Tax Policy and Administration: Review of Studies of the Effectiveness of
the Research Tax Credit (Letter Report, 05/21/96, GAO/GGD-96-43).
Pursuant to a congressional request, GAO reviewed eight studies of the
research tax credit, focusing on the: (1) adequacy of the studies' data
and methods to determine the amount of research spending stimulated per
dollar of foregone tax revenue; and (2) other factors that determine the
credit's value to society.
GAO found that: (1) four studies supported the claim that, during the
1980s, the research credit stimulated research spending that exceeded
its revenue cost, but the other four studies did not support the claim
or were inconclusive; (2) all of the studies had significant data and
methodological limitations that made it difficult to evaluate industry's
true responsiveness to the research tax credit; (3) the studies did not
use tax return data to determine the credit's incentive because the
authors did not qualify for access to such data; (4) publicly available
data were not a suitable substitute for the tax return data because
public sources used different definitions of taxable income and research
spending; (5) the studies' analytical methods, such as use of industry
aggregates and failure to incorporate important tax code interactions,
made their findings imprecise and uncertain; (6) there was little
research on the latest design of the credit to determine its effect on
incentives and costs; (7) the studies' evidence was not adequate to
conclude that a dollar of research tax credit would stimulate a dollar
of additional short-term research spending or about two dollars of
additional long-term research spending; and (8) to measure the credit's
true impact, the studies would need to assess the research's net benefit
to society, resource costs of research, and administrative, compliance,
and efficiency costs of funding the credit.
--------------------------- Indexing Terms -----------------------------
REPORTNUM: GGD-96-43
TITLE: Tax Policy and Administration: Review of Studies of the
Effectiveness of the Research Tax Credit
DATE: 05/21/96
SUBJECT: Tax credit
Statistical methods
Data collection operations
Research and development costs
Tax administration
Fiscal policies
Economic analysis
Research and development
Tax returns
IDENTIFIER: Research Tax Credit
******************************************************************
** This file contains an ASCII representation of the text of a **
** GAO report. Delineations within the text indicating chapter **
** titles, headings, and bullets are preserved. Major **
** divisions and subdivisions of the text, such as Chapters, **
** Sections, and Appendixes, are identified by double and **
** single lines. The numbers on the right end of these lines **
** indicate the position of each of the subsections in the **
** document outline. These numbers do NOT correspond with the **
** page numbers of the printed product. **
** **
** No attempt has been made to display graphic images, although **
** figure captions are reproduced. Tables are included, but **
** may not resemble those in the printed version. **
** **
** Please see the PDF (Portable Document Format) file, when **
** available, for a complete electronic file of the printed **
** document's contents. **
** **
** A printed copy of this report may be obtained from the GAO **
** Document Distribution Center. For further details, please **
** send an e-mail message to: **
** **
** **
** **
** with the message 'info' in the body. **
******************************************************************
Cover
================================================================ COVER
Report to the Honorable
Robert T. Matsui, House of Representatives
May 1996
TAX POLICY AND ADMINISTRATION -
REVIEW OF STUDIES OF THE
EFFECTIVENESS OF THE RESEARCH TAX
CREDIT
GAO/GGD-96-43
Research Tax Credit
(268715)
Abbreviations
=============================================================== ABBREV
IRS - Internal Revenue Service
NSF - National Science Foundation
OTA - Office of Technology Assessment
R&D - research and development
Letter
=============================================================== LETTER
B-270447
May 21, 1996
The Honorable Robert T. Matsui
House of Representatives
Dear Mr. Matsui:
During a May 1995 hearing on the future of the research tax credit,\1
you asked us to evaluate recent studies of the effectiveness of the
research tax credit, particularly those cited in a report prepared by
KPMG Peat Marwick,\2 to determine whether the studies provide
adequate evidence to conclude that each dollar taken of the tax
credit stimulates at least one dollar of research spending in the
short run, and about two dollars of research spending in the long
run. This report responds to your request and evaluates the six
studies cited in the KPMG Peat Marwick report as well as two other
recent studies of the credit. Our objectives were to (1) evaluate
the studies for the adequacy of the data and methods used to
determine the amount of research spending stimulated per dollar of
forgone tax revenue and (2) identify factors other than spending per
dollar of revenue cost that determine the credit's value to society.
--------------------
\1 The May 10, 1995, hearing was held by the Subcommittee on
Oversight, House Ways and Means Committee. We testified at that
hearing and at an April 3, 1995, hearing on the credit held by the
Subcommittee on Taxation and Internal Revenue Service Oversight,
Senate Committee on Finance. See Tax Policy: Information on the
Research Tax Credit (GAO/T-GGD-95-140, Apr. 3, 1995) and Tax Policy:
Additional Information on the Research Tax Credit (GAO/T-GGD-95-161,
May 10, 1995).
\2 KPMG Peat Marwick, LLP, Policy Economics Group, Extending the R&E
Tax Credit: The Importance of Permanence. Prepared for the Working
Group on Research and Development, Nov. 1994.
BACKGROUND
------------------------------------------------------------ Letter :1
HISTORY OF THE RESEARCH TAX
CREDIT
---------------------------------------------------------- Letter :1.1
In 1981, Congress created the research tax credit to encourage
business to do more research. The credit has never been a permanent
part of the tax code. Since its enactment on a temporary basis in
1981, the credit has been extended six times and modified four times.
The research tax credit has always been incremental in nature.
Taxpayers receive a credit only for qualified research spending that
exceeds a base amount. Beginning in 1981, taxpayers could reduce
their tax liability by 25 percent of qualified research that exceeded
a base amount that was equal to the average research expenditure of
the 3 previous years or a base amount that was equal to 50 percent of
the current year's expenditures, whichever was greater.
The Tax Reform Act of 1986 modified the credit by reducing the rate
to 20 percent of qualified spending above the base amount and more
narrowly defining qualified expenditures. The credit was changed
again in 1988 to require that taxpayers reduce their deductions for
research expenditures by an amount equal to 50 percent of the credit
they claim. In 1989, this amount was increased to 100 percent of the
credit they claim.
The Omnibus Budget Reconciliation Act of 1989 changed the method for
calculating the base amount. The base calculated as the average
expenditure of the 3 previous years was replaced by a base amount
equal to the ratio of total qualified research expenses to total
gross receipts for 1984 through 1988, multiplied by the average
amount of taxpayer's gross receipts for the preceding 4 years. This
base change removed the link between increases in current spending
and future base amounts that had reduced the incentive to undertake
additional research spending under the prior method for calculating
the base.
EARLY STUDIES OF THE
CREDIT'S EFFECTIVENESS
---------------------------------------------------------- Letter :1.2
The evaluation of the effectiveness of the credit requires first
estimating the additional research spending stimulated by the credit.
Ideally, this additional spending should then be evaluated according
to the net benefit it produces for society. However, this net social
benefit is difficult to determine because it depends on how the
research of some companies affects the costs and products of other
companies. Some researchers who have studied the credit have instead
calculated a "bang-per-buck" ratio, the amount of spending stimulated
per dollar of revenue cost. Once a decision has been made to provide
some form of credit, this ratio is a relevant criterion for assessing
alternative designs.
Most early studies of the research tax credit found that, although
the credit may have stimulated some additional research spending, the
effect on spending was relatively small. For example, Edwin
Mansfield in his 1986 study asked a random sample of corporate
officials to assess the effect of the credit on research spending and
estimated from their responses that the additional spending induced
by the credit equaled about one-third of the revenue cost.\3
Robert Eisner, et al., in 1984 compared the growth of research
spending that qualified for the credit and spending that did not
qualify in 1981 and 1982, and found no positive impact of the credit
on the growth of research spending.\4
Other early studies relied on estimates of the responsiveness of
research spending to reductions in its price to arrive at similar
conclusions. Because the credit is effectively a reduction in the
price of research, the greater the responsiveness of research
spending to price reductions, the more additional spending the credit
is likely to stimulate. Economists measure the responsiveness in
terms of the "price elasticity" of spending, which shows the
percentage increase in spending that would result from a 1-percent
reduction in the after-tax price of research and development (R&D).
In 1989, we reported that the best available evidence indicated that
research spending is not very responsive to price reductions. Most
estimates of the price elasticity of spending fell in the range of
-0.2 to -0.5, implying that a 1-percent reduction in the price of
research would lead to between a 0.2 percent and 0.5 percent increase
in spending.\5
In our 1989 report, we used Internal Revenue Service (IRS) data to
estimate that between 1981 and 1985, the credit provided companies
with a benefit of 3 to 5 cents per dollar of additional spending.
This benefit to companies is equivalent to a reduction in the price
of research. Combining these price reductions with the range of
elasticity estimates, we estimated that the credit stimulated between
$1 billion and $2.5 billion of additional research spending between
1981 and 1985 at a cost of $7 billion in tax revenue. Thus, we
estimated that each dollar of taxes forgone stimulated between 15 and
36 cents of research spending.\6
--------------------
\3 Edwin Mansfield, "The R&D Tax Credit and Other Technology Policy
Issues," American Economic Review, Vol. 76, 1986, pp. 190-94.
\4 Robert Eisner, Steven H. Albert, and Martin A. Sullivan, "The
New Incremental Tax Credit for R&D: Incentive or Disincentive?,"
National Tax Journal, Vol. 37, 1984, pp. 171-83.
\5 Tax Policy and Administration: The Research Tax Credit Has
Stimulated Some Additional Research Spending (GAO/GGD-89-114, Sept.
5 ,1989). In this report, we summarized the estimates of the
elasticities found in the economic literature. We also noted that
limited dissent from this range of elasticity estimates came from
Baily and Lawrence who tested three models and obtained elasticity
estimates of -0.2, -0.3, and -1.2. We stated in our report that the
third, higher estimate was derived from a very simple model that
employed aggregate rather than firm level data. See Martin Neil
Baily and Robert Z. Lawrence, "The Need for a Permanent Tax Credit
for Industrial Research and Development." Paper commissioned by the
Coalition for the Advancement of Industrial Technology, Feb. 1985.
\6 Again, Baily and Lawrence dissented from these estimates of the
credit's effectiveness. Using a higher elasticity estimate of -1.0,
they calculated that the credit stimulated two dollars of additional
spending per dollar of revenue cost between 1982 and 1985. See
Martin Neil Baily and Robert Z. Lawrence, "Tax Policies for
Innovation and Competitiveness." Paper commissioned by the Council on
Research and Technology, Apr. 1987.
REVIEWS OF RECENT STUDIES OF
THE CREDIT'S EFFECTIVENESS
---------------------------------------------------------- Letter :1.3
Reports on the research tax credit by KPMG Peat Marwick and by the
Office of Technology Assessment (OTA)\7 include reviews of studies of
the credit's effectiveness that were issued since our 1989 report.
The KPMG Peat Marwick report concludes that the studies provide
evidence that the spending stimulated by the credit equals or exceeds
its revenue cost. Specifically, the report concludes that the recent
studies show that one dollar of credit stimulates about one dollar of
R&D spending in the short run, and as much as two dollars in the long
run. According to the KPMG Peat Marwick report, the recent studies
KPMG Peat Marwick reviewed provide better estimates of the
effectiveness of the credit than earlier studies because they analyze
longer data series and because they use what it considered to be
better methodologies for analyzing the effect of the credit. The OTA
report reviewed the same recent studies as KPMG Peat Marwick and
observed that the available literature generally reports that the
credit stimulates about one dollar of additional spending per dollar
of revenue cost. However, OTA pointed out that the studies contain
data and methodological uncertainties. For our review, we evaluated
the studies cited in these two reports as well as other studies not
included in either report. We also addressed some methodological
issues that were not addressed in these reports and provided a more
detailed evaluation of each study.
--------------------
\7 The Effectiveness of Research and Experimentation Tax Credits,
Office of Technology Assessment, OTA-BP-ITC-174, Sept. 1995.
RESULTS IN BRIEF
------------------------------------------------------------ Letter :2
The studies we reviewed generally indicated that the amount of
research spending stimulated by the research tax credit had been
larger than estimated by most of the studies of the credit published
during the 1980s. However, the studies provided mixed evidence on
the amount of spending stimulated by the credit per dollar of revenue
cost. Half of the studies provided estimates in support of the claim
that, during the 1980s, one dollar of research credit stimulated at
least one dollar of additional research spending. The estimates made
in the remaining studies either do not support that claim or are
inconclusive regarding the relationship between the revenue cost of
the credit and the amount of spending stimulated.
The estimates presented in the recent studies do not provide all of
the information needed to evaluate the effectiveness of the latest
version of the credit. The amount of research spending stimulated
per dollar of revenue cost depends not only on the responsiveness of
spending to a tax incentive, but also on how the design of the credit
affects (1) the incentive to increase spending and (2) the revenue
cost. There has been little research on how the latest design of the
credit has affected incentives and costs. As we reported in our May
1995 testimony, the redesign of the credit in 1989 should have
increased the size of the incentive provided per dollar of revenue
cost. However, as we also reported, there is evidence that the
incentive provided by the new credit had eroded over time and that
the revenue cost of the additional spending stimulated by the credit
had increased. Only one of the new studies used any data for tax
years after 1989, and the author of that study was not confident of
her results for the post-1989 period.
Although most of the studies we reviewed used more sophisticated
statistical techniques and more years of data than prior studies of
the credit, all of the studies have data and methodological
limitations that are significant enough to lead us to conclude that
much uncertainty remains concerning the true responsiveness of
research spending to tax incentives. The authors of these studies
themselves, in many cases, said that their results should be used
with caution and would benefit from further research.
The studies may not accurately measure the amount of spending
stimulated by the credit because they used publicly available data
rather than tax return data to determine the incentive provided by
the credit.\8 We concluded from our analysis, and studies by other
researchers confirm, that the publicly available data are not a
suitable proxy for tax return data when measuring this incentive
because the public sources use different definitions of taxable
income and research spending. The tax return data were unavailable
to the authors of the studies that we reviewed.
The following methods, which the studies used to analyze the credit,
also leave room for imprecision and uncertainty:
-- The estimates of three studies that analyzed the credit at the
industry level rather than the company level may not be very
precise because their analyses do not reflect the different
incentives that companies face and their different responses to
these incentives.
-- The estimates of several studies are uncertain because they
change significantly when different assumptions regarding
taxpayer behavior are used.
-- The studies all used measures of the tax incentive that did not
incorporate important interactions with other features of the
tax code, such as the rules for allocating research expenses
between foreign sources and the United States.
Given the uncertainties relating to (1) the responsiveness of
research spending to tax incentives and (2) the current credit's
effect on incentives and revenue cost, we were unable to conclude
from the recent studies that they provide adequate evidence that a
dollar of research tax credit would stimulate a dollar of additional
research spending and, in the long run, would lead to about two
dollars of research spending. The stimulative effect of the credit
could be significantly smaller or significantly greater.
Moreover, to fully assess the value to society of the research tax
credit, one would need to look at more than just the amount of
spending stimulated per dollar of revenue cost. A comparison would
have to be made between (1) the total benefits gained by society from
research stimulated by the credit and (2) the estimated costs to
society resulting from the collection of taxes required to fund the
credit. The social benefits of the research conducted by individual
companies include any new products, productivity increases, or cost
reductions that benefit other companies and consumers throughout the
economy. Although most economists agree that research spending can
generate social benefits, the effects of the research on other
companies and consumers are difficult to measure. We are not aware
of any studies that have empirically estimated the credit's net
benefit to society.
--------------------
\8 Section 6103 of the Internal Revenue Code requires that tax return
information be kept confidential and prohibits disclosure of this
information, with limited exceptions. The exceptions include
disclosure to the chairmen of certain committees of Congress and
their agents, and disclosure to certain federal officers and
employees for purposes of tax administration. The authors of the
studies we reviewed did not qualify under section 6103 for access to
tax return data.
OBJECTIVES, SCOPE, AND
METHODOLOGY
------------------------------------------------------------ Letter :3
Our first objective was to evaluate recent studies of the research
tax credit for the adequacy of the data and methods used to determine
the amount of research spending stimulated per dollar of revenue
cost. In particular, we were to determine if recent studies provided
adequate evidence to conclude that each dollar of tax credit
stimulates at least one dollar of research spending in the short run
and, over the long run, stimulates about two dollars of research
spending. Our second objective was to identify the factors other
than spending per dollar of revenue cost that determine the credit's
value to society.
To meet our first objective, we reviewed the six studies cited by the
KPMG Peat Marwick report and two studies that the report did not cite
that we identified from our review of the literature on the credit
and from our interviews with authors of research tax credit studies.
In general, these recent studies were published since our 1989
report, although one study cited by KPMG Peat Marwick was published
in 1987. The studies are listed in appendix I.
We used standard statistical and economic principles in our review
and evaluation of the studies of the research tax credit. We relied
upon internal economists to carry out this evaluation. In our
evaluation, we considered such factors as the adequacy of the data
used to estimate the effect of the credit, the adequacy of the
variables used to measure the incentive provided by the credit, and
the sensitivity of the estimates to assumptions about taxpayer
behavior.
We also interviewed the authors of the studies of the research tax
credit and requested comments on a draft of our evaluation of their
studies. We received comments from the authors of six of the eight
studies that we reviewed. All agreed that our report accurately
summarized their studies. However, not all agreed with the
importance of the data and methodological limitations that we
identified in their work. A summary of their comments appears on
pages 12 and 13. We also requested comments on a draft of our report
from the authors of the KPMG Peat Marwick report. They stated that
they appreciated the opportunity to comment on our report but that
after reviewing our report, they had no comments to submit.
To meet the second objective, we reviewed academic articles and
government studies about the determinants of the social benefits of
research spending. We also reviewed studies that describe the
difficulties encountered when attempting to measure the full social
costs and benefits of research.
We did our work in Washington, D.C., from December 1995 through
January 1996 in accordance with generally accepted government
auditing standards.
RECENT STUDIES PROVIDE MIXED
EVIDENCE OF THE CREDIT'S
EFFECTIVENESS
------------------------------------------------------------ Letter :4
The recent studies that we reviewed provided mixed evidence on the
amount of spending stimulated per dollar of revenue cost. Of the
eight studies we reviewed, three supported the claim that one dollar
of credit stimulated about two dollars of additional research
spending. Another study, which did not directly evaluate the
research tax credit, reported estimates of the responsiveness of
research spending to other tax incentives. These estimates appear to
be consistent with the claim that the credit stimulates spending that
exceeds its revenue cost.
However, two studies reported that the credit stimulated spending
that was less than its revenue cost, and another two of the studies
reported estimates of additional spending that do not appear to
support the claim that spending exceeded revenue cost. One of these
latter studies does not compare additional research spending to
revenue cost but does report an estimate of additional spending that
is likely to be less than the revenue cost. The other study reported
that additional spending exceeded revenue cost through 1985 but
reported estimates of additional spending that were likely to be less
than the revenue cost after 1985.
Most of the recent studies used more sophisticated methods than prior
studies when analyzing the effectiveness of the credit. For example,
the studies improved on prior studies by using methods that attempt
to distinguish the credit's effect from other influences on research
spending like the potential size of the market for the product of the
research. However, the studies have the following data and
methodological limitations.
LIMITATIONS OF THE DATA USED
BY THE STUDIES
---------------------------------------------------------- Letter :4.1
The most appropriate data for assessing the effect of the credit are
tax return data. These confidential tax return data were not
available to the authors of the studies. Instead, they used publicly
available data sources, chiefly the COMPUSTAT data service,\9 which
do not accurately reflect the incentive provided by the credit. This
incentive depends on a company's ability to earn credits by having
qualified research spending that exceeds the base amount and on a
company's ability to claim its credits by having sufficient taxable
income. We concluded from our own comparison of tax return data with
COMPUSTAT data and from studies by other researchers that differences
in the measurement of research spending and taxable income make
COMPUSTAT an unreliable proxy for tax return data when analyzing the
credit. Because studies that use the public data cannot accurately
determine the credit's incentive, they may not accurately measure the
amount of spending stimulated by the credit.
--------------------
\9 COMPUSTAT provides financial information on publicly traded
companies drawn from such sources as the companies' filings with the
Securities and Exchange Commission and their annual and quarterly
reports.
LIMITATIONS OF THE STUDIES'
METHODS FOR MEASURING TAX
INCENTIVES
---------------------------------------------------------- Letter :4.2
Three studies that analyzed the credit at the industry level may not
accurately measure the credit's incentive. Analysis at the industry
level of aggregation does not reflect the different incentives the
companies face and their different responses to these incentives.
Industries include firms that earn no credit because their spending
is less than the base amount or claim no credit because they have no
tax liability. An analysis at the industry level that assigns the
same incentive to all these firms would not capture these differences
and is not likely to produce very precise measures of the credit's
effect on research spending.
The eight studies all used measures of the tax incentive that did not
incorporate important interactions with other features of the tax
code. For example, studies that measured the tax incentives by
reductions in the cost of research and development due to tax policy
changes did not include all the research and development provisions
of the tax code. In addition to the credit, the cost of R&D depends
on other tax code provisions like those governing the allocation of
research expenses between foreign sources and the United States. The
studies included some of these provisions but not others. Including
all relevant provisions of the code may change the estimates of the
research credit's effectiveness.
The estimates in several of the studies were highly sensitive to
assumptions made about the data and taxpayer behavior. For example,
one study's estimate of the responsiveness of spending to tax
incentives was reduced by half when more firms were included in the
sample studied or the assumptions were changed on how taxpayers
allocate research and development expenses between domestic and
foreign sources. Other studies that differed in terms of how they
measured the tax incentive produced significantly different estimates
of the spending stimulated by the credit. This sensitivity to the
assumptions made by the authors leads us to conclude that much
uncertainty remains about the effect of the credit on research
spending.
THE AMOUNT OF SPENDING
STIMULATED PER DOLLAR OF
REVENUE COST DEPENDS ON THE
DESIGN OF THE CREDIT
---------------------------------------------------------- Letter :4.3
The estimates presented in the most recent studies do not provide all
the information needed to evaluate the effectiveness of the latest
version of the credit. The amount of spending stimulated per dollar
of revenue cost depends on how the design of the credit affects the
incentive to increase research spending and on how the design affects
the revenue cost. Only one of the recent studies estimated the
effectiveness of the credit for years after its redesign in 1989, and
the author of that study is not confident of her results for the
post-1989 period.
Some reviewers have implied that the recent studies' estimates of the
responsiveness of research spending to price reductions--the price
elasticity of spending--are equivalent to the amount of research
spending stimulated by the credit per dollar of revenue cost.\10 They
said that using an empirical estimate that a 1-percent reduction in
the price of R&D will lead to a 1-percent increase in research
spending implies that one dollar of credit will lead to one dollar of
additional spending. However, these may not be equivalent estimates
because the amount of research spending stimulated by the credit per
dollar of revenue depends on the design of the credit as well as the
responsiveness of spending to price reductions.
For example, the credit's effect on spending and revenue cost will
depend on whether it is designed as a flat credit, which applies to
total research spending, or as an incremental credit, which applies
only to spending that exceeds a base amount. For the same
responsiveness of spending to price reductions, a flat credit with a
10 percent rate should stimulate roughly the same amount of spending
as an incremental credit with the same rate because both credits
provide the same 10 percent effective reduction in the price of
research.\11 However, the flat credit would allow a company to earn a
credit equal to 10 percent of its total qualified research spending,
while the incremental credit would give the company a credit equal
only to 10 percent of the difference between its current qualified
spending and some base spending amount. Consequently, the 10 percent
flat credit would have a higher revenue cost and, therefore, a lower
bang-per-buck than the 10 percent incremental credit.
Incremental credits that differ from one another in terms of how base
spending is defined can also differ substantially in terms of how
much spending they stimulate per dollar of revenue cost. The
bang-per-buck of the current incremental credit may be significantly
different from that of the credit that existed prior to 1990. As we
reported in our May 1995 testimony, the redesign of the credit in
1989 should have increased the size of the incentive provided per
dollar of revenue cost. However, as we also reported in our
testimony, there is evidence that the incentive provided by the
redesigned credit had eroded over time and that the revenue cost of
the additional spending stimulated by the credit had increased.\12
--------------------
\10 See, for example, KPMG Peat Marwick, Extending the R&E Tax
Credit: The Importance of Permanence, Nov. 1994, pp. 24-25, and
The Effectiveness of Research and Experimentation Tax Credits, Office
of Technology Assessment, OTA-BP-ITC-174, Sept. 1995, p. 27.
\11 Besides affecting research spending by reducing its price, the
credit may also affect spending for some types of companies by
increasing their cash flow. A number of recent studies found that
cash flow has a significant effect on the investment of companies
that find external funds more expensive than internal funds. One
recent study found that cash flow is as important a determinant of
R&D investment as it is of ordinary investment. See Bronwyn Hall,
"Investment and R&D at the Firm Level: Does the Source of Financing
Matter?," National Bureau of Economic Research Working Paper No.
4096, Cambridge, Massachusetts, 1992. However, we are not aware of
any studies that have empirically estimated the size of this
cash-flow effect for the research tax credit.
\12 See GAO/T-GGD-95-161.
FACTORS DETERMINING THE
RESEARCH TAX CREDIT'S VALUE TO
SOCIETY
------------------------------------------------------------ Letter :5
The value of the research tax credit to society cannot be determined
simply by comparing the amount of research spending stimulated by the
credit versus the credit's revenue cost. To fully evaluate the
credit's effect, one would have to (1) estimate the total benefits
gained by society from the research stimulated by the credit; (2)
estimate the resource costs of doing the research; (3) estimate the
administration, compliance, and efficiency costs to society resulting
from the collection of taxes (or the borrowing of money) required to
fund the credit; and (4) compare the benefits to the costs.\13 Simply
knowing how much additional research spending the credit stimulates
does not tell you the value of that research to society. Similarly,
the amount of revenue needed to fund the credit does not tell you the
total cost to society of the credit.
There is a general consensus among economists that research is one of
the areas where some government intervention in the marketplace may
improve economic efficiency. From society's point of view,
individual companies may invest too little in research if the return
on their investment is less than the full benefit that society
derives from the research. If the research leads to new products,
reduces costs or increases productivity for other companies and
consumers throughout the economy, the benefits to society may exceed
the return on investment of the companies that conduct the research.
Therefore, companies may not do as much research as society finds
desirable, and government policy to encourage research may be viewed
as appropriate.
However, as the Joint Committee on Taxation and OTA have noted, it is
also possible to decrease economic efficiency by encouraging too much
spending on research.\14 Because not all research generates social
benefits that exceed the returns to companies conducting the
research, encouraging more research may not be economically
efficient. It would be very difficult to determine, given the
difficulty of measuring the social benefit, whether the research tax
credit increases or decreases economic efficiency. No one that we
are aware of, including the authors cited by KPMG Peat Marwick, has
undertaken a study that could answer that question conclusively.
--------------------
\13 Taxes generally distort the relative prices of resources in the
economy and, therefore, lead to a less efficient allocation of
resources. There is no consensus on the size of the efficiency cost
imposed by the U.S. tax system; however, most analysts agree that it
is significant. In its official guidance for benefit-cost analysis
of federal programs, the Office of Management and Budget recommends
that supplementary analyses be prepared in which costs in the form of
public expenditures are multiplied by 1.25 to reflect the efficiency
costs of collecting the taxes to finance those expenditures. See
Office of Management and Budget, "Guidelines and Discount Rates for
Benefit-Cost Analysis of Federal Programs," Circular No. A-94,
revised Oct. 29, 1992, p. 14.
\14 Joint Committee on Taxation, Description and Analysis of Certain
Tax Provisions Expiring in 1994 and 1995 (JCS-8-95), May 8, 1995, p.
38. Office of Technology Assessment, The Effectiveness of the
Research and Experimentation Tax Credits, OTA-BP-ITC-174, Sept.
1995, pp. 21-22.
AUTHORS' COMMENTS ON OUR REVIEW
OF THEIR STUDIES
------------------------------------------------------------ Letter :6
As previously discussed, we requested comments from the authors of
the KPMG Peat Marwick report. After reviewing a draft of our report,
they stated that they had no comments to submit. We also requested
that the authors of the eight studies of the research tax credit that
we reviewed provide comments on our evaluation of their studies. The
following summarizes the comments of the six authors who responded to
our request.
All of the authors we interviewed agreed that the publicly available
data contain measurement errors that may affect their estimates of
the credit's effectiveness. However, two of the authors said that
they believed that their estimates would not change significantly if
tax return data were used. They said that either the data problem
was minor or that statistical methods used to correct the measurement
error reasonably addressed the problem. Two authors also said that
they believed that their elasticity estimates would not change
significantly but noted that predicting what would happen to the
estimates when better data are used is difficult. Two authors agreed
with our assessment of the importance of the potential inaccuracies
from using COMPUSTAT data.
As explained more fully in appendix I, we have concluded that
COMPUSTAT data are not a suitable proxy for tax return data when
analyzing the credit. Although the authors agree that COMPUSTAT data
are not the best data, they disagree among themselves about the
importance of this issue. We acknowledge that statistical methods
can be used to help address this issue of measurement error, but the
success of these methods is difficult to assess. We conclude that,
because the most appropriate data were not used in these studies,
uncertainty remains about the responsiveness of spending to the
credit.
The methodological limitations that we identified were not addressed
in the comments of all the authors because they were not relevant to
every study. The authors who did comment disagreed about the
importance of the methodological limitations. One author who
addressed the importance of correctly incorporating the features of
the tax code said she believed that some of the studies' estimates of
the effect of the credit were overestimated because the method of
estimation excluded tax preferences available for investments other
than research. Another author commented that the sensitivity of the
estimates to assumptions about taxpayer expectations accounted for
the difference in estimates across the studies. However, two authors
who agreed with our identification of the methodological limitations
in their work did not believe that the limitations had a significant
effect on their estimates. Finally, the authors who commented agreed
that analyzing the credit at the firm level rather than at the
industry level produces more accurate estimates. However, one author
said that he did not believe that his industry level estimates would
change significantly if they were based on analysis at the firm
level.
As explained in appendix I, we found that estimates reported in the
studies varied significantly when authors employed different
assumptions about the data and taxpayer behavior. This sensitivity
of the estimates to authors' assumptions leads us to conclude that
much uncertainty remains about the effect of the credit on research
spending.
---------------------------------------------------------- Letter :6.1
We are sending copies of this report to pertinent congressional
committees, the Secretary of the Treasury, KPMG Peat Marwick, the
individual authors, and other interested parties. Copies will be
made available to others upon request.
The major contributors to this report are listed in appendix II. If
you have any questions, please call me on (202) 512-9044.
Sincerely yours,
Natwar M. Gandhi
Associate Director, Tax Policy
and Administration Issues
EVALUATION OF STUDIES OF THE
EFFECTIVENESS OF THE RESEARCH TAX
CREDIT
=========================================================== Appendix I
We classified the studies of the effectiveness of the research tax
credit according to the level of aggregation at which the data are
analyzed and the method used to measure the incentive provided by the
credit. The studies analyze the credit using firm level data or
using data aggregated to the industry level. The incentive provided
by the credit is measured by a categorical or "dummy" variable, or by
a variable measuring the "tax price" of research and development
(R&D). The categorical variable measures the change in R&D spending
due to the presence or absence of the tax credit or to the ability of
firms to use the credit, while the tax price variable measures the
change in spending due to the effect of tax policy on the cost of
R&D.
We reviewed the six studies cited by KPMG Peat Marwick in their
report.\15 We also reviewed two recent studies of the credit's
effectiveness that were not cited by KPMG Peat Marwick.\16 The
following summarizes the studies and presents our evaluation of them.
--------------------
\15 Baily, Martin Neil and Lawrence, Robert Z., "Tax Incentives for
R&D: What Do the Data Tell Us?" Study commissioned by the Council on
Research and Technology, Washington, D.C., January 1992; Baily,
Martin Neil and Lawrence, Robert Z., "Tax Policies for Innovation and
Competitiveness." Study commissioned by the Council on Research and
Technology, Washington, D.C., April 1987. This study is also
summarized in Martin Neil Baily and Alok K. Chakrabarti, Innovation
and the Productivity Crisis, (Washington D.C., The Brookings
Institution, 1988), pp. 123-29; Hines, James R., "On the Sensitivity
of R&D to Delicate Tax Changes: The Behavior of U.S. Multinationals
in the 1980s," from Studies in International Taxation, Alberto
Giovannini, R. Glenn Hubbard, and Joel Slemrod, eds., The University
of Chicago Press, 1993; Hall, Bronwyn H., "R&D Tax Policy During the
Eighties: Success or Failure?," NBER Conference Report, Washington,
D.C., November 17, 1992. Also printed in Tax Policy and the Economy,
Vol. 7, ed. by James Poterba, National Bureau of Economic Research,
1993; Berger, Philip G., "Explicit and Implicit Tax Effects of the
R&D Tax Credit," Journal of Accounting Research, Vol. 31, No. 2,
Autumn 1993, pp. 131-71; Swenson, C.W., "Some Tests of the Incentive
Effects of the Research and Experimentation Tax Credit," Journal of
Public Economics, Vol. 49, 1992, pp. 203-18.
\16 Mamuneas, Theofanis P. and Nadiri, M. Ishaq, "Public R&D
Policies and Cost Behavior of the U.S. Manufacturing Industries,"
Working Paper No. 5059, National Bureau of Economic Research,
Cambridge, Massachusetts, March 1995; Tillinger, Janet W., "An
Analysis of the Effectiveness of the Research and Experimentation Tax
Credit in a q Model of Valuation," The Journal of the American
Taxation Association, Fall 1991, pp. 1-29.
DESCRIPTIONS OF THE STUDIES
--------------------------------------------------------- Appendix I:1
INDUSTRY LEVEL STUDIES
------------------------------------------------------- Appendix I:1.1
Martin Neil Baily and Robert Z. Lawrence use National Science
Foundation (NSF) data to examine the effect of the credit for 1981
through 1985 in their 1987 study, and for 1981 through 1989 in their
1992 study. The 1987 study analyses the credit using a dummy
variable that indicates the years in which the credit was in effect,
while the 1992 study uses a variable that reflects changes in the
credit's incentive due to changes in the tax law. Both studies
produce essentially the same finding: the percentage increase in R&D
spending in response to each percentage decrease in the price of
R&D--the price elasticity of R&D--is approximately equal to one.
Using this elasticity, Baily and Lawrence estimate that the credit
generated about two dollars of R&D for each dollar of tax revenue
forgone.
Theofanis P. Mamuneas and M. Ishaq Nadiri use industry level data
for 1956 through 1988, chiefly drawn from the Bureau of Labor
Statistics and NSF. Their method is to construct a rental price
variable for R&D capital that reflects the research tax credit and
the provisions for the immediate expensing of research expenditures.
To construct this variable, the authors acknowledge that they assume
that the firms in their industries have sufficient tax liability to
claim the credit, that their spending exceeds the base amount, and
that spending is less than twice the base amount. Their estimates of
price elasticities range from -1.0 for the three aggregate industries
of textiles and apparel; lumber, wood products, and furniture; and
other manufacturing to -0.94 for scientific instruments. On the
basis of these elasticities, they calculate that the average
additional research spending stimulated per dollar of revenue cost
was about 95 cents for the period 1981 to 1988.
FIRM LEVEL STUDIES
------------------------------------------------------- Appendix I:1.2
STUDIES USING TAX PRICE
VARIABLES
----------------------------------------------------- Appendix I:1.2.1
James R. Hines' study uses firm level data from COMPUSTAT for 1984
through 1989. His method is to construct a tax price variable that
measures how the costs of R&D are affected by the rules for
allocating R&D expenses between U.S. and foreign sources under
section 1.861-8 of U.S. Treasury regulations.\17 His tax price does
not include the research tax credit or other R&D related features of
the tax code. Hines' preferred estimates of the R&D price elasticity
range from -1.2 to -1.6. However, when he increases his sample size
to include firms previously excluded due to merger activity, these
elasticity estimates drop to a range of -0.5 to -0.6. Also, the
elasticities decrease to -0.5 to -0.9 when Hines changes his
assumptions about how firms allocate their research expenses. Hines
does not apply these elasticities to the credit or calculate how much
spending is induced by the credit.
Bronwyn H. Hall uses firm level data from COMPUSTAT for 1977 through
1991. Her tax price variable measures how the research tax credit
and expensing provisions affect the cost of R&D. Hall estimates a
short-run price elasticity of R&D of -1.5 and a long-run price
elasticity of -2.7. However, she advises that the long run
elasticity be viewed with caution, as it is likely to be "quite
imprecise." Hall estimates that the additional spending induced by
the credit in the short run was $2 billion per year, while the tax
revenue cost was about $1 billion per year.
--------------------
\17 The section 1.861-8 regulations issued in 1977 require that
research expenses are to be allocated using either a sales or gross
income method. Under the sales method, 30 percent of expenses are to
be allocated to U.S. sources, with the remaining 70 percent
apportioned between foreign and domestic sources using either sales
or income as the method of apportionment. Under the gross income
method, research expenses are to be apportioned according to the
relative amounts of gross income from U.S. and foreign sources.
Since 1981, these allocation rules have been subject to repeated,
temporary revisions that have allowed companies to allocate less
research expense to foreign sources.
STUDIES USING DUMMY
VARIABLES
----------------------------------------------------- Appendix I:1.2.2
Philip Berger's study uses firm level data from COMPUSTAT for 1975
through 1989. He measures the effect of the credit using a dummy
variable that indicates the years in which a firm is able to use the
credit, i.e., the firm has a positive tax liability in the current or
preceding 3 years. Berger uses the results of this analysis to
estimate that the credit induced $2.70 billion of additional spending
per year from 1982 through 1985. He compares this yearly increase to
a yearly revenue cost of $1.55 billion to conclude that additional
spending per dollar of forgone revenue was $1.74 during 1982 through
1985. Although Berger does not calculate the amount of spending per
dollar of forgone revenue for years after 1985, his study shows that
the credit was less effective in later years.\18
C. W. Swenson's study uses firm level data from COMPUSTAT for 1975
through 1988. He also uses a dummy variable that indicates the years
in which a firm is able to use the credit. However, the ability to
use the credit in his study depends not only on current tax status
but also on future tax status and the firms' planned R&D spending.
Swenson estimates that total additional spending induced by the
credit was $2.08 billion during 1981 through 1985. Swenson does not
compare this estimate to the revenue cost.
Janet W. Tillinger's study uses firm level data drawn chiefly from
COMPUSTAT for 1980 through 1985. She measures the effect of the
credit using a dummy variable that indicates the years in which firms
have research spending that exceeds the base amount. Tillinger uses
the results of this analysis to estimate that the credit induced
about 19 cents of increased spending per dollar of forgone revenue
for 1981 through 1985, which she notes is at the lower end of the
estimates from our 1989 study. Tillinger also finds that the
effectiveness of the credit varies by the type of firm. When the
firms are classified according to the opportunity costs of
alternatives to R&D investment like the payment of dividends, she
finds that the additional spending ranges from 8 cents to 42 cents
per dollar of forgone revenue.
The studies reviewed above provide mixed evidence for claims about
the amount of spending induced by the credit per dollar of forgone
revenue. Of the six studies cited by KPMG Peat Marwick, three
studies (the two by Baily and Lawrence, and Hall's study) support the
claim that each dollar of tax revenue stimulated about two dollars of
additional research spending. Hines' study reports a price
elasticity of research spending that, if applied to the research tax
credit, is likely to be consistent with the finding that additional
spending exceeds the revenue cost.
Two studies cited by KPMG Peat Marwick, however, may not support the
claim that induced research spending exceeds the revenue cost of the
credit. Swenson's study estimates that the credit induced additional
spending of $2.08 billion from 1981 through 1985. He notes that his
estimate is "comparable to . . . GAO estimates of $1 billion to
$2.9 billion for the same period." Swenson states that he does not
calculate a bang-per-buck measure because he does not have access to
the taxpayer data necessary to make this calculation. However,
Swenson states that his estimate of additional spending is not likely
to support the claim that the spending stimulated by the credit
exceeded its revenue cost. Berger's study estimates that additional
spending exceeded revenue cost in the period 1982 through 1985, but
the study may not support this claim in the years after 1985. Berger
does not calculate a bang-per-buck measure for years after 1985.
However, his study does show that the credit was less effective in
these years and that the credit was not a statistically significant
determinant of R&D spending in the years after 1986.
The two studies that were not cited by KPMG Peat Marwick do not
support the claim that induced spending exceeded the revenue cost of
the credit. The Mamuneas and Nadiri study estimates that the credit
stimulated additional spending that was slightly less than the
revenue cost during 1981 through 1988, while the Tillinger study
estimates that additional spending was significantly less than
revenue cost during 1981 through 1985.
--------------------
\18 Berger estimates that the credit increased firms' R&D spending to
sales ratios by an average of 8.5 percent during 1982-85, while
increasing this ratio only 3.8 percent on average during 1986-89.
Furthermore, his study shows that the credit had no statistically
significant effect on R&D spending in the years after 1986.
OUR EVALUATION OF THE STUDIES
--------------------------------------------------------- Appendix I:2
Most of the studies we reviewed use more sophisticated statistical
methods and more years of data than prior studies. For example, most
of the recent studies use methods that attempt to distinguish the
credit from other factors that influence research spending like
market size and the availability of investment funds. Some studies
also include the influence of taxpayers' expectations about factors
like the future tax status of firms when determining the effect of
the credit on current spending. Nevertheless, despite these
advantages over prior studies, these studies have data and
methodological limitations that are significant enough to lead us to
conclude that much uncertainty remains about the true responsiveness
of research spending to tax incentives.
None of the studies use the best data for assessing the effect of the
credit. They all use publicly available COMPUSTAT or NSF data, which
are not the most appropriate data for this purpose. The incentive
provided by the credit depends on companies' ability to earn credits
by having qualified research spending that exceeds the base amount,
and to claim credits by having tax liabilities. Information on
qualified research spending and tax liabilities can be most
accurately determined from confidential IRS data. The publicly
available data will not be as accurate because they use definitions
of research spending and tax liabilities that are different from IRS.
These tax return data were unavailable to these researchers. In her
study, Hall recognizes the limitations of publicly available data and
attempts to correct the errors in her measurements. However, it is
difficult to determine how successful her efforts are without
repeating her analysis using the tax return data. In any case, the
estimates of all the studies that we reviewed would be more reliable
if they were based on IRS data.
The tax price variables and the dummy variables used in the studies
to capture the incentive provided by the credit depend on companies'
ability to earn credits and claim them against their tax liabilities.
COMPUSTAT and NSF data do not accurately reflect credits earned and
claimed.
The ability to earn credits depends on the relationship of qualified
research spending to the base amount. COMPUSTAT and NSF data do not
accurately reflect this relationship because both data sources
include spending that does not qualify for the credit. Most notably,
spending reported by COMPUSTAT includes spending overseas that would
not be qualified research spending. In our 1989 report, we compared
COMPUSTAT data with tax return data and concluded that COMPUSTAT data
are not a suitable proxy for tax return data when analyzing the
credit. For example, when we compared the growth rate of COMPUSTAT
research spending with qualified research spending for a sample of
firms contained in both the COMPUSTAT database and IRS files, we
found that the rates varied considerably over the period 1981 through
1985. Qualified spending grew 1.46 times as fast as COMPUSTAT
spending in the 1980 to 1981 period, but only 0.72 times as fast in
the 1983 to 1984 period. The relationship between spending and the
base using COMPUSTAT may not accurately reflect the relationship
using tax data, and, therefore, both tax price variables and dummy
variables are likely to be inaccurate.
The ability to claim credits depends on the tax status of the firms.
COMPUSTAT contains information on taxable income and loss
carryforwards, but studies have shown that COMPUSTAT does not always
accurately or consistently reflect IRS data.\19 Furthermore,
COMPUSTAT data contain no information on the general business credit,
which limits the ability of companies to claim the credit.\20 Again,
because both the tax price variables and the dummy variables depend
on the ability of firms to claim the credit, we conclude that they
will be measured inaccurately when based on COMPUSTAT data.
The reliability of the Baily and Lawrence studies and the Mamuneas
and Nadiri study is also limited by the level of aggregation at which
the data are analyzed. Their analyses of the credit at the industry
level are unlikely to produce very precise measures of the credit's
effect. Their analyses do not reflect the different incentives that
companies face and the different responses to these incentives.
Industries will include firms that earn no credit because their
spending is less than the base, firms that cannot claim the credit
because they have no tax liability, and firms subject to the
50-percent base limitation. A measure that assigns the same
incentive to all these firms will not capture these differences and
is not likely to yield precise or reliable estimates of the credit's
effect.
The reliability of the studies that we reviewed is also limited by
the methods used to measure the incentive provided by the credit.
The studies use measures of the tax incentive that do not incorporate
important interactions of the research tax credit with other features
of the tax code. For example, Hines studies the effect of the
section 1.861-8 allocation rules on research spending but does not
analyze the effect of other features of the tax code such as the
research tax credit. Hall, on the other hand, analyzes the research
tax credit but does not incorporate the section 1.861-8 allocation
rules in her study. Hall believes that including the rules in her
analysis would not make "an enormous difference" because the firms
subject to the rules probably represent only a small part of her
sample. However, she does say that including the rules would make
her estimates more precise. Hines states that it is "difficult to
know for sure" the effect on his estimates of including interactions
with other features of the code.
The estimates in some of these studies are also uncertain because
they are sensitive to assumptions made about the data and taxpayer
behavior. Hines' estimate of the effect of tax policy on spending is
reduced by half when he includes more firms in his sample or changes
his assumptions about how companies allocate R&D expenses. Hall
notes that estimation at the firm level involving investments like
R&D is difficult and sensitive to assumptions made when specifying
the models. This sensitivity of the results is also illustrated by
the three studies using dummy variables where differences in the
approach to modeling taxpayer behavior and measuring the effect of
the credit yield very different estimates. Although some of the
studies attempt to measure the degree of this sensitivity and correct
for it, the success of these efforts is difficult to assess.
The authors of these studies themselves, in many cases, advise that
their results be used with caution and recognize that their estimates
would benefit from further research. For example, when describing
her estimates of the spending induced by the credit, Hall states that
"it needs to be kept firmly in mind that my tax estimates are not
likely to be as good as those constructed using IRS data." She also
mentions, in the 1992 version of her paper, that her analysis "needs
more investigation for robustness over time and industry." When
discussing the limitations imposed by not including interactions with
other aspects of the tax code, Hines notes that his results should be
used with caution because of these "restrictive assumptions built
into the estimated R&D responses to tax changes."
The current version of the credit has not been studied extensively,
and little is known about the actual incentives provided by the
current credit. The research tax credit was fundamentally
restructured in 1989. Hall's study, which spans the years 1977
through 1991, is the only study we reviewed that covers any tax years
after the credit was changed. However, her data contain only 2
years--1990 and 1991--under the revised credit structure. Hall notes
that her estimate of additional spending for these years amounts to
about 10 percent of the total R&D and that this "amount is almost too
large to be credible . . . and deserves further investigation as
more data become available." She indicated that her estimates of
additional spending may be less reliable because she did not have
data on the tax status of firms after 1991 that were needed to
measure the incentive provided by the revised credit.
--------------------
\19 See Rosanne Altshuler and Alan J. Auerbach, "The Significance of
Tax Law Asymmetries: An Empirical Investigation," Quarterly Journal
of Economics, Vol. 105, 1990, pp. 63-86.
\20 The general business credit combines several tax credits,
including the research tax credit, for the purpose of computing an
overall dollar limitation on the reduction of a company's tax
liability. The general business credit cannot exceed net income tax
minus the greater of (1) the tentative minimum tax or (2) 25 percent
of the net regular tax liability above $25,000.
MAJOR CONTRIBUTORS TO THIS REPORT
========================================================== Appendix II
GENERAL GOVERNMENT DIVISION,
WASHINGTON, D.C.
James Wozny, Assistant Director, Tax Policy and Administration Issues
Kevin Daly, Senior Economist
Anthony Assia, Senior Evaluator
*** End of document. ***