Competing Retailers and Inventory: An Empirical
Investigation of U.S. Automobile Dealerships
Gérard P. Cachon Marcelo Olivares
The Wharton School, University of Pennsylvania, Philadelphia PA, 19104
cachon@wharton.upenn.edu maolivar@wharton.upenn.edu
opim.wharton.upenn.edu/~cachon opim.wharton.upenn.edu/~maolivar
January 12, 2007
Abstract
In this study we empirically estimate the e¤ect of local market conditions
on inventory performance of U.S. automobile dealerships. We show that the
in‡uence of competition on a retailer’s inventory holdings can be separated
into two mechanisms. First, the entry or exit of a competitor can change a
retailer’s demand (a sales e¤ect). Second, the entry or exit of a competitor
can change the amount of bu¤er stock a retailer chooses to hold, which in‡uences the probability a consumer …nds a desired product in stock (a service
level e¤ect). The sales e¤ect can in‡uence inventory through the presence
of economies of scale. Theoretical arguments of inventory competition are
ambiguous on the expected sign of the service level e¤ect: there are models
suggesting that service level increases, decreases or remains constant with
entry. We obtained data (via a web crawler) on inventory and sales of auto
dealerships of a large manufacturer. Using cross-sectional variation of dealers in isolated markets, we estimated the e¤ect of market structure (number
and type of competitors), local demographics and sales on inventory levels.
We used market population as an instrumental variable to control for the
endogeneity of market structure. Our results suggest a strong positive nonlinear e¤ect of the number of rivals on service levels. Furthermore, we …nd
that the impact of competition on inventory via the service level e¤ect is
comparable to the impact via the sales e¤ect. Counterfactual experiments
indicate that reducing the dealership network of this manufacturer (thereby
reducing competition) could reduce the remaining dealers’ days-of-supply
(inventory divided by average sale rate) from 13% to 24%.
Keywords: Inventory competition, empirical, entry, supply chain management, automobile industry.
1.
Introduction
There exists a substantial literature on how …rms should manage inventory (e.g. Zipkin
(2000)), but less is known about how …rms actually manage inventory. In particular, why
does there exist considerable heterogeneity in inventory holdings, even across …rms within
a single industry? For example, Table 1 displays data on inventory holdings in the U.S.
automobile industry supply chain. During the years 1999-2004, Chevrolet held on-average
a 73 day supply of inventory in its supply chain whereas Toyota held only about a 42
day supply. Cachon and Olivares (2006) demonstrate that di¤erences in manufacturing
‡exibility and product assortments across auto makes (i.e., brands) explains nearly half of
this gap, but, due to the nature of their data, they are unable to assess whether di¤erences
in dealership structure is also a factor. Table 1 indicates that there are indeed dramatic
di¤erences across makes. For examples, Toyota has fewer than a third of the number of
dealerships as Chevrolet (1200 vs. 4227), but each of their dealership sells on average twice
as many vehicles per year (1251 vs. 627).
Inventory theory suggests that there generally are economies of scale in managing inventory. Hence, Toyota may carry less inventory in part because their dealerships operate at
a higher sales rate. Some Chevrolet dealers sell considerably more vehicles per year than
other Chevrolet dealerships, so economies of scale (with respect to sales) could even explain
heterogeneity in inventory holdings within a make. However, the U.S. auto makes exhibit
other signi…cant di¤erences across their dealership structures. For example, we would expect
that Chevrolet dealerships would be geographically located closer to each other than Toyota
dealerships because there are simply more of them in the U.S. Furthermore, dealerships
are not uniformally distributed across the country. Figure 1 plots the relationship between
the ratio of the number of General Motors (GM) dealerships to Japanese brand dealerships
(Toyota, Honda and Nissan) by state relative to each state’s population growth. In states
with large population growth from 1950 - 2004, such as California and Arizona, there is
approximately the same number of GM and Japanese brand dealerships, whereas in slow
growth states, such as Iowa and South Dakota, GM dealerships are much more numerous
in a relative sense. As a result there exists signi…cant variation in both the number and
type of dealerships in local markets. This variation regulates the degree to which dealerships compete and theoretical models predict that inventory is in‡uenced by the intensity
of competition. (Interestingly, as we elaborate on in section 2, there is disagreement in the
theoretical literature as to how competition in‡uences inventory, i.e., does competition lead
…rms to hold more or less inventory.) Finally, in addition to scale and competition e¤ects,
we expect that inventory holdings could vary due to di¤erences in consumer characteristics
across markets. For example, all else being equal, a …rm’s optimal inventory decreases
as its customers become more patient, i.e., more willing to wait for their prefered product
rather than choose a competitor’s product. Indeed, European consumers expect to wait
1
to receive their vehicles, so dealerships in Europe carry a very limited supply of on-hand
vehicles. Although the U.S. market operates in a very di¤erent fashion (most vehicles in
the U.S. are purchased from dealer inventory), there is signi…cant heterogeneity in consumer
demographics across the country, which could lead to di¤erences in buying behavior.1
The objective of this paper is to identify how inventory holdings of U.S. automobile
dealerships are in‡uenced by economies of scale with respect to sales, market structure (the
term we use to describe local competition) and consumer characteristics. We developed
a web-crawler to collect data on GM dealerships located in more than 200 isolated markets. Because there is little heterogeneity in dealership structures across time, our empirical
strategy is to exploit the cross-sectional variation in these markets to identify the e¤ects of
interest. We use instrumental variables to control for the endogeneity of market structure
with respect to unobserved market characteristics. We focus on the auto industry because
it is economically signi…cant and detailed data on local inventory holdings are available (via
our web-crawler). Although our results are speci…c to this industry, our econometric methods could be applied to study inventory in other retail industries. Furthermore, some of our
…ndings may apply broadly to other forms of retailing.
Our study is related to the growing empirical literature on inventory. Wu et al. (2005)
study the relationship between …rm inventory holdings and …nancial performance, while
Hendricks and Singhal (2005) study the impact of supply chain disruptions (including problems with inventory) on short term …nancial and accounting measures. Gaur et al. (2005)
…nd that retailers’ with higher margins and lower (or higher) assets tend to carry less inventory. Roumiantsev and Netessine (2006) use aggregate inventory data to measure the
relationship between demand uncertainty, lead times, gross margins and …rm size on inventory levels. These studies use data on publicly traded …rms and they do not measure the
e¤ect of market structure on inventory. Amihud and Mendelson (1989) use public data on
manufacturing …rms to estimate the e¤ect of market power (proxied by the …rms’margins
and market shares) on inventory levels and variability. They …nd that …rms lower their inventory as market power decreases, i.e., as competition intensi…es. Our study is di¤erent in
the sense that we track individual units of inventory and we are able to measure di¤erences
in local market structures.
Other empirical studies analyze the e¤ect of market structure on product variety. For
example, Berry and Waldfogel (2001) …nd that higher market concentration leads to higher
product variety in the radio broadcasting industry, showing evidence that …rms use product
line expansion to preempt entry. Alexander (1997) …nds a non-monotonic e¤ect of market
concentration on product variety in the music recording industry, where the highest level
of variety is achieved at a moderately concentrated structure. Inventory is not a primary
1
Holweg and Pil (2004) report that 89% of the sales in the U.S. are …lled from dealership stock, versus
38% in Europe.
2
concern with either of those industries. Watson (2004) also …nds a non-monotonic e¤ect of
competition on product variety among eyeglass retailers.2
In the next section, we provide a general econometric framework to measure the e¤ect of
sales and competition on inventory. Section 3 describes the data and the speci…cation of the
model. Section 4 shows our main results and section 5 provides a sensitivity analysis and
further empirical evidence. Section 6 analyzes a conterfactual experiment of changing the
dealership network. We conclude and discuss our …ndings in section 7.
2.
An empirical model of retail inventory
We use a basic single-item periodic review base-stock model to motivate our empirical framework. Orders are received at the beginning of each period with zero lead-time. Let D be
i.i.d. normal demand in each period with mean and standard deviation : Some fraction
of the demand that is not ful…lled from in-stock inventory is backordered; the remaining
demand is lost. Let Q be the order-upto level and z = (Q
)= . In this model the service
level is the probability that all demand within a period is satis…ed from inventory. The
service level is increasing in z; so for convenience we refer to z as the service level with the
understanding that it is really a proxy for the service level. The expected inventory at the
end of each period, I, is then
I = (z + L (z))
(1)
where L (z) is the standard normal loss function.
It is empirically inconvenient to work with (1) directly because demand is not directly
observable. However, it can be shown (see the Appendix for details) that (1) can be written
as
I = s K (z)
(2)
where s is the standard deviation of sales (minfQ; Dg) and K (z) is an increasing function.
Estimation of s requires an assumption of the relevant period length. Instead, as in van
Ryzin and Mahajan (1999), we use
(3)
s = As
to approximate the standard deviation of sales, where s is observed sales over a sample
period and A and are coe¢ cients. The coe¢ cient re‡ects the degree to which there are
economies of scale in inventory management with respect to sales. If = 1; then days-ofsupply (inventory divided by daily demand rate) is independent of expected sales whereas
2
He …nds that stores with two or three nearby rivals o¤er the highest level of variety, but product variety
falls as competition increases beyond that threshold. In auto dealerships, inventory levels can be viewed
as a measure of variety because it is rare to have two identical vehicles in stock. Watson (2004) does not
observe sales, so unlike in our model, he is unable to distinguish whether competition in‡uences his variable
of interest (variety/inventory) via a sales e¤ect or a service level e¤ect.
3
if < 1; then higher sales retailers carry a lower days-of-supply for the same service level.
Combining (2) and (3) and taking logarithms yields.
log I = constant+ log s + log K (z)
(4)
The above equation suggests that a …rm’s inventory level can be decomposed into two separate components: a sales component, log s; and a service level component, log K(z):
According to (4), market structure can in‡uence a …rm’s inventory either through its
sales or through its service level. Suppose the number of …rms competing in a market is
taken as the proxy for market structure. If a market’s potential sales is reasonably …xed,
then it is intuitive that entry could reduce each …rm’s sales (the …xed market potential is
allocated among more …rms). However, entry could increase a retailer’s sales either because
price competition is su¢ ciently severe to increase total sales (i.e., total potential demand
increases) or via a retail agglomeration e¤ect - consumers may be more likely to search a
retailer located near other retailers rather than an isolated retailer because the consumer
wishes to economize on search costs. In other words, all else being equal, the potential sales
of several …rms located near each other may be greater than their potential sales when they
are far apart.3 We are not directly concerned with the speci…c mechanism by which market
structure in‡uences sales because we conjecture that these mechanisms in‡uence inventory
only through their e¤ect on sales.
The other part of (4), the service level, is in‡uenced by several factors. Taking demand
to be exogenous, a retailer may choose a service level z to balance the cost of holding too
much inventory (overage cost) with the cost of holding too little inventory (underage cost).
The overage cost is mainly composed of the opportunity cost of capital, storage costs and
depreciation. We assume these are independent of market structure. The underage cost
depends on the behavior of consumers when they do not …nd their preferred product either
because it is not carried by the retailer or because it is temporarilly out of stock. In such
a situation a consumer could purchase some other product at the retailer (substitute), defer
purchase of the most preferred product to a later time (backorder) or leave the retailer without making a purchase (the no-purchase option). A retailer’s underage cost is increasing in
the retailer’s margin - the larger the margin on each sale, the more costly it is to lose a sale.
Furthermore, the underage cost is decreasing in consumers’propensity to substitute or backorder but increasing in the consumers’propensity to choose the no-purchase option. These
behaviors probably depend on numerous consumer characteristics, such as the intensity of
their preference for the products in the retailer’s assortment, their perception of the cost
to search/shop, and their ability and willingness to defer their purchase. Although most
inventory models assume demand is exogenous, the retailer’s service level may also in‡uence
3
See Dudey (1990), Eaton and Lipsey (1982), Stahl (1982) and Wollinsky (1983) for models of consumer
search in which …rm location decisions are endogenous.
4
its demand, i.e., a retailer’s demand may be endogenously determined by its service level.
As a result, in addition to balancing overage and underage costs, a retailer may use its target
service level to attract and/or retain customers. (Dana and Petruzzi (2001) and Gerchak
and Wang (1994) develop single-…rm models where a higher service level attracts more demand). Therefore, market structure can in‡uence service levels through two mechanisms:
overage/underage costs or endogenous demand.
Deneckre and Peck (1995) consider n retailers that compete on price and quantity where
consumers can observe both prices and quantities before making their …rm choice. As a
result, competition can in‡uence each …rms’underage cost (via price competition that may
reduce margins) and each …rm can increase its demand via o¤ering higher service (i.e., demand is endogenous because consumers are attracted to …rms that they believe to have
higher service levels). If an equilibrium exists, they …nd that equilibrium prices are decreasing in n; but service levels are nevertheless independent of n:4 This surprising result implies
that the two e¤ect of market structure on service levels are o¤setting each other in their
model. Dana (2001) modi…es the Deneckre and Peck (1995) model and indeed …nds that
entry can reduce service levels - the e¤ect of price competition on the …rms’underage costs
can dominate the endogenous demand e¤ect.5 The analogous conclusion can be inferred from
Bernstein and Federgruen (2005).6 Other models obtain the same result, but the causality
is reversed: competition induces …rms to reduce their service level, even if a 100% service
level is costless, because lower service levels dampen price competition (see Balachander and
4
There are other models that …nd the same result for a di¤erent reason. Suppose retailers face stochastic
demand, …xed retail prices and each retailer’s strategic decision is its initial stocking quantity (e.g., Lippman
and McCardle (1997), Mahajan and van Ryzin (2001), Netessine and Rudi (2003)) Furthermore, the a
priori allocation of demand is exogenous, so …rms only in‡uence each other through spillover demand, i.e.,
if a retailer’s initial demand exceeds its stocking quantity then some portion of this excess demand visits
another …rm in an attempt to …nd stock. In Deneckre and Peck (1995) there is no spillover demand …rms compete in the initial allocation of demand. Because competition does not impact price and each …rm
cannot in‡uence its initial demand via its stocking decision, each …rm’s optimal service level is independent
of market structure, where service level in this case is (as in our base-stock model) the probability all demand
is satis…ed immediately from stock. Hence, monopolists and duopolists choose the same service level. Note,
they do not choose the same stocking level because market structure also in‡uences inventory via sales - a
duopolist does not sell the same amount as a monopolist.
5
Dana (2001) shows that prices and service levels decrease in n with the following sequence: …rms choose
stocking quantities, …rms observe the quantity choices, …rms choose prices and …nally consumers make their
…rm choices based on observed prices and their expectations of the …rms’ stocking quantities given their
prices. Dana (2001) also studies a model in which prices are chosen …rst and observed by all …rms and
consumers and then …rms choose quantities. With that sequence service levels are independent of n; as in
Deneckere and Peck (1995). Note, in Dana (2001) …rms that deviate from equilibrium by o¤ering higher
availability do not attract more customers because availability is unobservable. Hence, incentives to increase
availability are softened relative to Deneckre and Peck (1995).
6
If prices decrease, Bernstein and Federgruen (2005) …nd that service levels decrease, but they do not
explicitly study the impact of market structure, and their model is ill suited to do so. To explain, their
model of retailer i’s demand is Di (p) = di (p)"i ; where p is the vector of retail prices, di (p) is a deterministic
demand function and "i is a stochastic shock. It is not clear how to modify "i to account for …rm entry. It
is possible to make assumptions regarding the impact of entry on "i but they do not do so.
5
Farquhar (1994) and Daugherty and Reinganum (1991)). Consistent with the hypothesis
that competition leads to lower service levels, Gaur et al. (2005) …nd that retailers with
lower margins carry lower inventory and Amihud and Mendelson (1989) provide evidence of
a direct link between market power and inventory levels.7
In Cachon (2003) …rms compete on their service levels but prices are …xed: demand is
allocated to n retailers proportional to their stocking quantities. (This is a special case of
the Deneckre and Peck (1995) model with …xed prices) There is only the demand attraction
e¤ect in this model (competition has no impact on the …rms’underage costs because prices
are …xed), so he …nds that service levels are increasing in n: each retailer ignores the marginal
impact its inventory decision has on the demand allocation of the other retailers, so increased
competition results in more aggressive stocking levels.
The Cachon (2003) model lacks product variety and consumer search. Now suppose
…rms compete both in terms of the breadth and depth of their assortment and consumers
must shop around to …nd a product that matches their preferences at a reasonable price.
Recall, if a consumer does not …nd their most preferred item at a retailer the consumer can
substitute, backorder or no-purchase. The no-purchase option can mean either shopping
another retailer, if one is available, or literally never buying an item. Hence, the no-purchase
option is less desirable to a consumer in a monopoly market than in a duopoly market - in
a monopoly market no-purchase means never-purchase whereas in a duopoly market the nopurchase option includes shopping and possibly buying something from the other retailer.
Because, all else being equal, a monopolist could expect its customers to be more likely to
substitute or backorder, the monopolist’s optimal service level is lower than the duopolist’s
- increased competition could lead to higher service levels due to consumer preferences for
variety, i.e., a demand retention e¤ect.8
To summarize, theoretical models of inventory competition predict service levels are either
decreasing, independent or increasing with respect to entry. Additional competition reduces
service levels if the impact of price competition on margins is severe, whereas additional
competition increases service levels if the endogenous demand e¤ect is strong (i.e., if higher
service levels either attract additional demand or help to retain demand).
Given this discussion, we now further elaborate on (4). For each retailer i and product
category b; based on the periodic review model, inventory, Iib; is determined by a combination
of sales, sib ; and the service level, zib . (We now distinguish products by category because it is
plausible that inventory levels across categories at the same retailer have di¤erent motivations
to hold inventory.) Market structure can in‡uence sales and possibly service level, but there
7
Gaur et al. (2005) do not directly link retail competition to inventory level - they only observe a correlation between margins and inventory turnover. Amihud and Mendelson (1989) also explore the variability
of inventory holdings.
8
See Cachon et al. (2006) and Watson (2006) for formal models in which …rms compete on prices and
product assortment. However, demand in their models is (e¤ectively) deterministic, so it is di¢ cult to
interpret their results with respect to the impact of entry on service levels.
6
are other factors that could potentially in‡uence service level (e.g., consumer characteristics).
We assume the following reduced form for the e¤ect of service level:
log K (zib ) = k + Cib + mib + Xib +
ib
(5)
The term mib captures the e¤ect of the margin (mib ) on service level. Based on Dana (2001)
and the empirical …ndings of Gaur et al. (2005), we would expect to be positive. Covariates
in Cib include di¤erent measures of competition and market structure, such as the number
of rival stores in the same market as …rm i o¤ering products that are substitutes for b. Cib
measures the impact of competition on service level, net of price competition. According to
either an inventory competition e¤ect (Cachon (2003)) or a consumer preference for variety
e¤ect, we expect to be positive. Xib and ib denote other observable and unobservable
factors, respectively, that a¤ect service level. Covariates in Xib include demographic characteristics of the markets that capture di¤erences in consumer characteristics which in‡uence
a retailer’s optimal service level. For example, if a certain demographic of consumers (e.g.,
college educated) has a particular a¢ nity for a retailer’s product (so that they are unwilling
to choose the no-purchase option) then retailers in markets with a high concentration of
these types of consumers may have a lower optimal service level.
In some applications, including the one studied in this paper, margins are unobserved.9
The linear projection of margin on market structure is de…ned as mib = Cib + ib , where ib
denotes other factors determining margins. We expect that competition reduces margins,
i.e., < 0: Replacing (5) in (4) and mib for its linear projection gives the following model,
which we seek to estimate using data from a cross section of retailers:
log Iib =
+ log sib + Cib + Xib +
ib :
(6)
The coe¢ cient
+
measures the overall e¤ect of market structure on the service
level, including price competition. Note, with this formulation we estimate the net e¤ect
of various mechanisms by which competition can in‡uence service level, ; but we do not
estimate each of them separately. The term ib = ib + ib captures factors unobserved by the
econometrician that a¤ect the service level, either directly or through margin. Other control
variables may be needed, depending on the application. For example, it may be necessary
to include controls for di¤erences in the retailers’supply processes (e.g., retailers may vary
in how frequently they are replenished and in the average size of each replenishment). The
parameters to be estimated are = ( ; ; ; ) : We are interested in the magnitude of
( = 1 means there are no economies of scale with respect to sales) and the sign and
magnitude of ( < 0 suggests that the margin e¤ect of competition dominates whereas
> 0 suggests that the demand attraction/retention e¤ects dominate).
9
An alternative is to use price as a proxy for margin. We do not have data on transaction prices of
automobiles.
7
There are several challenges associated with the the identi…cation of : It is important to
de…ne each retailer’s market appropriately, otherwise Cib may be a poor measure for competition and Xib may be a poor measure of consumer characteristics. We attempt to alleviate
this concern by identifying geographically isolated markets. To estimate precisely, it is
important that the selected markets have su¢ cient variation in market structure. The endogeneity of market structure is of particular concern with respect to the identi…cation of
: Retailers choose which markets to enter and they may do so based on market characteristics that they observe but are unobserved by the econometrician. Inventory costs a¤ect
dealership pro…ts, therefore entry decisions are a¤ected by factors that in‡uence inventory,
including ib . If such is the case, Cib and ib may be correlated and estimating (6) using
Ordinary Least Squares (OLS) would lead to biased estimates. We use Instrumental Variables (IV) to obtain consistent estimates of and assess the magnitude of the endogeneity
bias. We seek an observable instrument that is correlated with market structure, Cib ; but
uncorrelated with unobservable consumer characteristics in ib : We use market population as
our main instrument on the assumption that population is correlated with entry (more …rms
enter as a market’s population increases) and population is uncorrelated with unobserved
consumer characteristics that in‡uence service level conditional on the observed controls in
Xib :10
Sales can also be endogenous in (6). A product is likely to have higher sales in markets where consumers have high idiosyncratic valuation for the product. These consumers’
characteristics may also a¤ect their purchasing behavior, leading to a di¤erent service level.
To assess the magnitude and sign of this potential bias, we estimate including market
…xed e¤ects in (6). In this model, is identi…ed through di¤erences in sales volume across
stores located in the same market. This approach is feasible only in markets where two or
more retailers are observed. Furthermore, if Cib is common to all products and retailers
in a market, then is not identi…ed because the e¤ect of competition is absorbed into the
market …xed e¤ect.
3.
Data
This section provides a brief description of the U.S. auto industry and details the data in
our study.
Six companies account for about 90% of sales in the U.S. auto market: Daimler-Chrysler
(DC), Ford, GM, Honda, Nissan and Toyota. We refer to DC, Ford and GM as domestic
manufacturers. Each company o¤ers vehicles under several brands or auto makes. For example, GM makes include Chevrolet, GMC, Pontiac, Buick, Saturn, Cadillac and Hummer11 .
10
The assumption that larger markets lead to more entry can be veri…ed empirically when markets are
well de…ned.
11
GM closed its Oldsmobile brand in 2000.
8
Each auto make produces several models. Examples include the Chevrolet Malibu, the Toyota Camry and the Ford Explorer. Models can be classi…ed into vehicle classes, including
cars, sports cars, Sport Utility Vehicles (SUV) and pickups, among others. Each model
is o¤ered with multiple options, which include di¤erent body styles, engines, transmission
types and breaking systems, among other features.
In the U.S., more than 90% of new passenger vehicles are purchased from dealership
inventory. Auto distribution is regulated by franchise laws, which require that all new
vehicles must be sold through a network of dedicated franchised dealers.12 As of 2006,
there are approximately 22,000 dealerships in the U.S. The number of dealerships has been
declining in the U.S. since it peak in 1930 when there were about 50,000 dealerships (Marx
(1985)).
3.1
De…nition of Markets
Based on (6), we seek to de…ne isolated markets so that we can accurately proxy for the level
of competition within the market. Markets are de…ned by isolated Urban Areas (UA) of
population below 150,00013 . The criteria used to determine whether a UA is isolated depends
on the population of the UA. These criteria are summarized in Table 2. For example, a UA
with population between 50 and 100 thousand will qualify as an isolated market if it is at
least 50 miles away from the closest UA with population above 100 thousand and 30 miles
away from the closest UA with population above 50 thousand. Note, these criteria impose
minimum distance requirements to markets of equal or larger size. Our rationale behind
these criteria is that consumers not …nding their desired product inside the market will
consider travelling to the closest market which o¤ers the product. This tends to be a more
populous market. Similar rationale was utilized by Dranove et al. (1992) and Bresnahan and
Reiss (1991) to de…ne markets. From these markets, we selected those which had at least one
GM dealership. (As we describe later, our data are from GM dealerships.) Based on these
criteria, we obtained 235 isolated markets. We obtained demographic data and geo-coded
information (latitude and longitude) for these markets from the 2000 decennial census. 37%,
5%, 26% and 31% of the markets are located in the Mid-West, North-East, South and West
census regions, respectively. Data on new vehicle dealerships located in each market were
obtained from edmunds.com.14
Table 3 describes the selected markets. We grouped markets according to the total
number of Ford, DC, GM, Honda, Toyota and Nissan dealerships. The second column of
12
Smith (1982) provides a detailed analysis of dealership franchise laws.
We used UAs de…ned in the 2000 Census. These include: (i) urban areas consisting of territory with a
general population density of at least 1,000 people per square mile of land area that together have a minimum
residential population of at least 50,000 people; and (ii) urban cluster of densely settled territory with at
least 2,500 people but fewer than 50,000.
14
We matched dealers to UA based on 5 digit zipcodes. Matching tables were obtained from the Missouri
Census Data Center (http://mcdc2.missouri.edu).
13
9
the table shows the number of markets with the observed number of dealers. For example,
there are 17 monopoly markets with one GM dealership. In more than 90% of the markets,
the number of dealerships is in the range 1 to 10. The number of dealerships increases
with market size, measured by population (third column). The last three columns show the
percent of markets with at least one dealership of non-GM domestic manufacturers (Ford
or DC), Japanese manufacturers (Toyota, Honda or Nissan) and a second GM dealership,
respectively. The …rst competitor faced by a GM dealer is usually a non-GM domestic
dealership. Japanese dealerships usually enter markets with three or more dealerships. All
of the observed markets with six or more dealerships have at least one GM, one non-GM
domestic and one Japanese dealership. Several markets have more than one GM dealership.
In all but 5 markets, the GM dealers carry franchises of di¤erent brands. The table shows
that the selected markets have su¢ ciently rich variation in market structure, both in the
number and type of dealerships.
We obtained the following demographic data for each market: percent of population above
60 years old (OLD), African-American (BLACK ), with college (COLLEGE ), active in the
army (ARMY ), with farming occupation (FARMING) and using public transportation to
commute to work (PUBTRANS ). We also obtained median household income for each UA
(INCOME ) and the voting turnout and percent of republican votes for the county of each
market (TURNOUT and REPUBLICAN, respectively, obtained from uselectionatlas.org).
Summary statistics of these variable are shown in Table 4.
We included BLACK, INCOME, COLLEGE, REPUBLICAN, TURNOUT, OLD, and
FARMING in Xib because these variables have substantial partial correlation with the number of dealerships in a market (see Appendix Table 11). In addition, we included PUBTRANS
and ARMY to capture potential di¤erences in consumer characteristics and their a¢ nity to
GM brands.
3.2
Model speci…cation
We obtained inventory and sales data from a website o¤ered by GM that enables customers
to search new vehicle inventory at local dealerships15 . We developed a web-crawler that
monitors daily inventory in all the GM dealerships located in the selected markets. We
obtained data only for GM dealerships16 . For our empirical study, we collected data from
July 20 to November 15 of 2006. The web-crawler keeps record of the number and type of
vehicles available at each dealership. For example, it records the number of GMC Yukon 2007
15
http://www.gmbuypower.com
Developing web-crawlers for each manufacturer would require substantial additional e¤ort. Some websites o¤er inventory search for dealerships of several brands (e.g. nada.org) but many of these are not suitable
for the large-scale data collection that we require. We monitored our web-crawler frequently in case changes
were made to the website. In fact, during our study period GM did change its website. Substantial e¤ort
was required to repair the crawler.
16
10
4WD available at each dealer. The web-crawler also keeps track of speci…c information of
each vehicle, such as color, options, list price and, more importantly, the vehicle identi…cation
number (VIN) which uniquely identi…es all new vehicles in the U.S. By keeping track of the
speci…c VINs available at each dealership, we can identify replenishments and sales. We
also observe some vehicles that are transferred from one dealership to another. As we show
below, data on transfers are used in our estimation. Because in our markets there is only one
GM dealership o¤ering each brand, transfers are always received from a dealership outside
the market. To identify all vehicles transferred between dealers it is necessary to monitor all
dealerships in the U.S., which is infeasible. Instead, we monitored all dealerships in seven
states.17 To validate the data, we visited three dealerships in the Philadelphia area18 . Most
of the vehicles found at the three dealers lot during June 2, 2006 were posted in the website
during that day19 . The dealerships visited declined to provide data on the speci…c vehicles
sold.
To estimate model (6), we de…ned the dependent variable as the average vehicle inventory
of each brand b at a GM dealership i.20 We imputed total sales (SALES ) of each make during
the study period to measure sib .21 We include the demographic variables in Xib .
We estimated several speci…cations for the competition e¤ect, Cib : The simplest measure
is a linear e¤ect of the number of dealerships in the market (NC ). We restricted the dealership counts to the following manufacturers: GM, DC, Ford, Toyota, Honda and Nissan. The
square of this variable (NCSQ) is included in some speci…cations to test for non-linearities.
We also estimated the e¤ect of the number of rivals using a ‡exible non-parametric speci…cation, with indicator variables of the form x = 1 fN C xg ; with x 2 f1::Nmax g : x
measures the e¤ect of the xth entrant on service level. We restricted our sample to markets
with 10 or fewer dealerships to measure this e¤ect more precisely (Nmax = 10).
We also used an alternative measure of competition de…ned as the presence of dealerships
of di¤erent manufacturers. We de…ned three measures: HASDOM, HASJAP and HASGM2,
which capture the presence of non-GM domestic dealerships (Ford and DC), Japanese dealerships (Toyota, Honda and Nissan) and a second GM dealership in the markets (recall that
in all our markets there is at least one GM dealership).
To measure potential competition from outside the market, we included the driving time
to the closest GM dealership outside the UA (GMDRIVET ).22 Driving time was used to
17
The selected states are Colorado, Nebraska, Florida, Wisconsin, Maine, California and Texas. These
states are geographically relatively isolated (they border Mexico or Canada, they have a substantial coastline
and/or their border areas are sparsely populated) and exhibit variation in population growth (see Figure 1).
18
We selected this dealerships by convenience. None of the selected markets are in the Philadelphia area.
19
The dealership lots include many vehicles (sometime more than 100) and the authors could not verify
all of them.
20
We excluded the HUMMER brand from our analysis because it is present in only one of the markets in
our study.
21
Sales included vehicles transferred to other dealerships.
22
We used Rand McNally’s website to obtain driving distances and time.
11
capture the e¤ect of nearby highways on transportation costs. We also estimated models
with “bird-‡y”distance (using latitude and longitude data) and to GM dealerships carrying
the same brand. Our results were similar with these alternative measures.
GM dealerships can own multiple franchises of GM brands. If customers substitute
between di¤erent GM brands, a stock-out in one brand is less likely to become a lost sale for
a multi-brand dealership, because customers may buy a vehicle from another brand in the
lot. If inter-brand substitution within GM is substantial, we expect the number of franchises
carried by a dealership (NFRANCH ) to have a negative e¤ect on the service level.
Dealership inventories are a¤ected by other factors in addition to sales and service levels.
We include two measures that control for di¤erences in the dealerships’ supply processes.
Due to …xed transportation costs, dealerships receive vehicles in batches. Larger batch sizes
increase average inventory levels. Let Qibt be the total incoming orders (without transfers
from other dealerships) of product category b in week t to retailer i. To proxy batch size, we
calculated the coe¢ cient of variation of the volume of weekly incoming orders:
p
V (Qibt )
CV Qib =
E (Qibt )
where V ( ) and E ( ) are the variance and expectation operators, respectively. CVQ is
positively correlated with batch size, therefore we expect to see a positive e¤ect of this
variable on inventory levels23 .
It is common to observe vehicle transfers between dealerships that carry the same make.
Let Tib be the total amount of transfers received of product category b by dealership i. We
measure the percent of transfers received as:
TRANSF ib =
Tib +
Tib
P15
i=1
Qibt
(the denominator is the total number of vehicles received during the 15 weeks of the study
period, including transfers). Transfers enable dealerships to share inventory and bene…t from
risk pooling. Hence, we expected TRANSF to have a negative e¤ect on average inventory
levels.24 Recall, we are unlikely to observe all the transfers for all dealerships. We include a
dummy, ALLSTATE, indicating whether the dealership is located in one of the states where
23
Consider two dealership with the same sales volume S per week. Dealership A is replenished every
week, with a batch size of
p S. Dealership B is replenished every two weeks, with a batch size of 2S. Clearly,
CV QA = 0 < CV QB = 2: We used weekly replenishments to reduce measurement error.
24
Anupindi and Bassok (1992) show that centralization of inventory stocks of multiple retailers usually
decreases total inventory relative to the descentralized case where each retailer chooses their inventory level
independently. Rudi et al. (2001) analyze a model of two newsvendors with transshipments of left-over
inventory. It can be shown that their model implies a negative association between the average number of
transfers received by a retailer and its service level. Narus and Anderson (1996) report inventory reductions
from inventory sharing initiatives in several industries operating with descentralized distribution networks.
12
we monitored all dealerships.25
Table 5 shows summary statistics and correlation matrix of the main variables in the
econometric model.
The structural model underlying equation (6) suggest that coe¢ cient captures statistical economies of scale associated with sales volume. In auto dealerships, there are additional
economies of scale in sales volume. CVQ controls for economies of scale arising from …xed
ordering costs (such as order processing and transportation).26 We believe that statistical
economies of scale (i.e., the coe¢ cient of variation in sales decreases in the level of sales)
and …xed ordering costs are the main sources of economies of scale in auto distribution. If
CVQ is a good proxy for batch size, then the estimated can be interpreted as a measure
of statistical economies of scale.
3.3
Instrumental variables
We use two measures of market population to instrument for market structure (all based in
data from the 2000 decennial census). The …rst one is total population in the UA (UAPOP).
The second one is fringe population (FRINGEPOP). The fringe of a UA is de…ned as the
population of all zipcodes outside the UA within a 100 miles radius for which the UA is the
closest UA with dealerships27 .
In some speci…cations, we use additional instruments to increase e¢ ciency in the estimation. Franchising laws impose costs on the manufacturer to close existing dealerships.
Markets with current low population which had higher population in the past are likely to
have more dealerships than those which never had a large population. Due to this “stickiness” in dealership exit, past population has positive partial correlation (conditional on
current population) with the number of dealerships. We use county population in 1950 and
1970 (POP50, POP70 )28 . We also use county population in 2000 (POP2K ) as an additional
measure of current market size. All population measures were included with natural log
transformation because it provided better …t in the …rst step estimates of the IV regressions.
25
If the coe¢ cient on TRANSF is negative, we expect ALLSTATE to be positive because for the observations with ALLSTATE=0 a fraction of the transfers are unobserved. In all the speci…cations analyzed,
the coe¢ cient on ALLSTATE was positive and signi…cant. The average percent of transfers for dealerships
with ALLSTATE=0 and ALLSTATE=1 is 4.5% and 10%, respectively.
26
Indeed, there is substantial negative correlation between logSALES and CVQ.
27
A similar measure was used by Dranove et al. (1992). We calculated distances using latitude and
longitude. The census proxy of zipcodes (ZCTAs) were used.
28
Historical county population was obtained from the Inter-University Consortium for Political and Social
Sciences (ICPSR).
13
4.
Results
Table 6 shows the estimation results. In all the speci…cations the dependent variable is
logINV. Columns 1 through 6 include demographic control variables and the dummies for
make, region and ALLSTATE. The estimated coe¢ cients on these controls are ommited for
ease of visualization (a full description of the estimates of column 1 and 3 are shown in the
Appendix Table 10). Column 1 shows the OLS estimates of the regression including the
number of dealerships in the market (NC ) as a measure of competition. The coe¢ cient on
NC measures the average e¤ect of an additional dealership in the market, within the range
1 to 10. The coe¢ cient is positive and measured with precision. Increasing the number of
dealerships by one standard deviation (2.6 dealers, approximately) increases inventory by
approximately 6.8%. Batch sizes, proxied by CVQ, have a positive (and statistically signi…cant) e¤ect on inventory, as expected. The use of transfers from other dealerships, measured
by TRANSF, has a large economic (and statistically signi…cant) e¤ect in reducing inventory
levels. Increasing TRANSF by 0.1 (a 10% increase in the fraction of supply received from
transfers) reduces inventory by 8.9%. Similar estimates were obtained for the e¤ect of transfers and batch sizes across all the speci…cations analyzed. The coe¢ cients on GMDRIVET
and NFRANCH are small and not statistically signi…cant, suggesting that competition from
outside the market and substitution from other brands in multi-franchise dealerships is relatively small. Our market de…nition seems to capture well the local competition faced by a
dealership.
Sales volume has the highest explanatory power in describing inventory variations across
dealerships. The magnitude of the coe¢ cient suggest substantial economies of scale: a
10% increase in sales translates into a 2.6% decrease in days-of-supply ( (1
) 10% ). In
all the speci…cations, the estimated coe¢ cient is between .68 and .74.
Column 2 instruments for NC using population in the UA and fringe population as
instrumental variables. The point estimate of the NC coe¢ cient is positive and statistically
signi…cant. A Hausman test cannot reject that the estimates of columns 1 and 2 are equal
at any reasonable con…dence level (p-value 0.87). We also tested for the equality of the
single NC coe¢ cient and rejected the null (p-value less than 0.01)29 . There is some evidence
suggesting that OLS underestimates the e¤ect of competition. Nevertheless, the number of
dealerships has a positive e¤ect in the OLS and IV estimates. The …rst step regression has
an R-square of 0.76. The excluded variable UAPOP has the largest explanatory power in
describing variation in the number of dealerships. A full description of the …rst step estimates
are shown in Appendix Table 11.
In column 3, we estimate a non-linear e¤ect of competition, including a linear and a
quadratic term on the number of dealerships (NC and NCSQ). We …nd strong evidence
29
See Wooldridge (2002) pg. 120 for details on the test.
14
of decreasing marginal e¤ects of the number of dealerships on service level. The impact
of the …rst rivals is substantially larger than the average e¤ect estimated in column 1 and
2. The top part of Figure 2 illustrates the estimated impact of the number of dealerships
on inventory, measured by the percent change relative to a monopolist. Upper and lower
bounds of the 95% con…dence interval are illustrated with + and - symbols, respectively (the
standard errors are calculated using the delta method, see Hayashi (2000)).
Column 4 instruments NC and NCSQ using the instruments of column 2 (UAPOP and
FRINGEPOP) plus measures of past and current county population (POP50, POP70 and
POP2K ). The coe¢ cients on NC and NCSQ have the same sign as in column 3 and both
have high statistical signi…cance. While the magnitude of the estimates are di¤erent, the
marginal e¤ect at the mean is similar (aproximately 0.04). The IV estimates also suggest a
positive and marginally decreasing e¤ect of competition on service level.
In column 5 we estimate the e¤ect of competition using a ‡exible non-parametric speci…cation. We include seven indicator variables of the form 1 fN C xg with x 2 f2; 3; 4; 5; 6; 7; 9g :
The results provide strong evidence of a non-linear e¤ect of competition in service level. The
…rst rivals have the highest impact, and the e¤ect diminishes as the number of dealerships
in the market increases. The bottom part of Figure 2 illustrates the coe¢ cient estimates,
where the e¤ect is measured as the percent change in inventory relative to a monopolist30 .
Interestingly, the quadratic polynomial model estimated in column 3 approximates very
well the ‡exible non-parametric model. The large number of endogenous variables in the
non-parametric speci…cation makes IV estimation infeasible.
Column 6 uses indicators for the presence of di¤erent type of rivals in the market (HASDOMESTIC, HASJAPAN and HASGM2 ). The three measures are positive and highly
signi…cant. Non-GM domestic rivals have the largest e¤ect in increasing the service level:
GM dealerships in markets with Ford and DC dealerships have approximately 26% more
inventory (controlling for sales) compared to markets where these rivals are not present.
5.
Sensitivity analysis and further empirical evidence
In this section, we do a sensitivity analysis and provide additional empirical evidence to test
the robustness of our results.
Model (6) suggests a linear relationship between the logarithms of inventory and sales.
Furthermore, the model forces the slope to be constant across markets and una¤ected by
market structure. We analyzed a scatter plot of logINV versus logSALES for three types
of markets: GM monopoly markets, markets with GM and non-GM domestic dealers, and
markets with all kinds of dealerships (GM, non-GM domestic and Japanese).31 We …nd a
30
To make the …gure, we estimated the model with NC measured in levels. Standard errors, which are
used to plot the 95% con…dence intervals, were calculated using the delta method.
31
The scatter plot is shown in the Appendix.
15
strong linear relationship between the two variables. We regressed logINV on logSALES
allowing for di¤erent slopes and intercepts across the three groups. We cannot reject that
the slopes of the three series are equal (p-value= .4)32 . The R-square of the regression is
0.93. This analysis suggests that there are no interaction e¤ects between logSALES and
market structure, i.e., the e¤ect of competition on service level is separable from the e¤ect
of sales.
The regressions in Table 6 treat sales as exogenous. To test for potential biases in the
estimated coe¢ cient, we add market dummy variables to control for unobserved market
characteristics that could be correlated with sales. Note, market dummies control for all
market speci…c characteristics (including market structure), so in this regression the coe¢ cients of NC and the demographic controls in X are not identi…ed. We limit the sample to
markets with two or more GM dealerships and estimate using within-market variation in
dealership sales (373 observations in total). The estimated is 0:608 with standard error
SE ( ) = :037. A Hausman test rejects the estimated coe¢ cient from being equal to the one
obtained in column 3. The estimated reported in Table 6 seems to have a positive bias,
meaning that they under-estimate the e¤ect of economies of scale.
In regressions over the subsample of dealerships with ALLSTATE=1 the estimates of
and the competition e¤ect are similar in magnitude, sign and statistical signi…cance to those
reported in Table 6.
In section 4 we use market population as an IV to identify a causal e¤ect of competition
on service level. The main concern of using OLS is that the positive correlation between
competition and service level is driven by unobserved factors that a¤ect both variables rather
than a causal e¤ect of competition. For example, GM rivals may have higher incentives to
enter markets where consumer’s a¢ nity for GM brands is lower. These are markets where
GM is also likely to set a higher service level. Given all the demographic controls included in
Xib , we believe that it is unlikely that unobserved consumer characteristics that a¤ect service
level are correlated with market population. Hence, the IV estimates should be consistent.
Nevertheless, we provide additional results following a di¤erent identi…cation strategy which
corroborate our …ndings.
We conjecture that dealerships raise their service level when facing more intense competition to prevent losing customers to rival stores.33 If so, it is then reasonable to expect
that the e¤ect of entry on service level depends not only in the number of dealerships in a
market but also on the number and type of products that they o¤er. An entrant that o¤ers
more models which are close substitutes to the products o¤ered by the incumbents should
trigger a larger increase in the service level. Cachon and Olivares (2006) show that aggregate
32
We also conducted pairwise tests of the coe¢ cients. The smallest p-value among the three tests was .24
33
This e¤ect is suggested by Cachon (2003) and the consumer retention e¤ect that we propose in Section
.
2.
16
days of supply of a model tends to increase with the number of models o¤ered in the same
segment, providing some support for our conjecture.
To validate our conjecture, we estimate equation (6) using the number of models o¤ered
by rival dealerships as a measure of competition. We include several measures of competition
based on product characteristics. Following the literature of spatial competition (e.g. Seim
(2006)), we de…ne di¤erent bands where the products o¤ered by rival dealerships can be
located. These bands de…ne a measure of “distance”between product b and products o¤ered
by rivals. The de…nition of the bands is based on a market segmentation commonly used in
the auto industry. While these de…nitions can be subjective, they work reasonably well with
some product categories to capture the degree of similarity across products. We conjecture
that the number of products in closer bands should have a higher impact on the service level
than products located in the outer bands. On the contrary, if the positive correlation between
competition and service level is driven by unobserved characteristics related to a¢ nity for
GM brands, then products in all bands should have the same positive e¤ect on service level.
Let
be the set of all models o¤ered in model-year 2007. For a given product b, we
1
K
de…ne a partition
of the set of products . We refer to kb as the k th band of
b ::: b
product b. Bands are de…ned so that their distance to product b is increasing in k: Let Cibk
be the number of models in band kb o¤ered by the rivals of dealership i: The number of
models o¤ered is calculated based on the brands carried by each rival dealership and the list
of models o¤ered by each brand34 . We included dealerships of all manufacturers (not just
t
the six included in the previous estimation). De…ne the column vector Cib = Cib1 :::CibK and
the row vector of parameters = 1 ::: K : The parameter k measures the average e¤ect
of adding a model in the k th band to the assortment of a rival dealership in the market. This
model is di¤erent to the speci…cation de…ned in Section 3.2 because the e¤ect of competition
depends not only in the number of dealerships in the market but also the number and type
of models that they o¤er.
We de…ne product bands based on Ward’s model segmentation, which classi…es models
into several groups based on three dimensions. Models are partitioned into vehicle classes,
which include standard car, luxury car, sport utility vehicle, cross utility vehicle, van and
pickup. Within each class, models are classi…ed according to sizes (small, medium and large)
and prices (lower, middle, upper, etc.). This gives a total of 26 segments, which are shown
in the Appendix Table 13.
For our empirical analysis, we focused in groups of products for which at least three
product bands can be reasonably de…ned. We chose small and medium sized standard cars
(hereon SM cars, which exclude luxury and large cars) and light-trucks (hereon Trucks, which
include SUV, CUV and mini-van)35 . We de…ned bands for the segments on each of these
34
We do not know the actual number of models o¤ered since we do not observe inventory of dealerships
other than GM.
35
Full-sized vans and pickups are excluded because we could not obtain inventory data on them. We
17
two groups and ran two separate regressions. The dependent variable is the logarithm of
the average inventory level of models in a speci…c model segment o¤ered by each dealership.
For example, for the SM car regression, inventory of "Lower Small Car" and "Upper Middle
Car" of a speci…c dealership is counted as two di¤erent observations. For SM cars, four
bands were de…ned. The …rst band includes standard cars which have similar size or price
(B(PRICE,SIZE)). The second band includes all other standard cars (B(STDCAR)). The
third and fourth band includes luxury cars and light-trucks, respectively (B(ANYCAR) and
B(OTHER))36 . For trucks, we de…ned three bands. The …rst band includes vehicles within
the same class with similar size or price (B(PRICE,SIZE)). For example, if b ="Middle
SUV", band 1 includes vehicles in the segments "Middle Luxury SUV" and "Large SUV" but
not "Large Luxury SUV" nor "Middle CUV". Band 2 includes all other trucks (B(TRUCK)),
and band 3 all cars (B(OTHER)).
Table 7 summarizes the estimation results of the regressions for SM cars and Trucks.
All the speci…cations include dummies for region, price (based on the model segmentation)
and ALLSTATE and demographic characteristics37 . Columns 1 and 3 include the number
of vehicles on each band as the measure of competition. While none of the measures are
statistically signi…cant, the …rst band B(PRICE,SIZE) has the largest positive point estimate
of all the bands. In columns 2 and 4 we add a quadratic term on the number of vehicles
on the …rst band (SQ B(PRICE,SIZE)) to capture non-linearities. The results show that
the number of vehicles in the …rst band has a positive e¤ect on the service level, and the
marginal e¤ect is decreasing in the number of vehicles. The number of vehicles in the outer
bands have no signi…cant e¤ect on the service level (conditional on the number of models in
the …rst band). The results are similar in sign, magnitude and statistical signi…cance across
the two groups.
To compare the magnitude of the competition e¤ect between the product competition
and dealership competition models, we calculated the elasticities at the mean implied by each
model. For the dealership competition model estimated in Table 6, column 3, the implied
elasticity is 22%. For the SM car and Truck product competition models, the implied
elasticities are 32% and 20%, respectively. The average elasticities across the models are
similar in order of magnitude, suggesting that the estimated e¤ect of competition is robust
to di¤erent speci…cations and identi…cation strategies.
The empirical results can be interpreted as follows. First, the number of vehicles o¤ered
by rivals has a positive e¤ect on the service level of the products o¤ered by a dealership.
The e¤ect of competition in service level is increasing in the number of products o¤ered by
an entrant. Second, most of the e¤ect is captured by products which are closer substitutes.
excluded large and luxury cars because bands for this type of vehicles could not be reasonably de…ned.
36
We also ran regressions merging bands 3 and 4, and obtained similar results.
37
Given the detailed unit of analysis, there were few replenishments for each observation, making CVQ
very noisy. We excluded this measure from the regression.
18
The e¤ect of competition in service level depends not only on the number but also on the
type of products o¤ered by an entrant. Third, there is a saturation e¤ect: the …rst close
substitutes have a large impact in the service level, but the e¤ect becomes smaller as more
products enter the …rst band. Overall, these empirical results provide good support for our
conjecture that the intensity of inventory competition depends on the number and type of
products o¤ered in a market. This pattern is unlikely to be driven by unobserved market
characteristics a¤ecting service level.
6.
The e¤ect of reducing the dealership network
Domestic manufacturers started expanding their dealership networks in the early 1900’s. At
that time a large fraction of the U.S. population was living in rural areas and transportation
was di¢ cult. As a result, many dealerships were established so that they could be close to
population centers. The Japanese manufacturers started their dealership networks in the
second half of the century. Due to improved transportation and a greater concentration
of the population in urban areas, there are fewer Japanese brand dealerships and they are
concentrated in di¤erent regions of the country than the domestic manufacturers. (See Figure
1.) Due to rigid franchise laws that impose high costs on manufacturers for forcing the closure
of a dealership, the domestic manufacturers still have considerably more dealerships than the
Japanese brands. (GM paid more than one billion dollars to dealers to close its Oldsmobile
division (Welch (2006)).) Domestic manufacturers are concerned that the large number
of dealerships in their network is causing ine¢ ciencies. As a result, there is discussion on
the value of reducing the number of dealerships, despite the costs of doing so (Rechtin and
Wilson (2006)).
Closing a dealership in a market has two e¤ects on inventory of the remaining dealerships.
First, some of the sales of the closed dealership will be captured by the remaining dealerships.
Due to economies of scale, the remaining dealerships will reduce their days of supply. Second,
as shown in Table 6, column 6, the presence of another GM dealership in a market increases
a dealership’s service level. Hence, removing a dealership from a market is likely to decrease
the remaining dealer’s service level, further decreasing that dealers’days of supply.
We used the estimates of Table 6, column 6 to measure the e¤ect of closing some of GM’s
dealerships. We selected markets with seven or fewer dealerships. The maximum number of
GM dealerships in these markets is three. In the counterfactual, all but one of the dealerships,
the one with highest sales volume, remains in the market. To obtain a lower bound, we
assumed that all the sales of the dealerships closed are lost. This case considers inventory
reductions only due to lower service level. To obtain an upper bound, we assumed that all
the sales of the closed dealerships are captured by the remaining dealership. If the remaining
dealership carries di¤erent brands, the sales are allocated proportionally to the actual sales of
19
each brand. The following table summarizes the results from the counterfactual experiment.
No. dealers closed No. obs. Actual days of supply Reduction in days of supply
lower bound upper bound
1
79
122.1
15.0
27.0
2
5
126.2
15.5
37.9
Total
84
122.3
15.7
29.1
Actual days-of-supply and the two bounds on its reduction are reported separately for
markets in which one or two dealerships are closed. The improvement in inventory performance is substantial, between 15 and 29 days-of-supply on average. Interestingly, the service
level e¤ect (measured by the lower bound) is similar in magnitude to the potential gains
from economies of scale.38
7.
Conclusion
We develop an econometric model to estimate the e¤ect of market structure on inventory
holdings. We identify two drivers of inventory holdings: a sales e¤ect and a service level
e¤ect. We …nd that the sales e¤ect exhibits statistical economies of scale - increasing a
dealer’s sales reduces the dealer’s inventory when measured in terms of days-of-supply (i.e.,
inventory increases with sales, but at a slower rate). Based on our estimates, Chevrolet
could reduce its days-of-supply by 17% (a 12.3 decrease in days-of-supply) by matching
Toyota in sales volume per dealership. Our analysis suggests that, due to unobserved
market characteristics, this calculation may underestimate the inventory reduction due to
economies of scale in sales.
We are particularly interested in the impact of market structure (local competition) on
service levels (bu¤er inventory held by dealerships conditional on sales). Some theoretical
models predict that increased competition can lead to a decrease in service levels via a
price competition e¤ect - entry reduces margins, thereby reducing underage costs and the
incentives to hold inventory. Other models predict that entry can lead to higher service levels
- …rms may increase their service level to attract demand (as in Cachon 2003) or they may
increase their service level to better retain demand (i.e., to prevent customers from searching
other retailers). Among dealerships in the automobile industry we …nd that competition
increases service levels, i.e., any price competition e¤ect associated with entry is dominated
by the demand attraction/retention e¤ects. This result di¤ers from the …ndings in Gaur
et al. (2005) and Amihud and Mendelson (1989). Gaur et al. (2005) …nds that retailers with
lower margins tend to carry less inventory. (Amihud and Mendelson (1989) has a similar
38
An increase in a dealership’s sales volume is also likely to reduce its batch sizes, generating additional
improvements in inventory performance. Accounting for this reduction requires modeling the e¤ect of sales
on batching. Given that our measure of batch size, CVQ, is noisy and its e¤ect is measured imprecisely, we
ommited this e¤ect from the counterfactual analysis.
20
…nding but they study manufacturing …rms.) If low margins are taken as a proxy for higher
competition, then they …nd that inventory decreases with competition. They do not study
auto retailing, so it is possible that in other retail markets the price competition e¤ect of
entry dominates the demand attraction/retention e¤ects. Further research is needed to
reconcile these results.
Competition increases inventory in the auto industry, but we …nd that the marginal
e¤ect of competition is decreasing: the …rst entrant into a monopoly market causes a 17%
increase in inventory whereas entry beyond the seventh dealership has no positive e¤ect on
inventory (conditional on sales). We provide additional empirical evidence showing that
the service level of products depends on the number of close substitutes o¤ered by rivals,
but is insensitive to the number of disimilar products. Our results are robust to di¤erent
econometric speci…cations.
Our …ndings suggest that inventory may vary across automobile makes in part because
auto makes vary in their dealership structures. As the dealership network becomes more
dense there are two reinforcing e¤ects on inventory. One, if sales per dealership declines
as the number of dealerships increases (which is plausible), then the presence of statistical
economies of sales with respect to sales suggests that inventories will increase (measured in
days-of-supply). Second, an increases in the density of dealerships increases competition
among them (i.e., there will be more dealerships per market), which also increases inventory
via higher service levels. Thus, when comparing two automobile makes, we expect (all else
being equal) the one with the greater number of dealerships to carry more inventory. This
conjecture is consistent with aggregrate data for U.S. inventory holdings of the major automobile manufacturers. For example, Toyota has approximately half as many dealerships as
Chevrolet, and carries 42% less inventory (42 days of supply vs. 73 days of supply). (See
Cachon and Olivares (2006) for other factors that explain di¤erences in aggregate inventory holdings.) Furthermore, in the markets that we study, reducing the number of GM
dealerships reduces days-of-supply by at least 13% and by as much as 25%.
In the automobile industry, due primarily to restrictive franchise laws, it is di¢ cult for
an auto make to close dealerships. These restrictions therefore may force an auto make
to operate with more dealerships than it desires. As a consequence, the auto make would
carry more inventory than it would with a distribution network with fewer dealerships.
However, it is not possible with our analysis to conclude whether these franchise laws are
bad for consumer welfare - although they may lead to high inventory holding costs, they also
provide consumers with potentially lower prices (due to price competition) and a greater
selection of available products. Similarly, we are unable to conclude whether or not GM
would bene…t from reducing its dealership network - we do predict that GM would carry
less inventory, but this is only bene…cial if the impact on sales is su¢ ciently small.39 Thus,
39
Recently Ford attempted to consolidate dealerships in particular markets, such as Tulsa Oklahoma. They
21
an estimate of the impact of the dealership structure on sales is needed to obtain estimates
of the impact of a dealership network on consumer and …rm welfare. These estimates are
beyond the scope of this research, but worthwhile to pursue for future research.
In addition to the number and location of dealerships, we found that inter-dealership
transfers cause dealerships to lower their average inventory levels. Dealers may use transfers
to help retain customers that are not entirely satis…ed with the dealer’s assortment, i.e.,
transfers help to retain a customer by allowing the dealer to …nd a vehicle that better
matches the consumers’preferences. Alternatively, because dealers do not have full control
over which vehicles they receive from the manufacturer, dealers may use transfers to adjust
the mix of vehicles in their assortments. Additional research is needed to identify the
particular motivations for vehicle transfers between dealers.
Data Sources
http://www.gmbuypower.com
http://www.edmunds.com
http://www.randmcnally.com
U.S. Census Bureau (http://fact…nder.census.gov)
Inter-University Consortium for Political and Social Sciences (ICPSR)
Missouri Census Data Center (http://mcdc2.missouri.edu)
Ward’s Auto Datacenter
References
Alexander, P. 1997. Product variety and market structure: A new measure and a simple
test. Journal of Economic Behavior and Organization 32 207–214.
Amihud, Y., H. Mendelson. 1989. Inventory behavior and market power. International
Journal of Industrial Organization 7 269–280.
Anupindi, R., Y. Bassok. 1992. Centralization of stocks: Retailers vs. manufacturer. Management Science 45(2) 178–188.
Balachander, S., P.H. Farquhar. 1994. Gaining more by stocking less: A competitive analysis
of product availablility. Marketing Science 13(1) 3–22.
Bernstein, F., A. Federgruen. 2005. Descentralized supply chains with competing retailers
under demand uncertainty. Management Science 51(1) 18–29.
speculated that such a consolidation could lead to higher sales because customers a broader assortment of
inventory would be available to customers at the consolidated dealership than at any of the previous smaller
dealerships. This is plausible if consumer transportation/search costs are su¢ ciently low because the average
consumer must travel farther to a single dealership than to one of several dealerships in a market.
22
Berry, S., J. Waldfogel. 2001. Do mergers increase product variety? Evidence from radio
broadcasting. The Quarterly Journal of Economics 116(3) 1009–25.
Bresnahan, T., P. Reiss. 1991. Entry and competition in concentrated markets. The Journal
of Political Economy 99(5) 977–1099.
Cachon, G. 2003. Supply chain coordination with contracts. Handbooks in Operations
Research and Management Science: Supply Chain Management, S. Graves and T. de
Kok. 11:229-339.
Cachon, G., M. Olivares. 2006. Drivers of …nished goods inventory performance in the U.S.
automobile industry. Working paper, Wharton, University of Pennsylvania.
Cachon, G., C. Terwiesch, Y. Xu. 2006. The impact of consumer search, …rm entry and
competition on assortment and pricing. Working paper, University of Pennsylvania and
University of Miami.
Dana, J. 2001. Competition in price and availability when availability is unobservable. RAND
Journal of Economics 32(3) 497–513.
Dana, J., N. Petruzzi. 2001. Note: The newsvendor model with endogenous demand. Management Science 47(11) 1488–97.
Daugherty, A., J. Reinganum. 1991. Endogenous availability in search equilibrium. Rand
Journal of Economics 22(2) 287–306.
Deneckre, R., J. Peck. 1995. Competition over price and service rate when demand is
stochastic: A strategic analysis. RAND Journal of Economics 26(1) 148–162.
Dranove, D., M. Shanley, C. Simon. 1992. Is hospital competition wasteful? RAND Journal
of Economics 23(2) 247–262.
Dudey, M. 1990. Competition by choice: The e¤ect of consumer search on …rm location
decisions. American Economic Review 80(5) 1092–104.
Eaton, C., R.G. Lipsey. 1982. An economic theory of central places. The Economic Journal
56–72.
Gaur, V., M.L. Fisher, A. Raman. 2005. An econometric analysis of inventory turnover
performance in retail services. Management Science 51(2) 181–194.
Gerchak, Y., Y. Wang. 1994. Periodic-review inventory models with inventory-level dependent demand. Naval Research Logistics 41 99–116.
Hayashi, F. 2000. Econometrics. Princeton University Press, Princeton, New Jersey.
23
Hendricks, K.B., V.R. Singhal. 2005. Association between supply chain glitches and operating
performance. Management Science 51(5) 695–711.
Holweg, M., F. Pil. 2004. The second century. MIT Press, 5 Cambridge Center, Cambridge,
MA 02142.
Lippman, S.A., K.F. McCardle. 1997. The competitive newsboy. Operations Research 45
54–65.
Mahajan, S., G. van Ryzin. 2001. Inventory competition under dynamic consumer choice.
Operations Research 49(5) 646–657.
Marx, Thomas G. 1985. The development of the franchise distribution system in the U.S.
automobile industry. The Business History Review 59 465–474.
Narus, J.A., J.C. Anderson. 1996. Rethinking distribution. Harvard Business Review 112–
120.
Netessine, S., N. Rudi. 2003. Centralized and competitive inventory models with demand
substitution. Operations Research 51(2).
Rechtin, M., M. Wilson. 2006. Thinning the herd. Automotive News 81 82–84.
Roumiantsev, S., S. Netessine. 2006. What can be learned from classical inventory models:
A cross-industry empirical investigation. Working paper, http://www.netessine.com.
Rudi, N., S. Kapur, D.F. Pyke. 2001. A two-location inventory model with transshipment
and local decision making. Management Science 47(12) 1668–80.
Seim, K. 2006. An empirical model of …rm entry with endogenous product-type choices.
(forthcoming) RAND Journal of Economics.
Smith, R.L. 1982. Franchise regulation: An economic analysis of state restrictions on automobile distribution. Journal of Law and Economics 25(1) 125–157.
Stahl, K. 1982. Di¤erentiated products, consumer search, and locational oligopoly. Journal
of Industrial Economics 31 97–113.
van Ryzin, G., Siddarth Mahajan. 1999. On the relationship between inventory costs and
variety bene…ts in retail assortments. Management Science 45(11) 1496–1509.
Watson, R. 2004. Product variety and competition in the retail market for eyeglasses. Working paper, University of Texas at Austin.
24
Watson, R. 2006. Search, availability, and competition in product ranges. Working paper,
Department of Economics, University of Texas at Austin.
Welch, D. 2006. The other club battering GM. Businessweek 3974(38).
Wollinsky, A. 1983. Retail trade concentration due to consumers’ imperfect information.
Bell Journal of Economics 14(14) 275–82.
Wooldridge, Je¤rey. 2002. Econometric Analysis of Cross Section and Panel Data. MIT
Press, Cambridge, Massachusetts.
Wu, O.Q., H. Chen, M.Z. Frank. 2005. What actually happened to the inventories of American companies between 1981 and 2000. Management Science 51(7) 1015–1031.
Zipkin, P. 2000. Foundations of Inventory Management. McGraw-Hill.
25
Make
Days of
No. of
supply
dealers
Chevrolet
73
4227
Ford
74
3939
Honda
48
1001
Toyota
42
1200
Averages during years 1999-2004
Sales per
dealership
627
795
1077
1251
Table 1 – Comparison of inventory performance, number of dealerships and sales per
dealership across four auto makes. Days-of-supply is calculated based on all finished
vehicle inventory in the supply chain, including factory lots, ports of entry, in transit to
dealerships and in the dealerships.
Population
(thousands)
[100,150]
[50,100]
[25,50]
<25
Distance to closest UA (miles)
Pop>100
Pop>50 Pop>25 Pop>10
50
---50
30
--50
50
30
-50
50
30
30
Table 2 – Market selection criteria. Criteria vary depending on the population of the UA,
indicated in the first column. Columns two through five indicate the minimum distance
(in miles) required to the closest UA with population above the threshold shown in the
header.
# of
dealers
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
% of markets with dealers
# of
Median
non-GM
markets population domestic Japanese 2nd GM
17
24
43
29
21
22
14
21
9
13
9
3
6
1
1
3.91
6.20
9.76
13.05
26.16
36.48
38.37
62.63
68.22
61.47
78.50
100.32
122.00
105.36
122.98
0%
83%
100%
100%
100%
100%
100%
100%
100%
100%
100%
100%
100%
100%
100%
0%
8%
5%
55%
86%
100%
100%
100%
100%
100%
100%
100%
100%
100%
100%
0%
8%
2%
41%
57%
41%
71%
95%
100%
100%
100%
100%
100%
100%
100%
Table 3 – Summary statistics of the isolated markets. The last three columns show the
percent of markets with at least one dealership of non-GM domestic manufacturers,
Japanese manufacturers and a second GM dealership, respectively.
Variable
INCOME
OLD
BLACK
PUBTRANS
COLLEGE
REPUBLICAN
TURNOUT
FARMING
ARMY
Mean
32.6
15.5
5.4
0.8
18.5
0.6
0.4
1.3
0.7
Std.
Dev.
5.9
4.8
11.9
1.3
9.9
0.1
0.1
1.5
2.4
Min
19.7
4.7
0.0
0.0
1.5
0.3
0.1
0.0
0.0
Max
63.0
38.3
70.1
11.9
57.3
0.9
0.6
12.7
22.6
ANC
H
-0.21
-0.47 0.19
-0.28 0.07 0.21
-0.08 -0.06 0.04 0.04
Table 5 – Summary statistics and correlation matrix.
CVQ
0.49
-0.19
-0.43
-0.57
0.06
N FR
Std.
Min Max
Dev.
0.90 -0.24 5.14
0.98 0.00 5.74
2.63 1.00 10.00
0.55 0.05 6.50
1.29 1.00 5.00
0.47 0.42 3.04
0.08 0.00 0.50
RIVE
T
NDE
Mean
2.84
2.92
5.55
0.88
3.19
1.26
0.06
GMD
logS
ALES
0.89
0.51
-0.12
-0.43
-0.51
-0.02
V
logIN
Variable
logINV
logSALES
NDEALERS
GMDRIVET
NFRANCH
CVQ
TRANSF
ALER
S
Table 4 – Summary statistics for the demographic variables in the selected markets.
LOGSALES
NC
1
OLS
Linear
0.740***
(0.026)
0.026***
(0.009)
2
IV Linear
0.689***
(0.029)
0.077***
(0.015)
-0.02
(0.032)
-0.021
(0.014)
0.085**
(0.043)
-0.886***
(0.201)
0.001
(0.033)
0.011
(0.016)
0.085*
(0.044)
-0.914***
(0.206)
NCSQ
GMDRIVET
NFRANCH
CVQ
TRANSF
3
OLS
Square
0.728***
(0.025)
0.181***
(0.028)
-0.013***
(0.002)
-0.023
(0.031)
-0.024*
(0.014)
0.075*
(0.042)
-0.879***
(0.196)
NC>=2
NC>=3
NC>=4
NC>=5
NC>=6
NC>=7
NC>=9
4
5
Comp
IV Square Level
0.709*** 0.726***
(0.028) (0.025)
0.351***
(0.1)
-0.028***
(0.01)
-0.023
-0.019
(0.033) (0.032)
-0.024
-0.024*
(0.018) (0.014)
0.065
0.074*
(0.044) (0.042)
-0.875*** -0.889***
(0.202) (0.198)
0.170*
(0.091)
0.109*
(0.063)
0.099*
(0.053)
0.017
(0.061)
0.073
(0.064)
-0.014
(0.056)
-0.104**
(0.05)
HASDOMESTIC
HASJAPAN
HASGM2
Observations
R-squared
680
0.84
680
0.83
680
0.85
680
0.84
680
0.85
6
Comp
Type
0.715***
(0.025)
-0.025
(0.031)
-0.002
(0.015)
0.072*
(0.042)
-0.863***
(0.197)
0.259***
(0.067)
0.113***
(0.04)
0.132***
(0.042)
680
0.85
Table 6 – Main estimation results. Demographic variables and dummies for make, region
and ALLSTATE not shown (see Appendix for full estimation results). Column 2 uses
TOTPOP and FRINGEPOP (in logs) to instrument for NC. Column (3) uses the same IVs
plus POP50, POP70 and POP2K to instrument for NC and NCSQ. Standard errors shown
in parenthesis. ***,** and * indicate significance at the 1%, 5% and 10%, respectively.
1
SM car
0.5973***
(0.0275)
0.0070
(0.0057)
LOGSALES
N B(PRICE,SIZE)
2
SM car Sq.
0.5972***
(0.0274)
0.0360***
(0.0112)
-0.0005***
(0.0002)
-0.0080
(0.0076)
-0.0119
(0.0079)
NSQ B(PRICE,SIZE)
N B(STD_CAR)
-0.0005
(0.0072)
-0.0112
(0.008)
N B(ANY_CAR)
N B(TRUCK)
N B(OTHER)
0.0007
(0.0035)
-0.4529*
(0.2312)
818
0.62
PERTRANSF
Observations
R-squared
-0.0015
(0.0036)
-0.4654**
(0.23)
818
0.62
3
Truck
0.8051***
(0.025)
0.0068
(0.0062)
4
Truck Sq.
0.8026***
(0.0249)
0.0399**
(0.0157)
-0.0009**
(0.0004)
-0.0029
(0.0044)
-0.0017
(0.0051)
-0.5975**
(0.2578)
779
0.67
-0.0053
(0.0045)
-0.0013
(0.0051)
-0.6477**
(0.258)
779
0.67
Table 7 – Estimation results with number of models as a measure of competition.
Separate regressions were estimated for Small and Medium size cars (SM cars) and
Light-trucks (Truck). N B(.) measures the number of models in each product band. Other
controls include region and price dummies and demographics (not shown). Standard
errors shown in parenthesis. ***,** and * indicate significance at the 1%, 5% and 10%,
respectively.
Number of dealerships vs. population growth (1950-2004)
2.50
SD
IA
MI
WI
NE MT
log GM/Japan Dealers
1.50
OH
NY
0.50
-50%
0%
50%
AK
CO
NJ
100%
CA
FL
150%
NV
AZ
200%
250%
300%
-0.50
HI
-1.50
log(1+Growth)
Figure 1 – Scatter plot of the ratio of GM to Japanese dealerships versus population
growth between 1950-2004. Each observation is a state and both axes are with logarithm
transformation. Selected labels indicate US postal service state codes.
Service level effect of competition on inventory
(quadratic polinomial model)
100%
Inventory increase
(% above monopoly)
80%
60%
40%
20%
0%
1
2
3
4
5
6
7
8
9
10
-20%
# of dealerships
Service level effect of competition on inventory
(non-parametric model)
Inventory increase
(% above monopoly)
100%
80%
60%
40%
20%
0%
1
2
3
4
5
6
7-8
9-10
-20%
# of dealerships
Table 2 – Effect of competition on the targeted service level, measured as the percent
change in inventory relative to a monopolist GM dealership. Top (bottom) figure shows
the effect measured by the quadratic (non-parametric) model. + and – indicate upper and
lower bounds on 95% confidence interval, respectively.
Appendix
The following Lemma and Proposition provides a formal proof of equation (2) and the
properties of K (z) :
Lemma 1 Let D is a normally distributed random variable with mean and standard deviation :De…ne Q = + z and S = min fQ; Dg : Then, the variance of S, V (S), is given
by:
V (S) = 2 f (z)
(z) [z + (z)]g
where (z) and
respectively, and
z.
(z) are the density and distribution functions of the standard normal,
(z) = (z) = (z) is the hazard rate. Furthermore, V (S) is increasing in
Proof. We write the variance of sales V (S) as:
V (S) = V (SjD
Q) Pr (D
= 0 + V (DjD < Q)
Q) + V (SjD < Q) Pr (D < Q)
(1)
(z)
De…ne W
Q D; which is a normal r.v. with mean w = Q
and variance V (W ) =
2
:Note that w = = z. By construction, the events fD < Qg and fW > 0g are equivalent.
Using a result for the truncated normal distribution (see Olsen (1980)):
V (DjD < Q) = V (W jW > 0)
=
2
f1 + H (z) [ z
H (z)]g
where H ( ) = ( ) = ( ). Replacing this expression in 1 gives the formula for V (S).
Taking derivatives of the term in brackets and doing some algebra gives:
d
dz
(z)
(z) z +
(z)
(z)
=
(z) (z + (z))2
which is positive.
Proposition 2 In the periodic review base-stock inventory model with normally distributed
demand D with mean , standard deviation ; and order-upto level Q, the expected inventory
can be expressed as:
I = s K (z)
where
s
= V (S)1=2 and
K (z) = (z (z) + (z)) f (z)
Furthermore, K (z) is increasing in z:
1
(z) [z + (z)]g
1=2
:
Proof. The loss function for the standard normal is given by L (z) =
Combining this expression with the Lemma implies that
K (z) =
V (S)1=2
(z)
z (1
(z)).
(z + L (z))
Equation (1) implies I = s K (z) :
Denote f1 = (z (z) + (z)) and f2 (z) = (z)
(z) [z + (z)] : Note that f10 = (z)
and that f2 (z)
0 (otherwise, V (S) could be negative). Taking derivatives of K (z) =
1=2
f1 (z) = [f2 (z)] we obtain:
sign fK 0 (z)g = sign
(z) f2 (z)
= sign f2 (z)
= sign
which is positive given that f2 (z)
(z)
1 0
f (z) f1 (z)
2 2
1
(z) (z + (z))3
2
(z) [z + (z)]
1
1
(z + (z))2
2
0:
References
Olsen, R. 1980. Approximating a truncated normal regression with the method of moments.
Econometrica 48(5) 1099–1106.
2
APPENDIX
0
10
logINV
20
30
40
Scatter plot of aggregate inventory versus sales (by UA)
0
10
20
30
40
50
logSALES
GM monopoly
GM + other domestic + Japanese
GM + other domestic
Figure 3 – Scatter plot of log inventory (logINV) and log sales (logSALES) for three
types of market: monopoly markets (circles), markets with GM and non-GM domestic
dealers (squares) and markets with all types of dealers (GM, non-GM domestic and
Japanese). Each observation is a market, and both measures are aggregated across
dealerships in each market.
LOGSALES
NDEALERS
OLS
0.740***
(0.026)
0.026***
(0.009)
Beta
0.805
0.078
NDEALERSSQ
GMDRIVET
NFRANCH
CVQ
TRANSF
ALLSTATE
OLD
BLACK
PUBLICTRANS
LOGINCOME
COLLEGE
REPUBLICAN
TURNOUT
ARMY
FARMING
MAKENAME==BUICK
MAKENAME==CADILLAC
MAKENAME==CHEVROLET
MAKENAME==GMC
MAKENAME==PONTIAC
REGION==MW
REGION==NE
REGION==SO
CONSTANT
Observations
R-squared
-0.02
(0.032)
-0.021
(0.014)
0.085**
(0.043)
-0.886***
(0.201)
0.131***
(0.038)
-0.013***
(0.004)
0.001
(0.002)
-0.007
(0.012)
0.155
(0.116)
-0.001
(0.001)
0.074
(0.14)
-0.26
(0.159)
-0.004
(0.006)
-0.009
(0.015)
0.226*
(0.116)
0.227*
(0.119)
0.483***
(0.11)
0.326***
(0.118)
0.234**
(0.117)
-0.221***
(0.041)
-0.211**
(0.085)
-0.176***
(0.055)
-0.961
(1.21)
680
0.84
-0.012
-0.03
0.045
-0.076
0.064
-0.063
0.015
-0.01
0.028
-0.014
0.009
-0.032
-0.013
-0.012
0.096
0.09
0.243
0.133
0.108
-0.121
-0.048
-0.09
-1.074
680
0.84
OLS Sq
0.728***
(0.025)
0.181***
(0.028)
-0.013***
(0.002)
-0.023
(0.031)
-0.024*
(0.014)
0.075*
(0.042)
-0.879***
(0.196)
0.132***
(0.037)
-0.012***
(0.004)
0.002
(0.002)
-0.006
(0.012)
0.220*
(0.113)
-0.001
(0.001)
0.123
(0.137)
-0.261*
(0.155)
-0.008
(0.005)
-0.009
(0.014)
0.177
(0.113)
0.156
(0.116)
0.445***
(0.108)
0.278**
(0.115)
0.186
(0.114)
-0.196***
(0.04)
-0.183**
(0.083)
-0.127**
(0.054)
-1.943
(1.192)
680
0.85
Beta Sq
0.792
0.531
-0.465
-0.013
-0.035
0.04
-0.076
0.064
-0.06
0.029
-0.009
0.04
-0.013
0.016
-0.032
-0.025
-0.011
0.075
0.062
0.224
0.113
0.086
-0.107
-0.042
-0.065
-2.17
680
0.85
Table 10 – OLS estimates including all controls (intercept not shown). “Beta” column
shows standardized coefficients.
LOGUAPOP
LOGFRINGEPOP
LOGSALES
GMDRIVET
NFRANCH
CVQ
TRANSF
ALLSTATE
OLD
BLACK
PUBLICTRANS
LOGINCOME
COLLEGE
REPUBLICAN
TURNOUT
ARMY
FARMING
MAKENAME==BUICK
MAKENAME==CADILLAC
MAKENAME==CHEVROLET
MAKENAME==GMC
MAKENAME==PONTIAC
MAKENAME==SATURN
REGION==MW
REGION==NE
REGION==SO
REGION==WE
CONSTANT
Observations
R-squared
NDEALERS
1.616***
(0.089)
0.337***
(0.067)
0.206**
(0.097)
-0.071
(0.119)
-0.467***
(0.048)
-0.116
(0.155)
2.336***
(0.734)
0.084
(0.137)
0.033**
(0.015)
0.033***
(0.006)
-0.006
(0.045)
1.649***
(0.415)
0.017***
(0.005)
-0.805
(0.508)
-0.889
(0.591)
-0.005
(0.02)
-0.149***
(0.053)
0.044
(0.196)
0
(0)
-0.758***
(0.204)
0.087
(0.185)
0.067
(0.172)
-0.504
(0.43)
-0.233
(0.15)
-0.707**
(0.311)
-0.398**
(0.201)
0
(0)
-16.216***
(4.335)
679
0.76
Beta
0.607
0.103
0.076
-0.014
-0.228
-0.021
0.068
0.014
0.056
0.16
-0.003
0.103
0.063
-0.035
-0.037
-0.005
-0.063
0.006
0
-0.13
0.012
0.011
-0.026
-0.043
-0.055
-0.069
0
-6.167
679
0.76
Table 11 – Estimates from first step of IV regression. The dependent (endogenous)
variable is NC. Excluded exogenous variables are TOTPOP, FRINGEPOP and
COMMIN. (intercept not shown). “Beta” column shows standardized coefficients.
Manufacturer
GM
Ford
DC
Toyota
Nissan
Honda
Total
Standard
car
Luxury
car
Sport/Cross
Utility Veh.
Van
Pickup
56
38
25
26
16
14
15
9
6
9
5
3
8
10
2
5
2
4
20
14
12
9
6
5
6
2
3
1
1
1
7
3
2
2
2
1
Table 12 – Number of models offered by each manufacturer in model-year 2007. The
total number of vehicles is decomposed into multiple model segments, based on the
model segmentation of Ward’s Auto.
Segment
class
size
price
Lower Luxury
Luxury Specialty
Luxury Sport
Middle Luxury
Upper Luxury
Large Regular
Lower Middle
Lower Small
Middle Specialty
Small Specialty
Upper Middle
Upper Small
Large Cross/Utility Vehicle
Large Luxury Cross/Utility Vehicle
Middle Cross/Utility Vehicle
Middle Luxury Cross/Utility Vehicle
Small Cross/Utility Vehicle
Large Pickup
Small Pickup
Large Luxury Sport/Utility Vehicle
Large Sport/Utility Vehicle
Middle Luxury Sport/Utility Vehicle
Middle Sport/Utility Vehicle
Small Sport/Utility Vehicle
Large Van
Small Van
luxury car
luxury car
luxury car
luxury car
luxury car
standard car
standard car
standard car
standard car
standard car
standard car
standard car
CUV
CUV
CUV
CUV
CUV
pickup
pickup
SUV
SUV
SUV
SUV
SUV
van
van
large
middle
small
large
large
large
middle
small
middle
small
middle
small
large
large
middle
middle
small
large
small
large
large
middle
middle
small
large
small
lower
specialty
specialty
middle
upper
upper
lower
lower
specialty
specialty
upper
upper
std
luxury
std
luxury
std
std
std
luxury
std
luxury
std
std
std
std
Table 13 – Model segmentation (Source: Ward’s Auto).
no. of
models
17
5
14
11
6
10
6
7
9
6
16
20
7
6
20
12
7
9
11
8
9
5
16
1
4
12