P
HOENIX CENTER POLICY PAPER SERIES
Phoenix Center Policy Paper Number 48:
The Price Effects of Intra-Brand Competition in the
Automobile
Industry: An Econometric Analysis
T. Randolph Beard, PhD
George S. Ford, PhD
Lawrence J. Spiwak, Esq.
(March 2015)
© Phoenix Center for Advanced Legal and Economic Public Policy Studies, T. Randolph Beard,
George S. Ford and Lawrence J. Spiwak (2015).
Phoenix Center Policy Paper No. 48
The Price Effects of Intra-Brand Competition in the
Automobile Industry: An Econometric Analysis
T. Randolph Beard, PhD
George S. Ford, PhD
†
Lawrence J. Spiwak, Esq.
‡
(© Phoenix Center for Advanced Legal & Economic Public Policy Studies, T. Randolph Beard,
George S. Ford and Lawrence J. Spiwak (2015).)
Abstract: Cars are expensive and most consumers know to shop dealers
for the best prices. Yet, there is little to no empirical evidence on the
price effects of intra-brand competition among different dealer franchises
for the same automobile model. In this P
APER, using large samples of
transactions for ten of the most popular new cars purchased in the state
of Texas for the years 2011, 2012, and 2013, we estimate the effects of
intra-brand competition on new car prices. Intra-brand competition is
measured as the distance (in miles) to the nearest same-brand dealer.
Significantly, for all but one automobile model we consider in our
empirical analysis, we find that intra-brand competition does, in fact,
lower new car prices for consumers. For the popular Honda Accord, for
example, increasing the distance between Honda dealerships by thirty
miles raises the price paid by consumers by about $500. Given that retail
margins on auto sales are quite small (about 6% on average), the price
reductions resulting from intra-brand competition are substantial
relative savings for new-car consumers. Moreover, we find that the price
effects of intra-brand competition are relatively strong compared to
inter-brand competition—at the sample means, moving an intra-brand
dealer one mile closer reduces prices by the equivalent of an increase in
35 inter-brand rivals.
Senior Fellow, Phoenix Center for Advanced Legal & Economic Public Policy Studies;
Professor of Economics, Auburn University.
†
Chief Economist, Phoenix Center for Advanced Legal & Economic Public Policy Studies.
‡
President, Phoenix Center for Advanced Legal & Economic Public Policy Studies. The
views expressed in this paper are the authors’ alone and do not represent the views of the Phoenix
Center or its staff.
2 PHOENIX CENTER POLICY PAPER [Number 48
Phoenix Center for Advanced Legal and Economic Public Policy Studies
www.phoenix-center.org
TABLE OF CONTENTS:
I. Introduction ................................................................................................... 2
II. Background .................................................................................................... 4
III. Literature Review .......................................................................................... 7
IV. Empirical Model ............................................................................................ 9
A. Statistical Model ................................................................................... 14
B. Results .................................................................................................... 15
1. Ordinary Least Squares ................................................................... 16
2. Robust Regression ............................................................................ 18
3. Finite Mixture Model ....................................................................... 19
C. Summary ............................................................................................... 21
V. Conclusion ................................................................................................... 22
I. Introduction
Outside of buying a home, an automobile will be the most expensive item
ever purchased by many Americans. Annual transportation expenditures (17.6%
of total) are second only to housing costs (33.5% of total) for the average
consumer, exceeding food costs by 40% and healthcare expenditures by 150%.
1
As we all know from personal experience, buying and maintaining a car is a
significant challenge.
Fortunately, the automobile industry is today workably competitive by
almost any standard, thereby keeping prices low and quality high. Rivalry in the
industry exists at two levels: inter- and intra-brand competition. Inter-brand
competition occurs between different manufacturers such as Ford, Toyota,
Honda, and others, with each offering differentiated products that attempt to
satisfy the varied preferences of consumers. Intra-brand competition, in contrast,
occurs among dealers of the same brand and thus tends to emphasize price.
Intra-brand rivalry can be fierce: for example, Toyota’s website returns
seventeen (17) dealers for the city of Houston, Texas.
2
1
Consumer Expenditures 2013, Economics News Release, Bureau of Labor Statistics
(September 9, 2014) (available at: http://www.bls.gov/news.release/pdf/cesan.pdf
).
2
See http://www.toyota.com/dealers.
Winter 2015] INTRA-BRAND COMPETITION IN THE AUTO INDUSTRY 3
Phoenix Center for Advanced Legal and Economic Public Policy Studies
www.phoenix-center.org
In the U.S., competition among dealerships is made more interesting and
arguably more intense by the fact that dealerships are independent franchisees of
the manufacturers. Manufacturers do not own dealerships and are mostly
prohibited from doing so by state laws.
3
The franchise system has probably led
to a greater number of dealerships—especially for domestic brands—than would
a more vertically-integrated structure. In turn, the large numbers of same-brand
dealers in many local markets may lead to intense price competition.
Despite this important structural feature of the retail automobile industry, we
are unable to find any formal empirical analysis of the price effects of intra-brand
competition. In an effort to shed light on this important issue, in this P
APER we
conduct regression analysis on large samples of transactions for ten of the most
popular cars sold in the state of Texas during the years 2011, 2012 and 2013.
Given the strengths and weaknesses of the data we use, econometric methods
applied include ordinary least squares, robust regression, and finite mixture
models. We measure intra-brand competition as the distance (in miles) to the
closest intra-brand rival. While inter-brand competition, measured as the
number of new or used car dealerships within a thirty-mile radius of each dealer,
is generally found to reduce prices, the evidence suggests, as expected, that intra-
brand competition also reduces prices for consumers. Our modeling approach
allows these effects to vary by make and model and we do find some variation in
these respects.
Holding inter-brand competition constant, the price of the popular Honda
Accord changes by about $220 for every ten miles of distance between Honda
dealerships (at the mean distance). In contrast, the price of the lower priced
Honda Civic changes by about $70 for every ten miles of distance between
Honda dealerships. Given that retail margins on auto sales are quite small
(about 6% on average), the price reductions resulting from intra-brand
competition are substantial relative savings for new-car consumers.
While we hope that these results are an important addition to the analysis of
the retail auto industry, our findings may also have important policy
implications. The number of new car dealers in the U.S. has been in a steady
3
A discussion of the auto franchising laws is provided by F. Lafontaine and F.S. Morton,
Markets: State Franchise Laws, Dealer Terminations, and the Auto Crisis, 24 JOURNAL OF ECONOMIC
PERSPECTIVES 233-250 (2010). See also E.P. Kerrigan, Econ 101: Dealership Supply and Demand,
D
EALER-MAGAZINE (June 2009) (available at: http://autostarrealty.com/articles/erin_06-2009.pdf).
4 PHOENIX CENTER POLICY PAPER [Number 48
Phoenix Center for Advanced Legal and Economic Public Policy Studies
www.phoenix-center.org
decline for decades, reflecting economic forces such as increasing inter-brand
competition. Reductions in the number of dealerships, like those imposed by
Chrysler and General Motors in their 2009 reorganization,
4
could soften intra-
brand price competition in some markets. Also, the dealer franchise laws that
have historically tended to support large-numbers intra-brand competition are
presently being challenged in many states and at the federal level. Of course, the
effects of intra-brand competition are only a part of these very complex forces
impacting the structure of the automobile industry. Reductions in the number of
brand-specific dealers are driven by many factors including declining market
shares. Reforming state franchise laws has spurred a highly contentious debate
and involves numerous tradeoffs for existing dealers, new entrants like Tesla
Motors, local and state governments, and regional labor markets. We make no
effort to address all these varied concerns, but instead focus on a piece of this
very complex puzzle: intra-brand price competition.
Our P
APER is organized as follows. In Section II we offer some background
on automotive retailing to provide context. After a review of the economic
literature on competition in the industry in Section III, we turn to the empirical
analysis in Section IV. Conclusions are provided in Section V.
II. Background
In the early days of the car business, getting around wasn’t so easy, so
manufacturers wanted large dealer networks that would be close to the customer
and thus facilitate sales. In 1950, there were 47,000 new car dealers in the U.S., or
about 308 dealers per million persons.
5
Given improvements in transportation,
however, the number of retail automobile dealerships has declined steadily since
that time. By 2013 the number of new car dealers had fallen to 17,540, or about
55 dealers-per-million persons. This reduction has impacted mostly domestic
brands that have been shedding market share to international nameplates for
4
J. Puzzanghera and K. Bensinger, GM Proposes Painful Downsizing in Bid for Survival, LOS
ANGELES TIMES (April 28, 2009) (available at: http://articles.latimes.com/2009/apr/28/business/fi-
gm28); C. Clifford, Watchdog: Auto Dealers Shut Down Too Fast, CNN MONEY (July 18, 2010)
(available at: http://money.cnn.com/2010/07/18/smallbusiness/auto_dealership_report/
); R.
Reuteman, What's the Road Ahead for Shutdown U.S. Auto Makers, F
OX BUSINESS: SMALL BUSINESS
CENTER (October 27, 2009) (available at:
http://smallbusiness.foxbusiness.com/entrepreneurs/2009/10/27/rip-rebirth-detroit
).
5
NADA Historical Dealer Count (available at:
http://www.nada.org/Publications/NADADATA/historical_dealer_count.htm
).
Winter 2015] INTRA-BRAND COMPETITION IN THE AUTO INDUSTRY 5
Phoenix Center for Advanced Legal and Economic Public Policy Studies
www.phoenix-center.org
decades. The decline continues: over the past ten years, dealership counts have
fallen by about 2% annually. While the data show a substantial and continued
reduction in dealer counts, some analysts claim that there are still too many new
car dealers, especially for domestic brands with legacy dealer networks like GM,
Chrysler, and Ford.
6
Indeed, the established domestic brands have far more
dealerships than the more recent, and mostly foreign, entrants. For example, in
2008 Ford had 3,430 dealers to Toyota’s 1,235 dealers.
7
The average Ford dealer
sold 470 cars per year while the typical Toyota dealer sold 1,585, a difference
roughly proportionate to the dealer counts.
8
While consumers certainly benefit from inter-brand competition, that
competition is of a differentiated-products sort. Intra-brand competition, in
contrast, is over homogeneous or near-homogenous goods (e.g., a white Ford
Focus SE), so intra-brand rivalry focuses on price. Economic theory suggests and
consumer experience demonstrates that more dealers in close proximity mean
better deals for consumers. As one industry analyst observed,
dealers compete with each other within the brand. One Ford
dealer competes with the Ford guy a few miles away, rather than
with Toyota or Volkswagen. This intra-brand competition always
is on price. The customer runs from one dealer to another
dickering for a lower price ….
9
The National Automobile Dealers Association (“NADA”) puts it this way:
The ability of consumers to choose to comparison shop between
different dealerships in selling the same brand keeps prices
competitive and low. A Ford dealer’s biggest competitor is the
6
E.P. Kerrigan, Econ 101: Dealership Supply and Demand - The New Market Equilibrium,
D
EALER-MAGAZINE.COM (June 2009) (“We are still over-dealered. While the number of new car
dealerships has declined, the reduction is certainly not in keeping with the steep decline in new car
sales. In fact, since 2000, new car sales have declined five times more than the number of
dealerships.”) (available at: http://autostarrealty.com/articles/erin_06-2009.pdf
).
7
C. Isidore, GM Whacks 1,100 Dealers, CNNMONEY.COM (May 15, 2009) (available at:
http://money.cnn.com/2009/05/15/news/companies/gm_dealers
).
8
Id.
9
J. Flint, Too Many Dealers, Again?, WARD’S AUTOWORLD (September 2007) (available at:
http://wardsauto.com/news-amp-analysis/too-many-dealers-again
).
6 PHOENIX CENTER POLICY PAPER [Number 48
Phoenix Center for Advanced Legal and Economic Public Policy Studies
www.phoenix-center.org
Ford dealer down the street or in the next town. Because of broad
access to vehicle pricing, consumers can bargain with multiple
dealers to get a great deal.
10
In these views, as dealer counts fall, intra-brand price competition should soften
thereby raising new car prices for consumers.
The intensity of intra-brand price competition has become a significant policy
issue of late as a result of Tesla Motors’ efforts to own and operate its own
dealerships, an approach which violates the laws of almost all states.
11
Tesla’s
plan has met strong resistance from existing car dealers and some state
governments, not so much because of the threat of competition from Tesla
(which sells only about 30,000 extremely high-end electric cars per year), but
because of the threat of abandoning existing franchise laws protecting local car
dealers from manufacturer-driven shut downs.
12
In response to such challenges,
NADA lists, among other things, intra-brand competition as a benefit of such
laws.
13
Implicit in NADA’s defense of franchise laws is the seemingly reasonable
assumption that a weakening of dealer protections in the franchise laws will
result in a reduction of dealerships and, in turn, a lessening of intra-brand price
competition.
14
While the reform of the franchise laws and prohibitions on
10
Why Franchise Dealer Laws? Why States Promote the Buying and Selling of Cars through Local
Dealers, National Automobile Dealers Association (2014) (available at:
http://www.nada.org/NR/rdonlyres/39A36250-8BBA-41C0-9B15-
E1BB1A2E9EA9/0/NADA_Why_Dealer_Franchise_Laws_20140616.pdf).
11
See, e.g., C. Trudell and C. Smithe, U.S. Dealer Group Seeks Tesla Meeting on Retail Plans,
BLOOMBERG NEWS (October 24, 2012) (available at: http://www.bloomberg.com/news/2012-10-
23/dealer-group-leaving-tesla-retail-challenge-to-states.html); A. Ohnsman and M. Niquette,
Tesla’s Direct-Sales Push Raises Auto Dealers’ Hackles, B
LOOMBERG NEWS (March 10, 2014) (available
at: http://www.bloomberg.com/news/2014-03-10/tesla-s-direct-sales-push-raises-auto-dealers-
hackles.html); S. Alcorn, Car Dealers Sue Tesla, Citing State Franchise Laws, NRP.ORG (November 9,
2012) (available at: http://www.npr.org/2012/11/09/164736569/car-dealers-sue-tesla-citing-state-
franchise-laws).
12
A. Ohnsman, Tesla Rises After Model S Sales in 2013 Exceed Forecasts, BLOOMBERG NEWS
(January 15, 2015) (available at: http://www.bloomberg.com/news/2014-01-14/tesla-delivered-6-
900-cars-in-fourth-quarter-executive-says.html).
13
Get the Facts: The Benefits of Franchised Auto Dealers, National Automobile Dealers
Association (available at: http://www.nada.org/GetTheFacts).
14
According to Bill Wolters, President of the Texas Automobile Dealers Association
(TADA), easily two-thirds of Texas car dealerships
would be at risk if the law against manufacturer
(Footnote Continued. . . .)
Winter 2015] INTRA-BRAND COMPETITION IN THE AUTO INDUSTRY 7
Phoenix Center for Advanced Legal and Economic Public Policy Studies
www.phoenix-center.org
vertical integration involve many factors beyond the effects of such changes on
intra-brand competition, such competition is an important element of the debate.
Despite the evident relevance of intra-brand competition and the large
amount of empirical work that has been done on the automobile industry, there
is surprisingly little formal empirical analysis of the effects of intra-brand
competition. In fact, we were unable to find any formal statistical analysis—
published or otherwise—quantifying the price effect, if any, of intra-brand
competition. The absence of such evidence is a significant hole in the academic
literature on the automobile industry. As such, in what follows, we attempt to
shed some light on this important and interesting issue.
III. Literature Review
In this P
APER, we attempt to assess and quantify the effect of intra-brand
competition on the transaction prices for automobiles. Despite a wide search, we
were unable to find any empirical evidence focused specifically on this issue.
Earlier work has addressed the effects of competition (more broadly) on prices
and services, often using aggregate data. None of the studies, however,
attempted to quantify the effect of intra-brand competition using large samples
of actual customer transaction prices. While not directly relevant to our work,
the literature is nevertheless useful in some regards (e.g., model specification), so
a brief review is warranted.
An early paper on the topic by Bresnahan (1987) tests for competition and
collusion in the mid-1950s around an alleged price war in 1955, but the tests do
not involve intra-brand competition within a single market.
15
Berry, Levinsohn,
and Pakes (1995) develop an empirical model for analyzing demand and supply
for differentiated goods and applies it to the automobile industry.
16
While a
ownership of dealerships is eliminated. B. Blanchard, Tesla Motors, Car Dealers Clash Over Franchise
Law, THE TEXAS TRIBUNE (January 28, 2015) (available at:
https://www.texastribune.org/2015/01/28/tesla-motors-automobile-dealers-debate-franchise-l
).
15
T. Bresnahan, Competition and Collusion in the American Automobile Industry: The 1955 Price
War, 35 J
OURNAL OF INDUSTRIAL ECONOMICS 457-482 (1987).
16
S. Berry, J. Levinsohn, and A. Parkes, Automobile Prices in Market Equilibrium, 63
E
CONOMETRICA 841-890 (1995). On the issue of differentiation, see also P. Golberg, Product
Differentiation and Oligopoly in International Markets: The Case of the U.S. Automobile Industry, 63
E
CONOMETRICA 891-951 (1995); L. Thomas and K. Weigelt, Production Location Choice and Firm
(Footnote Continued. . . .)
8 PHOENIX CENTER POLICY PAPER [Number 48
Phoenix Center for Advanced Legal and Economic Public Policy Studies
www.phoenix-center.org
sophisticated and interesting study, it does not consider intra-brand competition
but focuses more on the attributes of cars (e.g., miles-per-gallon). Using data for
years 1981 through 1990, Sudhir (2001) considers strategic pricing among major
brands for compact and midsize market segments using a structural econometric
model. This paper is focused on competitive interactions across segments
(compact, midsize) and not intra-brand competition.
17
Alley (1997) uses a
conjectural variations model to test for collusion in the Japanese automobile
industry facilitated by partial-ownership arrangements over the period 1979
through 1994.
18
Closest to our analysis is a paper by Olivares and Cachon (2009), though their
analysis focuses on dealer inventories rather than prices and considers the
inventories of GM dealers only.
19
Competition is measured by a count of all
(major brand) dealerships in a market, though the authors test for a unique effect
of the presence of other GM dealers in the same market. The study finds that
competition generally, and intra-brand competition in particular, increases
inventory levels. This result is of interest in light of the claim that a reduction in
intra-brand competition will facilitate a rise in the quality of services offered by
dealerships.
20
The study provides evidence counter to that claim, but quality has
multiple dimensions beyond inventory. In any event, these findings motivate us
by highlighting important potential differences in the effects of inter- and intra-
brand rivalry in automobile retailing.
Studies that evaluate directly the determinates of automobile prices include
(but are not limited to) Eckard (1985), Rogers (1986), Mathewson and Winter
(1989), Walden (2005).
21
The focus of these studies is the price effects of franchise
Capabilities: Evidence from the U.S. Automobile Industry, 21 STRATEGIC MANAGEMENT JOURNAL 897-909
(2000).
17
K. Sudhir, Competitive Pricing Behavior in the Auto Market: A Structural Model, 20
M
ARKETING SCIENCE 42-60 (2001).
18
W. Alley, Partial Ownership Arrangements and Collusion in the Automobile Industry, 45
J
OURNAL OF INDUSTRIAL ECONOMICS 191-205 (1997).
19
M. Olivares and G. Cachon, Competing Retailers and Inventory: An Empirical Investigation of
U.S. Automobile Dealerships, 55 M
ANAGEMENT SCIENCE 1586-1604 (2009).
20
Olivares and G. Cachon, id.
21
E. Eckard Jr., The Effects of State Automobile Dealer Entry Regulation on New Car Prices, 24
ECONOMIC INQUIRY 223–42 (1985); R. Rogers, The Effects of State Entry Regulation on Retail Automobile
Markets, Bureau of Economics Staff Report to the Federal Trade Commission (1986); F. Methewson
(Footnote Continued. . . .)
Winter 2015] INTRA-BRAND COMPETITION IN THE AUTO INDUSTRY 9
Phoenix Center for Advanced Legal and Economic Public Policy Studies
www.phoenix-center.org
laws and not competition directly, an empirical analysis made possible by
variations in state laws regarding franchises in the past. Today, almost all states
prohibit manufacturer-owned dealerships, so this question is no longer amenable
to robust statistical testing. The results from these studies on franchise laws are
varied, and the methods of some of the earlier studies subject to potential
criticisms (Walden 2005). The latest study by Walden (2005) finds that the
franchise laws did not raise prices. While these studies, like ours, are focused on
price, our analysis is not directly concerned with the role of franchise laws on
prices but on intra-brand competition. Nevertheless, the empirical methods of
these earlier studies are useful in formulating an empirical model since they used
individual transaction data. Generally, these earlier studies model the
transaction price for an automobile as a function of market and car characteristics
and we do the same.
Recent research has focused on the role of the Internet on auto sales. By most
accounts, Internet use in the U.S. has done much to increase intra-brand
competition in the automotive industry. A study by Sewell and Bodkin (2009)
shows that the Internet does facilitate price shopping by consumers.
22
Unsurprisingly, with improved communications technologies, intra-brand
competition has intensified.
IV. Empirical Model
Our interest is in the role of intra-brand competition on transactions prices
for new cars sold in the United States. Detailed data on such transactions is not
readily available, but we were able to obtain data on transactions in the state of
Texas for three years (2011, 2012, and 2013) from IHS Automotive.
23
This data is
collected when an automobile is registered within the state, so it includes nearly
every automobile purchased and registered in the state. While some automobiles
are purchased in other states but registered in Texas, we limit our attention to
sales by Texas dealers to Texas customers (or customers registering their
and R. Winter, The Economic Effects of Automobile Dealer Regulation, 15/16 ANNALES D’ECONOMIE ET
DE
STATISTIQUE 409-426 (1989); M. Walden, Do Geographic Entry Restrictions Increase Car Prices, 35
REVIEW OF REGIONAL STUDIES 231-245 (2005).
22
E. Sewell and C. Bodkin, The Internet's Impact on Competition, Free riding and the Future of
Sales Service in Retail Automobile Markets, 35 E
ASTERN ECONOMIC JOURNAL 96-114 (2009).
23
https://www.ihs.com/industry/automotive.html. We are grateful to the National
Automobile Dealers Association for making the data available for our research.
10 PHOENIX CENTER POLICY PAPER [Number 48
Phoenix Center for Advanced Legal and Economic Public Policy Studies
www.phoenix-center.org
purchases in the state). Our measures of competition, however, may include
dealerships in neighboring states near a Texas border (New Mexico, Oklahoma,
Arkansas, and Louisiana). Rival dealership counts and distances are based on
the dealerships listed in the IHS sample; we assume it is inclusive of all
dealerships (at least those of any competitive significance). The data does not
indicate whether a car is new or used, so in an effort to limit our attention to new
cars we exclude any car with more than 200 miles on the odometer when sold.
The IHS Automotive data includes a variety of information about each
transaction including price, dealership name and address, make and model,
model year, and model specifics such as engine cylinders, number of doors, and
so forth. When available and relevant, our empirical analysis accounts for such
differences. The IHS data are not constructed specifically for the type of
empirical work we seek to perform so there are, unfortunately, some details
about the car that are not provided, including color, the presence of options like a
moonroof or special wheels. The transactions prices appear to include the effects
of a trade-in value but offer no details about those values. As with options and
premiums, we assume that the trade-in values are the source of a mis-
measurement in prices, but that such measurement problems are idiosyncratic
across dealerships. As such, the measurement error does not lead to inconsistent
or inefficient estimates of the model’s coefficients, with the exception perhaps of
the constant term, though the error variance of the model will be higher.
24
As far
as we can determine, these shortcomings in the data lead to undesirable but non-
fatal “noise” in the price series, although we utilize several estimation techniques
in an effort to reach conclusions on this point. This is explained in greater detail
below.
The brand of each dealership listed in the sample is determined from the
dealer’s name. Any dealership with a particular brand in the name (e.g., Toyota,
Honda, and so forth) is counted as a branded dealership.
25
Using the dealership
addresses provided in the data, we geocode each location using the program
geocode3 in STATA 13. This program links the address to the geocode
information using Google Maps. Distances between dealerships are computed
using the program geodist in STATA 13. The distances are based on surface
24
We have no reason to believe the measurement error has a mean of zero. J. Wooldridge,
E
CONOMETRIC ANALYSIS OF CROSS SECTION AND PANEL DATA (2002) at pp. 70-6.
25
As an alternative, we included any dealer that sold 50 more new cars of a brand
regardless of the dealer’s name. The results were not materially different from those reported here.
Winter 2015] INTRA-BRAND COMPETITION IN THE AUTO INDUSTRY 11
Phoenix Center for Advanced Legal and Economic Public Policy Studies
www.phoenix-center.org
curves, not driving distance. Using this data, we compute both the intra-brand
and inter-brand competition variables. Intra-brand competition (INTRACOMP)
for dealership j is measured as the distance (in miles) to the nearest same-brand
dealership. We cannot use distance to each inter-brand competitor because of
the large number of such rivals. Thus, inter-brand competition (INTERCOMP)
for dealership j is a count of rival dealerships within a thirty mile radius of the
address of dealership j, excluding same brand dealerships (Olivares and Cachon
2009). We view INTERCOMP as a measure of the general competitiveness of the
retail automobile market in the vicinity of dealership j.
The difference in the definition of the competition variables is also
statistically helpful (if not essential) in that the number of intra- and inter-brand
dealerships within a given mileage radius are highly correlated.
26
Dealership
counts across the brands appear to be driven by essentially the same factors,
which is not surprising. Given the high correlation, it is nearly impossible to get
an independent estimate of the price effect of the two when using a count
variable for both types of competition.
27
Ideally, the two measures of
competition would be similarly defined so that direct comparisons could be
made and relative impacts compared. The nature of the data, however, prohibits
us from doing so.
There are thousands of different automobile models included in the dataset.
We choose to limit attention to the ten most popular automobiles purchased over
the relevant time period. The ten most popular cars (excluding trucks) in the
dataset (not in order) are the Toyota Camry and Corolla, the Nissan Altima and
Rogue, the Honda Accord, CRV and Civic, the Hyundai Elantra, the Ford Focus,
and the Mazda 3. In choosing these models, we limit our attention to the most
popular sub-models in an effort to reduce variations in features for which we
have no data.
26
For Toyota dealers, the correlation coefficient between intra- and inter-brand rivals is 0.93.
The correlation coefficient between INTERCOMP and INTRACOMP for Toyota dealers is -0.49,
which while still high, is more manageable. The variance inflation factors for the count variables
are in excess of 5.0, indicating very high collinearity.
27
The Variance Inflation Factors were very large when using counts for both measures of
competition, and, as expected, the estimated coefficients very unstable.
12 PHOENIX CENTER POLICY PAPER [Number 48
Phoenix Center for Advanced Legal and Economic Public Policy Studies
www.phoenix-center.org
Table 1. Make, Model, and Prices
Make/Model Obs. Price
(Mean)
Price
(St Dev)
MSRP less
Invoice*
Price-Cost
Margin*
Toyota Camry 42,728 23,801 2,420 2,100 9%
Toyota Corolla 35,266 18,502 1,812 1,300 7%
Nissan Altima 2.5 33,258 23,875 3,204 1,500 6%
Nissan Rogue 24,372 23,605 2,926 1,100 5%
Hyundai Elantra Sedan 19,163 19,744 2,088 900 4%
Honda Civic LX 22,400 18,986 1,564 1,300 7%
Honda Accord EX 21,925 27,695 2,935 2,200 8%
Honda CRV EX 20,546 27,214 2,398 1,500 6%
Ford Focus SE 16,291 18,555 1,986 1,100 6%
Mazda 3i 14,737 18,855 2,370 1,100 6%
* Source (
www.cars.com), Year 2012. Price-cost margins based on MSRP and dealer invoice and not
actual transactions prices.
Table 1 lists the models included in the sample and some descriptive
statistics. All ten cars are mid-sized and moderately priced. The standard
deviation in prices is about 10% of the mean price. With one exception, we do
not trim the data on prices.
28
While some prices appeared to be quite low, such
cases are relatively rare and could be explained by the trade-in values. Also, two
of our statistical methods address extreme values (i.e., Robust Regression and the
Finite Mixture Model), so there is less motivation to trim for outliers in this data.
With this in mind, we believe the results from these alternative estimation
methods are more reliable.
The final two columns of Table 1 provide rough estimates of the dealer’s
typical price-cost margin on each model.
29
All the data is for the 2012 model
year. The margin data is illustrative and based on the MSRP and Invoice data for
models most closely matching those in our sample. This margin data is not
based on actual sale prices and the invoice is not a perfect indicator of actual
dealer costs. Dealers are also multi-product firms and so the sale of an
automobile may lead to subsequent service revenues, though the buyer is not
limited to any particular dealer for service. There also may be other dealer
28
One observation for a Toyota Camry priced at over $200,000 was excluded on the grounds
it was a coding error.
29
All data is from www.cars.com for the Year 2012.
Winter 2015] INTRA-BRAND COMPETITION IN THE AUTO INDUSTRY 13
Phoenix Center for Advanced Legal and Economic Public Policy Studies
www.phoenix-center.org
incentives from the manufacturer. Used cars often have higher margins than
new cars. AutoNation, a large publicly-traded dealer conglomerate, reports an
average gross margin on new car sales of about 6% (on an average sales price of
$33,967), so the figures in Table 1 are plausibly representative.
30
We note,
however, that industry data suggests that the profitability of new car sales is not
robust and that the contribution to the dealers’ profits is often negative.
31
Automobile dealers make most of their profits from the service department and
from used car sales. This fact by itself highlights the competitive nature of
automobile retailing.
Table 2. Descriptive Statistics, Competition Variables
Make INTRACOMP (Miles)
Mean, St Dev, Min, Max
INTERCOMP (Count)
Mean, St Dev, Min, Max
Toyota 18.7 , 25.8 , 3.9 , 137.1 157.2 , 96.8 , 0 , 313
Nissan 23.40 , 28.6 , 4.5 , 137.6 159.3 , 102.3 , 0 , 314
Honda 19.2 , 28.9 , 3.5 , 212.1 180.0 , 90.3 , 5 , 306
Hyundai 21.2 , 21.8 , 2.0 , 116.5 144.7 , 92.2 , 13 , 301
Ford 12.2 , 8.2 , 2.4 , 74.7 134.5 , 103.4 , 0 , 310
Mazda 26.7 , 41.4 , 3.9 , 236.5 149.3 , 97.5 , 8 , 304
Table 2 summarizes the competition variables for each brand. For most
brands, same-brand dealerships (measured using INTRACOMP) are similarly
spaced geographically at about 16 to 20 miles, on average. Ford is the exception,
with an average distance between dealerships of about 11 miles. The maximum
distance between Ford dealerships is 75 miles, but for the other brands it is
nearly double that amount. The rough average number of different brand rivals
within a thirty-mile radius for dealers in the sample is about 150 rivals. There
are instances of zero rivals in the radius and the maximum for each is about 300
dealers.
32
Additional variables include dummy variables for the model year (D2012
and D2013). For some models, there are dummy variables to account for a six
30
AutoNation S.E.C. Form 10-K (2013), at p. 26. The average margin on a used car is about
9% (for an average sales price of $18,079).
31
Annual Financial Profile of America’s Franchised New–Car Dealers, National Automobile
Dealers Association (2014) (available at: http://www.nada.org/NR/rdonlyres/DF6547D8-C037-
4D2E-BD77-A730EBC830EB/0/NADA_Data_2014_05282014.pdf) at p. 7.
32
Counting only new dealers had little impact on the econometric estimates in light of the
high correlation of that count with the count of all dealers (which is very close to 1.0).
14 PHOENIX CENTER POLICY PAPER [Number 48
Phoenix Center for Advanced Legal and Economic Public Policy Studies
www.phoenix-center.org
cylinder engine (6CYL), a sedan or hatchback body style (SEDAN, HBACK), and
an all wheel drive transmission (AWD). Last year’s models are often discounted,
so we include a variable measuring the difference between the year purchased
less the model year (BUYLATE). Larger values of this variable should be
associated with lower prices. Most cars in the sample are purchased in the
model year (82%), with about 15% purchased in the year prior and 3.4% in the
year after the model year. Finally, using the zip code of each dealership, we
added data on median income to measure the opportunity cost of driving and
shopping for a car (INCOME) and the preferences for lower- or higher-priced
automobiles.
A. Statistical Model
Given the large sample sizes we estimate the competition effects for each car
independently. We note that all our sales occur in Texas, and are therefore
subject to uniform regulations and legal requirements. This approach also
permits us to easily measure differences in the competitive effects across brands
and models, if any. Using the price for each car sale (P), the econometric model
takes the general form
BUYLATEINCOMEDD
INTERCOMPINTRACOMPP
5432
10
ln20132012
lnlnln
(1)
where is the econometric disturbance term and “ln” indicates the natural log
transformation. In some cases, the regression will include additional variables to
account for differences like engine size and so forth. The model is in log-log
format so the coefficients are estimates of elasticities. Given the short-run nature
of our study, we view industry structure as exogenous.
The coefficient of primary interest is , which measures the effect, if any, of
intra-brand competition. Intra-brand competition is measured as miles to nearest
same-brand dealer, so if intra-brand competition lowers prices then the
coefficient will be positive (the greater the distance, the higher the price). We
expect
1
< 0 (markets will more inter-brand competition have lower prices) and
5
< 0 (buying “last year’s” model will lower price). Based on opportunity costs,
we would expect
4
to be positive (higher opportunity costs lead to less search),
but since income may also influence preferences for particular cars we make no
prediction about the sign of the variable. We also make no predictions about the
signs of the coefficients on the year dummies. Additionally, six-cylinder engines
and all wheel drive transmission should sell for higher prices. For some models,
Winter 2015] INTRA-BRAND COMPETITION IN THE AUTO INDUSTRY 15
Phoenix Center for Advanced Legal and Economic Public Policy Studies
www.phoenix-center.org
hatchback models are dearer, so we have no general expectation for the sedan
and hatchback dummy variables.
As a benchmark, we estimate the model using Ordinary Least Squares
(“OLS”). Given evidence of heteroskedasticity, we compute robust standard
errors for testing. Also, given heteroskedasticity and the fact that trade-ins and
upgrades give some sales the appearance of outliers, we also estimate the model
using Robust Regression (“RREG”). The RREG procedure first screens for gross
outliers using Cook’s distance and then performs Huber iterations followed by
biweight iterations to produce the estimated coefficients.
33
As an alternative to
robust regression, we estimate the equation as a Finite Mixture Model
(“FMM”).
34
Given the nature of the data, we believe there are transaction prices
that are somewhat “normal” and others that are “extreme,” perhaps reflecting
special circumstances related to trade-in values, the addition of premium
upgrades, and so forth. The FMM divides the sample into sub-populations
without any sub-population identity information, relying solely on statistical
computations. We permit two components expecting the procedure will
separate the extreme prices in the distribution from the more “normal” prices.
This expectation is largely confirmed, though the bulk of the data is typically
placed into one component. In most cases, the “extreme” component is
identified to be very small, and we limit our attention the “normal”
subpopulation. In others, however, both subpopulations are sizeable shares of
the whole. In these cases, we provide the results for both subpopulations.
B. Results
All statistical estimates are produced using STATA 13. While we begin with
the Ordinary Least Squares estimates, we believe the nature of the data makes
the results from Robust Regression more reliable. Detailed estimates of the
models are provided in an Appendix. In the text, we present the results in a
more useful form by using the estimated coefficients for intra-brand competition
to compute the price differences based on mileage differences. For these tables,
33
This approach is available as a default in STATA 13 and follows G. Li, Robust regression, in
E
XPLORING DATA TABLES, TRENDS, AND SHAPES (D.C. Hoaglin, C.F. Mosteller, and J.W. Tukey eds.) at
pp. 281–340.
34
Estimates are produced using the “fmm” package in STATA 13. See also G.J. McLachlan
and D. Peel, F
INITE MIXTURE MODELS (2000); P. Deb and P. K. Trivedi, Demand for Medical Care by the
Elderly: A Finite Mixture Approach, 12 J
OURNAL OF APPLIED ECONOMETRICS 313-326 (1997).
16 PHOENIX CENTER POLICY PAPER [Number 48
Phoenix Center for Advanced Legal and Economic Public Policy Studies
www.phoenix-center.org
we set the minimum distance to 2.5 miles and compute the predicted average
transaction price at this distance. The other values in the table measure the price
difference from the price at 2.5 miles for other distances. For example, if the table
shows $100 for 5 miles, then the price at a dealer with its closest same-brand rival
at 5 miles would be $100 more than the price at a dealer with a same-brand rival
at 2.5 miles. We choose 2.5 miles based on the minimum values reported in
Table 2.
We note that for nine of the ten cars we consider the coefficient on intra-
brand competition is positive and statistically-significant at standard levels (5%
or better). The only exception is the Ford Focus, where the coefficient on intra-
brand competition is not statistically different from zero. We had other problems
with the Ford Focus data (e.g., non-convergence of the FMM), so we suspect
there may be some unmeasured factor with a strong influence on the observed
prices (e.g., fleet sales). The coefficient on intra-brand competition is also not
statistically different from zero for the Honda Civic in the OLS model, but it is
statistically significant in the RREG and FMM estimations, perhaps revealing the
impact of outliers on the OLS results.
1. Ordinary Least Squares
The results from the OLS estimation of Equation (1) are summarized in the
Appendix as Table A-1. Robust t-statistics are provided in parenthesis. The F-
statistics of all regressions are statistically-significant at better than the 1% level.
For the most part, all signs are as expected. Of most interest is the consistently
positive and statistically significant sign on the INTRACOMP variable
(coefficient from Equation 1), with the exception of the Ford Focus and the
Honda Civic. The estimates indicate that intra-brand competition lowers prices.
Also, the coefficient on INTERCOMP is negative in most cases and statistically
different from zero for eight of ten cars. Positive signs are seen for the CR-V and
the Mazda 3, with generally weak and variable statistical significance. Buyers
consistently pay less when buying last year’s model (BUYLATE), with negative
and statistically-significant coefficients on the variable in nine of ten cases. A six
cylinder engine and an all wheel drive transmission bring higher prices. Sedans
sometimes sell for more, but not always. The coefficients on the model year
dummy variables vary by model and year. No regular pattern emerges. Income
also has both positive and negative signs, and is statistically different from zero
in almost all cases. Areas with higher incomes tend to pay more for a Honda or a
Toyota Camry, but less for Nissans and the other cars in the sample. More
details on the marginal effects of these variables in provided in the discussion of
the RREG results.
Winter 2015] INTRA-BRAND COMPETITION IN THE AUTO INDUSTRY 17
Phoenix Center for Advanced Legal and Economic Public Policy Studies
www.phoenix-center.org
Table 3. Marginal Effects Intra-brand Competition (OLS)
(Values measure differences in dollars from price at 2.5 miles)
Miles to Nearest Intra-brand Rival
Price
at 2.5
miles
5 10 15 20 25 30 40 50 60
Camry 23,584 43 86 111 129 143 154 172 186 197
Corolla 18,227 89 178 231 269 297 321 359 387 411
Altima 22,823 341 684 891 1,038 1,150 1,243 1,392 1,506 1,600
CR-V 26,778 143 285 370 430 476 514 574 620 658
Elantra 19,386 100 200 259 301 333 360 402 434 461
Rogue 23,000 166 332 432 502 556 600 671 725 770
Accord EX 27,195 155 310 402 468 518 559 625 675 716
Civic LX* 18,915 3 6 8 9 10 11 12 13 14
Focus SE* 18,386 29 59 76 88 98 106 118 127 135
Mazda 3i 18,279 183 367 477 556 615 664 743 804 853
* Regression coefficient on intra-comp is not statistically-significant.
Table 3 summarizes the marginal effect on price from differences in the
INTRACOMP variables. Prices are computed at ten different mileages and then
subtracted from the computed price at 2.5 miles, which is approximate to the
minimum distances in Table 2. So, a Hyundai dealer with a same-brand rival 2.5
miles away will charge a price $461 less for an Elantra than a dealer with a same-
brand rival 60 miles away. This is a 2.4% discount off the sales price of the
Elantra. Considering the low margins on an Elantra (about 4%, from Table 1),
this discount is sizeable for the dealer as well as the customer. Of these ten
models, the discounts are largest for the Altima ($1,600 at the 60-mile mark),
which is one of the higher-priced cars in the group. For the Altima, the
difference between MSRP and Invoice is about $1,500, so the effect of
competition is substantial. Mileage-based discounts are the lowest for the Civic
and Focus, but the difference between MSRP and Invoice is only about $1,200 for
these models.
For both the Honda Civic and Ford Focus, the sign on the INTRACOMP
variable is positive. Thus, price declines if an intra-brand competition is more
proximate. Neither coefficient is statistically different from zero at standard
levels. For the Civic, the predicted price differentials in Table 3 are quite small.
We note, however, that the results from the Civic are much different in the RREG
and FMM estimation approaches, suggesting outliers are influencing the OLS
results. In none of the estimations is intra-brand competition statistically-
significant for the Ford Focus. However, the Focus is an outlier in the statistics in
many respects (perhaps due to fleet sales).
18 PHOENIX CENTER POLICY PAPER [Number 48
Phoenix Center for Advanced Legal and Economic Public Policy Studies
www.phoenix-center.org
2. Robust Regression
Estimating the model using Robust Regression renders some changes relative
to OLS; some of these changes are material, some are not. A summary of the
estimates are provided in Tables A-2. The largest changes of interest are for the
INTRACOMP variable for the Honda Civic, the Honda Accord, and the Ford
Focus. For the Civic, the coefficient on INTRACOMP increased by a factor of
about 26 (0.0002 from 0.006), and for the Accord the coefficient rises from 0.008 to
0.014. The coefficient on the Focus is nearly halved (0.0023 from 0.0014), but it
remains poorly estimated and not statistically different from zero. Other than
these three cases, the estimates are quite comparable across RREG and OLS.
Given the nature of the data, we believe the RREG results are more reliable so
we provide additional details on the marginal effects of the non-competition
related variables. The BUYLATE variable is in all regressions so we start there.
For the Honda Civic (which has roughly the mean coefficient for BUYLATE),
buying last year’s model (BUYLATE = 1) saves the buyer about $850. The buyer
would pay about $810 more to buy next year’s model.
The coefficients on the dummy variables can be converted into price effects
using the formula exp() – 1. A six cylinder in an Accord will cost you 12%
more, or about $3,100. In 2012, the six cylinder Accord EX had an MSRP of about
$3,000 more than the four cylinder, so the estimated differential seems accurate.
35
For the Toyota Camry, however, the six cylinder model is estimated to run about
$6,300 more than the four cylinder model. This differential is most consistent
with the MSRP spread between the lowest four-cylinder and highest six-cylinder
models, so we suspect this price differential reflects the reality that the six
cylinder models typically include features such as leather interiors, moonroofs,
and other premium features that are unaccounted for in the data.
36
35
http://www.cars.com/honda/accord/2012.
36
http://www.cars.com/toyota/camry/2012.
Winter 2015] INTRA-BRAND COMPETITION IN THE AUTO INDUSTRY 19
Phoenix Center for Advanced Legal and Economic Public Policy Studies
www.phoenix-center.org
Table 4. Marginal Effects Intra-brand Competition (RREG)
(Values measure differences in dollars from price at 2.5 miles)
Miles to Nearest Intra-brand Rival
Price
at 2.5
miles
5 10 15 20 25 30 40 50 60
Camry 23,507 55 111 144 167 185 199 223 240 255
Corolla 18,176 98 197 255 297 328 354 396 428 454
Altima 22,756 347 697 908 1058 1173 1267 1420 1536 1632
CR-V 26,623 189 378 491 571 632 683 763 825 875
Elantra 19,276 100 199 258 300 332 359 401 433 459
Rogue 22,895 170 340 442 514 569 614 687 742 787
Accord EX 26,810 275 550 715 832 922 995 1113 1204 1278
Civic LX 18,613 82 164 213 248 275 296 331 358 379
Focus SE* 18,342 18 35 46 53 59 64 71 77 81
Mazda 3i 18,220 178 357 464 540 598 645 722 781 829
* Regression coefficient on intra-comp is not statistically-significant.
Table 4 provides the marginal price effects of intra-brand competition. The
estimated price effects are very similar to those from the OLS estimates
summarized in Table 3 for all but the Civic, Accord, and Focus. For the Civic,
the 60-mile price increase is now $379 where under OLS it was only $14 (the
latter being statistically insignificant). Likewise, for the Accord, the price
differential at 60 miles rises from the OLS-based $716 to the RREG-based $1,278.
In contrast, the OLS-based $135 price increase at 60 miles for the Ford Focus is
now only $81. We generally view the RREG estimates are more reliable than the
OLS estimates given the influence of trade-ins and other factors the model cannot
account for given the data. The differences between OLS and RREG suggests
these factors are perhaps material, but even with these few large differences the
overall influence of intra-brand competition is unchanged—intra-brand
competition lowers new car prices for consumers.
3. Finite Mixture Model
The Finite Mixture Model is the most complex of our three estimation
methods. This approach divides the sample into two components. A review of
the estimates indicates that the procedure typically divided the sample into an
“extreme” price and “normal” price components, with the extreme group
including both very low and very high prices. In many cases, one component
contains almost all of the data. We reviewed the distribution of prices across the
two components to confirm that the procedure effectively parsed the data into a
“normal” and “extreme” sub-populations and the procedure was generally
effective.
20 PHOENIX CENTER POLICY PAPER [Number 48
Phoenix Center for Advanced Legal and Economic Public Policy Studies
www.phoenix-center.org
The econometric estimates for the “normal” component of the sample are
similar to the estimates from both the OLS and RREG procedures. We expect the
results to be more similar to the RREG approach, since both procedures address
outliers. For many of the cars, the extreme component was very small and full of
outliers, leading to poor estimates and nonsensical forecasts of prices. We ignore
those results. For two models—the Nissan Rogue and the Hyundai Elantra—the
observations were more evenly divided between the components, but still about
two-thirds of the data was assigned to one component, so we report the results
from the larger component. The FMM did not converge for the Ford Focus, so
we exclude those results from further discussion. A summary of the estimates
are provided in Table A-3, and the marginal effects are for the “normal”
component are provided in Table 5.
Table 5. Marginal Effects Intra-brand Competition (FMM)
(Values measure differences in dollars from price at 2.5 miles)
Miles to Nearest Intra-brand Rival
Price
at 2.5
miles
5 10 15 20 25 30 40 50 60
Camry 23,585 48 95 123 143 159 171 191 206 219
Corolla 18,225 88 175 227 264 292 315 352 380 403
Altima 22,842 333 668 870 1,014 1,123 1,213 1,359 1,470 1,561
CR-V 26,376 247 494 641 757 827 892 998 1,079 1,145
Elantra 19,972 122 243 315 367 406 438 489 529 561
Rogue 24,097 136 272 353 410 454 490 548 592 628
Accord EX 26,690 305 611 794 925 1,025 1,106 1,238 1,339 1,421
Civic LX 18,319 142 284 369 429 475 513 573 620 657
Focus SE* … … … … … … … … … …
Mazda 3i 17,728 245 491 639 745 825 891 998 1079 1146
* Model would not converge.
There are some changes across the estimation approaches. The marginal
price effect for Civic has risen again to $657 at the 60-mile mark, up from $378
under RREG, again suggesting the Civic data contains a good deal of noise.
Convergence of the FMM was not obtained for the Ford Focus; the estimates for
the Focus also changed substantially between the OLS and RREG estimates. We
suspect that fleet car sales for the Focus may be influencing the results, but are
unable to test this hypothesis with our data. The marginal effect for the Accord
under FMM ($1,421) is close to the RREG value ($1,278), which was much higher
than the OLS value. For both the Elantra and the Rogue, both of which have a
larger second component, the marginal price effects are similar to those found for
RREG.
Winter 2015] INTRA-BRAND COMPETITION IN THE AUTO INDUSTRY 21
Phoenix Center for Advanced Legal and Economic Public Policy Studies
www.phoenix-center.org
C. Summary
On balance, we feel it is safe to conclude that intra-brand competition is
effective in reducing the prices consumers pay for the most popular cars.
Further, the extent of these effects often varies substantially with the distance the
buyer must travel to visit same make rival dealer. When distances increase,
consumers can pay hundreds of dollars more for the same car. This density
effect is strong and robust across all makes and models in our sample.
While it is not possible to make a direct comparison of the relative effects of
intra- versus inter-brand competitors, we can use the estimates to compute a
rough equivalence across mileage and inter-brand competitors. To do so, we
focus on the RREG results since we believe them to be the most meaningful. The
question we seek an answer for is this: if the closest intra-brand competitor was
one mile closer, how many inter-brand competitors could be lost without any
change in price? Given the log-log form of the regression, the answer depends
on the starting point for both variables, so we choose the means for each variable
for our computations. These computations are based on non-linear marginal
changes (i.e., linear in the logs) from the mean and thus are approximations.
Table 6. Tradeoff between Intra- and Inter-
brand Competition
Car Model
Inter-brand Rivals:
Intra-brand
Mileage
Toyota Camry 4:1
Toyota Corolla 43:1
Nissan Altima 85:1
Hyundai Elantra 9:1
Nissan Rogue 19:1
Honda Accord EX 78:1
Honda Civic LX 8:1
The results of these computations for each car model are summarized in
Table 6. In cases where the competitive effect of either intra-brand competition
or inter-brand was not statistically significant (the Ford Focus and Mazda 3i), or
inter-brand competition had the wrong sign (the Honda CR-V), the tradeoff
calculation is excluded from the table. For the Camry, moving the nearest intra-
brand competitor one-mile closer reduces prices by the equivalent of an increase
in four inter-brand competitors. The tradeoff is much larger for the budget-
conscious Toyota Corolla, where bringing a Toyota dealer one mile closer has the
same effect as adding 43 inter-brand dealers to the market. As shown above, the
marginal effect of intra-brand competition is very large for the Honda Accord.
22 PHOENIX CENTER POLICY PAPER [Number 48
Phoenix Center for Advanced Legal and Economic Public Policy Studies
www.phoenix-center.org
Locating a dealer one mile closer to a Honda dealer reduces price by as much as
adding 78 inter-brand rivals. Weighting the tradeoffs by sample sizes to
approximate an “average” effect, the average tradeoff between intra- and inter-
brand competition is about 35 inter-brand competitors (about 25% of the average
inter-brand count) for every one-mile of distance.
V. Conclusion
Cars are expensive and most consumers know to shop dealers for the best
prices. Yet, there is little to no empirical evidence on the price effects of intra-
brand competition. Using large samples of transactions for ten of the most
popular new cars purchased in the state of Texas for the years 2011, 2012, and
2013, we estimate the effects of intra-brand competition on prices. Intra-brand
competition is measured as the distance (in miles) to the nearest same-brand
dealer. Causal observation suggests and basic economic logic predicts that intra-
brand competition should lower prices; for all automobiles we consider in our
empirical analysis, we find that intra-brand competition does, in fact, lower new
car prices for consumers. The price effects of intra-brand competition is
relatively strong—at the sample means, moving an intra-brand dealer one mile
closer reduces prices by the equivalent of an increase in 35 inter-brand rivals, or
about 25% of the inter-brand competitors in the average market.
Given the paucity of empirical evidence on intra-brand competition, this
P
APER is much needed addition to the economic analysis of the automobile
industry. The results are also of economic and public policy relevance given the
continuing decline in dealerships over time and the present calls to reform the
state franchise laws which tend to increase the number of dealerships above
what manufacturers may prefer. Our findings suggest that reductions in the
number of dealerships are likely to lead to higher prices for new cars. We stress,
however, that the dynamics of the automobile industry are complex and our
findings shed light on but a sliver of the relevant factors impacting the industry.
Winter 2015] INTRA-BRAND COMPETITION IN THE AUTO INDUSTRY 23
Phoenix Center for Advanced Legal and Economic Public Policy Studies
www.phoenix-center.org
Table A-1. OLS Results
Variable Camry Corolla Altima CR-V Elantra
lnINTRACOMP 0.003**
(2.82)
0.007***
(7.87)
0.022***
(16.02)
0.008***
(5.12)
0.007***
(8.59)
lnINTERCOMP -0.011***
(-13.92)
-0.001*
(-1.65)
-0.002*
(-1.77)
0.002
(1.24)
-0.007***
(-7.10)
D2012
0.068***
(54.45)
-0.032***
(-23.74)
-0.016***
(-7.50)
0.025***
(14.26)
-0.001
(-0.32)
D2013
0.026***
(18.88)
-0.037***
(-28.24)
0.045***
(22.42)
0.031***
(17.60)
-0.029***
(-13.23)
lnMEDIAN 0.005***
(3.65)
-0.008***
(-5.29)
-0.039***
(-16.72)
0.031***
(17.69)
0.003
(1.24)
BUYLATE
-0.014***
(-9.44)
-0.022***
(-14.65)
-0.037***
(-22.82)
-0.030***
(-14.46)
-0.038***
(-23.62)
6CYL
0.235***
(5.93)
… … … …
SEDAN
… … … … …
HBACK
… … … … …
AWD
… … … … …
Constant 10.02***
(640.8)
9.923***
(585.8)
10.425***
(373.5)
9.819***
(445.5)
9.864***
(351.5)
Obs. 42,728 36,065 33,203 20,546 19,163
R
2
0.10 0.03 0.09 0.05 0.06
Stat. Significance: *** (1%); ** (5%); * (10%)
24 PHOENIX CENTER POLICY PAPER [Number 48
Phoenix Center for Advanced Legal and Economic Public Policy Studies
www.phoenix-center.org
Table A-1. OLS Results
Variable Rogue Civic Accord Focus 3i
lnINTRACOMP 0.010***
(7.63)
0.0002
(0.17)
0.008***
(4.64)
0.002
(1.43)
0.014***
(11.00)
lnINTERCOMP -0.005***
(-4.06)
-0.009***
(-8.36)
-0.005***
(-3.71)
-0.003***
(-3.30)
0.003**
(2.30)
D2012
0.028
(1.39)
0.029***
(14.53)
-0.023***
(-13.27)
0.126***
(47.43)
0.046***
(20.69)
D2013
-0.044***
(-22.63)
0.055***
(27.34)
0.051***
(32.29)
0.088***
(32.32)
0.019***
(7.57)
lnMEDIAN 0.032***
(12.82)
0.034***
(21.69)
0.018***
(10.74)
-0.023***
(-8.20)
-0.030***
(-7.80)
BUYLATE
-0.038***
(-15.84)
-0.044***
(-28.09)
-0.031***
(-18.86)
-0.037***
(-22.29)
0.003
(1.29)
6CYL
… … -0.008***
(-5.31)
… …
SEDAN
… 0.003**
(2.29)
0.112***
(95.74)
… …
HBACK
… … … 0.044***
(27.31)
0.115***
(49.26)
AWD
0.075***
(22.17)
… … … …
Constant 10.41***
(348.8)
9.481***
(492.9)
9.971***
(434.9)
9.960***
(336.3)
10.061***
(251.8)
Obs. 22,294 22,400 21,925 16,163 14,737
R
2
0.08 0.11 0.34 0.24 0.17
Stat. Significance: *** (1%); ** (5%); * (10%)
Winter 2015] INTRA-BRAND COMPETITION IN THE AUTO INDUSTRY 25
Phoenix Center for Advanced Legal and Economic Public Policy Studies
www.phoenix-center.org
Table A-2. RREG Results
Variable Camry Corolla Altima CR-V Elantra
lnINTRACOMP 0.003***
(3.65)
0.008***
(8.18)
0.022***
(16.22)
0.010***
(6.68)
0.007***
(7.87)
lnINTERCOMP -0.011***
(-14.40)
-0.002***
(-2.71)
-0.002**
(-1.99)
0.003**
(2.17)
-0.006***
(-6.28)
D2012
0.069***
(57.40)
-0.033***
(-22.55)
-0.016***
(-7.23)
0.026***
(14.91)
0.000
(-0.12)
D2013
0.025***
(19.02)
-0.039***
(-28.42)
0.044***
(20.69)
0.032***
(18.24)
-0.030***
(-12.53)
lnMEDIAN 0.005***
(3.66)
-0.009***
(-6.03)
-0.037***
(-15.46)
0.030***
(16.98)
0.002
(0.83)
BUYLATE
-0.013***
(-9.42)
-0.023***
(-15.87)
-0.038***
(-22.78)
-0.029***
(-13.86)
-0.036***
(-22.05)
6CYL
0.65***
(8.34)
… … … …
SEDAN
… … … … …
HBACK
… … … … …
AWD
… … … … …
Constant 10.02***
(667.2)
9.939***
(588.8)
10.405***
(365.4)
9.811***
(433.7)
9.868***
(341.8)
Obs. 42,728 36,065 33,203 20,546 19,163
Stat. Significance: *** (1%); ** (5%); * (10%)
26 PHOENIX CENTER POLICY PAPER [Number 48
Phoenix Center for Advanced Legal and Economic Public Policy Studies
www.phoenix-center.org
Table A-2. RREG Results
Variable Rogue Civic Accord Focus 3i
lnINTRACOMP 0.011***
(8.18)
0.006***
(5.65)
0.018**
(13.86)
0.001
(0.87)
0.014***
(11.09)
lnINTERCOMP -0.004***
(-3.54)
-0.008***
(-7.96)
-0.002***
(-2.41)
-0.004***
(-4.95)
0.001
(1.01)
D2012
0.002
(1.17)
0.031***
(19.18)
0.022***
(14.88)
0.131***
(50.41)
0.044***
(18.25)
D2013
-0.047***
(-23.52)
0.057***
(34.72)
0.053***
(37.64)
0.089***
(34.80)
0.014***
(5.67)
lnMEDIAN -0.027***
(-10.76)
0.032***
(23.57)
0.018***
(11.78)
-0.024***
(-10.70)
-0.025***
(-6.34)
BUYLATE
-0.039***
(-15.87)
-0.046***
(-39.95)
-0.030***
(-19.37)
-0.039***
(-24.21)
0.004
(1.36)
6CYL
… … -0.006***
(-5.03)
… …
SEDAN
… 0.004***
(3.26)
… … …
HBACK
… … … 0.045***
(29.49)
0.114***
(46.45)
AWD
0.079***
(22.94)
… … … …
Constant 10.36***
(342.2)
9.467***
(549.0)
9.938***
(554.3)
9.969***
(421.4)
10.014***
(248.5)
Obs. 24,294 22,400 23,832 16,163 14,737
Stat. Significance: *** (1%); ** (5%); * (10%)
Winter 2015] INTRA-BRAND COMPETITION IN THE AUTO INDUSTRY 27
Phoenix Center for Advanced Legal and Economic Public Policy Studies
www.phoenix-center.org
Table A-3. FMM Results
Variable Camry Corolla Altima CR-V Elantra (1) Elantra (2)
lnINTRACOMP 0.003***
(3.31)
0.007***
(7.45)
0.021***
(15.83)
0.013***
(7.91)
0.002***
(2.59)
0.009***
(6.43)
lnINTERCOMP -0.010***
(-14.60)
-0.002**
(-2.08)
-0.003**
(-2.29)
0.003**
(2.39)
-0.011***
(-5.03)
-0.007***
(-3.34)
D2012
0.069***
(58.69)
-0.032***
(-22.49)
-0.016***
(-7.64)
0.027***
(15.14)
-0.013**
(-2.12)
0.000
(0.01)
D2013
0.026***
(20.07)
-0.038***
(-27.81)
0.045***
(22.55)
0.034***
(18.34)
-0.031***
(-2.91)
-0.035***
(-4.73)
lnMEDIAN 0.005***
(4.06)
-0.002***
(-5.60)
-0.039***
(-16.87)
0.028***
(15.12)
-0.006**
(-1.99)
0.009**
(2.17)
BUYLATE
-0.013***
(-9.87)
-0.023***
(-16.16)
-0.037***
(-23.85)
-0.028***
(-13.08)
-0.043***
(-5.20)
-0.046***
(-8.90)
6CYL
0.237***
(5.79)
… … … … …
SEDAN
… … … … … …
HBACK
… … … … … …
AWD
… … … … … …
Constant 10.02***
(688.3)
9.93***
(600.2)
10.43***
(375.3)
9.816***
(416.8)
9.945***
(333.0)
9.836***
(232.0)
Obs. 42,728 35,065 33,203 20,546 19,163 19,163
Comp. Prop. 0.9985 0.9989 0.999 0.932 0.343 0.657
Var. Diff 34.08 17.38 19.74 10.85 -5.74
Stat. Significance: *** (1%); ** (5%); * (10%)
28 PHOENIX CENTER POLICY PAPER [Number 48
Phoenix Center for Advanced Legal and Economic Public Policy Studies
www.phoenix-center.org
Table A-3. FMM Results
Variable Rogue (1) Rogue (2) Civic Accord Focus 3i
lnINTRACOMP 0.014***
(6.73)
0.008***
(3.99)
0.011***
(4.01)
0.019***
(14.18)
… 0.020***
(8.67)
lnINTERCOMP -0.002
(-1.02)
-0.007***
(-3.93)
-0.008***
(-6.28)
-0.003***
(-2.72)
… 0.003*
(1.77)
D2012
0.000
(-0.07)
0.005*
(1.69)
0.033***
(11.20)
-0.023***
(-14.79)
… 0.029***
(3.18)
D2013
-0.060
(-17.99)
-0.032***
(-9.75)
0.062***
(13.63)
0.055***
(37.75)
… -0.016*
(-1.82)
lnMEDIAN -0.011***
(-3.01)
-0.042***
(-11.20)
0.028***
(10.62)
0.017***
(11.35)
… -0.024***
(-5.47)
BUYLATE
-0.060***
(-13.44)
-0.026***
(-7.02)
-0.049***
(-17.39)
-0.030***
(-19.42)
… 0.003
(0.85)
6CYL
… … … 0.112***
(102.2)
… …
SEDAN
… … 0.005***
(3.34)
-0.007***
(-5.17)
… …
HBACK
… … … … … 0.130***
(20.06)
AWD
0.066***
(9.86)
0.075***
(15.89)
… … … …
Constant 10.095***
(222.5)
10.576***
(229.3)
9.491***
(429.7)
9.837***
(544.7)
… 9.984***
(222.7)
Obs. 24,294 24,294 22,400 23,832 … 14,737
Comp. Prop. 0.341 0.659 0.817 0.940 … 0.783
Var. Diff -6.31 2.23 23.2 … 3.83
Stat. Significance: *** (1%); ** (5%); * (10%)
