The Effect of Retail Competition on Relationship-specificInvestments: Evidence from New Car Advertising

Charles Murry∗

The Pennsylvania State University

July 12, 2017

Abstract

Longstanding state regulations restrict car manufacturers from terminating relationshipswith dealers, creating differences in retail competition across brands and markets. I use thisvariation to identify the causal effect of dealer competition on dealer and manufacturer advertis-ing. I find that greater intra-brand dealer competition is associated with lower dealer advertisingand greater local advertising by manufacturers. The results are evidence that manufacturerscan encourage retail relationship-specific investments by providing downstream market power.Additionally, the results provide novel evidence on the substitution of selling effort within ver-tical relationships and the optimal design of retail networks. I discuss the relevance of thesefindings to the effects of state automobile franchise regulation and the recent financial troublesof US car manufacturers.

KEYWORDS: franchise regulation, non-price competition, automobile retailing, vertical rela-tionships

1 Introduction

In many retail markets, local competition is a key determinant of the price a retailer charges andthe selling effort it provides. Manufacturers want their downstream counterparts to charge lowprices to avoid double marginalization, but they also want retailers to provide high levels of sellingeffort. These two goals may be at odds with each other. For example, retailers may have incentivesto provide adequate effort only if they face little intra-brand competition and thus capture highrents from effort by charging high prices. If retailers do not provide enough selling effort, themanufacturer may have to invest in retail selling effort to help sell the product. However, there arecompeting theories about how retailers respond to different levels of intra-brand competition, andthere is little empirical evidence on how upstream and downstream firms substitute relationshipspecific investments.

∗Department of Economics, 508 Kern Building, University Park, PA, 16802, cmurry@psu.edu. Naibin Chen andWenjing Ruan provided excellent research assistance. I am thankful for feedback from various seminar participants.All errors are my own.

1

In this paper, I estimate the causal effect of new car dealer intra-brand competition on localmarket advertising spending by dealers and manufacturers. This is an interesting setting to considerbecause new car dealers and manufacturers are independently responsible for large amounts of localadvertising in the United States, making advertising a very visible and economically meaningfulrelationship-specific investment. Additionally, intra-brand retail competition in the automobileindustry has recently received public policy attention. During the U.S. financial crisis in 2009-2010,two U.S. manufacturers proposed closing thousands of dealers in order to limit competition betweenretail stores so that remaining dealers could better survive the recession. However, ubiquitous statefranchise laws prohibit manufacturers from terminating dealer franchise contracts. The effects ofdealer closures were the subject of Congressional hearings, policy analysis and public comment.

I use the number of same-brand dealers in a local market to represent the level of intra-brandcompetition in order to estimate the effect of competition on advertising. Estimates of the correla-tion between dealer competition and advertising may not be causal if both the number of dealersand advertising levels are chosen optimally by firms in response to demand and supply conditions.For example, manufacturers may decide to establish more dealers in markets where they face afavorable demand or cost environment, which may also imply a higher marginal benefit of adver-tising and therefore greater levels of optimal advertising. I deal with this endogeneity issue byusing a novel instrument based on the enactment of automobile franchise regulations by US statesover the past half century. From the 1950s through the 1990s, US states universally adopted newcar dealer franchise regulations that restrict the ability of automobile manufacturers to terminateexisting relationships with their franchised dealers. Improved technology of car distribution andincreased competition from foreign brands made it ideal for manufacturers to utilize smaller retailnetworks than they originally set up in the early 20th century. Because manufacturers could notadjust their dealer networks after the adoption of termination regulations, markets with differenthistorical population growth have drastically different numbers of dealers, especially for US brands.For example, US cities with recent population growth have far fewer US brand dealers than “older”cities with historically large (and low growth) populations. US manufacturers are essentially stuckwith too many dealers in older cities. On the other hand, foreign brands entered after the adoptionof these laws and have much more balanced dealer networks across markets.

Specifically, I use historical population growth as an instrument for the competitiveness of mar-kets to estimate the effect of intra-brand dealer competition on advertising. The results imply thatincreased competition is associated with lower dealer advertising. The point estimate suggests thatone additional intra-brand rival leads to a decrease in advertising of about one-fifth of advertisingexpenditures for the average dealer that sells a US brand. Additionally, the estimates imply thattotal local dealer advertising falls with the addition of a dealer, on average.

Additionally, I estimate that the presence of an additional dealer leads to an increase in lo-cal manufacturer advertising for the same brand. The point estimate implies that the averagemanufacturer would increase advertising by about one-fifth. Also, the estimates imply that man-ufacturer advertising per dealer would increase. This result along with the result from the dealer

2

analysis suggests a net substitution between dealer and manufacturer advertising. In other words,manufacturers spend more on advertising as dealers decrease advertising due to increased dealercompetition. One policy relevant implication of these results is that manufacturers could encouragemore dealer advertising, and spend less themselves, by reducing the size of dealer networks in theabsence of state termination laws.

Early work by Telser (1960) identified the problem of downstream moral hazard, where retailerslack incentives to provide adequate effort from the upstream firm’s perspective.1 However, thereis very little work, both theoretical and empirical, on how upstream and downstream firms jointlydecide relationship specific investments. One exception is Lafontaine and Slade (2007), who presenta model of two-sided moral hazard where the principal (manufacturer) and agent (dealer) bothmake a relationship-specific investment in effort. In the model, a single manufacturer charges arisk averse retailer a two-part revenue sharing tariff. The retailer’s effort is increasing in the shareof rents relative to the manufacturer, and manufacturer effort is decreasing in the retailer’s share.Additionally, the greater the return on effort, the more effort is exerted by either party. This simpleintuition suggests a net substitution between retailer and manufacturer effort, and maps nicely tomy empirical example, where I interpret the share of retailer rents as the strength of intra-branddealer price competition, and the effectiveness of effort as potential spill-overs (or lack of spill-overs)in advertising among same-brand dealers.

The effect of competition on retailers’ advertising effort depends on different mechanisms. Twolines of thought from the theoretical literature predict that advertising decreases as competition in-creases. First, Dorfman and Steiner (1954) suggest that because firms that face greater competitionmake lower margins and sell fewer products, they have a lower marginal benefit of advertising andinvest less in advertising. Second, Telser (1964) develops a framework where retailers will advertiseless as markets become more competitive if advertising partially spills over to rivals. The morerivals, the greater the likelihood advertising positively affects rival demand instead of own demand.2

An alternative relationship between competition and advertising is provided by Becker and Murphy(1993), who predict that advertising is under-supplied by firms with market power if advertisingitself is a complementary good to the advertised product. Although there is a well established the-oretical literature on horizontal competition and advertising, much less is known about the verticalsubstitution of advertising or relationship-specific investments. This is an important considerationfor total advertising supplied in a market because changes to dealer advertising might be offset, orexacerbated, by changes in manufacturer advertising.

I contribute to the empirical literature of vertical relationships by providing direct evidence ofthe role of non-price selling effort. An empirical challenge when studying non-price decisions in

1See Tirole (1988) for a summary of moral hazard in vertical markets and Winter (1993) for an example ofdownstream moral hazard with downstream competition.

2The idea of advertising spillovers has a long history in the theory of relationship-specific investments in verticalrelationships. For example Mathewson and Winter (1984) and Perry and Porter (1990) both analyze how manufac-turers discipline advertising investment by retailers when advertising partially benefits rival retailers under the samemanufacturer. Also, see Chandra and Weinberg (2017) for empirical evidence that markets with more concentrationhave higher advertising because of fewer spillovers. See Bagwell (2007) for an overview of theories of advertising.

3

vertical relationships is the difficulty of quantifying these decisions. Consequently, there is limitedempirical work in this area. For example, how does one quantify the aggressiveness or helpfulness ofsales people or the attractiveness and comfort-level of a showroom? However, advertising providesan ideal measure of effort because it is both an important business decision and easily quantifiable.Three recent papers examine vertical externalities and non-price decisions. Conlon and Mortimer(2015) consider the restocking and display of candy in vending machines, but do not consider anyeffort by the upstream firm. Their focus is on the ability of the manufacturer to foreclose rivals. Xuet al. (2014) estimate the effectiveness of price advertising for consumer trucks by manufacturersand dealer associations, and Murry (2015) uses advertising and car transactions data from a singleUS state to estimate a structural model of advertising and vertical relationships in the automobileindustry.3

My analysis also contributes to the literature on the effect of market structure on advertising.4

Chandra andWeinberg (2017) find a similar result to my dealer result: increased concentration leadsto greater advertising, where they measure concentration using HHI. They interpret their result assupporting the spillover theory of advertising from Telser (1964). As Chandra and Weinberg (2017)point out, there are a limited number of other studies that examine the effect of market structureon advertising, and in general, these studies use aggregate data and do not address the endogeneityof market structure.For example, Buxton, Davies and Lyons (1984) estimate a positive relationshipbetween concentration and advertising using industry level adversing to sales ratios. However, theydo not account for the endogeneity of market structure. See Bagwell (2007) for a comprehensiveoverview of empirical and theoretical advertising literature.

2 Industry Background and Data Description

In the United States, new cars are sold through networks of independent franchised retailers, calleddealers.5 Manufacturers sell inventory at linear wholesale prices to dealers, who sell the inventorythrough showrooms at (typically) negotiated retail prices with consumers. New car manufacturersand dealers are huge advertisers at the national and local levels. Total auto advertising is morethan any other sales category, including retail and pharmaceuticals. In 2013, car dealers andmanufacturers combined to spend about $16 billion on advertising, or about $1,000 per unit sold.6

About 65% of advertising spending is from manufacturers, and the rest is from dealers and dealerassociations.

The large amount of advertising spending, along with the fact that advertising is not coordinated3Somewhat related, there is a more established empirical literature that examines the importance of downstream

and upstream effort and the likelihood of vertical integration. Some of these studies use advertising to describeupstream effort, for example, Lafontaine and Shaw (2005).

4There are, of course, numerous studies about pricing and market structure. A particularly related paper isBrenkers and Verboven (2006), who consider how the liberalization of the European auto distribution system ofexclusive territories affected prices and consumer welfare. However, they do not incorporate advertising or sellingeffort into their analysis.

5For a more detailed overview of the industry, see Lafontaine and Morton (2010) and Murry and Schneider (2015).6Aggregate ad statistics from adage.com.

4

within the vertical channel, or among rivals, makes advertising an ideal subject to study sellingeffort in vertical relationships.7 Along with national level advertising, manufacturers typically placeadvertisements directed at brand promotion in local markets. These advertisements are familiarmainstays on broadcast and cable television, and national and local print periodicals, among othermedia outlets. In addition, individual dealers advertise on their own, using their own creativematerial. These advertisements tend to focus on dealer characteristics like service, selection, andtrust, and they are typically lower production quality. These ads are also familiar mainstays ontelevision and radio, and in local newspapers, and represent a large portion of advertising revenuesfor local media.8

The data I use include the locations of all automobile dealerships from all major manufacturersselling in the US at the end of 2013, the local media market advertising expenditures by dealersand manufacturers in 2013, and demographic information, including historical, aggregated to themedia market geographic level. The cross-sectional unit of observation is a car brand (or make)in a Nielsen media market Designated Market Area (DMA). In total the data include 32 brandsand 99 DMAs for a total of 3168 observations. See Table A.1 in the appendix for a list of brandsincluded in the analysis.

2.1 Car Dealers and Markets

Data on the locations of automobile dealerships come from an on-line marketing firm, AggDataLLC. I observe the street address for every dealership listed on the manufacturer’s website at agiven point in time. Using the zip code of the dealership, I assign each dealership to a Nielsenmedia market using information from Nielsen and GIS mapping software. There is a total of 27,221dealers represented in the 99 DMAs, which is about 75% of all the dealers in the US reported fromthe data provider.9

Using a mapping from zip codes to counties to DMAs, I construct DMA level demographicinformation from the 2010 Decennial US Census publicly available files. The DMA information fromNielsen also includes the number of TV households per DMA. Additionally, I construct historical

7Manufacturers run co-op advertising programs for their dealers. This is a type of vertical restraint that aimsto increase dealer ad spending by matching (or partially matching) dealer advertising spending up to some level.However, these programs are not heavily used by dealers; for example, an industry professional told me about 15%of co-op funds are used every year. Reasons for low take-up by dealers include the fact that advertising needs tobe approved by the manufacturer and typically focuses on the brand, thus potentially spilling over to rivals. Otherrequirements of co-op advertising may be costly for dealers to meet.

8In addition to manufacturer and dealer advertising, manufacturers fund “dealer associations.” These typicallytake the form of a group of same-brand dealers who meet two to four times a year to decide on joint marketingstrategy. However, the funds for these marketing campaigns are almost always provided by the manufacturer, andthe particular advertising content is usually similar, or identical, to the content used by the manufacturer. I do notinclude dealer association advertising because in many cases these funds are based directly on dealer revenue at arate determined nationally, so there is a mechanical relationship between sales and ads as opposed to an optimaladvertising decision by the manufacturer or dealer.

9This number is somewhat larger as than the number of dealers reported in 2010 in Lafontaine and Morton (2010).My data come from a different source than theirs, and there was likely a small increase in the number of dealers from2010 (the last year of their data) to 2013 (the year of my data). Also, I define a dealer as a single franchise. So asingle location that sells both Dodge and Chrysler is considered as two dealers.

5

Table 1: Variable Defintions

Market (DMA) Level Description Source

Population 2010 Decennial Census in millions USCTVHH Number of households with a TV KMMedInc 2010 Decennial Census Median Household

Income, in thousands USCPopGrowthX Population growth from Decennial Census in year X USCTotalDealers Total number of new car dealer in a market AD

Market-Brand Level

IntraBrandDealers Number of dealers in the market, by brand ADUSbrand =1 if brand is historically associated with the Big 3

auto manufacturers ADAdsDealer Average advertising expenditures of

dealers for a particular brand KM/ADAdsManufacturer Advertising expenditures of manufacturers

for a particular brand KM/ADdivided by IntraBrandDealers

Note: USC = US Census; AD = AggData LLC; KM = Kantar Media

DMA level population going back to 1900 using publicly available historical county level decennialcensus information from the US Census.

In Table 2, I display descriptive statistics for those variables that vary at the DMA level, namelydemographic information and counts of car dealerships. On average there are 276 car dealershipsin a DMA, and 141 (51%) of them are US brand franchises. The average population of the DMAsis about 2.5 million, and the average DMA has nearly 1 million households. On average, there isa substantial amount of variation in population growth. For example, Youngstown, Ohio has sawa decrease in population of nearly 15% since 1970, and Las Vegas had an increase of over 600%since 1970. This will be important when I discuss how state franchise regulations have shaped themarket structure of new car dealers across markets.

Table 2: Descriptive Statistics: Market (DMA) Level Variables

variable Mean SD Min Median Max Obs.

Num. of Dealers 276.455 193.936 72.000 218.000 1341.000 99Num. of US Brand Dealers 141.242 82.271 31.000 122.000 505.000 99

Population 2.699 3.097 0.661 1.723 21.254 99TVHH 0.960 1.040 0.143 0.636 7.392 99PopGrowth1930 421.772 1353.925 -2.965 126.878 12773.154 99PopGrowth1970 77.718 102.283 -14.270 50.199 637.415 99PopGrowth1990 28.673 27.502 -8.381 22.683 172.001 99MedInc 51.299 85.148 37.044 49.935 85.222 99Note: Variable construction detailed in the text. P opulation and T V HH in millions; Growth in percent;

MedInc in thousands.

6

2.2 Advertising

Kantar Media, an advertising industry consulting firm, collects advertising expenditures for adver-tisements in the most populous DMAs. The advertising expenditures are broken down by singleproducts, advertiser type (i.e. dealer or manufacturer), market, and media. Local market adver-tising is all advertising purchased from local media, for example from the local network televisionstation or local newspaper (e.g. WNBC-TV New York, or the Baltimore Sun). This is in contrast tonational advertising, which is purchased and “broadcast” nationwide, for example from the nationaltelevision network (e.g. NBC Universal), or a national print publication (e.g. Time Magazine).I compute the total yearly local advertising expenditures for all dealerships of a particular brandby summing dealer expenditures across all products that mention the brand on all media. Somedealers own multiple franchises and advertise multiple brands, and some manufacturers producemultiple brands and place ads that mention multiple brands. For example this happens often forFord and Lincoln, which are manufactured by the same parent company and are often sold from adealer who has a franchise contract with both companies. In these cases, I divide the advertisingexpenditures equally among all brands.10

For the analysis, I define dealer advertising (AdsDealer) in a given market for a given brand asthe sum of all dealer advertising in that market for that brand for the year 2013, divided by thenumber of dealers for that brand. In other words, AdsDealer is average dealer advertising for abrand in a market. I define manufacturer advertising (AdsManufacturer) as the total advertisingexpenditures of a manufacturer for a particular brand in a local market. I do not include nationaladvertising because this spending is likely not driven by local market conditions. I also do notinclude dealer association advertising because, as stated above, this type of spending typically hasa mechanical relationship with sales, which in the short run is not the result of optimal decisionsof either the manufacturer or the dealer.

I display the summary statistics for the advertising variables in Table 3. On average, averagedealer advertising spending is $85,253 in 2013. On average manufactures spend $562,554 permarket in local market advertising. However, there is substantial variation across brands andacross markets. One natural way to dichotomize brands in this industry is by US versus non-USbrand. Non-US brand dealer advertising is nearly twice as much as US brand dealer advertisingspending on average. Non-US manufacturers also spend about twice as much as US manufacturerson average in local market advertising. The fact that that non-US brands typically have fewerdealers that sell more cars per dealers is preliminary evidence that less intra-brand competition isassociated with higher advertising.

In the top panel of Table 3 I display summary statistics for the variable I use to measurecompetition, the number of intra-brand dealers in each market, IntraBrandDealers. On average,there are about 7 dealers per brand in a DMA. However, there is substantial variation across brands.For example, There are about 4 dealers per brand in a DMA for non-US brands on average, and

10Sovinsky Goeree (2008) faces a similar issue in the PC market, but uses a structural model to estimate theweights on how advertising are split between products in multi-product ads.

7

Table 3: Summary Statistics: Market-Brand Variables

Variable Mean SD Min Median Max

USBrand 0.281 0.450 0.000 0.000 1.000

IntraBrandDealers

US Brands 15.7 11.7 0 13 77Non-US Brands 4.1 6.0 0 2 64All Brands 7.3 9.5 0 4 77

AdsDealer

US Brands 53.062 100.598 0.000 21.836 1103.390Non-US Brands 97.850 167.579 0.000 41.833 2819.525All Brands 85.253 153.074 0.000 33.043 2819.525

AdsManufacturer

US Brands 874.037 1,987.959 0.000 275.809 27,908.148Non-US Brands 440.670 1,708.014 0.000 7.303 44,098.988All Brands 562.554 1,801.425 0.000 33.903 44,098.988

Summary statistics for those variables that vary at the market-brand level. Sampledescribed in the text. Advertising in thousands of US dollars.

about 16 dealers per brand for US brands. I interpret these differences across different types ofbrands as preliminary evidence that state franchise termination regulations created differences inintra-brand competition across different types of brands. I take this up in more detail in the nextsection.

3 Dealer Competition and Franchise Regulation

My empirical strategy relies on instrumenting for the number of dealers in local car markets. Theinstruments are based on the historical regulatory environment in the automobile industry anddifferences in population growth across different geographic markets. In this section I provideinstitutional background that motivates the empirical strategy and I provide analysis that amountsto a first stage to the primary analysis of dealer and manufacturer advertising in the next section.

Throughout the latter half of the 20th century, all US states adopted dealer franchise regulationsthat restrict manufacturers from unilaterally closing dealers. In general, manufacturers must give“good cause” to terminate a dealer relationship, and even then may be subject to settlementpayments. In practice, unilateral termination is rare. In this section, I provide evidence that thereare substantial differences in intra-brand dealer competition across brands and markets. I arguethat dealer termination regulations contributed to these differences for three reasons: changes indemand in local markets over the past half century, changes in the technology of new car retailingover the past half century, and the recent emergence of competition from foreign brands who enteredmarkets after the adoption of termination regulations.

In the early 20th century US manufacturers aggressively expanded their dealer networks. How-ever, because of termination laws, they are unable to adjust these retail networks in the face of

8

changing population and demand. The result is that cities that experienced the bulk of their pop-ulation growth within the past few decades (like many cities in the south and southwest UnitedStates) tend to have fewer US branded dealers per capita than cities that experienced populationgrowth earlier in the 20th century and have not grown very much in the past half century (orexperienced population decrease, like cities in the “Rust Belt” region of the United States). How-ever, foreign brands entered the US market recently and set up retail networks based on currentpopulation trends.

To provide suggestive visual evidence of the effect of termination laws, I map Toyota (Japaneseheadquartered) and Ford (US headquartered) dealers in two large American cities that have expe-rienced markedly different population growth paths over the past century. Specifically, I comparePittsburgh, an “old” city with small or declining population growth, to Phoenix, a “new” city thathas recently experienced rapid population growth. The maps of these two cities with the locationsof Ford and Toyota dealers is in Figure A1 in the Appendix.11 Although Phoenix is larger thanPittsburgh, it is striking that there are far fewer dealers in Phoenix than Pittsburgh, and the mixof Toyota and Ford dealers is much more balanced in Phoenix.

In addition to long-term changes in demand due to population growth, the technology of sellingcars has changed since the adoption of dealer termination laws. In Congressional testimony, USmanufacturers argued that the efficient scale of retail networks is smaller than their current networksize, (see SIGTARP, 2010). In fact, US manufacturers have been trying to decrease the size of retailnetworks for decades through organic means such as (a) allowing financially distressed dealers toclose, and (b) permitting the consolidation of dealers by dealer conglomerates. In contrast, foreignmanufacturers that entered the US market after the adoption of dealer termination laws set upmuch smaller dealer networks.

Lastly, the entry of foreign brands into the US market provided an impetus for US manufacturersto adjust their dealer networks in response to competition by having fewer dealers that were largerin the most desirable areas of the market. However, in “older” cities US manufacturers were unableto respond to competition by adjusting the size of their dealer network. As seen in Figure A2 inthe Appendix, competition from foreign brands started to rise in the 1980s and 1990s, after theadoption of dealer termination laws. In those markets with more recent population growth, theretail networks of US and foreign brands look more similar than in “older” markets. For example,see figure A1 in the appendix.

I display how intra-brand dealer market structure varies across brands in Table 4. The tableshows the density of DMAs that have varying numbers of dealers for different brands. Overall,DMAs are much more competitive for US brands than non-US brands. There are 4 DMAs inmy sample with 3-5 Chevrolet dealers, and there are 85 DMAs with 11+ Chevrolet dealers. Incontrast, there are 34 DMAs with 3-5 Toyota dealers and only 28 DMAs with 11+ Toyota dealers.The difference between US and non-US brands is an empirical regularity across all major brands.There is a greater disparity between the number of US and non-US dealers for luxury brands because

11The map of Pittsburgh is meant to be comparable to a similar map in Lafontaine and Morton (2010).

9

Table 4: Tabulation of DMAs with Different Counts of Dealers, by Brand

Number of Dealers

Make 0-2 3-5 6-10 11+

US Brands

Buick 0 7 25 67Chevrolet 0 4 10 85Chrysler 0 5 21 73Ford 0 5 9 85

Non-US Brands

BMW 69 21 5 4Honda 5 43 27 24Toyota 0 34 37 28Volkswagen 32 38 20 9Note: Table displays the density of dealer counts across brands.Each cell is the count of DMAs that have the number of dealers inthe column bin for each major brand. For example, Ford has 3-5dealers in 5 DMAs, and Toyota has 3-5 dealers in 34 DMAs.

US luxury brands often accompany non-luxury brands under a dual franchise. For example, thereare only four DMAs with 11+ BMW dealers, but there are 67 DMAs with 11+ Buick dealers, aBMW competitor.

Next, I present more formal evidence that the number of intra-brand dealers correlates withpopulation growth and there are significant differences in this relationship across brands. To doso, I run a regression where the left-hand side variable is the number of dealers for each brand in agiven market, and the right-hand side variables include population growth rate, a dummy for USbrand, an interaction of the two, and additional market controls. I present the results in Table 5.12

Each column represents a separate base year which I use to calculate the population growth rate,where I calculate all growth rates to the 2010 decennial census. For example, the column header“1960” uses growth rates calculated from the 1960 census to the 2010 census. As can be seen by thecoefficient on the US brand dummy, between 12 and 13 for all base years, there are many more USbrand dealers than non-US brand dealers. The estimate of the growth and brand interaction termfor the 1990 base year is −0.040(0.010), which implies that the relationship between the number ofdealers and population growth is more negative for US brands than non-US brands. For example,a 28% population growth since 1990 (the mean in the sample) is associated with nearly 2 fewer USbrand dealers in a market than foreign brand dealers, conditional on other factors – this can beseen from the PopGrowthY EAR and USBrandXPopGrowth rows of the “1990” column in Table5. This confirms the hypothesis that newer cities tend to have relatively fewer US brand dealerscompared to foreign dealers than older cities. 13 The negative relationship between population

12These regression are exactly the “first-stage” in the instrumental variables analysis in the next section.13In general population growth is negatively associated with the number of dealers. One reason for this is may

10

growth and the number of US brand dealerships is strongest for markets that have more recentlyexperienced population growth; the coefficient on the interaction term USBrandXPopGrowth ismonotonically decreasing with the growth base years.

Table 5: Regression: # of Dealers per Market and Population Growth

Growth Base Year

DepVar: IntraBrandDealers 1930 1940 1950 1960 1970 1980 1990

USBrand 12.099** 12.190** 12.413** 12.626** 12.940** 12.888** 12.778**(0.668) (0.674) (0.691) (0.704) (0.748) (0.773) (0.862)

PopGrowthYEAR -0.000** -0.001** -0.001** -0.004** -0.009** -0.021** -0.045**(0.000) (0.000) (0.001) (0.002) (0.004) (0.007) (0.013)

USBrand X PopGrowth -0.001** -0.002** -0.004** -0.008** -0.017** -0.027** -0.040*(0.000) (0.001) (0.001) (0.002) (0.004) (0.009) (0.021)

log(Population) 6.330** 6.352** 6.405** 6.490** 6.665** 6.804** 6.829**(0.859) (0.854) (0.845) (0.829) (0.814) (0.810) (0.806)

log(MedInc) 0.675 0.666 0.569 0.591 0.101 0.543 0.293(1.870) (1.850) (1.821) (1.788) (1.753) (1.749) (1.812)

Constant -2.632 -2.576 -2.147 -2.098 -0.030 -1.595 -0.328(7.104) (7.028) (6.919) (6.807) (6.687) (6.677) (6.925)

R2 0.544 0.546 0.549 0.552 0.558 0.560 0.557Observations 3168 3168 3168 3168 3168 3168 3168Note: Observation is a DMA-brand. Dependent variable is the number of dealers. Each column uses a different base year tocalculate population growth. Robust standard errors shown in parentheses. * and ** represent statistical significance at the 5%and 1% levels respectively. Population is in millions, and income is in thousands.

These results document how dealer franchise regulations affect the behavior of US car man-ufacturers. It is clear that manufacturers were less aggressive at opening dealers after the stateregulations went into effect, which was also at the same time they started facing fiercer competitionfrom imports and potential changes in the technology of retailing cars, all the while dealing withchanging demand and populations. This is systematic evidence that confirms evidence provided byLafontaine and Morton (2010), who display a map of Toyota and GM dealerships in Pittsburgh, an“old,” low growth city. Their map shows that there are many more GM dealerships than Toyotadealerships in Pittsburgh. And some of the GM dealerships are in depressed parts of the city.They also document national trends in the dynamics of dealers by brand, and from that, one couldreasonably infer that US manufacturers did not open as many dealers in new growth markets.

4 Empirical Strategy and Results

Next, I present the results of estimating the effect of downstream retail competition on the advertis-ing expenditures of car dealers and manufacturers. The empirical strategy is to regress dealer andmanufacturer advertising, separately, on the number of intra-brand dealers present in the market,market characteristics, and brand dummies. Intra-brand competition is a choice of the firms andis likely correlated with unobserved features of brands and markets that affect optimal advertising

be that setting up new dealers requires high fixed costs and substantial market research. Therefore “new” cities willhave fewer dealers in general.

11

decisions. Therefore, I instrument for competition by exploiting the way in which state franchiseregulations affect markets and brands differently. I described the institutional details that justifythis strategy in the previous sections. Next, I describe the empirical strategy in more detail, andafter that, I present and discuss the results.

4.1 Market Structure Endogeneity

In my empirical application, I represent the level of intra-brand competition in a local market bythe number of same brand dealers. In franchise industries, the number of retailers in a local area isa joint decision of the manufacturer/franchisor and willing entrepreneurs. For an entrepreneur tostart a new car dealer franchise, the manufacturer must select the application of the entrepreneurand the two parties must sign a contract. The manufacturer has the final say on entry, but sincenew car manufacturers cannot sell directly to consumers, the manufacturers rely on talented andwilling entrepreneurs in local markets to effectively run their retail establishments.

It may be the case that unobserved market-brand characteristics correlate with the joint decisionto open a franchised car dealer. For example, if the manufacturing facility of the car is in the samemarket, demand for the car might be high, and costs might be low in that particular market. Ifdemand is particularly high or costs particularly low, the manufacturer might franchise additionaldealers, but this may also be why advertising in that market is high. Therefore, OLS estimates ofadvertising on competition would not represent the causal effect of competition. I deal with theendogeneity problem by instrumenting for market structure. I do so by exploiting variation in theeffect of state franchise regulations on market structure.

Specifically, I instrument for the number of intra-brand dealers, SameBrandDealers usingmarket population growth and the population growth interacted with a US brand dummy. Asexplained above in Section 3, there are three reason I agree that this instrument has explanatorypower for SameBrandDealers. First, for technological reasons auto manufacturers needed moredealers to sell cars in the first half of the 20th century before state franchise regulations wereadopted, than are needed presently. Second, population growth throughout the country happenedin a way that does not reflect the initial dealer network choices of US manufacturers. Third, non-USbrands entered the market after state franchise laws were adopted, at a time when it was optimalfor manufacturers to have fewer dealers in retail networks because of the first point. The story isthat “older” cities (those with little recent population growth, or decline) tend to have many moreUS dealers because dealer franchise regulations restrict the ability of manufacturers to terminatedealer relationships. These differences in the number of dealers across cities will also differ by brandbecause foreign brands entered the US after states adopted franchise laws.

In my main specifications I instrument for market structure using population growth since 1990.By 1990 nearly every state adopted auto franchise regulations and foreign brands were beginningto increase their presence in the US. For example, US voluntary export restraints (VERs) expiredin 1994 and around that time US manufacturing plants of foreign brands had started to open. Thefirst stage regression results are presented in the “1990” column of Table 5 and those first stage

12

results are discussed in detail in the accompanying text of Section 3. The excluded instruments,PopGrowthY EAR and USBrandXPopGrowth have the expected signs and are statistically sig-nificant. I present robustness results to the choice of base year in the appendix.

4.2 Results

I present results for the relationship between dealer advertising and intra-brand competition inTable 6. I report OLS estimates in column (1), IV estimates in column (2), OLS with brand fixedeffects in column (3), and IV with brand fixed effects in column (4). Standard errors are clustered atthe market level and are reported in parentheses. My preferred specification is column (4) becausein that specification I account for the endogeneity of competition and control for unobserved brandlevel effects. The results suggest that there is a negative economically and statistically significantaverage effect of the number of intra-brand dealers on advertising. This is consistent with thetheoretical market power hypothesis of Dorfman and Steiner (1954) or the spillover hypothesisof Telser (1964). The point estimate implies that for the average market-brand, average dealerspending decreases by about $12,000 with an additional intra-brand dealer. This is substantial, asUS dealers average about $50,000 and non-US dealers average about $97,000.

I use the log of Population and MedInc to control for market characteristics, such as consumerdemand and the market for advertising, that are associated with advertising decisions of dealers.For example, controlling for population is important because larger markets can naturally supportlarger retail networks, so competition, as expressed in terms of the number of retail outlets, isrelative to market size. Also, in larger markets, a single advertisement reaches more eyeballs, sothe marginal benefit of a dollar of advertising is different than in smaller markets, so the rawdollars of advertising will mean something different across markets with different populations. Thecoefficient on log(Population) is positive and significant in all specifications, which is the expectedsign. The coefficient on log(MedInc) is negative and estimated imprecisely. In specifications (1)and (2) I control for USBrand, but I cannot in the second two specifications because I includebrand fixed effects.

Next, I consider the effect of intra-brand dealer competition on the advertising of the manufac-turer of the same brand. In general, the benefits to the manufacturer of having an additional retailestablishments are greater sales because of (1) wider geographic coverage and (2) lower retail pricesdue to increased dealer competition. Both of these effects imply a greater marginal benefit of ad-vertising for the manufacturer. For example sending out an advertisement is more likely to result inan additional sale the more dealers in the market. However, there is a counteracting effect. As seenabove, greater dealer competition leads to lower levels of dealer advertising. Holding everythingelse constant, this should decrease sales and implies the manufacturer has a lower marginal benefitof advertising when there are more dealers. Therefore, it is an empirical question as to whetherthe manufacturer will substitute the decrease in dealer advertising with its own advertising, or ifmanufacturer advertising is complementary and will decrease.

I present results for manufacturer advertising in Table 7. I report OLS estimates in column (1),

13

Table 6: IV Regression Results: Dealer Advertising

(1) (2) (3) (4)

SameBrandDealers 1.02 -11.59** -1.55 -11.59**(0.91) (5.57) (1.29) (5.57)

log(Population) 43.84** 122.63** 59.90** 122.63**(10.49) (34.82) (11.22) (34.82)

log(MedInc) -35.96 -29.65 -34.67 -29.65(45.60) (46.57) (44.92) (46.57)

USbrand -56.61** 90.05 – –(11.73) (61.37) – –

Constant 205.84 180.14 180.81 144.40(173.92) (176.93) (171.47) (178.04)

Brand Dummies No No Yes YesIV No Yes No Yes

Observations 3168 3168 3168 3168Note: Observation is a DMA-brand. Dependent variable is average dealer advertising.Standard errors in parentheses clustered at the DMA level. * and ** represent statisticalsignificance at the 5% and 1% levels respectively. Advertising expenditures in thousandsUSD

Table 7: IV Regression Results: Manufacturer Advertising

(1) (2) (3) (4)

SameBrandDealers 111.08** 104.79** 125.03** 104.79**(24.17) (36.04) (30.62) (36.04)

log(Population) 488.47** 527.78** 401.32** 527.78**(89.67) (159.05) (108.55) (159.05)

log(MedInc) 29.41 32.56 22.44 32.56(326.43) (329.43) (326.23) (329.43)

USbrand -858.57** -785.39* – –(274.71) (415.75) – –

Constant -449.60 -462.42 -459.61 -532.99(1217.93) (1228.76) (1221.25) (1237.86)

Brand Dummies No No Yes YesIV No Yes No Yes

Observations 3168 3168 3168 3168

Manufacturer ads per dealer $76.54 $76.57 $76.47 $76.57at mean of SameBrandDealers (7.86) (6.28) (6.03) (6.28)Manufacturer ads per dealer $80.70 $79.97 $82.32 $79.97with one additional dealer (7.86) (8.73) (8.25) (8.73)Note: Observation is a DMA-brand. Dependent variable is local market manufacturer advertising. Standard errorsin parentheses clustered at the DMA level. * and ** represent statistical significance at the 5% and 1% levelsrespectively. Advertising expenditures in thousands USD.

14

IV estimates in column (2), OLS with brand fixed effects in column (3), and IV with brand fixedeffects in column (4). In all specifications, the coefficient on SameBrandDealers is positive andeconomically and statistically significant. The point estimate (Column 4) implies that adding anadditional dealer would increase manufacturer advertising by $104,790. This is significant, as aver-age manufacturer advertising is $562,554. The results support the hypothesis that manufacturersincrease advertising as their downstream retail network becomes more competitive, substituting forthe decrease in dealer advertising estimated in Table 6. The other covariates have the expected sign,but like the dealer regressions, log(MedInc) is not significantly different from zero at conventionallevels of significance.

There could be two reasons for a positive relationship between the number of same-branddealers and total manufacturer advertising. First, an additional dealer in a market implies thatthe manufacturer serves a geographically larger area. Serving a larger geographic area within themarket should increase the manufacturer’s marginal benefit of advertising because advertising canbe seen by all consumers in the market.14 Second, if incumbent dealers decrease advertising withincreased competition as found above, the manufacturer may substitute its own advertising effort.To provide evidence that the second mechanism is relevant, I compute the predicted change inmanufacturer advertising per dealer. If there is a diminishing marginal benefit of manufactureradvertising from adding dealers to a market, then manufacturer advertising per dealer shoulddecrease as the number of dealers increase. If manufacturer advertising per dealer increases withthe number of dealers, this would be evidence that the manufacturer is substituting advertisingin the vertical relationship for each dealer. Using the estimates from specification (4) in Table 7,I predict the manufacturer advertising per dealer at the average number of dealers (7.3), and inthe case with one additional dealer in the market. The estimates imply the average manufactureradvertising per dealer increases by about $3.5 thousand (the bottom panel of Table (7). However,given the standard errors for the predictions, I cannot reject the hypothesis that manufactureradvertising per dealer does not change after an additional dealer is added to the market.

I conduct robustness checks to the main analysis and present the results in Appendix B. Ireport results using different base years for the population growth instruments PopGrowth andUSBrandXPopGrowth for dealer advertising in Table A2 and for manufacturer advertising inTable A3. In general, the results are consistent with the main results, although there is a lower effectfor manufacturer advertising with earlier growth base years. A little less than 10% of observationshave zero advertising in the data. To check if my results are sensitive to this censoring, I run Tobitand IV Tobit analysis that mimics the main analysis. The results are in Table A4, where for eachspecification in the main analysis I present the coefficient on SameBrandDealers for the associatedTobit specification.15 The results are generally similar to the main results.

14Others have estimated that consumers have a large distaste for traveling to purchase cars, see Albuquerque andBronnenberg (2012) and Murry (2015).

15For the IV Tobit specification I estimate the model using STATAs implementation of the non-linear IV estimatorproposed by Newey (1987).

15

4.3 Instrument Validity

Here I discuss two concerns that may invalidate the instruments, both of which have to do withhow consumer preferences may be correlated with market population growth. Newer cities maybe populated with younger consumers and immigrants who are more likely to prefer foreign carsto domestic cars. However, if this were the case, US brands would likely advertise less in newermarkets. These newer markets, in turn, are markets that tend to have fewer US brand dealers. Thiscorrelation between negative preferences for U.S. brands, the number of dealers, and advertisingwould bias the results in the opposite direction of my findings. Specifically, if population growthand preferences are correlated in this manner, we should expect less advertising by US brands inmarkets where we observe fewer dealers, which is the opposite of what I estimate. Because ofthis, this concern would suggest my results underestimate the true effect of dealer competition onadvertising.

Alternatively, new growth cities might be populated by older retiree transplants from northernstates. These consumers likely have have preferences that favor US brands. This correlation betweenpopulation growth and consumer preferences would bias my results in the direction of my findings.To alleviate this concern, I conduct a robustness to my main results by dropping markets that aremost likely to have northern retiree populations: markets in southeastern coastal states, Arizona,and Southern California. I drop 23 markets in total. Examples of dropped markets include Norfolk,VA, Charlotte, NC, Ft. Meyers, FL, and Tucson, AZ. I present the results from this selected samplein Table 8. In column (1) I present the results for dealer advertising that are analogous to column(4) in Table 6. In column (2) I present the results for manufacturer advertising that are analogousto the specification in column (4) of Table 7. The results in both columns for the effect of theSameBrandDealers on advertising are very similar to the results using the full sample.

4.3.1 Discussion: Competition, Advertising, and Vertical Relationships

I show that local market dealer advertising decreases with intra-brand competition. This is con-sistent with two classic theories of competitive advertising. The classic theory of advertising ofDorfman and Steiner (1954) predicts that the marginal benefit of advertising depends on the profitmargins of firms and the elasticity of sales to advertising: higher margins and a higher elasticityimply a higher marginal benefit from a dollar of advertising expenditures which, in turn, impliesgreater optimal advertising. Extrapolating the theory to my application, my results suggest thatdealers in highly competitive markets set lower prices in order to steal business from rivals, orexpand the market, and margins and advertising will be relatively low. On the other hand, if adealer faces very little intra-brand competition, it can set high prices, make high margins, face littlebusiness stealing from competitors, and has a high marginal benefit of advertising.16

16I assume advertising increases demand for the product, whether through business stealing or from marketexpansion. The extent to which advertising steals business from rivals, versus expanding the market, would affectthe magnitude of the results, but I cannot separately identify these two channels. See Roberts and Samuelson (1988)for a discussion of this trade-off.

16

Table 8: IV Regression Results: Non-Southern Sample

Dealer Ads Manufacturer Ads

SameBrandDealers -15.46** 113.80**(6.46) (41.09)

log(Population) 157.42** 548.43**(43.68) (223.04)

log(MedInc) -63.90 -181.31(59.51) (422.51)

Constant 260.16 274.29(225.37) (1583.74)

Brand Dummies Yes YesIV Yes Yes

Observations 2432 2432Note: Observation is a DMA-brand. Dependent variable is average dealeradvertising. Standard errors in parentheses clustered at the DMA level. *and ** represent statistical significance at the 5% and 1% levels respectively.Advertising expenditures in thousands USD. Sample does not include marketsin southeastern coastal states, Arizona, and Southern California.

An alternative interpretation of the dealer advertising result comes from Telser (1964), whosuggests that in cases where advertising spills over to rivals, or in other words positively affectsrivals’ demand, firms will under-advertise. As the number of rivals increase, the larger the spilloverexternality becomes, and the less advertising. This story is not inconsistent with the institutionaldetails of automobile advertising. Advertising by new car dealers will typically mention the commonbrand, so there is clearly a mechanism for advertising to spill over to intra-brand rivals. Otherstudies have found that distance is a large factor in dealer choice (see Albuquerque and Bronnenberg,2012; Murry, 2015), so it could be reasonable to think that even if a consumer sees an advertisementfrom one dealer, she might instead visit a closer dealer who sells that same brand. As such, wecould expect lower dealer advertising as competition increases, depending on the magnitude ofthe spillover. Chandra and Weinberg (2017) find evidence to support this theory of competitiveadvertising in the beer market.

On the manufacturer advertising side, the results provide novel evidence for how verticallyrelated firms make decisions about relationship-specific investments, in this case, advertising. I showthat local market advertising by car manufacturers increases (both total advertising and advertisingper dealer) with the number of intra-brand dealers, on average. This result implies a trade-off facedby manufacturers – limiting downstream market power comes at the cost of decreased downstreaminvestment in selling effort. My results suggest that manufacturers optimally substitute a decreasein downstream selling effort with upstream selling effort. Even though there is a rich literature thatdiscusses how vertically related firms coordinate prices, there is little, if any, empirical evidence onhow selling effort is substituted within vertical relationships.

To provide intuition for this result, I relate it to the model proposed in Lafontaine and Slade

17

(2007) and Lafontaine and Slade (2001). In their model, a manufacturer and dealer exert effort togenerate sales. Dealer effort is not observable by the manufacturer, so effort cannot be contracted.Instead, the manufacturer charges a two-part tariff, where the linear tariff is a share of revenues.The optimal choice of effort for either party depends on the marginal value of a sale (i.e. the shareof revenue) multiplied by the marginal value to the sales process of effort. If the revenue sharefor the dealer decreases, for example because of increased intra-brand dealer competition, then thedealer decreases effort and the manufacturer increases effort. There is an analogous result if thereturn on effort decreases, for example because of effort spillovers. 17 The intuition for these resultsis that parties within the vertical relationship consider their private benefit of effort when decidingoptimal effort. If dealer’s private benefit decreases at the expense of the manufacturer’s, then therewill be substitution of effort.

The results highlight that the choice of the number of retailers in a local market is a crucialdecision facing a manufacturer. A highly competitive downstream market implies too little retailinvestment in advertising by dealers. One suggestion from the theoretical IO literature is thatmanufacturers can give retailers exclusive territories to encourage relationship-specific investments.My results confirm that this tool could be effective at promoting downstream investments.

However, there is a cost to giving downstream firms too much market power. In the absenceof price restraints, retailers with market power will set a high price, and the vertical structurewill suffer from double marginalization, which will lead to lower total welfare for consumers andall firms. Also, with fewer dealers, manufacturers would cover a smaller geographic area within agiven market. The trade-off implies that there is an optimal level of downstream competition fromthe manufacturer’s point of view. One that encourages dealer investment in advertising, but on theother hand limits dealer market power to avoid pricing externalities. My results provide evidencethat there exists a relationship between market structure and the level of downstream and upstreameffort. However, one would need to empirically understand how downstream market structure affectsprices and sales to draw inferences about the optimal level of intra-brand competition.

On the policy side, my results support US auto manufacturers’ arguments to close thousandsof dealerships during the 2009 TARP bailout. During Congressional Testimony, GM and Chryslerargued that closing dealers would leave remaining dealers stronger and able to invest more intheir franchises. GM argued that dealers would invest more in promotion and advertising (seeSIGTARP, 2010). It is clear from the data and my first stage results in Section 3 that US branddealer networks are much larger than foreign brands. Although I cannot make a statement aboutthe optimal number of dealers, there is evidence to suggest that state regulation has distorted thedealer networks of US brands. My results confirm the intuition of US manufacturers that closingdealers would lead to existing dealers investing more in advertising, this taking this burden fromthe manufacturer.

17The model presented in Lafontaine and Slade (2007) and Lafontaine and Slade (2001) assumes that dealer effortand manufacturer effort are perfect substitutes in the sales process. Adding complementarity of effort in the salesprocess would naturally lead to complementarity of optimal effort decisions. See Murry (2015) for another exampleof a model of two-sided vertical effort where optimal effort between manufacturers and retailers is a substitute.

18

5 Conclusion

In this paper, I use cross market and brand variation in the effect of dealer franchise regulationsas an instrument to estimate the effect of downstream competition on dealer and manufactureradvertising. I find that dealer advertising decreases with the number of intra-brand competitors.This result is consistent with two theories of advertising: (i) the theory that firms with lowermargins have a lower marginal benefit of advertising (Dorfman and Steiner, 1954) and (ii) the ideathat advertising might spill over to rivals, and this externality increases as the number of rivalsincrease (Telser, 1964). Additionally, I estimate that manufacturer advertising increases in thenumber of same brand dealers, which is evidence that dealer and manufacturer advertising aresubstitutes within the vertical channel. As dealers decrease advertising, manufacturers make upfor this with their own advertising.

My findings have implications for the recent policy debate concerning the size of US brand dealernetworks. Some have suggested (for example Lafontaine and Morton, 2010) that dealer franchiseregulations contributed to the weakness of US manufacturers during the last decade, and especiallyduring the 2009 financial crisis. One way this may happen is by forcing manufacturers to maintainlarger-than-optimal selling networks. My results suggest by reducing selling networks, remainingdealers would advertise more and manufacturers less. However, my analysis cannot speak to howequilibrium prices would change in the vertical relationship, and hence the surplus of manufacturersand remaining dealers.

More broadly, my results have implications for how upstream firms should design selling net-works. My results suggest there is a very clear reason why upstream firms would want to limitdownstream competition, that is to encourage selling effort. However, my analysis cannot speakto how the double marginalization problem might be exacerbated as downstream market power in-creases. But in many industries upstream first design wholesale contracts to help eliminate doublemarginalization. Whether these types are contracts affect selling effort as well is a question foreach specific situation.

The result I present about dealer advertising is important because there is limited literatureon how market structure affects advertising that deals with the endogeneity of market structure.A notable exception is Chandra and Weinberg (2017), who also find that greater competitionleads to less advertising. The results about manufacturer advertising are important because thereis very little literature on the substitutability of downstream and upstream selling effort, bothempirical or theoretical. However, this is a phenomenon that is common in many industries.Examples include advertising by computer chip producers and computer and software producers,clothing producers and retailers, and durable home goods producers and home improvement stores.Developing a theoretical framework to understand the substitutability of selling effort in verticalrelationships, and how this is related to classic theories of vertical externalities, is a potentiallyinteresting direction for future research.

19

ReferencesAlbuquerque, P., and B. Bronnenberg. 2012. “Measuring the impact of negative demandshocks on car dealer networks.” Marketing Science, 31(1): 4–23.

Bagwell, Kyle. 2007. “The Economic Analysis of Advertising.” In Handbook of Industrial Organi-zation. Vol. 3 of Handbook of Industrial Organization, , ed. Mark Armstrong and Robert Porter,Chapter 28, 1701–1844. Elsevier.

Becker, G.S, and Kevin M. Murphy. 1993. “A Simple Theory of Advertising as Good or Bad.”The Quarterly Journal of Economics, 941–964.

Brenkers, Randy, and Frank Verboven. 2006. “Liberalizing a distribution system: the Euro-pean car market.” Journal of the European Economic Association, 4(1): 216–251.

Buxton, Anthony J, Stephen W Davies, and Bruce R Lyons. 1984. “Concentration andadvertising in consumer and producer markets.” The Journal of Industrial Economics, 451–464.

Chandra, Ambarish, and Matthew Weinberg. 2017. “How Does Advertising Depend onCompetition? Evidence from US Brewing.” working paper.

Conlon, Christopher T, and Julie Holland Mortimer. 2015. “Efficiency and foreclosureeffects of all-units discounts: Empirical evidence.” working paper.

Dorfman, Robert, and Peter O Steiner. 1954. “Optimal advertising and optimal quality.” TheAmerican Economic Review, 44(5): 826–836.

Lafontaine, Francine, and Fiona Scott Morton. 2010. “Markets: State Franchise Laws, DealerTerminations, and the Auto Crisis.” Journal of Economic Perspectives, 24(3): 233–50.

Lafontaine, Francine, and Kathryn L Shaw. 2005. “Targeting managerial control: evidencefrom franchising.” RAND Journal of Economics, 131–150.

Lafontaine, Francine, and Margaret E. Slade. 2001. “Incentive Contracting and the FranchiseDecision.” Game Theory and Business Applications, , ed. Kalyan Chatterjee and William F.Samuelson, 133–188. Boston, MA:Springer US.

Lafontaine, Francine, and Margaret Slade. 2007. “Vertical integration and firm boundaries:the evidence.” Journal of Economic Literature, 45: 629–685.

Mathewson, G Frank, and Ralph Albert Winter. 1984. “An economic theory of verticalrestraints.” Rand Journal of Economics, 15(1): 27–38.

Murry, Charles. 2015. “Advertising in Vertical Relationships: An Equilibrium Model of theAutomobile Industry.” mimeo, The Pennsylvania State University.

Murry, Charles, and Henry Schneider. 2015. “The Economics of Retail Markets for Newand Used Cars.” In Handbook on the Economics of Retail and Distribution. , ed. Emek Basker.Edward Elgar.

Newey, Whitney K. 1987. “Efficient estimation of limited dependent variable models with en-dogenous explanatory variables.” Journal of Econometrics, 36(3): 231–250.

20

Perry, Martain K, and Robert H Porter. 1990. “Can resale price maintenance and franchisefees correct sub-optimal levels of retail service?” International Journal of Industrial Organization,8(1): 115–141.

Roberts, Mark J., and Larry Samuelson. 1988. “An Empirical Analysis of Dynamic, NonpriceCompetition in an Oligopolistic Industry.” RAND Journal of Economics, 19(2): 200–220.

SIGTARP. 2010. “Factors Affecting the Decisions of General Motors and Chrysler to Reducetheir Dealership Networks.” Office The Special Inspector General for the Troubled Asset ReliefProgram SIGTARP-10-008.

Sovinsky Goeree, Michelle. 2008. “Limited Information and Advertising in the US PersonalComputer Industry.” Econometrica, 76(5): 1017–1074.

Telser, Lester G. 1960. “Why should manufacturers want fair trade?” Journal of law and eco-nomics, 86–105.

Telser, Lester G. 1964. “Advertising and competition.” The Journal of Political Economy, 537–562.

Tirole, Jean. 1988. The theory of industrial organization. MIT press.

Winter, Ralph. 1993. “Vertical Control and Price Versus Non-price Competition.” QuarterlyJournal of Economics, 108(1): 61–76.

Xu, Linli, Kenneth C Wilbur, S Siddarth, and Jorge M Silva-Risso. 2014. “Price Adver-tising by Manufacturers and Dealers.” Management Science.

21

AppendixA Data Supplement

Table A1: List of Brands

Acura GMC MitsubishiAudi Honda NissanMercedes-Benz Hyundai PorscheBMW Infiniti Land RoverBuick Isuzu ScionCadillac Jaguar SmartChevrolet Jeep SubaruChrysler Kia ToyotaDodge Lexus VolkswagenFiat Lincoln VolvoFord MazdaNote: US brands in bold.

B Robustness Results

Table A2: Dealer Advertising Robustness to Different Base Years for Population Growth

Population Growth Base Year

1930 1940 1950 1960 1970 1980 1990SameBrandDealers -15.295** -14.604** -13.008** -12.496** -11.057** -11.426** -11.594**

(4.030) (4.151) (4.028) (4.216) (4.225) (4.798) (5.566)log(Population) 145.760** 141.439** 131.466** 128.267** 119.275** 121.583** 122.631**

(28.162) (28.650) (27.376) (27.592) (27.122) (30.529) (34.824)log(MedInc) -27.802 -28.148 -28.946 -29.202 -29.922 -29.737 -29.653

(48.697) (48.168) (47.068) (46.872) (46.150) (46.390) (46.569)Constant 130.984 133.491 139.278 141.134 146.352 145.013 144.405

(183.792) (181.983) (178.229) (178.050) (175.941) (177.002) (178.038)Observations 3168 3168 3168 3168 3168 3168 3168* indicates p-value<0.05. ** indicates p-value < 0.01.

22

Table A3: Manufacturer Advertising Robustness to Different Base Years for Population Growth

Population Growth Base Year

1930 1940 1950 1960 1970 1980 1990SameBrandDealers 37.420** 39.259** 44.329** 62.002** 77.109** 86.168** 104.791**

(17.714) (18.954) (21.412) (27.009) (29.673) (32.083) (36.038)log(Population) 948.762** 937.271** 905.587** 795.155** 700.756** 644.152** 527.784**

(175.442) (171.798) (164.402) (155.372) (149.292) (153.276) (159.047)log(MedInc) 66.250 65.330 62.794 53.957 46.402 41.872 32.559

(368.095) (366.600) (362.573) (349.316) (339.960) (335.050) (329.429)Constant -777.278 -770.609 -752.224 -688.142 -633.365 -600.518 -532.992

(1351.079) (1346.403) (1334.014) (1291.587) (1262.515) (1246.576) (1237.858)Observations 3168 3168 3168 3168 3168 3168 3168* indicates p-value<0.05. ** indicates p-value < 0.01.

Table A4: Robustness Check: Tobit Specification

Dealer(1) (2) (3) (4)

SameBrandDealers -0.019 -2.602** -6.255** -12.642**(0.346) (0.491) (0.808) (1.739)

Brand Dummies No No Yes YesIV No Yes No YesObservations 3168 3168 3168 3168

Manufacturer(1) (2) (3) (4)

SameBrandDealers 89.234** 115.965** 57.374** 96.232**(3.127) (4.667) (7.096) (15.633)

Brand Dummies No No Yes YesIV No Yes No YesObservations 3168 3168 3168 3168* indicates p-value<0.05. ** indicates p-value < 0.01. Additional control variables

log(P opulation), log(medInc) and a constant are included in all regressions, but not re-ported. IV Tobit specifications are estimated using STATA’s implementation of Newey(1987).

23

C Additional Figures

(a) Pittsburgh, PA (b) Phoenix, AZ

Figure 1: Toyota and Ford Dealers in Two Major Cities

24

Year

1961 1966 1971 1976 1981 1986 1991 1996 2001 2006 2011

Ma

rke

t S

ha

re (

%)

0

10

20

30

40

50

60

Ford

General Motors

Honda

Toyota

Volkswagen

Figure 2: Market Shares of Major Brands, 1961-2013

25