data, which were
withheld to protect the confidentiality of NSF survey respondents. BEA also makes an
adjustment to put the depreciation of structures and equipment used in the production of
R&D on a basis appropriate for national economic accounting rather than tax accounting.

Real R&D investment
An important methodological change is the introduction of “chain-type” measures
of real output and prices. This major methodological improvement was introduced into
the NIPAs as part of the 1996 comprehensive revision of the accounts. The real R&D
measures are “chain-type” measures that are based on the “Fisher ideal” formula that
incorporates weights from the two adjacent years rather than the weights of a single base
year.32 Thus, when the base year is updated, the levels of the estimates change but the
growth rates of the various components do not, and the revisions to the growth rates that
result from updating the base period of a fix-weighted index are avoided.
The rest of this section has two parts. First, it describes the construction of the
input-cost-based indexes used in scenario A. Next, it describes the construction of the
three high-productivity proxy indexes used in scenarios B, C, and D.

The Fisher quantity index relative is geometric mean of Paasche and Laspeyres quantity relatives. The

quantity index relative is calculated as Fisher =

Paasche =

L * P ,where Laspeyres =

t −1

∑ PQ
∑ PQ
t

∑P Q
∑P Q

t −1

and

Input-cost Price indexes

Ideally, BEA would develop measures of real R&D investment based on market
prices for R&D. Without these market prices, BEA has developed a methodology similar
to that used for other nonmarket output. This methodology uses input-cost price indexes
to deflate output. BEA would like to deflate all R&D expenditures disaggregated by
major cost components, such as compensation, materials and supplies, CFC, and other
overhead costs. In practice, only industry expenditures are deflated based on detailed cost
components because the survey of industrial R&D performers is currently the only survey
that provides the information needed to develop this detailed split of R&D costs.
Consistent with NIPA practice, the estimates do not include any additional adjustment,
such as an assumed rate of increase in productivity. Appendix Table 6 summarizes the
main deflation measures used for each performer in the input cost indexes used in the
2006 preliminary R&D satellite account.
The procedure above results in the performer-based price indexes presented in
Appendix Table 6.1. For funder-based estimates, since not all the detailed cost data are
available in scenario A, the input cost-based price indexes require additional steps.
Currently, BEA can identify only two of the necessary price indexes. One is a NIPA price
index for total Federal defense and nondefense purchases of R&D. The other is for
Federal intramural expenditures that are already derived as part of the performer-based
calculations.33 that are already derived as part of the performer-based calculations.
For funder-based estimates of investment, BEA uses the Federal price indexes
described above and the total (performer-based) price index to derive a non-Federally
33

Intramural expenditures are expenditures for R&D performed within a sector of the economy, in this case
within the Federal government. Extramural expenditures are expenditures for R&D performed by another
sector.

funded price index. Thus, when the Federal price indexes are combined with this derived
non-Federal index using a chain-type formula, the total funder-based price index will
match the total performer-based index.
BEA has introduced some important improvements to the real measures of R&D
output that were not available in the prior versions of BEA-produced R&D satellite
accounts. For the business sector, salaries for engineers in R&D organizations are used to
deflate compensation costs for R&D personnel (American Association of Engineering
Societies Annual Salary Surveys, various years). Input price indexes from costs incurred
by the Scientific Research and Development Services industry (NAICS 5417) are used to
deflate materials and supplies, overhead, and R&D CFC. These prices are based on
detailed data for intermediate input costs now available in BEA’s industry accounts.34
For R&D performed by colleges and universities, BEA incorporates a new
methodology for deflating academic R&D expenditures based on an academic R&D price
developed by the National Center for Education Statistics for the years 1960–95.35 This
index is extrapolated for the other years based on the BEA price index for personal
consumption expenditures on other education and research.36

BEA’s earlier work assumed that intermediate inputs to business R&D had a pattern of price change
more like that of the overall intermediate inputs used by the industry performing the R&D. The current
choice assumes that intermediate inputs to business R&D have a pattern of price change more like that of
the inputs to the R&D services industry. BEA and NSF plan to continue to work together to improve these
measures.
35
Deflation of academic data in the 1994 satellite account relied heavily on the use of BEA’s biomedical
research and development price index (BIRDPI), which BEA and NSF believed was not a desirable price
measure to use for the entire academic sector.
36
National Center for Education Statistics, Digest of Education Statistics, 2004, Table 35.

High-Productivity Price Indexes

While scenario A uses three price indexes to deflate R&D investment, the other
three alternative scenarios apply one price index to all investment. The price indexes used
to deflate nominal investment in Scenarios B, C, and D are constructed in various ways.
Direct High MFP (scenario B). The high MFP index is created by subtracting a

gross-output weighted average of the top 5 MFP growth rates of 2-digit SIC
manufacturing industries, as computed by the BLS , from the growth rate of the Fisher
aggregate of the three input-cost based indexes of scenario A. To create this combined
index, BLS multifactor productivity data for 2-digit manufacturing industries are
compared for each year. In each year, annual productivity growth from the five
industries that experienced that year’s highest MFP growth are weighted by the share of
gross output of each industry to the sum of the five industries’ gross output. Two
industries, “electronic and other electrical equipment, except computer equipment,” and
“industrial and commercial machinery and computer equipment” consistently appear in
the top measures since 1974. The resulting productivity series is used to adjust the inputcost based price index used in scenario A. As noted earlier, the resulting R&D price
index implies MFP growth in R&D conduct averaging 3.8 percent a year, and reflects the
well-known downturn in 1973 and acceleration since the mid-1990s.

High-Productivity Service Industries (scenario C). Service industries with high

average annual real value added growth over the period 1959–2002, (high-productivity
service industries) are used to create a price index that replicates the output growth of
R&D prices. The industries used for this index represent 5 of the 6 fastest growing

services-producing industries during the period.37 The value added price indexes from
BEA’s GDP-by-industry program are combined using Fisher aggregation (see footnote
29) of the price indexes for air transportation (NAICS 481); broadcasting and
telecommunications (NAICS 513); securities and commodity brokers (NACIS 523);
computer systems design and related services (NAICS 5415); and information and data
processing services (NAICS 514). This index was constructed using published currentdollar value added and value added price indexes for 1977–2002.38 Unpublished data for
these industries was used to compute the series for 1959–1976.

R&D performing industries (scenario D). In this scenario, the output measure

for R&D is based on the assumption that the unobserved R&D prices are similar to the
prices of industries that perform R&D; it is constructed using BEA’s GDP by industry
value-added price indexes. The R&D performers price index is a Fisher aggregate
consisting of the price indexes for four of the top industrial R&D performers, in terms of
current-dollar expenditures. These top performers were identified based on NSF data on
industry performance of R&D and weighted annually based on their shares of performed
R&D.39 The four industries that consistently make up the vast majority of industry
spending are radio and TV receiving equipment (SIC 366,377); drugs and medicine (SIC

Since the use of these industry price indexes is intended to proxy a pattern of output growth similar to
R&D, one fast-growing industry was excluded; this was Social Assistance which was the fourth fasted
growing industry over the period from 1959 to 2002.
38
Current-dollar value added can be found at: http://www.bea.gov/bea/industry/gpotables/gpo_action.cfm
(lines 37, 48, 49, 53, 62).
Chain-type price indexes for value added can be found at:
http://www.bea.gov/bea/industry/gpotables/gpo_action.cfm (lines 37, 48, 49, 53, 62).
39
National Science Board (2006).

283); office, computing, and accounting machines (SIC 357); and aircraft and missiles
(SIC 372, 376).40

Stocks of R&D Capital

Estimates of net stocks of R&D capital and depreciation are calculated for each
type of funder. Real net stocks for each institutional type are constructed using a
perpetual inventory method: the constant-cost net stock at the end of a year is estimated
as the stock at the end of the previous year, times one less a depreciation rate, plus
constant-cost investment during the year, multiplied by one less half the depreciation rate.
This half is a compromise between assuming all investment happens at the start of the
year versus the end. The depreciation rate is a flat 15 percent in scenario A. In scenarios
B, C, and D, the depreciation rate is a varying time series, using the NIPA-based rate for
the basket of assets comprising information-processing equipment and software for
1987–2002, extrapolated using the NIPA-based rate for the basket comprising private
fixed investment in nonresidential equipment and software for earlier years. This
alternative depreciation series is intended to capture accelerated depreciation of R&D in
recent years, consistent with faster technological change and shorter-lived assets.
Current-cost net stocks of R&D, which value the stock at the replacement cost,
are derived by pricing the constant-cost net stocks by an end-of-year deflator. Historical-

Average current-dollar weights for the R&D performers index are provided below.
Average weights per time period
195919731973
1995
Radio and TV receiving equipment (SIC 366,377)
30.3
29.4
Drugs and medicine (SIC 283)
10.6
20.1
Office accounting and computer machinery (SIC 357)
30.8
29.5
Aircraft and missiles (SIC 372,376)
28.2
21.0

19952002
40.4
28.8
18.3
12.5

cost net stocks of R&D, which are analogous to estimates used on company reports, are
constructed like constant-cost stocks, but with no deflation. Historical-cost and currentcost aggregates of net stocks and depreciation are obtained by directly summing the
estimates. Aggregate real-cost estimates are derived using the Fisher chain formula.

IV. Assumptions for the Satellite Account and Alternative Scenarios

In developing its preliminary estimates of the impact on the U.S. economy of
recognizing R&D as investment, BEA has balanced two distinct, but interrelated needs:
1) advancing the process of developing a methodology for repeatable estimates that will
be both consistent with the NIPAs and internationally comparable and 2) capturing
results that include the impact of R&D on the economy based on a set of macroeconomic
results reflecting existing consensus about the characteristics of R&D activity. These
needs are, of course, interrelated because the NIPAs are dynamic. The NIPA
comprehensive revision process provides the means for incorporating new concepts into
the national accounts after careful testing and the development of broad consensus among
national accountants and users of the accounts.

The Input-Cost Approach and Three High-Productivity Scenarios

It is useful to compare the assumptions used in each of the alternative scenarios of
the R&D satellite account as well as the results that other researchers have produced in
their estimates of the impact of treating R&D as investment. Each scenario makes
assumptions about: 1) the rate at which R&D depreciates; 2) the deflators used to
estimate real (inflation-adjusted) R&D; 3) the treatment of capital services for R&D
investment, 4) the scope of R&D considered investment; and 5) the assumptions made
about externalities or spillovers from the conduct of R&D.
In general, the measure of GDP will rise by adding capital services for
government and nonprofit R&D, broadening the scope for what is considered R&D, more

rapidly decreasing deflators, and including positive spillovers in the conduct of R&D. All
of the estimates in this paper are based on the same definition of R&D. They reflect only
the direct effects of R&D investment and do not include any additional explicit impacts
based on spillovers.
As this paper has described, BEA’s current estimates have been constructed with
a cost-based approach that starts by summing the expenditures for R&D and adjusting
these expenditures to make the measure of output consistent with national accounting
treatment of own-account, or internally-used investment. These cost-based nominal
estimates are then deflated to develop real, or inflation-adjusted measures of R&D
investment. Real R&D investment is transformed into stocks of R&D capital by
depreciating the cumulative investments of prior years to account for the loss in value
over time.
The two basic approaches used in this paper to produce measures of real R&D
investment produce noticeably different results. The conceptually simplest approach is to
deflate R&D investment with an input cost price index; this is the method used in
Scenario A. Since this method has the effect of not recognizing any productivity
improvement in the conduct of R&D, its results should be considered a lower bound on
the measure of real R&D investment. The other approach is to find an appropriate way to
estimate the real output of R&D. The three alternatives for this estimate; a highproductivity manufacturing industries price index, a high-productivity service sector price
index, and a R&D performing industries price index, have been described in Sections II
and III.

Table 5 summarizes the assumptions for each of the four scenarios reported on in
this account. Scenario A is consistent with most of the current accounting conventions of
the NIPAs. This scenario assumes a constant 15 percent geometric rate of depreciation
for all types of R&D. It also incorporates a set of input deflators that are developed from
the most detailed component costs available. As noted above, the use of these input price
indexes has the effect of not recognizing any productivity improvement in the conduct of
R&D, since the measure of R&D output is simply the sum of the inputs.
Table 5: Assumptions for the Scenarios in the R&D Satellite Account
Parameter
depreciation

A
15 percent

of R&D

B
before 1987: change in
private fixed investment
in nonresidential
equipment and software
depreciation
after 1987: information
processing equipment
depreciation
cost-based price index
adjusted to proxy highproductivity growth in
manufacturing

price index

Input costcomponent
based

net return to
business
R&D (capital
services)
net return to
government
and nonprofit
R&D (capital
services)

same as to
other fixed
assets

average net rate of 15
percent

none

estimated net return
based on long-term
average in the real 10year treasury rate, plus a
higher premium for
R&D investment

Same as B

composite
price index
based on the
value added
of five
highproductivity
service
industries
Same as B

composite price
index based on
the value added
of the four
industries that
perform the
most R&D

Same as B

Scenario A also reflects the current convention of the NIPAs that investment by
governments and nonprofit institutions earn no net return to capital. Further, it assumes
52

no additional net return to R&D relative to other types of assets. This scenario produces
the smallest impact of GDP of all the alternatives tested. Over the period from 1959 to
2002, the average impact on the level of GDP is 2.3 percent for scenario A and the
contribution to the growth in real GDP is 2.17 percent.

The High-Productivity Approach, Scenarios B, C, and D

Scenarios B, C, and D all share a set of assumptions about the characteristics of
R&D as an investment good with a rapid pace of technological change that leads to an
accelerating pattern of obsolescence and the quality of uncertainty surrounding its
outcome that leads to a higher rate of return to R&D investments.
Scenarios B, C, and D each employ a different approach to estimate real R&D
output using price indexes for related industries. Scenario B uses a high-MFP,
manufacturing industry-based, price index. Scenario C uses a composite highproductivity service sector price index; scenario D uses a composite R&D performing
industries price index.
These three high-productivity scenarios all include the depreciation profile
consistent with accelerating obsolescence. As described in Section II, it is based on the
NIPA depreciation series for information processing equipment after 1987 and the
change in the depreciation series for private fixed investment in nonresidential equipment
and software prior to 1987.
Scenarios B, C, and D each also incorporate a set of assumptions about rates of
return to R&D for business, government, and nonprofit institutions: 1) The net rate of
return to business sector R&D is assumed to be higher than the net return to non-R&D

assets; 2) government and nonprofit R&D are assumed to earn a net rate of return in
addition to simply the consumption of fixed capital; and 3) a a higher net rate of return to
government and nonprofit R&D capital is assumed compared with what non-R&D
government and nonprofit capital would earn.
Despite the different methods used to estimate the unobserved real measure of
R&D, all three all yield similar results with respect to the level of GDP and other
macroeconomic effects. The increase in nominal level of GDP is an average of 2.6
percent when R&D is treated as investment in each of the scenarios in table 6. The larger
impact compared with scenario A (2.3 percent) is mainly due to the addition of a net
return to government and nonprofit R&D capital.

Table 6: Overall Impact of Capitalized R&D on GDP Level
NIPA
treatment
GDP
[billions of
dollars]
1960
1970
1980

Scenario B
level

526.4
1,038.5
2,789.5
5,803.1
10,469.6

1990
2002
average
change, all
years

with R&D capitalized

537.4
1,068.6
2,859.3
5,963.3
10,751.5

Scenario B
percent
difference

2.1
2.9
2.5
2.8
2.7
2.6

Scenario C

537.8
1,067.1
2,856.0
5,961.6
10,743.5

Scenario C
percent
difference

Scenario D
level

Scenario D
percent
difference

2.2
2.8
2.4
2.7
2.6

538.0
1,069.0
2,857.5
5,961.9
10,747.3

2.2
2.9
2.4
2.7
2.7

2.6

By recognizing a new class of assets, the satellite account shows a rise in the
measured value of investment. The increase in private gross domestic investment ranges
from 7.5 percent in 1960 to 11.3 percent in 2002 (table 7). The increase is the same for

each scenario, because it is measured the same way in each scenario, is estimated in
nominal dollars, and is not affected by the adjustments to net returns.

Table 7: Impact on Gross Private Domestic Investment and the Saving Rate
Gross Private Domestic Investment
GPDI
NIPA
With R&D as
treatment
investment
Period
(millions)
All Scenarios

National Saving Rate (Note 1)
NIPA treatment
saving rate

with R&D
as investment
Percentage point difference
All Scenarios

1960

78,891

7.5

21.0

2.1

152,378

7.1

18.6

2.0

479,252

6.8

19.7

1.8

860,968

9.8

16.3

2.2

1,582,129

11.3

14.2

2.2

1970
1980
1990
2002

Note 1. Calculated as the ratio of the sum of Gross Saving (NIPA Table 5.1) to the sum of GNI for each time period,
expressed as a percent.

Similarly, the impact on the national savings rate is consistent across all the
scenarios; it increases by approximately two percentage points over the period. This
change is the result of recognizing business sector R&D expenditures as investment
rather than as expenses, and for government and nonprofit institutions, the shift of R&D
expenditures from consumption to investment.
Scenarios B, C, and D produce a relatively tight range of contributions to the real
growth in GDP. The largest contribution to growth comes from scenario B, the high-MFP
index; the contribution to growth was 4.9 percent over the period 1959 to 2002. In
scenario C, the scenario that includes the composite price index from the highproductivity service industries, 1959 to 2002 is an average 4.3 percent. For the late 1990s,
the contribution is 6.3 percent. Using the composite price index for R&D performing
industries yields a similar overall contribution, 4.6 percent, as well as a similar

contribution for the late 1990s, 6.7 percent. All three of these scenarios are characterized
by a bump up in the contribution of R&D investment to growth in the late 1990s. These
results are compared in the table below.

Table 8. Contribution of R&D investment to growth in adjusted real GDP

Years
1959-1973
1974-1994
1995-2002
1959-2002

Scenario B
High-MFP
adjustment
to input cost
price index

Scenario C
Highproductivity
service
industries
price index

Scenario D
R&D
performing
industries
price index

Contribution
4.46
4.68
6.77
4.94

Contribution
3.88
3.86
6.25
4.28

Contribution
4.03
4.33
6.69
4.61

BEA reports the estimates based on scenario D as the preliminary estimates for
the 2006 R&D satellite account. These results represent a mid-range of the three highproductivity options. Line 6 of Appendix Tables 1.1 through 1.4 provide the annual
estimates for real GDP, GDP (nominal), real GDI, and the saving rate based on scenario
D.

V. Looking forward to the 2007 R&D Satellite Account

The discussion in the previous section suggested areas where alternative
assumptions produce significantly different impacts on GDP and growth. BEA intends,
with the support of NSF, to explore the more important issues related to R&D investment
and capitalization in preparing the R&D satellite account in 2007 and in future work on
R&D. These issues include the overlap between R&D and software investment, the
international flows of R&D transactions, improved output measures, improved input
deflators, the treatment of spillovers or externalities associated with R&D activities, and
the ownership of R&D. Each of these issues is discussed briefly below.

Overlap Between R&D and Software

The NIPAs currently recognize software as investment that yields an intangible,
produced fixed asset. This investment includes prepackaged software, custom software,
and own-account software. Because software is currently treated as investment in the
NIPAs, the treatment of R&D expenditures as investment implies that R&D used to
create own-account software may be counted once as investment in an R&D asset, and
again as part of a software asset. This overlap exists because the value of own-account
software is built up from production costs, which primarily consist of the compensation
costs of computer programmers and systems analysts. For computer programmers and
systems analysts who work in R&D facilities, compensation is counted once as an input
to own-account R&D and again as an input to own-account software, resulting in a
double count.

R&D expenditures used to create software assets are currently included as
investment in the R&D satellite account. In 2007, BEA plans to address this overlap
between own-account software and R&D by first adjusting investment in software. This
adjustment includes a subtraction and an addition. The subtraction to investment in
software would be the compensation costs and intermediate expenses associated with the
creation of own-account software by computer programmers and systems analysts in
R&D facilities, or a reclassification of these expenditures from software investment to
R&D investment. The addition to investment in software will be for the consumption of
fixed capital of the new R&D asset used to create the software.
A second step is estimating the size of this overlap. BEA does not currently have
a good measure of the overlap in part because of differences in definition of own-account
software. The NIPA definition is somewhat different from the definition of own-use
software in the Census Survey of Industrial R&D (SIRD). This survey includes R&D
expenditures for software development produced for sale to others, and identifies it
separately in a line item. The survey specifically excludes expenditures for software
development intended for within-company use only. This latter category is defined
differently from the own-account definition in the SNA and implemented in the NIPAs.
In addition to software created for own use within an enterprise, the NIPA measure of
own-account software includes own account originals. These are software originals
created and retained within the same enterprise for use in further production.41 R&D to
create these software originals for further production is likely to be included in the R&D
totals despite the exclusion of expenditures for software for within company use. For the
41

A software original is a unique piece of software created by computer programmers and systems
analysts. Both this original and the copies that can be made from it and sold are treated as investment in
software.

2007 R&D Satellite Account, BEA plans to create estimates of this overlap, based on
data from Census and other sources on purchases of capitalized and expensed software by
industry.

International Transactions in R&D

There are two dimensions of international transactions for R&D and a related
category of payments for the use of R&D: international trade in research, development,
and testing services; business funding of foreign performed R&D; and royalties and
licensing fees for the use of industrial processes, presented in a detailed table in
Appendix Table 8. BEA’s R&D satellite account presents estimates of the stock of R&D
located in the U.S., regardless of the nationality of the owner. The satellite account treats
all domestically-performed business R&D as U.S. assets and excludes R&D performed
abroad by foreign affiliates of U.S. companies. This treatment implicitly assumes that the
private benefits of R&D are obtained in the country where the R&D is performed. The
stock estimates presented in this paper are not yet adjusted for R&D investment by U.S.
and foreign multinationals, or the exports and imports of research, development, and
testing services. Adjustment for exports and imports of research, development, and
testing services is planned for the 2007 R&D satellite account. Including R&D
investment by multinationals requires collection of data not currently available, and is a
longer term project.
The R&D surveys that provide the primary source data for the R&D satellite
account do not currently measure exports and imports of R&D, but they do provide
information on funding for foreign-performed R&D. Complete source data are not

currently available, and the discussion below describes the existing data on international
R&D transactions for the U.S., primarily available from BEA’s international survey
programs, and how they can be used in future versions of the R&D satellite account to
adjust the domestic expenditures.
Currently, trade in R&D services represents a relatively small component of
domestic R&D activity. Exports of R&D services (described in more detail below) of
$7.6 billion dollars in 2002 is small compared to $170.8 billion dollars in businessfunded investment in R&D in the U.S (Table 3.1) The production and sale of R&D
services are a result of R&D activity, and thus, are not directly comparable to domestic
R&D spending.42 However, comparing the volume of trade in R&D services to R&D
spending provides an indicator of its relative importance.
Data on international trade in R&D services between unaffiliated or unrelated
firms are available beginning in 1986 from BEA; for trade in R&D services between
affiliated entities (both are part of the same firm but are located in different countries),
data are available since 2001.43 Although these data represent trade in R&D services for
all industries, not only the R&D services industry (NAICS 5417), they represent a
somewhat broader scope of activities from Frascati based-R&D.
The greater magnitude of affiliated trade for imports and exports highlights the
role of multinational corporations in the international exchange of R&D services.
Affiliated trade is comprised of transactions between 1) U.S. parent companies and their
42

Trade in research and development and testing services in the BEA international services data covers
“laboratory and other physical research, product development services, and product testing services.” The
inclusion of product testing services may make its scope a somewhat broader than other measures of R&D
activity collected by BEA and the National Science Foundation, where the Frascati-based definition is
generally understood to exclude some types of testing services.
43
Prior to 2001 affiliated trade in R&D services was included in larger aggregates; other private services
between 1992–1996; and other business, professional, and technical services between 1997–2000.

foreign affiliates and 2) the U.S. affiliates of foreign parents. Most of the growth in
exports of R&D services is transactions between the U.S. affiliates of foreign firms and
their foreign parents; exports from U.S. parents to their foreign affiliates have been level
during this 4-year period.
Although trade in business R&D services from the U.S. is relatively small, R&D
is increasingly an international activity. This growth is reflected in a number of
dimensions of R&D-related activity, including R&D performed by multinational
corporations outside of their home countries, international trade in R&D services,
international transactions for the purchase and use of patents and other types of
technology, and international collaboration on R&D. For BEA’s R&D satellite account,
isolating the transactions in R&D and identifying both the owner of the R&D investment
and where the R&D is used are key pieces of information necessary to determine the
impact of these international activities on the stock of U.S. R&D.
R&D performed outside of the U.S. that is ultimately used in the U.S. would be
measured as domestic investment in a framework that recognizes R&D as investment. As
shown in the last two columns of Appendix Table 9, BEA and NSF surveys currently
provide information on R&D performed abroad that can be identified with U.S.
ownership or funding; however, they do not include information on where this R&D is
ultimately used. Without the latter information, the U.S. expenditures for foreignperformed R&D cannot be characterized as U.S. assets; the R&D may be used in the
country where it is performed, and thus represent an asset for that country.
To include international transactions in R&D, additional data are needed on where
R&D is used. Moris (2006) calls for collection of these international R&D data that

would identify where the R&D is used. This information would be particularly valuable
for developing stocks of capitalized R&D for the U.S. because this information about use
would serve as a reasonable proxy for the location of the asset—a somewhat inexact
concept when applied to intangibles.

Improved Output and Input Deflators

In 2007, BEA plans to refine its methodology for measuring R&D output. BEA
expects to refine the approaches used in the 2006 preliminary R&D satellite account; in
particular, developing a methodology for weighting the relative importance of high R&D
performing industries. BEA also plans to develop improved R&D price deflators for the
largest input cost: compensation of R&D personnel in business.
Over the longer term, a framework for including R&D in the U.S. industry
accounts needs to be constructed. The goal is to develop a more detailed look at the
makeup of R&D costs across industries, and correspondingly improved R&D deflators
for compensation and the other input costs with an emphasis on certain key industries
such as the manufacturing of computers, electronic products, and pharmaceuticals. For
example, the makeup of R&D personnel (scientists, engineers, technicians, and clerical
support) or the nature of R&D physical capital investment and its depreciation may vary
significantly across industries. The makeup of an industry’s R&D funding may also be
used to develop improved R&D deflators, especially for those industries which have a
high portion of their R&D funded by the Federal government.

Treatment of R&D Spillovers

Any additional impact of R&D investment beyond its share of nominal GDP
would arise from spillovers or externalities. These spillovers or externalities exist when
the social cost (or benefit) exceeds the private cost (or benefit). In the case of R&D, the
gap between the private and social cost creates a spillover that consists of benefits or
losses beyond the control of the R&D stocks’ owner, or the effects of products using new
technology and innovations, for example, production of products that use computers and
software. The assumed existence of positive spillovers from R&D, and the inability of the
owner to appropriate a payment for the R&D, has long been considered a justification for
government support of R&D activity. The 2006 R&D satellite account does not consider
any externalities from R&D activity, and in 2007, different techniques and methods will
be tested and evaluated to determine the feasibility of including spillovers from R&D
investment.

Ownership of R&D Assets

The estimates of R&D stocks and their macroeconomic impacts presented in the
satellite account assume that the funder of R&D owns the R&D. As part of its work to
develop improved concepts for treating R&D as investment, BEA plans to develop a
standard that can be applied to the existing survey data. This standard would use existing
information about the assignment of intellectual property rights—who has the right to
patent and collect royalties— to adjust the funder-based estimates of ownership of R&D.
This information about the assignment of intellectual property rights would then also be
used to begin to develop a measure of “freely available” R&D.

The concept of “freely available” is currently not well defined and thus is difficult
to apply to existing survey data. In the longer term, BEA will work with its data providers
to align the survey questions to the economic concepts necessary to identify ownership
and location of use. Since “freely available” R&D, often produced by governments,
nonprofits, and academic institutions, is likely to have measurable impacts on economic
activity, it is important that this type of R&D be reported separately.

VI. Summary and Conclusion

The recognition of R&D as investment in the NIPAs would represent a major
change in BEA’s treatment of intangible assets. The R&D satellite account, which can be
seen as a step toward that goal, presents preliminary estimates for its impact on GDP,
GDI, contributions to growth, and investment.
Several additional steps to refine the estimates are needed; some steps are slated
for release in September of 2007, and others are long-term projects. Most significant
among these additional steps are the following:
•

An adjustment for imports and exports of R&D services and the overlap
between R&D and software.

•

Development of R&D stocks by type of R&D; that is basic research, applied
research, and development of new products and processes. If BEA is able to
create consistent time series of these stocks, an improved set of estimates
could include depreciation rates that differ by type of R&D.

•

Improvement of the price indexes used for the estimates.

•

Estimates of capital services for R&D. While this would enhance the
usefulness of the R&D capital stocks for productivity analysis purposes,
preliminary capital services estimates are likely to be somewhat speculative
because of the limited availability of price data for the use of R&D. Given the
future directions planned for BEA’s national account in harmonizing BEA
data with the productivity program of BLS and others, developing capital
services estimates for R&D has a high priority. However, the public goods
qualities of R&D and the potential for spillovers complicate the process of

estimating the private rental value of R&D assets necessary from capital
services. BEA staff members are currently engaged in research to develop
econometric estimates of the private and social rates of return to domestic
industrial R&D. Ongoing work at BEA and BLS will develop insights into the
return received from R&D performed outside of the U.S. This work on social
returns might form the basis of a set of approximations of the magnitude of
the externalities or spillovers associated with R&D. These approximations
could be presented as an overlay to the basic satellite account.

In the longer term, the enhancements that BEA is most interested in developing
would require improved survey data. For example, the estimates of consumption of fixed
capital used to create R&D would be greatly improved by survey data on expenditure for
structures, equipment, and software used in the production of R&D. Similarly, better data
on the nature of the transaction between the funder and the performer of R&D would
improve the assignment of R&D to sectors and the separation of domestic R&D
investment from foreign R&D investment.
As BEA considers incorporating R&D as intangible investment in the NIPAs, a
series of estimating challenges emerge. One of the most immediate challenges will be the
alignment of NSF and other data sources with the industry classification systems used for
enterprise and establishment data at BEA. BEA is currently working on developing an
industry framework for R&D that will lead to industry-based estimates for R&D.
The additional dimensions of geographic and time measurement will need to be
addressed before R&D can be fully incorporated into the NIPAs and other BEA accounts.

Current R&D survey data are conducted annually or less frequently, and the publication
lag is usually between 1 and 2 years. For the NIPAs, BEA needs to produce quarterly
estimates with an initial lag of 1 month after the end of the quarter; quarterly R&D
investment estimates will require a methodology based on higher frequency data with
shorter lags than are currently available.

References

Ahmad, N. “Introducing Capital Services into the Production Account.” Issue
Paper SNA/M2.04/15 (2004).

American Association of Engineering Societies, Inc. (AAES), Engineering
Workforce Commission. “Engineers Salaries: Special Industry Report.”
Washington, D.C.

Australian Bureau of Statistics. National Accounts Research Section.
“Capitalising Research and Development in the National Accounts.” (March
2004).

Bernstein, Jeffery and Theofanis P. Mamuneas, “Depreciation Estimation, R&D
Capital Stock, and North American Manufacturing Productivity Growth”
Presented at the Bureau of Economic Analysis, (October 2004).

Bernstein, Jeffrey, and M. Ishaq Nadiri. “Interindustry Spillovers, Rates of
Return, and Production in High-Tech Industries. American Economic Review
Papers and Proceedings 78:429–34 (1988).

Bernstein, Jeffrey, and M. Ishaq Nadiri. “Product Demand, Cost of Production,
Spillovers, and the Social Rate of Return to R&D.” NBER Working Paper No.
3625. Cambridge, MA: National Bureau of Economic Research. (1991).

Caballero, Ricardo J. and Jaffe, Adam B. “How High are the Giants Shoulders:
An Empirical Assessment of Knowledge Spillovers and Creative Destruction in a
Model of Economic Growth.” In Blanchard, O., and Fischer, S., eds., NBER
Macroeconomics Annual 1993, Volume 8. Cambridge MA: The MIT Press,
(1993).

Carson, Carol S., Bruce T. Grimm, and Carol E. Moylan. “A Satellite Account for
Research and Development.” Survey of Current Business, (November, 1994):
37–71.

Clark, B. and Z. Griliches, “Productivity Growth and R&D at the Business Level:
Results from the PIMS Data Base.” R&D, Patents, and Productivity, University
of Chicago Press, Chicago, (1984):339–374.

10.

Corrado, Carol and Hulten, Charles and Sichel, Daniel. “Measuring Capital and
Technology: An Expanded Framework” in Measuring Capital in the New
Economy, editors Carol Corrado, John Haltiwanger, and Daniel Sichel. University
of Chicago Press. (2005): 11–45

11.

Corrado, Carol and Hulten, Charles and Sichel, Daniel. “Intangible Capital and
Economic Growth.” NBER/CRIW Paper, (April 2006).

12.

de Haan, Mark and Myriam van Rooijen–Horsten “Measuring R&D Output and
Knowledge Capital Formation in Open Economies.” 28th General Conference of
the International Association for Research in Income and Wealth, Cork, Ireland,
(August 22–24, 2004).

13.

Draft Report of the Third Meeting of the Advisory Expert Group on National
Accounts, Bangkok (July 18–22, 2005).
http://unstats.un.org/unsd/nationalaccount/AEG.htm

14.

Executive Office of the President Office of management and Budget. “North
American Industry Classification System”. United States (2002)

15.

Fraumeni, Barbara M., and Okubo, Sumiye. “R&D in the National Accounts: A
First Look at Its Effect on GDP.” In Measuring Capital in the New Economy,
editors Carol Corrado, John Haltiwanger, and Daniel Sichel. University of
Chicago Press. (2005): 275–321.

16.

Goto, Akira, and Kazuyuki Suzuki. “R&D Capital, Rate of Return on R&D
Investment, and Spillover of R&D in Japanese Manufacturing Industries.”
Review of Economics and Statistics 71 (November 1989):555–64.

17.

Griliches, Zvi, “Returns to Research and Development Expenditures in the Private
Sector,” in New Developments in Productivity Measurement, Kendrick, J.W. and
Vaccara, B. (Eds.), National Bureau of Economic Research, University of
Chicago Press, Chicago (1980):419–454.

18.

Griliches, Zvi, “Productivity, R&D, and Basic Research at the Firm Level in the
1970’s,” American Economic Review 76:141–54. (1986).

19.

Griliches, Zvi and Mairesse Jacques, “Comparing Productivity Growth: An
Exploration of French and U.S. Industrial and Firm Data,” European Economic
Review 21:89–119 (1983).

20.

Griliches, Zvi and Mairesse Jacques, “Productivity and R&D at the Firm Level,”,
Griliches, Zvi (Ed.), R&D, Patents, and Productivity, National Bureau of
Economic Research, University of Chicago Press, Chicago (1984):339–374.

21.

Griliches, Zvi and Mairesse Jacques, “R&D and Productivity Growth: Comparing
Japanese and U.S. Manufacturing Firms,” National Bureau of Economic Research
Working Paper #1778 (1986).

22.

Griliches, Zvi and Mairesse Jacques, “R&D and Productivity Growth: Comparing
Japanese and U.S. Manufacturing Firms,” Productivity Growth in Japan and the
United States, C. Holten, Ed., University of Chicago Press, Chicago (1990): 317–
348.

23.

Jaffe, Adam B., “Technological Opportunity and Spillovers of R&D: Evidence
from Firm’s Patents, Profits, and Market Value”, American Economic Review
(December 1986):984–1001.

24.

Jorgenson, Dale W. and Griliches, Zvi. “The Explanation of Productivity
Change.” Review of Economic Studies 34 (July 1967): 349–83.

25.

Jorgenson, Dale W., Mun S. Ho, and Kevin J. Stiroh. 2005. Information
Technology and the American Growth Resurgence, 2005 (Cambridge, MA: MIT
Press), pp. 38–39.

26.

Jorgenson, Dale, and Landefeld, Steve. “Blueprint for Expanded and Integrated
U.S. Accounts: Review, Assessment, and Next Steps;” “A New Architecture for
the U.S. National Accounts.” Edited by Jorgenson, Landefeld, and Nordhaus.
NBER Studies in Income and Wealth, Volume 60. University of Chicago Press,
(2006).

27.

Lev, B. and Sougiannis, T. “The Capitalization, Amortization and ValueRelevance of R&D” Journal of Accounting and Economics 21 (1996): 107–138.

28.

Lally, Paul R. “Note on the Returns for Domestic Nonfinancial Corporations in
1960-2005.” Survey of Current Business (May 2006): 6–10.

29.

Lichtenberg, Frank R. and Siegel, Donald, “The Impact of R&D Investment on
Productivity: New Evidence Using Linked R&D-LRD Data,” Economic Inquiry
(April 1991):203–229.

30.

Link, Albert N., “Inter-Firm Technology Flows and Productivity Growth,
Economics Letters (1983) 11:179–84.

31.

Mansfield E., Rapoport J., Romeo A., Wagner S., and Beardsley G., “Social and
Private Rates of Return from Industrial Innovations,” Quarterly Journal of
Economics, vol. 91, no. 2, (May 1977): 221–240.

32.

Mansfield, Edwin, “Industrial R&D in Japan and the United States: A
Comparative Study.” American Economic Review, Papers and Proceedings
(1980) 78:223–28.

33.

Mead, Charles Ian, Karen Moses, and Brent R. Moulton. “The NIPAs and the
System of National Accounts.” Survey of Current Business (December 2004), 17–
31.

34.

Minasian, J.R., “Research and Development, Production Functions and Rates of
Return,” American Economic Review. (1969) 59:80–85.

35.

Moris, Francisco. U.S. International Trade in R&D-Related Services and a
Transactions-based Profile of Business R&D. Prepared for the NBER-CRIW
Conference, Bethesda, MD (April, 2006).

36.

Moulton, Brent R. The Expanding Role of Hedonic Methods in the Official
Statistics of the United States. Paper presented to the BEA Advisory Committee,
November 30, 2001.

37.

Nadiri, M. Ishaq, “Innovations and technological spillovers.” NBER Working
Paper no. 4423. Cambridge, MA: National Bureau of Economic Research,
(August 1993). http:// papers.nber.org/papers/W4423.

38.

Nadiri, M. Ishaq and George Bitros, “Research and Development Expenditures
and Labor Productivity at the Firm Level. Studies in Income and Wealth, Vol. 44,
Conference on New Developments in Productivity Measurement, Edited by John
W. Kendrick and Beatrice N. Vaccara. University of Chicago Press, Chicago
(1980).

39.

Nadiri, M.I., and I. R. Prucha. “Estimation of the Depreciation Rate of Physical
and R&D Capital in the U.S. Total Manufacturing Sector,” Economic Inquiry,
(1996): 34,43–56.

40.

National Center for Education Statistics. “Digest of Education Statistics” (2004)
Table 35.

41.

National Science Board. 2006. Science and Engineering Indicators 2006. Two
volumes. Arlington, VA: National Science Foundation (volume 1, NSB 06-01;
volume 2, NSB 06-01A).

42.

National Science Foundation. Division of Science Resources Statistics. “National
Patterns of R&D Resources: 2002 Data Update (current to October 2002)”.
Arlington, VA (NSF 03-313) (March 2003).

43.

National Science Foundation and U.S. Department of Commerce. “Instructions
for Survey of Industrial Research and Development During 2001” SIRD or Form
RD-1. Form 2-8-2002. (2002)
<http://www.nsf.gov/sbe/srs/sird/form2001/rd1i.pdf>

44.

OECD. Frascati Manual 2002: Proposed Standard Practice for Surveys on
Research and Experimental Development, OECD Publications, Paris, France
(2002).

45.

Pakes, A, and Schankerman, M., “The Rate of Obsolescence of Patents, Research
Gestation Lags, and the Private Rate of Return to Research Resources” in Z.
Griliches ed., R&D, Patents, and Productivity, University of Chicago Press,
Chicago, U.S.A. (1984)

46.

Peleg, Soli: “Harmonization between R&D Statistics and the National Accounts.”
Central Bureau of Statistics, Israel. Paper presented at the NESTI/Canberra II
meeting in Berlin, Germany, (May 2006).

47.

“R&D Salaries Grow Despite Economic Doldrums”, R&D Magazine’s R&D
Industry Guide—August 2003–July 2004:8–11.

48.

Robbins, Carol. “Linking Frascati-based R&D Spending to the System of
National Accounts.” BEA Working Paper, (February, 2006).

49.

“Satisfaction and Salaries Growing in the Lab”, R&D Magazine’s R&D Industry
Guide—August 2002–July 2003:8–11.

50.

Schankerman, Mark, “The Effects of Double-Counting and Expensing on the
Measured Returns to R&D,” Review of Economics and Statistics (August
1981):454–458.

51.

Schankerman, Mark and M. Ishaq Nadiri, “A Test of Static Equilibrium Models
and Rates of Returns to Quasi-Fixed Factors, with an Application to the Bell
System, Journal of Econometrics (1986) 33:97–118.

52.

Scherer, Frederick, “Inter-Industry Technology Flows and Productivity Growth.
Review of Economics and Statistics (November 1982) 64:627–34.

53.

Schultz, Laura. “Estimated the Returns to Private Sector R&D Investment.”
Presented at the Western Economic Association International Annual Conference,
San Diego, CA (June 2006).

54.

Siddiqi, Yusuf and M. Salem: “Treating Research and Development as Capital
Expenditure in the Canadian SNA.” System of National Accounts Statistics
Canada. (March 2005).

55.

Robert Solow, “Technical Change and the Aggregate Production Function,”
Review of Economics and Statistics, 39 (August 1957), pp. 312–20.

56.

Sveikauskas, Leo, “Technology Inputs and Multifactor Productivity Growth.
Review of Economics and Statistics, (1981) 63:275–82.

57.

(SNA 93) Commission of the European Communities -Eurostat, International
Monetary Fund, Organization for Economic Co-operation and Development,
United Nations, World Bank . System of National Accounts 1993,
Brussels/Luxembourg, New York, Paris, Washington, DC.

58.

Terleckyj, Nestor, “Effects of R&D on the Productivity Growth of Industries: An
Explanatory Study. Washington, DC: National Planning Association. 1974.

59.

Triplett, Jack E. and Bosworth, Barry. Productivity in the U.S. Services Sector:
New Sources of Economic Growth, 2004 (Washington, D.C.: Brookings
Institution Press).

60.

U.S. Department of Commerce. Bureau of Economic Analysis. Fixed Assets and
Consumer Durable Goods in the United States, 1925–99. Washington, DC: U.S.
Government Printing Office, September, 2003.

61.

“U.S. R&D Salary Survey Data”, R&D Magazine’s R&D Industry Guid—August
June 2001–May 2002:20–22.

Full text of Bureau of Economic Analysis (BEA) News Releases : Preliminary Study Shows Research and Development Contribution to Economic Growth, BEA 06-44 : Full Report

FRASER