How effective are fiscal incentives for R&D? A …bhhall/papers/HallVanReenan...3 In addition to the features of the tax system targeted toward R&D equipment expenditures at the federal

Ž .Research Policy 29 2000 449–469www.elsevier.nlrlocatereconbase

How effective are fiscal incentives for R&D? A review of theevidence

Bronwyn Hall a,b,c, John Van Reenen c,)

a UC Berkeley, Berkeley, CA, USAb Nuffield College Oxford and NBER, UK

c Institute for Fiscal Studies, UniÕersity College London and CEPR, 7 Ridgmount Street, London, WC1E 7AE, UK

Abstract

This paper surveys the econometric evidence on the effectiveness of fiscal incentives for R&D. We describe the effectsof tax systems in OECD countries on the user cost of R&D — the current position, changes over time and across differentfirms in different countries. We describe and criticize the methodologies used to evaluate the effect of the tax system on

Ž .R&D behaviour and the results from different studies. In the current imperfect state of knowledge we conclude that adollar in tax credit for R&D stimulates a dollar of additional R&D. q 2000 Elsevier Science B.V. All rights reserved.

JEL classification: H32; 031; O38Keywords: Tax credits; R&D; international

1. Introduction

Economists generally agree that the market willfail to provide sufficient quantities of R&D as it hassome characteristics of a public good. But howshould policy bridge the gap between the private andsocial rate of return? A tax-based subsidy seems themarket-oriented response as it leaves the choice ofhow to conduct and pursue R&D programs in thehands of the private sector. There are several draw-backs to this tool, however, compared with govern-ment financing andror conducting the R&D pro-

Ž .gram directly see Klette et al., 2000 . Perhaps theprimary objection is that fiscal incentives are simplyineffective in raising private R&D spending — theresponse elasticity is so low it would take a huge tax

) Corresponding author. E-mail: [email protected]

change to generate the socially desirable level ofspending. This was the conventional wisdom amongeconomists until recently, so it is the key focus ofthis paper. We address the issue of how governmentsŽ .sometimes inadvertently have used the tax systemto promote R&D, how researchers have evaluatedthese effects, and what are the results of their evalua-tions.

There are other objections to the use of the taxsystem to which we will be paying less attention.First, the projects that should be promoted from asocial view are those with the largest gaps betweenthe social and private return. Yet private sector firmswill use any credits to first fund R&D projects withthe highest private rates of return. In principle thestate could do a lot better by targeting the projectswith the highest spillover gap. In practice this maybevery hard to deliver because of the intrinsic uncer-tainty of knowledge creation and because of the

0048-7333r00r$ - see front matter q 2000 Elsevier Science B.V. All rights reserved.Ž .PII: S0048-7333 99 00085-2

( )B. Hall, J. Van ReenenrResearch Policy 29 2000 449–469450

tendency of states to reward lobbyists and bureau-crats rather than take the optimal decisions. 1 In theface of pervasive government failure to implementthe optimal subsidy policy, tax credits appear moreattractive.

Using the tax system to stimulate R&D is farfrom the ultimate panacea for failures in the marketfor knowledge. Implementation in the existing politi-cal and tax environment has meant that there arefrequent changes in the fiscal incentives faced byfirms that affect the costs of performing R&D indifferent ways for different companies at differenttimes. This heterogeneity is a burden for companiesand policy makers but is a boon for social scientists.A long-standing problem in the investment literatureis the intractability of finding exogenous variation inthe user cost of capital. The heterogeneity acrossfirms and time in the cost of capital for this type ofinvestment has the potential to help identify parame-ters of the underlying R&D investment demandequation. The frequent changes of government policyoffer a rare opportunity to generate some exogenous

Ž .movement in the price of R&D even across firmsthat could be used to identify a key part of theneoclassical model. What’s bad for the economymay be good for the econometricians!

The paper is structured as follows. In Section 2we examine the tax treatment of R&D in an interna-tional context and introduce the major issues. InSection 3 we critically outline the methodologiesresearchers have used to examine the effects of taxincentives on R&D. In Section 4 we present thesurvey of results and in Section 5 we offer someconcluding comments.

2. The tax treatment of R&D across countries

2.1. The current position

The treatment of R&D by the tax system variousextensively between countries and over time. Table

1 Ž .On this point, see Cohen and Noll 1991 for discussion of theissue and a series of examples drawn from the U.S. experience ofthe past 30 years. They demonstrate that large federal R&Dprojects have frequently been continued well past the point whereexpected costs exceeded expected benefits due to the existence ofstakeholders that had legislative influence.

1, which is drawn from many sources, summarizesthe position in approximately 1995 to the best of ourknowledge. 2 The second column of the table at-tempts to give the definition of R&D that is used forthe purpose of the tax credit, which is often some-what more restrictive than the Frascati manualŽ .OECD, 1980 definition, but not always. The nexttwo columns give the rates at which non-capitalR&D and capital R&D are depreciated for taxpurposes. One hundred percent means that the quan-tity is expensed. In most cases it is also possible toelect to amortize R&D expenditure. This might con-ceivably be an attractive option if operating loss

Žcarryforwards are not available to use the R&Dexpense as a deduction even if no current tax is

.owed , but in most cases tax losses can be carriedŽ .forward and back see column 7 .

Given that R&D capital expenditure is typicallyonly 10–13% of business R&D, and that the busi-

Žness R&D–GDP ratio is typically 1–2% OECD,.1984 , implying an R&D capital equipment–GDP

ratio of 0.1–0.2%, a remarkable amount of time hasbeen spent in many of these countries tinkering withthe expensing and depreciation rules for capitalequipment used in R&D activities. 3 Although al-

Ž .most all countries except for the UK treat this kindof capital expenditure somewhat like ordinary invest-ment, many have used complex speeded-up deprecia-tion schemes at one time or another to give a boostto a R&D capital equipment investment; this can beoften be justified by the simple fact that the eco-nomic life of this kind of specialized equipment islikely to be shorter than that for other types ofcapital. Frequently the depreciation involved is also

2 Ž .Sources include Asmussen and Berriot 1993 , AustralianŽ . Ž .Bureau of Industry Economics 1993 , Bell 1995 , Bloom et al.

Ž . Ž . Ž .1998, 2000 , Griffith et al. 1995 , Harhoff 1994 , HiramatsuŽ . Ž . Ž .1995 , Leyden and Link 1993 , McFetridge and Warda 1983 ,

Ž . Ž . Ž .Seyvet 1995 , Warda 1994 , and KPMG 1995 .3 In addition to the features of the tax system targeted toward

R&D equipment expenditures at the federal level in many coun-tries, in many U.S. states there is a special sales tax provisionwhich exempts firms from paying sales tax on purchases orrepairs of this kind of equipment. This amounts to an additionaltax credit of about 4–8% in the states that have this provision.

( )B. Hall, J. Van ReenenrResearch Policy 29 2000 449–469 451

subject to the R&D tax credit. Normally buildings orplant for use by an R&D laboratory do not partici-pate in these schemes.

Columns 5 and 6 characterize the tax credit, ifthere is one. The rate and the base above which therate applies are shown; when the base is zero, thecredit is not incremental, but applies to all qualifyingR&D expense. At the present time, it appears thatonly France, Japan, Korea, Spain, the US, and Tai-wan have a true incremental R&D tax credit, andthey each use a slightly different formula for thebase. Canada has a non-incremental credit and Brazilhas a non-incremental credit that is restricted tocomputer industry research. Column 9 shows thatmany countries also have provisions that speciallyfavor R&D in small and medium-sized companies.In France, for example, this takes the form of aceiling on the credit allowed that is equal to 40

Žmillion francs in 1991–1993 approximately US$6.7.million . The effect is to tilt the credit toward smaller

firms, whereas direct R&D subsidies in France go toŽ .large firms to a great extent Seyvet, 1995 . An

exception to this rule is Australia, which has aminimum size of research program to which the taxpreference of 150% expensing applies: US$20,000.This seems to be related more to the administrativecost of handling the R&D tax concession than to

Žany policy decision Australian Bureau of Industry.Economics, 1993; Bell, 1995 .

Columns 10 and 11 give any differences in taxtreatment that apply to R&D done abroad by domes-tic firms or R&D done in the country by foreign-owned firms. For the first type of R&D, any special

Ž .incentives beyond 100% deductibility will typicallynot apply, except that up to 10% of the project costfor Australian-owned firms can be incurred outsideAustralia. For the second type of R&D, it is fre-quently difficult to tell from the summarized taxregulations. In Korea and Australia, foreign firms donot participate in any of the incentive programs. Inthe US and Canada, they are treated like domesticfirms, except that they do not receive an R&D grantin Canada when their tax liability is negative.

Column 8 details whether the incremental taxcredit is treated as taxable income, that is, whetherthe expensing deduction for R&D is reduced by theamount of the tax credit. Whether or not this is truetypically has a major effect on the marginal incentive

faced by a tax-paying firm, but it is somewhat hardto ascertain in many cases whether this feature ap-plies.

2.2. Changes oÕer time

Reforms of systems of taxing corporate incomeover the past decade have tended towards loweringstatutory rates and broadening the tax base. What hashappened to the tax treatment of R&D over thattime period? This section documents some of themain changes in the tax treatment of R&D in eight

Žcountries over the period 1979 to 1994 see Bloom etŽ . .al. 1998, 2000 for more details . It is worth noting

that the cost of R&D figures reported in this sectionare calculated assuming that the R&D investmentqualifies for any credit, that the amount of credit isnot constrained by any capping rules and that thefirm has sufficient tax liability against which tooffset the credit. In the next section we investigatehow the various credits affect firms in differentpositions.

The following assumptions are made concerningthe type of R&D investment to be analysed. Weconsider a domestic investment, financed from re-tained earnings, in the manufacturing sector anddivided into three types of asset for use in R&D —current expenditure, buildings, and plant and ma-chinery. An important assumption in the modellingstrategy used here is that current expenditure onR&D is treated as an investment — that is, its fullvalue is not realised immediately but accrues overseveral years. Current expenditure on R&D is as-sumed to depreciate at 30% a year, buildings at3.61% and plant and machinery at 12.64%.

Fig. 1 shows how the tax treatment of R&D haschanged over time. This graph shows the tax compo-nent of the user cost of R&D for a typical R&D 4

investment in Australia, Canada, France and the US.These are the four countries that had the most gener-

4 ‘‘Typical’’ means a domestic investment financed from re-tained earnings for a firm which is not tax exhausted or hittingany maximum tax credit caps.


Tab

le1

The

tax

trea

tmen

tof

R&

Dar

ound

the

wor

ld

Ž.

Ž.

Ž.

Ž.

Ž.

Ž.

Ž.

Ž.

Ž.

Ž.

Ž.

12

34

56

78

910

11

Cou

ntry

R&

DR

&D

capi

tal

Bas

efo

rC

arry

back

Spe

cial

For

eign

R&

DR

&D

byŽ

Ž.

date

Def

init

ion

ofR

&D

depr

ecia

-de

prec

iati

onT

axcr

edit

incr

emen

tal

CB

and

Cre

dit

trea

tmen

tfo

rby

dom

esti

cfo

reig

n.

enac

ted

for

tax

cred

itti

onra

tera

tera

teta

xcr

edit

carr

yfor

war

dta

xabl

e?S

ME

sfi

rms

film

sŽ

.C

F

Can

ada

Fra

scat

i,ex

cl.

soc

sci.

100%

100%

or20

%20

%0

3yr

CB

,ye

s40

%to

Rs

C$2

00K

expe

nse

20%

only

?Ž

.19

60s

mar

keti

ng,

rout

ine

DB

,20

%IT

C,

10yr

CF

gran

tif

nota

xli

ab.,

noIT

C,

etc.

test

ing,

etc.

not

buil

ding

s35

%ca

peq

ITC

to$2

Mw

Ž.

Fra

nce

Fra

scat

i,in

cl.

pate

ntde

p.10

0%3-

yrS

L50

%R

y1

5-yr

CF

,no

yes

noac

cel

dep

?Ž

.Ž

.Ž

.xŽ

.19

83co

ntra

ctR

,ex

cl.

offi

ceor

5yr

cap.

not

buil

ding

sq

Ry

2r

25-

yrfo

rO

L,

reca

ptur

edT

C-

50M

FF

unle

ssco

ns.

Ž.

expe

nses

&su

ppor

tac

cele

rate

dre

alT

Cre

fund

edno

cred

itpe

rson

nel

incl

.up

grad

es,

SW

,ov

erhe

adG

erm

any

Fra

scat

i,in

cl.

Dev

elop

-10

0%30

%D

B,

none

NA

1r5

yrN

Aas

sist

ance

via

25%

onm

ent,

impr

ovem

ents

,ca

p.If

acq.

4%S

L—

bldg

s,ca

shgr

antr

ITC

roya

ltie

sso

ftw

are

cash

gran

ts?

Ital

yF

rasc

ati,

incl

.S

oftw

are

100%

acce

lera

ted

none

NA

NA

?ye

s,ce

ilin

gor

5yr

cap.

Japa

nF

rasc

ati,

incl

.de

prec

iati

on10

0%ac

cele

rate

d20

%m

axR

sinc

e5-

yrno

6%R

inst

ead

6%cr

edit

for

20%

onŽ

.Ž

Ž.

1966

ofP

&E

,de

ferr

edch

arge

sor

5yr

cap.

5%T

C—

bldg

sm

axat

10%

1966

usua

lbu

tcr

edit

cap

-Y

100

m,

coop

wit

hro

yalt

ies

.be

nefi

t)1

yr,

incl

.ta

xli

ab.

lim

ited

to10

%6%

for

envi

r.r

fore

ign

labs

Sof

twar

edi

seas

eU

Kno

spec

ial

defi

niti

on;

100%

100%

ifno

neN

A5-

yrC

FN

A25

%on

trea

ted

asan

expe

nse,

‘‘sc

ient

ific

roya

ltie

sho

wev

erre

sear

ch’’

Ž.

US

excl

.co

ntra

ctR

for

doer

,10

0%3-

yr.,

20%

avg

of84

–88

R3r

15yr

yes

R&

Dto

Sal

es3%

not

elig

ible

sam

eas

Ž.

July

1981

rev.

engi

neer

ing,

prod

.15

-yr.

for

bldg

sfo

rst

artu

psdo

mes

tic

impr

ov.,

35%

cont

ract

RA

ustr

alia

Fra

scat

i,ex

cl.

soc

sci,

150%

3-yr

SL

none

NA

3r10

yrN

Ace

ilin

g;re

duce

dup

to10

%of

nosp

ecia

lŽ

.Ž

.Ju

ly19

85so

me

test

ing,

mar

keti

ngno

tbu

ildi

ngs

cred

itfo

rsm

all

proj

ect

cost

prov

isio

nsov

erhe

ad,

soft

war

eR

&D

prog

ram

sin

clin

1995

?A

ustr

iaD

ev.

and

impr

ov.

of10

5%ac

cele

rate

dno

neN

A5-

yrC

FN

Ava

luab

lein

vent

ions

Bel

gium

incl

.S

oftw

are

100%

or3-

yrS

Lno

neN

A5-

yrC

FN

A10

–15

%ad

dl3

yrca

p.20

-yr

—bl

dgs

capi

tal

dedu

ctio

n


Bra

zil

R&

Din

com

pute

rin

d.10

0%li

kein

vest

men

tno

neN

A4-

yrC

F10

0%of

com

p.Ž

.C

hina

PR

CN

Ano

neD

enm

ark

Spe

cial

tech

prog

ram

mes

100%

?10

0%?

?5-

yrC

F?

wit

hE

Cre

sear

cher

sIn

dia

scie

ntif

icre

sear

ch10

0%10

0%no

neN

A?

NA

30–

50%

onro

yalt

ies

orkn

owho

wex

cept

land

Ž.

Irel

and

scie

ntif

icre

sear

ch10

0%10

0%no

tre

late

d,

upto

400%

???

???

TC

ceil

ing

of27

%on

roya

ltie

s;in

cl.

soft

war

e15

%ot

herw

ise

525

000

tax

trea

ties

Kor

eaex

peri

men

tal

and

100%

18–

20%

depr

ec,

10%

0av

gof

last

2yr

s?no

yes;

spec

ial

10–

16%

onno

spec

ial

rese

arch

expe

ndit

ure

5.6%

—bl

dgs

25%

rule

sfo

rst

artu

psro

yalt

ies

prov

isio

nsM

exic

o10

0%3-

yrS

L,

none

NA

?N

A20

-yr

—bl

dgs

Net

herl

ands

Wag

esof

R&

Dle

adin

gto

100%

like

inve

stm

ent

12.5

–25

%0

8-yr

CF

noye

s;ce

ilin

gon

ITC

nota

xon

roya

ltie

sŽ

.Ž

.19

94pr

od.

dev.

not

serv

ices

or5

yrca

p.m

axon

R&

Dw

ages

Nor

way

prod

.de

v.,

capi

tali

zed

100%

like

inve

stm

ent

none

NA

10-y

rC

FN

Ano

tax

onro

yalt

ies

Ž.

know

how

cap

ifpr

od.

res.

rese

rve

Por

tuga

lus

ual

100%

orno

neN

A?

NA

does

not

0–

27%

on3

yrca

p.ap

ply

roya

ltie

sS

inga

pore

excl

.so

c.sc

i.,

qual

ity

cap.

exce

ptde

prec

.as

addl

dedu

ctio

nN

A?

NA

yes

Ž.

cont

rol,

soft

war

eso

me

R&

Dus

ual

200%

Sou

thA

fric

asc

ient

ific

rese

arch

100%

for

25%

dep

none

NA

?N

Ade

velo

pmen

tof

tech

.R

for

cap

cap.

for

DS

pain

excl

.ro

utin

epr

od.

amor

tize

100%

15%

r30

%,

avg

ofla

st2

yrs

5-yr

CF

-OL

,N

A5

–25

%on

roya

ltie

sŽ

.im

prov

e.in

cl.

soft

war

eov

er5

yrs

orde

prec

iate

30%

r45

%fo

rhi

gher

rate

3-yr

CF

-TC

onF

.A.

Sw

eden

100%

30%

DB

,no

neN

Ata

xli

abil

ity

NA

Ž dis

cont

inue

d4%

SL

—bl

dgs

.19

84S

wit

zerl

and

none

100%

like

inve

stm

ent

subc

ontr

acte

d?

2-yr

CF

?35

%on

roya

ltie

sin

cl.

soft

war

eor

5yr

cap.

rese

arch

Tai

wan

usua

l10

0%de

prec

.as

15%

2%re

venu

e,4-

yrC

FN

A3.

75–

20%

onro

yalt

ies

usua

l20

%3%

reve

nue

Not

esto

Tab

le1:

1.S

itua

tion

in19

95un

less

else

whe

resp

ecif

ied.

2.A

bbre

viat

ions

:R

sre

sear

ch,

NA

sno

tap

plic

able

,K

Cs

incr

emen

tal

tax

cred

it,

TC

sta

xcr

edit

.


Ž .Fig. 1. Tax component of R&D user cost — four most generous countries. Source: Bloom et al. 2000 .

ous treatment of R&D. The tax component user costmeasures the generosity of the tax system in subsi-

Ž .dising R&D see Appendix A . In general, the fulluser cost depends on differential inflation and inter-est rates, but we have set the real interest rate to be10% across all countries and years to highlight thetax element of the user cost. The user cost is weighted

Žacross assets 90% current expenditure, 3.6% build-.ings, and 6.4% plant and machinery . A value of

unity signals that the tax system is broadly neutralwith respect to R&D. This can occur if all R&Dwas fully written off and there were no special taxcredits.

Taking any year in isolation, it is clear that largedifferences exist among countries, a feature high-lighted in previous studies. It appears that Canadahas the most generous treatment of R&D, exceptduring 3 years in the mid 1980s when Australia gavea larger subsidy. Furthermore, in all of these coun-tries the tax treatment of R&D has become moregenerous since the early 1980s, although there hasbeen considerable turbulence. The relative positionof countries has moved around and there are substan-tial changes in the tax wedge on R&D due tochanges in tax policies. The mid to late 1980s was aperiod of particular change. This variation illustratesthe difficulty for firms considering long term invest-ment plans, that there may be considerable uncer-tainty about the permanence of fiscal incentives.

The reasons for the periods of large change in thecost of R&D vary across countries. In Australia, thelarge drop in 1985 was due to the introduction of a150% ‘‘superdeductibility’’ for R&D. The subse-quent increase was due to the lowering of Australia’sstatutory rate of corporation tax. The generosity ofthe Canadian system is driven by the fact that thecredit rate is relatively high on the incrementalamount of R&D. The fall in the cost of R&D in1988 was precipitated by the introduction of a sec-

Žond credit in Ontario the province which we model.here . In France, the introduction of the credit in

1983 had much less effect than the redefinition ofŽthe base from a moving base to a fixed base and

.then back again which occurred between 1987–1990. Similarly in the USA, the base re-definition in1990 had as large an effort as the introduction of thecredit in 1981. These points illustrate that the statu-tory credit rate is not of over-riding importance tothe cost of R&D. The design and implementation of

Ž .the schemes such as the definition of the base andŽthe effects of other parts of the tax system such as

.the statutory tax rate are at least of equal importancein explaining the trends over time.

Fig. 2 shows the tax wedge in the four lessgenerous countries. In these countries the tax sys-

Žtems are broadly neutral to R&D i.e., the tax wedge.is close to zero . There have not been many changes

in the tax treatment of R&D in these countries over


Ž .Fig. 2. Tax component of the user cost of R&D — four least generous countries. Source: Bloom et al. 2000 .

this period. Japan occupies an intermediate position,however, as the only country in this group which hasan R&D tax credit although the UK does also givean allowance for R&D capital expenditure.

Another striking feature of Figs. 1 and 2 takentogether is that the range of the user costs at the endof the period is greater than at the start. In 1979 themean effective marginal tax wedge on the typical

Ž .Fig. 3. Distribution of the effective R&D tax credit — U.S. Source: Hall 1993 .


Ž .Fig. 4. Distribution of the effective R&D tax credit — Canada. Source: Dagenais et al. 1997 .

R&D investment was 0.953 with a standard devia-tion of 0.098. By 1994 the mean had fallen to 0.857and the standard deviation increased to 0.163.

2.3. Heterogeneity of the effects of the tax system

One of the striking findings of the flourishing ofmicro-economic studies in the last two decades is thehuge heterogeneity between different firms. The wayin which the R&D tax credit creates heterogeneousand often perverse incentives has been a key feature

Ž .of the debate on the un desirability of R&D taxcredits. The heterogeneity emerges in many ways.First, unless there is a full refund then many firmswill not be able to use the full value of the tax creditbecause they do not have sufficient taxable profitsŽ .e.g., young firms or firms in recession . Carryfor-wards and carrybacks will compensate for this to anextent depending on interest rates and expectationsof future taxable profits. Second, there are usuallycaps limiting the maximum credit available. Third,the definition of the base will affect firms in differ-ent ways. A moving base will mean that firms whoare intending to increase their R&D may be put offbecause their current increases increase the size ofthe base which will limit their future tax rebatesŽ .Eisner et al., 1982 .

To illustrate the importance of heterogeneity, Fig.3 shows the distribution of the user cost of R&D in

Žthe US over time The user cost is defined slightly.differently from Figs. 1 and 2 – see Hall, 1993 .

There is considerable heterogeneity for most of theperiod. The reduction in the 1990s is due to movingfrom a moving base to a fixed base in 1989. Asimilar graph for Canada is given in Fig. 4. Thisvariation between firms is almost certainly an addi-tional source of uncertainty facing firms. It offers apotential source of identification in firm panel stud-ies of R&D.

3. Effectiveness of the R&D tax credit

There are two approaches to evaluating the effec-tiveness of any tax policy designed to correct theinsufficient supply of a quasi-public good. The firstasks whether the level of the good supplied after theimplementation of the policy is such that the socialreturn is equal to the social cost. In this situation,that would involve comparing the marginal return toindustrial R&D dollars at the societal level to theopportunity cost of using the extra tax dollars inanother way, for example, in deficit reduction. This


is a very tall order, and policy evaluation of the taxcredit usually falls back on the second method,which is to compare the amount of incrementalindustrial R&D to the loss in tax revenue. Theimplicit assumption in this method is that the size ofthe subsidy has been determined and that the onlyquestion to be answered is whether it is best adminis-tered as a tax credit or a direct subsidy. Obviously,this kind of benefit–cost ratio is only very looselyconnected with the magnitude of the gap between thesocial and private returns to R&D, if at all. It mightbe that the social return from additional industrialresearch is very high. If it is very high one may bewilling to give up more tax dollars than the actualresearch induced by the tax subsidy. On the contrary,if the social return is only slightly higher than theprivate return, lowering the cost of research mightcause the firm to do too much. 5 In this case, eventhough the tax credit induces more industrial R&Dthan the lost tax revenue, it would not be a goodidea, because one could have spent that tax revenueon some other activity which had a higher socialreturn. Fortunately, the available evidence on thesocial return to R&D suggests that the first case ismore likely than the second.

Most evaluations of the effectiveness of the R&Dtax credit have been conducted using the secondmethod, that is, as benefit–cost analyses. We need tocalculate both the amount of R&D induced by thetax credit, and computing the costs requires estimat-ing how much tax revenue is lost due to the presenceof the credit. The ratio of these two quantities is thebenefit–cost ratio; if it is greater than one, the taxcredit is a more cost-effective way to achieve thegiven level of R&D subsidy; if it is less than one, itwould be cheaper to simply fund the R&D directly.This part of the paper critically reviews the method-ology underlying these evaluations and surveys theresulting evidence, including the small number ofstudies that have been conducted using data fromoutside the US.

5 Some government policies towards R&D are explicitly aimedat reducing duplicative R&D — for example, in the US, govern-ment sponsored consortia such as SEMATECH, as well as theantitrust exemption contained in the National Cooperative Re-search Act of 1982.

3.1. Costs of R&D tax support

The first ingredient in doing a benefit–cost analy-sis of the tax credit is the computation of total cost.The total social cost consists of the net tax revenueloss due to the credit plus the costs of administeringit, both to the firm and to the taxing authority. Inpractice, the cost computed has been simply thegross tax credit claimed. At best this has been doneby simply adding up the credits claimed by the firms

Ž .that use the credit Mansfield, 1986; Hall, 1993 ,sometimes adding in the unused credits that have

Žbeen used to offset prior-year liabilities U.S. Gen-.eral Accounting Office, 1989 . Occasionally esti-

mates have been produced relying only on represen-tative or average firm behavior; this method is likelyto produce erroneous results given the extreme het-erogeneity in the data. Either way, this type ofanalysis ignores the fact that the existence and use ofthe R&D tax credit may have implications for theoverall tax position of the firm, so that the netchange in tax revenue because of the credit is notcaptured by simply adding up the credits. It is likelythat these other effects are relatively small, but by nomeans certain.

The second omission in the conventional compu-tation is the administrative cost of the tax credit. The

ŽGAO Study of 1989 U.S. General Accounting Of-.fice, 1989 , updated in 1995, makes it clear that

these costs can be high, but offers no estimate oftheir magnitude. Difficulties arise in two areas: thedefinition of eligible R&D, which typically requiresa distinction between routine and innovative re-search, and may be more restrictive than the defini-tion used by the firm’s accountants, and the perfor-mance of research by outside subcontractors. Forexample, the U.S. Internal Revenue Service appearsto have taken the position that the tax credit shouldflow to the organization that will pay for the R&D‘‘in the normal course of events’’, rather than to theorganization that bears the risk of the investment.

Ž .Stoffregen 1995 argues that these ambiguities ininterpretation of the law also impose costs on thefirms, in that they will be unsure whether the R&Dthey are undertaking will fall within the area delim-ited by the tax regulations as legitimate qualifiedexpenditures. The GAO reports that almost 80% ofreturns claiming R&D credits are audited in the US


with an average net adjustment downward of about20% of the credits claimed.

3.2. The benefits of R&D support: eÕaluation meth-ods

Can the R&D tax credit stimulate as much re-search per dollar as funding the R&D directly?Conceptually, measuring the amount of R&D in-duced by a tax credit is a ceteris paribus exercise, inwhich we attempt to ask the question: ‘‘How muchmore R&D did firms do given the existence of a taxcredit than they would have done if there had beenno credit?’’ The counterfactual is never observed,and researchers fall back on a variety of methods totry to estimate the level of R&D without the sub-sidy. We consider three evaluation methods.

3.2.1. EÕent and case studiesEvent studies typically rely on the assumption that

Žthe event being studied such as the introduction of a.tax credit is a surprise to the economic agents it

affects. They are usually conducted using financialmarket data, although this is not necessary. Themethod involves comparing behavior before a sur-prise change in policy is announced with behaviorafter the announcement in order to deduce the effectof the policy change. In this instance, such a compar-ison can take the form of comparing the marketvalue of R&D-oriented firms before and after thetax credit legislation was considered and passed, orof comparing R&D investment plans for the same

Žtime period before and after the legislation. Anexample of the former method is Berger, 1993 and of

.the latter is Eisner, as reported in Collins, 1983. Aproblem with many of these studies is that other

Ževents are not conditioned out such as demand.growth accompanying the policy change .

A case study is essentially a retrospective eventstudy. You simply ask the senior managers of indus-trial firms how their R&D spending has been af-

Žfected by the introduction of an R&D tax credit for.example, Mansfield, 1986 . These are often com-

Žbined with an econometric analysis e.g., Mansfieldand Switzer, 1985a,b who looked at 55 Canadian

. Ž .firms . These have the advantage that in principlethe manager controls for other factors when sheanswers the question. The main problem is that

managers may not give the right answer to thequestion, for subjective or perceptual reasons. 6 Fur-thermore, event and case studies tend to be focusedon rather small samples of firms, due to the cost ofcollecting the data to perform them.

3.2.2. Natural experiments: R&D demand equationwith a shift parameter for the credit

Here one constructs as well as possible an equa-Ž .tion that predicts the level of R&D investment r ,i t

ts time as a function of past R&D, past output,expected demand, perhaps cash flow and price vari-

Žables, and so forth different studies have different.conditioning variables — call these x . A dummyi t

Ž .variable is included C , equal to one when thei t

credit is available and zero otherwise. For example:

r sa qbC qgX x qu 3.1Ž .i t 0 i t i t i t

Where u is a stochastic error term. The magni-i tŽ .tude of the estimated coefficient of the dummy b

is equal to the amount of R&D induced by thepresence of the credit. If this exercise is conducted

Ž .using firm-level data is firm , the best method is tomeasure the availability of the credit at the firmlevel, that is, taking account of the usability of thecredit. If it is conducted at the macro-economic orindustry level, the identification of the credit effectwill generally come from the variation in R&D

Ž . Ždemand over time C sC examples: Baily andi t t

Lawrence, 1992; Swenson, 1992; Berger, 1993; Eis-.ner et al., 1983; McCutchen, 1993 . The advantage

of this method is its relative simplicity; it eliminatesthe need to perform the relatively complex computa-tions to determine the actual level of the tax creditsubsidy for each firm. The disadvantage is that themeasurement is relatively imprecise, because there isno guarantee that all firms are facing the samemagnitude of credit at any given point in time. Infact, we have seen how great the variation in the usercost has been after the credit was introduced in Fig.Ž . Ž .3 for the US and Fig. 4 for Canada . In addition, if

the variation in the credit dummy is over time, it is

6 There is a general tendency in surveys for managers to focusŽ .on their firm’s or their own individual idiosyncratic brilliance

rather than general features of the economic environment as thesource of positive change.


very possible that other forces which increase aggre-Žgate industrial R&D spending such as global eco-

.nomic conditions, trade, etc. and that are not in-cluded in the R&D equation may lead to a spuriousconclusion about the effectiveness of the tax credit.In other words the credit dummy is not separatelyidentified from a set of time dummies.

3.2.3. Quasi-experiments: price elasticity estimationThis method is similar to the previous method, in

that an R&D equation that controls for the non-taxdeterminants of R&D is estimated, but in this case aprice variable — the user cost of R&D — thatcaptures the marginal cost of R&D is included in the

Ž .equation. As with Eq. 3.1 lags may be introducedinto the explanatory variables. The estimated re-sponse of R&D to this price variable is converted toan elasticity of R&D with respect to price. If theprice variable includes the implicit subsidy given bythe tax system to R&D, this is a direct measure of

Žthe response of R&D to its tax treatment examples:.Hall, 1993; Dagenais et al., 1997 .

r sa qbr qgX x qu 3.2Ž .i t 0 i t i t i t

Even if the price variable does not contain ameasure of the tax subsidy, it is possibly to use themeasured elasticity of R&D with respect to price toinfer the response induced by a tax reduction of agiven size. This involves the step of estimating the

Žeffect of a given policy change such as an increase.in the credit rate on the user cost of R&D which is

a mechanical exercise given one’s definition of theprice. The second step is using the estimates of themodel to predict what will happen to R&D follow-ing a change in the price. In the most simple case,holding all else constant, if we estimate a priceelasticity of y0.5 and the effective marginal R&Dtax credit is 0.05, or a 5% reduction in cost, then theestimated increase in R&D from the tax credit will

Žbe 2.5% examples: Collins, 1983; Mansfield, 1986;.U.S. General Accounting Office, 1989 . Of course

this is too partial, a reduction in costs will also affectthe firm’s output and if output is in the equation, thefull effects are likely to be larger as output will riseas costs fall. There will also be possible spillovereffects, and so on. However, researchers have tended

Žto focus on the output–constant price effects see.below for more ‘‘structural’’ approaches .

The advantage of this method is that it is bettergrounded in economic theory and estimates the priceresponse of R&D directly. Thus it will be somewhatmore accurate than the previous method. Using the

Ž .tax price elasticity of R&D the first variant has acouple of disadvantages: First, because the firm ben-efits directly from the amount of R&D qualified toreceive the tax credit, it is possible that it will relabel

Žsome expenses as R&D legitimately or illegiti-.mately and the ‘‘true’’ induced R&D will therefore

be an overestimate. Secondly, and perhaps mostseriously, because the tax credit depends on a varietyof firm characteristics, such as its operating lossposition, how much foreign income it repatriates,and so forth, the R&D investment level and the taxprice faced by the firm are simultaneously chosen,and ordinary regression methodology is inappropri-ate in this situation. For this reason, some re-searchers have relied on instrumental variables toestimate the price elasticity, with both the attendantloss of precision in estimation and problems withfinding appropriate instruments to identify the en-dogenous variable. 7

The second variant of the quasi-experimental ap-proach suffers from deeper disadvantages. Absentvariations in tax treatment across firms and time, oneis forced to use a constructed R&D price deflator asthe price variable in an R&D demand equation.These deflators typically are a weighted average ofR&D inputs, of which around half is the wages andsalaries of technical personnel, and the other half issome kind of research materials and equipment in-dex. The only real variation in this variable is overtime. This is a very thin reed on which to rest theestimation of the price elasticity of R&D demand;the estimates will depend strongly on the othertime-varying effects included in the model.

We finish this section with some general method-ological problems. First, the theoretical justification

Ž .of Eq. 3.2 is unclear. Some writers have argued fora much more ‘‘structural’’ approach to the R&Dequation. This is more easily said than done, how-

7 Ž . Ž .See Hall 1993 and Hines 1993 for examples. Possibleinstruments are the lags of the user cost variables and the industrylevel deflators, as well as lagged values of firm characteristics inthe case of micro data.


ever. Structural investment models for physical capi-tal have had a poor record of success in empiricaltesting whether of q-models, Euler equations or

ŽAbel–Blanchard variety see Bond and Van Reenen,.1998, for a survey . Although various attempts have

been made to estimate these more structural formsŽnone have been conspicuously successful e.g., Hall,

. 81993; Harhoff, 1994 . A simple way of motivatingthe R&D investment equation is to treat it symmetri-cally to fixed investment. If the production function

Žcan be approximated as a CES constant elasticity of.substitution then the first-order condition under per-

fect competition would have the following form

g sa qbr qg y qu 3.3Ž .i t 0 i t i t i t

Ž . Ž .where g s the log R&D stock , y s log outputi t i tŽ .and r s log user cost of R&D . Under this modeli t

bs the Hicks–Allen elasticity of substitution. Con-stant returns implies that gs1. The stock is gener-ally calculated using the perpetual inventory method

Ž .where G sR q 1yd G , capital letters denot-t t ty1Ž .ing the levels not logs of g and r, and d is the

knowledge depreciation rate. Unfortunately, unlikephysical capital there is little information upon whichto base the initial condition in constructing thismeasure.

Several studies specify the R&D equation inŽterms of a stock rather than a flow measure e.g.,

.Shah, 1994; Bernstein, 1986 . It is important to beaware of this difference when examining the empiri-cal studies as the stock will be much higher than theflow. However, when the equation is specified in

Ž .logarithms as it usually is then the difference is notso clear. To see this assume that the R&D stock

Ž .grows at rate y , we have G s 1qy G so thati i t i i ,ty1

dqÕiR s dqÕ G s GŽ .i t i i , ty1 i tž /1qÕi

and

dqÕir s ln qg syh qgit i t i i tž /1qÕi

8 Ž .Hall 1993 is the only one of the studies in Tables 2 and 3 touse an Euler equation model for R&D investment demand, buteven she is unwilling to trust the estimates and also reports thesimple double log specification of the equation as well.

Ž .Substituting this equation into Eq. 3.3 gives

r sa qbr qg y qh qu 3.4Ž .i t 0 i t i t i i t

This implies that we have to allow for firm fixedeffects in the R&D equation, but that otherwise theestimates will be approximately the same, whetherwe use the log of the stock of R&D or its flow asthe dependent variable. 9 That is, as long as R&D isgrowing at approximately a constant rate at the firmlevel and we include fixed effects in the R&Dequation, the interpretation of the coefficients is the

Ž .same as it was in Eq. 3.3 .A deeper problem relates to the adjustment cost

function of R&D. ‘‘Reduced form’’ approaches willŽ .usually use a general dynamic form of Eq. 3.4 to

capture these. The problem is that adjustment costsfor R&D are likely to be large and this will bereflected in a large value for the lagged dependentvariable. Temporary shocks to the price are unlikelyto have very large effects and even permanent shockswill take a long time before their full effect is felt.This is compounded by the fact that R&D is charac-terised by large fixed and sunk costs so the linear

Ž .form of Eq. 3.4 may be inappropriate. At the leastone might consider modelling the decision to partici-pate in R&D separately from the amount of R&D

Ž .conditional on participation e.g., Bond et al., 1999 .

4. Econometric evidence

Since the preponderance of work has been doneon the US we focus first on the results of this workbefore surveying the smaller number of internationalstudies.

4.1. Studies on the US

Table 2 presents a summary of the results of themany studies of the US R&E tax credit that havebeen performed since its inception in 1981. In thistable we report an attempt to ascertain two standard-

9 Of course, the fixed effects will also control for many othervariables which have been omitted from the specifications such asfirm specific knowledge depreciation rates, so they would proba-bly also be useful in the version with the stock of R&D.


Tab

le2

Em

piri

cal

stud

ies

ofth

eef

fect

iven

ess

ofth

eR

&D

tax

cred

it—

Uni

ted

Sta

tes

Ž.

Ž.

Ž.

Ž.

Dat

eof

stud

y,C

olli

ns19

83;

Eis

ner

etal

.M

ansf

ield

Sw

enso

nB

erge

r19

93B

aily

and

Hal

l19

93M

cCut

chen

Hin

es19

93N

adir

ian

dŽ.

Ž.

Ž.

Ž.

Ž.

Ž.

auth

ors

Eis

ner

1983

1983

1986

1992

Law

renc

e19

93M

amun

eas

Ž.

Ž.

1987

;19

9219

97

Per

iod

ofcr

edit

1981

:219

81–

1982

1981

–19

8319

81–

1988

1981

–19

8819

81–

1989

1981

–19

9119

82–

1985

1984

–19

8919

56–

1988

Con

trol

peri

od19

81:1

1980

Not

rele

vant

1975

–19

8019

75–

1980

1960

–19

80?

1980

1975

–19

80N

otre

leva

ntN

otre

leva

ntD

ata

sour

ceM

cGra

w-H

ill

McG

raw

-Hil

lS

trat

ifie

dC

ompu

stat

Com

pust

atN

SF

R&

DC

ompu

stat

IMS

data

Com

pust

atq

surv

eys

surv

eys

Com

pus-

rand

omby

ind

and

10K

sta

t,IR

Sin

d.su

rvey

Dat

aty

pe99

firm

s;

600

firm

sfo

r11

0fi

rms

263

firm

s26

3fi

rms

122-

digi

tin

ds.

800

firm

s20

larg

edr

ug11

6m

ulti

na-

15in

dust

ries

Ž.

Ž.

Ž.

R&

D3,

4-di

git

bala

nced

bala

nced

unba

lanc

edfi

rms

tion

als

ind

for

tax

Ž.

Ž.

Ž.

Ž.

Ž.Ž

.Ž.Ž

.Ž.

Ž.

Ž.

Met

hodo

logy

3ev

ent

1du

mm

y4

surv

ey1

dum

my

1,

31

,2

2el

asti

city

1du

mm

y2

elas

tici

tyE

last

icit

yC

ompa

repr

e-R

&D

equa

tion

Log

R&

DR

esea

rch

R&

Dde

man

dC

ost

func

tion

ER

TA

est.

R&

Dco

mpa

red

pre-

Log

R&

Dde

man

deq

nin

tens

ity

eqn

eqn

wit

hta

xap

proa

chin

cto

post

-ER

TA

and

post

-ER

TA

Ask

edif

R&

DL

ogR

&D

R&

Din

tens

ity

dem

and

eqn

wit

hta

xpr

ice

byst

rate

gic

pric

efo

rse

cpu

blic

ally

spen

ding

for

R&

Dab

ove r

tax

ince

ntiv

ede

man

deq

uati

onw

ith

tax

pric

eor

var.

grou

pw

ith

861-

8fi

nanc

edR

&D

belo

wba

sein

crea

sed

equa

tion

cred

itdu

mm

yta

xcr

edit

Con

trol

sR

&D

lag

1&2,

Log

S,

Lag

Rr

S,

Ind.

Lag

R&

D,

Lag

R&

D,

Pas

tN

CE

s,D

om.

&fo

r.O

utpu

t,pu

blic

curr

ent

&la

gch

ange

Rr

S,

curr

ent

and

lag

curr

ent,

lag

dive

rs.,

sale

s,ta

xpr

ice

&R

&D

,ot

her

Ž.

Ž.

sale

s,C

Fin

LT

Deb

tIn

vrS

Ind.

outp

utlo

gsou

tput

logs

%dr

ugsa

les

sale

s,In

d,fi

rmfa

ctor

pric

esla

g1&

2In

vrS

,C

Fr

S,

dum

mie

sT

obin

’sq,

GN

PE

stim

ated

Ž.

elas

tici

tyIn

sig.

Insi

g.0.

35?

?1.

0–

1.5

0.75

0.25

1.0

–1.

50.

28–

10.0

?1.

2–

1.6

0.95

–1

Est

imat

ed-

1.0

NA

0.30

–0.

60N

A1.

741.

32

0.29

–0.

351.

3–

2.00

bene

fit-

cost

Com

men

tsA

lso

used

Not

ago

odIn

crea

ses

get

Cre

dit

dum

-U

sabi

lity

Tax

pric

eR

espo

nse

Hig

her

res-

Com

pare

sC

ompa

res

surv

eyex

peri

men

t;to

ola

rger

asti

me

mie

sde

pend

mea

sure

sas

sum

esfi

rmla

rger

inpo

nse

for

low

firm

sw

ith

and

opti

mal

ity

ofev

iden

ce,

OT

Aea

rly,

insu

ff.

pass

eson

usab

ilit

y;pr

oble

mat

icis

taxp

ayer

86–

91;

CF

firm

s;w

itho

utdi

ffer

ent

R&

Dco

mpu

tati

ons

cont

rol

for

TC

,st

rati

fied

byIV

esti

mat

ion

prob

lem

wit

hfo

reig

nta

xpo

lice

s:di

rect

poor

func

tion

alta

xst

atus

eqno

nhom

o-cr

edit

s—

diff

e-gr

ant

vs.

form

thet

icre

ntex

peri

men

tR

&D


ized results from these quite disparate studies: theŽprice elasticity of R&D for a typical firm in the

.sample and some kind of estimate of the benefit–costratio of the credit. In many cases, the data that wouldallow us to compute these numbers were not reallycomplete in the paper, and we were forced to givenothing, or a rough approximation to the quantitydesired. It is apparent from looking at the table that

Žthe first wave of estimates those using data through. Ž1983 differ substantially from the second those

.using data through 1988 and later in two respects.First, the early studies tend to have lower or non-re-ported tax price elasticities of R&D; only the laterstudy by McCutchen of large pharmaceutical firms isan exception, and the R&D equation in this studyappears to be misspecified. Secondly, they are typi-cally not based on the publicly reported 10-K datamaintained by Compustat, but on internal U.S. Trea-sury tax data, surveys and interviews, and, in onecase, an early Compustat file. This makes it difficultto ascertain whether the differences in results arebecause the response to the credit varied over time,or because the type of data used was substantiallydifferent.

Unfortunately, the only early study that used aŽlarge set of firms from Compustat Eisner et al.,

.1983 , contains an R&D equation that is not wellspecified, and does not contain any variable to cap-ture the effect of the tax credit. Thus it is notpossible to draw any conclusion about the incentiveeffect from the regressions published in this report.In order to investigate results using Compustat data

Ž .in the earlier period, Hall 1995a; b re-estimated theŽ .equations in Table 6 of Hall 1993 for the time

period 1981–1982 using ordinary least squares. Shefound that the estimated tax price elasticity for thisearlier period using Compustat data was slightlylower than that using Compustat data for the entire1980s, but still very significant. In either levels orgrowth rates, it is approximately y0.6 instead of they0.85 that was obtained for the whole period. If wemultiply this elasticity times the weighted averageeffective credit rates for 1981 and 1982 shown in

Ž .Table 3 of Hall 1993 , we obtain projected increasesin R&D spending during these 2 years of 2.1% and2.3% respectively; consistent with the relatively lowincreases reported by Eisner and Mansfield usingsurvey data that covered the same period.

As indicated above, later work using US firm-leveldata all reaches the same conclusion: the tax priceelasticity of total R&D spending during the 1980s ison the order of unity, maybe higher. This result was

Ž .obtained by Berger 1993 using a balanced Compus-Ž .tat panel, Hall 1993 using an unbalanced Compus-Ž .tat panel, Hines 1993 using a balanced Compustat

panel of multinationals and a tax price derived fromthe foreign income allocation rules for R&D rather

Žthan the credit, and by Baily and Lawrence 1987;.1992 using aggregate two-digit level industry data.

All of these researchers specified an R&D demandequation that contained lagged R&D, current andlagged output, and occasionally other variables suchas cash flow. Hall and Hines used instrumentalvariable techniques to correct for simultaneity in theequation. 10

Thus there is little doubt about the story that thefirm-level publicly reported R&D data tell: the R&Dtax credit produces roughly a dollar-for-dollar in-crease in reported R&D spending on the margin.However, it took some time in the early years of thecredit for firms to adjust to its presence, so theelasticity was somewhat lower during that period.Coupled with the weak incentive effects of the earlydesign of the credit, this low short run elasticityimplied a weak response of R&D spending in theinitial years, causing researchers to interpret it aszero or insignificant. Thus there is no actual contra-diction in the evidence.

However, most of the solid evidence we have todate rests upon the response of total R&D spendingto changes in the tax price of ‘‘qualified’’ R&E.This qualified R&E typically accounts for anywherefrom 50% to 73% of total R&D spending. It alsorests on rather shaky tax status data, where theeffective tax credit rate faced by the firm is inferredusing information in the Compustat files on operat-ing losses and taxable income over the relevantyears; where aggregate data is used, no attempt hasbeen made to correct for the usability of the credit.There is reason to believe that inferring the qualifiedR&E spending by multiplying total R&D on the

Ž .10-K by a common correction factor such as 0.6

10 Hall uses lags of the endogenous variables in a GMM estima-tor.


and inferring the tax status by looking at the 10-Knumbers is somewhat unreliable. The only study that

Ž .has used the true confidential corporate tax data isŽ .that by Altshuler 1989 and unfortunately for our

purposes here, it focuses on the weak incentive effectimplied by the credit design rather than evaluatingthe actual R&D induced.

Basing our conclusions on the response of totalR&D spending to a tax price inferred from Compus-tat data may suffer from two quite distinct problemsthat deserve further investigation: First, as discussedabove, the estimates based on public data may bequite noisy, and even misleading. Second, becausethese estimates are based on the response of reportedR&D to the credit itself, they may overestimate thetrue response of R&D spending to a change in price.This is sometimes called the ‘‘relabelling’’ problem.If a preferential tax treatment for a particular activityis introduced, firms have an incentive to make surethat anything related to that activity is now classifiedcorrectly, whereas prior to the preferential treatment,they may have been indifferent between labelling thecurrent expenses associated with R&D as ordinaryexpenses or R&D expenses. There is some sugges-

Ž .tive evidence reported in Eisner et al. 1986 con-cerning the rate of increase in qualified R&E expen-ditures between 1980 and 1981, when the credit tookeffect. Using a fairly small sample of firms surveyedby McGraw-Hill, they were able to estimate that thequalified R&D share grew greatly between 1980 and1981, less so between 1981 and 1982. This is consis-tent with firms learning about the tax credit, andshifting expenses around in their accounts to maxi-mize the portion of R&D that is qualified. It is alsoconsistent with the tax credit having the desiredincentive effect of shifting spending toward qualifiedactivities, although the speed of adjustment suggeststhat accounting rather than real changes are responsi-ble for some of the increase.

One way around the relabelling problem is to usea method of estimating the inducement effect thatdoes not rely directly on the responsiveness of R&Dto the tax credit. This is the method used in U.S.

Ž .General Accounting Office 1989 and in the Bern-Ž .stein 1986 study of the Canadian R&D tax credit.

One takes an estimated price elasticity for R&D,estimated using ordinary price variation and not taxprice variation, and multiplies this elasticity times

the effective marginal credit rate to get a predictedincrease in R&D spending due to the credit rate. Forexample, if the estimated short run price elasticity is

Ž .y0.13 as in Bernstein, 1986 , and the marginaleffective credit rate is 4%, the estimated short runincrease in R&D spending from the credit would be

Ž0.5%. With a long-run elasticity of y0.5 Bernstein.and Nadiri, 1989 and a marginal effective credit rate

of 10%, the estimated increase would be 5%. Inpractice, the difficulty with this method has been thatmost of the elasticity estimates we have are based ona few studies by Bernstein and Nadiri that rely onthe time series variation of an R&D price deflatorthat evolves as a fairly smooth trend and so iscorrelated with many other changes in theeconomy. 11 In addition, they are based on eitherindustry data from the 1950s and 1960s or a verysmall sample of manufacturing firms, so they maynot generalize that easily.

It is unlikely that the R&D demand elasticitywith respect to price is constant over very differenttime periods or countries, so it would be desirable tohave more up-to-date estimates in order to use thismethod. Obviously, one can never be sure that firmswill actually respond to a tax incentive in the wayimplied by the price elasticity and measured creditrate, but it would be useful to have this methodavailable as a check on the more direct approachusing tax prices.

4.2. Non-US studies

Few countries have performed as many studies oftheir incremental R&D tax credit programs as the

Ž .US. There are several reasons for this: 1 Most ofthese schemes have been in place for a shorter time

Ž .period. 2 They have relied on the US evaluationsŽ .for evidence of effectiveness. 3 Internal govern-

ment studies may have been done, but these are hardto come by if you are not connected with researcherswithin the government in question. The only studieswe have been able to find are displayed in Table 3.They cover Australia, Canada, France, Japan, andSweden.

11 Ž . Ž .See also Goldberg 1979 , Nadiri 1980 , Cardani and MohnenŽ . Ž .1984 , Mohnen et al. 1990 .

()

B.H

all,J.Van

Reenen

rR

esearchP

olicy29

2000449

–469

464

Table 3w Ž . Ž .xStudies of the of the R&D tax credit — other countries see text for a more complete of methodologies 1 – 4

Country Canada Canada Sweden Canada Japan Australia Canada G7 and Aus- France Canadatralia

Ž .Author s , McFetridge Mansfield Mansfield Bernstein Goto and Australian Bernstein Bloom et Asmussen Dagenais etŽ . Ž . Ž . Ž . Ž .date of study and Warda and Switzer 1986 1986 Wakasugi Bureau of 1998 al. 1998 and Berriot al. 1997

Ž . Ž . Ž . Ž .1983 1985a; b 1988 Industry 1993EconomicsŽ .1993

Period of 1962–1982 1980–1983 1981–1983 1981–1988 1980 1984–1994 1964–1992 1979–1994 1985–1989 1975–1992creditControl period NA Not relevant Not relevant 1975–1980 Non-users

Data source Statistics Stratified Stratified Prior esti- ABS R&D Canadian Manufactu- DGI, and CanadianCanada survey inter- random sur- mates survey IR&D manufacturing sector MRT data Compustat

Žview vey board ring panel esti- Statcan de-.mates flators

Data type Aggregate 55 firms 40 firms Firms? )1000 firms Sector 9 countries 339 firms 434 firmsŽ .30% of R

Ž . Ž . Ž . Ž . Ž . Ž . Ž . Ž .Methodology 2 Elastic- 4 Survey, 4 Survey, 2 Elastic- 1 , 4 Log Elasticity Elasticity R 1 Demand 1 Demand,ity, use elas- asked if R& asked if R& ity multiply R&D de- cost func- &D demand R&D de- log R&Dticity of 0.6 D tax incen- D tax incen- prior elastic- mand eqn tion eqn with mand eqn stock eqnand tax price tive in- tive in- ity estimate with credit approach tax-adjusted with with

U UŽ . Ž .of R&D creased creased times credit dummy con- user cost log credit , og credit ,spending spending rate trolrno con- Indicator for sample sel.

trol ceiling model

()

B.H

all,J.Van

Reenen

rR

esearchP

olicy29

2000449

–469

465

Controls NA No control NA Lag R&D, Output other Lagged R& Logs of gov Log sales,years, un- log size factor prices D, output subsidy, log capital,clear if these growth, tax country and size, size sq, ind. R stock,are total in- loss dummy, time dum- concentra- lag R stockcreases from gov support mies tion, fixed effectstax credit dummy

EstimatedŽ . Ž .Elasticity 0.6 0.04–0.18 small 0.13 y1.0 0.14 in 0.16 in 0.26 .08 0.40 .25

short-run short-runEstimated 0.30 in 1.1 in long-

long-run runŽ .Benefit-cost 0.60 0.38–0.67 0.3 to 0.4 0.83–1.73 0.6–1.0 ? 0.98 LR

Comments Elasticity Elasticity Increases Larger fig- Increased Elasticity is Find effect Estimated Includes acomes from estimated get larger as ure includes R&D by comb. of of tax cred- elasticity is selectionNadiri from McF& time passes. outout ef- 1% survey evi- its on relo- credit elas- eqn for do-Ž .1980 ‘‘ten- Warda tax fects dence and cation deci- ticity di- ing R&D;tative’’ cr. of 20% control sion vided by elasticity

and obs. R group analy- elasticity of derivedincrease sis tax price wrt from stock

credit est. C-B in-cludes out-put


There have been several studies of Canadian data.Ž .Dagenais et al. 1997 analyse Canadian firms using

the substantial variation in the R&D tax credit toconstruct a measure of the user cost. They estimate ageneralised Tobit model for the R&D stock whichallows the tax price to affect the amount of R&Dperformed as well as whether firms conduct R&D atall. They find a weakly significant effect on theformer with a long-run effect almost 20 times theshort-run effect. Through a simulation exercise theyfind that a 1% increase in the federal tax creditgenerates an average of US$0.98 additional R&Dexpenditure per dollar of tax revenues foregone.

One of the most comprehensive and carefullydone of these studies is that by the Australian Bureauof Industry Economics. It is noteworthy that theconclusions reached with respect to the tax priceelasticity and benefit–cost ratio are similar to thosein the recent US studies. The methodology usedcompares the R&D growth rates for firms able andunable to use the tax credit for tax reasons. This hasthe obvious disadvantage that assignment to a con-trol group is endogenous, and that the full marginalvariation of the tax credit across firms is not used,only a dummy variable. In general, the survey evi-dence that asks firms by how much they increasedtheir R&D due to the tax credit is consistent withthe econometric evidence.

Ž .The French study by Asmussen and Berriot 1993encountered some data difficulties having to do withmatching firms from the enterprise surveys, R&Dsurveys, and the tax records, so the sample is some-what smaller than expected, and may be subject toselection bias. The specification they used for theR&D demand equation includes the magnitude ofthe credit claimed as an indication of the cost reduc-tion due to the credit. If all firms faced the sameeffective credit rate on the margin, it is easy tocompute the tax price elasticity from the coefficientof this variable. Unfortunately, this is typically nottrue in France, so that this equation is not ideal forthe purpose of estimating the tax price elasticity.Even so, Asmussen and Berriot obtain a plausible

Ž .estimate of 0.26 0.08 , which is consistent withother evidence using similar French data and a truetax price.

Few studies have attempted to systematicallycompare the effectiveness of various R&D tax in-

centives across countries, partly because of the form-idable obstacles to understanding the details of each

Ž .system. McFetridge and Warda 1983 and WardaŽ .1994 have constructed estimates of the cost ofR&D for large numbers of major R&D-doing coun-tries. Like the Bloom 1998, 2000 study discussed inSection 2 they found that Japan, Germany, Italy,Sweden, and the UK had the highest tax cost ofR&D projects and the US, France, Korea, Australia,and Canada the lowest. Bloom 1998, 2000 use theuser-cost calculated over eight countries in Section 2to analyse the effect on R&D. Like the micro studiesthey also find a long-run elasticity of about unity but

Ž .a very low short run elasticity 0.16 . More interest-ingly they identify significant effects of the foreignuser cost of capital which they interpret to mean thatchanges in R&D tax credits can stimulate firms intorelocating their R&D across borders. This raises anew dimension in the debate over the efficacy of taxcredits. If some of the estimated increase comes frommultinationals relocating their R&D laboratories itraises the question of tax competition over‘‘footloose’’ R&D.

The central conclusion at present from studies inother countries is not different from those using USdata: the response to an R&D tax credit tends to befairly small at first, but increases over time. Theeffect of incremental schemes with a moving average

Ž .base France, Japan is the approximately the sameas in the US: they greatly reduce the incentive effectof the credit.

5. Conclusions

In this paper we have considered the tax treatmentof R&D and its effect on firm’s decisions. BecauseR&D is expensed it is tax privileged compared tofixed investment. There are also a host of special taxbreaks, such as the US R&E credit that furthersubsidise R&D activities. These have varied exten-sively over time and across countries to a muchgreater extent than physical capital. Our sense is thatthe tax treatment of R&D is becoming more lenientand it is likely that countries will increasingly turn tothe tax system and away from direct grants.


One feature of the existing schemes is that theyimply very heterogenous prices facing firms. Thisvariation is a useful source of identification of theeffect of price changes on quantity demanded, al-though there are still relatively few studies that haveused this. Taken as a whole there is substantialevidence that tax has an effect of R&D performed,the most compelling evidence coming from thequasi-experimental approach of calculating a usercost of R&D and estimating an explicit econometricmodel. A tax price elasticity of around unity is still agood ballpark figure, although there is a good deal ofvariation around this from different studies as onewould expect.

Looking ahead there are several ways in whichthe literature could grow. First, expanding beyondthe US to other countries is a trend which clearlyneeds to be encouraged. International firm leveldatasets are becoming more widely available and wewould emphasis to policy makers the imperative ofhaving more open, objective, statistical evaluationsof their policies. Secondly, there has been littleattempt to use the variation in tax prices as aninstrument for R&D in examining other variables ofinterest. For example we are interested in the ques-tion of the productivity effect of R&D and whetherthe tax credit could be used as a quasi-experiment toget better calculations of the return to R&D invest-ments. Finally, the issue political economy cutsthrough many of the issues here. Why and when dogovernments introduce tax breaks? Are they reactingto policies in other countries as the theory of taxcompetition suggests they will? Understanding theprocess by which different policies are conceivedand come to life is as important as evaluating theireffects once they are born and grown up.

Acknowledgements

This paper draws on previous work by the authorsŽespecially Hall, 1995a,b and Bloom et al., 1998,

.2000 and their colleagues. Special thanks to LucyChennells, Nick Bloom and Rachel Griffith. Thesecond author thanks funding from the ESRC Centrefor Microeconomic Analysis of Fiscal Policy at theInstitute for Fiscal Studies.

Appendix A. Measuring the user cost of R&D

The user cost of R&D is calculated using theŽ .standard approach of Hall and Jorgenson 1967 and

Ž .King and Fullerton 1984 and that was extended toŽ .the international setting in OECD 1991 and Dev-

Ž .ereux and Pearson 1995 . The aim of this approachis to derive the pre-tax real rate of return on themarginal investment project that is required to earn aminimum rate of return after tax. This will be afunction of the general tax system, economic vari-ables and the treatment of R&D expenditure inparticular.

We consider a profit maximising firm which in-creases its R&D stock by one unit in period one,then disposes of that unit in the second period. Thetax system affects the cost of making this investmentin two ways. First, the revenue earned from theinvestment is taxed at rate t . Second, the cost of theinvestment to the firm is reduced by depreciationallowances and tax credits.

Assuming that depreciation allowances are givenon a declining balance basis at rate f and begin int

the first period the value of the depreciation al-lowance will be t f in period one, and in subse-t t

Ž . 12quent periods the value falls by 1yf . Denotet

the net present value of the stream of these deprecia-tion allowances Ad,t

2t f 1yf t f 1yfŽ . Ž .i t t i t tdA st f q q q PPPt i t 1qr 1qrŽ . Ž .t t

t f 1qrŽ .t t ts

f qrŽ .t t

where r is the discount rate and the asset andt

country subscripts have been omitted for simplicity.Similarly we can calculate the net present value of

the tax credit, Ac, which will depend on the type oft

tax credit available on R&D expenditure. The mainfeatures that affect the value of a tax credit arewhether the credit applies to total or incremental

12 In practice depreciation allowances generally begin in thesecond period, or are given at half the rate in the first period. Thisis taken account of in the empirical application. Depreciationallowances may also be given on a straight line basis, in whichcase the expression for Ad is slightly different.t


expenditure, how the base level of expenditure isdefined in the incremental case and whether thecredit is capped on a firm by firm basis.

Under the assumption of perfect foresight and notax exhaustion the net present value of an incremen-tal tax credit with a base that is defined as thek-period moving average is

is11 yic cA st B y 1qr B A.1Ž . Ž .Ýt t t t tqiž /k k

where t c is the statutory credit rate, B is ant t q i

indicator which takes the value 1 if R&D expendi-ture is above its incremental R&D base in period tand zero otherwise. If the credit has an absolute firmlevel cap, as in France, then Ac is assumed to be ast

above for firms below the credit caps and zero forthose above the cap.

The depreciation allowances and tax credits varyacross types of asset, countries and time. We con-sider investment in the manufacturing sector intothree types of asset for use in R&D — currentexpenditure, buildings, and plant and machinery. Animportant assumption in the modelling strategy usedhere is that current expenditure on R&D is treated asan investment — that is its full value is not realisedimmediately. We also assume that domestic invest-ment is financed by retained earnings.

In an individual country, the user cost of a domes-Ž .tic investment in R&D for each asset indexed by j

is given by

1y Ad qAcŽ .jt jtr s r qd A.2Ž .Ž .jt t j1yt t

where d is the economic depreciation rate of thej

asset. The economic depreciation rates used are 30%for current expenditure on R&D, 3.61% for build-ings and 12.64% for plant and machinery. The do-mestic user cost of R&D for an individual country isthen given by

3dr s w r A.3Ž .Ýt j jt

js1

where w are weights equal to 0.90 for currentj

expenditure, 0.064 for plant and machinery and 0.036Ž Ž ..for buildings see OECD 1991 . The tax compo-

nent of the user cost of R&D is constructed using a

constant real interest rate across countries and overŽ .time 10% .

1y Ad qAcŽ .jt jttr s A.4Ž .jt 1yt t

References

Altshuler, R., 1989. A dynamic analysis of the research. NationalTax Journal XLI, 453–466.

Asmussen, E., Berriot, C., 1993. Le credit d-impot recherche, coutet effet incitatif. Ministere de l’Economie et des Finances,

Ž .Paris, France Avril 1993 . Photocopied.Australian Bureau of Industry Economics, 1993. R&D, Innovation

and Competitiveness: An Evaluation of the R&D Tax Conces-sion. Australian Government Publishing Service, Canberra.

Baily, M.N., Lawrence, R.Z., 1987. Tax Policies for Innovationand Competitiveness. Study commissioned by the Council on

Ž .Research and Technology, Washington, DC, April 1987 .Photocopied.

Baily, M.N., Lawrence, R.Z., 1992. Tax Incentives for R&D:What Do the Data Tell Us? Study commissioned by theCouncil on Research and Technology, Washington, DC,Ž .January 1992 . Photocopied.

Bell, J., 1995. The Australian 150% tax concession for R&D.Presentation of the Australian Delegation to the OECD AdHoc Meeting on Fiscal Measures to Promote R&D and Inno-vation, Paris, France.

Berger, P., 1993. Explicit and implicit effects of the R&D taxcredit. Journal of Accounting Research 31, 131–171.

Bernstein, J.I., 1986. The effect of direct and indirect tax incen-tives on Canadian industrial R&D expenditures. CanadianPublic Policy XII, 438–448.

Bernstein, J.I., Nadiri, M.I., 1989. Rates of return on physical andR&D capital and structure of the production process: cross-

Ž .section and time series evidence. In: Raj, B. Ed. , Advancesin Econometrics and Modelling. Kluwer Academic Publishers,Amsterdam, pp. 170–187.

Bloom, N., Griffith, R., Van Reenen, J., 1998. Do R&D taxcredits work? Evidence from an international panel of coun-tries 1979–1994, Institute for Fiscal Studies Working PaperW99r8.

Bloom, N., Chennells, L., Griffith, R., Van Reenen, J., 2000. In:Ž .Lawton-Smith, H. Ed. , The Economics of Regulation of

High Technology Industries. Oxford Univ. Press, Oxford.Bond, S., Van Reenen, J., 1998. Econometric models of factor

demand. Mimeo, IFS.Bond, S., Harhoff, S., Van Reenen, J., 1999. Investment, R&D

and financial constraints in Britain and Germany. Institute forFiscal Studies Working Paper Series W99r8.

Cohen, L.R., Noll, R.G., 1991. The Technology Pork Barrel.Brookings Institution, Washington, DC.

Collins, E., 1983. An early assessment of three R&D tax incen-tives provided by the Economic Recovery Tax Act of 1981.


Vol. PRAReport 83-7. National Science Foundation, Washing-ton, DC.

Dagenais, M., Mohnen, P., Thierrien, P., 1997. Do CanadianFirms Respond to Fiscal Incentives to Research and Develop-ment? Tilburg University mimeo.

Eisner, R., Steven, H.A., Sullivan, M.A., 1983. Tax incentives andR&D expenditures. In: Proceedings of the Conference onQuantitative Studies of Research and Development in Indus-try, edited by Ecole Nationale de la Statistique et de l’Admin-istration Economique and National Bureau of Economic Re-search, Vol. 2. CNRS, Paris, France, 1983, pp. 375–466.

Eisner, R., Albert, S.H., Sullivan, M.A., 1986. The new incremen-tal tax credit for R&D: incentive or disincentive. National TaxJournal XXXVII, 171–183.

Goto, A., Wakasugi R., 1988. Technology Policy. In: Komiya, R.,Ž .Okuno, M., Suzumura, K. Eds. , Industrial Policy of Japan.

Academic Press, Tokyo, 1988.Griffith, R., Sandler, D., Van Reenen, J., 1995. Tax incentives for

R&D. Fiscal Studies 16, 21–44.Hall, B.H., 1993. R&D tax policy during the eighties: success or

failure? Tax Policy and the Economy 7, 1–36.Hall, B.H., 1995. Fiscal policy towards R&D in the United States:

recent experience. Presentation to the OECD Working Groupon Fiscal Measures to Promote R&D, Paris, France, January19, 1995.

Hall, B.H., 1995. Effectiveness of research and experimentationtax credits: critical literature review and research design. Re-port to Office of Technology Assessment, Congress of theUnited States.

Hall, R., Jorgenson, D., 1967. Tax policy and investment be-haviour. American Economic Review 57, 391–414.

Harhoff, D., 1994. The tax treatment of R&D expenditure. ZEWand University of Mannheim, Research Report No. 94-02,

Ž .Mannheim, Germany, 1994 . Photocopied.Hines, J.R., Jr., 1993. On the sensitivity of R&D to delicate tax

changes: the behavior of U.S. multinationals in the 1980s. In:Ž .Giovannini, A., Hubbard, R.G., Slemrod, J. Eds. , Studies in

International Taxation, Ill. University of Chicago Press,Chicago, pp. 149–94.

Hiramatsu, H., 1995. Japanese fiscal measures to promote R&D.Presentation of the Japanese Delegation to the OECD Ad HocMeeting on Fiscal Measures to Promote R&D and Innovation,Paris, France, January 19.

Ž .King, M.A., Fullerton, D. Eds. , 1984. The taxation of incomefrom capital: a comparative study of the United States, theUnited Kingdom, Sweden, and West Germany. University ofChicago Press.

Klette, T., Møen, J., Griliches, Z., 2000. Do subsidies to commer-cial R&D reduce market failures? Microeconomic evaluationstudies. Research Policy 29 2000, 473–497.

KPMG, 1995. Tax Treatment of Research and Development Ex-penses. KPMG International Tax Centre, Amsterdam.

Leyden, D.P., Link, A.N., 1993. Tax policies affecting R&D: aninternational comparison. Technovation 13, 17–25.

Mansfield, E., 1986. The R&D tax credit and other technologypolicy issues. AEA Papers and Proceedings 76, 190–194.

Mansfield, E., Switzer, L., 1985a. The effects of R&D tax creditsand allowances in Canada. Research Policy 14, 97–107.

Mansfield, E., Switzer, L., 1985b. How effective are Canada’sdirect tax incentives for R&D?. Canadian Public Policy 11,241–246.

McCutchen, W.M. Jr., 1993. Estimating the impact of the R&Dtax credit on strategic groups in the pharmaceutical industry.Research Policy 22, 337–351.

McFetridge, D.G., Warda, J.P. 1983. Canadian R&D incentives:their adequacy and impact. Vol. Canadian Tax Paper No. 70,Canadian Tax Foundation, Toronto.

Nadiri, M., 1980. Contributions and determinants of research anddevelopment expenditures in the US manufacturing industries.

Ž .In: von Furstenburg, G. Ed. , Capital, Efficiency and Growth.Ballinger, Cambridge.

OECD, 1984. Science and Technology Policy: Review and Out-look. OECDrOCDE, Paris, France.

OECD, 1980. The Measurement of Scientific and Technical Ac-tivities. General Guidelines for Statistical Summaries of Re-search and Experimental Development, OECDrOCDE, Paris,France.

OECD, 1991. Taxing Profits in a Global Economy: Domestic andInternational Issues, Paris.

Ž .Seyvet, J., 1995. Le Credit d’Impot Recherche CIR en France.Ministere de l’Enseignement Superieur et de la Recherche.Presentation to the OECD Working Group on Fiscal Measuresto Promote R&D, Paris, France, January 19, 1995.

Shah, A., 1994. The economics of research and development: howresearch and development affects production and markets andis affected by tax incentives. Working Paper 1325, WorldBank, Policy Research Institute.

Stoffregen, P.A., 1995. Giving credit where credit is due: a briefhistory of the administration of the R&D tax credit. Tax Notes1995, 16 January 1995, pp. 403–16.

Swenson, C.W., 1992. Some tests of the incentive effects of theresearch and experimentation tax credit. Journal of PublicEconomics 49, 203–218.

U.S. General Accounting Office, 1989. The research tax credit hasstimulated some additional research spending. Vol.GAOrGGD-89-114, U.S. General Accounting Office, Wash-ington, DC.

Warda, J., 1994. Canadian R&D tax treatment: an internationalcomparison. Vol. Report No. 125-94. The Conference Boardof Canada, Ottawa, Ontario.

Documents

How effective are fiscal incentives for R&D? A …bhhall/papers/HallVanReenan...3 In addition to the features of the tax system targeted toward R&D equipment expenditures at the federal