THE EFFECTS OF ENTRY ON INCUMBENT INNOVATION AND

PRODUCTIVITY
Philippe Aghion, Richard Blundell, Rachel Griffith, Peter Howitt, and Susanne Prantl*

Abstract—How does firm entry affect innovation incentives in incumbent found in industries that lag behind. This is illustrated in figure
firms? Microdata suggest that there is heterogeneity across industries.
Specifically, incumbent productivity growth and patenting is positively
1, where we plot the annual rate of greenfield foreign firm
correlated with lagged greenfield foreign firm entry in technologically entry in each industry-year against the respective average of
advanced industries, but not in laggard industries. In this paper we provide subsequent total factor productivity growth in incumbent es-
evidence that these correlations arise from a causal effect predicted by
Schumpeterian growth theory—the threat of technologically advanced tablishments. The sample is split at the median distance to the
entry spurs innovation incentives in sectors close to the technology technology frontier, as measured by a labor productivity index
frontier, where successful innovation allows incumbents to survive the that relates incumbents in U.K. industries to their U.S. industry
threat, but discourages innovation in laggard sectors, where the threat
reduces incumbents’ expected rents from innovating. We find that the equivalent.
empirical patterns hold using rich micro panel data for the United Our explanation for this variation follows from Schum-
Kingdom. We control for the endogeneity of entry by exploiting major peterian growth theory—threat from frontier entrants in-
European and U.K. policy reforms, and allow for endogeneity of addi-
tional factors. We complement the analysis for foreign entry with evidence duces incumbents in sectors that are initially close to the
for domestic entry and entry through imports. technology frontier to innovate more, and this triggers
productivity growth, but entry threat reduces the expected
I. Introduction rents from doing R&D for incumbents in sectors further
from the frontier. In the former case, incumbent firms close

T HERE is a long-standing interest in the effects of entry,

which are widely recognized as major drivers of eco-
nomic growth. Entry can induce reallocation of inputs and
to the frontier know that they can escape and survive entry
by innovating successfully, and so they react with more
intensive innovation activities aimed at escaping the threat.
outputs, trigger knowledge spillovers, and affect innovation In the latter case, incumbents further behind the frontier
incentives in incumbent firms. The desire to induce entry by have no hope to win against an entrant. The escape-entry
foreign firms has spurred widespread policy reforms, par- effect in advanced industries is similar to the escape-
ticularly in countries or industries behind the technology competition effect developed in Aghion et al. (2001). The
frontier. However, empirical studies of the effects of market discouragement effect in lagging industries is similar to the
liberalizations and inward direct investment from foreign Schumpeterian appropriability effect of product market
firms provide mixed results on incumbent reactions.1 In this competition. Systematic variation of innovation activity
paper we explore systematic variation in the response of with distance to the technology frontier was introduced into
incumbent firms to entry. Schumpeterian theory by Howitt and Mayer-Foulkes (2005)
We are motivated by the following empirical regularity—we and, more closely related to this paper, by Acemoglu,
see substantial heterogeneity in the correlation between green- Aghion, and Zilibotti (2006).
field foreign firm entry and incumbent productivity growth Building on this theoretical background, we provide an
when we look across industries in the United Kingdom. In
empirical analysis of the variation of incumbent reaction to
industries close to the technology frontier there is a strong and
entry with distance to the technology frontier. We investi-
positive correlation, while a weak or even negative one is
gate how incumbent (labor and total factor) productivity
growth and patenting reacts to entry and find results that
Received for publication January 31, 2006. Revision accepted for
publication October 25, 2007. mirror the theoretical predictions. The main identification
* Aghion: Harvard University and Institute for Fiscal Studies (IFS); problem that we address arises because entry threat is not
Blundell and Griffith: University College London and IFS; Howitt: Brown observable and it is endogenous to incumbent performance.
University; and Prantl: Wissenschaftszentrum Berlin and IFS.
We are grateful to Daron Acemoglu, Francis Kramarz, Stephen Redding, We use actual foreign firm entry as a proxy for the unob-
Fiona Scott-Morton, Helen Simpson, Reinhilde Veugelers, seminar par- servable entry threat, which, if anything, exacerbates the
ticipants at Brown University, IFS, Stanford GSB, Yale University, and endogeneity problem (see discussion in section IIB). To
NBER, the Zvi Griliches conference in Paris, the EEA conference, the ES
World Congress, and the WZB-CEPR conference, and two anonymous tackle this we exploit variation in U.K. entry conditions that
referees for valuable comments and suggestions. This work contains arises from a major policy reform in the European Union,
statistical data from the Office of National Statistics (ONS) which is
Crown copyright and reproduced with the permission of the controller of
the Single Market Program, and from a series of U.K.
HMSO and Queen’s Printer for Scotland (under license number product market reforms in combination with rich micro
C02W002702). The use of the ONS statistical data in this work does panel data. We provide two interesting insights. First, we
not imply the endorsement of the ONS in relation to the interpretation
or analysis of the statistical data. This work uses research data sets find a consistent pattern of variation in incumbents’ reac-
which may not exactly reproduce National Statistics aggregates. Sup- tions to foreign firm entry using either U.K. policy reforms
plementary material is provided in a Web appendix available at or EU-wide policy reforms—a finding that may reduce
http://www.mitpressjournals.org/doi/suppl/10.1162/rest.91.1.20.
1 See, inter alia, Aitken and Harrison (1999), Pavcnik (2002), and political-economy concerns about using country-specific
Javorcik (2004). policy instruments in our context. Second, while our main

The Review of Economics and Statistics, February 2009, 91(1): 20–32

© 2009 by the President and Fellows of Harvard College and the Massachusetts Institute of Technology
 THE EFFECTS OF ENTRY ON INCUMBENT INNOVATION AND PRODUCTIVITY 21

FIGURE 1.—REACTIONS TO ENTRY IN INCUMBENTS NEAR AND FAR FROM panel data instead of plant or firm panel data to investigate
THE TECHNOLOGY FRONTIER
how the effectiveness of the Indian liberalization reform
depends on the technological and institutional state-industry
.08
subsequent total factor productivity growth rate

environment, in particular labor market regulation. Second,

there is the empirical industrial organization literature fol-
.06

lowing the work of Bresnahan and Reiss (1990, 1991) and

Berry (1992). Berry and Reiss (2006) survey structural
.04

econometric models with entry in well-defined, mostly

oligopolistic markets and endogenous market structure and
.02

discuss the insights gained into the determinants of firms’

entry decisions, the importance of firm heterogeneity, and
0

the nature of competition. Olley and Pakes (1996) investi-

gate the effects of deregulation on aggregate productivity
−.02

growth and the underlying reallocation mechanism in one

0 .02 .04 .06 particular industry. Our emphasis is instead on within-firm
greenfield foreign firm entry rate
changes in innovation incentives and in variation of entry
near frontier far from frontier effects on incumbent performance across markets. Another
Notes: The figure plots spline estimates of the relation between the greenfield foreign firm entry rate
related strand is the literature on product market competi-
and subsequent total factor productivity growth of domestic incumbent establishments in U.K. four-digit tion, firm performance, and innovation, in particular Nickell
industries, 1987 to 1993. Each dot represents the productivity growth estimate for establishments in one
industry-year cell. The spline regression includes time and industry fixed effects. Three spline points are (1996), Blundell, Griffith, and Van Reenen (1999), and
chosen such that all establishment observations in industry-year cells with nonzero entry are grouped into
four equally sized classes. The distance to the technology frontier is a labor productivity index relating Aghion et al. (2005a). Aghion et al. present evidence on an
incumbents in four-digit U.K. industries to their U.S. industry equivalent. The top curve (with dots) is for
establishments in industry-year cells near the technology frontier (less or equally distant to the frontier
inverted-U relationship between product market competi-
as the year-specific median of the distance distribution in the sample). The bottom curve (with triangles) tion and innovative activity and find this to be steeper in
is for establishments further behind the technology frontier (more distant to the frontier than the sample
median). neck-and-neck industries. Aghion and Griffith (2005) sur-
vey recent theoretical and empirical literature on competi-
tion, entry and growth, and the relevance of distance to
model specifications include distance to frontier and control frontier. Aghion and Howitt (2006) focus on policy impli-
variables such as import penetration and profitability that cations.
are assumed exogenous, we find similar effects of entry The paper is organized as follows. In section II we
when we allow for endogeneity of these variables. These discuss the theoretical background and empirical modeling.
findings relate our work to the literature on competition and In section III we describe our data. Empirical results are
trade. We complement our main analysis by considering presented in section IV and section V concludes.
whether different forms of entry have different impacts,
specifically entry by domestic firms or entry through import, II. Variation of Entry Effects with Distance to Frontier
and we explore why the two most likely alternative inter-
pretations—based on knowledge spillovers—are not consis- A. Theoretical Background
tent with the full pattern of our empirical results.
Our analysis relates to several different strands of empir- Variation of entry effects on incumbent performance,
ical work. First, there is the empirical literature on the depending on distance to the technology frontier, follows
effects of trade liberalization and inward direct investment from Schumpeterian growth models with escape-entry and
from foreign firms. Studies including Aitken and Harrison discouragement effects:
(1999), Pavcnik (2002), and Javorcik (2004) are, as ours,
based on plant or firm panel data and exploit variation of ● Increasing the threat of frontier entry induces incum-
trade or foreign firm activity across industries and time.2 bents in sectors that are close to the technology frontier
Aghion et al. (2004), Griffith, Redding, and Simpson to innovate more in order to escape entry. It reduces
(2002), and Haskel, Pereira, and Slaughter (2007) show for incumbents’ incentives to innovate in sectors that are
U.K. industries positive correlations of (increases in) the further behind the frontier, where there is little hope of
employment share in foreign firms and average growth of surviving entry.
total factor productivity in domestic producers. In contrast ● Increasing frontier entry threat has a more positive
to this, we use direct entry measures and focus on investi- effect on incumbent productivity growth in sectors that
gating systematic variation in incumbent innovation and are closer to the frontier than in sectors further behind
productivity reactions to entry with distance to the technol- the frontier.
ogy frontier. Aghion et al. (2005b, 2006) use state-industry
The idea of the escape-entry effect is similar to the idea
2
See also, for example, Amiti and Konings (2005), Bertschek (1995), of the escape-competition effect as developed in Aghion et
and Keller and Yeaple (2007). al. (2001). In that model each industry is assumed to be a
22 THE REVIEW OF ECONOMICS AND STATISTICS

duopoly, with two permanent rivals, and the degree of mance equations. We use lagged actual entry to proxy the
competition is measured by the elasticity of demand be- unobservable entry threat and, in doing so, we face the same
tween the rivals’ products. Here we consider a model in endogeneity problem as with entry threat, if anything in
which the rivals are constantly threatened with extinction by aggravated form.4 We discuss the endogeneity of entry
frontier innovators and there is an infinite cross-elasticity of below and outline our identification strategy. Second, we
demand. This leads to the escape-entry effect. Both of these focus on how the effects of frontier entry vary with distance
models assume step-by-step innovation instead of the leap- to the frontier. We measure technologically advanced entry
frogging assumed in earlier Schumpeterian models. To de- by considering foreign firm entry, and to measure distance
rive the escape-entry effect what is needed is that the to the frontier we use a labor productivity index that relates
probability that a frontier incumbent survives frontier entry incumbents in U.K. industries to their U.S. industry equiv-
is higher than the probability that a lagging incumbent alent. In our preferred specification the two continuous
survives.3 The discouragement effect in lagging industries is measures enter linearly and with an interaction. We also
similar to the Schumpeterian appropriability effect of prod- consider endogeneity of the distance to the frontier, check
uct market competition. Systematic variation of innovation whether the distance measure may capture other industry-
activity with distance to the technology frontier was intro- specific influences, and provide results for other forms of
duced into Schumpeterian theory by Howitt and Mayer- entry. Third, there are important covariates that may deter-
Foulkes (2005) and Acemoglu, Aghion, and Zilibotti mine the performance of incumbents in addition to entry—
(2006). most important, we think of effects triggered by trade
Related theory papers are, in particular, Gilbert and New- relations and other factors that affect competition, market
bery (1982) on preemption and Laffont and Tirole (1993) on structure, and the rents earned by incumbents. We control
the welfare effects of entry regulation in a model of product for these using observable and unobservable characteristics
differentiation. Laffont and Tirole concentrate on how reg- in our main empirical specifications, and in extended spec-
ulation affects the size of innovations, or the extent of ifications we allow for endogeneity of our main control
differentiation, by entrants and do not allow for incumbent variables. Fourth, to measure incumbent performance we
innovation. Fudenberg and Tirole (1984) and Tirole (1988) use two measures of productivity growth, as well as a count
analyze strategic interaction between entrants and incum- of patents.
bents. Barro and Sala-i-Martin (2004) and Aghion et al. To start with we measure incumbent performance as
(2005a) focus on competition among incumbent firms and growth of labor productivity at the establishment level
its effects on growth, but do not consider entry. Reallocative (⌬LPijt) and specify the following relation:
effects of globalization and trade liberalization are analyzed
in Melitz (2003) and Bernard et al. (2003). We focus instead ⌬LP ijt ⫽ ␣ ⫹ ␤ 1 E jt⫺1 ⫹ ␤ 2 D jt⫺1 ⫹ ␤ 3 E jt⫺1 D jt⫺1
(2)
on entry effecting growth through within-firm innovation
⫹ X⬘i jt⫺1 ␥ ⫹ ␶t ⫹ ␩i ⫹ uijt ,
effects in incumbents.
where i indexes incumbent establishments, j indexes indus-
B. Empirical Modeling tries, t indexes years, and E is actual greenfield foreign firm
The descriptive evidence in figure 1 is clearly not suffi- entry. To control for different permanent levels of produc-
cient to establish a causal relationship from entry to inno- tivity growth across establishments we include individual
vation and productivity growth or that it depends on dis- fixed effects ␩i. Common macro shocks are captured by
tance to the technology frontier. The central empirical time dummies ␶t. We also use growth of total factor pro-
relationship we are interested in is of the following form: ductivity, which may account for systematic variation in
factor inputs not captured in labor productivity growth.
Y ⫽ f共P, D, X兲, (1) Both measures of productivity growth could, however,
also reflect advances due to imitation of entrants with
where Y is a measure of incumbent performance, P is entry superior technologies rather than innovative activity. Thus,
threat, D is the distance to frontier, and X is a vector of we also use a count of patents as a measure of incumbent
further covariates. performance to check more directly whether our results are
We address a number of issues that arise when exploring picking up changes in firms’ innovative activity. There are a
this relationship empirically. First, entry threat P is unob- large number of firm-year observations with zero patents in
servable and potentially endogenous in incumbent perfor- our data, so we estimate a zero-inflated Poisson model

3 In section 1 of the Web appendix, we derive the escape-entry effect 4 In our theoretical context actual entry and entry threat are identical in

from the extreme assumption that a frontier incumbent survives frontier industries far behind the frontier, where entrants can never lose against
entry with probability 1, while a lagging incumbent survives with prob- incumbents. In industries closer to the frontier actual entry and entry threat
ability 0. A simplified version of this model with a fixed entry probability differ in situations where entrants may lose, but they will be positively
is sketched in Aghion et al. (2004), Aghion and Griffith (2005), and correlated as long as incumbent innovation aimed at escaping entry is not
Aghion and Howitt (2006). Aghion et al. (2005b) present a closely related too successful in the sense of entry prevention (Web appendix, sections
model. 1.2 and 1.6).
 THE EFFECTS OF ENTRY ON INCUMBENT INNOVATION AND PRODUCTIVITY 23

(Greene, 1994; Lambert, 1992). This also relaxes the re- tial portion of government-owned assets were sold and, in
strictive feature of the Poisson distribution that imposes most cases, the privatization decisions resulted in opening
equality of the variance and the mean. We model the up markets to firm entry. We use instruments that reflect the
probability of being granted at least one patent as a function respective stock market sales in directly affected industries.
of a firm’s presample stock of patents. Conditional on The U.K. reforms also include merger and monopoly cases
having at least one patent, we specify the innovation rate as where investigations of the U.K. Competition Authority
culminated in policy interventions. For each affected indus-
n ijt ⫽ exp共␣ ⫹ ␤1 Ejt⫺1 ⫹ ␤2 Djt⫺1 ⫹ ␤3 Ejt⫺1 Djt⫺1 try we construct a variable that indicates the dates on which
(3)
undertakings of inquiries were first publicly announced.8
⫹ X⬘i jt⫺1 ␥ ⫹ ␶t ⫹ ␩j ), In extended model specifications we allow for endogene-
where i indexes incumbent firms, j indexes industries, t ity of the distance to the technology frontier. This is to
indexes years, and ␩j indicates industry fixed effects. All address the concern that imposing exogeneity of that vari-
other variables and parameters are defined as above. To take able may affect findings on the variation of entry effects
unobservable firm-specific, time-invariant heterogeneity in with distance to frontier. We augment the set of instruments
patent behavior into account we follow Blundell et al. with U.S. variables on production inputs that correlate with
(1999) and use presample information on firms’ patent the distance to frontier, but should not depend directly on
stocks. anticipated developments in U.K. incumbent performance.
The policy interventions that we use for instrumenting
entry may also affect innovation incentives and productivity
C. Identification and Instruments
growth through other channels, especially through changing
The key identification issue that we tackle in this paper is trade relations or the competitive environment. Thus, we
the fact that entry can be endogenous to innovation and pay attention in our main specifications to controlling for
productivity growth. When considering entry into a new trade and competition, and assume that the instruments have
market potential entrants are likely to take into account the no additional impact on incumbent performance, after con-
productivity and innovative activity of local incumbents. ditioning on these covariates. We test the overidentifying
We expect a negative covariance between actual entry and restrictions in these specifications and experiment with
the error term in incumbent performance equations when using subsets of our policy instruments. In addition, we
industries are close to the frontier, but not necessarily in allow for endogeneity of the trade and competition covariate
industries far from the frontier.5 by adding U.S. trade and competition variables as instru-
We use two broad sets of policy reforms for instrument- ments, and relying on the additional assumption that the full
ing entry—reforms at the European level and reforms at the set of instruments affects entry, its interaction with the
U.K. level that changed entry costs and effected entry distance to frontier, and the instrumented covariate differ-
differentially across industries and time.6 In our main model entially.
specifications we endogenize the linear entry term as well as
its interaction with the distance to frontier. We show results III. Data and Descriptive Statistics
using different sets of policy instruments: instruments that
capture the EU Single Market Program only, U.K. policy A. Data
instruments only, and these instruments pooled.7 The Euro-
We combine microdata from several sources. Most im-
pean policy instruments indicate industries in which reforms
portant, we use comprehensive establishment-level panel
undertaken as part of the SMP were ex ante expected to
data for Great Britain from the U.K. Office for National
reduce medium or high entry barriers. The U.K. reforms
Statistics (ONS) Annual Respondents Database (ARD) for
include privatization cases—the Thatcher government em-
estimating productivity growth models. It is a legal obliga-
barked on a large-scale privatization program before similar
tion for firms to report these data. Innovation models are
programs were implemented in other countries. A substan-
estimated using firm-level accounting data from Data
5 In line with our theory framework foreign firms are more likely to enter
Stream that are matched to patent data from the NBER/Case
industries that are close to frontier if their relative advantages are high and Western Patent Database for a panel of firms listed on the
they anticipate this correctly. London Stock Exchange (LSE). These firms account for a
6 Controlling for unobservable individual, time-invariant heterogeneity
large proportion of U.K. R&D activity.9
and for time effects, as we also do, is unlikely to be sufficient to address
entry endogeneity, since industry-specific, time-varying changes of in- Productivity growth: Our key performance indicator is
cumbent performance should affect entry. Even lagging entry measures productivity growth, which we measure using the disaggre-
will not fully solve the problem if entrants anticipate changes to the gated ARD panel data on establishment inputs and outputs.
distribution of U.K. incumbent performance and this leads to relative
changes in entry across industries.
7 We apply IV estimation techniques in linear models, and a control 8 See section 2.2 and table A.4, both in the Web appendix, for further

function approach in nonlinear patent count models. In linear models, details.

control function and IV coefficient estimates coincide. See Wooldridge 9 Section 2 in the Web appendix provides further details on the data and

(2002) or Blundell and Powell (2003). the construction of variables.

24 THE REVIEW OF ECONOMICS AND STATISTICS

We calculate growth of labor productivity (LP) as growth in industries because they most often represent the world
real output per employee. To determine growth in total technological frontier, or are at least ahead of the United
factor productivity (TFP) we implement a superlative index Kingdom.13 Thus, U.S. industries can trigger technologi-
number approach, smoothing observed factor shares in cally advanced entry into the United Kingdom, and a large
order to mitigate potential consequences of measurement share of foreign entrants in Britain are indeed U.S.-owned.14
error. We check that our empirical results are robust to not The distance calculation uses U.S. industry panel data from
smoothing factor shares and to not imposing perfect com- the NBER manufacturing productivity database and U.K.
petition.10 data from the ARD. In addition to using a moving average,
Innovation: We measure innovation using the count of we also try alternative measures with other technology
patents firms take out in the U.S. Patent Office. Focusing on metrics and we use discrete versions of the variables to
U.S. patents of U.K. firms to measure innovation is advan- address concerns about measurement error.
tageous in our context, since U.K. firms are unlikely to Other variables: To measure trade activity we use the
patent low-value inventions in the United States. ratio of industry import over output from OECD STAN
Entry: Our focus in this paper is on technologically panel data. To capture the variation of competitive condi-
advanced entry, which we measure by greenfield entry of tions across industries and time we calculate an index of
foreign firms. This captures entry from firms that set up new average profitability in incumbent establishments based on
production facilities in Great Britain, and which operate ARD panel data. The index varies between 0 and 1 and
internationally and are thus most likely to produce at the takes the value of 1 in case of perfect competition. Pre-
technological frontier.11 Using panel data at the plant level sample information on patenting activity is summarized
from the ARD we calculate the annual greenfield foreign using a continuous patent stock variable based on patents
firm entry rate as the share of industry employment in dating back to 1968 along with a simple indicator of
entrants that meet the following conditions: the entering presample patent activity for firms in the panel of LSE-
firm (i) starts producing in one or more new British produc- listed firms.
tion sites in the year considered, (ii) is foreign owned, and
(iii) did not already operate beforehand in the respective
industry in Great Britain. B. Descriptive Statistics
Our measure has several advantages over other foreign To estimate productivity growth models we use an un-
entry measures that are commonly used. In contrast to balanced panel of 25,388 observations on 5,161 domestic
counting the number of foreign entrants, it takes the size of incumbent establishments in 180 four-digit industries
entrants into account. Compared with financial flows of (based on U.K. SIC 80) over the period 1987 to 1993.15 Of
inward direct investment the pattern of new real production these, 81% are older than ten years when entering the
activity in foreign firms is directly reflected. In contrast to sample. They have on average 263 employees between 1987
our earlier work (Aghion et al., 2004) and related literature, and 1993 and real output of £9m in 1980 pounds. Growth of
which use industry-level measures of employment in for- LP is on average 0.9% and TFP growth is ⫺1.1%. This
eign firms or equity owned by foreign investors,12 we focus reflects the recession in the early 1990s; the corresponding
on greenfield entry. This has the advantage of reflecting the figures for the years 1987 and 1990 are 2.3% and 1.0%,
scale of entry, but not reallocation between domestic and respectively.
foreign owners via acquisition, takeover, or merger activi- Innovation models are estimated using an unbalanced
ties. panel of 1,073 observations on 174 incumbent firms in 60
Greenfield entry of domestic firms is calculated in a three-digit industries between 1987 and 1993. Seventy-four
similar way and used below to proxy entry further behind percent of these firms were listed on the LSE for more than
the frontier. The value range for our entry measures is 0 a decade at sample entry. On average, they employ 8,286
to 100. people during the period 1987 to 1993 and have real sales of
Distance to the technology frontier: We capture the £433m in 1980 pounds. About 60% take out at least one
distance of incumbents in each U.K. industry to its U.S. patent at the U.S. Patent Office. As typically found, the
industry equivalent using a three-year moving average of sample distribution of patent counts is highly skewed—
U.S. industry labor productivity relative to labor productiv- many firms do not patent, some patent a little, and a small
ity in the respective incumbent U.K. industry. We average number of firms are granted many patents per year.
over the current and the two preceding years. We use U.S.
13 See, inter alia, Griffith, Redding, and Van Reenen (2004).
10 See table A.7 in the Web appendix. 14 For the time period 1986 to 1992 the ARD data show that, on average,
11 Multinational firms have been shown to be more productive on 36% of all greenfield foreign entrants in British manufacturing industries
average than firms that operate only nationally. For the United Kingdom are under U.S. ownership.
see Bloom, Sadun, and Van Reenen (2007), Criscuolo and Martin (2005), 15 As we would expect in line with the theory framework we find that our

and Griffith and Simpson (2004), among others. main results are stronger when we restrict the sample to incumbents that
12 See, among others, Aitken and Harrison (1999), Griffith et al. (2002), are more likely to be industry leaders than they are in the complementing
Haskel et al. (2007), and Javorcik (2004). subsamples (Web appendix, table A.7).
 THE EFFECTS OF ENTRY ON INCUMBENT INNOVATION AND PRODUCTIVITY 25

Foreign entryjt⫺1⫻

Notes: The table shows OLS estimates of basic entry models and of first-stage equations for the sample of 25,388 observations on 5,161 domestic incumbent establishments between 1987 and 1993. Bold numbers indicate coefficients. Standard errors in parentheses and
italics are robust, and observations are weighted by employment and the inverse of their sampling probability. In rows with F-test statistics, numerator degrees of freedom are in parentheses. The policy instruments that we use in first-stage equations are EU-SMP instruments
only in columns 5 and 6, U.K. policy instruments only in columns 7 and 8, and these instruments pooled in columns 9 and 10. Industry-specific reform variables are used for the following SIC-80 industries: 248 (refractory and ceramics), 2565 (explosive chemical products),
Given our interest in studying how entry effects vary with

371, 432/438,
Distancejt⫺1
distance to the industry-specific technology frontier, an

47.41(12)
3204, 361,
(0.006)
(0.013)

(0.013)
(0.049)

25.27(5)
15.09(2)
74.50(5)
0.019

0.070

0.004
ⴚ0.044

25,388
OLS

248, 2565,
(10)

Yes

Yes
Yes
important prerequisite for our empirical analysis is substan-

475
tial variation in the sample distance distribution. Thus, note
that about 20% of all four-digit industry-years in our data
are at or close to the frontier (less than 3.5% behind their

371, 432/438,
Foreign entryjt⫺1⫻ Foreign entryjt⫺1

53.03(12)
3204, 361,
U.S. industry equivalent), while another 20% are at least

19.99(5)
27.56(2)
92.60(5)
(0.014)
(0.029)

(0.049)
(0.102)

25,388
0.031

0.204

0.127
0.032
OLS

248, 2565,

Yes

Yes
Yes
(9)
50% behind. It is also crucial that we have variation in entry

475
rates at different distances to the frontier: there are four-digit
industry-years with no, some, or substantial greenfield for-
eign firm entry in each quartile of the distance distribution.

Distancejt⫺1

(0.013)

(0.013)
(0.050)

13.66(2)
77.35(5)
62.31(7)
In addition, comparing quartile-specific distributions we see

2565, 3204,
0.064

ⴚ0.001
ⴚ0.045

25,388
361, 475
OLS

Yes

Yes
Yes
(8)
considerable overlap of these for the entry rate, the number

3204 (fabricated constructional steel work), 361 (shipbuilding and repairing), 371 (precision instruments), 432/438 (cotton and silk, carpets and other textile floor coverings), and 475 (printing and publishing).
of employees in entering firms, and entrants’ size.16

Foreign entryjt⫺1⫻ Foreign entryjt⫺1

IV. Empirical Results

25.69(2)
94.16(5)
77.24(7)
(0.030)

(0.047)
(0.103)
2565, 3204,

25,388
361, 475
0.201

0.090
0.024
OLS

Yes

Yes
Yes
(7)
To investigate the economic mechanism behind the de-
scriptive evidence in figure 1 we analyze how the effects of

FIRST-STAGE EQUATIONS
frontier entry on incumbent innovation and productivity

248, 361, 371,

Distancejt⫺1
vary with the distance to the technology frontier, allowing

27.61(5)

27.61(5)
(0.007)

25,388
0.031

432/438
OLS

Yes

Yes
Yes
for endogeneity of entry and controlling for possible con-
(6)
founding factors. We address a number of potential robust-
ness concerns, take other forms of entry into account, and
conclude by explaining why the most likely alternative Foreign entryjt⫺1

248, 361, 371,

interpretations do not fully explain our empirical findings.

26.81(5)

26.81(5)
(0.019)

25,388
0.120

432/438
OLS

Yes

Yes
Yes
AND

(5)
TABLE 1.—ENTRY: BASIC MODELS

A. Entry
Foreign entryjt⫺1

The key identification issue that we address in our empirical (0.074)

77.46(1)
analysis is the potential endogeneity of entry to productivity
ⴚ0.651

25,388
OLS

Yes
Yes
(4)

growth and patenting. We instrument greenfield foreign firm

entry using major EU and U.K. policy reforms that aimed at
changing entry costs during the 1980s and early 1990s. In table
Foreign entryjt⫺1

1 we first show how the separate types of policy reforms relate

(0.030)

6.74(1)

25,388
0.078

to entry (columns 1–4), then we present our main first-stage

OLS

Yes
Yes
(3)

regressions for entry and the interaction of entry and distance

to the technology frontier (columns 5–10). These are used in
the second-stage estimations in table 2 and include all exoge-
Foreign entryjt⫺1

nous variables from the second-stage equations.

(0.102)

8.98(1)

25,388
0.305
OLS

Yes
Yes

In column 1 of table 1 we relate the EU-wide Single

(2)

Market Program (SMP) to greenfield foreign firm entry, and

constrain the EU-SMP coefficient to be common across all
affected industries. The positive and significant coefficient
Foreign entryjt⫺1

indicates that the EU-SMP led to increased entry—a result

28.33(1)
(0.029)

25,388
0.156
OLS

Yes
Yes
(1)

that is consistent with ex ante expectations. In column 2 we

use information on industries that are directly affected by
the U.K. privatization program, again constraining the co-
U.K. merger casesjtⴚ1 (MM)

efficient to be the same across industries, and find a positive

Additional industry-specific

import, and competition

U.K. privatization casesjtⴚ1

F-test, all excl. instruments

and significant coefficient, just as we do in column 3

U.K. monopoly casesjtⴚ1

Number of observations
Establishment effects
Programjtⴚ1 (SMP)

F-test, SMP variables

Distance to frontier,

F-test, MM variables

considering all U.K. merger cases where investigations of

reform controls
EU Single Market
Dependent variable

F-test, P variables

the U.K. Competition Authority triggered subsequent policy

interventions. In column 4 with U.K. monopoly cases that
Year effects
effects
(MM)
(P)

16 Tables A.2 and A.3 in the Web appendix provide more details on these

issues; descriptive statistics on the estimation samples are in table A.1.

26 THE REVIEW OF ECONOMICS AND STATISTICS

Bold numbers indicate coefficients. Standard errors in parentheses and italics are robust and allow for correlation between establishments within the same industry. Observations are weighted by employment and the inverse of their sampling probability. In the row with ␹2 test results, degrees
Notes: The table displays OLS and IV estimates of productivity growth models for the sample of 25,388 observations on 5,161 domestic incumbent establishments between 1987 and 1993. In columns 3–5 and 8–10 we allow for entry endogeneity in the linear and interacted entry terms.
triggered policy interventions the coefficient is negative and

SMP, MM, P
significant. However, if we allow the effect to vary in one

EF, EF ⫻ D
(0.024)

(0.004)

(0.020)

(0.039)

(0.040)

6.06(10)
ⴚ0.132

0.042

0.106

0.100

0.140

25,388
(10)

Yes
Yes
industry (printing and publishing) by adding an industry-
IV

specific reform variable to the vector of explanatory vari-

of freedom parameters are in parentheses. SMP indicates policy instruments capturing the EU Single Market Program, MM indicates policy instruments based on U.K. Competition Authority merger and monopoly cases, and P indicates U.K. privatization instruments.
ables then we find a positive and significant coefficient of
0.041 (standard error: 0.024) for the term that aggregates
Growth of total factor productivityijt

EF, EF ⫻ D
over all U.K. monopoly cases. In fact, when we look across
(0.026)

(0.004)

(0.022)

(0.040)

(0.041)

MM, P
4.59(5)
ⴚ0.130

0.041

0.106

0.100

0.140

25,388
Yes
Yes
(9)
IV

all policy reforms we find that their impact on entry (and on

the entry-distance interaction) is very pronounced in some
additional industries.
In the first-stage regressions in columns 5 and 6 we

EF, EF ⫻ D
(0.034)

(0.010)

(0.021)

(0.036)

(0.046)

1.23(3)
ⴚ0.123

0.041

0.103

0.100

0.143

25,388
include a common EU-SMP effect across affected industries

SMP
Yes
Yes
(8)
IV

and allow for four additional industry-specific effects.17 In

columns 7 and 8 we use the U.K.-based policy reforms and
allow the effects to vary in one industry affected by privat-
(0.015)

(0.003)

(0.019)

(0.037)

(0.044)

25,388 ization and three industries with merger and monopoly

ⴚ0.054

0.020

0.089

0.100

0.151
OLS

Yes
Yes
(7)

investigations.18 In columns 9 and 10 we include all of these

SECOND-STAGE EQUATIONS

instruments. For all first-stage regressions we report F-tests

on the joint significance of the instruments excluded from
(0.003)

(0.019)

(0.036)

(0.050)

25,388
0.007

0.077

0.099

0.162
OLS

Yes
Yes

the second-stage equations. Overall, the empirical evidence

(6)

suggests that the set of EU-wide and U.K.-based policy

reforms, which liberalized product markets and affected
SMP, MM, P

entry costs, have led to more greenfield foreign firm entry in

EF, EF ⫻ D

13.65(10)
(0.023)

(0.004)

(0.030)

(0.034)

(0.033)
ⴚ0.073

0.029

0.093

0.084

0.074

25,388

the United Kingdom.

Yes
Yes
(5)
IV
AND

B. Growth of Productivity
TABLE 2.—PRODUCTIVITY GROWTH: OLS ESTIMATES

We start by considering the average effect of entry on

EF, EF ⫻ D

subsequent productivity growth in incumbents, and then

(0.020)

(0.002)

(0.030)

(0.034)

(0.034)

MM, P
7.96(5)
ⴚ0.055

0.028

0.088

0.084

0.081

25,388
Growth of labor productivityijt

Yes
Yes
(4)
IV

focus on how entry effects vary with the distance to the

technology frontier. All regressions in table 2 include year
dummies and establishment effects. Standard errors allow
for correlation between establishments within the same
EF, EF ⫻ D
(0.031)

(0.010)

(0.029)

(0.033)

(0.033)

5.67(3)
ⴚ0.071

0.022

0.095

0.085

0.067

25,388

industry, and observations are weighted by employment and

SMP
Yes
Yes
(3)
IV

the inverse of their sampling probability.19

In columns 1 to 5 in table 2 we explain labor productivity
(LP) growth in incumbents. In column 1 we show OLS
(0.013)

(0.002)

(0.028)

(0.033)

(0.030)

25,388
ⴚ0.043

0.018

0.087

0.084

0.076

estimates using the lagged levels of foreign entry, distance

OLS

Yes
Yes
(2)

to frontier, import penetration, and competition as explan-

atory variables. We see a positive and significant correlation
of greenfield foreign firm entry with subsequent LP growth
(0.004)

(0.029)

(0.033)

(0.034)

25,388
0.008

0.077

0.084

0.084
OLS

Yes
Yes

in domestic incumbent establishments.20 High values of the

(1)

lagged distance measure indicate incumbent establishments

in U.K. industries in years where they are far from their
Foreign entryjtⴚ1 ⴛ distancejtⴚ1(EF ⴛ D)

17 We include industry-specific reform variables for the following

␹2 test of overidentifying restrictions

SIC-80 industries: 248 (refractory and ceramics), 361 (shipbuilding and

repairing), 371 (precision instruments), and 432/438 (cotton and silk,
carpets and other textile floor coverings).
Distance to frontierjtⴚ1(D)

18 These SIC-80 industries are 2565 (explosive chemical products), 3204

(fabricated constructional steel work), 361 (shipbuilding and repairing),

Number of observations
Import penetrationjtⴚ1

Establishment effects
Foreign entryjtⴚ1(EF)

and 475 (printing and publishing).

Type of instruments
Dependent variable

Instrumented terms

19 We find similar results if we estimate model specifications using

Competitionjtⴚ1

unweighted data, four-digit industry instead of establishment effects, and

standard errors that allow for correlation between establishments within
Year effects

the same industry-year. See table A.9 in the Web appendix.

20 This is in line with related findings in Aghion et al. (2004), Griffith et

al. (2002), and Haskel et al. (2007). The theory framework discussed
above yields this prediction for plausible parameter assumptions (Web
appendix, section 1.8).
 THE EFFECTS OF ENTRY ON INCUMBENT INNOVATION AND PRODUCTIVITY 27

industry-specific U.S. technology frontier. The positive and found to be important.24 We find evidence for negative
significant coefficient suggests higher LP growth rates for covariance between actual foreign entry and the error term
incumbents in industries far from the frontier. Another form in our productivity growth models in industries close to the
of entry we control for is entry into local product markets frontier. This is in line with our expectation derived from
through imports. We use a lagged measure of import pene- theory. However, our findings are also consistent with at-
tration and find a positive and significant effect on subse- tenuation bias towards 0 in OLS regressions, caused by
quent LP growth in incumbents. To capture the variation of measurement error.
competitive conditions across industries and time, we in- So far we have focused on results for model specifica-
clude a lagged measure of average profitability in incum- tions with growth in LP as the dependent variable. One
bents’ industries. In line with previous work (such as Nick- concern with this might be that our results are affected by
ell, 1996), we find a positive and significant coefficient.21 unaccounted for systematic variation in factor usage. In
Since our focus in this paper is on the variation of entry columns 6 to 10 of table 2 we reestimate using growth in
total factor productivity (TFP) as dependent variable. The
effects with the distance to the technology frontier, we now
results and conclusions hold up, and the precision of the
turn to more flexible empirical models where we interact
estimates tends to increase.
foreign entry and distance to frontier (see equation [2] in
To investigate the economic significance of our estimates
section IIB). The OLS estimates in column 2 show a
we calculate the growth impact of increasing entry by one
negative and significant coefficient on the interaction term, sample standard deviation (0.5 percentage points) at the
while the coefficients on the linear entry and distance terms 10%, 50%, and 90% percentile of the sample distribution of
remain positive and significant. The negative interaction the distance to the frontier, using the estimated coefficients
effect counteracts the positive effect of entry in industries from column 5. In industries at median distance, increasing
that are far away from the frontier.22 Thus, the OLS results greenfield foreign firm entry by one standard deviation
suggest that incumbent productivity growth responds more raises subsequent LP growth in domestic incumbent estab-
positively to technologically advanced entry in industries lishments by 0.7 percentage points. In industries far from
close to the technology frontier than in industries farther the frontier (90% percentile), the growth effect is ⫺0.6
below the frontier. percentage points, and close to the frontier (10% percentile)
In columns 3 to 5 we address the issue of entry endoge- it is 2 percentage points. Since the sample mean of incum-
neity in the linear and in the interacted entry terms. We use bent growth in LP is 1.1% and the standard deviation is 13.8
alternative identification strategies with different sets of percentage points, these estimates are economically signif-
policy instruments as shown in table 1 and discussed above. icant and their variation along the distance distribution
In column 3 we use the set of EU-SMP policy instruments— seems reasonable.25
the corresponding first-stage regressions are shown in col- All in all, the above OLS and IV estimates strongly
umns 5 and 6 of table 1. In column 4 of table 2 we use the indicate heterogeneity in the effects of greenfield foreign
set of U.K. policy instruments, and in column 5 the full set firm entry on subsequent LP and TFP growth of domestic
of EU and U.K. policy reforms. The exclusion restrictions in incumbent establishments as predicted from theory: techno-
these models are not rejected—the ␹2 test results are re- logically advanced entry in industries close to the technol-
ported near the bottom of table 2. All three IV regressions ogy frontier triggers subsequent productivity growth among
show negative and significant interaction effects, and posi- incumbents and can discourage it in industries that are far
tive and significant linear effects and, thus, confirm the main from the frontier.
conclusion from the OLS evidence.23 Instrumenting is

21 The results in this paragraph all hold as well in models where we C. Patenting
include each of the four explanatory variables separately. We checked that
the coefficient of the entry rate is driven by changes in the numerator, the
number of new employees in entering foreign firms, rather than by The evidence on productivity growth provides support for
changes in the denominator. See table A.6 in the Web appendix. the idea that frontier entry spurs incumbents to invest in
22 We find this pattern confirmed when we use a discretized model
innovation, particularly when they are near the technology
specification involving a different technology metric for measuring the
distance to frontier (Web appendix, table A.7). frontier. A lingering concern is, however, that productivity
23 Further variation of the set of policy instruments has also been growth may reflect not only entry-induced innovative activ-
investigated. When we restrict the set of instruments to either instruments ity, but also entrant imitation or growth driven by realloca-
capturing U.K. privatization cases only or U.K. merger and monopoly
investigations only, we get qualitatively similar, but noisier second-stage
results than those reported in table 2. However, in all these LP or TFP 24 We can reject the null hypothesis of exogenous entry terms using

growth regressions the entry terms remain jointly significant at the 1% F-tests on the control function: the F-test statistic for the LP model in table
significance level. This is also the case if we restrict the instrument set to 2, column 5, is 3.07 with two degrees of freedom, and the one for the TFP
four variables that aggregate, respectively, EU-SMP industries, U.K. model in table 2, column 10, is 7.92.
privatization cases, U.K. merger cases, and U.K. monopoly cases. In 25 Using the estimated coefficients from column 10 with total factor

addition, both entry terms are then also individually significant in the TFP productivity gives more pronounced, but qualitatively similar, economic
growth regression. See table A.8 in the Web appendix for further details. effects.
28 THE REVIEW OF ECONOMICS AND STATISTICS

TABLE 3.—PATENT COUNTS: ZERO-INFLATED POISSON ESTIMATES

(1) (2) (3) (4) (5)
ZIP ZIP ZIP-CF ZIP-CF ZIP-CF
Dependent variable Number of patentsijt
Foreign entryjtⴚ1 ⴛ distancejtⴚ1(EF ⴛ D) ⴚ0.557 ⴚ1.933 ⴚ3.238 ⴚ1.665
(0.237) (0.679) (1.618) (0.583)
Foreign entryjtⴚ1(EF) 0.107 0.245 0.608 0.437 0.506
(0.059) (0.078) (0.227) (0.216) (0.171)
Distance to frontierjtⴚ1(D) 0.582 0.652 0.852 0.753 0.825
(0.250) (0.251) (0.300) (0.254) (0.277)
Import penetrationjtⴚ1 1.746 1.692 1.937 1.957 1.834
(0.817) (0.770) (0.794) (0.759) (0.771)
Import penetrationjtⴚ1 squared ⴚ0.567 ⴚ0.542 ⴚ0.616 ⴚ0.605 ⴚ0.600
(0.309) (0.287) (0.297) (0.287) (0.291)
Competitionjtⴚ1 31.876 33.950 32.003 28.790 32.231
(16.764) (16.641) (17.200) (16.308) (17.054)
Competitionjtⴚ1 squared ⴚ17.722 ⴚ18.885 ⴚ17.910 ⴚ15.733 ⴚ18.007
(9.413) (9.340) (9.667) (9.240) (9.617)
Patent stocki, presample 0.005 0.005 0.005 0.005 0.005
(0.001) (0.001) (0.001) (0.001) (0.001)
D (patent stocki, presample > 0) 1.490 1.502 1.515 1.503 1.515
(0.317) (0.318) (0.319) (0.316) (0.317)
Year effects Yes Yes Yes Yes Yes
Industry effects Yes Yes Yes Yes Yes
Inflation Equation
Patent stocki, presample ⴚ0.207 ⴚ0.207 ⴚ0.207 ⴚ0.207 ⴚ0.207
(0.037) (0.037) (0.037) (0.037) (0.036)
D (patent stocki, presample > 0) ⴚ0.558 ⴚ0.554 ⴚ0.550 ⴚ0.553 ⴚ0.552
(0.175) (0.175) (0.175) (0.175) (0.175)
Control function EF, EF ⫻ D EF, EF ⫻ D EF, EF ⫻ D
Type of instruments SMP MM, P SMP, MM, P
Number of observations 1,073 1,073 1,073 1,073 1,073
Notes: The table displays zero-inflated Poisson estimates (ZIP) of patent count models; in columns 3 to 5 we allow for entry endogeneity in the linear and interacted entry term by including the respective first-stage
residuals as control function terms. Estimates are for the sample of 1,073 observations on 174 incumbent firms listed at the London Stock Exchange between 1987 and 1993. Bold numbers indicate coefficients.
Standard errors in parentheses and italics are robust and allow for correlation between firms within the same industry-year. SMP indicates policy instruments capturing the EU Single Market Program, MM indicates
policy instruments based on U.K. Competition Authority merger and monopoly cases, and P indicates U.K. privatization instruments.

tion between plants within incumbent establishments. To In column 1 greenfield foreign firm entry and distance to
address this issue we explore the relation between entry and the frontier enter in levels, while import penetration and
innovation more directly in patent count models. While we competition enter as quadratic functions. Entry is positively
use a different data set, we find a strikingly similar pattern correlated with the patenting activity of U.K. incumbent
of entry effects as for productivity growth. firms, as is distance to the technological frontier. For import
In table 3 we present estimates from a zero-inflated penetration, the effects are increasing until above the 90%
Poisson model.26 For comparison we also show results for a percentile of the sample distribution and positive for the
linear model in column 9 of table 4 and for a generalized whole distribution. We find an inverted-U relationship be-
negative binomial model in column 10. All specifications in tween competition and patent counts, in line with Aghion et
table 3 include year effects, dummies for (grouped) three- al. (2005a).
digit industries, and firm-specific presample patent stock In column 2 we include the interaction between foreign
variables to capture unobservable firm-specific, time- entry and distance to the frontier, and find this is negatively
invariant heterogeneity of patenting behavior (see Blundell correlated with patenting in correspondence to our produc-
et al., 1999). We show sandwiched estimates of the standard
tivity growth results. In columns 3, 4, and 5 we allow for
errors, which allow for correlation between firms within the
endogeneity of the linear and interaction terms by using the
same industry-year. The probability of being granted zero
residuals from the first-stage regressions for entry and
patents is modeled as a function of a firm’s presample stock
entry-distance interaction as control function corrections.
of patents.27
As in table 2, we first show results using EU-SMP instru-
26 The firm panel that we use in this section provides industry informa- ments only, then instruments reflecting the U.K. policy
tion on the three-digit industry level only, whereas all estimations dis- reforms only, and then using all these instruments together.
cussed so far involve using information on the four-digit industry level. The pattern of estimates holds across all three instrumenting
27 When testing the inclusion of additional variables into the inflation

equation, especially measures of entry, distance to frontier, trade, or strategies. Taken together, the patent count results suggest
competition, these turned out to be irrelevant. that a major driving force behind the entry effects in our
 THE EFFECTS OF ENTRY ON INCUMBENT INNOVATION AND PRODUCTIVITY 29

productivity growth estimations is incumbent innovation, reported, linear (and quadratic) specifications for import
not just imitation or reallocation. penetration indicate positive and significant level effects.
The coefficients of the interactions with the distance to
D. Extensions and Further Robustness Checks frontier in our performance regressions are negative and
significant in the LP growth and the patent count regressions
Greenfield domestic firm entry and entry through imports. (table 4, columns 2 and 6).31 In all these regressions our
In our empirical analysis we focused so far on foreign firm main findings for the linear and interacted terms of green-
entry, which is from firms that operate internationally, and field foreign firm entry remain very stable.32
are more likely to produce at the technological frontier than
other entrants in the United Kingdom. This accords well Endogeneity of distance to frontier, competition, and
with introducing entry threat at the new technological fron- import penetration. We augment our main model specifi-
tier into the framework of Aghion et al. (2001). cations from table 2, column 5, and table 3, column 5, to
In that theoretical context, we can also explore the case allow for endogeneity of the distance to the technology
where entry takes place behind the new frontier. If entry frontier. We add the industry-level U.S. capital-labor ratio
takes place one step behind the new frontier, then increasing and the industry-level U.S. ratio of skilled over all workers
entry threat encourages innovation and productivity growth
to the set of instruments. These are significantly correlated
in sectors that are at intermediate distance from the frontier;
with the distance measure and we assume that they do not
it discourages innovation and productivity growth in sectors
depend directly on anticipated shocks to incumbent perfor-
that are far below the frontier; and it has little effect close to
the frontier. In the case where entrants threaten to enter two mance in the United Kingdom. We estimate three first-stage
or more steps behind the new frontier, no incumbent reac- regressions: one for entry, one for the distance, and one for
tions are to be expected.28 their interaction. The findings for LP growth and patenting
In table 4 we show that greenfield domestic firm entry has in table 4, columns 3 and 7, show that our second-stage
no impact on incumbent LP growth (column 1) or patenting entry, distance, and interaction results remain robust.33
(column 5). The linear effects are insignificant, as are the In addition, we test the robustness of our findings to
interactions with the distance to frontier.29 These results allowing for endogeneity of import penetration or competi-
correspond with the predictions above, since greenfield tion. When treating import penetration as potentially endog-
domestic entry rates are likely to reflect entry behind the enous in the LP growth model we use the industry-level
frontier. Typical findings in the literature are that the aver- U.S. import penetration as an additional instrument and
age domestic entrant struggles for survival during the first estimate three first-stage regressions: one for entry, one for
years after market entry, is occupied with learning about its the entry-distance interaction, and one for import penetra-
own productivity and market conditions, and is very small tion.34 The second-stage results in table 4, column 4, pro-
compared with foreign entrants or incumbents in the same vide support for a positive level effect of import penetration
industry. The number of innovative domestic entrants is on labor productivity growth and, most important, for het-
usually small, they are often found to be hampered by erogeneous effects of greenfield foreign firm entry along the
financial constraints or immature technologies and, thus, distance to the frontier distribution. These findings are
even innovative domestic entrants are unlikely to challenge confirmed in the corresponding patenting and TFP growth
incumbents shortly after their market entry.30 regressions.35 To address potential endogeneity of our com-
Entry through imports is another form of entry into petition covariate, we add an industry-level index of U.S.
product markets, and industry-level import penetration rates profitability to the set of instruments, estimate the extended
into the United Kingdom partly reflect entry of new prod-
set of first-stage equations, and find our main results in LP
ucts. Among these may be technologically advanced prod-
growth, TFP growth, and patent count regressions again
ucts, but also less advanced products. Our import penetra-
confirmed.36
tion variable is, thus, a much noisier measure of frontier
entry than greenfield foreign firm entry. In accordance with
this we find similar, but weaker, effects for import penetra- 31 In the TFP growth regression we find a positive and insignificant

tion than for greenfield foreign firm entry. As already coefficient for the import-distance interaction (Web appendix, table A.9).
32 If we interact competition with the distance to frontier, these interac-

tions remain insignificant in the LP growth, TFP growth, and patent count
28 In this case actual entry would not occur since entry is optimal only regressions. The estimates for the linear distance terms get noisy, but the
when the entrant can take away market shares from the incumbent. coefficients for the entry terms and the entry-distance interactions remain
29 The corresponding TFP growth regression confirms (Web appendix, stable and significant at the 1% or 5% significance level (Web appendix,
table A.9). table A.10).
30 See, for example, Caves (1998), Disney, Haskel, and Heden (2003), 33 Results are similar if we use TFP growth as dependent variable (Web

Dunne, Roberts, and Samuelson (1988), Geroski (1995), and Gompers and appendix, table A.9); first-stage results are in table A.5 in the Web
Lerner (1999). In our data the average plant size of domestic entrants is appendix.
about ten times smaller than that of foreign entrants in their industry and 34 See table A.5 in the Web appendix for first-stage results.

about seven times smaller than that of incumbent plants in their industry 35 See table 4, column 8 and table A.9 in the Web appendix.

that are at least five years old. 36 See table A.10 in the Web appendix.
 30

TABLE 4.—ROBUSTNESS OF PRODUCTIVITY GROWTH AND PATENT COUNT RESULTS

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
IV IV IV IV ZIP-CF ZIP-CF ZIP-CF ZIP-CF OLS GNB
Dependent variable Growth of labor productivityijt Number of patentsijt

Foreign entryjtⴚ1 ⴛ ⴚ0.074 ⴚ0.052 ⴚ0.075 ⴚ0.098 ⴚ1.723 ⴚ1.438 ⴚ1.899 ⴚ1.664 ⴚ7.715 ⴚ0.546
distancejtⴚ1 (EF ⴛ D) (0.021) (0.023) (0.019) (0.026) (0.558) (0.552) (0.861) (0.593) (3.675) (0.285)
Foreign entryjtⴚ1(FF) 0.028 0.022 0.020 0.032 0.514 0.410 0.475 0.501 3.206 0.296
(0.003) (0.005) (0.009) (0.004) (0.168) (0.170) (0.208) (0.172) (1.497) (0.082)
Domestic entryjtⴚ1 ⴛ 0.003 0.137
distancejtⴚ1 (0.005) (0.085)
Domestic entryjtⴚ1 ⴚ0.002 0.004
(0.002) (0.040)
Distance to frontierjtⴚ1 (D) 0.087 0.166 0.231 0.092 0.721 2.278 1.127 0.785 8.181 1.294
(0.033) (0.047) (0.104) (0.028) (0.322) (0.811) (0.480) (0.273) (2.575) (0.327)
Importjtⴚ1 ⴛ distancejtⴚ1 ⴚ0.077 ⴚ1.117
(0.035) (0.576)
Import penetrationjtⴚ1 (I) 0.084 0.085 0.067 0.201 1.823 1.825 1.664 1.807 5.509 1.558
(0.034) (0.034) (0.033) (0.073) (0.820) (0.740) (0.803) (0.781) (5.816) (0.794)
Importjtⴚ1 squared ⴚ0.607 ⴚ0.531 ⴚ0.560 ⴚ0.650 ⴚ1.190 ⴚ0.452
(0.288) (0.276) (0.291) (0.306) (1.937) (0.321)
Controls as in table 2 Yes Yes Yes Yes
Controls as in table 3 Yes Yes Yes Yes Yes
Inflation model as in table 3 Yes Yes Yes Yes
Competition and year Yes
controls as in table 3,
firm effects
Instrumented terms/Control EF, EF ⫻ D EF, EF ⫻ D EF, EF ⫻ D, D EF, EF ⫻ D, I EF, EF ⫻ D EF, EF ⫻ D EF, EF ⫻ D, D EF, EF ⫻ D, I
function
Type of instruments SMP, MM, P SMP, MM, P SMP, MM, P, SMP, MM, P, SMP, MM, P SMP, MM, P SMP, MM, P, SMP, MM, P,
U.S. Input U.S. Import U.S. Input U.S. Import
THE REVIEW OF ECONOMICS AND STATISTICS

␹2 test of overidentifying 13.18(10) 10.17(10) 8.81(11) 13.92(10)

restrictions
Number of observations 25,388 25,388 25,388 25,388 1,073 1,073 1,073 1,073 1,073 1,073
Notes: The table displays IV estimates of productivity growth models in columns 1, 2, 3, and 4, and OLS estimates of a patent count model in column 9. In these columns standard errors in parentheses and italics are robust and allow for correlation between establishments
within the same industry. Zero-inflated Poisson estimates with control function terms are shown in columns 5, 6, 7, and 8; the inflation model is specified as in table 3. Column 10 provides results from a generalized negative binomial model without control function where
the shape parameter alpha is specified as a function of firm-specific presample patent stock information. Standard errors in these columns are robust and allow for general correlation between firms within the same industry-year. Bold numbers indicate coefficients. In the row
with ␹2 test results degrees of freedom parameters are in parentheses. We use the following abbreviations for policy instruments: SMP: EU Single Market Program; MM: U.K. Competition Authority merger and monopoly cases; P: U.K. privatization instruments; U.S. Input:
industry-level U.S. capital-labor ratio and ratio of skilled over all workers; U.S. Import: industry-level U.S. import penetration.
 THE EFFECTS OF ENTRY ON INCUMBENT INNOVATION AND PRODUCTIVITY 31

Alternative omitted effects and knowledge spillovers as incentives with the distance to the technology frontier.
explanations. We investigate the possibility that the inter- Threat of technologically advanced entry encourages in-
action between the distance to the technology frontier and cumbent innovation and productivity growth in sectors that
foreign entry may simply reflect alternative omitted inter- are initially close to the technological frontier, whereas it
action effects. We expand the covariate vectors of our main may discourage incumbents in sectors further behind the
model specifications with additional industry characteristics frontier. We use rich micro panel data and address the
that might affect incumbents’ abilities and incentives to problem of endogeneity in foreign entry by exploiting
react to entry. When adding, for example, a lagged industry- variation in entry conditions that arises due to major Euro-
level measure of average establishment size and its interac- pean or U.K. policy interventions. Endogeneity of distance
tion with entry to the labor productivity growth model we to frontier, competition, and trade are also considered, and
find similar effects for the linear entry term, the linear results for domestic firm entry and entry through import
distance term, and their interaction as before.37 Using in- complement our analysis of foreign firm entry.
stead the industry employment share in establishments with Our findings have implications for the policy debate on
working owners to capture the variation of ownership struc- market (de)regulation, competition policy, privatization, and
tures across industries and time, or using an industry-level trade liberalization. This debate underlies the consideration
measure of capital per worker, does not lead to any insta- of costs and benefits of globalization and the discussion on
bility of our main findings. entry regulation in different countries and industries (Ace-
Finally, we consider the extent to which alternative the- moglu et al., 2006; Bertrand & Kramarz, 2002; Djankov et
oretical explanations may also be consistent with the pattern al., 2002; Nicoletti & Scarpetta, 2003). Policies aiming at
of empirical results reported above. Potential candidates are decreasing or removing product market barriers to entry
theories that focus on the role of knowledge spillovers alone may not be sufficient to foster growth of incumbent
instead of innovation incentives. Consider the widely estab- firms in all sectors of an economy, even if such policies are
lished idea that firms and sectors further from the technol- found to be growth-enhancing on average. This, in turn,
ogy frontier should benefit most from knowledge spillovers, suggests the need for complementary labor and capital
since the scope for learning is highest there.38 This suggests market institutions that facilitate the reallocation of factors
positive coefficients on the linear distance to frontier terms, and resources from less to more technologically developed
as well as on their interactions with entry. We find, however, sectors where incumbent firms respond more positively to
a different pattern, namely negative interaction effects and higher entry threat.
positive level effects of the distance to frontier.39
Another idea prevalent in the existing literature on
knowledge spillovers argues that firms in industries closer REFERENCES
to the technology frontier have higher absorptive capacity Acemoglu, Daron, Philippe Aghion, and Fabrizio Zilibotti, “Distance to
and may benefit more from spillovers. If so, then firms in Frontier, Selection, and Economic Growth,” Journal of the Euro-
pean Economic Association 4:1 (March 2006), 37–74.
industries closer to the frontier should react stronger to Aghion, Philippe, Nick Bloom, Richard Blundell, Rachel Griffith, and
general spillovers, as well as to knowledge transfers from Peter Howitt, “Competition and Innovation: An Inverted-U Rela-
entrants, than firms in industries further behind the frontier. tionship,” Quarterly Journal of Economics 120:2 (May 2005a),
701–728.
Our finding of negative and significant coefficients on the Aghion, Philippe, Richard Blundell, Rachel Griffith, Peter Howitt, and
interaction terms is consistent with this. But the positive and Susanne Prantl, “Entry and Productivity Growth: Evidence from
significant coefficients for the linear distance to frontier Microlevel Panel Data,” Journal of the European Economic Asso-
ciation 2:2–3 (April–May 2004), 265–276.
terms are not in line with this explanation. Aghion, Philippe, Robin Burgess, Stephen Redding, and Fabrizio Zili-
botti, “Entry Liberalization and Inequality in Industrial Perfor-
mance,” Journal of the European Economic Association 3:2–3
V. Conclusions (April–May 2005b), 291–302.
“The Unequal Effects of Liberalization: Evidence from Disman-
In this paper we provide comprehensive empirical evi- tling the License Raj in India,” NBER working paper no. 12031
dence on substantial heterogeneity of productivity growth (February 2006), forthcoming in American Economic Review.
Aghion, Philippe, and Rachel Griffith, Competition and Growth: Recon-
and patenting reactions in incumbent firms to foreign firm ciling Theory and Evidence (Cambridge, MA: MIT Press, 2005).
entry. This corresponds to Schumpeterian growth theory Aghion, Philippe, Christopher Harris, Peter Howitt, and John Vickers,
suggesting systematic variation of incumbent innovation “Competition, Imitation and Growth with Step-by-Step Innova-
tion,” Review of Economic Studies 68:3 (July 2001), 467–492.
Aghion, Philippe, and Peter Howitt, “Appropriate Growth Policy: A
37 Labor productivity growth regressions are shown in table A.11 in the Unifying Framework,“ Journal of the European Economic Asso-
Web appendix. Corresponding TFP growth or patent count regressions ciation 4:2–3 (April–May 2006), 269–314.
provide similar insights. Aitken, Brian J., and Ann E. Harrison, “Do Domestic Firms Benefit from
38 Griffith et al. (2004) find empirical support for such consequences of Direct Foreign Investment? Evidence from Venezuela,” American
general spillovers looking across a panel of OECD industries and coun- Economic Review 89:3 (June 1999), 605–618.
tries. Griffith et al. (2002) find similar evidence at the establishment level Amiti, Mary, and Jozef Konings, “Trade Liberalization, Intermediate
in the United Kingdom. Inputs and Productivity: Evidence from Indonesia,” IMF working
39 The theoretical framework we rely upon generates predictions in line paper no. 146 (July 2005), forthcoming in American Economic
with both these results (Web appendix, sections 1.4 and 1.7). Review.
32 THE REVIEW OF ECONOMICS AND STATISTICS

Barro, Robert J., and Xavier Sala-i-Martin, Economic Growth, 2nd ed. Gompers, Paul A., and Josh Lerner, The Venture Capital Cycle (Cam-
(Cambridge, MA: MIT Press, 2004). bridge, MA: MIT Press, 1999).
Bernard, Andrew B., Jonathan Eaton, J. Bradford Jensen, and Samuel Greene, William H., “Accounting for Excess Zeros and Sample Selection
Kortum, “Plants and Productivity in International Trade,” Ameri- in Poisson and Negative Binomial Regression Models, “Stern
can Economic Review 93:4 (September 2003), 1268–1290. School of Business working paper no. 94-10 (March 1994).
Berry, Steven T., “Estimation of a Model of Entry in the Airline Industry,” Griffith, Rachel, Stephen Redding, and John Van Reenen, “Mapping the
Econometrica 60:4 (July 1992), 889–917. Two Faces of R&D: Productivity Growth in a Panel of OECD
Berry, Steven T., and Peter Reiss, “Empirical Models of Entry and Market Industries,” this REVIEW 86:4 (November 2004), 883–895.
Structure,” Yale University and Stanford University mimeograph, Griffith, Rachel, Stephen Redding, and Helen Simpson, “Productivity
(June 2006). Convergence and Foreign Ownership at the Establishment Level,”
Bertrand, Marianne, and Francis Kramarz, “Does Entry Regulation Hinder CEPR discussion paper no. 3765 (February 2002).
Job Creation? Evidence from the French Retail Industry,” Quar- Griffith, Rachel, and Helen Simpson, “Characteristics of Foreign-Owned
terly Journal of Economics 117:4 (November 2002), 1369–1413. Firms in British Manufacturing (pp. 147–179), in David Card,
Bertschek, Irene, “Product and Process Innovation as a Response to Richard Blundell, and Richard B. Freeman (Eds.), Seeking a
Increasing Imports and Foreign Direct Investment,” Journal of Premier Economy: The Economic Effects of British Economic
Industrial Economics 43:4 (December 1995), 341–357. Reforms, 1980–2000 (Chicago: University of Chicago Press,
Bloom, Nick, Rafaella Sadun, and John Van Reenen, “Americans Do I.T. 2004).
Better: US Multinationals and the Productivity Miracle,” NBER Haskel, Jonathan E., Sonja C. Pereira, and Matthew J. Slaughter, “Does
working paper no. 13085 (May 2007). Inward Foreign Direct Investment Boost the Productivity of Do-
Blundell, Richard, Rachel Griffith, and John Van Reenen, “Market Share, mestic Firms?” this REVIEW 89:3 (August 2007), 482–496.
Market Value and Innovation in a Panel of British Manufacturing Howitt, Peter, and David Mayer-Foulkes, “R&D, Implementation and
Firms,” Review of Economic Studies 66:3 (July 1999), 529–554. Stagnation: A Schumpeterian Theory of Convergence Clubs,”
Blundell, Richard, and James L. Powell, “Endogeneity in Nonparametric Journal of Money, Credit and Banking 37:1 (February 2005),
and Semiparametric Regression Models” (pp. 312–357), in 147–177.
Mathias Dewatripont, Lars Peter Hansen, and Stephen J. Tur- Javorcik, Beata Smarzynska, “Does Foreign Direct Investment Increase
novsky (Eds.), Advances in Economics and Econometrics: Theory
the Productivity of Domestic Firms? In Search of Spillovers
and Applications, Eighth World Congress, Volume II (Cambridge,
Through Backward Linkages,” American Economic Review 94:3
U.K.: Cambridge University Press, 2003).
(June 2004), 605–627.
Bresnahan, Timothy F., and Peter C. Reiss, “Entry in Monopoly Markets,”
Review of Economic Studies 57:4 (October 1990), 531–553. Keller, Wolfgang, and Stephen R. Yeaple, “Multinational Enterprises,
“Entry and Competition in Concentrated Markets,” Journal of International Trade, and Productivity Growth: Firm-Level Evi-
Political Economy 99:5 (October 1991), 977–1009. dence from the United States,” University of Colorado and Uni-
Caves, Richard E., “Industrial Organization and New Findings on the versity of Pennsylvania mimeograph (April 2007).
Turnover and Mobility of Firms,” Journal of Economic Literature Laffont, Jean-Jacques, and Jean Tirole, A Theory of Incentives in Procure-
36:4 (December 1998), 1947–1982. ment and Regulation (Cambridge, MA: MIT Press, 1993).
Criscuolo, Chiara, and Ralf Martin, “Multinationals and U.S. Productivity Lambert, Diane, “Zero-Inflated Poisson Regression, with an Application
Leadership: Evidence from Great Britain,” CEP discussion paper to Defects in Manufacturing,” Technometrics 34:1 (February
no. 672 (January 2005), forthcoming in this REVIEW. 1992), 1–14.
Disney, Richard, Jonathan Haskel, and Ylva Heden, “Entry, Exit and Melitz, Marc J., “The Impact of Trade on Intra-Industry Reallocations and
Establishment Survival in UK Manufacturing,” Journal of Indus- Aggregate Industry Productivity,” Econometrica 71:6 (November
trial Economics 51:1 (March 2003), 91–112. 2003), 1695–1725.
Djankov, Simeon, Rafael La Porta, Florencio Lopez-de-Silanes, and Nickell, Stephen J., “Competition and Corporate Performance,” Journal of
Andrei Shleifer, “The Regulation of Entry,” Quarterly Journal of Political Economy 104:4 (August 1996), 724–746.
Economics 117:1 (February 2002), 1–37. Nicoletti, Giuseppe, and Stefano Scarpetta, “Regulation, Productivity and
Dunne, Timothy, Mark J. Roberts, and Larry Samuelson, “Patterns of Firm Growth: OECD Evidence,” Economic Policy 18:36 (April 2003),
Entry and Exit in U.S. Manufacturing Industries,” RAND Journal 9–72.
of Economics 19:4 (Winter 1988), 495–515. Olley, G. Steven, and Ariel Pakes, “The Dynamics of Productivity in the
Fudenberg, Drew, and Jean Tirole, “The Fat Cat Effect, the Puppy Dog Telecommunications Equipment Industry,” Econometrica 64:6
Ploy and the Lean and Hungry Look,” American Economic Review (November 1996), 1263–1297.
74:2 (May 1984), 361–366. Pavcnik, Nina, “Trade Liberalization, Exit, and Productivity Improve-
Geroski, Paul A., “What Do We Know about Entry?” International ments: Evidence from Chilean Plants,” Review of Economic Stud-
Journal of Industrial Organization, 13:4 (December 1995), 421– ies 69:1 (January 2002), 245–276.
440. Tirole, Jean, The Theory of Industrial Organization (Cambridge, MA:
Gilbert, Richard J., and David M. G. Newbery, “Preemptive Patenting and MIT Press, 1988).
the Persistence of Monopoly,” American Economic Review 72:3 Wooldridge, Jeffrey M., Econometric Analysis of Cross Section and Panel
(June 1982), 514–526. Data (Cambridge, MA: MIT Press, 2002).