Pricing and Market Concentration in Oligopoly Markets

Pricing and Market Concentration in Oligopoly Markets
Author(s): Vishal Singh and Ting Zhu

Source: Marketing Science, Vol. 27, No. 6 (Nov. - Dec., 2008), pp. 1020-1035
Published by: INFORMS
Stable URL: https://www.jstor.org/stable/40057161
Accessed: 13-06-2020 22:27 UTC
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms
INFORMS is collaborating with JSTOR to digitize, preserve and extend access to Marketing
Science
This content downloaded from 132.174.254.252 on Sat, 13 Jun 2020 22:27:53 UTC
All use subject to https://about.jstor.org/terms
Marketing Science InfWH.
Vol. 27, No. 6, November-December 2008, pp. 1020-1035
issn 0732-2399 1 eissn 1526-548X 1 08 1 2706 1 1020 doi 10.1287/mksc.l070.0357
©2008 INFORMS
Pricing and Market Concentration in

Oligopoly Markets
Vishal Singh
Stern School of Business, New York University, New York, New York 10012, vsingh@stern.nyu.edu
Ting Zhu
Graduate School of Business, The University of Chicago, Chicago, Illinois 60637, tzhu@chicagogsb.edu
paper investigates the relationship between prices and market concentration in the auto rental industry.
We assemble an original database that includes the number of auto rental operators and other exogenous
demand and cost conditions at every commercial airport in the country. The data are interesting because we
observe a large variation in market structure, ranging from more than 100 monopoly and duopoly markets to
several competitive airports with more than eight firms. In addition, we collect daily rental prices in each market
that are regressed against the number of operating firms and other control factors. Due to potential biases in
treating market participants as exogenously assigned, we employ a two-stage estimation procedure in which
an equilibrium model of endogenous market structure provides correction terms for the second-stage price
regression. Results show that ignoring the endogeneity of market structure severely underestimates the impact
of additional competitors on prices, with the competitive interaction parameters doubling in magnitude after
the correction procedure. The downward bias in the competitive parameter can have important implications for
horizontal mergers, which may incorrectly appear innocuous when using a model that ignores the endogeneity
of market structure. More generally, our results serve as a warning on the potential biases to a large number of
applications in marketing and economics that attempt to relate outcome variables such as prices, markups, or
profits to the observed market structure.
Key words: pricing; market structure; entry; horizontal mergers; endogeneity

History: This paper was received July 14, 2006, and was with the authors 10 months for 2 revisions; processed
by Miguel Villas-Boas.
1. Introduction market concentration could be due to the competitive

A long stream of literature in economics and mar- of a few firms.
superiority
Over the past several decades, the profit concen-
keting strategy examines the relationship between
tration studies have been replaced by a stream of
competitive characteristics of market and profitabil-
research
ity. In the structure-conduct-performance paradigm of that examines the relationship between mar-
ketcross-
industrial organization, this literature relies on structure and prices rather than profits. An advan-
sectional data across industries to document the tage of using prices as opposed to profits is that they
impact of market concentration on profitability. are easier to obtain and are not subject to accounting
A general finding in this literature is that conventions.
higher Weiss (1989) provides a large number of
market shares and seller concentration are associ- price concentration studies and argues that because
ated with higher profitability (see, e.g., Buzzell prices
and are determined in the market, they are not sub-
Gale 1987, Schmalensee 1989). However, the profit
ject to superiority criticism, as is the case with profits.
concentration studies have been criticized on several Furthermore, the majority of the price concentration
grounds. First, these studies are plagued by studies use data across local markets within an indus-
mea-
try rather than across industries and incorporate
surement problems because, in general, accounting
profits are poor indicators of economic profits. more
Sec- industry- and firm-specific details. These stud-
ond, the cross-sectional data from different industries
ies include a wide range of industries such as grocery
(Cotterill 1986), banking (Calem and Carlino 1991),
used in these studies are problematic because of large
differences in demand and supply conditions acrossairlines (Borenstein and Rose 1994), hospitals (Keeler
industries. Finally, these studies are subject toet al. 1999), driving lessons (Asplund and Sandin
the
"efficiency" critique offered by Demsetz (1973), 1999),
who cable television (Emmons and Prager 1997), and
argued that positive correlation between profits movie
and theaters (Davis 2005).
1020
Singh and Zhu: Pricing and Market Concentration in Oligopoly Markets
Marketing Science 27(6), pp. 1020-1035, ©2008 INFORMS 1021
A general finding in this literature

In this paper, weis that
study high
the relationship between
concentration is associated with significantly higher
prices and the number of firms in the auto rental in-
prices (Weiss 1989; see also various
dustry. Westudies cited
develop an original in
data setathat includes
recent survey by Newmark 2004). However, as both
the number of car rental firms at every commercial
Bresnahan (1989) and Schmalensee
airport in the(1989) point
United States. The dataout
are unique in
that we
in their chapters in the Handbook ofobserve a wide cross-section
Industrial Organi-of market struc-
zation, price concentration regressions,
tures, ranging from such as those
several monopoly and duopoly
used in the literature, suffer markets
from serious endogene-
to many airports with more than eight firms.
ity issues. The fundamentalFor problem arises
each of these markets, because
we collect an extensive set
market structures are not randomly assigned,
of variables that captures the demand a nec- and cost con-
essary condition for a standard regression
ditions. model
These variables include to
both airport-specific
yield consistent estimates. Instead, the
factors (e.g., airlineobserved mar-
passenger traffic) and local demo-
ket structures are the outcome of strategic decisions
graphics in the city in which the airport is located
by firms that evaluate demand and
(e.g., cost
income, retailconditions
wages). In addition, aswe collect
well as potential competitors data
inonthe market in mak-
daily rental prices from every firm and for
ing their entry decisions. As each
such, there
car type (economy, are likely
full-size, to our data
etc.). Thus
be unobserved demand and cost shocks in a market
set consists of approximately 450 markets (airports)
that influence not only prices but also the underlying
for which we observe the number of firms that serve
market structure. For example, markets with unob-
each market and the prices they charge.
served high costs are likely to have higher prices, but
Our primary objective is to test how the prices
these markets are also likely to attract fewer entrants.
change with the number of competitors in the market.
In this case, a regression of prices on the number of
However, in doing so, we account for the endogeneity
firms may lead to the inference that high prices are
of market structure. Because we observe only a cross-
associated with a low number of firms, but this find-
section of markets, we cannot use the Evans et al.
ing may partially be driven by the unobserved costs.
(1993) approach, which requires a panel data struc-
Similarly, unobserved positive demand shocks may
ture with sufficient variation in the number of players
result in higher prices and an unusually large num-
over time. Such data, with substantial entry and exit
ber of firms in a market, in which case the impact
over time, are rarely available. In addition, even with
of competition on prices may be understated. Evans
a panel data structure, it is still necessary to worry
et al. (1993) formally address this issue and propose
a combination of fixed effects and instrumental vari- whether the changes in the error term for the price
regressions are correlated with the changes in mar-
able procedures that are applicable when panel data
are accessible. They study the price concentration ket in structure. As Manuszak and Moul (2007, p. 98)
the airline industry and find that the impact of con- point out, "the panel data approach shifts the concern
about correlation between unobservables and market
centration on price is severely biased using ordinary
structure to differences in those two items over time."
least-squares (OLS) procedures.
Recent structural work in marketing (see In this paper we use a two-stage estimation proce-
Chintagunta et al. 2006 and the accompanying com- dure to address the endogeneity of market structure.
ments) has pointed out the endogeneity problems In the first stage, we estimate an equilibrium model
associated with short-term marketing activities, such of entry that predicts the number of competing firms
as pricing and promotions in both aggregate (e.g., in a market. In particular, we follow the literature
Besanko et al. 1998) and individual panel data (e.g., advanced by Bresnahan and Reiss (1987, 1990, 1991)
Villas-Boas and Winer 1999). This literature uses and Berry (1992) that endogenizes competitive struc-
ture characteristics such as the number of firms and
variants of the methods proposed by Berry et al.
(1995) to examine issues such as brand value cre- the degree of concentration in the market. The key
ation and competitive advantage (Besanko et al. insight underlying this stream of research is that we
1998), product line competition (Kadiyali et al.can infer features of a firm's latent profit functions
1999), channel power (Kadiyali et al. 2000), retailerby observing its entry decisions because it will enter
pricing (Sudhir 2001, Villas-Boas and Zhao 2005, if it expects positive profits, but it will not other-
Villas-Boas 2007, Manuszak 2005), and price discrimi-wise. Because market entry decisions are discrete, the
nation (Chintagunta et al. 2003, Khan and Jain 2004).model suggests using a discrete-choice framework to
However, given the nature of the industry (mostly draw inferences about the factors that affect firms'
packaged goods) and relatively short time periods,actions. The approach parallels the standard single-
these studies take the existing market structure (e.g.,agent discrete-choice models in which the researcher
the number of firms in a market) as exogenouslymakes inferences about unobserved latent utility on
determined. the basis of the inequality restrictions. However, in
1022 Marketing Science 27(6), pp. 1020-1035, ©2008 INFORMS
the case of entry games, payoffswriting pads, and envelopes).

and resultant behav- The extent to which
an aggregate
ior reflect the interaction of decisions price index
of multiple or prices from a handful of
indi-
vidual agents. Thus, the estimation products is based
capture on an
competition in these industries is
oligopolistic equilibrium concept rather
subject than on an
to debate.
individual utility maximization. The auto rental industry that we analyze herein
In the empirical application, the parameter
overcomes the problemsesti-
associated with market defi-
mates from this first-stage entry model
nition and are used
price measurement to a large extent. First,
to derive correction terms that are inserted in the our data are complete in that we have collected
information on the number of firms and prices for
price equation to correct for the correlation between
the price errors and the market structure variables.every commercial airport in the country. Second, the
The procedure is similar to the two-step estimation market in this industry is reasonably well defined
used widely in labor econometrics (Heckman 1976). because the primary clientele for auto rentals at the
Mazzeo (2002a) uses this procedure in a similar con-airport locations is the passengers flying into the
text as this paper to study the relationship betweenairport (according to various firms' annual reports).
prices and market structure for motels. To check the
Although airport locations might compete with the
sensitivity of the correction procedure to the first-
outlets in downtowns or other city locations that we
stage model specification, we estimate three models do not consider, the problem is less severe than mar-
of endogenous market structure: (1) a model with- ket definitions in the previous studies. Third, the price
out firm heterogeneity, (2) a model with only unob- information is reasonably easy to collect and compare
served heterogeneity, and (3) a model with across both firms and markets because, at least at the out-
observed and unobserved firm heterogeneity. Aslet dis-
level, these are single-product firms and car rental
cussed below, the results are robust directionallyisand the primary business. Finally, the product is fairly
homogenous in that the quality of a particular car
in relative magnitude across all model specifications.
type tends to be similar across firms, although these
We apply this procedure to the auto rental industry,
which offers several advantages over previous appli- firms may differ in terms of service quality.
cations in this stream of research. As we discuss in The results from the first-stage model show that
the data section, we observe a wide range of market many demand and cost factors, such as airport traf-
structures, ranging from 68 monopoly marketsfic, and wages, and local demographics, are important
66 duopoly markets to more than 50 very competitive determinants in firms' entry decisions. In addition,
markets with more than eight firms. Such largeparameters
vari- of the latent payoff functions show that
ation in market structure provides a unique setting profits
to are higher in holiday markets and markets
study the relationship between prices and the num- with headquarters for large firms, whereas markets
with better public transportation infrastructure are
ber of firms. In addition, most previous price concen-
tration applications suffer from problems related lessto
profitable. Entry of additional competitors in the
market and product definition as well as price mea- market is found to reduce profits significantly. For
surement, which is much less of a concern in thethe cur-
pricing equation, several demand and cost con-
trol variables, service quality (in-terminal counters),
rent application. For example, consider the retailing
sector, such as grocery or office supply stores as and
stud-product quality (car size) are found to be sig-
ied in Cotterill (1986) and Manuszak and Moul (2007),
nificant. In terms of the primary focus of the paper,
we find that concentrated markets (those with fewer
respectively. In these industries, the market definition
and trading areas are difficult to define. The prob-than three firms) have prices that are approximately
lem is magnified because product assortments tend 30% higher than competitive markets (with more than
to overlap across a wide range of retail formats.eightFor firms). More important, we find that ignor-
example, supermarkets, discount stores, dollar stores,
ing the endogeneity of market structure in the price
price clubs, and supercenters all tend to haveregressions
sig- severely underestimates the competition
nificant overlap in product assortment, which makes parameters. In particular, the competitive interaction
the industry and the resultant competition difficultparameters
to are doubled in magnitude when we insert
define. In addition, retailers are multiproduct firms
the correction terms in the price equation. Thus, the
that carry thousands of products, which makesimpact the of market concentration on prices appears
collection and comparison of prices for all productsless severe if the endogeneity of market structure is
difficult, if not impossible. The common approach ignored.
to In the empirical analysis, we test the sensi-
deal with the problem is to create some aggregate tivity of the results to various first-stage model spec-
ifications.
price indexes or to rely on the prices from a handful of The direction and magnitude of bias is
products. For example, Cotterill (1986) uses an aggre-
consistent across model specifications and similar to
gate price index for grocery products, and Manuszak those in Evans et al. (1993), who find that OLS param-
and Moul (2007) use data from three products (paper,eters are biased in the magnitude of 150%.
The bias in the parameters capturing

at the airports, such the com-
as the number of passengers fly-
petitive interactions underscores
ing intothe fundamental
the airport, local demographics, and other
problem in the vast literature controls,
thate.g., uses regression
firm and car size fixed effects. Because
we do not observe market
technique to understand the relationships shares in our data, we can-
between
variables such as prices, markups, revenues,
not use concentration ratio or theor
Herfindahl index.
profits and market structure. In addition,
Instead, we capture theit has
competitive structure in our
important policy implications because
application price the
by including con-
total number of car rental
centration approaches are commonly used
firms that offer econo-
services at the airport. In the empir-
metric techniques employed byical the Federal
application, we alsoTrade
run models using a flexible
Commission (FTC) to analyze horizontal
dummy specification mergers
for monopoly markets, duopoly
(Whinston 2003). As Baker andmarkets, and so forth.(1999,
Rubinfeld
p. 405) note, "[R]educed form priceThe price equation in
equations are(1) represents
the a typical model
workhorse empirical methods for used in the pricelitigation."
antitrust concentration literature (see, e.g.,
variouswere
For example, such price regressions studiesused
in Weiss
in 1989).
the As is well known,
the OLS
highly litigated case on the merger estimator Staples
between applied to such a model is bi-
ased and inconsistent
and Office Depot, which was eventually declined if thebyunobservable epmk are
correlated
the FTC. Using a similar two-stage with explanatory
approach as in this variables in the regres-
paper, Manuszak and Moul (2007) sion. Of particular
revisit this concern
case and in (1) are the variables
that capture
point out the bias in price regressions the competitive
employed bystructure, Nm, because
therethe
the FTC. In the current application, are likely to be unobserved
negative rami- demand and cost
fications of market concentration are underestimated,
conditions in a market that influence not only prices
implying that horizontal mergers but alsointhe number
this of firms that operate in the
industry
may incorrectly appear innocuous market. For example,
using a model markets
that with unusually high
ignores the endogeneity of marketcostsstructure.
are likely to have higher prices, but these mar-
kets are as
The rest of the paper is structured alsofollows:
likely to attract
In fewer entrants. Simi-
the next section we describe thelarly,
price unobserved positive or
regression, negative demand shocks
the
first-stage entry model, and the can influence a procedure.
correction firm's pricing and its decision to
In §3, we discuss the data from theoperate
auto in the market.
rental market.For instance, suppose that
f(Nm,8p)
Section 4 presents the results; the = 8pNm and that
implications of Nm
the and Zmk are not per-
main findings of the paper, along with directions for
fectly correlated; then, the OLS estimator of 8p is 8p =
future research, are discussed incov(Nm,
§5. lnpwjk)/var(NJ = 8p + cov(Nm, epmk)/var(Nm).
If there are unobserved demand shocks that influence
2. Model prices and the number of firms in the same manner

such that cov(Nm, epmk) > 0, the parameter estimate 8p
2.1. Price and Market Concentration
will be biased downward (i.e., 8p > 8p/ or \8p\ < \8p\
Consider a typical price concentration regression
because we would expect 8p to be negative).1
model in which the relationship among the prices, ex-
A possible solution to this econometric problem is
ogenous market characteristics, and the markettostruc-
use instrumental variable techniques. For example,
ture variables can be specified as follows: there could be variables that affect firms' long-term
]npmk = Zmke + f(Nm,Sp) + ef'mk, (1)

1 The atheoretical nature of the typical price regression makes it
difficult to infer the nature of biases that may occur. For instance,
where pmk are the observed prices in marketconsider
m fora simple cournot model with linear cost and demand:
firm k, and Zmk are all observed firm and market char-
P(Q) = a-Q and C,-(<fc) = c<fc + F, for all i.
acteristics except the market structure variables that
affect prices. The function f(Nm,8p) represents theare n firms in the market, the equilibrium price is p* =
If there
impact of the underlying market structure ona prices.
- no* = (a + nc)/(n + 1). With free entry, the number of firms is
determined by II* = 0, and we obtain n* = (a - c)/Vf - 1. Note
In empirical applications, the market structure vari-
that if the variation in the number of firms is determined by some
ables are typically captured using measures unobserved
such as factors in a, then in the price regression, the endogene-
concentration ratio or the Herfindahl index. Finally,
ity effect would be in the direction of greater n being associated
epmk are firm-market-specific unobservables thatwith
influ-
higher price, whereas the opposite is true if the variation in
ence prices. n is determined by some unobserved factors in c. If the variation
in n is determined by variation in only F, then there would not be
In the current context, our dependent variable, pmk,
any endogeneity effect. The empirical results in this paper seem to
would be the price of a particular car type (e.g., econ- show that in the particular industry considered, the endogeneity
omy) at each airport location, and the exogenous vari- effect due to the demand shocks (a) seems to be most important.
ables Zmk would include demand and cost conditions We thank the associate editor for this insight.
Assume
entry decisions but do not affect the that firm
short-term k's latent profits in mar
prices.
In our study, we use a variation
with N ofentrants
instrumental
(including itself) can be spe
follows:
variable approach by first estimating a model that
directly describes the determinants of market struc-
ture. Formally, this model considers the observed
market structure as the outcome of a strategic game
= Xmk(3 + 8(N-l) + emk, (2)
between potential entrants. Estimates from the first-
where
stage model are then used to derive correction terms
that are inserted in the price equation to emk = yum0 the
alleviate + <rumk. (3)
correlation between the price errors and the market
In the profit function (2), 7f%k(Xmk, j8, S) captures the
structure variables. The procedure is payoffs
expected similar as ato the of exogenous demand
function
two-step estimation used widely in labor economet-
and cost shifters as well as the number of competi-
rics (Heckman 1976). Mazzeo (2002a)
tors in the and
market.Manuszak
The term 8(N - 1) captures how
and Moul (2007) use this procedure
a firm's profit decreases con-
in a similar as more competitors enter
text as ours to study the relationship
the market.2between
The vector prices
Xmk represents the observed
and market structure for motels, and of
component office supply
the profit function and includes mar-
stores, respectively. ket characteristics that are common across players as
well as firm-specific observed components. Note that
2.2. Model of Endogenous Market Structure
several variables in Xmk describing firms' entry behav-
To model the number of car rental
ior can companies oper-
be the same as those that affect the prices;
ating at an airport, we follow the literature on multi-
that is, Zmk in (1). When all exogenous variables in (1)
agent discrete games, which provides an (i.e.,
and (2) are identical empirical
Xmk = Xm = Zmk), identifica-
approach to analyze game-theoretic tion of models in procedure
the correction which we describe subse-
agents make discrete choices, such as entry
quently is through a nonlinearandfunctional form of the
exit (for a survey, see Reiss 1996). The approach
entry model alone. However, our is empirical applica-
advanced by Bresnahan and Reiss (BR) (1987, 1990,
tion includes several variables in the entry model that
1991) and Berry (1992). The authors are not includeddevelop
in the econo-
price regression. For exam-
metric models to investigate how ple, variables such asof
the number firms
annual population growth, real
varies across markets as a result of various demand
estate value, traffic in past years, and distances to
and cost factors. The approach endogenizes the com- firms' headquarters are likely to affect market struc-
petitive structure in the market by implicitly ana- ture, but not short-term pricing, and serve as exclu-
lyzing the first stage of a two-stage game in which sion restrictions.
firms first decide whether to enter and then deter-
The second component in (2), emkf summarizes
mine price or quantity competition. The key insight the unobserved characteristics. In (3) we decompose
underlying the research is that features of latent prof-
these unobservables into two terms, where um0 rep-
its can be inferred by observing firms' entry decisions
resents unobserved market characteristics that are
because they will enter if they expect positive profits,
common across all players, and umk is firm-specific
but they will not enter otherwise. unobservables. Thus, this model allows for hetero-
Unlike the structural models of supply and demand geneity across firms due to firm-specific observed
that provide marginal conditions, the discrete deci- components in Xmk and unobserved heterogeneity
sions imply threshold conditions for players' unob- across firms. The terms umk and um0 are assumed to be
served profits. Econometrically, this feature of the independently distributed with standard normal dis-
model suggests using a discrete-choice framework to
tribution across firms and markets. For identification,
make inferences about the factors that affect firms'
we impose the traditional constraint that the vari-
actions. However, unlike the single-agent discrete-
ance of emk equals one, via the restriction a = y/l - rj1,
choice models that have been used extensively inwhere
the 17 is the correlation of the unobservables emk in
marketing literature to model consumer choice, ina the
given market.
case of a discrete game, payoffs and resultant behav-
ior reflect the interaction of multiple agents' decisions.
2 Note that there is a structural relationship between 8 in the
Thus, estimation is based on an oligopolistic equi-
entry model and 8p in price regression that we do not model.
librium concept rather than on an individual utility
For example, if demand is linear and firms compete in quantities
maximization. In what follows, we build on the exist-
(as in the example presented in Footnote 1), (dU* / dn) / (dp* / dn) =
2(fl - c)/(n + 1), where n is the number of firms in a market. It
ing literature to develop a model that describes firms'
would be nice to know the conditions under which there would
entry decisions. The model is general and includes
be a structural model that would be consistent with both the price
simpler setups of no firm heterogeneity as special
and profit equations used in the paper. We would like to thank the
cases. associate editor for pointing this out.
To identify the parameters of a the

closed-form solution. Berry
model, we (1992) proposes a simu-
impose
an equilibrium assumption thatlation
allmethod
active to solve firms
this problem, expect
which involves
nonnegative profits and that defining
an additional
a prediction error as theentrant
difference between
the unprofitable.3
would find entry into the market observed number of firms andIn the expected
the num-
empirical application, we estimate three
ber of firms E(N \Xm; j8,models with
8, 17) from the model:
different specifications on firms' payoff functions:
vm = Nm-E(N\Xmipt8tr,).
(1) no firm-specific heterogeneity, (2) unobserved firm (6)
heterogeneity, and (3) observed and unobserved firm
By construction, vm is mean independent of the exoge-
heterogeneity.
nous data when it is evaluated at the true parameter
2.2.1. Symmetric Firms (emk = um0; Xmk = Xm;
values:
ir%k = tt^ for All it). Bresnahan and Reiss (1990)
describe a special case of the model E[i/JXM;j8
(2) that= assumes
j8*, 5 = 5*, r/ = r/*] = 0. (7)
symmetric firms. In this framework, the equilibrium
Given the condition,
assumption implies that the probability either nonlinear regression or
of observing
n active firms in a market is generalized method of moments (GMM) can be
used to estimate the model, where E(N|Xm;/3,5)
P(Nm = n) = Pr« > 0 and <+1 < 0) is obtained using simulation. In particular, given
S draws for the underlying random variables and
= Pr(-<<Wm0<-7fr1)
guesses of j8, 8, and rj, we can construct the equi-
= F(-7f»m+1)-F(-<), (4) librium number of firms N(Xm, 5, j8, us0, u\, ... , usR),
where us0 is the unobserved market characteristics,
where F(.) is the cumulative distribution function
u\, ... ,us^ represents the firm-specific unobservables,
of um0. If a firm does not enter a market, its payoff
and in
N is the number of potential entrants in the mar-
that market is normalized to zero. Therefore,
ket. The estimate of E(N | Xw; j8, 8) is
P(Nm= 0) = PrK<0)
E(N\Xm;Pf8) = WN(Xm/8/p/r1/us0/u\
= F(-«*).
Although restrictive, the model generates closed-form In our application, we use this "frequency estima-
solutions for the probability of each market configu- tor" to construct the predicted number of entrants
ration and is straightforward to estimate. for each market and use nonlinear regressions to esti-
mate the parameters that minimize the distance bet-
2.2.2. Unobserved Firm Heterogeneity (emk ^ smk,
ween the predicted and the observed number of firms
if k^ k'; Xmk = Xm for All k). In this specification, in the data.
each firm is allowed to be different but only because
of unobserved factors. Similar to the previous model, 2.2.3. Observed and Unobserved Firm Hetero-
no firm-specific observed information is incorporated geneity (smk ^ emk,; Xmk f Xmk, for k # k'). This final
in the model. This model is able to predict specification only allows for firm-specific observed charac-
the number of entrants in the market (but not teristics their to enter the payoff functions. For example,
identities): we may expect firms to enter markets that are closer
to their headquarters with a higher probability. Sim-
Nm = max(n: #{/c: irmk{nf smk)}) > 0. (5)
n ilarly, firms targeting a specific segment - say, leisure
travelers - may be more likely to enter holiday mar-
Because the number of entrants is uniquely deter- kets. Besides complication in estimation, the cost of
mined, we can base an estimation strategy on this allowing such firm heterogeneity is that there might
unique number of firms. However, even with this be multiple equilibria in a simultaneous-move game.
simple form of firm heterogeneity, we no longer have We follow Berry (1992) and assume that firms make
decisions sequentially and that the order of moves is
3 The literature on entry games focuses on pure strategy Nash equi-determined by firms' profitability (i.e., the firm with
libria for simplicity. However, it may be important to allow forthe highest profit moves first). Given the sequential-
mixed-strategy equilibria when making inferences about the payoffmove assumption, we can construct moment condi-
structure of a game in certain applications. For instance, supposetions on the basis of the identities of entrants in the
there are two potential entrants in a market. If we observe that
market:
both firms choose to enter, we may infer that the competition effect
is small or the market conditions are good. However, with mixed-
strategy equilibria, the entry by both may merely be the result of Z[Vmk I Xmit/ )8 = )S*, 5 = 5*, 77 = 77*1 = 0, where
a randomization by firms. Therefore, ruling out the mixed-strategy
equilibria may underestimate the competition intensity. *mk = Vmk ~ E(ymk I Xmk; )8 = /3*, 5 = 5*, 77 = 77*). (9)
Note that the observed marketdata Xmk and with

structure ymM independent
is now markets; the sam-
ple analog is
an N x 1 vector, where N is the number of potential
entrants in the market. Each element of ym, ymk, is an
indicator variable that equals 1 if firm k enters
(*m\ the
market m and 0 otherwise. The term £(ym | Xmk; j3, 8)
is the predicted market structure. With M indepen-
dent markets, a sample analog of the GMM estimation
procedure based on the derived moment restriction
that vmk is uncorrelated with the exogenous data
Xra,is
W/
is the frequency estimator of the prediction errors. An
1(0) = ^£ . ®x»*' <10)
tn • estimate of 6 = (j8, 8, fj) is chosen to minimize
\VmN / G(0) = g(0)'Ag(0).
and the estimate of 6 = (/§, 8, fj) is chosen to minimize Following the traditional method of moments tech-
nique, we first set A equal to the identity matrix and
then update it using the sample variance of the indi-
G(6) = g(0yAg(0)> (11) vidual moment conditions.
where A is the weighting matrix. 2.3. Correcting for Endogenous Market Structure
This model discussed above is quite general, inallow-
the Price Equation
ing for firm-specific profit functions, and estimated
Having can be the parameters from the latent pay-
directly applied in industries with a fewoff firms. For we derive correction terms that will be
functions,
example, Manuszak and Moul (2007) study office
inserted insup-
the price regression to correct for the po-
ply stores and focus on three firms: Staples, tentialOffice
correlation between the price errors and the
Depot, and OfficeMax. Similarly, Zhu market et al. structure
(2008) variables. More formally, we specify
analyze entry decisions in the retail discount the correlation
indus- of the error terms for the prices and
try and explicitly model firm identities for the entry payoffs as follows:
Wal-Mart,
Target, and Kmart. However, in the current applica-
tion, we have a total of 17 distinct firms. Restricting
attention to national players and lumping all regional
firms as one still leaves us with nine firms, which
where epmk and um0 represent the error terms from
makes the estimation cumbersome. Our approach to
the price and entry models, respectively, and p is
deal with the problem is to classify the eight national
the covariance between the two. With normally dis-
firms into either "business" or "leisure" on the basis
tributed error terms, the conditional distribution
of their stated primary business segment from their
of epmk given um0 is also normal, and the mean is equal
annual reports. In addition, all smaller regional play-to pum0. Note that um0 is not observable, but we are
ers are classified as "other." Thus, our empirical appli-
able to infer the range of um0 given the observed mar-
cation involves three estimation equations for ket
eachcharacteristics and the market structure at mar-
market:
ket m. We can represent the expectation of the error
term in price regression by using iterated expectation
v°±NZ-E(N°\Xmk;fi,8,r,), as follows:
vBm±N*-E(NB\Xmk;P,8,V),
E[e"mk\Xmk,Nm] = pE[um0\Xmk,Nm].
vLm±K-E(NL\Xmk;p,8,r,),
Given the observed number of firms Nm, we can
specify the price regression as follows:
where v% is the prediction error of "other" entrants,
and vBm and vLm are prediction errors of the num-
lnpmt=Zm^+/(Nm/8p)+p£[Mm0|Xm,,Nm] + I;pmt, (13)
ber of entrants targeted at business travelers and
leisure travelers, respectively. The GMM estimation
where tfmk = evmk - E[epmk \ Xmk,Nm] is now the
procedure is based on the derived moment restriction
pure idiosyncratic error term affecting prices. In the
empirical application, we insert the estimates of
that (i?°, v*m, v^Y are uncorrelated with the exogenous
high (H) or low

E[um0 \Xmk,Nm] from the first-stage (L). In model4
entry addition, there
in are market-
the price regression such that the unobservables
specific covariance um0 between
that impact firms' entry
um0 and epmk, p, becomes an decisions and are assumed
additional to take one of two
parameter tovalues:
be estimated. 0 or -1. Suppose that the observed prices, market
With no firm heterogeneity, the expression ofstructure, traffic, and unobservables from the three
markets are as follows:
E[um0 | Xmk, Nm] has a closed form and can be easily
obtained as follows:
Number
Market Traffic of firms Unobservable um0 Price
E[UmolNm,XJ = /_^\^^}ds
All 0 pA
_ 4>(^")-<K^+1) ( n,. }
~<D(^)-<t>(^+Y _ ( n,. } B
C
H 1 -1 pB
H 2 0 pc
We can derive a similar error correction term from
the entry model that allows for firm heterogeneity. Here, Market B h
Note, however, that with firm heterogeneity, there isenters this marke
no closed form solution for E[um0 \ Xmk, Nm] so that we (um0 = -1). These
rely on simulation techniques. In particular, givencost factors (e.g.,
a set of parameter values (j8, 8) from the first-stage(e.g., good public
entry model and a set of random draws capturinglikely to deter ent
the unobservables, we solve for the equilibrium mar- shocks could influ
ket structure in market m. Let <orm denote the solvedever, note that th
equilibrium (i.e., the predicted entry decisions) for thebe different on m
rth draw. We repeat the procedure for R sets of ran-ing on whether th
dom draws to obtain a consistent estimate of the con- For example, nega
ditional expectation of the market unobservables: entry and lower p
lower probability o
ure 1 illustrates th
E[um0 | Xmk, Nm] = ^ AA r=l
£ urmOl(a>rm = ym \ Xmk; j§, 8, t)),
(15) Figure 1 Illustration of
where ym is a variable representing the observed entry
decisions of the firms, and l{(orm = ym \ Xmk; J3, 8, 17)
is an indicator function that equals 1 if the predicted
entry outcomes from the rth draw are consistent with
the observed outcome and 0 otherwise. This simulator
yields an estimator that is consistent as the number of
simulations grows large.5
The two-stage procedure described above is rela-
tively straightforward to account for the endogeneity
of market structure in regressions attempting to study
the relationship between prices or profits and mar-
ket concentration. To illustrate how the competitive
interaction parameters may be biased by ignoring the
endogeneity of market structure, consider the follow-
ing simple example: Suppose our sample consists of
three airport markets for which we observe the num-
ber of car rental firms and corresponding prices. For
simplicity, assume that the only observed market con-
dition in our data is airport traffic, which is either
4 Note that if a firm's payoff is not af

entry decisions, i.e., 8 = 0 in the first-stag
reduces to the Heckman-type single-agen
5 An additional concern when constructi
parameter estimates in price regressions
about the uncertainty around our first-s
we rely on a bootstrap technique.
correlation, which is also summarized

outlets inin the follow-
downtowns or other city locations that we
ing table. The correlation between the
do not unobserved
consider, fac-
the market definition is significantly
tors influencing prices and entry is represented by p,
cleaner than previous applications (e.g., in retailing
or banking industries).
and 5true and 0true represent the competition Finally, the price informa-
and traffic
parameters in the price regression. When
tion in the is
this industry correla-
also reasonably easy to col-
tion is positive, both 5true and 0^ are underestimated,
lect and compare across firms and markets because
whereas the opposite is true for these
p < are
0. single-product firms, with car rental being
the primary business. However, we should point out
True parameter Biased that these firms also generate revenues by providing
values estimates p > 0 p < 0 ancillary products and services, such as supplemen-
tal equipment (e.g., child seats, ski racks, cell phones,
Competition 5true = pB, - pc 8 = pB- pc 5true > 8 8tTue < 8
effect navigation systems), insurance, and gasoline payment
options, for which we do not have any information.
Sensitivity 0true = ^-^ H~L
to traffic H~L 6 = ^-^
H~LH~L 0true > d 0true < 6 Nevertheless, the problem is not as severe as with
multiproduct operations such as supermarkets and
hospitals.
3. Data Description Our data collection strategy relies on various
Our application in this paper is to the
sources. First,auto rental
we use the list of commercial airports
industry, which began as early as 1918
provided with
by the small
Federal Aviation Administration.
operators providing car rentals to people in local
This list consists of themar-
airport codes, full names,
kets. Currently, the industry is fairly consolidated,
and addresses (city, state, and zip codes) for all com-
with eight major firms and a hostmercial
of regional oper-including Alaska and
airports in the country,
ators that conduct their operations through
Hawaii. For each of a com-
these airports, we use the avia-
bination of company-owned andtion
licensee-operated
data provided by the Bureau of Transportation
locations. These firms serve both business and leisure Statistics to collect information on the air traffic vari-
travelers, although different firms may focus more ables. These include measures on the total number of
on a particular segment. These firms offer services inpassengers and information on the number of major
local markets as well as commercial airports. The local
airlines flying into the airport. Finally, we use the
city locations typically target individuals who need aairport addresses to collect information on various
vehicle for special occasions, insurance replacements,demand and cost factors using data from the U.S.
or repairs, whereas the airport locations are primarily
Census Bureau, the Bureau of Labor Statistics, and the
geared toward business and leisure travelers. With theCOMPUSTAT.
deregulation in the airline industry, the number of air- The full list of variables describing each airport
port locations has grown dramatically over the years, appears in Table 1. The first set includes the
now accounting for a significant proportion of rev-standard variables that capture the demand (e.g.,
enues for most companies. For example, Hertz gener-population) and cost (e.g., wages) conditions in a mar-
ates 90% of its revenues from its airport locations inket. The variable "Traffic_2004" is the total number of
the United States (2003 annual report). In this study,
passengers flying into the airport in 2004. Because the
we focus on the airport car rental locations, for which
primary clientele for car rentals at the airport loca-
we develop an original database that consists of three
tions tends to be passengers flying into the airport,
major pieces of information: (1) a list of all domes-
we expect this variable to play a major role in deter-
tic commercial airports and the exogenous variables
mining the number of car rental firms that operate.
describing each airport, (2) the number and identities
of all car rental companies operating at those airports,
and (3) the rental prices for each car type (e.g., econ-
Table 1 Summary Statistics
omy, full-size).
Variables Mean Std. dev.
Given the three pieces of information, the auto
rental industry seems quite suitable for the purposePopulation 412,269.97 966,758.39
of this study for several reasons. First, our data arePop. growth 0.14 0.16
complete in that we have been able to collect infor- Poverty ratio 0.13 0.05
Retail wages 21,800.54 3,211.57
mation on the number of firms and prices for everyHousing value 79,350 49,777
commercial airport in the country. Second, our focusTraffic_2004 1 ,267,1 36.46 3,633,1 97.91
on only airport locations makes the market definitionNo.ofHQ 6.69 19.53
reasonably clean because these locations tend to tar- Pub_Ratio 0.02 0.05
Holiday 0.09 0.29
get customers who fly into the airport. Although to a Hub 0.08 0.27
certain extent the airport locations compete with the
Table 2 Airport Traffic and Prices by Market Structure
Market structure N Traffic Economy Compact Mid Standard Full
No firms 19 2,637 na na na na na
Monopoly 68 6,401 45 49 53 56 61
Duopoly 66 16,927 43 45 49 52 56
Oligopoly-3 49 38,656 41 43 47 51 53
Oligopoly-4 33 60,582 40 42 45 50 51
Oligopoly-5 33 146,821 37 39 44 47 49
Oligopoly-6 49 731,332 35 37 42 46 48
Oligopoly-7 37 702,145 34 36 40 44 45
Oligopoly-8 53 2,747,626 35 36 41 45 47
The variable "No. of HQ"In is Table

the number 2, we display of large the cor-av
porations (with annual sales car type over under $100 million) different that comp
have their headquarters located
patterns in are the apparent
market, and from the th
variable "Pub_Ratio" is the shown percentage in Table of 2. The price
population
in the market that uses from public transportation
economy to full-size, to getand p
to work. This variable servescompetition. as the proxy However, for public in our
transportation infrastructurepersion in in a market,
prices within whichmark can
the need for
be considered as a substitute other control
for rentingvariables. We cars.
discuss theseFinally,
issues further
"Holiday" represents holiday in the next section. such as Miami,
airports,
Las Vegas, and Atlantic City, and "Hub" is a dummy
variable that indicates whether the airport is a hub of
4. Results
any major airlines.
Our next major piece of Wedatanow present the resultsthe
includes from number
the model and data
and identities of all car rental companies thatfrom
discussed above. We provide estimates the first
oper-
stage entry model,
ate at each airport. We collected this information fromfollowed by the price regressions,
Orbitz and Expedia websitesand then aswe well
discuss theas implications
from individ- of the bias when
treating market structure
ual firms. In Table 2, we report the observed market variables as exogenous.
structures in this industry, along with a distribution
4.1. First-Stage Entry Estimates
of airport traffic for each market structure. We use a
We rely on the variables outlined in the data section
total of 407 airports in the analysis and exclude air-
to capture the factors that affect firms' entry deci
ports from Alaska and an additional nine markets
sions. Profits and entry costs at any airport location
with missing variables. As Table 2 shows, we observe
depend on a large number of factors, including agree-
a wide range of market structure in this industry,
ments between car rental firms and airport authori
ranging from more than 130 airports with either a
ties through negotiations and /or bidding processes.
monopoly or duopoly to more than 50 very compet-
These agreements provide for concession payments
itive markets with eight firms. We can also observe
that the number of firms based upon a specified
increases percentage
with of revenueas
demand, gener-
ated at the airport, subject to a minimum annual fee,
measured by the airport traffic.
and often include fixed rent for terminal counters or
The final piece of information in our database is the
other leased properties and facilities. Unfortunately,
daily rental prices of the cars at all the airports. The
we do not observe many of these factors and rely pri-
prices are for the rental period from 10 a.m. Monday,
marily on the observed market characteristics.
March 21, 2005, to 10 a.m. Tuesday, March 22, 2005.
We estimate three alternate models to describe the
The price information was collected on Monday,
factors that affect firms' entry decisions: (1) no firm
March 14, exactly one week in advance.6 We collected
heterogeneity, (2) only unobserved heterogeneity, and
the prices on five popular car types: economy, com-
(3) observed and unobserved firm heterogeneity. The
pact, mid-size, standard, and full-size; we excluded
left column of Table 3 presents the parameter esti-
special vehicles, such as minivans and SUVs, because
these tend to be availablemates based
in on the modelathat
only ignoresmarkets.
few firm hetero-
geneity and is similar to the model considered in
BR and subsequent applications (e.g., Manuszak and
6 Note that the actual prices paid might not be the same as the
Moul 2007). Most of the parameter estimates seem
posted prices, particularly for corporate clients that may have spe-
reasonable and in the correct direction. The profitabil-
cial contracts with individual firms. We repeated our entire exercise
using weekend rental prices (with probably few corporate travel- ity of a market is increasing in market size, which
is captured by annual airport traffic, population in
ers), and the findings are similar to those reported in the paper.
Table 3 Parameter Estimates from Entry Model
With heterogeneity
With unobserved With unobserved and

Without heterogeneity heterogeneity only observed heterogeneity
Variable Parameter est. Std. error Parameter est. Std. error Parameter est. Std. error
Intercept 2.674 0.182 3.025 0.155 0.695 0.170

No.ofHQ 0.120 0.045 0.121 0.041 0.155 0.047
Holiday 0.740 0.231 0.743 0.212 0.726 0.216
Hub 0.117 0.276 0.099 0.257 0.184 0.261
Poverty ratio 0.078 0.060 -0.068 0.058 -0.008 0.070
Pub.Ratio -0.305 0.070 -0.306 0.067 -0.284 0.044
Log(Traffic_2004) 1.757 0.097 1.647 0.082 1.577 0.090
Population growth 0.147 0.060 0.100 0.058 0.103 0.046
Log(Population) 0.406 0.096 0.417 0.093 0.343 0.069
Retail wages -0.136 0.086 -0.143 0.082 -0.119 0.094
Housing value -0.034 0.071 -0.038 0.068 -0.052 0.079
Midwest -0.042 0.173 -0.069 0.170 -0.061 0.172
South 0.265 0.169 0.309 0.165 0.455 0.252
West 0.099 0.1715 0.160 0.167 0.117 0.179
Observed heterogeneity
Business 1.087 0.123
Business x no. of HQ 0.093 0.094
Leisure 0.141 0.039
Leisure x holiday 1.786 0.811
Dummy_carHQ 0.224 0.106
Number of firms (8) -0.702 0.0273 -0.650 0.016 -0.346 0.024
Correlation (rj) 0.399 0.027 0.506 0.014
the market in which heterogeneity.

the airport The resultsisare more or less consis-
located, and
population growth rate tentinwiththat
those we market.
discussed previously.
The However,
local
we estimate
ulation, airport traffic, and onepopulation
more parameter, ry, that captures
growth r
how the unobservedimpact
have a statistically significant component in firms'
on profitabil-
demand
ity is correlated
capture the costs of entry, we withuse
each other and is found
retail to be
wages,
estate prices in the market, and regional shows
highly significant. The right column of Table 3 dumm
the results ofdifferences
that capture any systematic the entry model with both inobserved
fixed c
and unobserved
across airports in different parts heterogeneity
of the discussed in §2. UnlikeFir
country.
entry profits decrease the other two
with models, this
retail wagesspecification
and, allows for
given
firm-specific observed
same market characteristics, are higher in airporvariables to enter the latent
profit function. In particular, using the firms' annual
the South. The estimate for number of HQ of la
reports, we create dummy variables for "Business" or
companies is positive, suggesting that the presenc
"Leisure" depending on whether a particular firm pri-
headquarters captures attractiveness of market co
marily serves business or leisure clientele. These two
tions beyond what local demographic characteris
variables also interact with "No. of HQ" and "Holi-
capture. Similarly, the positive coefficient for the H
day" variables discussed above. Finally, we create a
iday dummy suggests that profitability is highe
dummy variable for all airports that are located in
holiday markets. Finally, the estimate for Pub_R
the state or in surrounding states in which the firm's
a variable we include to capture the convenience
headquarters is located.
public transportation, Comparing
has a the strong negative
estimates from the middle columnimp
which is intuitive because a good public transpor
with the results from the full model, we find that most
tion infrastructure can be considered as a substi-
of the parameters are more or less comparable in both
tute for renting cars. The estimate of the competition
direction and magnitude. On average, the business-
effect, 5, indicates that the number of rivals in type
the firms earn higher profits because the parame-
market is an important determinant of profitability,
ter estimate for the business dummy is positive and
because entry by additional competitors reduces larger
prof- than the estimate for the leisure dummy. Exam-
its significantly. ining the interactions between market conditions and
The middle column of Table 3 shows the results
firm types, we find that leisure-type firms benefit
of the entry model that only allows for unobserved
more if they are present in holiday markets, which is
Table 4 Pooled Estimation Results from Price Regression
Estimates without correction Estimates with correction
Model 1 Model 2 Model 3 Model 4
Variable Parameter Std. error Parameter Std. error Parameter Std. error Parameter Std. error
Intercept 3.651 0.017 3.475 0.025 3.562 0.022 3.188 0.044

LogfMarch traffic) 0.019 0.002 0.022 0.002 0.040 0.004 0.043 0.003
PopulationD 1.154 0.262 1.124 0.263 1.585 0.270 1.509 0.266
Retail wages 0.010 0.004 0.011 0.004 0.013 0.004 0.014 0.004
Pub.Ratio 0.007 0.005 0.005 0.005 -0.009 0.005 -0.010 0.005
No.ofHQ -0.006 0.003 -0.006 0.003 0.002 0.003 0.003 0.003
Poverty ratio -0.001 0.003 -0.001 0.003 0.000 0.003 -0.001 0.003
Holiday -0.081 0.008 -0.076 0.009 -0.052 0.010 -0.046 0.009
In-terminal 0.059 0.007 0.056 0.007 0.060 0.007 0.057 0.007
Economy -0.148 0.007 -0.148 0.007 -0.147 0.007 -0.147 0.007
Compact -0.103 0.007 -0.103 0.007 -0.103 0.007 -0.103 0.006
Standard 0.092 0.007 0.092 0.007 0.092 0.007 0.092 0.007
Full-size 0.115 0.006 0.115 0.006 0.115 0.006 0.115 0.006
Midwest -0.020 0.007 -0.025 0.007 -0.026 0.007 -0.033 0.007

South 0.024 0.007 0.024 0.007 0.038 0.007 0.040 0.007
West 0.004 0.007 0.003 0.007 0.006 0.007 0.003 0.007
No. of firms -0.018 0.002 na na -0.041 0.004 na na
Monopoly na na 0.136 0.017 na na 0.311 0.028
Duopoly na na 0.105 0.014 na na 0.254 0.023
Oligopoly-3 na na 0.123 0.011 na na 0.247 0.019
Error correction na na na na 0.036 0.006 0.038 0.005
Notes. All regressions include firm fixed effects that are suppre
intuitive. Finally, we find

a competitive that
market and, say, afirms are mo
monopolist's payoff
to enter the markets that
is driven not onlyare closer
by the intensity to but
of competition their
ters. The competition effect estimate from
also the superiority of the monopolist. Therefore, we t
model is smaller than the estimates from the other
may expect a lower competition parameter estimate
two models. Note that the magnitude of the competi-
after we control for such superiority in the full-entry
tion parameter is not directly comparable across mod-
model.
els because of different normalization. However, we
can compare the relative importance by calculating
4.2. Price Regressions
the ratio of the competition parameter to some esti-
The first-stage entry estimates, although interesting
mate of demand, e.g., annual traffic, and the compe-
tition effect estimate from the third model is smaller on their own, serve mainly as a tool for correcting
than the other two models. A potential explanationpotential endogeneity in the price regression. Our
primary objective in the paper is to study the rela-
for this is that the full model explicitly accounts for
tionship between prices and market structure. Our
the "superiority" of the early entrants in the market.
Recall that the full model allows for the more prof- data include the prices from every firm serving the
itable firm to enter first due to the sequential movemarket for five car types: economy, compact, mid-
assumption. The competition parameter in these mod- size, standard, and full-size. In what follows, we
els is inferred from the difference between the mar- pool the data across firms and car types and regress
ket characteristics in concentrated (e.g., monopoly) the log of prices against firm and car type fixed
markets and those in the more competitive markets. effects, other market-specific covariates, and a market
If firms' payoffs are the same, i.e., there is no het-structure variable, which is captured using two spec-
erogeneity across potential entrants, firms' profits are
ifications: (1) total number of firms7 and (2) an indi-
lower in the competitive markets only due to the cator variable for monopoly, duopoly, and so on. We
competition. However, if we allow for heterogeneity
across firms and assume that the most profitable firm
moves first, the difference between a firm's payoff in 7 Using log (number of firms) produces similar results.
subsequently present the results in

forTable 3). The
each last type
car row in sep-columns 3 and 4 shows
arately to determine whether the thecompetition
estimated parameterdiffers
that represents the correla-
across product types. tion between the unobservables that affect the prices
The results from the regression appear
and the in Table
payoff functions 4. firms' entry deci-
underlying
sions. The parameter
The unit of analysis in these regressions is the is similar
log of in magnitude in both
columns
price for each firm and car type. We useand is precisely
seven market-estimated. The estimate is
specific covariates: Log(March traffic), PopulationD
positive, suggesting that the unobserved factors affect
(population density in the market),
both observedRetail
prices and wages,
probability of firm entry in
Pub_Ratio (percentage of population
the same that uses
way. Recall from public
our discussion in §2.3 that
transportation to get to work), aNo.positive
of coefficient
HQ (the results in an underestimation of
num-
the competition
ber of headquarters for major firms in the parameters.
market), Comparing the control
variables across
Poverty ratio (measure of income), and theHoliday
models that correct
(an for this endo-
indicator variable for holiday destinations). geneity and the modelsNote in that
columns 1 and 2, we find
the airport traffic in the price regression that the parameters for some
is the of the market-specific
total
number of passengers flying into variables change somewhat,
the airport in March but there is little change
2005, the month in which our price in the data areorcollected,
In-terminal car-type fixed effects. The great-
but not the annual airport traffic est change,
in 2004, however,
as is in thein
used parameters of primary
the first-stage model. Finally, note interest
that - those
all that capture the competitive interac-
regressions
include 16 firm fixed effects that tions.are suppressed
In particular, examiningforthe No. of firms param-
brevity. eter in Model 3, we find that the impact of an addi-
The left panel in Table 4 shows the results from the tional entrant in the market is almost twice that sug-
models that ignore the endogeneity of market struc- gested by Model 1. A similar picture appears when
ture variables. The first specification uses the total we use a dummy variable specification. For example,
number of firms in the market to capture the com- the monopoly prices are approximately 30% higher
petitive structure, and the second uses a more flexible than the base of the eight-firm oligopoly versus 14%,
indicator variable specification to capture the nonlin- as suggested by Model 2. As we discussed in §2, this
earities in the relationship between prices and the finding is consistent with the notion that unusually
number of firms in the market. In general, the con- attractive demand conditions encourage entry while
trol variables have the expected sign. Prices tend to also supporting higher prices.
be higher in markets with greater traffic, population We next explore how competition effects vary by
density, and higher retail wages, whereas markets that car types. For this we run separate regressions for
are holiday destinations tend to have lower prices. the five car types in our data: economy, compact,
The variable "In-terminal" is an indicator variable mid-size, standard, and full-size. These regressions
that takes a value of 1 if the car is offered inside the
include the same set of explanatory variables as those
airport terminal and 0 otherwise. The parameter esti- in Table 4, including firm fixed effects, and we run
them separately for each car type. Since the results
mate shows that the prices are significantly higher for
such convenience. In terms of car type, the parameterswere similar for "Large" (mid-size, standard, and
are of the expected sign. Economy cars are approxi- full-size) and "Small" (economy and compact) cars,
mately 15% cheaper and full-size cars are 12% more we pool the data for these products for brevity. The
expensive compared with mid-size cars. Finally, the results are presented in Table 5. We present the results
regional indicator variables suggest that prices are from regressions without correction for endogeneity
highest in airports in the South and lowest in the Mid-of market structure, and for comparison we report
west. With respect to the more interesting parameters the estimates using correction terms derived from the
capturing the competitive structure, the results from first-stage model that ignores firm heterogeneity and
the first column indicate that adding an equivalentthe of full model as in Table 4.8 Looking at the parameter
one additional firm at the airport reduces the price estimates, a few patterns are apparent. First, in terms
by approximately 2%. Column 2 uses a more flexible of control variables, the coefficients are in the same
specification and shows that monopoly markets have direction and in similar magnitude for both car types.
prices that are approximately 14% higher than highly In terms of the variables capturing the market struc-
competitive markets with eight firms. ture, it appears that the competition is more intense
The right panel in Table 4 shows the regression for the smaller (economy and compact) cars, which is
results after we insert the term that corrects for the probably driven by higher price sensitivity for these
endogeneity of market structure. Note that this cor- customers. Comparing the correction results from the
rection term is derived from the first-stage model
that incorporates observed and unobserved firm het- 8 Results from model with only unobserved heterogeneity are sim-
erogeneity in entry decisions (i.e., the full model ilar to the full model.
Table 5 Estimation Results from Price Regression for Different Car Types
Estimates with correction
Estimates without correction Symmetric firms Full model
Small cars Large cars Small cars Large cars Small cars Large cars
Variable Est. Std. error Est. Std. error Est. Std. error Est. Std. error Est. Std. error Est. Std. error
Intercept 3.292 0.042 3.580 0.031 2.915 0.073 3.263 0.054 2.983 0.074 3.309 0.055
Log(March Traffic) 0.025 0.003 0.021 0.003 0.052 0.006 0.044 0.004 0.047 0.005 0.040 0.004
PopulationD 1.068 0.444 1.157 0.329 1.028 0.442 1.128 0.328 1.482 0.450 1.522 0.334
Retail wages 0.006 0.007 0.014 0.005 0.002 0.007 0.010 0.005 0.010 0.007 0.017 0.005
Pub_Ratio 0.004 0.008 0.006 0.006 -0.006 0.008 -0.003 0.006 -0.012 0.009 -0.009 0.007
No.ofHQ -0.008 0.005 -0.004 0.004 0.003 0.005 0.005 0.004 0.001 0.005 0.004 0.004
Poverty ratio -0.001 0.004 -0.001 0.003 0.002 0.004 0.001 0.003 -0.001 0.004 -0.001 0.003
Holiday -0.061 0.014 -0.085 0.011 -0.061 0.014 -0.084 0.011 -0.030 0.016 -0.057 0.012
In-terminal 0.058 0.012 0.056 0.009 0.061 0.012 0.059 0.009 0.058 0.012 0.056 0.009
Midwest -0.019 0.012 -0.028 0.009 -0.033 0.013 -0.041 0.009 -0.028 0.013 -0.036 0.009
South 0.023 0.012 0.026 0.009 0.024 0.012 0.028 0.009 0.039 0.012 0.041 0.009
West 0.005 0.012 0.001 0.009 -0.001 0.012 -0.003 0.009 0.005 0.012 0.002 0.009
Monopoly 0.175 0.030 0.115 0.021 0.420 0.049 0.321 0.036 0.362 0.047 0.280 0.034
Duopoly 0.135 0.024 0.086 0.017 0.335 0.040 0.255 0.029 0.296 0.039 0.227 0.029
Oligopoly-3 0.164 0.020 0.096 0.014 0.319 0.032 0.227 0.023 0.297 0.033 0.214 0.024
Oligopoly-4 0.137 0.019 0.092 0.014 0.271 0.029 0.205 0.021 0.249 0.029 0.191 0.022
Oligopoly-5 0.124 0.017 0.101 0.013 0.248 0.026 0.206 0.019 0.216 0.025 0.182 0.018
Oligopoly-6 0.074 0.013 0.034 0.010 0.164 0.019 0.109 0.014 0.140 0.018 0.092 0.014
Oligopoly-7 0.028 0.012 -0.012 0.009 0.081 0.015 0.033 0.011 0.068 0.015 0.023 0.011
Error correction na na na na 0.046 0.007 0.039 0.005 0.041 0.008 0.036 0.006
Notes. "Large" refers to mid, standard, and full-size cars; "Small" refers to economy and c
model with homogenous firms regressions

and of thethis full
type are
model,
problematicwebecause mar-
find that directionally they are consistent
ket structures in that
are not randomly thea necessary
assigned,
competitive parameters increase conditioninfor a standard regression
magnitude after model to yield
the correction term is inserted. All estimates for the consistent estimates of the relationship between mar-
parameter representing the correlation between theket structure and an outcome variable. The results
unobservables that affect the prices and the payoffpresented above show that the resulting bias in the
functions are positive, suggesting that the unobserved competitive interaction parameters can be severe and
factors affect both observed prices and probability of can bring into question the conclusions drawn from
firm entry in the same way. While the magnitude ofthe previous studies using such regression techniques.
change is slightly greater for the homogenous model, Furthermore, the magnitude of bias in the OLS regres-
there are large changes in the competitive parameters sions can have important policy implications. As dis-
under all model specifications and for both car types. cussed in §1, such price regressions are often used by
Thus the specification of the first-stage entry modelthe FTC to predict price and welfare effects of merg-
does not seem to impact the correction procedure iners. Because the negative ramifications of market con-
our application. Nevertheless, it may be important incentration on prices are underestimated, our results
other industries and applications to test the robust-suggest that horizontal mergers in this industry may
ness of results to the first-stage model specification. incorrectly appear innocuous when using a model
that ignores the endogeneity of market structure.
5. Discussion The two-stage estimation procedure outlined in the
paper canand
We investigate the relationship between prices also be useful in making predictions due
to the changes
the number of firms in the auto rental industry. Our in exogenous variables. For example,
consider
work follows the long stream of literature the change in prices due to an exogenous
in eco-
nomics and marketing that has examined relation-
change in airport traffic. According to the price regres-
ships between market structure measures sions reported above, the coefficient of traffic is posi-
and prices,
markups, revenues, or profits (e.g., Weiss tive,
1989,indicating
Buzzel that an increase in demand results in
and Gale 1987). Typically, this type of higher
application
prices. However, it is important to understand
the source of
involves a regression of the outcome variable of the increase in traffic - that is, whether it
interest on a market structure variable such as con-
is a short-term demand shock or a long-term change
centration ratios or the number of firms. However,
of market condition. If it is a short-term event, e.g.,
References
a sports competition that attracts a lot of travelers
to the airport, we expect prices to go
Asplund, M., R.upSandin.because of in interrelated markets:
1999. Competition
An empirical
the higher demand. However, if the higher traffic isstudy. Internat. ]. Indust. Organ. 17 353-369.
Baker, J., D. L. Rubinfeld. 1999. Empirical methods in antitrust lit-
driven by a long-term factor, e.g., a new resort, the
igation: Review and critique. Amer. Law Econom. Rev. 1(1/2)
answer becomes subtle. In particular,
386-535. according to the
parameter estimates from the entry
Bass, F. 1969.model,
A simultaneous a perma-
equation regression study of adver-
nent increase in traffic (which is tising
an and important deter-
sales of cigarettes. /. Marketing Res. 6(3) 291-300.
minant of the number of firms) may
Berry, induce
S. T. 1992. Estimationadditional
of a model of entry in the airline indus-
competitors to enter the market, leading
try. Econometrica to lower
60(4) 889-917.
prices. Berry, S. T., J. Levinsohn, A. Pakes. 1995. Automobile prices in mar-

ket equilibrium. Econometrica 63(4) 841-890.
From a methodological point of view, our approach
Besanko, D., S. Gupta, D. Jain. 1998. Logit demand estimation
provides an alternative to the methods used by Evans under competitive pricing behavior: An equilibrium frame-
et al. (1993), who study the price concentration in work. Management Sci. 44(11) 1533-1547.
the airline industry. These authors observe a panel Borenstein, S., N. Rose. 1994. Competition and price dispersion in
data structure and use a combination of fixed effects the U.S. airline industry. /. Political Econom. 103 653-683.
and instrumental variable procedures to account for Bresnahan, T. F. 1989. Empirical studies in industries with market
the endogeneity of market structure. As noted in §1, power. R. Schmalensee, R. D. Willig, eds. Handbook of Industrial
Organization, Vol. II. Elsevier Science Publishing, Amsterdam,
such panel data with substantial entry and exit over 1011-1057.
time are rarely available. In light of the cross-sectional
Bresnahan, T. F., P. C. Reiss. 1987. Do entry conditions vary across
data used in majority of applications in this area, our markets? Brookings Papers Econom. Activity (3) 833-881.
approach, in which we explicitly model the determi- Bresnahan, T. F., P. C. Reiss. 1990. Entry in monopoly markets. Rev.
nants of underlying market structure, can provide aEconom. Stud. 57(4) 531-553.
Bresnahan, T. F., P. C. Reiss. 1991. Entry and competition in con-
useful alternative. However, the validity of the cor-
centrated markets. /. Political Econom. 99(5) 977-1009.
rection procedure depends on the correct specifica-
Buzzell, R. D., B. T. Gale. 1987. The PIMS Principles- Linking Strategy
tion of the first-stage entry model. In this paper, weto Performance. The Free Press, New York.
tried three model specifications and found similar Calem, P. S., G. A. Carlino. 1991. The concentration /conduct rela-
results. Depending on the specific application, mod- tionship in bank deposit markets. Rev. Econom. Statist. 73(2)
268-276.
els used in Mazzeo (2002b) for discrete heterogene-
ity, Seim (2006) and Orhun (2006) for firm location, Carpenter, G. S. 1989. Perceptual position and competitive brand
strategy in a two-dimensional, two-brand market. Management
and Zhu and Singh (2007) and Zhu et al. (2008) thatSci. 35(9) 1029-1044.
allow for firm identities can be considered. All of
Chintagunta, P., J. P. Dube, V. Singh. 2003. Balancing profitability
these papers analyze the underlying market struc-
and customer welfare in a supermarket chain. Quant. Marketing
ture variables but have limited or no information on Econom. 1(1) 111-147.
Chintagunta,
price, quantity, and costs. In the marketing literature, P., T. Erdem, P. Rossi, M. Wedel. 2006. Structural mod-
researchers have looked at issues such as reactions to eling in marketing: Review and assessment. Marketing Sci. 25(6)
604-616.
entry (Singh et al. 2006) or changes in the market-
Cotterill, R. 1986. Market power in the retail food industry: Evi-
ing strategies of competitors (Steenkamp et al. 2005),
dence from Vermont. Rev. Econom. Statist. 68(3) 379-386.
implications of brand-positioning strategies on other
Davis, P. 2005. The effect of local competition on retail prices in
marketing mix elements (e.g., Hauser and Shugan
the U.S. motion picture exhibition market. /. Law Econom. 48(2)
677-707.
1983, Carpenter 1989), and the impact of short-term
Demsetz, H. 1973. Industry structure, market rivalry, and public
marketing activities such as price and advertisement
policy. /. Law Econom. 16(April) 1-9.
on sales (e.g., Bass 1969). More recent structural work
Emmons, W. R., R. A. Prager. 1997. The effects of market structure
has tried to endogenize such marketing mix activities and ownership on prices and service offerings in the U.S. cable
of firms while treating the market structure variablestelevision industry. RAND J. Econom. 28 732-750.
as exogenous. With availability of data sets with suf- W. N., L. M. Froeb, G. J. Werden. 1993. Endogeneity in the
Evans,
ficient variation in the number of firms over time or concentration-price relationship: Causes, consequences, and
cures. /. Indust. Econom. 61(4) 431-438.
across markets, an interesting area for future inquiry
Hauser, J., S. Shugan. 1983. Defensive marketing strategies. Market-
could involve estimating supply and demand func- ing Sci. 2(4) 319-360.
tions with endogenous market participants to pro-Heckman, J. J. 1976. Common structure of statistical-models of trun-
vide a more complete characterization of the long-run cation, sample selection and limited dependent variables and
equilibrium in a marketplace. a simple estimator for such models. Ann. Econom. Soc. Measure-
ment 5(4) 475-492.
Kadiyali, V., P. Chintagunta, N. Vilcassim. 2000. Manufacturer-
Acknowledgments retailer channel interactions and implications for channel
The authors are listed in alphabetical order and contributed power: An empirical investigation of pricing in a local market.
equally. Marketing Sci. 19(2) 127-148.
Kadiyali, V., N. Vilcassim, P. Chintagunta.

Seim, 1999.
K. 2006. AnProduct
empirical modelline
of firmexten-
entry with endogenous
sions and competitive market interactions: Anchoices.
product-type empirical anal-
RAND J. Econom. 37(3) 619-640.
ysis. /. Econometrics 89(1-2) 339-363. Singh, V. P., K. T. Hansen, R. C. Blattberg. 2006. Market entry
Khan, R., D. Jain. 2005. An empirical analysis
and consumerof price
behavior: discrimi-
An investigation of a Wal-Mart super-
nation mechanisms and retailer profitability.
center. Marketing
/. Marketing
Sci. 25(5) 457-476. Res.
42(4) 516-524. Steenkamp, J. E. M., V. R. Nijs, D. M. Hanssens. 2005. Competitive
Keeler, E. B., G. Melnick, J. Zwanziger. 1999.
reactionsThe changing
to advertising effects
and promotion attacks. Marketing Sci.
of competition on non-profit and for-profit24(1) 35-54. hospital pricing
behavior. /. Health Econom. 18(1) 69-86.
Sudhir, K. 2001. Structural analysis of manufacturer pricing in the
Manuszak, M. D. 2005. Firm conduct in the presenceHawaiian retail
of a strategic retailer. Marketinggaso-
Sci. 20(3) 244-264.
line industry. Working paper, Carnegie Mellon
Villas-Boas, S. University,
B. 2007. Vertical relationships between manufacturers
Pittsburgh. and retailers: Inference with limited data. Rev. Econom. Stud.
Manuszak, M. D., C. C. Moul. 2007. Prices and endogenous market 74(2) 625-652.
structure in supply superstores. /. Indust. Econom. 56(1) 94-112. Villas-Boas, J. M., R. Winer. 1999. Endogeneity in brand choice mod-
Mazzeo, M. J. 2002a. Competitive outcomes in product-differenti- els. Management Sci. 45 1324-1338.
ated oligopoly. Rev. Econom. Statist. 84(4) 716-728. Villas-Boas, J. M., Y. Zhao. 2005. Retailer, manufacturers, and indi-
Mazzeo, M. J. 2002b. Product choice and oligopoly market struc- vidual consumers: Modeling the supply side in the ketchup
ture. RAND ]. Econom. 33(2) 221-242. marketplace. /. Marketing Res. 42(1) 83-95.
Newmark, C. 2004. Price concentration studies: There you go again. Weiss, L. 1989. Concentration and Price. MIT Press, Cambridge, MA.
Manuscript prepared for the DOJ/FTC Merger Workshop, Whinston, M. D. 2003. Lectures on Antitrust Economics, Chapter 3:
"Concentration and Market Shares" panel. Horizontal Mergers. Working Paper 0041, The Center for the
Orhun, Y. 2006. Spatial differentiation in the supermarket industry. Study of Industrial Organization, Northwestern University,
Working paper, Graduate School of Business, The University Evanston, IL.
of Chicago, Chicago. Zhu, T., V. Singh. 2007. Spatial competition with endogenous loca-
Reiss, P. C. 1996. Empirical models of discrete strategic choices. tion choices: An application to discount retailing. Working
Amer. Econom. Rev. 86(2) 421^26. paper, Carnegie Mellon University, Pittsburgh.
Schmalensee, R. 1989. Inter-industry studies of structure and per- Zhu, T., V. Singh, M. Manuszak. 2008. Market structure and
formance. R. Schmalensee, R. D. Willig, eds. Handbook of Indus- competition in the retail discount industry. /. Marketing Res.
trial Organization, Vol. 2. North-Holland, Amsterdam, 951-1009. Forthcoming.

Pricing and Market Concentration in Oligopoly Markets

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Pricing and Market Concentration in Oligopoly Markets

Uploaded by

Copyright:

Available Formats

Pricing and Market Concentration in Oligopoly Markets

Author(s): Vishal Singh and Ting Zhu

Pricing and Market Concentration in

Key words: pricing; market structure; entry; horizontal mergers; endogeneity

1. Introduction market concentration could be due to the competitive

A general finding in this literature

the case of entry games, payoffswriting pads, and envelopes).

The bias in the parameters capturing

2. Model prices and the number of firms in the same manner

]npmk = Zmke + f(Nm,Sp) + ef'mk, (1)

To identify the parameters of a the

Note that the observed marketdata Xmk and with

\VmN / G(0) = g(0)'Ag(0).

high (H) or low

4 Note that if a firm's payoff is not af

correlation, which is also summarized

Table 2 Airport Traffic and Prices by Market Structure

Market structure N Traffic Economy Compact Mid Standard Full

The variable "No. of HQ"In is Table

Table 3 Parameter Estimates from Entry Model

With unobserved With unobserved and

Intercept 2.674 0.182 3.025 0.155 0.695 0.170

the market in which heterogeneity.

Table 4 Pooled Estimation Results from Price Regression

Estimates without correction Estimates with correction

Model 1 Model 2 Model 3 Model 4

Intercept 3.651 0.017 3.475 0.025 3.562 0.022 3.188 0.044

Midwest -0.020 0.007 -0.025 0.007 -0.026 0.007 -0.033 0.007

intuitive. Finally, we find

subsequently present the results in

Estimates with correction

Estimates without correction Symmetric firms Full model

model with homogenous firms regressions

prices. Berry, S. T., J. Levinsohn, A. Pakes. 1995. Automobile prices in mar-

Kadiyali, V., N. Vilcassim, P. Chintagunta.

You might also like