A Poisson Probability Model of Entry and Market Structure with an Application to U. S.
Industries during 1972-77
Author(s): William F. Chappell, Mwangi S. Kimenyi, Walter J. Mayer
Source: Southern Economic Journal, Vol. 56, No. 4, (Apr., 1990), pp. 918-927
Published by: Southern Economic Association
Stable URL: http://www.jstor.org/stable/1059881
Accessed: 26/06/2008 11:32
Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at
http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless
you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you
may use content in the JSTOR archive only for your personal, non-commercial use.
Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at
http://www.jstor.org/action/showPublisher?publisherCode=sea.
Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed
page of such transmission.
JSTOR is a not-for-profit organization founded in 1995 to build trusted digital archives for scholarship. We work with the
scholarly community to preserve their work and the materials they rely upon, and to build a common research platform that
promotes the discovery and use of these resources. For more information about JSTOR, please contact [email protected].
http://www.jstor.org
A Poisson Probability Model of Entry and
Market Structure with an Application to
U.S. Industries during 1972-77
WILLIAMF. CHAPPELL
MWANGI S. KIMENYI
WALTERJ. MAYER
Universityof Mississippi
University,Mississippi
I. Introduction
The performancecharacteristicsof an industryare closely linked to the natureof entry and exit
in the industry.If entry barriersare low, the threatof potential entry can effectively constrain
the ability of incumbentfirms to raise price above the competitive level. On the other hand, as
entrybarriersrise and the probabilityof entrydiminishes,the potentialfor monopolisticpractices
increases. Priorempiricalstudies of entryhave focused mainly on its determinants,emphasizing
industrycharacteristicsas entrybarriers.ExamplesincludeMcGuckin [19], Orr [21], McDonald
[17], and Duetsch [7], which analyzethe numberof new entrants,and Berry [4], McDonald [17;
18], and Masson and Shannon[15; 16] which focus on the marketshareof enteringfirms.
Our study uses the model of Orr, and its later extensionby Duetsch, as its initial reference
point. Like the Orr-Duetschstudies we estimate a model for entry determinantsacross industries based on the numberof new firms. Our contributionis two-fold. First, we analyze a new
sample period, 1972-77. Second, we provide a methodologicalimprovementover the logarithmic regressionapproachesof Orrand Duestch. Because the observationson entry are count data
(non-negativeintegers), our model is developedfrom the premise that entry requiresa statistical
frameworkbased on a discreteprobabilitydistribution.To meet this requirement,we specify and
estimate an econometricmodel of entrybased on the Poisson distribution.Our methodology is in
the spirit of Hausman,Hall, and Griliches [11] who apply the Poisson distributionto count data
on patent applicationacross firms. Hausman,Hall and Grilichespoint out that the Poisson model
offers an improvedmethodologyfor a wide rangeof economic applicationsthatfeaturedata in the
form of repeatedcounts. This observationmotivatedour applicationof the Poisson distributionto
the entry problem.
The Poisson approachadmitsa richeranalysisof the entry data than the logarithmicregression approachin two ways. First, the logarithmicspecification,while computationallyconvenient,
provides a ratherincompletedescriptionof the entry data. The log of entry is only well defined
1. Other applicationsof the Poisson distributionto economic data can be found in Cameronand Triveda [5], and
Mullahy [20].
918
POISSON PROBABILITY MODEL OF ENTRY
919
for those industriesshowing positive entry;it is undefinedfor industriesin which entry is absent.
The flaw, of course, is that "zero entry" industries,which are not capturedin the specification,
are also importantto the study of entry behavior.It is these industriesthat reveal the configurations of profitsand other industrycharacteristicsthatprecludeentry.If observationson the "zero
entry" industriesare deleted from the sample to facilitatethe estimation, as is sometimes done
in practice, then relevantinformationis lost. For the Poisson approach,in contrast, both types
of observations(positive and zero entry) are naturaloutcomes of the specification;the Poisson
approachincorporatesall relevantinformationin a logically consistentmanner.
The second limitation of the log specification,which is overcome by the Poisson, is that
maximum likelihood estimationis precluded.Obviously,the errorterm for a logarithmicregression equationwith a discretedependentvariable(the log of entry)cannotbe normallydistributed,
which is requiredfor the OLS estimatesto coincide with the ML estimates. More generally, ML
estimationis precludedfor this specificationbecause it is unclearwhat distributionalassumptions
are appropriate.In contrast, for the Poisson approach,which specifies the distributionof entry,
ML estimationalong with its well known desirablepropertiesis readily available.2
With the distributionof entry specified as Poisson we are also able to estimate the probabilities of entry across industries, and thus direct the study towardslimit pricing in a new and
interesting way. Indeed, the probabilityof entry has become a central feature of recent limit
pricing models. In the classic static limit pricing model of Bain [2], dominantfirms can either
engage in limit pricing and deter all entryor they can set a short-runprofitmaximizing price that
results in unimpededentry.Entrybarriersdeterminethe optimal strategy;with high barriers,the
price decrease needed to deter entry is likely to be small enough to make limit-pricing a viable
alternative. The probabilityof entry is crucial to the recent extensions of Bain's work to a dynamic setting by Kamien and Schartz [13], Gaskins [9], Baron [3], and Flaherty [8]. In these
models, firmsface a continuumof pricingstrategieslinkedto the probabilityof entry:as the price
rises above that which completely blocks entry,the probabilityof entry graduallyrises. In sum,
dynamic limit-pricingmodels predictthatpricingdecisions acrossdifferentmarketstructuresvary
with the probabilityof entryacross markets.Consequently,estimatesof these probabilitiesare of
interestfor limit pricing.
Section II presentsa generalspecificationof entrybased on the workof Orr [21] and Duetsch
[7], and describes the Poisson estimationapproachin more detail. Section III describes the data
on the 330 four-digit SIC industriesto be analyzed. Section IV presents the ML estimates of
the Poisson model, and OLS estimatesof log and linearentry models. These estimates are compared with each other and with those found in the literature.Finally, we present estimates of the
probabilityof entry across industries.
II. A Poisson Model of Entry
The first task is to identify the relevantentry-variables.Our selection of variables was guided
primarily by argumentsgiven in Orr [21] and Duetsch [7] which we briefly summarize here.
2. In particular,undercertainregularityconditions,the maximumlikelihoodestimatoris the most efficient estimator in the sense that its asymptoticcovariancematrixreachesthe Cramer-Raolower bound. Least squaresestimators, on
the other hand, share this propertyonly if the errorterm in the equationis normallydistributed.For furtherdetails, see,
for example, Johnston[12, 274-9].
920
WilliamF. Chappell,MwangiS. Kimenyi,and WalterJ. Mayer
(The next section defines the operationalversionsof the variablesactuallyused in the estimation.)
Having identifiedthe relevantvariables,we then turnto the problemof estimation, and introduce
the Poisson model.
The relevantentry-variablesfollow from an assessmentof the entry decision as a function
of perceived benefits and costs. On the benefit side, entry is expected to be higher in industries
in which incumbentfirms earn relativelyhigher profits, and in large or growing markets. The
costs of entry are capturedby variablesreflectingentry barriers.Three such variables traditionally emphasized are capital, advertising,and industryconcentration.High capital requirements
can discouragepotential entrants,as can advertisingexpendituresby existing firms. The role of
advertisingexpenditures, however, is somewhatambiguoussince entrantsmay use advertising
as a vehicle for increasingthe chances of successful entry. The level of industry concentration
is negatively related to entry if potentialentrantsrecognize the threatof collusive retaliationby
dominantfirms in concentratedmarkets.
Finally, entry conditions are related to productioncharacteristicsacross markets. Large
economies of scale in an industrycan discourageentry: unless large scale entry is feasible, an
entrantwill produce at a highercost thanexisting firms. Multiplantoperationscan also discourage entry. Duetsch [7] suggests two reasons for an entry-deterringeffect. Multiplantoperations
may be caused by geographicor productsegmentationof markets.A new entrantis likely to be
at a cost disadvantageif it is only able to produceone productline or in one region. Established
multiplantfirms can also more readilyinitiatea price reductionin an individualmarketsegment
when entry occurs.
These considerationslead to the following specification:
where
Ei =f(Xi),
Xi = (PRi, GRi, VSi, Ki, ASRi, CRi, SCALi, MULT),
i = 1,2,...
.
,n.
Ei denotes the numberof new entrantsin the ith industry;PRi is a profit index; GRi is industry
growth;VSi is the value of shipmentsin the industry;Ki is the capitalrequirementper plant;ASRi
is the advertising-salesratio;CRi is the industryconcentrationratio;SCALimeasures economies
of scale; and MULTiis an index of multiplantactivity.
Orr [21] and Duetsch [7] base their estimationson logarithmicregression versions of the
specification.The difficultywith this approachis thatentry,Ei, is an integer-valuedcount variable
with a range that includes Ei = 0 (no entry). Thus, the problemarises of how to handle log(Ei)
for Ei = 0. If one wishes to retainthe logarithmicspecificationthere are two options, neither of
which is completely satisfactory.The first option is to exclude the "zeroes" from the sample.
However, a smallerand less informativesamplemeans less efficientestimates. Alternatively,one
can "solve" the zero value problemby adoptingan ad hoc proceduresuch as setting the zeroes
to one and adding a dummy variableto the equation as is done in Pakes and Griliches [22].
(Neither Orrnor Duetsch discuss the zero value problem,and thus it is not clear how they handle
it.) Of course, the magnitudeof the problemvaries directlywith the numberof industriesin the
sample characterizedby no entry.For our samplesuch industriesare significant(38 percent), and
consequentlythe zero value problemmeritsattention.
The necessity of ad hoc proceduresor excludingobservationsbrings into question the appropriatenessof applying the logarithmicapproachto the entry data. Recently, Hausman, Hall,
and Griliches [11] addressthe problemof estimationwith count data, and the related zero value
POISSONPROBABILITYMODELOF ENTRY
921
problem. As an alternativeto the logarithmicregressionapproach,they propose modeling count
data with the Poisson distribution,and obtainingthe estimatesby the method of maximum likelihood. Unlike the logarithmicregressionspecification,the Poisson explicitly reflects the integer
natureof count data. The range of a Poisson randomvariableis the set of nonnegative integers,
which obviously matches the possible values of Ei. As a result, by specifying entry as a Poisson
randomvariable,the zero problem,Ei = 0, becomes a naturaloutcome of the specification;there
is no need to either exclude observationsor incorporatead hoc proceduresinto the estimation.
For these reasons, this is the approachtakenhere.
The data on entry are assumedto be generatedby the following Poisson distribution:
Prob(Ei)= tiEi exp(-ti)/Ei
!,
Ei = 0, 1, 2,...
where /,i is the firstmomentof the Poissondistribution,and thus is the expected value of entry in
industryi. The explanatoryvariables,Xi, enter the model by specifying the Poisson parameter,
,ti, as a functionof Xi and an unknownparametervector, P, to be estimated. Consequently,the
Poisson model is analogousto the familiarregressionspecificationin that E(Ei Xi) = g(Xi, P),
where g(Xi, 1P)= xi.
Following Hausman,Hall, and Griliches[11], we specify:
=/i
=exp(ZiP3),
where Zi = (1, Xi). Augmentingthe vector Xi with "1" allows for an interceptterm.
To obtain the maximumlikelihood estimateof 1 (and hence /i), P is chosen to maximize
the log likelihood of the sample of n industries:
n
[Ei ! - exp(zi ) + Eizip].
L(P) =
i=l
The maximumlikelihood estimate, S, is obtainedby solving the firstorderconditionwhich takes
the form:
E[Zi'(Ei - exp(Zi 3)]= 0.
Because B enters nonlinearly,a numericalalgorithmsuch as Newton-Raphsonis requiredto iterate to a solution. Hausman, Hall, and Griliches [11] point out that the likelihood function is
globally concave which ensuresconvergenceto a uniquesolution. This propertyfollows from the
negative-definiteHessian matrixof the likelihood:
E[-
(Zi'Zi) exp(Zi P1)]
Estimates of the asymptotic variances of the ML estimates, which are needed to compute tstatistics, are obtained from the negative inverseof this matrixevaluatedat P.
As discussed in the introduction,the probabilityof entryis of interestto contemporarylimit
pricing theory. Once 13is obtained, estimatesof these probabilitiesacrossindustriescan be easily
generatedfrom
Eo
Prob(Ei > Eo) = 1 - E ir exp(-i,i)/r!.
r=O
922
William F. Chappell, Mwangi S. Kimenyi, and Walter J. Mayer
Industries
TableI. VariableDefinitionsandMeans:330 Four-digit
Manufacturing
Definition
Mean
E
Net entry of firms:numberof firms in 1977 minus the the number
of firms in 1972
92.064
258
PR
Profit index: the residualof 1972 industryprice-cost margins
regressedon capital-salesratios and advertisingsales ratios
0.000
.073
GR
Growth:value of shipmentsin 1977 minus value of shipmentsin
1972 divided by 1972 shipments
.715
.427
VS
Value of 1972 shipments, millions of dollars
1705
3412
K
Average capital requirementper plant:Fixed assets in 1972
(million of dollars) divided by numberof plantsin 1972
2.994
8.396
ASR
Advertising-salesratio for 1972
.015
.027
CR4
1972 four firm concentrationratio
.397
.212
SCAL
Value-addedper employee of the four-largestfirmsdividedby
value-addedper employee of all smallerfirms, 1972 values
1.393
.479
MULT
Number of plants in 1972 minus the numberof firmsin 1972
divided by the numberof firms in 1972
.242
.366
Variable
St. Dev.
III. Data Description
The sample to be analyzed consists of observations on 330 four-digit SIC industries for the years
1972 and 1977. Summary statistics and variable descriptions are given in Table I. Data on all
variables except ASR are from the 1972 and 1977 census of manufacturers; data on the advertising
variable, ASR, are from the 1972 Input-Output tables.3
Following Orr and Duetsch we adopt "net entry" as an operational definition of E:
Ei =
i77
= 0,
-
i72,
if ci77 > Ci72,
if otherwise;
where, for example, ci77 is the number of firms reported in the ith industry in 1977. For the reasons discussed in Duetsch [7, 480-1], net entry is likely to underestimate the number of entering
3. There are 450 four-digitSIC industriesin manufacturing.Input-outputtables providedata at the three digit level
for a numberof industries;these were deleted. The census of manufacturersdoes not providedata on concentrationratios
in a few industries,resultingin their deletion also.
Our variableswere generatedfrom thirteenvariables.The numberof companies in 1972 and 1977, value of shipments in 1972 and 1977, value-addedin 1972, total employmentin 1972, numberof plants in 1972; fixed assets in 1972,
and total payroll in 1972 are from the four-digitindustrytables entitled "HistoricalStatistics for the Industry:1977 and
EarlierYears," in VolumeII, Part 1, 2, and 3, Censusof Manufactures.Fixed assets values were not availablefor 1972 for
seventeen industries.In those cases the nearestASM estimatewas used. Values for the proportionof shipments, employment, and value-addedaccountedfor by the four largest firms in each industryare from 1972 Census of Manufactures,
Special ReportSeries, ConcentrationRatios in Manufacturing,MC72(SR)-2, Table8. Values for advertisingexpenditures
are from 1972 Input-OutputTables.A list of the SIC industriesused is availableon request.
POISSONPROBABILITYMODELOF ENTRY
923
firms. Ideally, one should measureE directly with data on gross entry; unfortunately,such data
are unavailable.
Following Duetsch [7], the profit indices, PR, are the residuals from regressing industry
price-cost margins on capital-salesratios and advertising-salesratios. The reason for the adjustment is to obtain a measureof profitabilitymore closely associated with the entry incentive. As
Duetsch [7, 481] points out, high price-cost marginsdue to high advertisingand capital intensity
will be largely discountedby potentialentrants.
For an index of scale economies in an industry,SCAL, we use the productivityper worker
of large firms relative to the productivityper workerof small firms. A similar index has been
used by Allen [1], and Chappeland Cottle [6]. Descriptionsof the remainingvariablesare left to
Table I.
IV. Empirical Results
For comparisonpurposes, three versions of the entry specificationwere estimated: a regression
model with E as the dependentvariable(the linear model), a log model (the Orr-Duetsch approach), and the Poisson model. The first two versionswere estimatedby OLS, and the Poisson
by maximum likelihood. The log model excluded the "zero entry" industriesfrom the sample,
and thus was based on 203 observations.In contrast,estimationof the othertwo models used 330
observations. The likelihood functionfor the Poisson model was maximized by a SAS program
written with "Proc Matrix". The programuses the Newton-Raphsonalgorithm. Our experience
with maximizingthe likelihoodwas similarto thatreportedby Hausman,Hall, and Griliches [11];
convergence to the global maximumwas fairlyrapid. The results are reportedin Table II.
A comparison of the three columns of Table II reveals differences among the estimated
models, and thus indicatesthe sensitivityof conclusionsdrawnfrom the entry-datato the econometric specification.In particular,as in Hausman,Hall, and Griliches[11], the estimated standard
errorsof the Poisson model areconsiderablysmallerthanthose of the OLS estimates. While there
are some similarities in results, there are also notable differences. Going from the linear to the
Poisson model, the coefficients of PR, ASR, CR4, SCAL, and MULTbecome significant. Going
from the log to the Poisson model, the coefficientof ASR switches sign and becomes significant,
and the coefficients of CR4 and SCALbecome significant.We have arguedthat the most appropriate specificationis the Poisson model. The remainderof the paperconcentrateson the Poisson
estimates.
The Poisson estimatesaregenerallyquitereasonable.All of the variablesare statisticallysignificant, and the signs of the coefficientsfor eight out of nine variablescome out as predictedby
theory. (The one exception,SCAL,is discussedbelow.) We now turnto a discussionof the Poisson
estimate for each variableindividually,and drawcomparisonswith the resultsobtainedby Orrand
Duetsch. Interestingly,the Poissonestimatesconformmoreclosely, overall, to economic-theoretic
predictionsthan do the results of either study.
Of the entry determinantsin TableII, the estimatesof the Orrand Duetsch studies agree on
only market size, InVS, and plant capital requirement,lnK.4 The Poisson estimates also agree
with these findings. Marketsize is positively and significantlyrelated to entry, while the effect
of plant capital requirementsis negative and significant.The Orr-Duetschestimates conflict for
4. Both Orrand Duetsch use the averagecapitalrequirementof the largestplants in an industryas a measureof K.
We have used the averagecapitalrequirementof all plantsin an industrywith the same result.
924
WilliamF. Chappell,MwangiS. Kimenyi,and WalterJ. Mayer
Table II. Determinantsof Entry*
Variable
Constant
Linear
Model
-454.296a
(-3.861)
PR
113.971
(.586)
GR
99.616a
(3.247)
In VS
67.922a
InK
(4.830)
-50.035 a
(-2.960)
ASR
-403.035
(-.772)
CR4
-115.047
(-1.144)
SCAL
MULT
48.899
(1.495)
-15.551
(-.377)
Log
Model
Poisson
Model
- 1.586b
(-2.299)
- 1.509a
(-24.849)
2.441 b
1.500a
(2.168)
(16.453)
.621a
(3.456)
1.255a
(103.313)
.796a
.783a
(9.844)
- .699a
(119.925)
(-7.265)
.474
(.159)
-.750
(-1.360)
-.152
(.690)
-.558b
(-2.312)
-.387 a
(-39.603)
-.755a
(-3.605)
-2.692a
(-49.792)
.300a
(2.929)
-3.551
(-38.674)
*t-ratios are in parentheses. Following Orr [21] and Duetsch [7], the logarithmicvalues of VS and K are used in
these regressions. The logarithmicmodel is estimatedfrom 203 observations,while the other two models are estimated
from 330 observations.
a. Significantat one percent level, two tail test.
b. Significantat five percent level, two tail test.
three of the variablesused in the Poisson model: PR, ASR, and CR4. Orr did not find profits,
PR, significantlyrelatedto entry,while Duetsch found it to be positive and significant.The Poisson estimate for PR agrees with Duetsch, and thus also conformswith the traditionalnotion that
profits play a crucial role in the allocationof resources. Orr found both advertising,ASR, and
concentration,CR4, to be significantbarriersto entry, but Duetsch's results were insignificant.
The Poisson estimatesforASR and CR4 agree with Orr,and thus with the roles typically assigned
to advertisingand concentrationas entrybarriersby economic theory.
There are threevariablesthatarecomparableonly betweenthe Duetsch study and the present
study. Growth, GR, is a positive and significantentry determinantin both studies. The Poisson
estimate for multi-plantactivity,MULT,is negativeand significantwhich corroboratesDuetsch's
estimate and multi-planthypothesis. Finally, the economies of scale proxy was insignificantin
Duetsch, which contrastswith the positive and significantPoisson estimate for SCAL.5Thus, the
sign of the Poisson estimate for SCAL is contraryto a priori expectations. This nontraditional
result may indicatean aspect of the relationshipbetweenefficiencydifferentialsof firms within an
industryand entry that has generallybeen ignored. Recall that the proxy SCALis the per worker
5. The scale proxy in Duetsch differsfromthe one used here. He used a measureof the cost disadvantageassociated
with operatingsub-optimalplants;we use a productivitydifferentialbetween large and small firms.
POISSON PROBABILITY MODEL OF ENTRY
925
of EntryacrossMarket
Structures
TableIII.TheProbability
ESTIMATED
PROBABILITIES
OFENTRY:
Numberof EnteringFirmsGreater
Than
0
1
5
10
50
100
150
200
300
<.20
.21-.30
.31-.50
1.00
.98
.98
.99
.97
.95
.99
.95
.83
.99
.93
.71
.83
.61
.29
.67
.34
.12
.56
.21
.05
.42
.13
.02
.25
.06
.01
.51-.70
>.71
.89
.72
.84
.57
.66
.21
.49
.10
.12
.03
.05
.00
.03
.00
.02
.00
.02
.00
CR4
SAMPLE AVERAGES OF ACTUAL AND PREDICTEDENTRY:
Concentration
Ratio,CR4
<.20
Numberof
Industries
71
Actual
Entry
257
Predicted
Entry
250
61
86
99
102
40
43
63
38
30
33
11
6
mean=. 13
.21-.30
mean= .24
.31-.50
mean= .40
.51-.70
mean= .57
>.71
mean= .81
productivitydifferentialbetween small and large firms. If there is a large differencebetween the
productivity per workerof large and small firms, and hence a large value for SCAL, there are
two possible effects on entry.To the extent thatenteringfirms must compete with large, efficient
firms, there is a negative effect on entry.This is the traditionalinterpretation.However, if there is
a relativelyinefficientfringe thatenteringfirmsare going to compete against, entry will be stimulated ratherthan suppressed.Our results suggest that when there is a significantcost difference
among firms in an industry,there is a net positive effect on entry.
Next we consider the issue of limit pricing which was not addressed in the Orr-Duetsch
studies. Limit pricing models are predicatedon the assumptionthat dominant firms base their
pricing decisions on the probabilityof entry. These models demonstratethat small probabilities
of entry are suggestive of limit pricing. As mentionedearlier,an advantageof the Poisson model
is that estimates of the probabilitiesof entry across industriescan be computedfrom it. The estimated probabilitiesof entry are reportedin TableIII. Five marketstructureswere created based
on industryconcentrationlevels. Actual and predictedentry (based on estimates of /i) are also
reportedfor each concentrationgroup.
In the lowest concentrationgroup (four-firmconcentrationless than or equal to 20 percent)
the estimated probabilityof more than 10 and 50 firmsenteringis .99 and .83 respectively. This
result is as expected: there is little basis for believing that limit pricing can occur in low concen-
926
WilliamF. Chappell,MwangiS. Kimenyi,and WalterJ. Mayer
trationindustries. Similarly,the estimatedprobabilitiesof significantentry (more than 10 firms)
are large for the next two groups:21-30 percentand 31-50 percentconcentration.
When concentrationis greater than 50 percent, the results are more suggestive of limit
pricing. In the 51-70 percentgroup, the estimatedprobabilityof entry by more than five firms is
.66, drops to .49 for more than 10 firmsentering,and is only .12 for more than 50 firms entering.
The results for the highest concentrationcategory (greaterthan 71 percent) indicate conditions
amenable to limit pricing by dominantfirms:the estimated probabilityof more than one firm
enteringis only .57 and the estimatedramaticallydropsto .21 for more than five firms entering.
V. Concluding Remarks
We have presented maximumlikelihood estimatesof an entry model based on the Poisson distribution. This methodology reflects the discrete natureof the dependent variable. The sample
analyzed was for a more recenttime period, and includedconsiderablymore industriesthan prior
studies. The estimatesgenerallyconformto a prioriexpectations.We found profitability,growth,
marketsize, and intra-firmefficiencydifferentialsto be positively relatedto entry.Four indices of
entry barrierswere foundto have a significant,negativeimpact:capitalrequirements,advertising,
concentration,and multi-plantproduction.These results are especially importantgiven the conflicting results of priorstudies, most notably,those of Orrand Duetsch. Moreover,we presented
the first estimates of the probabilityof entry across various marketstructures;a concept that is
fundamentalto contemporarylimit pricingmodels.
References
1. Allen, RobertF., "Efficiency,MarketPowerand Profitabilityin AmericanManufacturing."SouthernEconomic
Journal, April 1983, 933-39.
2. Bain, Joe S. Barriersto New Competition.Cambridge:HarvardUniversityPress, 1956.
3. Baron, David P., "LimitPricing, PotentialEntry,and Barriersto Entry."AmericanEconomicReview, September 1973, 666-74.
4. Berry,CharlesH. CorporateGrowthand Diversification.Princeton:PrincetonUniversityPress, 1975.
5. Cameron,A. C. and P. K. Triveda,"EconometricModels Based on CountData:Comparisonsand Applications
of Some Estimatesand Tests."Journalof AppliedEconometrics,1986, 29-53.
6. Chappell, William F. and Rex L. Cottle, "Sourcesof Concentration-Related
Profits."SouthernEconomicJournal, April 1985, 1031-37.
7. Duetsch, Larry L., "Entryand the Extent of MultiplantOperations."Journal of IndustrialEconomics, June
1984, 477-88.
8. Flaherty,M. Therese, "Dynamic Limit Pricing, Barriersto Entry,and RationalFirms." Journal of Economic
Theory, October 1980, 160-82.
9. Gaskins, Darius, "DynamicLimit Pricing:OptimalPricing Under the Threatof Entry."Journal of Economic
Theory, September1971, 306-22.
10. Gorecki, Paul K., "The Determinantsof Entryby New and DiversifyingEnterprisesin the U.K. Manufacturing
Sector, 1958-63." AppliedEconomics,June 1975, 139-47.
11. Hausman, Jerry,BronwynH. Hall, and Zvi Griliches, "EconometricModels For Count Data With An Application To the Patents-R&DRelationship."Econometrica,July 1984, 909-38.
12. Johnston,John. EconometricMethods.New York:McGraw-HillCompany,1984.
13. Kamien, Mortonand Nancy Schwartz,"LimitPricingand UncertainEntry."Econometrica,May 1971, 441-54.
14. Maddala,G. S. LimitedDependentVariablesin Econometrics.Cambridge,Mass.: CambridgeUniversityPress,
1983.
15. Masson, RobertT., and Joseph Shaanan, "Stochastic-DynamicLimit Pricing:An EmpiricalTest." Review of
Economicsand Statistics, August 1982, 413-22.
POISSONPROBABILITYMODELOF ENTRY
927
16.
and
, "Excess Capacity and Limit Pricing: An Empirical Test." Economica, August 1986,
365-78.
17. McDonald, JamesM., "Diversification,MarketGrowth, and Concentrationin U.S. Manufacturing."Southern
EconomicJournal, April 1984, 1098-111.
18.
, "Entryand Exit on the CompetitiveFringe."SouthernEconomicJournal, January1986, 640-52.
19. McGuckin, Robert, "Entry,ConcentrationChange, and Stabilityof MarketShares."SouthernEconomic Journal, January1972, 363-70.
20. Mullahy, John, "Specificationand Testingof Some Modified Count Data Models." Journal of Econometrics,
December 1986, 341-65.
21. Orr, Dale, "The Determinantsof Entry:A Study of the CanadianManufacturingIndustries."Review of Economics and Statistics, February1974, 58-66.
22. Pakes, A. and Zvi Griliches, "Patentsand R&D at the FirmLevel: A First Look." EconomicLetters, 5 (1980),
377-81.