Information about http://www.bea.gov/papers/pdf/Mead_RD_Paper_wp.pdf
R&D Depreciation Rates in the 2007 R&D Satellite Account …
Tags: background paper, bea, bureau of economic analysis, carlos m gutierrez, chief statistician, cynthia a glassman, department of commerce, depreciation rates, deputy director, economic affairs bureau, literature, mead, moylan, national science foundation, preface, research and development, rosemary, satellite, statistics administration, u s department,Pages: 13
Language: english
Created: Fri Feb 8 13:56:40 2008
Display cached document
Page 1
Page 2
Page 3
Page 4
Page 5
Page 6
Page 7
Page 8
Page 9
Page 10
Page 11
Page 12
Page 13
R&D Depreciation Rates in the 2007 R&D Satellite Account
Bureau of Economic Analysis/National Science Foundation
2007 R&D Satellite Account Background Paper
By Charles Ian Mead
U.S. DEPARTMENT OF COMMERCE
Carlos M. Gutierrez
Secretary
ECONOMIC AND STATISTICS ADMINISTRATION
Cynthia A. Glassman
Under Secretary for Economic Affairs
BUREAU OF ECONOMIC ANALYSIS
J. Steven Landefeld
Director
Rosemary D. Marcus
Deputy Director
November 2007 www.bea.gov
Preface
This paper discusses how the industry-specific rates of depreciation for R&D used to
construct the 2007 R&D Satellite Account were determined after a review of the
literature.
Acknowledgements
Dennis J. Fixler, Chief Statistician of the Bureau of Economic Analysis oversaw the
development of the research presented in this paper. Carol A. Robbins, Carol E.
Moylan, Laura I. Schultz, and Brian K. Sliker provided value comments.
Abstract
This paper is part of a series that provides the details behind the Bureau of Economic
Analysis's (BEA) satellite account on research and development (R&D) activity. In the
current work, the focus is on industry-specific depreciation rates for business R&D
capital. This paper begins by discussing the literature on R&D depreciation rates. It then
describes how the rates in the 2007 satellite account were chosen from the related
findings.
November 2007
R&D Depreciation Rates the 2007 R&D Satellite Account
Bureau of Economic Analysis/National Science Foundation
R&D Satellite Account Background Paper
By Charles Ian Mead
One of the improvements introduced in the 2007 Research and Development
Satellite Account (R&DSA) are industry-specific depreciation rates that affect the
measures presented for rates of return, capital stocks, and consumption of fixed capital.
This change not only increases the quality of the measures presented in the main satellite
account, but it also allows for the presentation of meaningful estimates on industry-
specific rates of return to R&D capital. This paper discusses how the industry-specific
depreciation rates used in the R&D satellite account were determined.
I. Background
The 2006 R&DSA was based on two sets of assumptions about the rate at which
R&D capital depreciates. In the "base case," all R&D capital was assumed to depreciate
(obsolesce) at an annual rate of 15.0 percent. In the "alternative scenarios," all R&D
capital was assumed to depreciate at the same rate as that of nonresidential equipment
and software before 1987 and at the same rate as that of information processing
equipment afterwards. These rates were based on a literature review on that focused on
an aggregate rate of depreciation for all business R&D capital, which was the level of
aggregation presented in the 2006 estimates.
In 2007, a review of literature was undertaken that resulted in the choice of the
following set industry-specific depreciation rates for R&D capital in the 2007 R&DSA:
Transportation equipment, 18.0 percent; computer and electronics, 16.5 percent;
chemicals, 11.0 percent, and "all other," 15.0 percent. 1 In the featured results, these
individual rates are assumed to be constant over time. However, some evidence suggests
that a more appropriate assumption might be R&D depreciation rates that increase over
time. To see the effect that increasing rates would have on the measures in the satellite
account, alternative measures based on this assumption are presented in the R&D
background paper on rates of return to capital. 2
II. Review of the Literature
Since studies that calculate industry-specific depreciation rates are limited and
have varying results, this section begins by focusing on depreciation rates for the R&D
capital of business. This initial focus allows the industry-specific rates that are discussed
afterwards to be effectively benchmarked to a reasonable rate for the R&D capital of
business in the final section of the paper.
A) Empirical methods
Four types of empirical specifications are used to estimate R&D depreciation
rates--production functions, amortization models, patent renewal models, and market
valuation models. However, the most appropriate choice among these models is still an
unresolved issue: None seem completely satisfactory because they are based on strong
identifying assumptions or applied to data that lack sufficient variation to separately
identify R&D depreciation rates.
Hall (2007) clearly illustrates some of the issues associated with estimating R&D
depreciation rates using a production function by discussing the types of identifying
assumptions that are often needed to separately identify R&D depreciation rates. The
first of these models assumes that firms exist in a perfectly competitive market place,
which Hall mentions is inconsistent with the notion that R&D is often conducted to
generate monopolistic returns. The second assumes that the output elasticities of
ordinary capital and R&D capital are proportional to their input shares, which Hall
1
Since the literature examines businesses, the R&D of government and nonprofit institutions is assumed to
depreciate at 15 percent in the 2007 R&DSA, which is the same assumption used in the 2006 R&DSA.
This lower depreciation rate reflects the likelihood that the R&D activity of government and nonprofit
institutions is more heavily weighted toward basic research, and thus depreciates more slowly.
2
See Brian K. Sliker, "Rates of Return to R&D Capital," 2007 R&D Satellite Account Methodology Paper,
BEA Website Publication, forthcoming.
characterizes as a "heroic" assumption that also may also introduce a notable amount of
specification error into estimation results.
Other assumptions are also used in the production function approach framework.
For instance, the empirical results presented in Nadiri and Prucha (1996) and in Bernstein
and Mamuneas (2004; 2006) are based on assumptions about future price expectations. It
is unclear how much specification error these assumptions may introduce into the
estimates. In addition, some economists, such as Huang and Diewert (2007), have more
broadly argued that many of these models may inappropriately model the role of R&D in
production by treating it in the same as ordinary physical capital. In particular, since
R&D capital does not lose value in the same manner as physical capital (wearing out
from general use in production), some argue that R&D capital should be treated as a
factor that increases the production possibilities faced by a firm rather than an input in
production.
Perhaps the most intellectually appealing of these models is Huang and Diewert.
Not only does it treat R&D capital as a factor that shifts out the production possibilities
faced by a firm, but it also considers firms that can experience monopolistic markups
associated with R&D investment. However, Huang and Diewert state that the estimation
results are preliminary because their model still excludes some important features
associated with R&D investment, such as the role of uncertainty, and their results have
not been checked against the results from other functional forms for a production
function. Their model also does not fit the data well in some industries.
Results from amortization models, such as those presented in Lev and Sougiannis
(1996) and Ballester, Garcia-Ayuso, and Livnat (2003), are based on more general set of
models that attempt to explain the returns on R&D investment. However, the resulting
estimates are subject to similar concerns as those raised about results from production
function models. For example, the results of Lev and Sougiannis are based on an
assumed relationship between the amortization rate of R&D capital and earnings that
these assets generate. The results are also based on the assumption that operating income
serves as a good proxy for R&D benefits.
Results from patent renewal models, such as those presented in Pakes and
Shankerman (1984), are subject to a different set of concerns. These models estimate the
rate of obsolescence associated with R&D capital by using information on renewed
patents to estimate a model in which firms maximize the present discounted value of their
returns to R&D investment. Yet patent renewals are not necessarily a good measure of
the value of the knowledge created by R&D because the value of this knowledge may not
be well approximated by the price of the renewal. Even when attempts are made to
address this consideration, another limitation is that not all R&D activity is associated
with the filing of patents.
Market valuation models, such as those presented in Hall (2006), estimate R&D
depreciation rates from model related to the market value of a firm. A potential problem
related to such models is that capital markets may be inefficient, which is inconsistent
with what these models assume.
Although none of the empirical models are completely satisfactory or totally
preferable to the others, there is one data issue that affects the selection of the
depreciation rates used in the R&D satellite account. Specifically, results from studies
that rely on data collected at the industry level, such as those in Bernstein and Mamuneas
(2006) and in Huang and Diewert, are more applicable than results from studies that rely
on data collected at the firm level, such as those in Lev and Sougiannis and in Ballester,
Garcia-Ayuso, and Livnat, because the 2007 R&DSA plans to present industry-level data.
Yet firm-level results may be driven by a large number of firms that actually contribute
little to industry-level capital stocks. R&D capital may also obsolesce quicker for
individual firms than for an industry as a whole. 3
B) Estimates for business R&D
Studies of business R&D find depreciation rates that range from 12 percent to 29
percent. However, the three highest rates in this range should be discounted in the
consideration of the appropriate rates for the 2007 R&DSA for a variety of reasons (table
1). 4
3
The notion that R&D capital may obsolesce quicker for individual firms than for an industry as a whole is
discussed in Sviekauskas (2004).
4
The study of Cabellero and Jaffe (1995) is excluded from this section because the results are difficult to
relate to those of the other studies and are more closely relate to the depreciation rates that increase over
time, which are discussed later in the paper.
The European-based rate found in Pakes and Shankerman should be discounted
for two reasons. First, it is based on a sample of patent returns over the period of 1930-
1939. The other studies use data related that are related to the more recent periods of
time that are covered in the 2007 R&DSA. Second, the European-based depreciation rate
for R&D capital may differ from those in the United States.
Although the U.S. rates found in Bernstein and Mamuneas (2004) and in Hung
and Diewert are based on industry-level data and the most applicable to the 2007
R&DSA, the results of these two studies are discounted for their own separate reasons.
The empirical model in Bernstein and Mamuneas produces industry-level depreciation
rates that are notably higher than those of studies based on firm-level data. Although this
result might be the result of industry-level data, one would expect the estimated rates in
Bernstein and Mamuneas to be lower than those usually found in the studies that use
firm-level data because R&D capital is more likely to obsolesce at slower rate within an
industry than within an individual firm. As previously mentioned, the empirical results
of Huang and Diewert are preliminary, and sometimes their model has difficulty
separately identifying depreciation rates for the R&D capital of some industries.
In light of these considerations, the 15 percent depreciation rate for R&D capital
that is commonly assumed in studies of the net return to R&D capital is consistent with
the empirical evidence, which seems to indicate that the range of 15 to 20 percent is
correct for the depreciation rate of business R&D.
C) Estimates for industry-specific R&D
Studies of industry-specific R&D find depreciation rates that range from -11
percent to 52 percent (table 2). However, the rates found in Hall are dropped from
consideration in the 2007 R&DSA because the rates that are presented in the first portion
of the study, which are based on a production function seem unreasonably low, and the
rates that are presented in the second portion of the paper, which are based on a market
valuation model, seem unreasonably high. Once these rates are dropped, the industry-
specific depreciation rates more reasonably range from 1 percent to 29 percent.
The industry-specific rates found both in Bernstein and Mamuneas (2006) and in
Huang and Diewert are also discounted. In addition to the reasons mentioned above, a
few of the industry-specific rates in Huang and Diewert are estimated with poorly-
behaving data, which may explain the questionable magnitude of the estimates for
industrial machinery and chemicals.
Two notable characteristics that are associated with the relative magnitudes of the
depreciation rates are found in the remaining studies. First, with the exception of
Ballester, Garcia-Ayuso, and Livnat, the chemical industry always has the lowest
estimated depreciation rate when compared to other industries within a study. Second,
with the exception of Bernstein and Mamuneas, the transportation equipment and
scientific instrument industries always have depreciation rates that are higher than
average within a study.
III. Estimates for the Satellite Account
The industry-level estimates featured in the 2007 R&DSA are based on a two-step
process. In the first, the midpoints of the range of estimates given by all the studies other
than Hall are calculated for each industry. In the second step, these midpoints are scaled
down so that the recommended rates are more closely centered on a value of 15 and that
the overall ranking of industry-level rates suggested by the literature is preserved.
Although more detailed estimates for R&D depreciation rates are provided in the
literature, the rates in the 2007 R&DSA are limited to a few 3-digit North American
Industry Classification System codes.
Since there is no clear empirical evidence to indicate otherwise, R&D capital is
assumed to decline in a geometric pattern, which is consistent with the use of constant
depreciation rates. Further, related measures for consumption of fixed capital are
calculated with the standard half-year adjustment used by BEA. Both of these practices
also maintain consistency with the treatment of most other types of assets in the BEA's
accounts. 5
Although the featured measures in the 2007 R&DSA are based on constant
depreciation rates for R&D capital, some argue that R&D depreciation rates should
5
For more information on the conventions used by BEA to calculate capital stocks and consumption of
fixed capital estimates, see the chapter entitled "Concepts and Methods" in Bureau of Economic Analysis,
Fixed Asset and Consumer Durable Goods in the United States, 1925-97 (Washington, DC: U.S.
Government Printing Office, September 2003).
increase over time. For example, Cabellero and Jaffe (1995) find increasing
obsolescence rates for all R&D capital over time. However, it is unclear whether its
results are primarily driven by changes in the composition of R&D assets across
industries or "true" increases in obsolescence rates because the study does not estimate
industry-level depreciation rates. Regardless, many practitioners in the field also believe
that R&D depreciation rates have increased in more recent years within their own
industries.
To estimate the effect that increasing rates have on industry-level rates of return,
the industry-level depreciation rates from the 2007 R&DSA are assumed to all increase
over time at the same rate as the depreciation rate of nonresidential equipment and
software before 1987 and at the same rate as that of information processing equipment
afterwards. These alternative conventions, which are less than perfect due to the lack of
more appropriate data, are based on the notion that the value of R&D investment is
heavily embedded in these types of assets, and the resulting measures are presented in a
supplemental table in the R&D background paper on rates of return to R&D capital,
which is available at .
References
Ballister, Marta, Mannuel Garcia-Ayuso, and Joshua Livnat. 2003. "The Economic Value
of the R&D Intangible Asset." European Accounting Review, Vol. 21, No. 4, 605-
633.
Bernstein, Jeffery I., and Theofanis P. Mamuneas. 2006. "R&D Depreciation, Stocks,
User Costs and Productivity Growth for US R&D Intensive Industries."
Structural Change and Economic Dynamics, Vol. 17, 70-98.
_____, and _____. 2004. "Depreciation Estimation, R&D Capital Stock, and North
American Productivity Growth." Paper presented at the Bureau of Economic
Analysis, Washington, DC, November 4.
Caballero, Ricardo J., and Adam B. Jaffe. 1993. "How high are the Giant's Shoulders: An
Empirical Assessment of Knowledge Spillovers and Creative Destruction in a
Model of Economic Growth." In NBER Macroeconomics Annual 1993, eds. O.
Blanchard and S. Fischer. Cambridge MA: The MIT Press.
Hall, Bronwyn H. 2007. "Measuring the Returns to R&D: The Depreciation Problem."
NBER Working Paper 13473. Cambridge, MA: National Bureau of Economic
Research.
_____. 2006. "R&D, Productivity, and Market Value." Manuscript, University of
California at Berkeley.
Huang, Ning, and Erwin Diewert, 2007. "Estimation of R&D Depreciation Rates for the
U.S. Manufacturing Sector and Four Knowledge Intensive Industries." Paper
prepared for the Sixth Annual Ottawa Productivity Workshop held at the Bank of
Canada, May 14-15.
Lev, Baruch, and Theodore Sougiannis. 1996. "The Capitalization, Amortization, and
Value-relevance of R&D." Journal of Accounting and Economics, Vol. 21, 107-
138.
Nadiri, M. Ishaq, and I. R. Prucha. 1996. "Estimation of the Depreciation Rate of
Physical and R&D Capital in the U.S. Total Manufacturing Sector." Economic
Inquiry, Vol. 34, 43-56.
Pakes, Ariel, and Mark A. Shankerman. 1986. "Estimates of the Value of Patent Rights in
European Countries During the Post-1950 Period," The Economic Journal, Vol.
96, No. 384, 1052-1076.
Sviekauskas, Leo. 2004. "R&D and Productivity Growth: A Review Article."
Manuscript, Bureau of Labor Statistics.
Table 1. Depreciation rates for all R&D capital
Author (Year) Annual rate of depreciation Estimation technique Comments
(R&D)
Pakes and 0.25 Patent renewal model Patent renewal rates
Shankerman (1984) over the period of
1930-1939 for
France, UK,
Netherlands, and
Switzerland
Nadiri and Prucha 0.12 Production function U.S. manufacturing
(1996) (simultaneous industries over the
estimation of labor and period of 1960-1988
material demand
functions)
Lev and Sougiannis 0.15 Amortization model 825 U.S. firms over
(1996) (average annual rate) the period of 1975-
1991
Ballester, Garcia- 0.12 Amortization model 652 U.S. firms over
Ayuso, and Livnat the period of 1985-
(2003) 2001 for preferred
specification
Bernstein and 0.25 for U.S. Production function Manufacturing
Mamuneas (2004) 0.21 for Canada industries over the
period of 1953-1998
for U.S. and over the
period of 1963-1995
for Canada
Huang and Diewert 0.29 Production function Manufacturing
(2007) (R&D shifts production industries over the
function, and model period of 1953-2001
includes a monopolistic for U.S.
markup)
Table 2. Depreciation rates for industry-level R&D capital
Author (Year) Annual rate of depreciation Estimation technique Comments
(R&D)
Lev and Sougiannis · Scientific instruments, 0.20 Amortization model 825 U.S. firms over
(1996) · Transportation equipment, 0.14 the period of 1975-
· Industrial machinery, 0.14 1991
· Electrical equipment, 0.13
· Chemicals, 0.11
Ballester, Garcia- · Transportation equipment, 0.17 Amortization model 1092 U.S. firms over
Ayuso, and Livnat · Chemicals, 0.14 the period of 1985-
(2003) · Industrial machinery, 0.14 2001 for preferred
· Scientific instruments, 0.14 specification
· Electrical equipment, 0.13
Bernstein and · Electrical equipment, 0.29 Production function U.S. industries over
Mamuneas (2006) · Industrial machinery, 0.26 the period of 1954-
· Transportation equipment, 0.21 2000
· Chemicals, 0.18
Hall (2006) Production function: Production function 16750 U.S. firms
· Metals and machinery, - 0.02 (first set of results); over the period of
· Miscellaneous, - 0.02 Market valuation model 1974-2003
· Chemicals, - 0.02 (second set of results)
· Electrical equipment, - 0.03
· Computers and scientific
instruments, - 0.05
· Drugs and medical
instruments, - 0.11
Market valuation model:
· Electrical equipment, 0.52
· Metals and machinery, 0.43
· Computers and scientific
instruments, 0.42
· Miscellaneous, 0.24
· Chemicals, 0.22
· Drugs and medical
instruments, 0.16
Huang and Diewert · Transportation equipment, 0.27 Production function U.S. industries over
(2007) · Electrical equipment, 0.14 (R&D shifts production the period of 1953-
· Industrial machinery, 0.03 function, and model 2001
· Chemicals, 0.01 includes a monopolistic
markup)