Close
About
FAQ
Home
Collections
Login
USC Login
Register
0
Selected
Invert selection
Deselect all
Deselect all
Click here to refresh results
Click here to refresh results
USC
/
Digital Library
/
University of Southern California Dissertations and Theses
/
Quality investment and advertising: an empirical analysis of the auto industry
(USC Thesis Other)
Quality investment and advertising: an empirical analysis of the auto industry
PDF
Download
Share
Open document
Flip pages
Contact Us
Contact Us
Copy asset link
Request this asset
Transcript (if available)
Content
QUALITY INVESTMENT AND ADVERTISING:
AN EMPIRICAL ANALYSIS OF THE AUTO INDUSTRY
by
Linli Xu
A Dissertation Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(BUSINESS ADMINISTRATION)
May 2012
Copyright 2012 Linli Xu
ii
DEDICATION
I would like to dedicate this dissertation to my family, especially…
my beloved husband, Yi, for his infinite support, understanding and patience;
my Mom and late Father for instilling the importance of hard work and higher
education;
my brother for encouraging me to realize my dreams.
iii
ACKNOWLEDGEMENTS
I would like to thank all of those people who have helped make this dissertation possible.
I am especially grateful to my advisors, Dr. S. Siddarth and Dr. Kenneth C. Wilbur, for
their guidance, encouragement, support, and patience. As mentors they have taught me
more than I could ever give them credit for here. Their sincere interests in research have
been a great inspiration to me. I would also like to thank my committee members Dr.
Anthony Dukes, Dr. Lan Luo, Dr. Hai Che, and Dr. Guofu Tan for their very helpful
insights, comments and suggestions during this process. Additionally, I would like to
acknowledge the generosity of Dr. Che-Lin Su for allowing me to use the software
KNITRO free of charge.
iv
TABLE OF CONTENTS
Dedication ........................................................................................................................... ii
Acknowledgements ............................................................................................................ iii
List of Tables ...................................................................................................................... v
List of Figures .................................................................................................................... vi
Abstract ............................................................................................................................. vii
Overview ............................................................................................................................. 1
Chapter One: Price Advertising By Multiple Channel Members ....................................... 4
Chapter One: Introduction ........................................................................................... 4
Chapter One: Experimental Evidence ......................................................................... 8
Chapter One: Econometric Evidence ........................................................................ 16
Chapter One: Practical Importance ........................................................................... 27
Chapter One: Discussion ........................................................................................... 32
Chapter Two: Predicting Incumbents’ Product Quality Reactions To Entry ................... 34
Chapter Two: Introduction ........................................................................................ 34
Chapter Two: Model ................................................................................................. 38
Chapter Two: Estimation and Simulation ................................................................. 45
Chapter Two: Empirical Application ........................................................................ 49
Chapter Two: Results ................................................................................................ 55
Chapter Two: Discussion .......................................................................................... 62
Bibliography ..................................................................................................................... 65
Appendices ........................................................................................................................ 71
Appendix A: Truck Prices ......................................................................................... 71
Appendix B: Model Selection ................................................................................... 75
Appendix C: Nonparametric Identification, Endogeneity and Slope Endogeneity ... 79
Appendix D: Cross-Partial Derivatives of Market Shares ........................................ 84
Appendix E: Alternate Explanations for Estimated Advertising Interactions .......... 86
Appendix F: Policy Function Derivation .................................................................. 92
v
LIST OF TABLES
Table 1: Prevalence of Advertising by Type ...................................................................... 7
Table 2: Advertising Interaction Estimates....................................................................... 23
Table 3: Price-Advertising Interactions ............................................................................ 25
Table 4: Advertising Main Effects .................................................................................... 25
Table 5: Gas Price and Holiday Effects ............................................................................ 26
Table 6: Aggregate Channel Profit Gains with Channel Advertising Coordination ........ 32
Table 7: Simulation Results .............................................................................................. 49
Table 8: Model Parameter Estimates without Truck Features .......................................... 56
Table 9: Prediction Accuracy Comparison ....................................................................... 61
Table 10: Price Regression Results................................................................................... 73
Table 11: Advertising Carryover Measure Comparisons ................................................. 75
Table 12: Fit Statistics Comparison across Model Specifications .................................... 76
Table 13: F-series Transaction Characteristics across Advertising Intensity Levels ....... 90
Table 14: Price Advertising Content Analysis.................................................................. 91
vi
LIST OF FIGURES
Figure 1: Screenshot of the Two Price Advertisement Stimuli ........................................ 10
Figure 2: Average Perceived Quality Ratings by Treatment ............................................ 14
Figure 3: Average Perceived Quality Ratings by Stated Attribution ................................ 14
Figure 4: Truck Prices 1984 – 2009 (in 1999 dollars) ...................................................... 52
Figure 5: Market Shares by Truck Model 1984 – 2009 .................................................... 53
Figure 6: Estimated Perceived Quality after Discretization ............................................. 57
Figure 7: Quality Investment in Equilibrium .................................................................... 58
Figure 8: Value Functions in Equilibrium ........................................................................ 59
Figure 9: Advertising Expenditure over Weeks of Each Calendar Year .......................... 79
Figure 10: Advertising Expenditure over Weeks of Each Calendar Year ........................ 80
Figure 11: Distribution of TV Advertising Expenditures over Program Genres .............. 87
Figure 12: Distribution of TV Advertising Expenditures over Half-hours ....................... 87
Figure 13: Distribution of TV advertising expenditures over networks and affiliates ..... 88
vii
ABSTRACT
Quality is one of the most important factors that drive the market position of a product,
which determines the success of the firm. Firms may be able to influence how consumers
perceive their product quality through product design, advertising, promotion, pricing,
and distribution channel.
This dissertation first investigates how price advertising by different channel
members would affect consumer quality perceptions. It suggests that manufacturer price
advertising leads to lower perceived product quality than price advertising by dealers, and
such effect has significant impact on total channel profits. This implies that
manufacturers and their retailers might benefit from coordinating their price advertising
messages and could consider to pulsing their brand advertising and price advertising
simultaneously.
The second chapter of this dissertation examines the effects of competition and
product entry on firms’ investment in perceived product quality. It shows that
competition has a greater impact on quality laggard’s investment, and that a new product
entry might cause a more serious threat to quality laggards and could drive them out of
business. It also demonstrates that the recently developed dynamic structural models
could be used to predict competitive reactions in product quality and could improve the
prediction accuracy comparing with a reduced-form descriptive model.
1
OVERVIEW
Quality might be the most important factor that determines the success of a product or a
company. Studies have shown that product quality has positive influence on market share
(Phillips et al. 1983), return-on-investment (Phillips et al. 1983), premium prices (Tellis
and Wernerfelt 1987), advertising (Tellis and Fornell 1988), and stock market return
(Aaker and Jacobson 1994). However, it has been established that it is consumers’
perceptions of product quality that drive product preferences, satisfaction, loyalty, sales,
and profitability (Mitra and Golder 2006).
There are various marketing strategies that firms could adopt to influence how
consumers perceive their product quality. Among these marketing strategies, advertising
plays an essential role in forming quality perceptions. Nelson (1970) proposed that high
level of advertising spending may signal high product quality for experience goods.
Sridhar and Zhao (2009) showed that there generally is a positive relationship between
advertising expenditure and perceived product quality, even after controlling for
objective product quality, price and market share. The first chapter of my dissertation
investigates this relationship for one particular type of advertising, price advertising, in a
channel setting.
“Price advertising” communicates product price and availability (Mela et al. 1997,
Jedidi et al. 1999). In many industries, manufacturers advertise brand and price
information while retailers or dealers advertise price information. For example,
automobile manufacturers typically use advertising to communicate brand messages such
as “Ford Tough” or product characteristics like “V-8 engine.” They also use advertising
2
to send price messages such as temporary manufacturer rebates or “employee pricing.”
Dealers, on the other hand, typically advertise price terms such as “low APR” or
announce sales events such as a “Chevy Presidents’ Day Sale.”
It is known that prices may signal quality, that prices are often communicated by
advertising, and that the source of an advertisement is a key determinant of its effect.
Taken together, these three facts suggest price advertising by different channel members
may have different effects. In particular, price advertising by manufacturers should result
in lower product quality perceptions than price advertising by retailers or dealers. The
first chapter of my dissertation uses content analysis, an experiment, and econometrics to
establish the internal and external validity of this prediction. To assess the potential
importance of the effect, counterfactual analysis suggests that full coordination of the
channel’s price advertising messages could raise channel profits by about 3%.
Besides advertising, firms could make direct investment in physical product
features to influence consumer perceptions about product quality. For example, high tech
companies like Apple constantly spend billions of dollars on product design in hope for
bringing superior quality to satisfy the market need and gain competitive advantage in the
market place. Due to competition, the amount of investment for a firm may depend not
only on the current quality of its own product, but also on the perceived quality of its
competitors’ products. This implies the importance of firms being able to predict their
competitors’ reactions in product quality in order to maintain their competitive
advantages.
3
The second chapter of my dissertation aims to investigate the usefulness of
dynamic structural models in predicting such competitive reactions. This framework has
attracted broad attention in both industrial organization and marketing literature and has
proven its value repeatedly by enabling researchers to understand the impacts of
competition and governmental policies in a variety of settings. However, whether it could
be useful to make predictions on competitive reactions in product quality is an empirical
question that, to my knowledge, has not been investigated.
Applying this framework could also help firms understand how perceived quality
competition would influence their investment decisions and how a new product entry
would affect incumbents’ investment reactions. The analysis reveals three insights about
how competition in product quality influences firms’ equilibrium investments. First, the
quality leader’s optimal investment depends less on the quality differential than the
quality laggard’s. Second, having the quality laggard as a competitor has a greater impact
on the firm’s optimal investment. Third, the quality laggard has less incentive to invest
than the quality leader when the quality differential gets bigger. The results also indicate
that the structural model’s prediction hit-rate doubles that of the benchmark model, and
its mean squared prediction error is 22% lower.
4
CHAPTER ONE: PRICE ADVERTISING BY MULTIPLE CHANNEL
MEMBERS
Chapter One: Introduction
It is well known that prices may signal product quality. Wolinsky (1983) and Milgrom
and Roberts (1986) proved that this may occur in equilibrium when consumers are
imperfectly informed about products’ characteristics. Laboratory experiments have
repeatedly confirmed this prediction. Deal offers have been found to undermine the
perceived quality of the discounted item and lower the probability of future purchases.
Rao and Monroe’s (1989) meta-analysis of 36 laboratory studies showed that price
consistently increased consumers’ product quality perceptions across a variety of
experimental designs.
In practice, firms often use advertising to send price cues to consumers. Kaul and
Wittink (1995) synthesized a large econometric literature by defining two types of
advertising content. “Price advertising” communicates product price and availability
(Mela, Gupta and Lehmann 1997, Jedidi, Mela and Gupta 1999). It is usually delivered
through local media due to geographical variation in prices and availability. “Nonprice
advertising” decreases price sensitivity by communicating brand positioning and unique
brand attributes. It is typified by manufacturers’ use of national media and is often
referred to as “brand advertising.”
It is also well known that the source of an advertisement—that is, the perceived
sender of the message—is a key determinant of advertising response. A leading
advertising textbook (Belch and Belch 2007) identifies the source, the message and the
5
medium as the three key advertising characteristics that determine advertising
effectiveness. The perceived source of an advertisement may be either the organization
that paid for the ad, or a character or spokesperson appearing within the ad. It is known
that, when consumers attribute a discount offer to the brand or product, it has a negative
effect on perceived quality (Folkes 1988, Lichtenstein et al. 1989, Burton et al. 1994),
consistent with the quality-signaling function of price.
If price signals product quality, if price is communicated via advertising, and if
the source of an advertisement influences consumer response, then manufacturer price
advertising should be less effective than price advertising by retailers or dealers. This is
because the manufacturer’s price communications send a stronger signal about product
quality, since the manufacturer makes the product. This prediction is the central theme of
the current chapter.
This theme is important because manufacturers frequently use price advertising in
addition to brand advertising. Dealers or retailers typically use price advertising, and
seldom use brand advertising for the products they sell. To gain some insights into the
extent to which manufacturers and dealers used these different advertising stratetgies, we
started with the 22 product categories studied by researchers cited in Kaul and Wittink
(1995). We used the TNS Stradegy database to identify whether or not any manufacturer
in each category had spent money on TV or newspaper advertising between 2005 and
2008. We also used the same database to identify whether or not manufacturers in each
category had expenditures on free standing inserts (FSI) during the same time, which we
took as indicating that the manufacturer had undertaken price advertising. For single
6
category retailers (14% of the sample), we randomly watched dozens of ad creatives paid
for by each retailer (downloaded from the same database) and classified the retailer as
having used price advertising if the watched ads contained price information. If the ads
did not contain price information, we classified the retailer to be doing brand advertising.
Because the other multi-category retailers (86% of the sample) mostly sell consumer
packaged goods, we could safely classify them as doing mostly price advertising and not
brand advertising. Table 1 shows that manufacturer brand advertising is prevalent in all
categories, while manufacturer price advertising was used in 64% of categories.
Correspondingly, while dealer price advertising was prevalent in all categories, retailer
brand advertising for manufacturer products was found in none of the categories.
1
1
Retailer advertising of manufacturer brands cannot be ruled out completely: some manufacturers’
cooperative advertising programs require that retailers communicate differentiating messages about the
manufacturer’s product. Still, retailer advertising of manufacturer brands appears to be the exception rather
than the rule, in contrast to manufacturer price advertising, which appears to be quite common.
7
Table 1: Prevalence of Advertising by Type
The current chapter investigates the relationship between manufacturer price
advertising and retailer price advertising in three ways. First, an experiment strongly
supports the prediction that manufacturer price advertising lowers consumers’ product
quality perceptions relative to dealer price advertising. This establishes the internal
Product Category
Manufacturer
Brand
Advertising
Manufacturer
Price
Advertising
Dealer
Brand
Advertising
Dealer
Price
Advertising
Yogurt Y Y N Y
Confectionary/Candy Bars Y Y N Y
Ground Coffee Y Y N Y
Soft Drinks Y Y N Y
Frozen Waffles Y Y N Y
Ready-to-Eat Cereals Y Y N Y
Sparkling Wine Y Y N Y
Aluminum Foil Y Y N Y
Hair Spray Y N N Y
Detergents Y Y N Y
Insecticides Y N N Y
Deoderants Y N N Y
Suntan Lotions Y Y N Y
Liquid Household Cleansers Y Y N Y
Bath Tissue Y Y N Y
Ketchup Y N N Y
Disposable Diapers Y N N Y
Cat Litter Y N N Y
Dry Dog Food Y Y N Y
Cigarettes Y N N Y
Electric Shavers Y N N Y
Television Sets Y Y N Y
Proportion "Yes" 100% 64% 0% 100%
Notes
Product categories studied by papers cited in Kaul and Wittink's (1995) meta-analysis are
included, except three undisclosed CPG product categories and categories that used
franchisees or vertically integrated retailers to distribute goods (e.g., gasoline, banks,
airlines, law firms).
8
validity of the prediction discussed above. Second, an econometric analysis of the market
for pick-up trucks shows a pattern of results consistent with the experimental evidence,
establishing the external validity of the experimental result. Third, a counterfactual is
analyzed to assess the potential importance of this prediction in practice. It finds that full
coordination of the channel’s price advertisements could raise channel profits by about
3%. To the best of our knowledge, and despite an extensive search, all three of these
points are novel contributions to the literature.
This prediction is tested in a category with frequent price advertisements: full-size
pick-up trucks. This product category was more heavily advertised than any other in 2007.
Automobile manufacturers spent about $6 billion on truck advertising, including more
than $2 billion on price advertising. Dealers associations spent an additional $3 billion on
price advertising. About one in every twelve US television commercials advertised a
pick-up truck.
Chapter One: Experimental Evidence
Perhaps surprisingly, no previous paper has directly manipulated the channel member as
the source of a price advertisement. This section shows that consumers exposed to price
advertising from dealers rate truck quality more highly than those exposed to price
advertising from manufacturers.
Stimuli
To create the stimuli, a manufacturer price advertisement was modified so that its source
was either Ford (a pick-up truck manufacturer) or the Texas Ford Dealers Association. In
the manufacturer source condition, the audio of the commercial was replaced with an
9
audio clip in which a researcher recited the script from the original ad with a Ford
commercial song playing in the background. This ensured that the narrator’s voice and
background music could be kept constant across the two price advertisements in the
experiment.
Three changes were made to manipulate the source of the advertisement. First,
source subjects and pronouns in the script were appropriately modified. Second, the
geographic frame of reference was changed from the national market to the specific local
market. Third, the onscreen logo and call to action were changed appropriately to
highlight that the commercial came from the Texas Ford Dealers Association rather than
Ford. All three of these changes are consistent with dealers associations’ typical price
advertisements for pick-up trucks.
Figure 1 provides download links to watch the two commercials, complete scripts,
and pictures of the logos in the ads. In sum, this manipulation was rather subtle. 87% of
the words and 90% of the video frames are identical in the two stimuli.
A pretest with 147 subjects was conducted to judge whether the manipulation was
effective. Each person was randomly assigned to watch one of the two price
advertisements and then answered a multiple choice question about who they thought had
paid for the advertisement: (a) Ford, (b) the Texas Ford Dealers Association, (c) other or
(d) don’t know/not sure. 75% of the subjects who watched the manufacturer source
stimulus chose option (a) while 74% of the subjects who watched the dealer source
stimulus chose option (b). Both figures are different from 50% at the 99% confidence
level, thus confirming that the source manipulation of the price advertisement is valid.
10
Subject Pool
The survey link was posted on eight popular internet forums devoted to pick-up trucks. It
was also advertised on Facebook, targeting users whose profiles indicate that they “like”
pick-up trucks or major pick-up truck brands. 337 truck enthusiasts completed the
experiment, taking an average of about three minutes each.
Experimental Design
The experiment was conducted online using Qualtrics software. On the first page of the
survey, each subject was presented with a manufacturer brand advertisement, followed by
Revised price advertisement script with revisions
indicated by added emphasis:
“Since Ford introduced the Ford family plan, hundreds
of thousands of Texans have joined in on the savings.
Now it’s been extended to next September 6. Get
employee pricing on F-series, Texas’ best selling
truck, including F-150, the highest ranked light-duty
full-size pick-up in initial quality by JD Power &
Associates. Now get a built Ford tough F-150 for just
15,270. That’s over 6,000 savings on a new F-150. No
hassles, no gimmicks. Visit your Texas Ford Dealer
today!”
http://www.youtube.com/watch?v=lCxHNOteVgs
Original manufacturer price advertisement (emphasis
added):
“Since we introduced the Ford family plan, hundreds of
thousands of Americans have joined in on the savings.
Now we’re extending the plan until next September 6.
Get employee pricing on F-series, America’s best
selling truck, including F-150, the highest ranked light-
duty full-size pick-up in initial quality by JD Power &
Associates. Now get a built Ford tough F-150 for just
15,270. That’s over 6,000 savings on a new F-150. No
hassles, no gimmicks. Visit your local Ford store
today!”
http://www.youtube.com/watch?v=i2zmiJ1HgWg
Manufacturer Attribution Dealers Association Attribution
Figure 1: Screenshot of the Two Price Advertisement Stimuli
11
a price advertisement. All subjects saw the same brand ad but were then randomly
assigned with equal probability to see either the manufacturer price ad or the dealers’
price ad. 94% of all subjects stayed on the page for at least 90 seconds or clicked at least
twice (the minimum number of clicks needed to view both advertisements), verifying that
they were exposed to the stimuli.
After viewing both ads, subjects clicked to a second page, where they were asked
to indicate their agreement (on a 1-10 scale) with five statements:
“Ford trucks have high quality.”
“Ford trucks are a good value.”
“Ford trucks are tough.”
“I would test drive a Ford truck.”
“I would consider buying a Ford truck.”
The first two statements are intended to measure product quality perceptions directly. The
third statement measures subjects’ agreement with the branding claim. The last two
indicate the likely behavioral impact of product quality perceptions.
Subjects were also asked who they thought had paid for the brand advertisement
and, separately, who they thought had paid for the price advertisement, using the same
multiple choice question as the pretest. Finally, each subject indicated all truck brands he
or she had owned.
Hypotheses
Based on the discussion in the introduction, manufacturer price advertising can be
expected to lower product quality perceptions more than dealer price advertising, for two
12
reasons. First, the manufacturer is more immediately responsible for the product’s quality.
Second, the retailer’s price cuts may be more easily attributable to non-quality-related
factors such as excess inventory or local competition. Thus,
H1. Subjects exposed to dealer price advertising should perceive trucks to have higher
quality than subjects exposed to the manufacturer price advertising.
The discussion in the introduction is based on the idea that consumers make inferences
about product quality based on the source of the price advertisement. If this is true, it
should be that consumers’ source perceptions are consistent with their quality inferences.
That is, if consumers think the manufacturer is the source of the price advertisement they
were exposed to, it should have a larger attendant harm to perceived brand quality than if
they think the dealers association was the source of the price advertisement. This leads to
a second hypothesis.
H2. Subjects who perceive the dealer as the source of a price ad should perceive trucks
to have higher quality than subjects who perceive the manufacturer as the source of the
price ad.
H1 tests the effect of exposure on perceived quality and is identified by the random
assignment within the experimental design. H2 tests the effect of source perception on
perceived quality and is identified by consumers’ self-reported source attributions.
13
Experimental Findings
Figure 2 presents consumers’ quality ratings on all five measures for both advertisement
conditions. Subjects in the dealer price advertising condition gave higher ratings than
those who saw manufacturer price advertising. Specifically, the perception of Ford
quality, value and toughness, and the willingness to test drive and to buy a Ford truck are
all significantly different between the two groups at the 95% confidence level. Taken
together these results strongly support H1 on all five dimensions of perceived quality.
The point estimate of the effect size is considerable, about 0.75 on a 10-point scale.
To test H2, we group subjects based on who they thought had paid for the price
advertisement: Ford Motor Company or a Ford Dealer Association. Figure 3 shows that
subjects who said that the price advertisement came from the dealer association gave
higher quality ratings than subjects who attributed the manufacturer as the source of the
price advertisement. This difference is significant at the 95% confidence level for quality,
value, toughness and willingness to test drive, and significant at the 90% confidence level
for willingness to consider purchase. Thus the results provide strong evidence supporting
H2 on four dimensions of perceived quality and weak evidence on the fifth dimension of
perceived quality.
14
Figure 2: Average Perceived Quality Ratings by Treatment
Figure 3: Average Perceived Quality Ratings by Stated Attribution
6.73
6.56
6.67
6.99
6.39
7.50
7.28
7.46
7.69
7.20
0.00
1.00
2.00
3.00
4.00
5.00
6.00
7.00
8.00
9.00
Ford trucks have
high quality.
Ford trucks are a
good value.
Ford trucks are
tough.
I would test drive
a Ford truck.
I would consider
buying a Ford
truck.
Manufacturer Treatment Dealers Treatment
(p < .05)
(p < .05)
(p < .05) (p < .05) (p < .05)
6.77
6.59
6.73
6.99
6.42
7.42
7.25
7.39
7.70
7.15
0.00
1.00
2.00
3.00
4.00
5.00
6.00
7.00
8.00
9.00
Ford trucks have
high quality.
Ford trucks are a
good value.
Ford trucks are
tough.
I would test drive
a Ford truck.
I would consider
buying a Ford
truck.
Manufacturer Attribution Dealers Attribution
(p < .05)
(p < .10)
(p < .05) (p < .05) (p < .05)
15
One would also expect the intersection of price advertising source and reported
attribution to significantly impact perceived product quality. This was true. Subjects
exposed to the dealer source ad who stated that they attributed it to the dealer indicated a
higher perceived quality than subjects exposed to the manufacturer source ad and stated
that they attributed it to the manufacturer. The differences were significant at the 95%
confidence level on all five quality measures.
Robustness Check on H2: Prior Ford Ownership
One might be concerned that prior Ford ownership may influence participants’
attributions and lead to spurious confirmation of H2. That is, subjects who owned Ford
trucks may perceive them to have higher quality, and therefore be more likely to attribute
price advertisements for Ford trucks to the dealers association rather than to the
manufacturer. If true, this could lead to a spurious positive correlation between stated
dealer attribution and perceived quality. Note that this alternate explanation would not
affect the evidence for H1, since identification of the exposure effect depends solely on
random assignment.
63% of the subjects indicated having owned a Ford truck. They rated Ford trucks
significantly higher on all five dimensions of perceived quality than those who had never
owned a Ford truck. However, their stated attributions were not statistically different
from those who had never owned a Ford truck. For example, 77% of Ford owners and 77%
of non-Ford owners receiving the dealer source treatment attributed it to the dealers
association.
16
The experimental manipulation produces the expected results on H1 even when
we restrict the sample to Ford owners. That is, Ford owners exposed to the dealer price
advertisement rate the truck higher on all five dimensions of perceived quality than Ford
owners exposed to the manufacturer price advertising. Differences in the first two
measures are significant at the 90% confidence level, and the last three are significant at
the 95% confidence level. It appears that prior Ford ownership did not contaminate the
stated attributions.
Summary
The experimental findings show that consumers are generally able to correctly identify
the source of a price advertisement, and that both exposure and attribution perceptions
lower perceived quality when the manufacturer is the source of the price advertising
relative to when the dealers association is the source of the price advertising. Next, we
investigate whether these effects are large enough to show up in market data.
Chapter One: Econometric Evidence
This section explores whether the theoretical predictions confirmed by the experiment are
large enough to be detected in market data. In so doing, it contributes to a large literature
in marketing on estimating consumer demand for automobiles.
2
To the best of our
2
Most recently, Sudhir (2001) examined competitive pricing behavior in different segments of the U.S.
auto market and found that automakers price aggressively in the entry-level segment but are more
cooperative in the larger car segment. Busse et al. (2006) found that “customer cash” results in larger pass-
through than “dealer cash” since customer cash is directly revealed to the consumer. Dasgupta et al. (2007)
showed that consumers are more likely to lease vehicles with high expected maintenance costs and to buy
cars that are seen as more reliable. Bucklin et al. (2008) developed a method to evaluate the impact of
dealer distribution intensity on consumers’ new car choices. Busse et al. (2010) found that employee
discount pricing promotions offered by automotive manufacturers led to simultaneous increases in prices
and sales by cannibalizing demand from future periods. Albuquerque and Bronnenberg (2010) proposed a
framework to estimate dealer costs and predicted the impact of a recession on dealer exit and post-exit
automobile prices.
17
knowledge, only three studies of automotive demand include advertising data. Kwoka
(1993) estimated returns to product redesigns and advertising, finding that redesigns’
effects on sales last longer than advertising’s effects. Barroso (2007) found that the
omission of advertising data biases automotive demand estimates. Srinivasan et al. (2009)
found that auto quality and advertising interact positively to increase automakers’ stock
prices, while price promotions reduce stock prices, showing that investors’ profit
expectations respond to automakers’ marketing actions.
Data
The econometric analysis combines two datasets. The first is transaction data on the sales
of full-size pick-up trucks in the Los Angeles metropolitan area collected by the Power
Information Network (PIN), a division of J.D. Power and Associates. Each observation
includes the transaction date, type (lease, finance, or cash), pricing terms,
3
vehicle
characteristics (make, model, model-year, options), an anonymous dealership number,
and customer gender and zip code.
When modeling advertising effectiveness, it is essential to ensure the data include
nonpurchases as well as purchases. Otherwise, one cannot fully identify advertising’s lift
over baseline sales. A number of recent studies (Bucklin et al. 2008, Chen et al. 2008)
have estimated auto demand using transaction level data, but a transaction-level approach
is not suitable to the purposes of this chapter. Since every transaction, by definition, is an
observation of positive demand, such an approach would not include nonpurchase data.
Instead, we estimate demand by modeling zip code purchases with nonpurchases
3
Appendix A shows how market prices are constructed from the data.
18
included in the outside option, as in Albuquerque and Bronnenberg (2010), Berry et al.
(1995, 2004), and Sudhir (2001). However, we use weekly data (rather than quarterly or
annual data) to increase the amount we can learn from intertemporal advertising variation.
The sample includes six trucks—Chevrolet Avalanche, Ford F-Series, Dodge
Ram, Chevrolet Silverado, Toyota Tundra, and GMC Sierra—accounting for 87% of
category sales from July 2
nd
, 2001 to December 31
st
, 2005.
The second data source is Kantar Media’s “Stradegy” database. It contains
estimated advertising expenditures of all truck manufacturers, dealers associations, and
individual dealerships. Advertising spending in all television and newspaper media are
included, accounting for 81% of all category advertising expenditures. 26 weeks of pre-
sample advertising data are used to construct the initial conditions for the advertising
goodwill stocks defined below.
Video files for 130 pick-up truck ads were obtained, approximately 17% of the ad
creatives aired during the sample period. A content analysis of these advertisements was
performed to identify the types of message conveyed by manufacturers and dealers
associations. Two independent coders rated each advertisement on a 0-100 scale, where 0
indicated all brand messages and no price messages, and 100 indicated no brand
messages and all price messages. In agreement with Kaul and Wittink (1995),
manufacturer advertising on national networks was found to focus primarily on the brand,
with an average score of 19. Local manufacturer advertising and dealers associations’ ads
19
are much more oriented toward price, with average scores of 54 and 58, respectively.
4
We therefore define the following three variables for consistency with the discussion in
the introduction and the previous literature. Manufacturer Brand Advertising (MBA) is
paid for by the manufacturer, carries primarily truck-specific branding messages, and is
conveyed by national television networks and national newspapers in many metropolitan
areas simultaneously.
5
Manufacturer Price Advertising (MPA) is paid for by the
manufacturer, carries primarily truck-specific price messages, and is conveyed by local
television stations and newspapers. Dealers’ Price Advertising (DPA) is paid for by local
dealers associations, carries primarily truck-specific messages about pricing terms and
holiday sales events, and is conveyed by local television stations and newspapers.
Model
A mixed logit model is used to estimate consumer demand for trucks. Each consumer
chooses among j=1,…,6 pick-up trucks and an outside option (j=0). Let s represent the
calendar month in which week t ends. Consumer i in zip code z gets utility
izjt
u by
choosing truck j in week t:
izjt zjt z zjt jt jt jt zjt z t js izjt
YC DPA MPA MBA p f X u ) ; , , , , (
, (1)
4
The coders’ percentage agreement was 83%. They were not allowed to resolve discrepancies through
discussion.
5
To equate manufacturer brand advertising expenditures to the other types of advertising, we deflate it by
the Los Angeles market’s share of the total US population. All advertising expenditures are expressed in
July 2001 dollars.
20
where
js
is a truck-month intercept that captures the mean level of all consumers’
valuations for truck j in month s,
6
t
X includes gas price and holiday week dummies, f
and
z
govern consumers’ response to marketing variables such as price
zjt
p and
advertising as described below, and
zjt
captures zip code z’s weekly deviation from
mean truck-month preferences.
izjt
is a “logit error” which is assumed to be i.i.d.
Gumbel with scale parameter one. The mean utility of the outside option is normalized to
zero, so predicted market shares within a zip code take the familiar mixed logit functional
form.
7
Many studies have shown that advertising effects may persist over time.
Advertising carryover is represented by an exponentially smoothed advertising carryover
variable (e.g. Jedidi et al. 1999, Erdem et al. 2008) for each advertising variable
} , , { DPA MPA MBA A . Thus,
1 1
) 1 (
jt A
A
jt A jt
A a A (2)
where
A
jt
a
1
is the log of one plus the expenditure of brand j in week 1 t on advertising
of type A, and ) 1 , 0 (
A
is a decay parameter.
6
Truck-month dummies are used instead of truck-week dummies because truck-week dummies would not
be separately identified from truck-specific advertising response parameters.
7
The market size is defined as the maximum number of unit sales across weeks within each zip code.
21
Consumer response to marketing instruments is modeled as:
.
zjt jt
DPp
zjt jt
MPp
zjt jt
MBp
jt jt
PP
j jt jt
BDP
j jt jt
BMP
j
zjt
Y
z jt
DP
zj jt
MP
zj jt
MB
zj zjt
p
z
p DPA p MPA p MBA
DPA MPA DPA MBA MPA MBA
YC DPA MPA MBA p f
(3)
Equation (3) allows consumer demand to be a function of direct effects and first-order
interactions among advertising variables and price.
8
The random coefficient structure
allows consumer preferences to vary over zip codes:
z z
~
~ ~
, (4)
where ) , , , , (
~
Y
z
DP
zj
MP
zj
MB
zj
p
z z
, and
z
~
and
z
~
are conformable vectors of the mean
and random effects, respectively. Holiday week and gas price response parameters are
similarly defined as
z z
, (5)
where
z
is a conformable vector of random effects and gamma is a vector of the
corresponding mean values.
9
Nonparametric Identification, Endogeneity and Estimation
Advertising and prices may be jointly determined by market forces and strategic
considerations, and may depend on unobserved factors. It is therefore important to
control for potential endogeneity when estimating their effects on market demand. Truck-
8
Following previous literature, the model includes a measure of each zip code’s recent truck sales to
control for persistence in unobserved demand shocks (Dasgupta et al. 2007).We do this by constructing
past sales carryover
1 1
) 1 (
zjt y zjt y zjt
YC y YC , where
1 zjt
y is the sales of truck j in zip code z in
week t-1 and
y
is a carryover parameter to be estimated.
9
A number of alternatives to equations (2)-(5) were considered. The one presented here minimized the AIC
and BIC. The interested reader is referred to Appendix B for further details.
22
month dummies are included in the model to completely remove any endogeneity due to
monthly fluctuations in market demand for trucks. The control function approach of
Villas-Boas and Winer (1999) is used to control for possible residual correlations
between marketing mix variables and weekly demand shocks. The interested reader is
referred to Appendix C for a detailed discussion of nonparametric identification,
endogeneity, and slope endogeneity.
The model is estimated via simulated maximum likelihood (Train 2003). The
weekly demand shocks
zjt
are distributed normal with common variance to be estimated,
and each element of the unobserved heterogeneity vector is distributed normal with an
element-specific variance to be estimated.
Advertising Source and Advertising Content Results
If manufacturer brand advertising communicates brand quality, and if manufacturer price
advertising diminishes consumers’ quality perceptions, then manufacturer price
advertising should reduce the effectiveness of manufacturer brand advertising. This
would imply that the cross-partial derivative of market share with respect to these two
variables is negative.
10
And, if dealers’ price advertising increases consumers’ quality
perceptions, then dealer price advertising should facilitate the effectiveness of
manufacturer brand advertising. This would imply that the cross-partial derivative of
market shares with respect to these two variables is positive.
10
Ai and Norton (2003) show that correct inference of interaction terms in choice models requires analysis
of cross-partial derivatives rather than linear parameter estimates. Appendix D shows the cross-partial
derivatives under our application.
23
Table 2 reports the cross-partial derivatives of market shares with respect to each
pair of the three advertising variables. There are three interesting results to note. First, all
interactions between manufacturer price advertising and manufacturer brand advertising
have negative signs, and three are statistically significant. This is consistent with the idea
that manufacturer price advertising may diminish the credibility of the manufacturer’s
brand advertising.
Table 2: Advertising Interaction Estimates
Second, all interactions between dealers’ price advertising and manufacturer
brand advertising have positive signs, and five are statistically significant. This suggests
that manufacturer brand advertising may work synergistically with dealer price
advertising.
While these interactions have the expected sign, it must be noted that one
MBA/DPA interaction and three MBA/MPA interactions are not significant. Two of
Truck
−8.0E-7 3.9E-6 1.8E-5
(2.6E-6) (2.1E-6) (1.8E-5)
−0.002 ** 0.002 ** −3.3E-4
(4.7E-4) (3.7E-4) (2.9E-4)
−2.9E-7 5.5E-6 ** −1.2E-6 **
(2.5E-6) (7.8E-7) (3.5E-7)
−0.009 ** 0.002 ** −0.001 *
(0.002) (0.001) (0.001)
−3.4E-6 * 5.0E-6 ** −2.5E-6
(1.7E-6) (7.7E-7) (1.6E-6)
−2.7E-4 4.4E-4 ** −1.0E-4
(1.8E-4) (9.0E-5) (8.6E-5)
MBA× MPA MBA× DPA
Avalanche
MPA× DPA
* significant at 95% confidence level; ** significant at
99% confidence level.
F-series
Ram
Silverado
Tundra
Sierra
24
these insignificant interactions were for Chevrolet Avalanche, which had the smallest
number of weeks with positive advertising expenditures in the data. The other
insignificant interactions might have been influenced by collinearity between advertising
variables and their interactions, which would inflate the standard errors and bias the
results against finding statistical significance.
Finally, two interactions between MPA and DPA are negative and significant, and
none were positive and significant. The prior literature does not make a clear prediction
about the sign of this effect, but the result could perhaps be interpreted as some evidence
of diminishing returns to price advertising.
Price Interactions, Advertising Main Effects, Gas Price and Holiday Dummies
Table 3 reports the estimated interaction effects between price and the three types of
advertising. Only the interaction between price and manufacturer brand advertising is
found to be statistically significant (and negative). This is consistent with the
“informative” view of advertising (e.g., Stigler 1961; Grossman and Shapiro 1984). This
literature argues that advertising increases consumers’ awareness of brand attributes,
enlarges consumers’ consideration sets, increases market competition, and therefore
makes demand more elastic. Gatignon (1984) found that brand advertising may increase
consideration set sizes, leading to increased demand responsiveness to price. Kanetkar et
al. (1992) showed that brand advertising may increase demand responsiveness to price
when it reinforces similarities with competitors’ brands. Mitra and Lynch (1995)
suggested that brand advertising may increase the number of alternatives considered by
providing recall cues, and therefore increase demand responsiveness to price. Erdem et al.
25
(2008) found the same effect because advertising brings in the marginal consumer (who
is, by definition, the most price sensitive).
Table 3: Price-Advertising Interactions
Table 4 presents the estimated main effects of advertising on consumer demand
for trucks. Manufacturer brand and price advertising are positively associated with sales
(independent of their negative interaction), but dealers’ price advertising is not.
Table 4: Advertising Main Effects
Interactions
Price× MBA −2.2E-4 **
(8.1E-5)
Price× MPA 6.3E-5
(8.5E-5)
Price× DPA −4.5E-5
(3.2E-5)
Estimate
(Std. Err.)
** significant at 99%
confidence level.
Truck
1.063 ** 1.479 ** −5.837
(0.292) (0.550) (3.888)
3.109 ** 6.354 ** −0.753
(0.629) (1.313) (0.503)
0.761 2.930 * 0.041
(0.645) (1.191) (0.679)
5.673 ** 7.088 ** 0.748
(0.799) (2.032) (0.943)
1.191 6.049 ** 0.232
(1.019) (2.161) (0.526)
1.317 ** 3.688 −1.199
(0.417) (1.996) (0.920)
DPA
Avalanche
MBA MPA
* significant at 95% confidence level; ** significant at
99% confidence level.
Silverado
Tundra
Sierra
F-series
Ram
26
Table 5 shows that other demand parameter estimates conform to intuition. Gas
prices reduce truck demand, with an effect that is significant at a very high confidence
level. Holiday departures from mean monthly demand for trucks are highest on Labor
Day and Memorial Day and lowest on Columbus Day.
Table 5: Gas Price and Holiday Effects
Price elasticities are in line with other studies of this industry at about −2.
However, the advertising interaction terms make advertising elasticities difficult to
interpret. Instead, counterfactuals reported in next section provide a better measure of the
empirical importance of these advertising interactions.
Summary
Manufacturer price advertising interacts negatively with manufacturer brand advertising,
while dealer price advertising interacts positively with manufacturer brand advertising.
This pattern is consistent with the experimental evidence.
Variable Variable
Gas price −0.90 ** Memorial Day Weekend 0.96 **
(0.14) (0.13)
Christmas 0.09 MLK Day Weekend −0.08
(0.07) (0.07)
Columbus Day Weekend −0.17 * New Years 0.47 **
(0.08) (0.08)
July 4 Weekend
0.19 *
Thanksgiving Weekend
−0.04
(0.08) (0.08)
Labor Day Weekend 1.15 ** Presidents Day Weekend 0.14
(0.11) (0.08)
* significant at 95% confidence level; ** significant at 99% confidence level
Estimate
(Std. Err.)
Estimate
(Std. Err.)
27
Since the econometric results are not based on a field experiment, they may be
attributable to factors other than the price-signaling explanation. A range of alternate
explanations were explored but little or no evidence was found to support them. The
interested reader is referred to Appendix E for details.
Chapter One: Practical Importance
It has so far been confirmed that manufacturer price advertising lowers perceived product
quality relative to retailer price advertising, and that this pattern of results may be found
in market data.
A straightforward reaction to these results would be to say that manufacturers
should leave the price advertising to their dealers. This intuition would only be correct if
the two parties’ interests were fully aligned. However, it is well known that misalignment
between manufacturers’ and dealers’ incentives may lead to channel conflict. In fact,
manufacturer price advertising may be a tool that the manufacturer uses to manage
channel conflict. By providing price incentives directly to the consumer, and announcing
them via price advertising, the manufacturer may be able to reduce the double-
marginalization that occurs when it gives discounts to the dealer (Busse et al. 2006)
Yet there is still an inefficiency in the sense that price advertising is more
effective when it comes from the dealer rather than the manufacturer. Ideally, the channel
members could contract to eliminate this inefficiency. The marketing literature has made
substantial progress toward characterizing these contracts; see Iyer (1998), Kolay (2010),
and He et al. (2011).
28
What is done here is to indicate the possible gains to the channel from
coordinating its members’ price advertising. That is, if the manufacturer and dealers
association were fully integrated, how much would their joint profit rise if they
coordinated their price advertising? This provides a rough estimate of the welfare gains
available from solving the channel inefficiency in price advertising. These results can
indicate the possible gains to the channel of taking the managerial action of coordinating
channel members’ price advertising.
It should be noted that we do not seek to show what managerial actions could
eliminate the price advertising inefficiency, as previous literature shows that optimal
channel contracts may depend on such factors as competition among retailers and
knowledge of local market conditions. Instead, we limit ourselves to indicating the
possible gains to resolving the inefficiency.
The question of interest is how much aggregate channel profits would increase
under channel advertising coordination. Since the empirical model did not assume
advertising optimality, this question is addressed by comparing two sets of
counterfactuals. The baseline counterfactual optimizes each channel member’s
advertising profits individually in a game-theoretic channel interaction for a single brand.
The treatment counterfactual optimizes joint channel profits under advertising
coordination.
Control Counterfactual: Advertising Optimization without Channel Coordination
The following procedure establishes the baseline of optimization without channel
coordination for each truck. A sequential game is assumed, in which the manufacturer
29
moves first without anticipating its dealers association’s action and the dealers
association moves second with knowledge of the manufacturer’s choices. This
assumption is made because dealers associations’ relationships with local media outlets
likely afford them greater flexibility in altering ad buys.
Specifically, manufacturer j chooses its price advertising to solve
t
MPA
j jt
t
jt jt
z
z zjt jt jt
MPA
A MPA t s
MPA MBA M s c w
jt
. .
) ( ˆ ) ( max
, (6)
where
MPA
j
A is the manufacturer’s observed price advertising budget.
jt
w is the wholesale
price and
jt
c is the marginal cost for truck j in week t. Automakers guard actual margin
data closely, so we assume t j c w
jt jt
, , and discuss below.
The dealers association then chooses its advertising expenditures to solve
t
DPA
j jt
t z
jt z zjt jt zjt
DPA
A DPA t s DPA M s w p
jt
. . ˆ ) ( max , (7)
where
DPA
j
A is the dealers association’s observed price advertising budget. Similar to the
calculation of manufacturers’ profits, dealers’ profits are calculated based on an average
profit assumption for dealers in the truck category.
Treatment Counterfactual: Optimization and Channel Coordination
The treatment counterfactual optimizes total channel profits by coordinating both channel
members’ price advertising. For each truck j, MPA and DPA are chosen to solve
30
, ) ( . .
) ( ˆ ) ( max
} , {
j
t
jt jt
t z
jt jt jt z zjt jt zjt
DPA MPA
A DPA MPA t s
DPA MPA MBA M s c p
jt jt
(8)
where
j
A is the observed total price advertising spending for truck j across weeks.
Caveats
Careful interpretation of the results requires explicit awareness of all assumptions.
Conservative assumptions are made to avoid inflating the profit gains from channel
advertising coordination.
Advertising message content does not change under channel coordination,
limiting channel members’ degrees of freedom in their choice of advertising
strategies.
The baseline counterfactual assumes a particular sequential channel interaction.
We have tested some alternate games (Vertical Stackelberg, simultaneous-move)
and found that while they affect the division of rents, the sum of channel profits is
relatively insensitive to assumptions about the game between channel partners.
Manufacturer brand advertising is assumed to remain unchanged since it must be
chosen to influence sales in many markets simultaneously. A channel advertising
coordination policy could potentially be adopted on a nationwide basis and
improve manufacturer brand advertising effectiveness as well as its price
advertising effectiveness.
Due to lack of data, the counterfactual ignores aftermarket dealer services, dealer
competition, and effects of pick-up truck ads on sales in other auto categories.
31
Each truck brand is assumed to operate independently. Allowing multibrand
manufacturers to realize cross-brand advertising synergies could yield larger gains.
There are no competitive reactions in advertising across manufacturers. If all
manufacturers coordinate channel advertising expenditures, each one’s predicted
profit gain would be attenuated by enhanced competition.
Results
Counterfactuals are computed for the three trucks (Ford F-Series, Chevrolet Silverado,
GMC Sierra) that spent the most on dealer price advertising during the sample.
11
Discussions with auto industry managers suggested that dealers’ profit margin per
truck sold is about 6%, while manufacturer margins range from 15% to 25%. The
counterfactual results, in Table 6, indicate that channel advertising coordination can
increase total channel profits by 1-7% relative to optimization alone, with the biggest
gains available to the market leader, Ford. These counterfactuals are also evaluated with
15% and 25% manufacturer margin assumptions, finding that total channel profits
increase by 2.8% - 3.5% on average. These gains came from reallocating advertising
dollars from manufacturer price advertising to dealers price advertising, leading
manufacturer brand advertising to be more effective.
11
26 weeks of data are used to limit the size of the optimization. Decision variables in (6)-(8) are
constrained to be nonnegative. Advertising expenditures are constrained at the observed advertising
budgets, but these constraints do not bind. Constrained profit functions were maximized directly due to
frequent corner solutions. Observed advertising levels were used as starting values.
32
Table 6: Aggregate Channel Profit Gains with Channel Advertising Coordination
Chapter One: Discussion
This was the first study to consider how the effects of a price advertisement depend on
whether the manufacturer or the dealer is the perceived source of the advertisement. An
experiment established that manufacturer price advertising lowers perceived quality
relative to dealer price advertising, as suggested by the extant literature. A similar pattern
of effects was found in an econometric model, establishing external validity.
Counterfactual analysis based on the econometric model suggests that channel
advertising coordination may increase total channel profits by about 3% in the pick-up
truck category.
These results show why a manufacturer might choose to start a cooperative
advertising program rather than engage in price advertising on its own. Further, they may
suggest that manufacturers would profit by framing price messages in ways that are
unlikely to send quality signals. For example, offering a free service (e.g. “free
maintenance for 3 years”) may be more profitable than a price cut of the same size. Or,
15%
Manufacturer
Margin
20%
Manufacturer
Margin
25%
Manufacturer
Margin
F-Series 5.4% 6.3% 7.0%
Silverado 1.3% 1.6% 1.9%
Sierra 1.9% 1.6% 1.4%
Average 2.8% 3.2% 3.5%
Note: in all cases, dealers' profit margin is assumed to be
6%.
33
manufacturers may wish to de-emphasize the source of their price ads to reduce harmful
inferences about product quality.
The results also have implications for the timing of messages of different types.
Perhaps manufacturers and dealers associations’ media buyers should coordinate to pulse
brand advertising and price advertising simultaneously. Further, package goods
manufacturers’ cooperative advertising programs could include time-dependent
reimbursement rates to encourage retailers to take advantage of advertising synergies.
This study could be extended in several directions. The effects measured here
come from a market setting in which manufacturers used exclusive dealers. Consumers’
quality inferences in other channel structures may depend on such factors as brand
strength, store traffic patterns, retailers’ bargaining power, or category management
strategies. Future research might also explore what communication techniques maximize
synergies between branding messages and price advertising and how price offer framing
changes quality inferences.
34
CHAPTER TWO: PREDICTING INCUMBENTS’ PRODUCT QUALITY
REACTIONS TO ENTRY
Chapter Two: Introduction
Launching a new product is the most costly marketing action a firm can undertake. Prior
to launch, managers have to forecast the potential demand and profitability of the new
product. But in markets with frequent innovations, forecasting initial sales is not enough.
If the new product is successful, competitors will react to its entry by changing their
product development strategy. They may accelerate their product quality improvements,
or they may abandon further product quality improvements. Each competitor’s choice of
strategy depends on the intensity of competition, its quality standing within the
marketplace, the demand for quality and the costs of research and development. And the
competitors’ strategies, in turn, will affect the long-run profitability of the new product.
Consider Nissan which introduced a new full-size pickup truck, the Titan, in 2004.
The Titan included a standard 300-horsepower V-8 engine, an automatic transmission
and 9,400 pounds maximum towing capacity. These were significant advantages over the
market-leading Ford F-150’s 2003 components: a 200-horsepower V-6 engine, manual
transmission and 8,800 pounds maximum towing capacity. However, Ford reduced
Titan’s advantage by upgrading its 2004 F-150 to the same engine size and transmission
type, and maximum towing capacity of 9,500 pounds.
Bringing the Titan to market was very expensive. Nissan spent $1.4 billion on the
production factory alone; however, Titan captured just 3% of the market in its first year
and sold about 84,000 units, falling short of its expected goal of 100,000 units
35
(Bloomberg Businessweek, 2006). If Titan’s market potential forecast neglected Ford’s
response in product quality, it would have seriously overestimated Titan’s post-launch
sales.
The purpose of this chapter is to investigate whether a dynamic game-theoretic
framework combined with empirical analysis can successfully predict incumbents’
competitive reactions in product quality to a new product entry. Ericson and Pakes (1995;
“EP” hereafter) developed an empirical framework to study firms’ dynamic entry, exit
and investment decisions (see Doraszelski and Pakes 2007 for a review). In this
framework, firms maximize their net present value of profits given expectations about
competitors’ equilibrium strategies. This framework has proven its value repeatedly by
enabling researchers to understand the impacts of competition and governmental policies
in a variety of settings such as hospital quality competition (Gowrisankaran and Town
1997), advertising expenditures (Dubé et al. 2005), generic drugs’ entry timing (Ching
2010), research joint ventures in product quality (Song 2010), impact of supercenter
elimination (Beresteanu et al. 2010) and effects of product innovation on market structure
(Goettler and Gordon 2011).
However, whether this dynamic structural model framework can be used to
predict competitive reactions in product quality is an empirical question that, to our
knowledge, has not been investigated. On one hand, it is documented that dynamic
structural models could be used to predict the effects of a policy change (Chintagunta et
al. 2006)by adjusting the effects of the entry based on a theoretically founded model of
behavior. On the other hand, the added complexity and structural assumptions of dynamic
36
structural models may limit its ability to provide good predictions. Therefore, it is an
empirical question whether a dynamic structural model could predict better than a
descriptive model.
We adapt the Markov Perfect Equilibrium (MPE) framework proposed in EP.
Firms simultaneously make investment decisions after observing quality levels of all
products available in the market, and consumers choose among the differentiated
products. Product quality is a subjective measure of consumer perceptions of both
objective and subjective product characteristics; it might be influenced by advertising
expenditure in addition to objective improvements in product attributes. More investment
today would likely result in higher product quality tomorrow, but the outcome of
investment is stochastic and is subject to common industry-wide shocks.
To estimate the model, one needs to solve the MPE for each firm for every set of
product quality levels. The complexity of the EP framework makes the model hard to
estimate and essentially limits researchers’ ability to apply it to industries with many
firms competing with each other. The traditional estimation method, the nested fixed
point (NFXP) approach of Rust (1987), requires solving the equilibrium many times
when searching for model parameter values. The major challenge of this approach is the
computational burden that it imposes when the state space gets large and when the
number of players in the game increases.
Recently developed two-step estimation techniques make the estimation feasible
in many dynamic oligopoly applications (Aguirregabiria and Mira 2007, Bajari et al.
37
2007, Pakes et al. 2007).
12
The essence of the two-step approach is to avoid the
computation of the fixed point by flexibly estimating state transition probabilities and
policy functions in the first step, assuming the observed data are generated by equilibrium
play. However, one general concern about this approach is that the first step estimates
may not be compatible with the equilibrium implied by the underlying model, resulting in
serious finite sample biases of the second-stage estimates of the structural parameters
(Agguiregabiria and Mira 2007).
This chapter employs the Mathematical Programming with Equilibrium
Constraints (“MPEC,” Su and Judd 2010), a mathematical programming method for
structural estimation. The idea behind MPEC approach is to choose the parameter values
and endogenous variables (value functions and policy functions in this case) that
maximize the likelihood of the data, subject to constraints under which the endogenous
variables satisfy the equilibrium conditions. It is a constrained optimization approach that
avoids solving the equilibrium for each guess of the parameter vector. Instead, the
equilibrium is solved exactly only once at the final iteration of the parameter value search
process, which therefore alleviates the computational burden of the NFXP approach. Su
and Judd (2010) used Monte Carlo experiments to show that the MPEC estimator has
superior finite sample properties than the two-step estimators.
The context we consider in this chapter is the incumbents’ competitive reactions
in product quality to the introductions of Toyota Tundra and Nissan Titan in the US full-
12
Weintraub et al. (2008) has introduced a new equilibrium concept, Oblivious Equilibrium, based on
which the dynamic games can be solved as single-agent problems and therefore it can resolve the
computation concern. However, this approach requires that no single firm has influence over the overall
market’s evolution.
38
size pickup truck market. This market was stable and dominated by the “Big Three” U.S.
auto manufacturers until the Toyota Tundra and the Nissan Titan entered in 2000 and
2004, respectively. We first estimate the proposed model using data prior to Tundra’s
entry, then use the parameter estimates as primitives of the dynamic game to compute
firms’ optimal investment and expected payoffs after a counterfactual entry, and finally
make predictions on incumbents’ product quality levels in time periods following the
entry.
The analysis reveals three insights about how competition in product quality
influences firms’ equilibrium investments. First, the quality differential has a bigger
impact on the investment level of low quality firms than on the investment level of high
quality firms. Moreover, second, as the quality differential gets larger, the quality laggard
has a lower incentive to invest than does the quality leader. Third, the presence of a low
quality competitor in a market has a greater impact on a focal firm’s optimal investment
decision than does the presence of a high quality competitor. The predictions about
incumbents’ product quality after the entry of Tundra and after the entry of Titan are
compared to a descriptive time-series model. The results indicate that the structural
model’s hit-rate doubles that of the benchmark model, and its mean squared prediction
error is 22% lower.
Chapter Two: Model
This section specifies a dynamic stochastic oligopoly game in product quality investment.
Time is discrete and the horizon is infinite, ,..., 2 , 1 t . Each firm j ) ,..., 1 ( J j is
39
described by its quality level, or state, } ,..., 2 , 1 {
jt
.
13
At any point of time, the
industry is completely characterized by the list of states for all firms (so-called the
“industry state”), } ,..., {
1 Jt t t
. At the beginning of each time period, firms
simultaneously make investment ) 0 (
jt
x decisions to improve their product quality.
Product competition in prices then takes place in the market once the investment decision
is made. We follow the literature and model price competition in a static fashion since
product prices have no effect on the evolution of the firms’ states. Firm j chooses its price
jt
p to maximize the single-period profit
jt
, before subtracting the investment cost. The
outcome of a firm’s investment process depends on its own investment level
jt
x and an
industry-wide shock that represents movements in the demand or cost conditions, and
is unknown to firms when making the investment decisions. At the end of each time
period, the industry state is updated.
14
Firms’ Dynamic Decisions
A firm is able to change its state over time through its investment
jt
x . At the beginning of
each time period, firms observe the industry state
t
and simultaneously choose
jt
x to
maximize the expected discounted value of net cash flows:
t
t
j j
t
x
x E
j
| ) ( max , (9)
13
It is assumed that each firm only sells one product. One firm selling multiple products could be
potentially allowed, but the state space will grow significantly, which leads to substantive increase in
computational burden.
14
Unlike many studies that follow the EP framework, we choose not to model firms’ entry decision, mainly
because the sample period only has two entry occurrences.
40
where is the discount factor and ) (
j
is firm j’s single-period profit before
deducting the investment cost at time .
Investments are risky and their outcomes are stochastic. The state transition
process depends on the current state
jt
, investment level
jt
x , and industry-wide shock
t
. Let
jt
v represents the outcome of firm j’s investment, and its probability distribution,
) (
jt
v p , is assumed to be:
otherwise 0
) 1 ( y probabilit with 1
jt jt
jt
ρx ρx
v , (10)
where 0
represents the effectiveness of investment.
15
Following Pakes and McGuire (1994), we assume that the state evolves as follows:
j j j
v
'
, (11)
where
'
j
is the realized state at the end of each time period. The state cannot improve
without investment, but may decay with positive realizations of the shock . Factors that
could cause such positive realization include development of substitute products or
macroeconomic variables that lead consumers to favor substitutable products. Having this
common shock can allow firms’ profits to be positively correlated, which is often
observed within an industry. For simplicity, we assume 1 exogenously with a fixed
probability and 0 with probability 1 .
15
This simple probability distribution is chosen because it leads to analytical solutions of firms’ investment
in equilibrium, which could reduce the computational burden.
41
A strategy or policy function for firm j specifies an action ) (
j
x for each
industry state . We restrict our attention to stationary Markovian strategies, which are
only a function of the current state and do not depend on calendar time, so the time
subscript is omitted in equation (11) and all equations hereafter in this section. Denote the
net present value of profits for firm j as ) , (
j j
V
. It could be defined recursively by
the solution to the following Bellman equation:
] | ) , ( [ ) , ( max ) , (
' '
j j j j j
x
j j
V E x V
j
, (12)
where
j
is the vector of product quality of all firms excluding firm j. The expectation
on the right hand side of equation (12) is taken over the probability distribution of the
values of firm j’s own state and its competitors’ states in the next time period. To
compute this expectation, each firm must have a perception of the likely future states of
its competitors conditional on different possible outcomes of the shock . We can
rewrite the expectation as:
j
v
j j j j
v p v W V E ) ( ) | ( | ) , (
' '
, (13)
where
, ) ( ) , | ( ) , ( ) | (
'
' '
j
p q v V v W
j j j j j
(14)
is firm j’s expected payoff conditional on the outcome of its investment, } 1 , 0 {
j
v .
) , | (
'
j
q
represents firm j’s beliefs about its competitors’ next period state
'
j
conditional on the common shock . These beliefs, ) , | (
'
j
q
, are consistent with
42
firm j’s rational expectation of its competitor’s current investment decision, which, in
turn, are the competitor’s optimal decisions in equilibrium.
Following the literature since Ericson and Pakes (1995), we only consider pure
strategy Markov Perfect Equilibrium (MPE). A strategy profile } ,..., {
1 J
x x x is a vector
of decision rules for all firms. A strategy profile
x is MPE if, given the competitor
strategy profile
j
x , firm j prefers its strategy
j
x to all alternative strategies
'
j
x , that is,
) , ; ( ) , ; (
'
j j j j j j
x x V x x V for all j. In other words, an MPE involves value function
V and policy function
x such that (i) given
j
x ,
j
V solves the Bellman equation (12)
for all j, and (ii) given
j
x and
j
V ,
j
x solves the maximization problem on the right-
hand side of equation (12) for all and all j.
Each firm solves its own equation (12) to obtain its optimal investment level
given the current industry state. Substituting (13) and (14) into (12) and taking the first-
order condition of (12) give us:
0 1
) (
) | (
j
v j
j
j
x
v p
v W . (15)
Given the assumption of ) (
j
v p and 0
j
x , we can analytically compute the optimal
investment level in equilibrium:
16
16
Please see Appendix F for the derivation.
43
,
)] | 0 ( ) | 1 ( [ 1
, 0 max ) , (
W W
x
j j j
(16)
if ) | 0 ( ) | 1 ( W W and 0 ) , (
j j j
x otherwise.
The functional form of the state transition probabilities (Equation (10)) satisfies
the unique investment choice admissibility condition derived in Doraszelski and
Satterthwaite (2010). It guarantees the uniqueness of each firm’s optimal investment
solution. We have also numerically verified that the equilibrium computation converges
to unique solutions of value functions and policy functions from many different initial
values.
Firm’s Single-Period Profit
In this section, we present firms’ profit functions in each time period, excluding the cost
of investment in quality. A typical assumption in the literature is that prices do not
influence the evolution of the state variable, product quality in this chapter, so they could
be determined in a static competition. That is, after making the investment decision in
each time period, each firm sets its price,
jt
p to maximize its single-period profit:
jt jt jt
p
Ms mc p
jt
) ( max , (17)
where mc
is the marginal cost of production, M is the market size, and
jt
s is the market
share of product j at time t. A simple marginal cost is assumed, , ) ln( mc
where will
be estimated along with other parameters. The F.O.C. of equation (17) gives us the
implicit solution of the optimal prices at each time period:
44
j
s
mc p
jt
jt
,
) 1 (
1
. (18)
Product market share
jt
s is derived from the multinomial logit demand system.
Consumer i purchasing truck j=1,…,J in time t gets indirect utility:
ijt jt jt ijt
p u
~
, (19)
where
jt jt jt
X
~
is the perceived quality of product j at time t. We focus on
perceived quality because 1) truck characteristics such as design, brand equity, toughness
or reliability may be difficult to quantify but important predictors of truck demand; 2) it
is the consumer perception of product quality that essentially determine preference,
satisfaction, loyalty and profitability (Mitra and Golder 2006). Perceived quality is
decomposed into a set of observed product characteristics
jt
X and an unobserved quality
term
jt
. In essence, the perceived quality is the market’s mean taste for truck j that
rationalizes its market share at its price
jt
p . The downside of this measure is that it could
pick up some demand-side preference shocks that are not product quality related.
Assuming consumer i’s idiosyncratic preferences
ijt
are independently
distributed extreme value, and normalizing the utility of the outside option to
t i0
, the
market share of product j at time t is:
17
k
kt kt
jt jt
jt
p
p
s
)
~
exp( 1
)
~
exp(
. (20)
17
With a larger sample size, we could accommodate consumer heterogeneity in the demand side.
45
Chapter Two: Estimation and Simulation
The model parameters that need to be estimated are } , , , , { .
18
Aside from these
parameters, we also solve the dynamic oligopoly game to get firms’ value functions
) ( V and policy functions ) ( x . In this section, we first describe the estimation
procedure to obtain the parameter estimates, and then present simulation exercises to
show that the estimation procedure can recover known parameters.
Estimation and Identification
Since the static parameters , , and do not enter the dynamic model and the single-
period profit is treated as a primitive of the dynamic game, they are estimated using a
standard generalized method of moments (GMM) approach. We incorporate the F.O.C.
of the single-period profit with respect to price (Equation (18)) in the estimation to
control for potential price endogeneity concern. This estimation procedure is standard and
widely applied in the marketing literature.
19
The main goal of this step is to get consistent
estimates of product quality
jt
~
for each firm.
The assumption on the state transition process (Equation (11)) implies that (i) a
firm’s quality from one period to the next cannot move more than one step, and (ii) if one
firm’s quality moves upward, none of the other firms’ quality could move downward in
the same time period. These two implications are used as criteria to discretize the
18
The value of the discount factor
is chosen to be 0.925, corresponding to an 8% annual interest rate
(Ching 2010).
19
We omit the details of this estimation procedure in the interest of space. Chintagunta and Nair (2010)
provide an extended discussion.
46
perceived quality estimates
jt
~
into discrete quality
jt
that is consistent with the
dynamic structural model. In the empirical application,
jt
~
is roughly evenly discretized
into eight levels.
20
We then solve the system of F.O.C.s for the optimal prices and calculate the
single-period profits for each firm at each industry state. Using the single-period profits
as primitives of the dynamic game, we estimate the dynamic parameters and using
the Mathematical Programming with Equilibrium Constraints (MPEC) approach.
Define
jt jt jt
d
1
as firm j’s observed quality change as a result of its
investment at time t and the realization of the industry-wide shock . According to the
state transition process, product quality changes can be positive ( 1
jt
d ), zero ( 0
jt
d ),
or negative ( 1
jt
d ).The observed quality changes overtime across firms allow us to
form the likelihood function to estimate the dynamic parameters. The likelihood function
is:
21
. Pr 1 1 Pr 0 0 & 0 1 Pr
, Pr Pr 1 1 0 Pr
, Pr 1 Pr 1 1 1 Pr
where
, 0 0 & 0 1 Pr 1 0 Pr 1 1 Pr , ;
1 1
) 1 (
1 1
) 1 (
1 1
0 0
1 1
j
jt
j
jt t t
j
I
jt
I
jt t
j
I
jt
I
jt t
t
t t t t jt
I I
I
I
I I I I d L
jt jt
jt jt
(21)
20
We tested the sensitivity of the choice of the total number of discrete quality levels. We found that the
pattern of the results do not vary much.
21
Note that if any firm’s quality is at either bound of the state space, we need to adjust the likelihood
function for that particular observation accordingly.
47
1 1 Pr
t
I is the joint probability that at least one firm’s quality improves, 1 0 Pr
t
I is
the joint probability that at least one firm’s quality deteriorates, 0 0 & 0 1 Pr
t t
I I is
the joint probability of no change in all firms’ quality levels. These three possible events
are mutually exclusive within each time period. ) 1 ( 1
1
jt jt
d I and ) 1 ( 1
0
jt jt
d I are
indicator variables that represent firm j’s product quality movement.
) 1 ( ) 1 Pr( Pr
1
jt jt jt jt
x x v is the probability that firm j’s investment succeeds.
In most empirical applications involving investment, it is often very hard, or
impossible, to obtain data on investment expenditure at the product level because firms
would not release such competitive information and might even have trouble measuring it
accurately. Instead, we estimate investment cost x for each firm at each state that is
implied by the equilibrium, and substitute the corresponding values into the above
likelihood function. Assuming rational investment behavior, value functions satisfy the
Bellman equation (Equation (12)), and policy functions are the solutions to the F.O.C.
equation (15). Therefore, we impose equation (12) and (15) as constraints of the objective
function. Eventually, we solve the following constrained optimization problem:
j. x
j
x
v p
v W
j V E x V
V x L
j
v j
j
j
j j j j j j j
V x
j
, 0
, 0 1
) (
) | (
], | ) , ( [ ) , ( ) , ( subject to
) , , , ( max
' '
, , ,
(22)
Although the constrained optimization problem increases the size of the parameter space,
the Jacobian matrix of the constraints is very sparse. This sparsity, along with analytically
48
tractable gradients, speeds optimization and enables (22) to be maximized using a
Newton-Raphson method rather than a Quasi-Newton algorithm. We call the KNITRO
optimization package from Matlab to solve this constrained optimization problem.
Note that the exogenous shock in the state transition process is assumed to occur
with probability and is common to all firms. Thus it is identified by the propensity of
all firms’ qualities to deteriorate in the same period. represents the effectiveness of
individual firm’s investment, so it is identified by any individual firm’s tendency to
improve its quality in a single period.
Simulation
Before estimating the model with the market data, we used it to recover known
parameters from simulated datasets. This ensures the model works as advertised and
guards against errors in the estimation code. We follow the Gauss-Jacobi algorithm
described in Pakes and McGuire (1994) to compute the value functions and policy
functions in equilibrium for a given set of parameter values. The algorithm is essentially
an iterative searching process: a guess of value functions
0
V and policy functions
0
x are
plugged into the right hand side of the Bellman equation to obtain a new set of value
functions
1
V ; analytical solutions on policy function give a new set of policy functions
1
x ; this procedure is repeated until it converges according to a stopping criterion. It was
run multiple times using different starting values to numerically verify equilibrium
existence and uniqueness.
Starting from a given initial quality state, we obtain the corresponding optimal
investment from the equilibrium policy functions, and then compute the next period’s
49
product quality by simulating the outcome of the investment and the realization of the
industry shock. This simulation process is repeated to generate the simulated datasets
with 50 or 100 time periods. In the simulated datasets, we observe both product quality
and investment overtime, so we estimate the model using data on both. We are able to
recover the parameter values and find that having longer time periods and/or more firms
in the data would allow us to obtain more accurate parameter estimates.
We then take away data on investment, to reproduce the scenario in the empirical
application, and try to recover parameter values by only using data on product quality.
Table 7 shows the true values of the parameters and the estimated values. It suggests that
we can get reasonable estimates with relatively small sample size even if we do not
observe firms’ investment expenditure.
Table 7: Simulation Results
Chapter Two: Empirical Application
The proposed model is tested using market data from the U.S. full-size pickup truck
category. In this section, we first briefly introduce the U.S. automotive industry and
describe the data, and then we discuss the procedure to make the quality predictions.
Institutional Context and Data
The U.S. automotive industry has been one of the most important sectors in the U.S.
economy. According to a recent study by Center for Automotive Research (CAR), the
auto industry has historically contributed 3% – 3.5% to the U.S. gross domestic product
Parameters True Value Estimates
ρ 2 1.99
δ 0.3 0.29
50
(GDP). The practice of annual model-year design changes has been maintained since
General Motors (GM) introduced it in early 1920s.The need to constantly innovate and
stay competitive in the market makes the automakers focus on R&D. The industry spends
$16 to $18 billion dollars a year on R&D, but it should be pointed out that, unlike many
other industries, R&D investment in the auto industry is largely funded by the industry
itself. Only 1% of the $16 billion in automotive R&D was funded through the federal
government in 2007.
22
North American International Auto Show (NAIAS) is one of the most important
annual events for automotive manufacturers. It is held each January at the COBO
Exhibition Center in downtown Detroit. Besides the debut of concept vehicles,
automakers often introduce their new production vehicles (i.e. the ones that are put into
mass production) to the auto show. It could be the introduction of a new model entering a
specific segment or a new model of existing vehicles. The model-year of these new
models could be the same as the calendar year when the auto show took place, or the
following calendar year.
23
Not all automakers participate in the auto show every year, nor
do they display all of their product lines on the show stage. The industry generally has
done a very good job at keeping the new introductions under wraps until they are brought
out on stage.
Among all automobile segments, we are interested in the full-size pickup truck
segment, because 1) it used to be the most profitable product category for automakers; 2)
22
Reported in “Contribution of the Automotive Industry to the Economics and All Fifty States and the
United States” by Center for Automotive Research, 2010.
23
www.naias.com
51
it has a small number of players in the market; 3) the definition of the full-size pickup
truck category is relatively clear and has a long history in the market. The market for full-
size pickup trucks has been stable and dominated by the “Big Three” U.S. automakers –
GM, Ford, Chrysler since the 1980’s. On average, the “Big Three” accounted for about
64% of the total U.S. truck market in the last 25 years.
There were two new entrants in the full-size pickup truck market within the past
decade. Toyota introduced its first full-size pickup truck, the Tundra, in 2000. At the time
of entry, it was considered as the first full-size import-brand truck built with an American
look and feel, but still suffered the perception of being too small and carlike to pose a
serious threat to the “Big Three”. However, it turned out to have the largest initial vehicle
sales in Toyota history, and captured 9.2% of the market by 2007. Nissan introduced its
brand new full-size pickup, the Titan, in 2004 with massive dimensions, best-in-class
interior room, and the most powerful standard V-8 engine in its class. At the time it
seemed to be a serious new threat to the “Big Three”, but it did not reach the same
success as the Tundra and finished 2007 with only 2.4% share of the market.
Data used for estimation include truck features, list prices, and unit sales for each
model-year from 1984 to 2009, collected from Ward’s Automotive Yearbook. The major
truck brands are Ford F-Series, Chevrolet Silverado/GMC Sierra, and Dodge Ram.
Chevrolet Silverado and GMC Sierra are two twin brands under GM and have shared the
same characteristics since the late 1980’s. The characteristics and prices correspond to
those of the base model for a given truck. All prices are transformed into 1999 dollars
using the Consumer Price Index (CPI). Two observed truck characteristics are included:
52
horsepower (a measure of the power capacity of the engine) and a dummy variable that
indicates whether an extended cab option is available, representing extra space of the
seating area inside a truck.
Figure 4 shows the prices of the five major trucks over time. In 1989, GM
renamed its GMC division truck as “Sierra”, which basically is the same truck as its
Chevrolet Silverado. This might be the reason that Sierra’s price dropped to the same
level as Silverado in the same year, and has remained close since then. Ford introduced
the new generation of its F-150 in 2004 with a large increase in its price, but then cut the
price dramatically in the following four years. The price reduction might be a strategic
action in price competition. Similarly, Toyota introduced the new generation of its
Tundra in 2007 with a large price increase as well. The last entrant, Nissan Titan, topped
all the other trucks with the highest price since its entry.
Figure 4: Truck Prices 1984 – 2009 (in 1999 dollars)
$11,000
$13,000
$15,000
$17,000
$19,000
$21,000
$23,000
Ram F-150 Silverado Sierra Tundra Titan
53
Figure 5 shows the trucks’ market shares, measured as a fraction of all full-size,
mid-size and compact pickups sales. The two market leaders are Ford F-150 and
Chevrolet Silverado, accounting for about 50% of the market on average over the sample
period. Dodge Ram caught up in mid 90s and remained the third leader in the market.
The market share of GMC Sierra remains nearly constant across all model-years and is in
the fourth place in the market. Tundra quickly grasped the market, especially after its
introduction of the new generation in 2007, and has closed the gap on Sierra. The latest
entrant, the Titan, only had a slight increase in market share the year after its entry, and
then it never caught up with other trucks in the market.
Figure 5: Market Shares by Truck Model 1984 – 2009
Benchmark and Prediction Procedure
The goal of the chapter is to test the usefulness of the structural model in predicting
incumbents’ competitive reactions in product quality, and it is done by comparing its
0.0%
5.0%
10.0%
15.0%
20.0%
25.0%
30.0%
35.0%
Ram F-150 Silverado Sierra Tundra Titan
54
predictions to a benchmark model. Similar to the benchmark model used by Ailawadi et
al. (2005), we consider a vector autoregression (VAR) model as the benchmark. It
predicts future product quality levels based on observed past ones and allows firms’
product quality to be correlated. Note that the benchmark model is estimated using the
same data as used in estimating the dynamic structural model. The data set consists of a
vector of discrete product quality J j
jt
,..., 1 , and T t ,..., 1 , where T is the last period
before Tundra entered the market. We estimate the following VAR model:
Jt
t
Jt
t
Jt
t
e
e
A
1
1
1 1 1
, (23)
where A is a JxJ matrix of parameters to be estimated. Note that the estimation is based
on data before the entry occurs, so the benchmark model cannot predict future qualities
for new trucks (Tundra and Titan). We will only compare the predictions of the
incumbent firms between the VAR model and the structural model.
After getting the parameter estimates of the structural model and the benchmark
model, the next question is how to use them to predict future product quality levels for
each incumbent in each time period after entry. For the dynamic structural model, we first
need to know how much each firm would invest in equilibrium after the entry occurs.
This requires solving the equilibrium of the stochastic dynamic oligopoly game following
Pakes and McGuire (1994) algorithm under the assumption of J+1 firms. Having the
optimal investment solutions to the dynamic game with J+1 firms, we can then simulate
55
the outcomes of firms’ investment based on the state transition process (Equation (11)).
The simulation is conducted for each incumbent in each time period following the entry.
There is an implicit assumption in this counterfactual entry analysis: the entry
occurs exogenously. It implies that the incumbents do not act to strategically deter entry.
It may be justified by the observation that, in a category with high fixed costs of entry
which is dominated by two firms, the incumbents are more likely to react to each other
than to react to the threat of potential entry. Still, it is a strong assumption and may not
apply in settings with lower fixed costs, so it would be a very interesting topic for future
research.
For the benchmark model, we plug the incumbents’ quality observed at the last
time period prior to the entry into the right hand side of equation (14), and forecast their
quality levels in the entry period by simulating parameter matrix A drawn from its
estimated. Since the predicted dependent variable of the VAR model is continuous, we
round them up to the nearest integer. This process is again conducted for each incumbent
in each time period after the entry occurs.
Chapter Two: Results
The two entrants, Toyota Tundra and Nissan Titan, entered the US full-size pickup truck
market in different time periods. We conduct two separate prediction exercises: predict
incumbents’ product quality levels after the Tundra’s entry in 2000 (labeled as the “Case
of Tundra”); and predict incumbents’ product quality levels after the Titan’s entry in
2004 (labeled as the “Case of Titan). The model is estimated using data prior to Tundra’s
56
entry and considering the market leaders, Ford F-150 and Chevrolet Silverado, as the
incumbents in both cases.
Static Parameter Estimates
Table 8 shows the estimates of the model parameters. Price coefficient is −0.194, and the
marginal cost coefficient is 2.141, translating to a marginal cost of about $8,400 to
produce a truck. Consumers prefer trucks with higher horsepower ( HP _ ) and extra
interior space at the back of the front seats ( ExtCab _ ). More specifically, the implied
willingness-to-pay (WTP) for upgrading from 200-HP to 300-HP is about $5,200 and the
WTP for an extended cab option is about $3,572.
Table 8: Model Parameter Estimates without Truck Features
The goal of estimating these static parameters is to obtain consistent estimates of
the perceived product quality for each incumbent over time
jt
~
. Figure 6 plots the
discretized quality levels for the two incumbents over the sample time period for
estimation. F-150 has higher quality than Silverado in early years until 1989 when
Silverado closed the gap. During the following ten years, the two remained close
Parameter Estimates
α 0.194
γ 2.141
θ_HP 0.010
θ_ExtCab 0.693
ρ 0.0004
δ 0.520
57
competitors by having the same quality levels, with the only exception in 1996. F-150
was back to be the quality leader again in 1999.
Figure 6: Estimated Perceived Quality after Discretization
Dynamic Parameter Estimates
The investment effectiveness parameter maps investment expenditure into the
probability of successful investment outcome. Its estimated value is 0.0004, suggesting
that the median of the probability of successful investment is 0.72 across all possible
industry states. The estimated probability of an industry-wide shock affecting the industry
state is 0.52.
In addition to the parameter estimates, we could also learn about how quality
competition influences firms’ equilibrium investments and expected payoffs. Figure 7
and Figure 8 plot the estimated optimal investment and net present value of profits at
0
1
2
3
4
5
6
Ford F-150 Chevy Silverado
58
each industry state under the case of Titan. The patterns of the graphs look the same for
the case of Tundra.
Figure 7: Quality Investment in Equilibrium
1
2
3
4
5
6
7
8
1
2
3
4
5
6
7
8
0
2000
4000
6000
8000
10000
12000
14000
Own Quality
Competitor's Quality
Optimal Investment
59
Figure 8: Value Functions in Equilibrium
Several features of the results are interesting. On the optimal investment side, first,
given a firm’s own quality state, its optimal investment is decreasing in its competitor’s
quality. But the marginal change in the optimal investment is smaller if the firm is a
quality leader. This suggests that the quality leader’s investment level depends less on the
quality differentials than does the quality laggard. Moreover, second, as the quality
differential gets larger, the quality laggard has a lower incentive to invest than does the
quality leader. The quality laggard only invests to compete with the leader, so the
marginal return to quality improvement is very small when the quality difference
becomes large. However, the quality leader competes with the outside good as well as the
laggard, so it still has incentive to invest even when it has big quality advantage over the
laggard. Third, given the competitor’s quality level, the firm’s optimal investment
1
2
3
4
5
6
7
8
1
2
3
4
5
6
7
8
0
1
2
3
4
5
6
7
8
x 10
5
Own Quality
Competitor's Quality
Discounted Present Value
60
presents an inverse-U shape in its own quality. But the marginal change in investment is
much smaller when the competitor is a quality leader. This is because the marginal
expected payoffs are lower when competing with the quality leader.
Looking at the expected payoffs, it is unsurprising to see that each firm’s profits
fall with its competitor’s quality. But the marginal loss is much higher when the firm is a
quality leader, because it is costly to maintain the high quality status. It implies that
quality competition has a greater impact on the quality leader’s expected profits. Second,
the marginal expected payoff is smaller when competing with a quality leader than when
competing with a quality laggard, which explains the second finding in the above
investment competition pattern.
Predicted Investment Reactions
To help the entrant make better entry and investment decisions, it is useful to understand
how a new product entry influences the incumbents’ optimal investment and how the
effects of entry depend on the entrant’s quality. We could answer these questions by
comparing the incumbents’ optimal investment before and after entry.
The results show that the incumbents’ reactions are different, depending on their
quality standing within the marketplace. First, when all product quality levels are located
at either extreme of the quality space, the incumbents will reduce their optimal
investment after entry. Second, the entry poses a greater threat to the quality laggard than
to the quality leader. Therefore, the quality laggard reacts by increasing the investment in
order to stay competitive, but it significantly reduces its investment if the entrant comes
in with a higher quality. Third, the quality leader increases its investment only when its
61
quality advantage over the entrant is small, otherwise it always reduces its investment
level in response to the entry.
Quality Predictions
The quality prediction procedure is done with 1000 simulated draws for both the
structural model and the benchmark model. In each round, we calculate the difference
between the predicted quality and the observed quality for each incumbent in each time
period after the entry.
We calculate two measures to evaluate the accuracy of the predictions: the mean
squared prediction error (MSPE) and the percentage of correct predictions, i.e. the hit-
rate. Table 9 shows that the proposed structural model could predict well and better than
the benchmark model. More specifically, the structural model improves the MSPE by 22%
under the case of Titan and 3% under the case of Tundra, and its average hit rates more
than double the hit rates of the benchmark model under both cases.
Table 9: Prediction Accuracy Comparison
Case of Titan Case of Tundra
12 8
Benchmark Model 2.08 4.51
Structural Model 1.62 4.38
Improvement 22% 3%
Benchmark Model 15% 13%
Structural Model 37% 27%
Improvement 147% 108%
Total # of Predictions
MSPE
Hit-Rate
62
Chapter Two: Discussion
Dynamic structural models have gained tremendous attention in the industrial
organization and marketing literatures, mostly to understand how competition or a policy
change would affect strategic interactions among competitors and the market structure.
This chapter provides an initial step to show that dynamic structural models could be
used to help managers understand how incumbents will react in product quality to a new
entry. Despite their apparent complexity, they produce predictions that are superior to
descriptive models. This is because the change in the market – a new product entry –
affects incumbents’ decision rules in a way which cannot be predicted by an atheoretical
model. Because the dynamic structural framework models how market conditions change
firms’ equilibrium investments, it can derive decision rules which reflect changes in
market conditions such as a new product entry. Therefore, the dynamic structural model
is more appropriate for predicting incumbents’ competitive reactions after entry.
This chapter also has important managerial implications to both incumbents and
entrants. First, it helps firms understand how quality differentials influence the quality
competition. Second, it offers insights on how the entrant’s quality affects incumbents’
competitive reactions in product quality. The implication is that the new entry causes a
more serious threat to quality laggards and could even drive them out of business.
Although the results are encouraging, we need to acknowledge the limitations and
directions for future research. One limitation of the model is that we do not consider
endogenous entry decisions. In our case, we simply do not have enough observations on
product entry, and one drawback of not modeling the entry decision is that incumbents
63
are not allowed to act to deter entry. But it would be fairly easy to accommodate
endogenous entry decision in the model. Another limitation is that the counterfactual
entry exercise has an implicit assumption that the same equilibrium has been played
before and after the entry, which may apply to counterfactuals in a number of policy
simulations in similar models. A new product introduction may dramatically change the
nature of the equilibrium, which might contaminate the validity of the predictions. Third,
we assume consumer demand to be static in a durable good market. Modeling forward-
looking behavior on both demand and supply sides adds substantially more complexity,
but may improve the model’s predictive ability even further.
We have shown the usefulness of the dynamic structural models in predicting
incumbents’ quality reactions in one application. There are many potential research
avenues to adjust the model to other applications. First, firms may have different abilities
to innovate, so it might be interesting to introduce firm heterogeneity into the model. For
example, Apple is known for its creativity and innovativeness. It might have a higher
probability to succeed in R&D than a small firm. Second, product life cycle may play a
role in determining the state transition process. In early stage of the product life cycle,
innovation may be less likely to succeed. But as the product and technology become
more mature, it becomes easier to improve product quality. This implies that the effects
of current investment on future product quality could be time dependent, which means a
non-stationary state transition process is needed. The challenge is that it might be very
difficult to solve a non-stationary infinite horizon dynamic problem. Third, the current
model is within a complete information scenario where each firm shares common
64
knowledge about the market. One might want to introduce private information into the
model in other applications. Although it is computationally demanding to estimate such
models, rapidly improvement in computing power and better available market data may
make it more feasible.
65
BIBLIOGRAPHY
Aaker, D. and R. Jacobson (1994), “The Financial Information Content of Perceived
Quality,” Journal of Marketing Research, 31(2): 191-201.
Ai, C. and E. C. Norton (2003), “Interaction Terms in Logit and Probit Models,”
Economics Letters, 80(1): 123-129.
Ailawadi, K. L., P. K. Kopalle, and S. A. Neslin (2005), “Predicting Competitive
Response to a Major Policy Change: Combining Game-Theoretic and Empirical
Analyses,” Marketing Science, 24(1): 12-24.
Aguirregabiria, V. and P. Mira (2007), “Sequential Estimation of Dynamic Discrete
Games,” Econometrica, 75(1): 1-53.
Albuquerque, P. and B. J. Bronnenberg (2010), “Measuring the Impact of Negative
Demand Shocks on Car Dealer Networks,” mimeo, Simon Graduate School of
Business, University of Rochester, Rochester.
Bajari, P., C. L. Benkard, and J. Levin (2007), “Estimating Dynamic Models of Imperfect
Competition,” Econometrica, 75(5): 1331-1370.
Barroso, A. (2007), “Advertising and Demand: Some Empirical Evidence in the Spanish
Automobile Market,” mimeo, Center for Monetary and Financial Studies (CEMFI),
Madrid.
Belch, G. E. and M. E. Belch (2007), Advertising and Promotion: an Integrated
Marketing Communications Perspective, 7
th
Ed. New York, NY: McGraw-Hill
Irwin.
Beresteanu, A., P. B. Ellickson, and S. Misra (2010), “The Dynamics of Retail Oligopoly,”
mimeo, University of Rochester.
Berry, S., J. Levinsohn, and A. Pakes (1995), “Automobile Prices in Market Equilibrium,”
Econometrica, 60(4): 889-917.
Berry, S., J. Levinsohn, and A. Pakes (2004), “Differentiated Products Demand Systems
from a Combination of Micro and Macro Data: The New Car Market,” Journal of
Political Economy, 112(1): 68-105.
Briggs, R. and G. Stuart (2006), What Sticks, Kaplan, Chicago.
Bucklin, R. E., S. Siddarth, and J. M. Silva-Risso (2008), “Distribution Intensity and New
Car Choice,” Journal of Marketing Research, 45(4): 473-486.
66
Burton, S., D. R. Lichtenstein, A. Biswas, and K. Fraccastoro (1994), “The Role of
Attributions in Consumer Perceptions of Retail Advertisements Promoting Price
Discounts,” Marketing Letters, 5(2): 131-140.
Busse, M. R., J. M. Silva-Risso, and F. Zettelmeyer (2006), “$1000 Cash Back: The
Pass-Through of Auto Manufacturer Promotions,” American Economic Review,
96(4): 1253-1270.
Busse, M. R., D. I. Simester, and F. Zettelmeyer (2009), “The Best Price You’ll Ever Get:
The 2005 Employee Discount Pricing Promotions in the U.S. Automobile Industry,”
Marketing Science, 29(2): 268-290.
Chang, K., S. Siddarth, and J. M. Silva-Risso (2004), “Developing Regional Promotional
Programs in the Automotive Industry Based on Transaction Data: A Hierarchical
Bayes Model with Zip Code-Level Response Heterogeneity,” mimeo, School of
Business Administration, University of Seoul.
Chen, Y., S. Yang, and Y. Zhao (2008), “A Simultaneous Model of Consumer Brand
Choice and Negotiated Price,” Management Science, 54(3): 538-549.
Ching, Andrew T. (2010), “A Dynamic Oligopoly Structural Model for the Prescription
Drug Market after Patent Expiration,” International Economic Review, 51(4): 1175-
1027.
Chintagunta, P., T. Erdem, P. E. Rossi, and M. Wedel (2006), “Structural Modeling in
Marketing: Review and Assessment,” Marketing Science, 25(6): 604-616.
Chintagunta, P. and H. S. Nair (2010), “Discrete-Choice Models of Consumer Demand in
Marketing,” mimeo, University of Chicago.
Clark, C.R., U. Doraszelski, and M. Draganska (2009), “The Effect of Advertising on
Brand Awareness and Perceived Quality: An Empirical Investigation Using Panel
Data,” Quantitative Marketing and Economics, 7(2): 207-236.
Dasgupta, S., S. Siddarth, and J. M. Silva-Risso (2007), “Lease or Buy? A Structural
Model of a Consumer’s Vehicle and Contract Choice,” Journal of Marketing
Research, 44(3): 490-502.
Doraszelski, U. and A. Pakes (2007), “A Framework for Applied Dynamic Analysis in
IO,” Handbook of Industrial Organization, Vol. 3, 2007, North-Holland,
Amsterdam, pp. 1887-1966.
Doraszelski, U. and M. Satterthwite (2010), “Computable Markov-Perfect Industry
Dynamics,” RAND Journal of Economics, 41(2): 215-243.
67
Draganska, M. and D. Klapper (2010), “Choice Set Heterogeneity and the Role of
Advertising: An analysis with Micro and Macro Data,” mimeo, Stanford
University, Stanford.
Dubé , J. P., G. J. Hitsch, and P. Manchanda (2005), “An Empirical Model of Advertising
Dynamics,” Quantitative Marketing and Economics, 3(2): 107-144.
Erdem, T., M. Keane, and B. Sun (2008), “The Impact of Advertising on Consumer Price
Elasticity in Experience Goods Markets,” Quantitative Marketing and Economics,
6(2): 139-176.
Ericson, R. and A. Pakes (1995), “Markov-Perfect Industry Dynamics: a Framework for
Empirical Work,” The Review of Economic Studies, 62(1):53-82.
Folkes, V. S. (1988), “Recent Attribution Research in Consumer Behavior: A Review and
New Directions,” Journal of Consumer Research, 14(4): 548-565.
Gatignon, H. (1984), “Competition as a Moderator of the Effects of Advertising on Sales,”
Journal of Marketing Research, 21(4): 387-398.
Goettler, R. and B. Gordon (2011), “Does AMD Spur Intel to Innovate More?” mimeo,
University of Chicago.
Gowrisankaran, G. and R. J. Town (1997), “Dynamic Equilibrium in the Hospital
Industry,” Journal of Economics and Management Strategy, 6(1): 45-74.
Grossman, G. and C. Shapiro (1984), “Informative Advertising with Differentiated
Products,” Review of Economic Studies, 51(1): 63-81.
He, X., A. Krishnamoorthy, A. Prasad, and S. P. Sethi (2011), “Retail Competition and
Cooperative Advertising,” mimeo, UT-Dallas.
Iyer, G. 1998. Coordinating Channels under Price and Nonprice Competition. Marketing
Science, 17(4) 338-355.
Jedidi, K., C. F. Mela, and S. Gupta (1999), “Managing Advertising and Promotion for
Long-run Profitability,” Marketing Science, 18(1): 1-22.
Kanetkar, V., C. B. Weinberg, and D. L. Weiss (1992), “Price Sensitivity and Television
Advertising Exposures: Some Empirical Findings,” Marketing Science, 11(4): 359-
371.
Kaul, A. and D. R. Wittink (1995), “Empirical Generalizations about the Impact of
Advertising on Price Sensitivity and Price,” Marketing Science, 14(3): 151-160.
68
Kolay, S. (2010), “Persuasive Advertisements in Distribution Channels,” mimeo, the Paul
Merage School of Business, University of California, Irvine.
Kuksov, D. and J. M. Villas-Boas (2008), “Endogeneity and individual Consumer
Choice,” Journal of Marketing Research, 45(6): 702-714.
Kwoka, J. E. Jr. (1993), “The Sales and Competitive Effects of Styling and Advertising
Practices in the U.S. Auto Industry,” The Review of Economics and Statistics,
75(4): 649-656.
Liaukonyte, J. (2008), “Is Comparative Advertising an Active Ingredient in the Market
for Pain Relief? A Structural Empirical Model of Advertising Content,” mimeo,
Department of Applied Economics and Management, Cornell University.
Lichtenstein, D. R., S. Burton, and B. S. O’Hara (1989), “Marketplace Attributions and
Consumer Evaluations of Discount Claims,” Psychology and Marketing, 6(3): 163-
180.
Luan, Y. J. and K. Sudhir (2010), “Forecasting Marketing-Mix Responsiveness for New
Products,” Journal of Marketing Research, 47(3): 444-457.
Mela, C. F., S. Gupta, and D. R. Lehmann (1997), “The Long-Term Impact of Promotion
and Advertising on Consumer Brand Choice,” Journal of Marketing Research,
34(2): 248-261.
Milgrom, P. and J. Roberts (1986), “Price and Advertising Signals of Product Quality,”
Journal of Marketing Research, 94(4): 796-821.
Mitra, A. and J. G. Lynch Jr. (1995), “Toward a Reconciliation of Market Power and
Information Theories of Advertising Effects on Price Elasticity,” Journal of
Consumer Research, 21(4): 644-659.
Mitra, D. and P. N. Golder (2006), “How Does Objective Quality Affect Perceived
Quality? Short-Term Effects, Long-Term Effects, and Asymmetries,” Marketing
Science, 25(3): 230-347.
Nelson, P. (1970), “Information and Consumer Behavior,” Journal of Political Economy,
78(2): 311-329.
Pakes, A. and P. McGuire (1994), “Computing Markov-perfect Nash Equilibria:
Numerical Implications of Dynamic Differentiated Product Model,” RAND Journal
of Economics, 25(4): 555-589.
69
Pakes, A., M. Ostrovsky, and S. Berry (2007), “Simple Estimators for the Parameters of
Discrete Dynamic Games (with Entry-Exit Examples)” RAND Journal of
Economics, 38(2): 373-399.
Phillips, L., D. Chang, and R. D. Buzzell (1983), “Product Quality, Cost Position and
Business Performance: a Test of Some Key Hypotheses,” Journal of Marketing,
47(2): 26-43.
Rao, A. R. and K. B. Monroe (1989), “The Effect of Price, Brand Name, and Store Name
on Buyers’ Perceptions of Product Quality: An Integrative Review,” Journal of
Marketing Research, 26(3): 351-357.
Rust, J. (1987), “Optimal Replacement of GMC Bus Engine: An Empirical Model of
Harold Zurcher,” Econometrica, 55(5): 999-1033.
Scott Morton, F., F. Zettelmeyer, and J. M. Silva-Risso (2003), “Consumer Information
and Discrimination: Does the Internet Affect the Pricing of New Cars to Women
and Minorities?” Quantitative Marketing and Economics, 1(1): 65-92.
Song, M. (2011), “A Dynamic Analysis of Cooperative Research in the Semiconductor
Industry,” International Economic Review, 52(4): 1157-1177.
Moorthy, S. and H. Zhao (2009), “Advertising Spending and Perceived Quality,”
Marketing Letters, 11(3): 221-233.
Srinivasan, S., K. Pauwels, J. M. Silva-Risso, and D. M. Hanssens (2009), “Product
Innovations, Advertising, and Stock Returns,” Journal of Marketing, 73(1): 24-43.
Stigler, G. J. (1961), “The Economics of Information,” Journal of Political Economy,
69(3): 13-26.
Su, C. L. and K. Judd (2011), “Constrained Optimization Approaches to Estimation of
Structural Models,” mimeo, University of Chicago.
Sudhir, K. (2001), “Competitive Pricing Behavior in the Auto Market: A Structural
Analysis,” Marketing Science, 20(1): 42-60.
Tellis, G. J., R. K. Chandy, and P. Thaivanich (2000), “Which Ad Works When, Where
and How Often? Modeling the Effects of Direct Television Advertising,” Journal
of Marketing Research, 37(1): 32-46.
Tellis, G. J. and C. Fornell (1988), “Advertising and Quality over the Product Life Cycle:
a Contingency Theory,” Journal of Marketing Research, 15(1): 64-71.
Tellis, G. J. and B. Wernerfelt (1987), “Competitive Price and Quality under Asymmetric
Information,” Marketing Science, 6(3): 240-253.
70
T. Peterson (2006), “Nissan’s Bulked Up Family Truck,” Bloomberg Businessweek,
April 12.
Thomadsen, R. (2005), “The Effect of Ownership Structure on Prices in Geographically
Differentiated Industries,” RAND Journal of Economics, 36(4): 908-929.
Train, K. (2003), Discrete Choice Methods with Simulation, Cambridge University Press,
New York, NY.
Villas-Boas, J. M., and R. S. Winer (1999), “Endogeneity in Brand Choice Models,”
Management Science, 45(10): 1324-1338.
Weintraub, G. Y., C. L. Benkard, and B. Van Roy (2008), “Markov Perfect Industry
Dynamics with Many Firms,” Econometrica, 76(6): 1375-1411.
Wolinsky, A. (1983), “Prices as Signals of Product Quality,” Review of Economic Studies,
50(4): 647-658.
71
APPENDIX A: TRUCK PRICES
It is critical to control for representative price levels when estimating price advertising
response. Three steps are used to remove consumer-, dealer-, product-, and transaction-
level heterogeneity from the truck transaction data.
The first step is to create a unified measure of price. There are three transaction
types observed in the data: lease, dealer finance, and “cash” in which customers finance
through an outside lender or pay in full. Following Dasgupta et al. (2007), for cash
transactions, price is the vehicle price less the manufacturer rebate. For dealer finance
transactions, price is given by
r
r
mpt rebate downpmt p
T
1
1
1
1 1
* ) (
, (24)
where downpmt is the initial down payment on the vehicle, rebate is the cash rebate
offered by manufacturer, T is the term of the loan in months, mpt is the monthly
payment that consumers have to pay for the entire duration of the term (a function of
APR), and r
is the market interest rate that discounts the payment. For lease transactions,
price is
T
T
r
resid
r
r
mpt rebate downpmt p
1
1
*
1
1
1
1 1
* ) (
, (25)
where resid is the residual value of the vehicle when the lease contract ends; that is, the
amount the customer would have to pay to purchase the car at the end of the lease term.
The Consumer Price Index is used to express all prices in July 2001 dollars.
72
The second step is to remove the impact of heterogeneity on prices. Let
jldsz
p be
the price of transaction (or sale) l of truck j on day d in month s in zip code z. Equation
(26) is used to estimate the impact of various sources of heterogeneity on transaction
prices.
jldsz jl d l z js jldsz
I H x p
2 1
, (26)
js
is a truck-month intercept to allow for unobserved demand shocks over months
within a truck.
z
is a zip code dummy variable to control for the effect of zip code
demographics on prices.
l
x includes dealership dummies to allow for the possibility that
some dealers charge higher prices than others; customer gender dummies to allow
negotiated prices to be different between male and female buyers (Scott Morton et al.
2003); transaction type dummies to allow some transaction types (lease, cash, dealer
finance) to result in higher total prices than others; weekday dummies to allow prices to
fluctuate throughout the week; product-level characteristics such as model year, cylinders,
displacement, doors, and drive type; and tenure on the dealer’s lot (daystoturn and
daystoturn squared).
d
H includes holiday dummies to control for price variation during
holidays.
jl
I is the weekly inventory level of truck j at the dealership where transaction l
takes place to control for price variation due to dealer inventory level.
The third step is to construct a panel of weekly market prices for each truck
within each zip code. The market price
zjt
p is the median of
2 1
ˆ ˆ ˆ ˆ
jl d z js
I H
among all sales of truck j in week t within zip code z.
73
The price regression in equation (26) controls for truck-month demand shocks,
dealer effects, customer and zip code effects, transaction type effects, weekday and
holiday effects, inventory effects, and product characteristics. The regression explains
most of the variation in truck prices, with an adjusted R
2
of 0.70. The product
characteristics effects can be intuitively summarized by saying that larger engines, bigger
cabs, and newer trucks cost more. Selected results are shown in Table 10.
Table 10: Price Regression Results
Variable Variable
Financed 5668.87 ** Columbus Day Weekend 129.35
(53.7) (188.6)
Leased 7803.55 ** July 4 Weekend −92.69
(80.8) (161.3)
Female 1003.43 ** Labor Day Weekend −560.53 **
(71.6) (152.4)
Male
a
523.66 ** Memorial Day Weekend −633.75 **
(60.8) (171.9)
Sunday −7.37 MLK Day Weekend −141.23
(67.3) (207.3)
Tuesday 125.12 New Year's Day −572.28
(68.5) (464.3)
Wednesday 106.31 New Year's Eve −359.89
(68.0) (233.6)
Thursday −21.16 Presidents Day Weekend −540.00 **
(64.1) (185.1)
Friday 200.82 ** −128.85
(61.8) (182.4)
Saturday 252.68 ** daystoturn 2.78 **
(63.8) (0.6)
Christmas Day 6814.71 daystoturn squared −0.01 **
(4,226.6) (1.7E-3)
Christmas Eve 253.79 Inventory −2.77 **
(394.8) (0.8)
* significant at 95% confidence level; ** significant at 99% confidence level.
a
There are three gender classifications: Male, Female, and Unknown, so we can
estimate both "Female" and "Male" gender effect on price.
Thanksgiving Weekend
Estimate
(Std. Err.)
Estimate
(Std. Err.)
74
The most expensive option for purchasing a truck is to lease it, followed by dealer
financing. Women pay about $480 more than men, holding other factors constant,
consistent with Scott Morton et al. (2003). Prices are significantly higher on Fridays and
Saturdays than on other days of the week. Holiday results show that prices tend to fall
significantly during Labor Day, Memorial Day, and Presidents’ Day, but do not have
significant changes around Thanksgiving, Christmas, New Year’s, or July 4th.
The estimated effect of daystoturn on price is an inverted “U” shape with a peak
at 220 days, but we exercise caution in interpreting this effect. While we have controlled
for model type and options, there may still be some unobserved differences across trucks.
If so, it is likely that vehicles produced early in the model year are more attractive and
priced higher than those produced later. While this suggests daystoturn may control for
unobserved heterogeneity, it muddles the interpretation of its effect. Inventory has a
significantly negative effect on prices, consistent with intuition that dealers would lower
price to sell off their inventories.
75
APPENDIX B: MODEL SELECTION
A number of model selection decisions can influence the estimates of the interaction
effects, the main objective of the study. Because we do not have strong priors and do not
want to impose an incorrect theory on the data, we use statistical model selection
techniques to govern which variables should enter the model and how they should
interact.
First, we test three different functional forms to account for advertising carryover
effects ( a variables measure raw advertising expenditure):
Alternative 1:
1 1
) 1 ( ) 1 ln(
jt A jt A jt
A a A ,
Alternative 2:
1 1
) 1 (
jt A jt A jt
A a A ,
Alternative 3:
S
s
s jt As jt
a A
1
.
24
Table 11 shows the fit statistics for each model under each alternative. Because
Alternative 1 has the lowest AIC and BIC values we use it as the basis for the models
reported in Chapter 1 and for all of the model specification tests that follow.
Table 11: Advertising Carryover Measure Comparisons
Next, different specifications on truck and time fixed effects and consumer
response to marketing instruments are tested. Starting from the simplest possible
24
Following Tellis, Chandy and Thaivanich (2000), S was chosen as the number of lags at which the next
lag effect was insignificant. It was allowed to vary across types of advertising.
–LL BIC AIC
Alternative 1 2499788 2499891 2499824
Alternative 2 2499807 2499910 2499843
Alternative 3 2499862 2499959 2499896
76
specification, model complexity is increased in order to minimize the AIC and BIC. First,
these model selection criteria suggest including truck-month dummies (Model 2), rather
than simpler fixed effect structures, such as a set of truck dummies and a separate set of
month dummies (Model 1). Second, aggregate local dealer advertising (LDA) is excluded
from the analysis (Model 3). This is not too surprising since LDA primarily
communicates dealership existence and location rather than information specific to the
truck category. Third, the model includes first-order interactions between price and
advertising variables (Model 4). Finally, observed zip code demographics (Model 5) are
excluded from explaining truck preference heterogeneity.
For Model 1-Model 5, we also test whether consumer response parameters should
be common across trucks, group-specific (High/Low advertising spending), or brand-
specific. Table 12 displays the log likelihood value (LL), AIC and BIC under each model.
These criteria suggest that Model 4 with brand-specific effects should be the final model,
so this specification is reported in Chapter 1.
Table 12: Fit Statistics Comparison across Model Specifications
Model 1:Separate truck and month dummies, linear effects only
izjt zjt z zjt jt jt jt zjt z t s j izjt
YC DPAC MPAC MBAC p f X u ) ; , , , , (
,
–LL BIC AIC –LL BIC AIC –LL BIC AIC
Model 1 2500690 2500788 2500724 2500147 2500267 2500189 2499346 2499489 2499396
Model 2 2499788 2499891 2499824 2499331 2499451 2499373 2498485 2498628 2498535
Model 3 2499848 2499882 2499946 2499256 2499376 2499298 2498409 2498547 2498457
Model 4 2499764 2499865 2499801 2499267 2499387 2499309 2498372 2498515 2498422
Model 5 2499777 2499881 2499813 2499411 2499531 2499453 2498661 2498799 2498709
Common marketing response
parameters across brands
Group specific marketing
response parameters
Brand specific marketing
response parameters
77
zjt
Y
z jt
DP
z jt
MP
z jt
MB
z zjt
P
z z
YC DPAC MPAC MBAC p YC DPAC MPAC MBAC p f ; , , , ,
Model 2: Truck-month dummies, no interactions
izjt zjt z zjt jt jt jt zjt z t js izjt
YC DPAC MPAC MBAC p f X u ) ; , , , , ( ,
zjt
Y
z jt
DP
z jt
MP
z jt
MB
z zjt
P
z z
YC DPAC MPAC MBAC p YC DPAC MPAC MBAC p f ; , , , ,
Model 3: Truck-month dummies, no interactions, local dealer advertising included
izjt zjt z zjt jt jt jt jt zjt z t js izjt
YC LDAC DPAC MPAC MBAC p f X u ) ; , , , , ( ,
.
) ; , , , , , (
zjt
Y
z jt
LD
z jt
DP
z jt
MP
z jt
MB
z zjt
p
z
z
YC LDAC DPAC MPAC MBAC p
YC LDAC DPAC MPAC MBAC p f
Model 4: Truck-month dummies, first-order interactions
izjt zjt z zjt jt jt jt zjt z t js izjt
YC DPAC MPAC MBAC p f X u ) ; , , , , ( ,
zjt jt
DPp
zjt jt
MPp
zjt jt
MBp
jt jt
PP
jt jt
BDP
jt jt
BMP
zjt
Y
z jt
DP
z jt
MP
z jt
MB
z zjt
P
z z
p DPAC p MPAC p MBAC
DPAC MPAC DPAC MBAC MPAC MBAC
YC DPAC MPAC MBAC p YC DPAC MPAC MBAC p f
; , , , ,
Model 5: Truck-month dummies, first-order interactions, random coefficients
izjt zjt z zjt jt jt jt zjt z t js izjt
YC DPAC MPAC MBAC p f X u ) ; , , , , ( ,
78
zjt jt
DPp
zjt jt
MPp
zjt jt
MBp
jt jt
PP
jt jt
BDP
jt jt
BMP
zjt
Y
z jt
DP
z jt
MP
z jt
MB
z zjt
P
z z
p DPAC p MPAC p MBAC
DPAC MPAC DPAC MBAC MPAC MBAC
YC DPAC MPAC MBAC p YC DPAC MPAC MBAC p f
; , , , ,
z z z
D
~
~ ~
, where
z
D contains observed zip code demographics such as income
and population density.
79
APPENDIX C: NONPARAMETRIC IDENTIFICATION, ENDOGENEITY AND
SLOPE ENDOGENEITY
The primary effects of interest are interactions among advertising variables. These can
only be identified if there is significant asynchronous variation among all three variables.
Figures 9 and 10 show plots of advertising expenditures over weeks of each calendar year
from 2002 to 2005. MPA and DPA often go to zero, providing many observations of high
MBA with low MPA, high MBA with high MPA, high MBA with low DPA, and high
MBA with high DPA. In sum, it appears there is enough variation in the advertising data
to identify the interaction effects, at least for most trucks.
Figure 9: Advertising Expenditure over Weeks of Each Calendar Year
0
2
4
6
8
10
12
14
16
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51
Avalanche 2002
log(MBA+1) log(MPA+1) log(DPA+1)
0
2
4
6
8
10
12
14
16
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53
Avalanche 2003
log(MBA+1) log(MPA+1) log(DPA+1)
0
2
4
6
8
10
12
14
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51
Avalanche 2004
log(MBA+1) log(MPA+1) log(DPA+1)
0
1
2
3
4
5
6
7
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51
Avalanche 2005
log(MBA+1) log(MPA+1) log(DPA+1)
0
2
4
6
8
10
12
14
16
18
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51
F-Series 2002
log(MBA+1) log(MPA+1) log(DPA+1)
0
2
4
6
8
10
12
14
16
18
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53
F-Series 2003
log(MBA+1) log(MPA+1) log(DPA+1)
0
2
4
6
8
10
12
14
16
18
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51
F-Series 2004
log(MBA+1) log(MPA+1) log(DPA+1)
0
2
4
6
8
10
12
14
16
18
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51
F-Series 2005
log(MBA+1) log(MPA+1) log(DPA+1)
0
2
4
6
8
10
12
14
16
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51
Ram 2002
log(MBA+1) log(MPA+1) log(DPA+1)
0
2
4
6
8
10
12
14
16
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53
Ram 2003
log(MBA+1) log(MPA+1) log(DPA+1)
0
2
4
6
8
10
12
14
16
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51
Ram 2004
log(MBA+1) log(MPA+1) log(DPA+1)
0
2
4
6
8
10
12
14
16
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51
Ram 2005
log(MBA+1) log(MPA+1) log(DPA+1)
80
Figure 10: Advertising Expenditure over Weeks of Each Calendar Year
Since advertising variables may be partially determined by variation in demand
conditions, truck-month fixed effects are included in the model specification to control
for unobserved monthly demand shocks. The correlation between unobserved monthly
demand shocks and advertising is then among observed variables, rather than between
observed variables and the error term, eliminating any endogeneity bias from monthly
demand shocks.
While the truck-month dummy variables completely control for monthly
fluctuations in demand shocks, there still may be weekly changes in unobserved variables
0
2
4
6
8
10
12
14
16
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51
Tundra 2002
log(MBA+1) log(MPA+1) log(DPA+1)
0
2
4
6
8
10
12
14
16
18
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53
Tundra 2003
log(MBA+1) log(MPA+1) log(DPA+1)
0
2
4
6
8
10
12
14
16
18
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51
Tundra 2004
log(MBA+1) log(MPA+1) log(DPA+1)
0
2
4
6
8
10
12
14
16
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51
Tundra 2005
log(MBA+1) log(MPA+1) log(DPA+1)
0
2
4
6
8
10
12
14
16
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51
Silverado 2002
log(MBA+1) log(MPA+1) log(DPA+1)
0
2
4
6
8
10
12
14
16
18
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53
Silverado 2003
log(MBA+1) log(MPA+1) log(DPA+1)
0
2
4
6
8
10
12
14
16
18
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51
Silverado 2004
log(MBA+1) log(MPA+1) log(DPA+1)
0
2
4
6
8
10
12
14
16
18
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51
Silverado 2005
log(MBA+1) log(MPA+1) log(DPA+1)
0
2
4
6
8
10
12
14
16
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51
Sierra 2002
log(MBA+1) log(MPA+1) log(DPA+1)
0
2
4
6
8
10
12
14
16
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53
Sierra 2003
log(MBA+1) log(MPA+1) log(DPA+1)
0
2
4
6
8
10
12
14
16
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51
Sierra 2004
log(MBA+1) log(MPA+1) log(DPA+1)
0
2
4
6
8
10
12
14
16
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51
Sierra 2005
log(MBA+1) log(MPA+1) log(DPA+1)
81
that influence advertising expenditures. Holiday week dummies are included in the model
to predict consumers’ propensity for increased shopping at those times. Yet we cannot
rule out the possibility that marketing variables may still be correlated with unobserved
weekly departures from the mean demand level within the month. One way to resolve
this would be to take a full-information approach and model the supply side. We are
hesitant to model automakers’ advertising decisions in this way, for four reasons. First,
Briggs and Stuart (2006) argue that automakers do not set optimal advertising budgets:
In Detroit, the big spenders are General Motors, Ford, and Chrysler—the “big 3”
as they are known in Motown. The media director of a major Detroit agency told
us that when he was working on one of the Big 3 accounts, they set the level of
TV spending based on the reported [past] TV spending of the big guy across town,
assuming they must have the magic formula for determining spending. A few
years later, he went to work [for] the very competitor across town and asked the
ad agency how they set their TV budgets. You’ll never guess: They said they look
at the other Big 3 auto brand’s [past] TV spending and set it at that! Billions spent,
with the blind seemingly leading the blind.
Second, a few years after the sample period, two major US automakers went bankrupt:
Chrysler filed for bankruptcy in April 2009, followed by GM in June 2009, suggesting
possibly imperfect profit maximization. Third, a search of the trade press and interviews
with several automotive advertising executives indicated that automakers and dealers are
not able to link customer purchases directly to advertising exposures. We were told that
advertising effectiveness is only measured ex post as the total number of visitors to
dealers’ lots, suggesting it may be difficult to link weekly ad expenditures directly to
weekly fluctuations in consumer demand.
Since the available information call automaker advertising optimization into
question, modeling the supply side poses a risk of model specification bias. Instead, we
82
use the limited-information approach of Villas-Boas and Winer (1999). Lagged prices are
used as instruments as in Villas-Boas and Winer (1999). Following the standard two-
stage IV estimation approach, we first regress prices on brand intercepts and lagged
prices, and then use the predicted prices in our truck demand model presented in Chapter
1.
25
While the limited-information approach controls for correlations between
marketing variables and additive demand shocks, previous research has also identified a
concern about slope endogeneity, that is, possible correlation between
z
~
and marketing
instruments such as advertising. The idea is that manufacturers and dealers associations
may set advertising decisions based on unobserved changes in advertising responsiveness
(Kuksov and Villas-Boas 2008, Luan and Sudhir 2010). Slope endogeneity requires two
conditions. First, it must be that consumer responsiveness to advertising varies across
trucks or across time periods. The model includes truck-specific advertising response
parameters, so it explicitly allows for advertising response variation across trucks.
However, it does not allow those parameters to vary over time. The second condition is
that firms must know how this responsiveness changes over time and set advertising
expenditures accordingly.
If the second condition for slope endogeneity holds in our data, we should see
positive covariation across different automakers’ advertising expenditures as they
advertise more during periods that consumers are most responsive to advertising, and
advertise less in periods that consumers are less responsive to advertising. We calculated
25
Advertising expenditures are explicitly constructed as depending only on past expenditures, so it was not
necessary to use lags to instrument for them.
83
six-way correlations within each advertising variable to check whether such positive
covariation occurs in the data. The formula is
6
1
6
1
) 5 (
) (
6
j
A
t j
j jt
A
j
s T
A A
corr , (27)
where j indexes trucks, t indexes weeks, } , , { DPA MPA MBA A indexes types of
advertising, and
j
A and
j
A
s are the sample mean and standard deviation of advertising
spending of type A for truck j. The second condition for slope endogeneity suggests that
we should find 0 6
A
corr for at least some A. However, we find that 15 . 0 6
MBA
corr ,
01 . 0 6
MPA
corr , and 15 . 0 6
DPA
corr . Empirical support for slope endogeneity
therefore appears to be absent.
84
APPENDIX D: CROSS-PARTIAL DERIVATIVES OF MARKET SHARES
Ai and Norton (2003) show that correct inference of interaction terms in choice models
requires analysis of cross-partial derivatives rather than linear parameter estimates. In this
study, the main interest is to investigate the interaction effect of MBA and MPA and that
of MBA and DPA. This appendix derives the cross-partial derivatives of market shares
with respect to MBA and MPA and those with respect to MBA and DPA.
The market share of truck j in zip code z in week t is:
k
zkt z zkt kt kt kt zkt z t ks
zjt z zjt jt jt jt zjt z t js
zjt
YC DPA MPA MBA p f X
YC DPA MPA MBA p f X
s
) ) ; , , , , ( exp( 1
) ) ; , , , , ( exp(
. (28)
The partial derivative of
jzt
s with respect to MBA is,
) 1 (
) (
zjt zjt
jt jt
zjt
s s
MBA
f
MBA
s
. (29)
The cross-partial derivative of
jzt
s with respect to MBA and MPA is given by,
) 1 (
) 2 1 (
) ( ) (
) (
2
2
zjt zjt
zjt
jt jt
jt jt
jt jt
zjt
s s
s
MPA
f
MBA
f
MPA MBA
f
MPA MBA
s
. (30)
Given equation (3), we have
zjt
MBp
jt
BDP
j jt
BMP
j
MB
zj
jt
p DPA MPA
MBA
f
) (
, (31)
zjt
MPp
jt
PP
j jt
BMP
j
MP
zj
jt
p DPA MBA
MPA
f
) (
, (32)
85
BMP
j
jt jt
MPA MBA
f
) (
2
. (33)
Plugging (31)-(33) into (30), we’ll have the cross-partial derivatives of market share with
respect to MBA and MPA. Similarly, we can calculate the cross-partial derivatives of
market share with respect to MBA and DPA as follows:
) 1 (
) 2 1 (
) ( ) (
) (
2
2
zjt zjt
zjt
jt jt
jt jt
jt jt
zjt
s s
s
DPA
f
MBA
f
DPA MBA
f
DPA MBA
s
. (34)
86
APPENDIX E: ALTERNATE EXPLANATIONS FOR ESTIMATED
ADVERTISING INTERACTIONS
This section considers alternate explanations for the econometric results in Chapter 1:
- Advertising Targeting
- Transaction Characteristics
- Trade Promotions
- Price Advertisement Content
No strong evidence is found in favor of any of these mechanisms.
Appendix E: Advertising Targeting
Manufacturers and dealers associations may target their price advertisements to different
groups of consumers, and this targeting may overlap differentially with manufacturer
brand advertising messages. For example, if MBA and MPA overlap to a greater extent
than MBA and DPA, then negative interactions between MBA and MPA could be due to
advertising wearout.
Figures 11, 12, and 13 show how advertising expenditures for each type are
distributed across program genre, hour of the evening, and network affiliate. There is
surprisingly little difference in how manufacturers’ and dealers’ price advertising
messages are targeted.
87
Figure 11: Distribution of TV Advertising Expenditures over Program Genres
Figure 12: Distribution of TV Advertising Expenditures over Half-hours
0%
5%
10%
15%
20%
25%
30%
MBA MPA DPA
0%
5%
10%
15%
20%
25%
30%
7:00pm 7:30pm 8:00pm 8:30pm 9:00pm 9:30pm 10:00pm 10:30pm
MBA MPA DPA
88
Figure 13: Distribution of TV advertising expenditures over networks and affiliates
Appendix E: Transaction Characteristics
It could be that MPA and DPA appeal to different types of consumers, lead to sales of
trucks with different characteristics, or are associated with different transaction types or
price structures. The approach employed is to compare how each factor differs as the
intensities of each advertising variable change. For each of the three advertising variables
for a given truck, each week is classified as “High” or “Low” if the observed ad spending
for that variable is above or below the sample median, giving 2x2x2=8 different
classifications. If the variable in question (e.g., customer gender) is relatively constant
across all eight classifications, or if changes in the variable accompanied with high MPA
intensity are in the same direction as changes accompanied with high DPA intensity, it is
not considered to be a likely driver of the interaction results.
0%
5%
10%
15%
20%
25%
30%
ABC CBS FOX NBC
MBA MPA DPA
89
Table 13 presents the comparison for Ford F-series. The ratio of female truck
buyers to males holds nearly constant under each of the eight classifications, so it does
not seem to explain the interaction results. Similarly, major truck characteristics—model,
drive type, engine size and door types—increase in the same direction when moving from
the low-MPA/low-DPA condition to either the high-MPA/low-DPA or low-MPA/high-
DPA conditions.
Next, consider pricing terms. Customer rebates are higher in the low-MPA/high-
DPA condition than in the high-MPA/low-DPA condition, but this difference is roughly
offset by higher down payments. In general, there are not clear patterns of movements in
transaction types or pricing terms across advertising conditions.
Appendix E: Trade Promotions
Busse et al. (2006) showed that $1 in “customer cash” lowers prices more than $1 in
“dealer cash.” Customer cash directly enters the price variable, as shown in Appendix A,
but dealer cash does not enter the demand model, since consumers typically are not
informed about this trade promotion, and the dealer’s “discount” passed to the consumer
is implicitly included in the transaction price.
90
Table 13: F-series Transaction Characteristics across Advertising Intensity Levels
It stands to reason that dealer cash may motivate dealers to engage in extra price
advertising. This suggests a possible positive correlation between dealer price advertising
and the latent variable of “dealer cash.” We collected weekly data on dealer cash from
2002 to 2005. For four out of six trucks, the correlation coefficient between DPA and
dealer cash was not statistically significant. The two significant correlations were for F-
MBA Low Low Low Low High High High High
MPA Low Low High High Low Low High High
DPA Low High Low High Low High Low High
# of weeks 42 28 29 19 30 18 17 53
F-150 75% 76% 75% 74% 73% 70% 72% 72%
F-250 25% 24% 25% 26% 27% 30% 28% 28%
Female 17% 17% 18% 18% 17% 18% 19% 17%
Male 83% 83% 82% 82% 83% 82% 81% 83%
2WD 78% 73% 76% 74% 72% 70% 70% 71%
4WD 22% 27% 24% 26% 28% 30% 30% 29%
6-cylinder 74% 84% 79% 85% 85% 88% 91% 89%
8-cylinder 17% 11% 15% 10% 8% 6% 5% 6%
10-cylinder 8% 5% 6% 5% 7% 7% 4% 4%
4D Ext Cab 5% 15% 8% 14% 18% 22% 25% 21%
Crew Cab 50% 56% 52% 56% 56% 58% 57% 57%
Ext Cab 36% 21% 31% 23% 19% 13% 9% 14%
Regular Cab 9% 9% 9% 8% 8% 7% 8% 7%
Cash 14% 15% 13% 14% 13% 15% 13% 13%
Dealer Finance 72% 77% 73% 77% 82% 78% 81% 81%
Lease 14% 9% 15% 9% 6% 7% 6% 6%
APR 6.36 6.68 6.54 6.73 6.00 6.29 6.47 6.27
Rebate $1,379 $1,809 $1,451 $1,927 $1,525 $1,654 $2,209 $1,855
Monthly Payment $527 $531 $536 $531 $541 $551 $550 $544
Term (months) 56 61 56 60 60 62 61 62
Down Payment $5,737 $6,136 $5,892 $6,700 $6,134 $6,211 $7,274 $6,504
Residual $13,810 $14,182 $14,193 $15,203 $14,666 $15,563 $16,279 $14,532
Note: For each advertising variable, "High" indicates a weekly expenditure above the median.
Transaction
Type
Pricing
Terms
Drive Type
Cylinder
Type
Door Type
Gender
Model
91
series and Silverado, 0.21 and −0.18 respectively. This pattern of results does not support
unobserved trade promotions as a driver of the econometric results.
Appendix E: Pricing Advertisement Content
Because manufacturers and dealers typically use different agencies to produce their price
advertisements, differences in advertising content may potentially explain how each type
of price advertising interacts with MBA. To gain insight into this issue, three independent
coders were hired to analyze the price advertising content. The results are presented in
Table 14. None of the content measures, except the extent to which financial terms were
included in the ad, was significantly different at the 95% level. Thus it does not appear
that differences in price advertising content can explain the results.
Table 14: Price Advertising Content Analysis
MPA DPA
Likeable (1-5 scale) 2.8 2.6
Funny/humorous (1-5 scale) 1.5 1.4
Comparative Information (1-5 scale) 2.1 1.8
Product Price (MSRP) 55% 50%
Promotions (Cash back, rebate, discount, etc.) 82% 76%
Financial Terms (APR, monthly, down payment, etc.) 17% 31%
Average Time Price Info. is Displayed Onscreen (seconds) 3.4 3.2
Relative Font Size of Onscreen Price Information (1-5) 2.9 2.7
If Price Information is Repeated, Number of Repetitions 4.0 3.7
Is Price Presented as a Math Problem? (0=no, 1=yes) 52% 62%
92
APPENDIX F: POLICY FUNCTION DERIVATION
In this appendix, we show how to solve the Bellman equation. Plugging equation (13)
and (14) into (12), we’ll get:
,
) ( ) ( ) , | ( ) , (
) , (
max
) ( ) | ( ) , ( max ) , (
'
' '
j j
j
j
j
v
j j j j j
j j j
x
v
j j j j j
x
j j
v p p q v V
x
v p v W x V
(35)
Note that ) | (
j
v W is not a function of
j
x , so the F.O.C. of this Bellman equation is:
0
) (
) | ( 1
j
v
j
j
j
x
v p
v W . (36)
Given the assumption on the distribution of ) (
j
v p , we can get
2
) 1 (
) 1 (
j j
j
x x
v p
and
2
) 1 (
) 0 (
j j
j
x x
v p
. Plugging them into equation (36), we’ll get
0 ) | 0 ( ) | 1 (
) 1 (
1
2
j j
j
v W v W
x
. (37)
Rearranging equation (37) and considering the non-negativity constraint 0
j
x , we have
the analytical solution for the policy function
) | 0 ( ) | 1 ( 1
, 0 max ) , (
W W
x
j j j
if ) | 0 ( ) | 1 ( W W and 0 ) , (
j j j
x otherwise.
Abstract (if available)
Abstract
Quality is one of the most important factors that drive the market position of a product, which determines the success of the firm. Firms may be able to influence how consumers perceive their product quality through product design, advertising, promotion, pricing, and distribution channel. ❧ This dissertation first investigates how price advertising by different channel members would affect consumer quality perceptions. It suggests that manufacturer price advertising leads to lower perceived product quality than price advertising by dealers, and such effect has significant impact on total channel profits. This implies that manufacturers and their retailers might benefit from coordinating their price advertising messages and could consider to pulsing their brand advertising and price advertising simultaneously. ❧ The second chapter of this dissertation examines the effects of competition and product entry on firms’ investment in perceived product quality. It shows that competition has a greater impact on quality laggard’s investment, and that a new product entry might cause a more serious threat to quality laggards and could drive them out of business. It also demonstrates that the recently developed dynamic structural models could be used to predict competitive reactions in product quality and could improve the prediction accuracy comparing with a reduced-form descriptive model.
Linked assets
University of Southern California Dissertations and Theses
Conceptually similar
PDF
Essays on the competition between new and used durable products
PDF
Essays on commercial media and advertising
PDF
Essays on product trade-ins: Implications for consumer demand and retailer behavior
PDF
Marketing strategies with superior information on consumer preferences
PDF
Essays on the luxury fashion market
PDF
Essays on the role of entry strategy and quality strategy in market and consumer response
PDF
Essays in retail management
PDF
Essays on consumer product evaluation and online shopping intermediaries
PDF
The essays on the optimal information revelation, and multi-stop shopping
PDF
Price competition among firms with a secondary source of revenue
PDF
Managing product variety in a competitive environment
PDF
Manipulating consumer opinion on social media using trolls and influencers
PDF
Essays on digital platforms
PDF
Three essays on industrial organization
PDF
Essays on dynamic control, queueing and pricing
PDF
Accrual quality and expected returns: the importance of controlling for cash flow shocks
PDF
Essays on firm investment, innovation and productivity
PDF
The impacts of manufacturers' direct channels on competitive supply chains
PDF
Three essays on agent’s strategic behavior on online trading market
PDF
Understanding virality of YouTube video ads: dynamics, drivers, and effects
Asset Metadata
Creator
Xu, Linli
(author)
Core Title
Quality investment and advertising: an empirical analysis of the auto industry
School
Marshall School of Business
Degree
Doctor of Philosophy
Degree Program
Business Administration
Publication Date
04/27/2012
Defense Date
03/22/2012
Publisher
University of Southern California
(original),
University of Southern California. Libraries
(digital)
Tag
advertising,automotive industry,choice model,dynamic structural models,empirical industrial organization,OAI-PMH Harvest,quality investment
Language
English
Contributor
Electronically uploaded by the author
(provenance)
Advisor
Siddarth, Sivaramakrishnan (
committee chair
), Wilbur, Kenneth C. (
committee chair
), Che, Hai (
committee member
), Dukes, Anthony (
committee member
), Luo, Lan (
committee member
), Tan, Guofu (
committee member
)
Creator Email
linlixu@usc.edu,linlixu2um@gmail.com
Permanent Link (DOI)
https://doi.org/10.25549/usctheses-c3-15541
Unique identifier
UC11288363
Identifier
usctheses-c3-15541 (legacy record id)
Legacy Identifier
etd-XuLinli-669.pdf
Dmrecord
15541
Document Type
Dissertation
Rights
Xu, Linli
Type
texts
Source
University of Southern California
(contributing entity),
University of Southern California Dissertations and Theses
(collection)
Access Conditions
The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...
Repository Name
University of Southern California Digital Library
Repository Location
USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA
Tags
automotive industry
choice model
dynamic structural models
empirical industrial organization
quality investment