University of Southern California Dissertations and Theses

Empirical essays on industrial organization

(USC Thesis Other)

Empirical essays on industrial organization

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content EMPIRICAL ESSAYS ON INDUSTRIAL ORGANIZATION
by
Yusun Hwang
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
(ECONOMICS)
August 2014
Copyright 2014 Yusun Hwang
Dedication
Tomybelovedwife,Mokyean
ii
Table of Contents
Dedication ii
List of Tables iv
List of Figures v
Abstract vi
Chapter 1: Vertical Integration in the Korean Movie Industry 1
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Korean Movie Industry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.3 A simple model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.4 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.5 Empirical Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
1.6 Robustness check . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
1.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
Chapter 2: Influence of Word-of-Mouth on Consumer Demand : Evidence from
Movie Industry 46
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
2.2 Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
2.3 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
2.4 Empirical Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
2.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
Appendix 77
iii
List of Tables
1.1 Market shares of distributors in 20062009 . . . . . . . . . . . . . . . . . 10
1.2 Number of Theaters and Screens by Exhibitor . . . . . . . . . . . . . . . . 10
1.3 Theater Characteristics by Integration Status . . . . . . . . . . . . . . . . 33
1.4 Movie Characteristics by Integration Status . . . . . . . . . . . . . . . . . 34
1.5 Screening times at the Opening Weekend (Saturday) (CJ-CGV) . . . . . . 35
1.6 Screening times (Lotte-Lotte Cinema) . . . . . . . . . . . . . . . . . . . . 35
1.7 The Effects of Vertical Integration on Screening Times: Full sample . . . 36
1.8 The Effects of Vertical Integration on Screening Times: Restricted sample 37
1.9 Total Days of Movie Running (CJ-CGV) . . . . . . . . . . . . . . . . . . . 38
1.10 Total Days of Movie Running (Lotte-Lottecinema) . . . . . . . . . . . . . 38
1.11 The Effects of Vertical Integration on the Length of Movie Run: Full sample 39
1.12 The Effects of Vertical Integration on Movie Stopping Decision using Probit 40
1.13 The Effects of Vertical Integration on Film Choice Decision using Logit . 41
1.14 Integration Effects on Screenings Times at the Opening week . . . . . . . 42
1.15 Change in screening times before and after disintegration . . . . . . . . . 43
1.16 Change in total days of movie run before and after disintegration . . . . 44
1.17 Disintegration ofShowbox andMegabox . . . . . . . . . . . . . . . . . . . 45
2.1 Descriptive Statistics for Variables related to Word-of-Mouth . . . . . . . 58
2.2 Summary statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
2.3 Regression of Log Weekly Box-office Revenues . . . . . . . . . . . . . . . 66
2.4 Regression of Log Weekly Box-office Revenues: movie fixed effects . . . 69
2.5 Who initiate WOM transters? . . . . . . . . . . . . . . . . . . . . . . . . . 70
2.6 The effect of controlling for the number of reviews . . . . . . . . . . . . . 71
A1 The Effects of Vertical Integration on the Length of Movie Run: Restricted
sample . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
A2 The Effects of Vertical Integration on Movie Stopping Decision using OLS 79
A3 The Effects of Vertical Integration on Film Choice Decision using OLS . . 80
iv
List of Figures
1.1 How box-office revenue is split . . . . . . . . . . . . . . . . . . . . . . . . 8
1.2 Effects of Vertical Integration on Screen Allocation . . . . . . . . . . . . . 13
1.3 Seasonality in the Korean Movie Industry . . . . . . . . . . . . . . . . . . 15
2.1 Weekly box-office revenues with the effect of word-of-mouth . . . . . . . 55
2.2 Histogram ofexante Expectation . . . . . . . . . . . . . . . . . . . . . . . 58
2.3 Gross box-office vs Volume of user ratings . . . . . . . . . . . . . . . . . . 59
2.4 Correlation between Box-office and Word-of-Mouth . . . . . . . . . . . . 60
2.5 Correlation between Box-office and User ratings by week . . . . . . . . . 61
2.6 Dynamics of Box office by satisfaction rate . . . . . . . . . . . . . . . . . . 62
2.7 Hit vs Flop . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
v
Abstract
This dissertation estimates the effects of vertical integration and word-of-mouth on
product sales using the data from the Korean movie industry. In chapter 1, I examine
exhibition behavior of movie theaters in the Korean movie industry in order to inves-
tigate the influence of vertical integration on competition. I focus specifically on the
choice of films, screen allocation, and movie run stopping over different vertical struc-
tures. Because, in the Korean movie industry, not only can we observe the same movie
being shown in both integrated theaters and unintegrated theaters but also observe the
same theater showing movies from distributors of different vertical structures, I use
movie and theater fixed effects to control for the unobserved quality of movies and the-
aters. The empirical results suggest that vertically integrated theaters are more likely to
choose their affiliated movies than other competing movies, and they choose them more
often than other competing theaters do. In addition, integrated theaters give their own
movies a greater number of screenings over longer time periods. This effect is mostly
restricted to company operated theaters, and it is greater when movies are expected to
get positive word-of-mouth as well as when underlying demand is high such as hol-
idays. I argue that these results are not driven by the matching between movie and
theater based on anything other than integration status, and that vertical integration
leads to the foreclosure, denial of access, of independent distributors to integrated the-
aters, to the detriment of consumers.
Chapter 2 analyzes the effects of word-of-mouth on box-office sales, another impor-
tant feature in movie industry. In general, consumers often rely on the information from
vi
their peers and other sources like internet sites when product quality is uncertain before
its purchase so that word-of-mouth is believed to be one of the most influential factors
in product sales. In this chapter, I quantify the effect of word-of-mouth on weekly box-
office revenues in the context of the Korean movie industry. Using online user ratings
and reviews as well as a new dataset including pre-expectation rate and satisfaction
rate for each movie, I find strong evidence that word-of-mouth is significant in movie
business. My estimates imply that word-of-mouth explains 24% of total box-office sales
and 68% of sales from the second week when we compare movies at 75th percentile of
satisfaction rate to movies at 25th percentile, assuming that they have the same level of
pre-expectation.
vii
Chapter 1
Vertical Integration in the Korean
Movie Industry
1.1 Introduction
The possibility of vertical foreclosure in vertical mergers has been a major concern
of antitrust authority investigations. Theories suggest that for the purpose of gaining
monopoly power, vertically integrated firms may deny an access of competing down-
stream (upstream) firms to intermediate goods (downstream outlets). They also sug-
gest that vertical foreclosure can survive as an equilibrium in an oligopoly setting,
indicating that vertical integration can harm consumer welfare by raising price of final
goods.(Ordover et al. (1990), Salinger (1988), Hart et al. (1990), Choi and Yi (2000), Chen
(2001))
Empirical studies have provided evidence that vertical integration gives rise to fore-
closure. For example, Ford and Jackson (1997), Waterman and Weiss (1996), and Chipty
(2001) found that vertically integrated cable operators in the cable television indus-
try were more likely to carry their affiliated networks. In particular, Chipty (2001)
demonstrated that Time Warner, which owns the premium movie service, HBO, tends
to exclude AMC, the basic movie service, from its basic package offer. In addition,
Goolsbee (2007) found that broadcast networks are more likely to carry their own shows
than independent programming. Regarding to movie industry, Gil (2008) and Fu (2009)
examined the effect of vertical integration between distributors and exhibitors in the
Spanish and in the Singapore movie industry respectively. Both studies found that
1
vertically integrated theaters showed their affiliated movies longer than unintegrated
theaters did.
However, the existence of vertical foreclosure is, by itself, not sufficient to allow for
the conclusion that vertical integration harms consumers. In fact, in economics, the
effects of vertical integration on consumer welfare have long been a source of debate.
Theories predict (Ordover et al. (1990), Salinger (1988)) that vertical integration may
have efficiency-enhancing effects by reducing transaction costs or eliminating succes-
sive monopoly mark-ups, and as a result, vertical integration can improve consumer
welfare by lowering prices. Hence, the welfare effect of vertical integration depends on
the relative importance of anti-competitive effect of vertical foreclosure and efficiency
gains.
A few studies have attempted to assess the consequences of a vertical merger, pro-
viding mixed results.
1
Goolsbee (2007) found that broadcast networks apply lower
standards to carrying their own shows than to carrying independent programming.
Specifically, independent programs need to generate over 15 percent higher revenues
from advertising than comparable in-house programs in order to get on the air, sug-
gesting that the foreclosure effect outweighs efficiency gains. On the other hands,
Chipty (2001), in her paper on the cable TV industry, concluded that vertical integra-
tion does not harm but rather benefits consumers. By comparing consumer welfare
across integrated and unintegrated markets, she argued that efficiency gains dominate
losses from foreclosure. Corts (2001) studied how vertical integration in the movie
industry between producers and distributors affected competition of movie release-
date scheduling. He demonstrated that integrated firms internalize the negative exter-
nality of close release dates, indicating that vertical integration improves the efficiency.
Gil (2008), also, interpreted his findings as efficiency gains in his investigation of ver-
tical integration between movie distributors and exhibitors. He argued that integrated
theaters run their own movies longer than other movies, and longer than unintegrated
1
For a survey of empirical studies, refer to Lafontaine and Slade (2007) and Rey and Tirole (2007)
2
theaters do, and concluded that vertical integration solves the distortion of movie run
length created by the revenue sharing contracts in the movie industry. However, this
kind of interpretation should be made with caution because theaters face capacity con-
straints caused by having a limited number of screens. To retain their own movies for a
longer period of time, integrated theaters should sacrifice revenues generated by other
movies that otherwise would have been shown, which could be interpreted as a reduc-
tion in total box-office revenues as well as consumer welfare.
A major obstacle in assessing the effects of vertical integration is that we hardly
notice that companies with different organizational forms handle the same set of prod-
ucts from both integrated firms and independent firms in the same market. When each
product is a differentiated good, which holds in many industries, controlling product
quality is crucial to demand estimation, but observables often explain little about prod-
uct quality. If we can observe that downstream firms do business with the same set of
products, we might attribute observed difference between integrated downstream and
unintegrated downstream to the effects of vertical integration by controlling product
quality. This is the case in the cable TV industry in which cable TV providers offer
different sets of channels chosen from the same set of channels available. However,
the cable TV industry is virtually monopolized in many markets. Several cable TV
providers operate nationwide, but it is common that a specific provider is the only
option that consumers can choose in their residential area. In that case, the comparison
of integrated markets to unintegrated markets could suffer from differences in underly-
ing demand over markets in the assessment of the consequences of vertical integration.
In this spirit, the Korean movie industry provides several advantages for the anal-
ysis of vertical integration. First, because two major domestic distributors own multi-
plex chains, it is possible to observe how integrated theaters give preferential treatment
to their own movies against other movies supplied by rival distributors compared to
unintegrated theaters. That is, we can observe four different combinations between
movies and theaters: (1) integrated movies shown in integrated theaters, (2) integrated
3
movies shown in unintegrated theaters, (3) unintegrated movies shown in integrated
theaters, and (4) unintegrated movies shown in unintegrated theaters. This circum-
stance enables us to control unobserved movie quality
2
as well as theater quality by
using movie-theater fixed effects. Theater fixed effects also control the difference in
underlying demand for movies over geographical markets. With these fixed effect, the
effects of vertical integration are determined by difference in differences approach in
the level of movie by theater.
Second, it is distributors and not movie theaters that promote movies nation-
wide, suggesting that theaters in the same geographical market face the homogeneous
demand for each movie. Although each theater might enjoy some degree of market
power because of its membership programs, it is difficult to conclude that potential
consumers at an integrated theater have strong preference for movies from its affiliated
distributor. Moviegoers are usually concerned about the contents they can see such as
trailers, casting, and directors, but not about what is happening behind the film like
which company distributes the film. In addition, movie theaters are located close to
each other in many markets, especially in urban areas in Korea. In the extreme case,
two different theaters are operating within 100-meter distance. It is hard to believe that
integrated theaters draw different sets of consumers based on their preference. Hence,
the observed differences in exhibition behavior between integrated and unintegrated
theaters can be attributed to the practice of discrimination by means of vertical inte-
gration under the assumption that in-house promotion by theaters has little impact on
movie demand.
3
Third, contracts between distributors and exhibitors are fairly standard and simple
in the Korean movie industry, contrary to the U.S. movie industry where contracts vary
with movies and theaters. Distributors and exhibitors make revenue sharing contract
for each movie, splitting the box office revenues that the movie earns. Each party’s
2
In this paper,moviequality means a movie’s box office appeal, not an artistic quality.
3
Survey shows that most moviegoers choose what movie to see before going to a movie theater.
4
share, however, does not vary across movies or during the weeks after the release with
a few exceptions. Therefore, this analysis does not suffer from a lack of data avail-
ability on contracts, which are often problematic in many other studies. If contracts
between distributors and exhibitors do vary, then the observed differences should not
be interpreted as the effects of vertical integration because the discrepancy in contracts
generates different incentives for theaters to each movie. With virtually no variations in
contracts, vertical integration between distributor and exhibitor does not generate cost
asymmetry between theaters, which are often considered as a main force in generating
the efficiency gains of vertical integration.
Lastly, movie ticket prices are quite uniform. Ticket prices are higher on the week-
end and lower for the matinee, but do not vary with movies and theaters.
4
With ticket
prices being fixed, total consumer welfare depends only on the total number in atten-
dance. As the capacity of every theater is constrained by a limited number of screens,
integrated theaters should reduce screens devoted to other movies in order to show
their movies more and for longer periods. Hence, there is no gain in consumer wel-
fare from vertical integration unless the increase of box office revenues from integrated
movies outweighs what rival movies would have earned.
With the consideration of these facts, this article examines the effects on competition
and market performance of vertical integration between distributors and exhibitors in
the Korean movie industry. I use the data for movies released from 2006 to 2008 in the
Korea movie industry where integrated firms are major players both in the distribution
sector and in the exhibition sector. Some anecdotes in this industry suggest that access
to integrated theaters may be restricted to some degree for independent distributors
when integrated firms have their own movies to show. In this paper, I focus on three
different aspects of the exhibition behavior of movie theaters that are crucial to box
4
Second-run theaters charge a lower price, but their market share is negligible, so I include only first-
run theaters in my sample. The prevalence of uniform pricing in the movie industry is, in fact, somewhat
puzzling. Orbach and Einav (2007) documented rationales for uniform pricing in the movie industry
5
office revenues: the allocation of screens (or screening times), the decision to stop movie
run, and film choice decisions.
My estimates suggest that integrated theaters discriminate against competing dis-
tributors in favor of their own movies in all three different aspects of exhibition prac-
tice. First, integrated theaters allow approximately two more additional screenings for
their own movies than for other competing movies, and than other unintegrated the-
aters do. This effect is mostly restricted to company operated theaters, and it is greater
when movies are expected to have positive word-of-mouth as well as when underlying
demand is high such as holidays. Second, integrated theaters are less likely to drop
their own movies than comparable movies from competing distributors. Integrated
theaters are also more likely to choose their own movies, suggesting that independent
distributors are partially foreclosed in the theatrical exhibition market. These results
indicate that vertically integrated theaters apply lower standards to their own movies,
which implies that vertical integration between distributor and exhibitor reduces total
box office sales, as a result, harming consumers.
This paper proceeds as follows. In Section 2 and 3, I provide an overview of the
Korean movie industry and a simple model to provide testable implications. In Section
4 and 5, I describe the data and the empirical findings on three different aspects of the
exhibition practice of movie theaters. In section 6, robustness check is provided and I
conclude the paper with the discussion on welfare implications in section 7.
1.2 Korean Movie Industry
Movie industry consists of three sectors such as production, distribution and exhi-
bition. Distributors supply films to exhibitors (theaters) and ancillary windows such
as DVD, cable, broadcast TV market and so on. Main decisions of distributors include
scheduling the release timing of movies to theatrical window as well as ancillary win-
dows, acquiring enough screens through negotiating with theaters, and promoting
6
movies nationwide. In the Korean movie industry, distributors do not have their own
production companies, rather they play an role as main stake holders in production
stage. Exhibitors maximize box office revenues from movies and revenues from other
sources such as concession sales. Since new movies are released almost every week
throughout the entire year, exhibitors must make decisions regarding the replacement
of movies playing in their screens every week. Theaters also promote movies through
in-theater advertisements, but its impact is limited because most of consumers make
decisions of what movie to watch before coming to theaters. Distributors and exhibitors
use revenue sharing contracts which specify the weekly share of box office revenues that
each party takes. Contracts generally do not specify either requirement days of movie
run or which screen the movie should be screened on, although theaters usually have
different size of screens in terms of the number of seats.
Korean movie industry is one of a few markets in which domestic movies are fairly
competitive against Hollywood movies. The market share of domestic movies fluctuate
across years, but they usually enjoy around 50% market share against foreign movies
including mainly Hollywood movies.
5
Market share of each distributor also fluctuates
across years. Three domestic distributors such as CJ Entertainment(hereafter CJ), Show-
box, andLotteEntertainment(hereafterLotte), and subsidiaries of Hollywood studios are
dominant players in distribution sector in Korea. CJ and Lotte own their multiplex
chains such as CGV and Lotte Cinema respectively. Another major domestic distributor
Showbox had its own theater chain,Megabox, but it was disintegrated on July, 2007.
6
No
subsidiaries of Hollywood major studios have their own theaters in Korea.
While declining revenue-sharing term is common in US movie industry, the Korean
movie industry observes the fairly fixed revenue sharing rate. Distributors take 50%
of total box office revenues for domestic movies and 60% for foreign movies regardless
5
In 2006, market share of domestic movies is 94% in India, 53.2% in Japan and 63.8% in Korea.
6
Showbox and Megabox are defined as integrated firms before disintegration and as independent firms
afterwards in the analysis
7
of expected demand each movie has.
7
This is an important feature which enables us
to attribute observed variation in exhibition practice between integrated theaters and
unintegrated theaters to the effects of vertical integration. If contracts vary over movie
by theater, there is no reason to believe that comparable theaters should make the same
decision regarding to film choice, screen allocation, and stopping movie run even when
market observes no integrated firms.
Figure 1.1: How box-office revenue is split
Figure 1.1 describes how box office revenues are split into players involved. For
example, suppose that a movie generates 100 million dollars as its total box-office rev-
enues. Movie theaters take 50 millions as its share according to revenue sharing con-
tract. For the rest of 50 millions, the distributor recoups its costs of prints and advertis-
ing as well as distribution fee
8
first, and the production company takes its production
7
50/50 split rule is applied to theaters located in regions other than Seoul for both domestic movies and
foreign movies.
8
Distribution fee is on average around 10% of total box-office revenues.
8
costs before residual holders claim their shares. Distributors generally play as major
stake holders, although their shares differ across movies.
Tables 1.1 and 1.2 show the overview of the Korean movie industry in each sector.
9
An integrated distributor, CJ is the leading company in distribution sector, account-
ing for around 30% of market share, while another integrated distributor, Lotte, is the
third player among domestic distributors in terms of market share. Lotte is a young
distributor which started its business in movie distribution in 2003. Table 2 shows that
distributor-owned theaters, CGV and Lotte Cinema were growing fast, accounting for
almost 50% of screens in theatrical exhibtion market in 2008. It is worthwhile noting
that vertical relationship is somewhat different in these two integrated firms. Lotte and
Lotte Cinema belong to the same corporate entity under the same CEO, and it is very
flexible in the transfer of personnel between two departments. Meanwhile,CJ andCGV
are subsidiary companies of the conglomerate, CJ Corporation, and there is no transfer
of workers between two companies. Therefore, the vertical relationship would be much
stronger between Lotte and Lotte Cinema than CJ-CGV. In fact, regression results below
suggest that the effects of vertical integration are larger for Lotte-Lotte Cinema than CJ-
CGV.
Given the fact that each movie is a differentiated product and the multiplex is a
dominant form of theaters in Korea, complete foreclosure by integrated firms is hardly
observed, while partial foreclosure is reported to be present. Market practitioners often
complain of unfair treatments by vertically integrated firms. For instance, independent
distributors insist that they have a difficulty in acquiring sufficient number of screens
because integrated theaters show their affiliated movies aggressively, providing less
screens to movies of competing distributors in order to protect their distributors’ profits.
9
A multiplex is generally defined as a theater having more than 5 screens. Multiplexes account for
around 90% of screens in 2008
9
Table 1.1: Market shares of distributors in 20062009
Market share (%)
Distributor 2006 2007 2008 2009
CJ Entertainment 23.2 29.7 30.1 29.1
Showbox 20.1 12.3 10.2 15.2
Lotte Entertainment 5.6 8.6 8.3 11.8
UPI Korea 7.6 3.5 10.0 2.1
Sony Pictures Releasing Buena Vista Korea 10.0 9.8 6.8 8.5
Warner Bros. Korea 5.8 11.3 6.1 5.6
20th Century Fox Korea 5.9 5.6 5.1 7.8
Others 21.8 19.2 23.4 19.9
Total 100.0 100.0 100.0 100.0
Table 1.2: Number of Theaters and Screens by Exhibitor
N of theaters N of screens
Exhibitor 2006 2007 2008 2006 2007 2008
Multiplex
CGV 44 57 63 351 461 511
Primus 33 38 36 226 276 259
Lotte Cinema 36 41 47 273 316 360
Megabox 20 15 13 166 123 116
Cinus 15 23 25 109 160 178
Others 50 36 39 437 344 352
Total 198 210 223 1562 1680 1776
Non-Multiplex 123 104 86 318 295 228
Total 321 314 309 1880 1975 2004
Since admission price is constant across movies within a theater, a main strategy
for distributors to maximize their profits is to show their movies longer in as many
of screens as they can. Besides movie quality, the number of screens and the length
of movie run also affect the overall box office. Getting more screens is important espe-
cially in the first week of its release when the information about movie quality is not yet
fully revealed and the highest revenues, around 40% of overall box office, are usually
achieved. Since 35 movies are released every week, theaters have to make decisions
about what movies to show and to drop. It is hard to believe that vertically integrated
10
theaters have the same pattern with what independent theaters do because indepen-
dent theaters only consider their own profits but vertically integrated theaters take into
account of joint profits with affiliated distributor.
1.3 A simple model
A theater needs to choose what movies to show and to replace with the limited
‘shelf space’ every week. For the tractability of model, I assume that the number of
screens allocated to each distributor (or movie) is a continuous variable chosen by the
theater. In this model, the theater maximizes its current profits, ignoring dynamics of
box-office revenues over the life of movie run. This setting can be justified by the fact
that a typical contract between distributor and exhibitor does not specify the minimum
length of movie run.
10
Consider the case where a multiplex chooses the combination of movies from two
distributors such asD
A
andD
B
under the capacity constraint withN screens.
=(p
) [
A

A
(N
A
)N
A
+
B

B
(N
B
)N
B
]
s:t N
A
+N
B
=N
: profit of the theater

i
: theater’s share of box-office revenue fromD
i
’s movies (i =A;B)
p : ticket price (fixed)

: marginal cost of theater (fixed)

i
: admissions per screen allocated toD
i
’s movie
10
In the US movie industry, minimum length of movie run is usually specified in the contract. This
difference may reflect the fact that distributors do not have strong bargaining power over movie theaters
in Korea while big studios - distributors - are dominant players in US.
11
N
i
: number of screens to showD
i
’s movie

i
is assumed to be a decreasing function of N
i
, implying that additional screen
dedicated to the same distributor generates less demand because it would be a less
popular movie. The optimal decision of film choice would be determined at

A

@
A
@N
A
N
A
+
A

=
B

@
B
@N
B
N
B
+
B

This condition implies that marginal revenues from showing the least popular
movie from each distributor would be equal at the optimal decision of the multiplex.
However, when the multiplex is merged with D
A
, vertical merger takes into account
revenues from both its distributor sector and its exhibitor sector, having an increased
share with
A
>
B
. Due to the increase in perceived marginal revenues from its own
movies, the multiplex has an incentive to choose its own movies more frequently than
D
B
’s movies compared to what it would be under independent ownership. As a result,
industry profits andD
B
’s profits decrease while vertical merger has increased its prof-
its.
In the previous literature, cost asymmetry through vertical integration and commit-
ment problem play a critical role to drive the results. However, with the fixed ratio
in revenue sharing contract across movies, vertical integration does not generate cost
advantage to movie theaters in acquiring movies. As screening schedule of each the-
ater is open to the public, there is no commitment problem in contracts between dis-
tributor and exhibitor either. With fixed price across movies and theaters, consumer
welfare depends only on the total number of admissions, implying that vertical merger
reduces consumer welfare in this setting. Vertical merger does not want to foreclosure
completely D
B
, but D
B
’s ability to reach customers would be limited by the vertical
merger.
The effect of vertical integration on the decision of screen allocation and stopping
movie run can be analyzed in the same manner. For example,N
i
can be thought to be
12
the number of screens to show theD
i
’s movie if each distributor has one movie. It is
common for a multiplex to show a movie on several screens when it is expected to have
higher demand like in the case of blockbusters. Additional screen dedicated to the same
movie can increase the total admission because it provides more frequent screening
times to attract moviegoers facing time constraints, but its additional increase would
decrease as the number of screens increases. As in the case of film choice, the increase
in perceived marginal revenues makes integrated theaters allocate more screens to their
own movies. This result is depicted in figure 1.2 whereN
A
0 is the number of screens
devoted toD
A
’s movie when the theater is integrated withD
A
.
Number of screens
MRA
NA
MRB
NB
MRA ’
NA ’
Figure 1.2: Effects of Vertical Integration on Screen Allocation
In summary, this model gives three testable implications about the effects of vertical
integration: (1) Integrated theaters show their own movies longer than other movies
13
and than other unintegrated theaters do. (2) Integrated theaters allocate more screens
to their own movies. (3) Integrated theaters are less likely to drop their own movies.
Externality through Word-of-Mouth
Movie, as a product in theatrical window, is stylized as having a short life, reaching
at its peak mostly at the opening weekend and declining over the rest of its life in terms
of its box-office revenues. It is rare that weekly box-office revenues are higher in the
second or third week than the first week, but some movies show relatively slow decay
rates as positive word-of-mouth enhances demand in subsequent weeks. This is not a
’buzz’ effect generated by marketing efforts or by other movie characteristics which are
known before its release such as star casting, director, sequel, and trailer etc. It is rather
a process of social learning under which potential consumers update their beliefs on
movie quality through the information from who experienced the movie.
Industry executives in movie business consider word-of-mouth as one of major
driving forces for the success of movie. In a companion paper, I showed that word-of-
mouth has a significant and considerable impact on box-office revenues. In the presence
of word-of-mouth, the return to attracting a consumer is greater than the direct effect -
ticket price - that the consumer has on box-office revenues because it may increase the
demand of other potential consumers. Since word-of-mouth can prevail through online
user ratings and reviews, this social multiplier effect can be even bigger than what is
measured within the geographical area. While independent theaters do not consider
this spillover effect as well as the multiplier effect, integrated theaters can internalize
the benefits from these effects nationwide that their own movies can generate. When
early box-office revenues are expected to boost positive word-of-mouth, integrated the-
aters have an incentive to sacrifice current revenues on behalf of future sales by allowing
more screens to their own movies. Hence, we should observe bigger impact of vertical
integration for movies with higher quality.
14
Seasonality and Competition
Movie industry observes a clear seasonality. Underlying demand for movies is rel-
atively high in holidays and school vacations as shown in Figure 1.3.
11
Figure 1.3: Seasonality in the Korean Movie Industry
In order to capture high demand, distributors tend to release their biggest hits in
peak seasons, resulting in an amplified seasonality in box office revenues as well as an
intense competition among strong movies.
12
It is not obvious how this intense competition and high underlying demand influ-
ence the effects of vertical integration on exhibition behavior of movie theaters. With the
fixed number of screens available, competing theaters would not allow many screens
to integrated distributors so that one might expect that integrated theaters have strong
11
Einav (2007) decomposed observed pattern of box-office revenues into underlying demand and ampli-
fied effects. He found that underlying demand has a similar pattern of observed one, but movies were too
clustered in high seasons compared to underlying demand, suggesting that studios could increase rev-
enues by adjusting release timing of their movies more into off-peak seasons.
12
Corts (2001) argued that vertical integration between production and distribution sector help movies
not being clustered too much although it does not achieve the optimal level.
15
incentives to show their own movies when competition is intense. However, integrated
theaters also face increased opportunity costs of showing their own movies because
of the presence of other good quality movies from competitors. On the other hands,
integrated firms have strong incentives to boost early box-office revenues by allowing
more screens to their own movies when positive word-of-mouth is expected. Higher
underlying demand can amplify multiplier effects, inducing integrated theaters to show
their own movies with an aggressive manner. Increased competition with good-quality
movies from other distributors, however, attenuates the possibility to top competitors
and capture higher demand, diminishing the incentive to sacrifice current revenues of
theaters for expected revenues from subsequent weeks. Therefore, the effects of verti-
cal integration can be either higher or lower depending on relative importance of each
incentive.
Company-operated theaters vs Dealer-run theaters
There are two different types of integrated theaters, company-operated theaters and
dealer-run theaters. Film choice decision is still under the control of parent company
which assigns their staffs to dealer-run theaters as theater managers. Hence, it is hard to
say that dealer-run theaters are independent from parent companies. However, because
the parent company takes a fixed share of ’total’ box-office revenues from dealer-run
theaters, dealer-run theaters have less incentives to favor movies from their affiliated
partners than company-operated theaters.
Regarding to differential effects of vertical integration between dealer-run and
company-op downstream firms, Hastings (2004) examined retail gasoline market in
Southern California. She found that vertical integration caused the increase in retail
gas price and this effect was not confined to company-op stations, concluding that it is
consistent to a model of differentiated products with consumer brand royalty. In other
16
words, it is the consumer brand royalty, not the vertical integration that derives the
surge of retail gas price in her paper.
In movie industry, it is also possible that some of consumers have a brand royalty
in favor of a distributor as well as a multiplex chain. Every multiplex chain encourages
consumers to be enrolled with its own reward program which makes program mem-
bers be locked in. It is also possible that some of consumers may prefer movies from a
specific distributor, if that distributor has good reputation on specific genre (e.g. Dis-
ney). However, for the brand loyalty to explain the favor of integrated theaters to their
own movies, a group of consumers should have brand loyalty for both sectors of the
integrated firm. That is, those who prefer Lotte Cinema should prefer movies of Lotte
Entertainment. This is not the case, I think, here because moviegoers are going to make
their movie-going decisions by what is presented in the marketing, not by what’s hap-
pened behind the camera. It is not likely that consumer brand loyalty plays a role in
observed differences between integrated theaters and independent ones in exhibition
behavior. Therefore, I expect that the effects of vertical integration would be stronger
for company operated theaters.
1.4 Data
The data comes from three different sources in this paper. First, I collected daily
screening records of theaters during 20062008 in the Korean movie industry from
Korea Film Council (KOFIC), a governmental agency. KOFIC collects daily screen-
ing records from registered theaters which consist of over 95% of theaters in Korea.
This data enables us to examine what movies were shown in each theater, how long
each movie was shown in a specific theater, and how many screens were allocated to
each movie within a theater. This paper focuses on first-run theaters and feature films
released nationwide. The full sample includes 590 movies and 250 theaters, counting
to 83419 observations of combinations between movies and theaters. For each movie, I
17
compiled information on its characteristics including its distributor, nationality, rating,
average score of online user ratings, and total box office revenues. I also matched infor-
mation on the number of screens, location and the identity of multiplex chain to each
theater.
Second, I use the unique data from a movie research marketing company which sur-
veys about expectation on coming movies as well as satisfaction rate on movies people
watched. In particular, the data provides the number of people who are aware of a
movie coming to be released, the number of people who answered that they would go
to see the movie in a theater, the number of people who are satisfied with a movie.
13
This data is available for movies released since 2008, so the sample of the analysis using
this data will be restricted to movies released in 2008 in this paper.
Weekly box office revenues each movie generates would be the most important fac-
tor to the decision of movie run stopping. With the help of two major distributors, I
collected the data of weekly box office revenues of each movie from these distributors
in every theater in 2008. Since one distributor is an integrated firm and the other uninte-
grated, I can still examine the effects of vertical integration on movie stopping decision
in the sense of difference in differences approach.
Descriptive statistics for theaters by integration status are presented in Table 1.3.
Both integrated theaters and independent theaters have statistically the same number
of screens and seats, and the same average days of movie running. Interestingly, inte-
grated theaters not only show larger number of movies, but also allocate more screens
to newly released movies than independent theaters do. Integrated theaters are also
less likely to show domestic movies than unintegrated theaters. This difference sug-
gests that integrated theaters are systematically different from unintegrated theaters
although they have similar characteristics in terms of size, highlighting the need to con-
trol theater quality in the analysis.
13
The sample of the data includes around 2,000 individuals every week, and the sample is replaced
every 6 months. The survey is made through online.
18
Table 1.4 indicates that a clear distinction exists between movies over integration
status. Movies of integrated distributors are shown longer in more theaters with more
screens. Average performance of movies by integrated firms overwhelms unintegrated
movies, but large variation in performances is also observed even among movies of one
distributor. This implies that estimation can be never successful without controlling for
movie quality. There is no significant difference in genre, suggesting that integrated
distributors are not specialized to a specific genre.
1.5 Empirical Results
In this paper, I focus on three important aspects of the decision by movie theaters:
allocation of screens, movie run stopping, and film choice decision. Consider the speci-
fication where
i
and
j
are unobserved characteristics for moviei and theaterj respec-
tively, and Y
ij
is the decision of theater j on movie i. D
ij
[OwnMovie] takes one if
moviei’s distributor is integrated with theaterj, zero otherwise. X
ij
includes movie
characteristics and theater characteristics.
Y
ij
= +
1
D
ij
[OwnMovie] +
2
D
ij
[OwnMovie]X
i
+
3
D
ij
[OwnMovie]X
j
+
i
+
j
+"
ij
(1.1)
Since equation (1.1) includes fixed effects for both movie and theater, the effects
of movie and theater characteristics cannot be determined. However, the existence of
both independent theaters and integrated theaters allows for the identification of the
effect of vertical integration because the same movies are observed showing under dif-
ferent vertical structures. I also investigate the differential effects of vertical integration
depending on movie and theater characteristics. For this analysis, I include interaction
terms of vertical integration with relevant variables (X
i
andX
j
) including a proxy for
movie quality, a dummy for high seasons, and a dummy for company-operated the-
aters.
19
With this fixed-effects specification, the effect of vertical integration is determined
by comparing the difference of exhibition behavior between integrated movies and
unintegrated movies in vertically integrated theaters to those in independent theaters.
Therefore, the identification comes from the omission of interactions between movie
and theater under the assumption that there is no matching between movies and the-
aters based on their characteristics other than whether they are integrated partners or
not. I will relax this assumption and show that regression results do not change quali-
tatively in the section of robustness check below.
Allocation of Screening Times
First, I examine the effects of vertical integration on the decision of allocation of
screening times at the opening weekend. I focus on screening times rather than the
number of screens devoted to a movie in order to consider the possibility that theaters
make multiple movies share a screen, by showing a movie after another movie in the
same screen.
14
How many screens each movie is shown on is important to the overall
success of the movie in its box-office revenues because almost every week observes
newly released movies so that movies’ popularity quickly declines after the first week.
Difference in differences estimators for two integrated firms are provided in Table
1.5 and 1.6 respectively. The results clearly show that integrated theaters give more
screening times to movies they own than movies they do not have ownership stake in
compared to other theaters. CJ is likely to give around one additional screening time
per day to its own movie while Lotte tends to increase screening times by around 1.5
per day for its own movies. It is worth noting that this effect is significant even for
Lotte andLotteCinema, among whichLotte does not have many hits in its line-up while
14
Distributors often complain that theaters do not allow entire time slots of a screen for a movie even
at the opening week. In addition, I assume that each screen has the same number of seats. However, it is
possible that integrated theaters locate their own movies to bigger screen having more seats and at show
times when demand is high such as evening. This kind of discrimination is not considered in this paper
because of the lack of data.
20
Lotte Cinema is the second largest multiplex chain.
15
This result provides the evidence
that observed pattern does not result from the matching between movies and theaters
depending on their quality.
Table 1.7 represents the results of estimating the effects of vertical integration on
allocation of screening times at the opening weekend (Saturday) using the full sam-
ple. Movie and theater fixed effects are included in the second half of the table to con-
trol unobserved characteristics of movie and theater as well as differences in underly-
ing demand by week and location. The result in column (1) indicates that integrated
theaters allocate approximately one more screening time to their own movies.
16
As
expected, the larger the number of screens is, the more screening times are. Column
(2) shows that company operated theaters are even more aggressive in the decision of
screen allocation in favor of their own movies than deal-run theaters. Integration effect
in deal-run theaters disappears when including the interaction of vertical integration
with dummies for high seasons in column (3), meaning that no integration effect is
observed in dealer-run theaters at low seasons. This is not surprising because inte-
grated firms take a fixed share of box office revenues from deal-run theaters based on
total box office revenues. Results do not change much, but get more significant when
movie and theater fixed effects are included, as shown in the columns (4)(6). Inte-
grated and company operated theaters allow around two screening times more to their
own movies at high seasons.
In order to see the relationship between integration effects and movie quality, I
restrict the sample including only movies released in 2008, for which I can observe
variables of pre-expectation and satisfaction rate. Satisfaction rate is used as a proxy
for movie quality, which can be justified by the fact that distributors have informa-
tion from pre-screening to their focused groups as well as to the public in some cases.
15
Market share is 30% forCJ and 8% forLotte in 2008
16
Results do not change when the number of screens is used as a dependent variable instead of the
number of screening times
21
Programmers of movie theaters are also able to watch and evaluate movies before the
decision of film choice. Both pre-expectation and satisfaction rate are included as the
deviations from their mean values throughout this paper. Regression results with this
restricted sample are presented in Table 1.8. Results in columns (1)(4) indicate that
both pre-expectation and satisfaction rate are positively correlated with the number of
screening times, while the relationship of each factor with integration differs. Columns
(4)(8) show that the interaction of integration with satisfaction rate is strongly sig-
nificant while the interaction with pre-expectation is not once unobserved movie and
theater characteristics are controlled. This implies that integration effects are concen-
trated for movies expected to generate positive word-of-mouth consistent to what the
model predicts. When satisfaction rate is evaluated at 75 percentile, the results show
that vertically integrated and company operated theaters allocate 2.4 screening times
per day more to their own movies at peak seasons when the underlying demand is
relatively high.
It is worth noting that the favor of integrated theaters in their own movies is also
stronger for domestic movies. Integrated distributors are all domestic firms, but they
distribute both domestic movies and foreign movies. While integrated distributors play
as major stake holders for most of domestic movies that they distribute, they take only
the distribution fee for most of foreign movies so that they take only small portion
of box-office revenues that foreign movies generate.
17
This provides the evidence that
observed favor of integrated theaters in their partners does not result from the reduction
of transaction costs which is often referred as a source of efficiency gains from vertical
integration. If the reduction of transaction costs is what drives the observed pattern,
then the effects of vertical integration should not depend on the nationality of movies
as long as movies are distributed by the same distributor. Therefore, the large impact
on domestic movies of vertical integration suggests that integrated theaters show their
17
In both cases, distributors recoup the costs of prints and advertising in advance.
22
own movies with larger amount of screening times because of its increased share in
perceived box-office revenues.
Movie Stopping Decision
In this section, I examine how organizational form affects the decision to stop movie
run.
Every week, movie theaters decide what movies to drop in order to show newly
released movies. From the model discussed earlier, I expect that integrated theaters are
less likely to stop their own movies, so that integrated theaters show their own movies
for a longer period of time.
Tables 1.9 and 1.10 suggest that integrated theaters show their own movies longer
than other movies when compared to unintegrated theaters and integrated rival the-
aters, but the extent of this effect is different across multiplex chains. LotteCinema shows
a stronger favor in its own movies, while the same pattern is observed for the largest
multiplex chain,CGV.
18
Table 1.11 examines the effect of vertical integration on the length of movie run using
the full sample where dependent varialbe is log of total days of movie run. In all of spec-
ifications, vertically integrated theaters show their own movies for a longer period of
time, at least two additional days. Columns (1)(3) present the regression results with
observed characteristics of movies and theaters. As expected, the number of screens in
theaters and average score of user ratings are positively related to the length of movie
run. Regarding to genre of movies, action, adventure and thriller movies tend to be
held longer in theaters.
19
Interestingly, company operated theaters are found to show
movies for relatively shorter time periods. It might imply that these theaters are system-
atically different from other theaters, highlighting the need to control theater quality. In
18
Results do not change when the dependent variables is defined as the number of weeks of movie-
running.
19
Each movie is classified as having multiple genre.
23
the second half of table, I re-estimate the effect of vertical integration on the length of
movie run with movie and theater fixed effects. The coefficient of own movie does not
change much in its magnitude, but the effect of vertical integration turns out to differ
depending on movie and theater characteristics, contrary to the regression results in the
first half of table in which unobserved characteristics are not taken into account. Results
from fixed effects models indicate that company operated theaters are more aggressive
in carrying their own movies like what we see in the analysis of screen allocation.
20
One might be concerned about the possibility that results are driven from the differ-
ence in revenues across movie by theater. If integrated theaters are better at promoting
own movies so that revenues their own movies generate are higher than competitors’
movies in integrated theaters, then we should also observe the same pattern in movie
stopping decision. Most of previous literatures try to confirm that integrated firms are
in favor of products or contents they have ownership stake in and to see whether it is
efficiency gains or strategic foreclosing move that determines observed pattern. There-
fore, it is very important to see if integration effect can be observed even after control-
ling weekly box office revenues. In fact, realized weekly box-office revenues would be
the most deciding factor when the decision to stop movie run is relevant. Movie’s true
quality is almost realized to the public through word-of-mouth so that multiplier effect
is not crucial to movie stopping decision.
CUT
ijt
= +
1
D
ij
[OwnMovie] +
2
D
ij
[OwnMovie]X
i
+
3
D
ij
[OwnMovie]X
j
+
4
Age
it
+)5Rev
ijt
+w
t
+"
ijt
(1.2)
Equation (1.2) considers the decision of movie theaters to stop movie run. CUT
ijt
takes one if theaterj drops moviei at weekt, zero otherwise. The regression includes
dummies for weeks, w
t
, the number of weeks since movie’s release, Age
it
, and most
importantly, weekly box-office revenues of moviei at theaterj at weekt,Rev
ijt
.
20
The estimation results with the restricted sample are provided in Appendix, confirming the same
results.
24
For the present purpose, I use the data for movies released in 2008 by two domestic
distributors, one integrated and the other unintegrated of which I can observe weekly
box office revenues in the level of theater. This restricts the sample into 38 movies, 17%
of movies released nationwide in 2008.
Table 1.12 shows the results of estimating a probit using this data.
21
The results
show that integration effect does not change when I control for weekly box office rev-
enues. The results in column (4) show that integrated theaters are 13% less likely to
stop their own movies than other theaters. This is almost the same as the effect of log
of box office revenues (measured in tickets sold) when evaluated at means. Given the
mean value of log weekly box office revenues (measured in ticket sold) equal to 6.47,
this result means that independent movies should sell approximately 1,100 tickets more
than integrated movies to be dropped at the same week.
Overall, the significant effect of vertical integration in this model suggests the evi-
dence of partial market foreclosure, but not the existence of efficiency gains. Integrated
theaters discriminate movies from other distributors in favor of movies they own in the
decision of movie run stopping. With the fixed number of screens available in the short
run, this foreclosure not only makes independent distributors difficult in their business,
but also reduces the total box-office revenues in theatrical market.
Film choice decision
Now, I turn to the relationship between organization structure and film choice deci-
sion. As pointed out earlier, vertically integrated theaters are more likely to show their
own movies as they internalize revenues from both distributor and exhibitor. How-
ever, this effect would be relevant mainly for movies at the margin which usually have
difficulty in finding theaters to be shown.
21
Results of linear probability model with the full sample are presented in Appendix. Weekly box-office
revenues are not controlled in that analysis because of the lack of data. The results confirm that integrated
theaters are less likely to stop their own movies.
25
Table 1.13 shows odds ratios from the results of estimating logit models for the deci-
sion of film choice.
22
Columns (1)(4) include all the movies released from 2006 to 2008,
while columns (5)(8) include movies released in 2008 only. In all of specifications, the
effect of vertical integration is significant, implying that integrated theaters are more
likely to choose their own movies. In columns (1) and (2), I present odds ratios from
a logit model that does not include movie and theater fixed effects. Results show that
the effect of vertical integration is higher in peak seasons, but negatively correlated
with online user ratings. I include average score of online user ratings as a proxy of
movie quality, but it seems not to be a good measure for two reasons in this specifica-
tion. First, average score of user ratings does not represent opening attractiveness of
the movie, but it is ex ante expectation about movie quality rather than true quality of
movie that determines the performance at opening weekend. Without specifying the
minimum length of movie run in the contract with distributor, movie theaters would
consider pre-expectation about movies as the most important factor for their film choice
decisions. Second, those who leave online ratings are not representative to the popu-
lation. It is often observed that movies receiving stellar reviews from critics fail to do
strong business in box-office. Likewise, online user ratings for some of movies might
represent preference of specific groups rather than overall popularity.
In columns (3) and (4), I employed fixed effect logit models to control unobserved
quality of movie.
23
Integration effect is still significant. Fixed effect models, however, show that differ-
ential effects of vertical integration depending on seasonality and online user ratings
reduce much, while company operated theaters are more likely to choose their own
movies than dealer run theaters. These results hold the same when I restricted the sam-
ple to movies released only in 2008 in order to use the data about pre-expectation and
22
Results from linear probability model are in Appendix.
23
Probit model with fixed effects suffer from incidental parameters problem, generating inconsistent
estimators, while the logit does not.
26
satisfaction rate for each movie from a movie marketing company as shown in columns
(5)(8). They increase the probability to be chosen, while the effect of integration does
not vary along the level of these variables once movie fixed effects are controlled. This
might be because the sample does not include movies failed to be marketed, but only
movies released nationwide. Movie characteristics such as casting, director and sce-
nario provide some degree of forecast for the demand of movie in advance, but the suc-
cess of a movie is hardly predictable before a film is made for the release.
24
As a result,
some of movies are not released to the public and the decision to market a movie would
depend on movie’s expected demand. Meanwhile, conditional that a movie is released
nationwide, whether to show it in its own theater does not depend much on movie
quality and its potential for opening scores. In fact, over 90% of movies are shown in
their own theaters in the sample. Movie’s expected popularity would be more relevant
to the decision of screen allocation which was investigated above.
On the other hands, these results contradict the view that integrated firms force
independent theaters to show their own movies when movies are expected to have
lower quality, by which integrated theaters can increase their profits by allowing more
screens to movies with better quality from other distributors. If integrated firms have
better bargaining power over independent theaters and the above view holds in the
market, then the lower movie quality is, the less the probability to show own movies
should be, which is not what we observe here. Overall, integrated and company oper-
ated theaters are at least 10% percent more likely to show their own movies when using
movie-theater fixed effects.
1.6 Robustness check
The identification of the effects of vertical integration is, so far, justified by the
assumption that there is no systematic matching between movies and theaters based
24
This is why we observe flops every year.
27
on their characteristics other than integration status. This assumption can be violated
if both integrated distributors and integrated theaters are of better quality so that inte-
grated theaters prefer their own movies because of higher movie quality their movies
have. This is not likely to drive the results for following reasons. First, integrated dis-
tributors show better performance on average, but each distributor has both hits and
bursts in its line-up. Second, among two integrated distributors,CJ is the leading com-
pany, explaining around 30% of market share, but the other integrated distributor,Lotte
has 8% of market share on average during the data period and does not have many hits
in its line-up.
Nevertheless, to address this concern, I re-estimate the effects of vertical integra-
tion on screenings times with the inclusion of interactions between movie and theater
characteristics in Table 1.14. Columns (1) and (4) repeat the estimation results without
interaction terms for the purpose of comparison. In columns (2) and (5), the number of
screens in the theater is interacted with measures of movie quality represented by pre-
expectation and satisfaction rate. In addition, I include the interactions between these
movie quality measures and dummies for multiplex chain in columns (3) and (6). Pos-
itive and significant effects of interactions between the number of screens and movie
quality measures indicate that there exists some degree of matching between movies
and theaters based on other than integration status. However, in all of specifications,
the effects of vertical integration are quite stable and significant, and the size of integra-
tion effects does not change much.
Product Differentiation One might still concern that observed patterns reflect the
results of product differentiation strategy of each theater. In fact, Chisholm et al. (2010)
found that the closer theaters are located to each other, they tend to choose different
film-choice programming to lessen competition in US motion picture industry. This
result can be seen as a strategic product differentiation, where the product is defined as
28
a set of movies to show. What we are observing in data might be a mixture of the effect
of vertical integration and product differentiation.
I address this issue with the use of the disintegration of a distributorShowbox and a
multiplex chainMegabox which happened in 2007. This disintegration betweenMegabox
andShowbox provides the opportunity to disentangle the pure effect of vertical integra-
tion from observed distribution of movie allocation. This disintegration was not forced
by any other policy like Paramount decree in U.S., which provokes the concern about
endogenous treatment problem. The use of movie and theater fixed effects is believed
to attenuate this concern.
Tables 1.15 and 1.16 describe howMegabox changes in its treatment ofShowbox after
the disintegration in the manner of DIDID approach. Before the disintegration,Megabox
allows 1.227 screening times more at the opening Saturday to their own movies com-
pared to other movies and other theaters. This tendency does not seem to change even
afterMegabox was disintegrated fromShowbox (1.339). The difference is very small and
insignificant (t=0.19). The length of movie run reduced after the disintegration, but it
is not significant either. To get more precise results, I estimate the effect of Showbox
movies inMegabox theaters on screening times at the opening Saturday before and after
the disintegration separately. Table 1.17 shows these results. In all of specifications, I
include movie and theater fixed effects to control unobserved characteristics of movie
and theater. My estimates suggest thatMegabox allocates more screening times toShow-
box movies even after the disintegration. However, the size of the effects is significantly
lower than what was before the disintegration and Wald test rejects that these effects are
equal before and after the disintegration at least 5% level, and at 0.1% level when inter-
acting with company operated theaters, suggesting that product differentiation does
not explain the entire picture of screen allocation.
Enodeneity The topic of vertical integration is related to firm boundaries, specifically
to the question of which transactions to carry out in-house and which to buy through
29
the market. Literature of transaction cost economics argues that vertical integration
may be an efficient way to organize when contracts are incomplete andexpost renegoti-
ation is costly. In the study of the US airline industry, Forbes and Lederman (2009) show
that airlines are more likely to use owned regionals on routes on which ex post adapta-
tion needs to be made frequently and the costs of adaptation are more costly. They use
the average weather patterns at the endpoint airports of a city pair to measure the prob-
ability to have adaptation on routes and the degree to which a given city is integrated
into the major’s overall network to measure costs of adaptation. Gil (2007) also shows
that vertically integrated distributors are more likely to distribute movies of contrac-
tual complexity in the Spanish movie industry, and more likely to show them in their
own theaters. In fact, the decision of vertical integration is endogenous, and movies
of integrated distributors may be systematically different from movies of independent
distributors.
To investigate this issue, I re-examine the decision of movie run stopping with the
consideration of renegotiation ex post. Box-office revenues are split into distributor and
exhibitor and its share is quite uniform in the Korean movie industry as pointed out
earlier. However, the distributor and exhibitor use an ex-post renegotiation to adjust
sharing terms. I have information about the renegotiation of one integrated exhibitor
with distributors in 2008. Regression results show that the effects of vertical integration
is quite robust to the inclusion of the interaction of vertical integration with dummies
for renegotiation, suggesting that the possibility of renegotiation ex post is not a major
issue in contracts between distributor and exhibitor.
Interviews with industry executives also reveal that no distributors prefer risky
movies. Vertical integration can reduce costs of ex post adaptation, but it does not
mean that integrated firms want to choose risky packages ex ante. In addition, renego-
tiation is not common and, if any, usually happens several weeks after the release when
movies generate typically very small amount of box-office revenues.
30
1.7 Conclusion
This paper studies the effects on exhibition behavior of movie theaters of verti-
cal integration between movie distributor and exhibitor in the Korean movie industry.
Specifically, I examine the effects of vertical integration on the decision of film choice,
allocation of screening times, and movie run stopping. The use of movie and theater
fixed effects control the variations in movie quality and underlying demand over mar-
kets.
Using a rich dataset on movies released during 20062008 in the Korean movie
industry, I find that integrated theaters are more likely to choose their own movies than
movies of other distributors, and than unintegrated theaters do. It implies that inde-
pendent distributors have a limited access to theatrical windows at the margin. I find
as well that integrated theaters allocate more screening times to their own movies at
the opening week which is the crucial timing for the success in box office revenues.
This tendency is getting larger when movies are expected to generate positive word-
of-mouth as well as when underlying demand for movies is high like holidays. In the
analysis of movie run stopping decision, it is shown that integrated theaters are less
likely to drop their own movies. This favor in own movies survives even after control-
ling the difference in weekly box office revenues of each movie over theaters. Hence,
it should be interpreted as that integrated theaters discriminate movies of independent
distributors in favor of movies they have ownership stake in. That is, integrated the-
aters use a lower standard to their own movies against other competing movies. I argue
that these results are not driven by the matching based on movie and theater quality or
other characteristics other than integration status.
Overall, combined with stylized facts in the Korean movie industry, these findings
suggest that independent distributors are partially foreclosed and that vertical integra-
tion harms consumer in theatrical exhibition market. Nevertheless, the assessment of
31
consumer welfare should be made with caution. As discussed earlier, integrated the-
aters can internalize the dynamic effect as well as the multiplier effect that early box-
office revenues generate, increasing overall performance of movies that are embedded
with good quality. The increase in future revenues from these movies is achieved at
the expense of current box-office revenues, but it might be possible that this effect out-
weighs the loss of current revenues. Future research requires precise evaluation of the
dynamic effect of vertical integration in order to measure the overall effect on consumer
welfare.
32
Table 1.3: Theater Characteristics by Integration Status
Total Nonintegrated Integrated Diff
Number of screens 7.743 7.510 7.868 -0.358
(1.789) (1.927) (1.708) (-1.13)
Number of seats 1439.2 1425.3 1446.6 -21.32
(573.0) (699.8) (495.8) (-0.21)
Number of movies 294.1 274.0 304.9 -30.90**
(59.70) (70.50) (50.15) (-3.00)
Days of movie run 19.99 20.05 19.97 0.0803
(1.314) (1.725) (1.038) (0.34)
Screenings at opening 7.736 7.455 7.887 -0.432*
(0.995) (0.735) (1.084) (-2.50)
Own movies 0.107 0 0.164 -0.164***
(0.0810) (0) (0.0247) (-46.40)
Movies per screen 38.44 36.83 39.31 -2.480**
(5.240) (6.709) (4.027) (-2.73)
Domestic movies 0.408 0.427 0.398 0.0284**
(0.0555) (0.0661) (0.0463) (2.97)
US movies 0.418 0.409 0.422 -0.0132
(0.0396) (0.0493) (0.0326) (-1.90)
Other countries 0.174 0.164 0.179 -0.0152***
(0.0256) (0.0285) (0.0222) (-3.49)
Observations 140 49 91 140
Sample includes theaters which do not have any interruption in their operation for the entire period
of the data. Parentheses include t-statistics for Diff, SD for mean values. *Significant at 5%, **at 1%,
***at 0.1%
33
Table 1.4: Movie Characteristics by Integration Status
Total Independent Integrated Diff
Number of theaters 141.4 135 154.1 -19.12***
Average weeks 2.62 2.459 2.941 -0.482***
Number of screens 1.304 1.25 1.41 -0.159***
Number of screenings 7.377 6.913 8.301 -1.388***
Days of movie run 18.03 16.91 20.27 -3.363***
Domestic movies 0.376 0.267 0.594 -0.327***
US movies 0.419 0.501 0.254 0.247***
Other countries 0.205 0.232 0.152 0.0793*
Boxoffice 769295.1 599116.4 1108788.6 -509672.2***
Naver rating 7.17 7.17 7.168 0.00203
Number of reviews 2478.9 2119.9 3195.1 -1075.3**
Pre expectation 11.005 9.847 14.476 -4.629**
Satisfaction rate 48.639 47.91 50.795 -2.879
Action 0.268 0.262 0.279 -0.0171
Adventure 0.158 0.181 0.112 0.0690*
Animation 0.078 0.084 0.066 0.018
Comedy 0.344 0.333 0.365 -0.0321
Crime 0.141 0.132 0.157 -0.025
Drama 0.447 0.438 0.467 -0.0293
Family 0.117 0.12 0.112 0.00792
Fantasy 0.115 0.122 0.102 0.0206
Horror 0.102 0.104 0.0964 0.00788
Romance 0.219 0.214 0.228 -0.0147
Sci-fi 0.0797 0.0967 0.0457 0.0510*
Thriller 0.273 0.285 0.249 0.0363
Observations 590 393 197 590
For pre-expectation and satisfaction rate, the sample includes 206 movies released in 2008. *Significant
at 5%, **at 1%, ***at 0.1%
34
Table 1.5: Screening times at the Opening Weekend (Saturday) (CJ-CGV)
Distributor
Total Non-CJ CJ Diff
Theater Non-CGV Mean 7.864 7.598 9.104 1.506
SD 4.13 3.89 4.88 0.04
Obs 50635 41677 8958
CGV Mean 8.052 7.607 10.050 2.442
SD 4.35 3.89 5.58 0.06
Obs 32784 26817 5967
Diff Mean 0.187 0.009 0.946 0.936***
SE 0.03 0.03 0.08 0.14
Table 1.6: Screening times (Lotte-Lotte Cinema)
Distributor
Total Non-Lotte Lotte Diff
Theater Non-Lotte Mean 7.950 8.070 7.057 -1.012
SD 4.17 4.30 2.83 0.05
Obs 66537 58652 7885
Lotte Mean 7.891 7.829 8.291 0.461
SD 4.40 4.51 3.64 0.1
Obs 16882 14611 2271
Diff Mean -0.059 -0.241 1.233 1.474***
SE 0.03 0.04 0.07 0.17
35
Table 1.7: The Effects of Vertical Integration on Screening Times: Full sample
(1) (2) (3) (4) (5) (6) (7)
Own movie 1.096*** 0.583*** -0.302 1.185*** 0.641*** 0.125 -0.265
(8.66) (4.21) (-0.94) (29.92) (13.83) (0.56) (-1.15)
Own X Comp op 0.897*** 0.939*** 0.957*** 0.969*** 0.974***
(3.80) (3.91) (14.65) (14.83) (14.92)
Own X High 1.602** 0.759*** 0.779***
(3.18) (8.50) (8.77)
Own X Domestic 0.514 0.399***
(1.26) (4.92)
Own X Online 0.0343 0.0534
(1.14) (1.79)
Company op 0.402*** 0.275* 0.271*
(3.39) (2.37) (2.33)
Online ratings 0.171 0.172 0.163
(1.89) (1.89) (1.80)
N of screens 0.455*** 0.454*** 0.453***
(10.42) (10.40) (10.38)
Constant 4.377*** 4.447*** 4.269*** 8.032*** 8.026*** 7.826*** 7.764***
(3.78) (3.85) (4.01) (22.76) (23.00) (22.75) (22.69)
FE Yes Yes Yes Yes Yes Yes Yes
Observations 83419 83419 83419 83419 83419 83419 83419
Adj R-squared 0.315 0.316 0.319 0.646 0.647 0.648 0.648
All of specifications count 83419 observations with 590 movies. The dependent variable is the number of screen-
ing times of moviei at theaterj on Saturday of its opening weekend. Robust standard errors are clustered by
movie and theater in columns (1)(3). Movie and theater fixed effects are employed in the second half of the
table. t statistics are in parentheses. *Significant at 5%, **at 1%, ***at 0.1%
36
Table 1.8: The Effects of Vertical Integration on Screening Times: Restricted sample
(1) (2) (3) (4) (5) (6) (7) (8)
Own movie 1.115*** 0.318 -0.242 -0.318 1.136*** 0.362*** 0.034 -0.0236
(5.80) (1.23) (-0.78) (-0.94) (16.53) (4.58) (0.37) (-0.25)
Own X Company op 1.435*** 1.447*** 1.452*** 1.398*** 1.402*** 1.394***
(3.33) (3.35) (3.33) (12.69) (12.76) (12.65)
Own X High season 1.330** 1.206** 0.778*** 0.682***
(2.96) -2.95 (5.60) (5.07)
Own X Pre Expectation -0.0019 0.00641
(-0.08) (0.66)
Own X Satisfaction rate 0.0325** 0.0163***
(2.79) (3.73)
Pre Expectation 0.226*** 0.226*** 0.225*** 0.225***
(12.06) (12.06) (12.15) (12.40)
Satisfaction rate 0.0197** 0.0196** 0.0196** 0.0173**
(3.00) (2.99) (3.01) (2.76)
Company op 0.340* 0.156 0.155 0.154
(2.28) (1.12) (1.11) (1.10)
Number of screens 0.594*** 0.594*** 0.593*** 0.592***
(11.31) (11.34) (11.32) (11.27)
Movie and Theater FE No No No No Yes Yes Yes Yes
Observations 33835 33835 33835 33835 33942 33942 33942 33835
Adj R-squared 0.541 0.543 0.545 0.547 0.65 0.652 0.653 0.654
All of specifications count with 206 movies released in 2008. Robust standard errors are clustered by movie and theater in columns
(1)(4). Movie and theater fixed effects are employed in the rest half of the table. Pre-expectation and satisfaction rate are measured as
deviations from their mean values. t statistics are in parentheses. *Significant at 5%, **at 1%, ***at 0.1%
37
Table 1.9: Total Days of Movie Running (CJ-CGV)
Distributor
Total Non-CJ CJ Diff
Theater Non-CGV Mean 19.515 18.458 24.434 5.976
SD 11.16 10.62 12.21 0.12
Obs 50635 41677 8958
CGV Mean 19.677 18.434 25.264 6.829
SD 11.36 10.62 12.872 0.16
Obs 32784 26817 5967
Diff Mean 0.162 -0.023 0.830 0.862**
SE 0.08 0.08 0.20 0.14
Table 1.10: Total Days of Movie Running (Lotte-Lottecinema)
Distributor
Total Non-Lotte Lotte Diff
Theater Non-Lotte Mean 19.632 20.118 16.016 -4.102
SD 11.33 11.65 7.69 0.13
Obs 66537 58652 7885
Lotte Mean 19.369 19.197 20.476 1.278
SD 10.84 11.04 9.42 0.24
Obs 16882 14611 2271
Diff Mean -0.262 -0.921 4.459 5.380***
SE 0.09 0.10 0.19 0.55
38
Table 1.11: The Effects of Vertical Integration on the Length of Movie Run: Full
sample
(1) (2) (3) (4) (5) (6)
Own 0.129*** 0.122*** 0.114*** 0.134*** 0.120*** 0.121***
(3.46) (5.05) (7.37) (39.93) (23.00) (17.37)
Own X Comp op 0.0107 0.0109 0.0236*** 0.0236***
(0.61) (0.61) (4.18) (4.18)
Own X High -0.0427 -0.042 -0.000602 -0.000684
(-0.89) (-0.88) (-0.09) (-0.10)
Company op -0.0470*** -0.0470*** -0.0454***
(-6.58) (-6.58) (-6.71)
Online ratings 0.105*** 0.105*** 0.105***
(8.30) (8.30) (8.28)
N of screens 0.0175*** 0.0175*** 0.0175***
(6.30) (6.29) (6.29)
Action 0.111** 0.111** 0.111**
(2.68) (2.68) (2.68)
Adventure 0.195*** 0.195*** 0.195***
(3.96) (3.96) (3.96)
Thriller 0.155** 0.155** 0.155**
(3.06) (3.06) (3.05)
Movie FE No No No Yes Yes Yes
Observations 83419 83419 83419 83419 83419 83419
Adj R-squared 0.397 0.397 0.397 0.787 0.787 0.787
All of specifications count 83419 observations with 590 movies. The dependent variable is log of total days
of moviei’s run at theaterj. Robust standard errors are clustered by movie and theater in columns (1)(3).
Movie and theater fixed effects are included in the second half of the table. t statistics are in parentheses.
*Significant at 5%, **at 1%, ***at 0.1%
39
Table 1.12: The Effects of Vertical Integration on Movie Stopping Decision
using Probit
(1) (2) (3) (4) (5)
Own movie -0.1124*** -0.1051*** -0.0982*** -0.1301*** -0.1964***
(0.0283) (0.0269) (0.0273) (0.027) (0.018)
Own X Company op -0.0136 0.0516*** 0.0287* 0.0072
(0.0117) (0.0133) (0.0142) (0.0182)
Log of weekly Box office -0.1625*** -0.1354*** -0.1047***
(0.0121) (0.0080) (0.0044)
Weeks since release 0.0621*** 0.0621*** 0.0292** 0.0351*** 0.2114***
(0.0132436) (0.0132) (0.0100) (0.0110) (0.0076)
Integ Movie -0.0056 -0.0056 -0.0075 0.0171 0.0221
(0.0505) (0.0505) (0.0320) (0.02470) (0.0096)
Number of screens -0.0140*** -0.0141*** 0.0017 -0.0027 -0.0115***
(0.0023) (0.0023) (0.0029) (0.0022) (0.0017)
Company op 0.0319*** 0.0348*** 0.0865*** 0.0810*** 0.0912***
(0.0047) (0.0058) (0.0085) (0.0093) (0.0089)
Pre Expectation -0.0125*** -0.0125*** -0.0015 -0.0055* -0.0240***
(0.0030) (0.0030) (0.0028) (0.0027) (0.0011)
Satisfaction rate -0.0078*** -0.0078*** -0.0039** -0.0020 0.0006
(0.0014221) (0.0014) (0.0013) (0.0014) (0.0004)
Week dummies No No No Yes Yes
RE(Movie*Theater) No No No No Yes
Obs 17629 17629 17629 17611 17611
This table reports average marginal effects in probit models. Sample includes movies released in
2008 by two major domestic distributors, one integrated and the other independent, counting up to
36 movies. Pre-expectation and satisfaction rate are measured as deviations from their mean values.
Robust standard errors are in parentheses and clustered by movie. Column (5) shows the results of
random effect models. *Significant at 5%, **at 1%, ***at 0.1%
40
Table 1.13: The Effects of Vertical Integration on Film Choice Decision
using Logit
(1) (2) (3) (4)
Own movie 0.0684759*** 0.0718607* 0.0056453*** 0.0048076**
(5.87) (2.54) (5.51) (2.94)
Own X Company op 0.0435569 0.0037047*
(1.81) (2.4)
Own X High season 0.0413672* 0.0014359
(2.36) (0.84)
Own X Domestic movie -0.0165941 -0.0015474
(-1.01) (-0.73)
Own X Online user ratings
Own X Pre Expectation 0.0068742* 0.000325
(2.51) (1.88)
Own X Satisfaction rate -0.000766 -0.0000246
(-1.36) (-0.43)
N of screens 0.0339106*** 0.0313946*** ***0.0029152 0.002944***
(10.56 (9.47) (7.87) (7.77)
Company op -0.0035635 0.0114261 -0.0003681 -0.0004681
(-0.28) (1.59) (-1.09) (-1.31)
Pre Expectation 0.0124917*** 0.0115974***
(14.47) (11.43)
Satisfaction rate 0.0006122*** 0.0006088***
(4.78) (4.91)
Action 0.0764796*** 0.0711499***
(10.8) (9.24)
Thriller 0.0942396*** 0.0889778***
(11.73) (10.54)
Movie FE No No Yes Yes
N 20795 20795 19346 19247
This table reports marginal effects of (conditional) logit models evaluated at means. The dependent
variable is a dummy variable that takes value one if the theater show the movie, and zero other-
wise. Sample includes 7 metropolitan cities only and 206 movies released in 2008. Pre-expectation
and satisfaction rate are measured as deviations from their mean values. Standard errors are clus-
tered by theater in regressions without fixed effects. Z statistics are in parentheses.
*Significant at 5%, **at 1%, ***at 0.1%
41
Table 1.14: Integration Effects on Screenings Times at the Opening week
(1) (2) (3) (4) (5) (6)
Own movie -0.353 -0.294 0.0227 -0.0236 0.0255 0.381***
(-1.09) (-0.91) (0.07) (-0.25) (0.29) (4.32)
Own X Company op 1.458*** 1.187** 1.002** 1.394*** 1.124*** 0.948***
(3.37) (3.11) (2.79) (12.65) (10.78) (9.32)
Own X High season 1.235** 1.262** 1.217** 0.682*** 0.712*** 0.621***
(2.87) (3.04) (3.05) (5.07) (5.63) (5.10)
Own X Pre expectation 0.0261 -0.00135 -0.0088 0.00641 0.00309 -0.00555
(1.16) (-0.06) (-0.42) (0.66) (0.37) (-0.70)
Own X Satisfaction rate 0.0330** 0.0309** 0.0163*** 0.0166*** 0.0111**
(2.78) (2.61) (3.73) (4.03) (2.73)
N of screens X Pre Exp 0.0429*** 0.0420*** 0.0434*** 0.0428***
(8.84) (8.99) (32.96) (33.18)
N of screens X Satis rate 0.00172 0.00151 0.00213*** 0.00187***
(1.26) (1.17) (4.45) (3.91)
Constant -4.381*** 1.372 -0.0175 2.430*** 3.803*** 4.254***
(-4.19) (1.25) (-0.02) (5.68) (11.46) (13.41)
Movie and Theater FE No No No Yes Yes Yes
Pre Exp X Exhibitor No No Yes No No Yes
Satis rate X Exhibitor No No Yes No No Yes
Observations 33835 33835 33835 33835 33835 33835
Adj R-squared 0.545 0.592 0.61 0.654 0.7 0.718
This table reproduces regression coefficients for the effects of vertical integration on screening times at the
opening week with the inclusion of interactions between movie and theater characteristics. The dependent
variable is the number of screening times dedicated to moviei in theaterj on Saturday of the opening week.
Pre expectation and satisfaction rate are measured as deviations from their mean values as before. The sample
includes movies released in 2008, counting to 206 movies. t statistics are in parentheses and standard errors are
clustered by movie and theater in columns (1)(3) while robust standard errors are used with movie-theater
fixed effects in columns (4)(6). *Significant at 5%, **at 1%, ***at 0.1%
42
Table 1.15: Change in screening times before and after disintegration
Before disintegration After disintegration
Distributors Distributors
Others Showbox Difference Others Showbox Difference DIDID
Theaters (1) (2) (2) - (1) (3) (4) (4) - (3)
Others (a) Mean 7.880 8.827 0.947 7.729 8.155 0.426
SD 4.307 4.469 0.068 3.906 4.135 0.063
Obs 28662 4806 39099 4384
Megabox (b) Mean 8.436 10.611 2.174 8.380 10.145 1.765
SD 5.237 6.276 0.268 4.873 5.981 0.293
Obs 2826 475 2842 325
DID (b)-(a) 1.227 (b)-(a) 1.339 0.111
SE 0.337 0.468 t=0.19
43
Table 1.16: Change in total days of movie run before and after disintegration
Before disintegration After disintegration
Distributors Distributors
Others Showbox Difference Others Showbox Difference DIDID
Theaters (1) (2) (2) - (1) (3) (4) (4) - (3)
Others (a) Mean 19.160 20.531 1.371 19.757 21.119 1.362
SD 10.348 12.521 0.167 11.464 13.815 0.187
Obs 28662 4806 39099 4384
Megabox (b) Mean 18.108 21.196 3.088 18.464 20.557 2.093
SD 10.178 12.498 0.523 10.252 11.393 0.607
Obs 2826 475 2842 325
DID (b)-(a) 1.717 (b)-(a) 0.731 -0.986
SE 0.459 0.729 t=1.15
44
Table 1.17: Disintegration ofShowbox andMegabox
3 months dropped 6 months dropped
(1) (2) (3) (4) (5) (6) (7) (8)
Show*Mega 1.640*** 1.005*** 0.498 1.061* 1.980*** 1.185*** 0.531 1.286*
() (11.72) (5.51) (1.94) (2.18) (11.43) (5.81) (1.61) (2.22)
Show*Mega(Op) 1.568*** -0.0641 1.934*** -0.114
(
) (5.29) (-0.12) (5.14) (-0.19)
Constant 1.419* 1.083 1.420* 1.083 -1.445* 0.122 -1.441* 0.122
(1.98) (1.67) (1.98) (1.67) (-2.21) (0.18) (-2.21) (0.18)
Observations 24999 28273 24999 28273 16929 23942 16929 23942
Adj R-squared 0.643 0.651 0.644 0.651 0.647 0.661 0.648 0.661
Wald test
(1)
=
(2)
( +
)
(3)
= ( +
)
(4)

(5)
=
(6)
( +
)
(7)
= ( +
)
(8)

2
5.56 11.45 6.00 11.10
P-value 0.0184 0.0007 0.0143 0.0009
This table reports regression coefficients from models with movie and theater fixed effects. Areas with noMegabox theaters are excluded
from the sample. 3 months before and after the disintegration are also dropped in columns (1)(4) and 6 months before and after
dropped in columns (5)(8) to reduce possible noise from the transition in integration status. Regression results using the sample
before the disintegration are shown in columns with odd numbers while results from the sample after the disintegration shown in
columns with even numbers. t statistics are in parentheses. *Significant at 5%, **at 1%, ***at 0.1%
45
Chapter 2
Influence of Word-of-Mouth on
Consumer Demand : Evidence from
Movie Industry
2.1 Introduction
Word-of-mouth involves informal communication among consumers about prod-
ucts and services, which is a common way to form or update their expectation about
products when product quality is uncertain. Consumers usually seek the information
from their peers, in particular from those who have experienced the product. It is
widely believed that word-of-mouth plays an important role in consumption decisions
and especially in most of experience goods whose quality is hardly discovered before
its purchase. For example, you never know how good a movie ‘Gravity’ is to you before
you actually see it. Once you, however, see it in a theater, you can’t get any refund even
though you think that Gravity is not worthy of 10 dollars of ticket price and one and a
half hour of your time.
The advent of information technology has expanded boundaries of word-of-mouth,
resulting that the influence of word-of-mouth is no longer bounded by physical distance
of social interactions. The use of the Internet allows consumers to share their thoughts
with other internet users and to influence decisions of other consumers through online
word-of-mouth. For example, yelp.com is a website where people leave their reviews
with ratings out of 5 stars for all kinds of products including restaurants, shopping
46
stores, grocery markets, theatres and so on. Because of the influence of this web-
site, many restaurants advertise themselves as having a good reputation in yelp.com.
imdb.com androttentomatoes.com are also websites where online users leave their reviews
and ratings for movies. It is often reported what movies receive in these websites after
movies’ release. Another example includes tripadvisor.com in which information about
travel has been accumulated by online users. Information provided by other consumers
could be more trustworthy than information by parties interested in business so that
these websites have been popular channels through which word-of-mouth is gener-
ated. With the growing use of internet, word-of-mouth is getting more important for
the success of product sales.
Despite of widespread belief that word-of-mouth is one of most influential factors
in product sales, it has received little attention for empirical analysis. Lack of relevant
data has been the biggest obstacle, but the availability of data on online user ratings and
reviews enable researchers to attempt to estimate the effect of word-of-mouth because
it is believed that online user ratings could be a good proxy for overall word-of-mouth,
providing the chance to analyze how word-of-mouth affects consumption decisions and
product sales. There are several studies investigating the influence of word-of-mouth
based on data of user reviews, among which some studies use online word-of-mouth
as a proxy for overall word-of-mouth, while others try to measure the direct effect of
online user reviews.
For the book industry, Chevalier and Mayzlin (2006) try to measure the direct
effect of online user ratings and reviews on book sales at that sites by employing the
difference-in-differences specification to control unobserved book fixed effect as well as
site fixed effect by comparing two different internet sites such asAmazon.com andbarne-
sandnoble.com. They find that an improvement in a book’s reviews leads to an increase
in relative sales at that sites. Zhu and Zhang (2010) studied game industry with the
same strategy (DID). Main findings are that consumer reviews have a greater influence
on the sale of games, but its influence depends on product characteristics and consumer
47
characteristics. For example, online reviews are significantly more influential in affect-
ing sales of less popular games than sales of more popular games, and that the influence
of online reviews becomes greater after the early introductory months.
Regarding to the movie industry, previous works, however, show somewhat inter-
esting results. Both Liu (2006) and Duan et al. (2008) investigate the importance of
online user reviews on box-office revenues from the data ofmovies.yahoo.com. They pro-
vide similar results that the volume of user reviews have an impact on box-office rev-
enues, but the valence of user ratings does not, contrary to what theory predicts. They
argue that consumers are affected by the awareness effect represented by the volume
of word-of-mouth generated, while they are not influenced by the persuasive effect
of online user ratings. Interpretation should be, however, cautious because causality
runs through both directions between the volume of word-of-mouth and box-office
revenues. While large number of user ratings could imply that word-of-mouth is more
active among potential consumers, it could be a result of movie sales in previous weeks.
Without controlling for unobserved movie quality, it is still possible that the relation-
ship between the volume and revenues is spurious.
1
Furthermore, the awareness effect
on product sales, if any, can go both directions depending on whether word-of-mouth
is positive or negative. If the volume of user reviews measures degree of intensity of
word-of-mouth, then we should observe that box-office revenues increase in the vol-
ume of user reviews when movies are expected to have good quality. However, it is
also possible that high degree of negative word-of-mouth can make box-office perfor-
mance even worse. On the other hands, Dellarocas (2007) showed that the inclusion
of online user ratings improve forecasting ability of their diffusion model. They, how-
ever, focus on the forecasting ability of models, requiring a caution in its interpretation
because online user ratings could be just predictor rather than influencer. If there is
1
Liu (2006) employs a simple OLS, so that it is plausible that his estimation suffers from omitted variable
bias. Taking into account of dual causality, Duan et al. (2008) estimate simultaneous equations system
using 3SLS with daily data, but they use daily data for the first two weeks. Although online user ratings
vary over time, it is strongly autocorrelated.
48
an unobserved factor correlated with product sales as well as online user ratings, then
observed correlation between online user ratings and product sales could be spurious.
Since most of experience goods are highly differentiated products, it is hard to believe
that the inclusion of observables can control product quality which is most likely to
be correlated with both demand and online user ratings. In this spirit, papers investi-
gating the influence of critic reviews on movie sales are closely related to the study of
word-of-mouth. Eliashberg and Shugan (1997) study the influence of critic reviews on
movie sales, finding that critic reviews are predictors rather than influencers. Reinstein
and Snyder (2005), however, find differential effect of critic reviews over movies. Using
the difference-in-differences approach due to the differential timing of reviews of influ-
ential critics, Siskel and Ebert, over movies, they find statistically significant influence
effect of critic reviews on opening weekend box-office revenues for narrowly-released
movies and for genre of drama, while no influence effect for widely-released movies
and for genres such as action and comedies. These results highlight the need to control
unobserved product quality.
Closely related to this paper, Moretti (2011) also examines the impact of word-of-
mouth using data from movie industry. Instead of using online user ratings or other
proxies for word-of-mouth, he identifies the effect of word-of-mouth by comparing
dynamic patterns of two different groups of movies classified as their performance
in opening weekend. Provided that the number of screens dedicated to a movie in
its opening weekend reflects the sales expectation, he identifies movies with positive
word-of-mouth from a regression of log sales on log number of screens in the opening
week. He defines a movie as ’positive surprise’ when it has a positive residual from
this regression, arguing that a positive residual implies that this movie achieved bet-
ter performance than its ex ante expectation and this is because the movie is of good
quality and consumers received a positive signal on it. He finds that the effect of word-
of-mouth is remarkably large, amounting to over 30% of total sales. One concern about
49
this approach is that the identification of movies with positive(or negative) word-of-
mouth is based on the misjudgement of theater programmers who decide the number
of screens devoted to each movie. Rather, there exists information asymmetry on movie
quality between movie theaters and moviegoers. With pre-screening or insider infor-
mation, movie theaters could have better knowledge about movie quality before its
release. Hence, it is possible that more screens are dedicated to a movie if it is expected
to have good word-of-mouth. If word-of-mouth play a little role in the opening week-
end, we would observe relatively lower box-office per screen for the movie which is
likely to have positive word-of-mouth but weak pre-expectation.
In this paper, I investigate the effect of word-of-mouth on box-office revenues in
the korean movie industry. Following to Moretti (2011), I focus on dynamic pattern
of weekly box-office revenues, especially on differential effect of movie quality on the
decline rate of weekly box-office revenues. This strategy of using panel data enables
us to control for unobserved product quality which is an issue for some of previous
studies.
I also use a new dataset from the korean movie industry containing ex ante expec-
tation on movie quality as well as ex post satisfaction rate. First, there are considerable
doubts about the reliability of online user ratings and reviews for the research of word-
of-mouth because of a selection issue on the sample of online reviewers. Anderson
(1998) shows that extremely satisfied and extremely dissatisfied customers are more
likely to initiate word-of-mouth transfers. To overcome these problems, I use a new
dataset on consumer satisfaction as well as ex ante expectation on movie quality in
movie industry. I will address these issues in detail later with the description of dataset.
It is important to control for pre-expectation on movie quality because movie’s per-
formance at the opening weekend is mostly determined by pre-expectation, not by
movie’s true quality. With the presence of word-of-mouth, we should observe slow
decay rate in weekly box-office revenues when movies are expected to be of good qual-
ity, holding pre-expectation constant.
50
2.2 Model
Following to Moretti (2011), I define the utility that individuali obtains from watch-
ing moviej as
U
ij
=q
j
+
ij
(2.1)
q
j
: the quality of moviej for the average consumer

ij
N(0;
1
d
) : tastes of individuali for moviej
Consumers have a prior on the quality of a movie before its release. I assume that a prior on
the quality of the movie is given by
q
j
N(X
0
j
;
1
h
j0
) (2.2)
X
j
are observables of characterisitics of movie j such as genre, budget, actors, directors,
nationality, ratings, the date of movie release, critics and marketing efforts by distributors. The
decision to watch a movie heavily relies on these factors at the opening week. After its release,
however, potential consumers receive a noisy signal about movie quality from their peers who
watched it as well as from movie ratings in internet movie sites. I assume that the signal is
unbiased.
s
ijt
=q
j
+u
ijt
(2.3)
whereu
ijt
N(0;
1
hjt
)
At week t, potential consumers would update their belief on movie quality based on their
prior and the signals they received. I assume that the expected value of movie quality at week t
is the weighted sum of the prior and the signals of movie quality through Bayesian updating.
51
E
it
[q
j
jX
0
j
;s
ij1
;s
ij2
;:::;s
ijt
] =w
jt
X
0
j
+w
j1t
s
ij1
+::: +w
jtt
s
ijt
(2.4)
=w
jt
X
0
j
+ (1w
jt
)q
j
+
t
X
s=1
w
jst
u
ijs
(2.5)
wherew
jt
=
hj0
P
t
s=0
hjs
andw
jst
=
hjs
P
t
s=0
hjs
fors = 1; 2;:::;t
A consumer decides to watch a movie when his expected utility from watching it is bigger
than the opportunity cost of not going to the movie. The opportunity cost of not going to see
moviej is defined as
r
ijt
=r
t
+
it
(2.6)
where
it
N(0;
1
k
)
Then, the probability that individuali goes to see moviej in week t is
P
jt
=Prob(E
t
[U
ij
jX
0
j
;s
ij1
;:::;s
ijt
]> 0) =

(1w
jt
)q
j
+w
jt
X
0
j
r
t

jt

(2.7)
where () is the standard normal cumulative function and
jt
=
q
1=d + 1=k +
P
t
r=1
h
2
jr
=(
P
t
s=0
h
js
)
2
Equation (2.7) shows that the purchase-probability is increasing in movie quality q
j
and
prior on movie qualityX
0
j
. (
@Pjt
@qj
> 0,
@Pjt
@X
0
j

> 0) Movie quality is getting more important in
determining purchase-probability as more information about movie quality is available through
word-of-mouth(
@wjt
@t
< 0).
2
If there is no word-of-mouth, then purchase-probability would be
constant because it depends solely on prior on movie quality which is predetermined before its
release.
2
For simplicity, I do not take into account change in the distribution of potential consumers due to the
truncation of the people who have already seen the movie.
52
Now assume that weekly box-office revenue is equal to the product of market size and
purchase-probability as
Q
j;t
=M
j;t
P
j;t
= (M
j
R
j;t1
)P
j;t
= (M
j

t1
X
k=1
Q
j;k
)P
j;t
=M
j
t1
Y
k=1
(1P
j;k
)P
j;t
(2.8)
whereQ
j;t
represents weekly box-office revenues,M
j;t
the potential consumer population
for moviej at weekt,R
j;t
the cumulative box-office revenues of moviej until week t,P
t;j
the
probability of purchase. Since most of consumers watch a movie once, the potential pool of
consumers is shrinking as people watch the movie. Log-transformation provides
lnQ
j;t
= lnM
j
+
t1
X
k=1
ln(1P
j;k
) + lnP
j;t
(2.9)
In the case of no word-of-mouth, soP
j;t
=P
j
for all t, log of weekly box-office revenues can
be characterized as
lnQ
j;t
= lnM
j
+Age
j;t
ln(1P
j
) + lnP
j
(2.10)
where Age
j;t
is the number weeks since the release of movie j at week t, which is set to
be (t 1). This result represents that decay rate of weekly box-office differs over movies even
under the assumption of no word-of-mouth. Higher probability of purchase contributes to box-
office revenues, but at the same time, decreases the pool of potential consumers quickly, pro-
viding a rapid decline in weekly box-office revenues. Under no word-of-mouth, the purchase-
probability P
j
would depend largely on prior on movie quality, so it would be expected that
movies with higher pre-release expectation show faster decline in box-office revenues. Hence,
in order to capture the effect of word-of-mouth, the pre-expectation on movie quality should be
taken into account.
53
Now, consider the case with the existence of the effect of word-of-mouth in which P
j;t
is
increasing inq
j
. Then, the change of log box-office revenues from weekt to weekt + 1 can be
found as;
lnQ
j;t+1
lnQ
j;t
= ln(1P
j;t
) + lnP
j;t+1
lnP
j;t
(2.11)
Suppose there are two movies,H;L, with identical prior, but with different movie quality,
q
H
>q
L
. Since they have the same prior on movie quality, purchase-probability would be same,
P
H;1
= P
L;1
= P
1
at week 1, but higher for movie ‘H’ (P
H;2
> P
L;2
) at week 2, assuming that
word-of-mouth is relevant to the second week and subsequent weeks. Then, we should observe
slower decline from opening week to the second week for movie ‘H
0
as shown below;
lnQ
H;2
lnQ
H;1
= ln(1P
H;1
) + lnP
H;2
lnP
H;1
= ln(1P
1
) + lnP
H;2
lnP
1
> ln(1P
1
) + lnP
L;2
lnP
1
= lnQ
L;2
lnQ
L;1
(2.12)
For subseqent weeks, it would be possible that we have either higher decline or slower
decline for movies with higher quality. This is because higher movie quality increases purchase-
probability, but at the same time, increased probability reduces potential size of moviegoers for
following weeks. It can be shown from equation (2.11) such as;
@ [lnQ
j;t+1
lnQ
j;t
]
@q
j
=
@Pj;t
@qj
1P
j;t
+
@Pj;t+1
@qj
P
j;t+1

@Pj;t
@qj
P
j;t
(2.13)
In equation (2.13), the second term on the right hand side measures the increase in box-
office revenues at weekt + 1 due to the increase in purchase-probability, but the first and the
last terms indicate that higher movie quality also increasesP
j;t
, resulting in reduced potential
pool for movie j at week t + 1. Overall, this analysis demonstrates that under the presence
of word-of-mouth, a movie with higher quality experiences slower decline at early stages of
movie run, but it might have rapid decline in box-office as potential consumer groups shrink,
suggesting that concave pattern would be observed for movies with higher quality. This is
shown in Figure 2.1.
54
What this model suggests is different from statistical herding and information cascade mod-
els of Banerjee (1992) and Bikhchandani and Hirshleifer (1992) in which consumers may ignore
personal information on movie quality to follow the public consensus represented by box-office
revenues in previous week. In these models, ex ante expectation on movie quality determines
box-office revenues for the entire life of a movie. These models predict that move quality does
not have net impact on box-office revenues and the decline rate of weekly box-office revenues
should be lower and concave for movies with higher pre-expectation.
Figure 2.1: Weekly box-office revenues with the effect of word-of-mouth
2.3 Data
Movie industry has some of advantages for the research of word-of-mouth. First, no other
industry is as under the strong influence of word-of-mouth as movie industry. Movie is one
of common topics of daily conversation. Newspapers, radio, and TV programs talks about
newly released movies every week. Industry executives in movie business consider word-of-
mouth as one of major driving forces for the success of movies. In fact, box-office success of
several movies like Paranormal Activity, My Big Fat Greek Wedding, and The Hangover has been
attributed to strong word-of-mouth that these movies generated. Second, movie ticket price
does not vary over movies, enabling us to focus on the effect of word-of-mouth free from price
55
mediation.
3
Third, weekly box office revenue data is available publicly, which allows us to
explore the dynamic pattern of movie sales as well as to control for unobserved movie effect.
In addition to these general features, the Korean movie market provides other advantages
when using online user ratings as a proxy for word-of-mouth. Unlike US market, ancillary
windows such as DVD, cable, and broadcast TV market have relatively small proportions com-
pared to theatrical window in Korea.
4
As a result, the majority of online user ratings come from
moviegoers who went to theaters. Since those who see a movie in theaters would be differ-
ent from other consumers who see it through DVD or TV in their taste on that movie, using
the overall user ratings in US market could bias the effect of word-of-mouth when exploring
box-office in theatrical window. Second, I use the data on online user ratings from a website
Naver.com whose market share is over 70% in search engine market in Korea. There is a saying
that data from Naver is not a sample, but a population because of its dominance. Naver also
provides detailed data on user ratings by gender, age groups, and weeks after movie release.
These attributes could help to reduce a bias driven from the discrepancy of distribution between
online users and overall moviegoers.
I define the effect of word-of-mouth as social learning which is the process to update one’s
belief on product quality from those who experienced it. The ‘buzz’ effect marketing efforts
generate before movie release is not captured as the effect of word-of-mouth here. Instead of
looking into opening weekend box-office revenues or gross revenues, I explore dynamic pat-
terns of weekly box office revenues. If word-of-mouth matters, we should see different patterns
of weekly box-office revenues between movies with positive word-of-mouth and movies with
negative word-of-mouth.
Weekly box-office data is obtained from Korean Film Council, a governmental agency. Char-
acteristics of movies such as ratings, nationalities, distributors, genre, release dates are also from
KoreanFilmCouncil. I also collect online user ratings fromNaver.com which is the leading search
engine in Korea. The influence of Naver is quite impressive, achieving 72% market share in
Korea as Feb, 2012. Naver is not a pure search engine, rather provides lots of its own services,
among which movie section allows users to leave their reviews with ratings out of 10 scale on
3
Uniform pricing in movie industry is somewhat puzzling, considering the differential popularity of
movies. For the discussion of uniform pricing in movie industry, refer to Orbach (2007).
4
Sales from theatrical window captured 70.4% in Korea, while 28.7% in US in 2007
56
movies. Because of Naver’s dominance, user ratings in Naver is also believed to reflect average
taste of the entire moviegoers. Indeed, market insiders consider user ratings in Naver as an
important indicator and influencer for the success of movies.
I also use the unique data from a research company in Korea, which surveys about expecta-
tion on coming movies as well as satisfaction rate on movies people watched. In particular, the
data provides the number of people who are aware of a movie coming to next week, the number
of people who answered that they would go to see the movie in a theater, the number of people
who are satisifed with a movie, and the number of people who want to recommend the movie
they saw to others. Table 2.1 shows that average attendance rate is 47% in this dataset, implying
that almost 5 out of 10 individuals watched a movie per week in this sample, which means that
people watch around 2 movies per month. According to Korean Film Council Year Book, peo-
ple watched around 9 movies in 2008, converting into less than one movie per month. Hence,
the sample in this dataset seems to include more frequent moviegoers compared to population,
but they would be those who generate early word-of-mouth.
5
This data is available for movies
released since 2008, so the sample of the analysis is restricted to movies released between 2008
and 2011 in this paper. The final sample includes 755 movies which opened on over 50 screens.
I take the share of people who answered ‘yes’ to the question ‘Are you going to watch this
movie in a theater’ as a measure ofexante expectation on movie quality.
6
Figure 2.2 shows that
the distribution of this measure is right-skewed. It is consistent to the fact that most of attention
is paid to only one or two movies when several movies are released at the same time. For a
proxy of movie quality, I use either the share of people who answered as being satisfied with
the movie or average score of user ratings inNaver.com. It is worthy to note that these variables
are not proxies for word-of-mouth, but for movie quality.
5
If frequent moviegoers have systematically different tastes from overall population, we might have
selection issue and estimated effect would be biased. It is possible that frequent moviegoers might have
more strong taste in favor of better-quality movie in terms of artistic value. With this distortion in the
sample, ex ante expectation and satisfaction rate would be measured relatively low for movies which did
not get good reviews from critics, but achieved commercial success. Then, the impact of word-of-mouth
will be underestimated so that estimated effect of word-of-mouth could be interpreted as a lower bound.
6
For another measure for ex ante expectation, I use the share of people who answered to go to see
this movie as well as to be well-aware of this movie, which is believed to represent the group of ’serious
consumers’. Empirical results are robust to this change
57
Table 2.1: Descriptive Statistics for Variables related to Word-of-Mouth
Mean S.D Min Max
Weekly Survey Data
Sample size 978 125.18 775 1381
Total Attendance 459 104.91 284 772
Total Attendance / Sample 0.47 0.07 0.34 0.66
Questions
For coming movies
Are you aware of this movie 40.81 21.25 4 95.1
Are you well-aware of this movie 15.86 10.60 1 64.7
Are you going to watch this movie 9.95 9.24 0.1 61
For movies running currently
Are you satisfied with this moive 52.92 20.43 0 100
Do you recommend this movie 53.52 20.53 0 100
Online user ratings fromNaver.com
Number of reviews 2724 4052 33 40341
Average ratings (out of 10) 7.29 1.27 2.53 9.43
The sample includes 755 movies released between 2008 and 2011. Mean value of each question
is the percentage of people who answered ’yes’ to corresponding question.
0 50 100 150
Frequency
0 20 40 60
Pre-Expectation (Willingness to Watch)
Figure 2.2: Histogram ofexante Expectation
58
Does WOM matter? : Some evidences from descriptive statistics
Previous studies provided mixed results about the effect of word-of-mouth. Although
some of papers from marketing literature tend to show that the inclusion of online user rat-
ings improve the accuracy of forecast on product sales, Duan et al. (2008) and Liu (2006), with
linear regression models, found that the volume of user ratings is significant, but online user
ratings was not significant in explaining box-office revenues in movie industry. This finding
contradicts a prevalent belief among industry executives that positive word-of-mouth is critical
for the success of movie sales, assuming that online user ratings can be a good proxy for overall
word-of-mouth.
10 12 14 16
ln of total boxoffice (ticket)
4 6 8 10 12
ln of number of user ratings
correlation = 0.82
Figure 2.3: Gross box-office vs Volume of user ratings
Consistent with these findings, Figure 2.3 shows a clear positive correlation between total
box-office revenues and number of user ratings. Valence of user ratings, however, do not have
such strong correlation with total box-office revenues as seen in Figure 2.4. Correlation between
average user ratings and total box-office revenues is found to be 0.28 in my sample. This might
59
10 12 14 16
Ln of total box-office
0 20 40 60 80 100
Satisfaction rate
10 12 14 16
Ln of total box-office
0 20 40 60 80 100
Recommendation rate
10 12 14 16
Ln of total box-office
2 4 6 8 10
Online user ratings
Figure 2.4: Correlation between Box-office and Word-of-Mouth
be because movie demand is involved with lots of other variables like genres, ratings
7
, nation-
alities, cast, director as well as unobserved movie quality.
8
In general, weekly box-office revenues reach the peak in the opening week and decline
over time. My sample finds that box-office revenues drop by around 50% from the opening
week to the second week, and by 64% from the second week to the third week. As the process
of social learning, word-of-mouth, however, takes place only after early moviegoers go to the-
aters, implying that the effect of word-of-mouth is limited in the opening weekend, but growing
over time. The correlation between box-office revenues and average use ratings gets larger over
weeks after movie release(Figure 2.5). If word-of-mouth matters and its effect grows over time,
then we should expect slower decline in box-office revenues for movies with positive word-of-
mouth compared to movies with negative word-of-mouth as we see from the model discussed
earlier. An ideal example to allow us to measure the impact of word-of-mouth would be two
simultaneously opened movies receiving the same pre-expectation on movie quality but dif-
ferent user ratings or satisfaction rate after their release. With the same pre-expectation levels,
7
In Korea, movies are classified into the following categories: Suitable for all audiences(ALL); suitable
for children 12 and older(12+); suitable for children 15 and older(15+); suitable for adults 18 and older(18+);
restricted to adults 19 and over(R)
8
In this paper, movie quality implies the attractiveness of movie regarding to consumer’s utility, not
the artistic value
60
the difference in their performances can be construed as the effect of word-of-mouth. Hence,
it is important to control for pre-expectation on movie quality in order to isolate the effect of
word-of-mouth.
0 5 10 15
2 4 6 8 10
Average user ratings
Log opening weekend box-office
0 5 10 15
2 4 6 8 10
Average user ratings
Log 2nd weekend box-office
0 5 10 15
2 4 6 8 10
Average user ratings
Log 3rd weekend box-office
0 5 10 15
2 4 6 8 10
Average user ratings
Log 4th weekend box-office
Figure 2.5: Correlation between Box-office and User ratings by week
Figure 2.6 provides suggestive evidence that word-of-mouth drives box-office revenues
after the first week. Movies with higher ratings clearly exhibit a slow decline in their sales.
Box-office revenues are even higher in the second week for movies getting above 9 as the aver-
age of user ratings. It is also important to note that box-office revenues in the opening weekend
do not vary much across user ratings because this evidence implies that slower decline in box-
office revenues of movies with higher ratings is not driven by theirexante attractiveness before
the release of movies.
Figure 2.7 highlights the evidence that word-of-mouth affects box-office performance. As
one of the biggest surprise in all time, a movieKingandtheCrown received relatively low atten-
tion at the opening week because of no star in its cast, but it experienced very slow decay rate
in box-office revenues over its life in theatrical window. Surprisingly, box-office revenues even
increased for 2 weeks after its release. On the other hands, movie Sector 7 had rapid drop in its
box-office performance while it achieved great box-office score at the opening weekend because
61
Figure 2.6: Dynamics of Box office by satisfaction rate
of enormous attention from the fact that this movie was the first 3D movie made in Korea. The
striking difference in their patterns of weekly box-office revenues can be hardly explained with-
out the existence of strong word-of-mouth.
Summary statistics are shown in Table 2.2. Box-office variables are measured as the number
of tickets sold. The average movie has 267 screens at the opening. It runs over 6 weeks in
theaters, and achieve 800,000 attendance in total. Market share of domestic movies is around
50% in terms of total box-office, but the number of domestic movies is quite short of that of
US movies which consist of almost 50%. Action, comedy, drama, and thriller are the four most
common genres.
Empirical strategy
To identify the effect of word-of-mouth, a following equation is considered;
ln(Q
jt
) =
0
+
1
Age
jt
+
2
(Age
jt
WOM
j
) +
3
(Age
jt
PreExp
j
) +m
j
+w
t
+
jt
(2.14)
whereln(Q
jt
) is the log of box-office revenues of moviej in weekt; Age
jt
is the number
of weeks of moviej since its release; WOM
j
is a proxy for the quality of moviej (average of
62
Table 2.2: Summary statistics
Variable Mean Std. Dev.
Weekend sales (thousand) 81.47 178.89
Number of screens 127.48 154.46
Number of screening times 1637.08 2339.74
Occupancy rate (tickets / seats) 0.19 0.14
Total boxoffice (thousand) 804.38 1360.22
Length of movie run (week) 6.05 3.15
Advertisement costs (million won) 269.10 273.22
Opening weekend sales (thousand) 217.47 278.83
Number of screens at opening 268.89 161.50
Number of screening times at opening 4141.62 2603.60
Occupancy rate at opening weekend 0.22 0.12
Rating
All 0.189
12+ 0.240
15+ 0.362
Teenager restricted 0.209
Nationality
Domestic 0.323
US 0.498
Other countries 0.179
Genre: selected
action 0.278
adventure 0.163
animation 0.114
comedy 0.262
horror 0.086
romance 0.197
thriller 0.269
sequel 0.087
Number of movies 755
Movie sales are reported as the number of tickets sold. Advertisement costs count
TV , radio, newspaper, magazine commercials. The sample includes 755 movies
that were released between 2008 and 2011. The total sample size is 4567. Adver-
tisement costs are available for 695 movies. Each movie is classified as multiple
genres.
63
Figure 2.7: Hit vs Flop
online user ratings, satisfaction rate of moviej);PreExp
j
is ex ante expectation on the quality
of moviej;m
j
is a movie fixed effect;w
t
calendar week dummies.
Since box-office revenues usually get the peak at the opening weekend and decline over time
(which is also consistent to what theory predicts), I expect that
1
is negative. The coefficient of
our interest is
2
.
2
> 0 implies that higher movie quality lowers the rate of decay in box-office
revenues. As pointed out earlier with the model, the decay rate could differ over movies even
without the word-of-mouth. If a strong performance in the first weekend could reduce the base
of potential consumers in the following weeks, then we should expect faster decline of box-office
revenues for movies that experienced strong sales in the opening week without word-of-mouth.
At the same time, it is, however, possible that movies with high ex ante expectation can expe-
rience slower decline in their box-office revenues because of their competitiveness in following
weeks against newly released movies. Observed movie demand can be decomposed into under-
lying demand and expansion effect of the movie, and its relative importance can influence the
dynamic pattern of box-office revenues.
9
Therefore, it is necessary to controlexante expectation
on movie quality in order to identify the effect of word-of-mouth. With this specification, iden-
tification of the effect of word-of-mouth comes from the difference in the change of box-office
9
Einav (2007) explores seasonality in U.S. movie industry.
64
revenues among movies with different degree of movie quality(measured by satisfaction rate or
online user ratings), holdingexante expectation equal.
2.4 Empirical Results
Table 2.3 shows estimates of equation (2.14) without movie fixed effects in order to see styl-
ized pattern in relationships of movie characteristics with box-office revenues. For the measure
of movie quality, I use the valence of online user ratings(denoted as WOM1) in the first half of
columns and satisfaction rate in the second half of columns(denoted as WOM2).
One of findings is that advertising costs are positively correlated with weekly box-office rev-
enues as expected.
10
This result, however, should not be interpreted as the advertising impact
on box-office revenues. The size of advertising costs is endogenously determined with the con-
sideration of several factors such as release timing, production budget and most importantly its
box-office appeal which is not controlled in these regressions. Hence, this result seems to reflect
the positive correlation of advertising costs with movie’s appeal in box-office revenues. Sec-
ond, stars and directors are found to be insignificant in explaining weekly box-office revenues
in all of specifications.
11
There is a belief that star and director might contribute to the increase
of pre-expectation. If it holds, then the effect of star power could be run through increased
pre-expectation. Comparison of column (1) and (2) or of column (5) and (6) suggests this pos-
sibility. Once pre-expectation is controlled as in column (2) and (6), the coefficients of star and
director decreases and become negative while they are insignificant. However, as they are still
insignificant in columns (1) and (5) even in the models without pre-expectation, the role of star
and director seem to be very limited.
12
Third, domestic movies outperform movies from other
countries including Hollywood movies. This result reflects the fact that Korea is one of a few
countries in which domestic movies are competitive to Hollywood movies.
10
In this paper, advertisement costs include TV , radio, newspaper, magazine commercials. Online and
other outdoor advertising costs(e.g. Billboard) are not included because of lack of data.
11
In this paper, each actor or director is considered to be ’star’ or ’director’ if he or she either has any
movie achieving over three million within 5 years or wins award of best actor or actress in leading role.
12
Ravid (1999) and De Vany (2004) examine the role of stars and both studies conclude that stars play no
role in the success of a film.
65
Table 2.3: Regression of Log Weekly Box-office Revenues
(1) (2) (3) (4) (5) (6) (7) (8)
Age -0.773*** -0.786*** -1.071*** -1.066*** -0.773*** -0.781*** -0.986*** -0.974***
(-14.92) (-15.25) (-32.71) (-31.24) (-14.93) (-15.09) (-18.81) (-19.41)
WOM1 0.601*** 0.564*** -0.490*** -0.501***
(11.70) (10.99) (-6.38) (-6.63)
Age * WOM1 0.289*** 0.291***
(12.00) (12.05)
WOM2 0.0429*** 0.0379*** -0.0107* -0.0212***
(11.11) (9.87) (-2.05) (-3.89)
Age * WOM2 0.0126*** 0.0151***
(7.99) (8.89)
Pre-Expectation 0.0760*** 0.0892*** 0.0958*** 0.0647*** 0.0739*** 0.122***
(6.74) (7.97) (6.46) (6.08) (7.44) (8.71)
Age * Pre-Exp -0.00144 -0.0103***
(-0.52) (-3.52)
Ads costs 0.0039*** 0.0028*** 0.0023*** 0.0024*** 0.0034*** 0.0026*** 0.0019*** 0.0020***
(8.81) (6.36) (5.51) (5.69) (8.15) (6.15) (4.94) (5.03)
Star 0.0604 -0.256 -0.0874 -0.0936 0.0908 -0.182 -0.007 -0.0256
(0.32) (-1.44) (-0.59) (-0.63) (0.50) (-1.05) (-0.05) (-0.17)
Director 0.215 -0.133 -0.0754 -0.0729 0.222 -0.0795 -0.0849 -0.0711
(0.72) (-0.41) (-0.28) (-0.27) (0.75) (-0.25) (-0.31) (-0.26)
Sequel 0.0822 -0.336 -0.221 -0.228 0.000318 -0.351 -0.279 -0.319
(0.41) (-1.66) (-1.01) (-1.05) 0.00 (-1.96) (-1.51) (-1.74)
Nationality
US -1.168*** -0.895*** -0.756*** -0.750*** -0.852*** -0.654*** -0.517** -0.461**
(-6.50) (-5.04) (-4.45) (-4.42) (-4.84) (-3.70) (-2.83) (-2.64)
Other countries -1.177*** -0.999*** -0.963*** -0.953*** -1.088*** -0.938*** -0.959*** -0.892***
(-5.38) (-4.70) (-5.05) (-5.13) (-4.71) (-4.13) (-4.04) (-4.06)
N 4261 4261 4261 4261 4245 4245 4245 4245
t statistics in parentheses. Standard errors are clustered by movie. WOM1 denotes average score of online user ratings from
Naver.com, while WOM2 satisfaction rate. WOM variables are used as deviation from its mean value. All of specifications
include calendar week dummies and genre dummies. *Significant at 5%, **at 1%, ***at 0.1%
66
Related to the main interest of this paper, regression results show that movie quality mea-
sured as WOM variables lowers the decline rate of weekly box-office revenue, consistent to what
the model predicts. In this paper, WOM variables and Pre-Expectation are used as demeaned
values. This result is robust to the inclusion of pre-expectation. In columns (1) and (5), I present
the coefficient that include only age variable, the number of weeks since movie’s release, with-
out any interaction with. The coefficient of age, -.773, is the rate of decline in box-office revenues
for the average movie. In column (3) and (7), the regression includes Age * WOM. The coeffi-
cient from this regression is equal to 0.289 and 0.0126 respectively. They are statistically different
from zero. The inclusion of interaction term of WOM variables and age significantly increase
the goodness of fitness of the model. The difference in magnitude between two regressions
is mainly from different scale of two measures and average. In column (3) where online user
ratings are used for movie quality, the rate of decay is -0.81 for the movie at 75th percentile of
movie quality, while it is -1.27 for the movie at 25th percentile. The regression in column (7) with
satisfaction rate also has similar pattern, -0.8 and -1.16 respectively. Notable difference between
models with user ratings in Naver.com and models with satisfaction rate is that the impact of
pre-expectation on decline rate is only significant in the model with satisfaction rate as shown
in column (8). The entry is negative, while pre-expectation is positively correlated with box-
office revenues, suggesting that higher pre-expectation could increase the level of box-office
revenues, but bring a rapid rate of decline over weeks.
Each movie is a unique product and observables might not explain all of its appeal in box-
office even though I included proxies for movie quality and pre-expectation in previous regres-
sions. To address this concern, I re-estimate the effect of word-of-mouth with movie fixed effects
model. The regression results are shown in table 2.4. In columns (2) and (5), WOM variables are
significant and positive on time trend of box-office revenues again. Its magnitude is not much
different from what we found in previous regressions without movie fixed effects, although the
effect of age is even larger when unobservables are controlled.
Now, I investigate the possibility that the rate of decay varies over weeks as the model pre-
dicts. I include the quadratic terms ofAge
jt
as the interaction withWOM
j
and withPreExp
j
.
Furthermore, interaction terms of movie characteristics such as genre, nationality, star, rating
67
and advertisement costs with age are included to capture the differential effect of each variable
on the rate of decay.
ln(Q
jt
) =
0
+
1
Age
jt
+
2
(Age
jt
WOM
j
) +
3
(Age
jt
PreExp
j
)
4
(Age
jt
X
j
)
+
5
(Age
2
jt
WOM
j
) +
6
(Age
2
jt
PreExp
j
) +m
j
+w
t
+
jt
(2.15)
Columns (4) and (7) present regression results of equation ((2.15)). Regardless of the choice
of WOM variable, quadratic term of age, Age sq, is statistically different from zero and posi-
tive. This means that weekly box-office revenues has convex curve for the average movie at
50th percentile of movie quality and pre-expectation because these variables are used as the
deviation from their mean values in this regression. The interaction of WOM variables and
pre-expectation withAge sq are also significant in both columns. Negative coefficients for these
interactions suggest that both positive word-of-mouth and higher pre-expectation contribute to
slower rate of decay and its marginal effect declines over weeks. For a movie with 50th per-
centile pre-expectation to have a concave pattern, WOM1 should be, however, more than 3.96
(out of 10 scale) and WOM2 should be 60.9 (out of 100 scale) and over, but this is impossible
because this level is out of range.
13
The same results hold for even movies at 75th percentile
pre-expectation. The entries for interaction terms withAge sq indicate that only movies with
over 95th percentile of both WOM and pre-expectation could have concave pattern in their log
of weekly box-office revenues, which do not exist in this sample.
De Vany (2004) shows that star movies have more staying power than opening power. It
this argument is also valid in the korean movie industry, then we should have positive effect
of stars on the rate of decay. This might be true if consumers are relatively reluctant to update
bad signal when movies have stars in their cast. Age*Star and Age*Director are included to
consider this possibility. In columns (3) and (6), I find that stars have no significant effect on
the decay rate of movies sales in theaters, while directors have positive effect at 5% significance
level. This is an interesting result, suggesting that casting stars do not have any major benefits
to the financial success of a film. The Korean movie industry is known to give directors more
flexibility in their directing, compared to Hollywood system. For movie ratings, movies with
rating ‘All’ has slower decay rate than other movies. Most of movies with rating ‘All’ tend to
target kids and their preference would be quite different from the average preference of other
13
Maximum value is 2.14 for WOM1 and 47 for WOM2 in the sample.
68
Table 2.4: Regression of Log Weekly Box-office Revenues: movie fixed effects
(1) (2) (3) (4) (5) (6) (7)
Age -1.013*** -1.274*** -1.652*** -1.923*** -1.196*** -1.631*** -1.898***
(-26.79) (-42.47) (-23.11) (-26.68) (-29.96) (-21.39) (-27.66)
Age * WOM1 0.291*** 0.196*** 0.334***
(15.42) (10.15) (9.35)
Age * WOM2 0.0148*** 0.0095*** 0.021***
(11.33) (7.41) (9.73)
Age * Pre-Exp 0.0024 0.0005 0.0198*** -0.0052* -0.0042* 0.0094
(1.01) (0.25) (4.67) (-2.45) (-2.04) (1.76)
Age sq 0.027*** 0.063*** 0.032*** 0.065***
(5.53) (10.88) (5.87) (13.21)
Age sq * WOM1 -0.016***
(-4.42)
Age sq * WOM2 -0.0011***
(-5.43)
Age sq * Pre-Exp -0.0016*** -0.0011*
(-4.75) (-2.30)
Age * Star -0.00434 -0.0719 0.0209 -0.0476
(-0.10) (-1.60) (0.46) (-0.99)
Age * Director 0.115* 0.0968* 0.117* 0.101
(2.53) (2.05) (2.22) (1.96)
Age * Sequel -0.155 -0.185* -0.166* -0.179**
(-1.87) (-2.43) (-2.39) (-2.73)
Genre
Age * Action -0.0857 -0.0497 -0.138** -0.108*
(-1.94) (-1.12) (-3.09) (-2.45)
Age * Adventure 0.00562 -0.0304 -0.0243 -0.0428
(0.12) (-0.67) (-0.46) (-0.98)
Age * Sci-fi -0.012 0.0414 0.00972 0.0438
(-0.20) -0.67 -0.15 -0.66
Ratings
Age * All 0.185** 0.102 0.202** 0.0791
(2.87) (1.71) (2.87) (1.27)
Age * 12+ 0.00388 0.0391 -0.0219 0.0171
(0.07) (0.71) (-0.37) (0.29)
Age* 15+ -0.0295 -0.0382 -0.0469 -0.0608
(-0.55) (-0.71) (-0.82) (-1.08)
Nationality
Age * Domestic 0.118 0.0974 0.0876 0.0788
(1.90) (1.70) (1.32) (1.38)
Age * US 0.0876 0.0649 0.109 0.0901
(1.53) (1.26) (1.80) (1.74)
N 4567 4567 4567 4567 4546 4546 4546
R-sq 0.739 0.805 0.835 0.845 0.786 0.827 0.843
t statistics in parentheses. WOM1 denotes average score of online user ratings from Naver.com, while WOM2
satisfaction rate. *Significant at 5%, **at 1%, ***at 0.1%
69
age groups who consist of the sample of survey data. Overall, there are not many variables hav-
ing significant impacts on the rate of decay among movie characteristics once pre-expectation
and WOM variables are included.
To evaluate the impact of word-of-mouth, consider two movies at the same level of pre-
expectation at its median and the same level of the opening weekend performance, but having
different WOM, one with 25 percentile and the other movie with 75 percentile of satisfaction
rate. Empirical results in table 2.4 show that the rate of decay is -1.4 for movies with 25 per-
centile, but -0.97 for movies with 75 percentile. Converted into the difference of total box-office
revenues, word-of-mouth raises total box-office sales by 24% for the entire movie run. When we
consider the contribution of word-of-mouth to box-office revenues from the second week to 8th
week, this effect of word-of-mouth is strikingly larger at 65% of box-office revenues.
14
These
numbers increase to 35% and 78% respectively when movies at 90th percentile are compared to
movies at 25% percentile but the same level of pre-expectation at 50%.
Lastly, I investigate what the volume of user ratings means related to the effect of word-of-
mouth. Previous studies find that volume of online user ratings has an significant impact on
box-office sales, arguing that the volume of online user has an awareness effect which increases
demand for movies. As pointed out earlier, there is no reason to believe that increased aware-
ness about a specific movie has only positive effect.
Table 2.5: Who initiate WOM transters?
Online user ratings (percentile)
Pre-Expectation 0<25 2550 5075 75<100 Total
0<25 289.45 386.63 327.57 526.45 354.21
2550 708.33 688.10 619.20 805.26 699.36
5075 1054.47 1000.29 862.73 2018.45 1208.72
75<100 2941.45 2198.76 2503.63 5734.76 3208.21
Total 893.84 913.34 1026.26 2186.72 1154.80
This table shows mean value of the number of reviews for movies grouped over pre-
expectation by ratings of online user ratings.
14
Since movie fixed effect is used in these models, the effect ofexante expectation and of word-of-mouth
on the first week is not identified. I assume that movies with the same level of ex ante expectation have
the same performance at the first weekend in this computation. Therefore, the difference would be in fact
larger than this calculation.
70
Table 2.5 presents the average of the volume of online user reviews over pre-expectation by
online user ratings. The volume of online reviews increases in both pre-expectation and user
ratings. However, there is not a big difference between lower quartile and middle quartiles
grouped by online user ratings. In fact, the volume of user reviews in lower quartile is even
higher for middle quartiles groups when we exclude lowest quartile group of pre-expectation.
For movies in highest quartile of pre-expectation, this tendency gets even clear. This statis-
tics describe that extremely satisfied consumers and extremely dissatisfied consumers are more
likely to generate WOM transfer than other groups. Hence, word-of-mouth would make movie
of good quality perform better in box-office sales and movies of bad quality perform even worse
in box-office sales.
Table 2.6: The effect of controlling for the number of reviews
(1) (2) (3) (4)
Age -1.192***
(-30.29)
Age * Satisfaction 0.0143***
(10.42)
Age * Pre-Exp -0.00685** -0.00559* -0.00601* -0.00636*
(-2.73) (-2.30) (-2.20) (-2.27)
Age * Satis(< 25 %) -1.470*** -1.420*** -1.445***
(-19.17) (-14.51) (-16.70)
Age * Satis(2550%) -1.161*** -1.157*** -1.154***
(-15.77) (-15.69) (-15.73)
Age * Satis(5075%) -1.082*** -1.081*** -1.079***
(-14.86) (-14.91) (-14.92)
Age * Satis(> 75%) -0.760*** -0.850*** -0.853***
(-15.92) (-10.93) (-13.99)
Age * Satis(> 75%) * WOM 0.142
(1.36)
Age * Satis(< 25%) * WOM -0.14
(-0.95)
Age * Satis(> 75%) * WOM(> 75%) 0.178
(1.90)
Age * Satis(< 25%) * WOM(> 75%) -0.153
(-1.01)
N 4443 4464 4464 4464
R-sq 0.786 0.779 0.78 0.781
t statistics in parentheses. Satisfaction represents satisfaction rate as demeaned variable. ’Satis’ denotes
dummies for each percentile. WOM is driven from the regression of the number of reviews at the opening
week on opening weekend box-office revenues. WOM takes one if the residual from this regression is
positive, zero otherwise. WOM(>75%) takes one if residual belongs to over 75 percentile. *Significant at
5%, **at 1%, ***at 0.1%
71
If the volume of user reviews implies the intensity of word-of-mouth, then we should
observe higher impact of word-of-mouth for movies receiving more reviews. Instead of using
the volume of user reviews for the intensity of word-of-mouth, I regress the volume of user
reviews at week 1 on box-office revenues at week 1 and take its residual as an variable to repre-
sent the intensity of word-of-mouth. Positive residual implies that a movie has relatively larger
volume of reviews compared to other movies having the same box-office sales at week 1.
Table 2.6 represents regression results with this residual denoted as WOM. To consider non-
linear effect of WOM regarding to satisfaction rate, I use dummies for quartile groups by satis-
faction rate. In column (3), I include interaction of age, dummies for highest and lowest quartile
groups of satisfaction rate with WOM. Although insignificant, sign of coefficient is consistent to
the theory. With higher volume of user reviews, highest quartile group of satisfaction rate has
even lower decay rate and lowest quartile has even faster decay rate, although both are insignif-
icant. This result suggests that the volume of user reviews has nonlinear effect on movie sales,
requiring a caution in the use of the volume of user reviews for the study of word-of-mouth.
2.5 Conclusion
This paper investigates the effect of word-of-mouth on box-office sales in the Korean movie
industry. Movie’s performance at opening weekend is largely determined by pre-expectation
which are often quite different from its true appeal. Under the presence of word-of-mouth,
movies with higher quality would have slower rate of decay in weekly box-office revenues than
movies with lower quality but with the same pre-expectation. Using online user ratings and
a new dataset about pre-expectation on coming movies and satisfaction rate on movies shown
in theaters, I find that movies of higher quality have lower rate of decay in weekly box-office
revenues than movies of lower quality but the same level of pre-expectation. I interpret this
result as the evidence of strong effect of word-of-mouth on movie sales in theaters. My estimates
show that word-of-mouth amounts to 24% of total box-office sales and 68% of revenues from
the second week to 8th week when movies at 75th percentile of satisfaction rate are compared
to those at 25th percentile of satisfaction rate, assuming that their pre-expectation is same at
50th percentile. This finding is consistent to what they predicts as well as widespread belief that
word-of-mouth plays an important role in movie sales.
72
The contributions of this paper are twofold. First, I use a new dataset from the Korean
movie industry enabling me to estimate the net effect of word-of-mouth under the control of
pre-expectation. Some of previous studies use online user ratings before its release as a proxy
for pre-expectation, but its volume is quite limited and more importantly it has a severe selec-
tion issue. Another contribution of this paper is that I find the valence of online user ratings
significant in explaining movie sales, which existing papers fail to show in movie industry. I
think this discrepancy results from the difference in methodologies used in papers, but it might
reflect the difference between the Korean movie industry and US market. It is desired to use the
same method for the study of US market in future research.
73
Bibliography
Anderson, E. W. (1998, August). Customer Satisfaction and Word of Mouth. Journal of Service
Research1(1), 5–17.
Banerjee, A. (1992). A simple model of herd behavior. TheQuarterlyJournalofEconomics107(3),
797–817.
Bikhchandani, S. and D. Hirshleifer (1992). A theory of fads, fashion, custom, and cultural
change as informational cascades. JournalofpoliticalEconomy100(5), 992–1026.
Chen, Y. (2001, January). On vertical mergers and their competitive effects. RAND Journal of
Economics32(4), 667–685.
Chevalier, J. A. and D. Mayzlin (2006). The Effect of Word of Mouth on Sales : Online Book
Reviews. JournalofMarketingResearchXLIII(August), 345–354.
Chipty, T. (2001). Vertical integration, market foreclosure, and consumer welfare in the cable
television industry. AmericanEconomicReview91(3), 428–453.
Chisholm, D. C., M. S. McMillan, and G. Norman (2010, April). Product differentiation and film-
programming choice: do first-run movie theatres show the same films? Journal of Cultural
Economics34(2), 131–145.
Choi, J. and S. Yi (2000). Vertical foreclosure with the choice of input specifications. RAND
JournalofEconomics (July 1996), 717–743.
Corts, K. (2001). The strategic effects of vertical market structure: Common agency and division-
alization in the US motion picture industry. JournalofEconomics&ManagementStrategy10(4),
509–528.
De Vany, A. (2004). Hollywood Economics : How Extreme Uncertainty Shapes the Film Industry.
Routledge.
Dellarocas, C. (2007). Exploring the value of online product reviews in forecasting sales: The
case of motion pictures. JournalofInteractiveMarketing21(4), 23–45.
Duan, W., B. Gu, and A. B. Whinston (2008, November). Do online reviews matter? An empir-
ical investigation of panel data. DecisionSupportSystems45(4), 1007–1016.
Einav, L. (2007, March). Seasonality in the U.S. motion picture industry. The RAND Journal of
Economics38(1), 127–145.
Eliashberg, J. and S. Shugan (1997). Film critics: Influencers or predictors? The Journal of Mar-
keting61(2), 68–78.
74
Forbes, S. J. and M. Lederman (2009, December). Adaptation and Vertical Integration in the
Airline Industry. AmericanEconomicReview99(5), 1831–1849.
Ford, G. and J. Jackson (1997). Horizontal concentration and vertical integration in the cable
television industry. ReviewofIndustrialOrganization12(4), 501–518.
Fu, W. W. (2009, June). Screen Survival of Movies at Competitive Theaters: Vertical and Hori-
zontal Integration in a Spatially Differentiated Market. JournalofMediaEconomics22(2), 59–80.
Gil, R. (2007, August). Make-or-Buy in Movies : integration and ex-post renegotiation. Inter-
nationalJournalofIndustrialOrganization25(4), 643–655.
Gil, R. (2008). Revenue sharing distortions and vertical integration in the movie industry.Journal
ofLaw,Economics,andOrganization25(2), 579.
Goolsbee, A. (2007). Vertical Integration and the Market for Broadcast and Cable Television
Programming.
Hart, O., J. Tirole, D. W. Carlton, and O. E. Williamson (1990). Vertical Integration and Market
Foreclosure. BrookingsPapersonEconomicActivity.Microeconomics1990(1990), 205.
Hastings, J. (2004). Vertical relationships and competition in retail gasoline markets: Empirical
evidence from contract changes in southern california. American Economic Review 94(1), 317–
328.
Lafontaine, F. and M. Slade (2007). Vertical integration and firm boundaries: the evidence.
JournalofEconomicLiteratureXLV(September), 629–685.
Liu, Y. (2006). Word-of-mouth for movies: Its dynamics and impact on box office revenue.
JournalofMarketing70(3), 74–89.
Moretti, E. (2011, January). Social Learning and Peer Effects in Consumption: Evidence from
Movie Sales. TheReviewofEconomicStudies78(1), 356–393.
Orbach, B. (2007). Uniform prices for differentiated goods: The case of the movie-theater indus-
try. InternationalReviewofLawandEconomics.
Orbach, B. and L. Einav (2007). Uniform prices for differentiated goods: The case of the movie-
theater industry. InternationalReviewofLawandEconomics (337).
Ordover, J., G. Saloner, and S. Salop (1990, February). Equilibrium Vertical Foreclosure.American
EconomicReview80(1), 127–142.
Ravid, S. (1999). Information, Blockbusters, and Stars: A Study of the Film Industry*. TheJournal
ofBusiness72(4), 463–492.
Reinstein, D. and C. M. Snyder (2005). The influence of expert reviews on consumer demand
for experience goods: a case study of movie critics. The journal of industrial economics 53(1),
27–51.
Rey, P . and J. Tirole (2007). A primer on foreclosure. Handbookofindustrialorganization3.
Salinger, M. (1988). Vertical mergers and market foreclosure. The Quarterly Journal of Eco-
nomics103(2), 345–356.
75
Waterman, D. and A. Weiss (1996, May). The effects of vertical integration between cable tele-
vision systems and pay cable networks. JournalofEconometrics72(1-2), 357–395.
Zhu, F. and X. Zhang (2010). Impact of online consumer reviews on sales: The moderating role
of product and consumer characteristics. JournalofMarketing74(March), 133–148.
76
Appendix
77
Table A1: The Effects of Vertical Integration on the Length of Movie Run: Restricted sample
(1) (2) (3) (4) (5) (6) (7) (8)
Own movie 0.0870* 0.122*** 0.100*** 0.114*** 0.124*** 0.0866*** 0.0984*** 0.0629***
(2.05) (3.76) (3.31) (5.23) (23.89) (10.22) (11.36) (6.09)
Own X Company op 0.0172 0.0162 0.0187 0.0341*** 0.0330*** 0.0338***
(0.97) (0.91) (1.02) (4.01) (3.91) (4.01)
Own X High season 0.0237 0.0201 0.0075 0.0451*** 0.0560*** 0.0597***
(0.40) (0.33) (0.12) (4.60) (5.62) (6.01)
Own X Pre-Expectation -0.00918* -0.00741** -0.00246*** -0.00416***
(-2.54) (-2.62) (-5.38) (-7.34)
Own X Satisfaction rate 0.00407* 0.00356* -0.000393 0.0000933
(2.24) (2.09) (-1.41) (0.33)
Own X Domestic movies 0.0783 0.0779***
(1.08) (6.50)
Pre-Expectation 0.0259*** 0.0258*** 0.0250*** 0.0250***
(11.01) (10.99) (10.50) (10.47)
Satsifaction rate 0.00828*** 0.00826*** 0.00851*** 0.00851***
(6.80) (6.79) (7.10) (7.10)
Company op -0.0544*** -0.0543*** -0.0545*** -0.0521***
(-5.14) (-5.13) (-5.14) (-5.24)
Number of screens 0.0279*** 0.0279*** 0.0278*** 0.0278***
(11.29) (11.29) (11.16) (11.15)
Movie and Theater FE No No No No Yes Yes Yes Yes
Observations 33835 33835 33835 33835 33942 33942 33835 33835
Adjusted R-squared 0.639 0.639 0.637 0.637 0.804 0.804 0.805 0.805
This table reports OLS coefficients where dependent variable is log of total days of movie run. All of specifications count with 206 movies
released in 2008. Robust standard errors are clustered by movie and theater in columns (1)(4). Movie and theater fixed effects are included in
the rest half of the table. Pre-expectation and satisfaction rate are measured as deviations from their mean values. t statistics are in parentheses.
*Significant at 5%, **at 1%, ***at 0.1%
78
Table A2: The Effects of Vertical Integration on Movie Stopping Decision using OLS
(1) (2) (3) (4) (5) (6) (7)
Own movie -0.0245

-0.0268

-0.0419

-0.939

-0.0558

-0.0556

-0.0441

(-7.72) (-7.88) (-8.55) (-12.91) (-10.82) (-11.58) (-5.60)
Number of screens -0.00484

-0.00796

-0.0148

-0.0150

(-5.51) (-9.78) (-10.49) (-10.76)
Pre-Expectation -0.00768

-0.0125

-0.0135

(-66.55) (-44.04) (-40.81)
Satisfaction rate -0.00316

-0.00622

-0.00608

(-47.50) (-52.08) (-50.55)
Weeks since release 0.112

0.112

0.149

0.142

0.141

(34.59) (34.13) (112.00) (73.22) (73.10)
Own X Pre Expecation 0.00482

0.000274
(10.04) (0.72)
Own X Satisfaction rate -0.000383 0.000842

(-1.88) (3.21)
Own X Company op 0.0174

-0.0184

(2.19) (-2.36)
Movie and Theater FE No No No No Yes Yes Yes
Week Dummies Yes Yes Yes Yes No Yes Yes
Observations 89334 89153 89153 89153 89334 89334 89153
AdjustedR
2
0.147 0.195 0.318 0.320 0.334 0.423 0.423
This table reports OLS coefficients. Dependent variable isCUT
ijt
taking one if theaterj drops moviei at weekt, zero otherwise. In
columns (2), I include variables related to movie quality, and these variables are found to reduce the chance to stop movie run. When
I add the age of movie - the weeks after movie’s release - and interactions of vertical integration with movie quality, the results do
not change. In columns (5)(7), I use movie and theater fixed effects to control for possible difference in revenues across movies and
theaters. Week dummies are also included in columns (6) and (7), enabling me to control for the variation in opportunity costs related
to newly released movies over weeks. All of specifications suggest that integrated theaters are less likely to stop their own movies than
nonintegrated theaters and other rival integrated theaters do. The sample includes movies released in 2008. Robust standard errors are
in parentheses and clustered by movie in columns (1)(4). *Significant at 5%, **at 1%, ***at 0.1% 79
Table A3: The Effects of Vertical Integration on Film Choice Decision using OLS
(1) (2) (3) (4) (5) (6) (7) (8)
Own movie 0.0608*** 0.213** 0.0612*** 0.0614*** 0.0716*** 0.125*** 0.0711*** 0.0953***
(5.48) (2.86) (18.58) (3.49) (4.90) (4.71) (12.87) (10.01)
Own X Company op -0.0317 -0.0225*** -0.0118 0.0005
(-1.90) (-3.99) (-0.55) (-0.06)
Own X Average of user ratings -0.0194 0.00165
(-1.86) (0.69)
Own X High season 0.0191 0.000886 -0.137*** -0.0162
(0.68) (0.13) (-3.56) (-1.57)
Own X Pre Expectation 0.00272 -0.00345***
(1.12) (-7.60)
Own X Satisfaction rate -0.00124 -0.000698*
(-1.16) (-2.41)
Average of user ratings 0.00503 0.00658 0.0179 0.018
(0.66) (0.85) (1.40) (1.43)
Number of screens 0.0580*** 0.0580*** 0.0532*** 0.0527***
(13.06) (13.06) (10.33) (10.29)
Company op 0.0197 0.0234* 0.00808 0.0104
(1.91) (2.11) (0.63) (0.75)
Action 0.107*** 0.106*** 0.110** 0.110**
(4.38) (4.37) (3.02) (3.09)
Thriller 0.0769** 0.0772** 0.115* 0.117**
(2.71) (2.72) (2.52) (2.58)
Movie and Theater FE No No Yes Yes No No Yes Yes
Observations 109935 109935 109935 109935 43015 42810 43015 42810
Adjusted R-squared 0.215 0.215 0.447 0.447 0.194 0.196 0.415 0.416
This table reposts OLS coefficients. The dependent variable is a dummy variable that takes value one if the theater show the movie, and zero oth-
erwise. The first half of table counts with all of 590 movies released during the entire data period, while the second half counts 206 movies released
in 2008. Pre-expectation and satisfaction rate are measured as deviations from their mean values. Robust standard errors are in parentheses and
clustered by movie in regressions without fixed effects (columns (1),(2),(5),(6)). *Significant at 5%, **at 1%, ***at 0.1%
80

Abstract (if available)

Abstract This dissertation estimates the effects of vertical integration and word‐of‐mouth on product sales using the data from the Korean movie industry. ❧ In chapter 1, I examine exhibition behavior of movie theaters in the Korean movie industry in order to investigate the influence of vertical integration on competition. I focus specifically on the choice of films, screen allocation, and movie run stopping over different vertical structures. Because, in the Korean movie industry, not only can we observe the same movie being shown in both integrated theaters and unintegrated theaters but also observe the same theater showing movies from distributors of different vertical structures, I use movie and theater fixed effects to control for the unobserved quality of movies and theaters. The empirical results suggest that vertically integrated theaters are more likely to choose their affiliated movies than other competing movies, and they choose them more often than other competing theaters do. In addition, integrated theaters give their own movies a greater number of screenings over longer time periods. This effect is mostly restricted to company operated theaters, and it is greater when movies are expected to get positive word‐of‐mouth as well as when underlying demand is high such as holidays. I argue that these results are not driven by the matching between movie and theater based on anything other than integration status, and that vertical integration leads to the foreclosure, denial of access, of independent distributors to integrated theaters, to the detriment of consumers. ❧ Chapter 2 analyzes the effects of word‐of‐mouth on box‐office sales, another important feature in movie industry. In general, consumers often rely on the information from their peers and other sources like internet sites when product quality is uncertain before its purchase so that word‐of‐mouth is believed to be one of the most influential factors in product sales. In this chapter, I quantify the effect of word‐of‐mouth on weekly box‐office revenues in the context of the Korean movie industry. Using online user ratings and reviews as well as a new dataset including pre‐expectation rate and satisfaction rate for each movie, I find strong evidence that word-of-mouth is significant in movie business. My estimates imply that word‐of‐mouth explains 24% of total box‐office sales and 68% of sales from the second week when we compare movies at 75th percentile of satisfaction rate to movies at 25th percentile, assuming that they have the same level of pre‐expectation.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Three essays on supply chain networks and R&D investments

PDF

Essays on pricing and contracting

PDF

Behavioral approaches to industrial organization

PDF

Three essays on agent’s strategic behavior on online trading market

PDF

Essays on economics of education

PDF

Essays on competition and antitrust issues in the airline industry

PDF

Vertical integration and two-sided market pricing: evidence from the video game industry

PDF

Essays on price determinants in the Los Angeles housing market

PDF

Essays on econometrics

PDF

Essays on development economics

PDF

Essays on competition between multiproduct firms

PDF

Essays in the economics of education and conflict

PDF

Essays on consumer conversations in social media

PDF

The EITC, labor supply, and child development

PDF

Essays on the empirics of risk and time preferences in Indonesia

PDF

Essays on causal inference

PDF

Two essays in economic policy: The influence of perceived comparative need on financial subsidy requests; &, Unintended consequences of the FOSTA-SESTA legislation

PDF

Essays on family and labor economics

PDF

The influence of new media marketing public relations on the South Korean film industry -- in relation to the U.S. film industry

PDF

Effects of economic incentives on creative project-based networks: communication, collaboration and change in the American film industry, 1998-2010

Asset Metadata

Creator Hwang, Yusun (author)

Core Title Empirical essays on industrial organization

School College of Letters, Arts and Sciences

Degree Doctor of Philosophy

Degree Program Economics

Publication Date 07/09/2014

Defense Date 04/22/2014

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag movie industry,OAI-PMH Harvest,vertical foreclosure,vertical integration,word‐of‐mouth

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Ridder, Geert (committee chair), Aronson, Jonathan (committee member), Wilkie, Simon J. (committee member)

Creator Email chehwang@gmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c3-432982

Unique identifier UC11286846

Identifier etd-HwangYusun-2637.pdf (filename),usctheses-c3-432982 (legacy record id)

Legacy Identifier etd-HwangYusun-2637.pdf

Dmrecord 432982

Document Type Dissertation

Format application/pdf (imt)

Rights Hwang, Yusun

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