You are on page 1of 21

The Strategic Effects of Vertical Market Structure: Common Agency and Divisionalization in the US Motion Picture Industry

Kenneth S. Corts
Harvard Business School, Harvard University, Boston, MA 02163

I examine the release-date scheduling of all motion pictures that went into wide release in the US in 1995 and 1996 to investigate the effects of vertical market structure on competition. The evidence suggests that complex vertical structures involving multiple upstream or downstream rms generally do not achieve ef cient outcomes in movie scheduling. In addition, analysis of the data suggests that the production divisions of the major studios act as integrated parts of the studio, rather than as independent competing rms.

1. Introduction
A wide variety of vertical structures arise in different industries: for example, a small set of highly integrated rms historically dominated the steel industry, a few big suppliers sell key components to many fragmented assemblers in the personal computer industry, and a few large Hollywood studios distribute the lms made by numerous small producers in the motion picture industry. The literature offers many possible explanations for why these vertical structures arise, including up- and downstream scale economies, the mitigation of transaction costs, strategic motivations, and so on. However, it is not clear whether or how the con guration of a vertical structure—a particular rm and its up- and downstream trading partners—affects competition. At one extreme, Coasian reasoning suggests that with complete contracts feasible among all parties in a vertical structure, the ef cient outcome (for that structure) should always prevail, regardless of the number of up- and downstream rms involved. At the other end of
I thank Rob Gertner, Bob Kennedy, Darwin Neher, Jan Rivkin, Fiona Scott Morton, Brian Silverman, Mike Whinston, and seminar participants at Harvard, Cornell, Yale, UCLA, the 1998 AEA meetings, the US Department of Justice, and the Canadian Competition Bureau for helpful comments on earlier versions of this work. Michael McCole and Michael Passante provided excellent research assistance.
© 2001 Massachusetts Institute of Technology. Journal of Economics & Management Strategy, Volume 10, Number 4, Winter 2001, 509–528


Journal of Economics & Management Strategy

the spectrum, extremely incomplete contracts would yield inef cient outcomes except possibly (ignoring intra rm agency issues) in fully integrated structures. Intermediate cases include a world in which only bilateral contracts are feasible, where a vertical structure composed of a single upstream rm and a single downstream rm would achieve the ef cient outcome (for that structure), though more complex structures would not. Empirically, little is known about how vertical market structure affects competition and limits or facilitates the implementation of ef cient outcomes. Moreover, it may not even be clear in practice what constitutes a distinct rm in a vertical structure, given the multitude of corporate structures complicated by various divisionalization schemes, alliances, and so on. This paper empirically explores competition in the US motion picture industry to illuminate the effect of vertical market structure on competition. In particular, it addresses two questions: First, do vertical structures that involve multiple upand downstream rms achieve ef cient outcomes for the structure; Second, do divisionalized rms act like fully integrated rms, or like competing independent rms? The empirical evidence suggests that, in this industry, these more complex vertical structures generally do not achieve ef cient outcomes for the structure, and that divisionalized rms generally behave like integrated rms, not like competitors. In their rst paper on common agency, Bernheim and Whinston (1985) argue that rms might implement a jointly ef cient outcome by delegating marketing and pricing decisions to a common sales agent, recognizing that such vertical contracts are more likely to escape antitrust scrutiny than horizontal contracts that achieve the same end. Bernheim and Whinston show that, in equilibrium, rms use sell-out contracts to give the common agent the full residual pro t claim, thereby inducing the common agent to internalize the competitive externalities and choose the joint pro t-maximizing outcome. Thus, Bernheim and Whinston show that bilateral vertical contracts with a common agent suf ce to implement the ef cient outcome. If horizontal contracts are prohibited, then their results suggest that vertical market structure does matter: rms without common agents do not achieve jointly ef cient outcomes, while rms with common agents do. In more general settings, bilateral contracts with a common agent may be insuf cient to achieve ef cient outcomes. In a second paper on common agency, in which the model includes noisy outcomes and risk-averse agents, Bernheim and Whinston (1986) show that when the second-best jointly ef cient contract is not a sell-out contract, an inef cient outcome results. Intuitively, this is because externalities are

Segal (1999) shows that when the terms of a contract with one rm affect the reservation values of other rms. O’Brien and Shaffer (1992). should not be scheduled in a way that internalizes the negative externalities and thus should be relatively close together. and should therefore be farther apart. speci cally on release-date scheduling. does not achieve an outcome that is ef cient for the vertical structure. because the contract-accepting rm cannot be compensated for externalities that may be imposed on it by other rms’ contracts.S. bilateral contracts will not lead to ef cient outcomes. bilateral contracting does not implement ef cient outcomes. since demand is highly variable (with big spikes at holidays. In all of these cases. Speci cally. should be scheduled in a way that is ef cient for the vertical structure. depending on how one counts). do divisions compete like independent rms. bilateral vertical contracting. is it enough to be jointly distributed or jointly produced? Is having one rm in common suf cient to implement the ef cient scheduling decision through bilateral vertical contracts. since the rm making public contract offers will distort the terms of its offers in order to induce contract acceptances by other rms on more favorable terms.U. Finally. or do they jointly maximize like an integrated rm? Is a vertical structure composed of Dis- . a pair of lms that are both jointly produced and jointly distributed should certainly have these externalities internalized. whose lms are distributed through 13 distributors. The fact that the production operations of most of the major studios are divisionalized and/or work in collaborative agreements with other production companies offers an opportunity to address an additional question about vertical market structures. Pursuing a similar theme. many of whom have their own (sometimes multiple) in-house production companies. and Segal (1999) demonstrate that when contract offers are unobservable to nonparties. as in Bernheim and Whinston’s rst commonagency model? The evidence suggests that it is not. McAfee and Schwartz (1994). Motion Picture Industry 511 present at the ef cient solution. even with common agents. It is easiest to conceptualize the test by focusing on two extreme cases: a pair of purely competitive lms. for example) and movies that are released close together are likely to impose negative externalities on each other’s revenues. which have neither a producer nor a distributor in common. in the US motion picture industry in 1995 and 1996. This paper explores the effect of vertical market structure on competition. This industry features many production companies (as many as 80 during this period. The basic empirical test in this paper focuses on how vertical market structure affects the scheduling of a pair of lms. The question that remains is then. One important strategic variable in this industry is the release date.

Section 2 describes the motion picture industry. Paramount. and also the contracts between them. write. Generally.. Universal. 2. In contrast. Section 3 describes the data and presents descriptive statistics. and that in fact the multiple divisions of a studio act more like a single integrated entity than like competing rms. and the importance of release-date scheduling.1 Types of Firms The process of making a major motion picture is a long and complicated one. They acquire. Baye et al. For present purposes. involving many different entities. because a signi cant theoretical literature (Schwartz and Thompson. the typical contracts between producers and distributors. . Had eld (1991) and Corts and Neher (2000) argue that this assumption may not be justi ed. Section 5 concludes and explains why some alternative explanations for these results are not satisfactory. a fully integrated structure or a structure of common agency? This is an important question. The evidence suggests that they are.512 Journal of Economics & Management Strategy ney’s distribution arm and the production companies Hollywood Pictures and Touchstone Pictures. The US Motion Picture Industry 2.). Some studios have several production companies that operate under different names—Sony’s Columbia and TriStar. and so on. An empirical test similar to the one described above sheds light on the divisionalization question by assessing whether lms that are jointly distributed and produced by different divisions of the same studio are scheduled more ef ciently than purely competitive lms. 1996) has argued that divisionalization may be an important commitment device because the divisions will compete ercely among themselves and not behave as an integrated rm. and edit a screenplay. or Disney’s Touchstone and Hollywood. and promotes and advertises the lm. 1986. physically reproduces and distributes the lm. Sony. they cast and contract with actors. and Section 4 presents the empirical analysis and results. they physically shoot the lm footage. especially when rms may renegotiate or sign unobservable contracts with their divisions. etc. production companies are responsible for the creative aspects of making a movie. it is important to understand broadly the roles of producers and distributors. Distribution companies are typically part of a major studio (Warner Brothers. for example. but all studios also run substantial production companies. both owned by Disney. The distributor contracts with theaters for exhibition of the lm.

S.nancing/distribution agreements. The distributor. Because the lm is a more well-de ned product at the time of the transaction under these two forms. the distribution agreement is not signed until the lm is “in the can. While the rst four forms involve print and advertising (P&A) funding from the studio/distributor. i. or an independent nancing entity is likely to retain control of the copyright if it provides production funding. also has a partial claim to pro ts through the structure of standard contracts. There are at least two distinct sets of rights to a lm that are at stake in such a relationship: the ownership of the copyright to the lm. an agreement to distribute the lm is reached before principal photography.. The copyright owner has a clear claim to residual pro ts. as it shares the net proceeds after distribution fees and expenses with actors. potentially securing more favorable terms. directors.” production nancing having been arranged through other sources. which becomes involved at a correspondingly later stage of the production process. The most common form for a distribution deal . Motion Picture Industry 513 2. an individual producer. and the rights to distribute it in a speci c territory.e. An independent production company. which are closely related to rst-run success in the U. This agreement is then used in negotiating production nancing from third parties. production. The rst two are distinguished by whether the lm is already “packaged” when acquired by the distributor. 1997): in-house production/distribution. The studio/distributor is more likely to control the lm’s copyright the earlier and more signi cant is its involvement in the lm’s production and funding. however. whether the underlying story has been acquired and a screenplay and outline prepared. The copyright owner also earns pro ts through distribution in foreign territories and subsequent fees for cable and broadcast television exhibition. In an acquisition deal. and other pro t participants. In this order. only the rst two forms involve production funding from the studio/distributor. and especially under the latter. which may be assigned to other parties by the lm’s owner. The third and fourth are distinguished by whether the lm is completed before the acquisition occurs.2 Contractual Relationships Financing and distribution arrangements between motion picture producers and distributors take one of ve basic forms (Cones. acquisition deals. negative pickups. these arrangements involve decreasing levels of nancial involvement from a major studio/distributor. it is easier for the producer to shop the lm around and solicit competitive offers from rival distributors. and rent-a-distributor deals.U. market.S. In a negative pickup deal.

this gure increased from 27% in 1990 to 34% in 1997. states atly: “If you don’t pick the right release date. leading to crowding of releases in peak periods and a dearth of lms in off-peak periods. Barry Reardon.. Trade journals and industry executives term this clustering of lms “self-destructive” and “a nightmare for all parties” (Anon. Large seasonal uctuations in demand. As discussed in the introduction. Over that same period. then recoups its print and advertising expenses from the remaining sum before distributing the net proceeds to the production company and other pro t participants. Financial agreements with theaters normally give the lmmaker a greater percentage of the box of ce during the rst weeks of release. 2. 1996b). except of course in the case of a vertically integrated studio that both produces and distributes a particular lm. studio executives also worry that theaters will replace a lm with another if it doesn’t win audiences quickly. 1997)..514 Journal of Economics & Management Strategy (the contract governing theatrical exhibition in a speci c territory) is the net deal. sellout contracts allocating the full residual pro t claim to a single party are basically nonexistent. in part because a lm’s opening weekend is usually the most lucrative one for its studio. the fraction of a lm’s box of ce take accounted for by its opening week increased steadily throughout the 1990s. In fact. And in this glutted market. The conventional wisdom on release-date competition is limited to the observation that this competition seems excessive from a jointpro ts point of view. 1991) asserts that studio executives insist the release date is critical. And the Wall Street Journal (Anon. 1991). Thus.. According to Variety (Anon. and favorable contractual provisions for opening weekends together lead to intense competition for the best weekend release slots. the total accounted for by the opening weekend alone rose from 20% to 25%.3 Demand Cyclicality and Release-Date Competition I focus speci cally on one important dimension of competition in this industry: the battle for favorable release dates. you can destroy a movie” (Anon. the absence of such contracts tends to undermine the role of common agency in achieving ef cient outcomes for the rms. Warner Brothers’ president of distribution. in which the distributor collects a distribution fee of (typically) 30% of gross rentals.. The Wall Street Journal came to the following conclusion when confronted with weak box of ce totals in . media attention to hits.

and Christmas–New Year. compared to a competitive benchmark. Labor Day.. and generally falls in the 10–20% range. the period to be studied here begins with the fth weekend of 1995 and extends through the fourth weekend of 1997. Because the window containing the Christmas and New Year’s holidays extends through week 4 of the following year. Method 1 partitions the weeks of the year into 10 annual windows by starting a new window every time total revenue reaches a trough in Murphy’s data. but less so than if they . Thanksgiving. One could consider every pair of lms in the data.1 Seasonality Attendance patterns in the US motion picture industry are highly seasonal. Such perspectives support the idea that ef cient outcomes for a vertical structure involve reduced clustering of its lms. the most salient case of competitive business-stealing occurs when two lms are released on the same weekend. The ratio of the standard deviation to the mean ranges from 3% for fall window 1 in 1995 to 41% for Christmas–New Year 1996–1997. This division of the year into demand windows is important because the empirical analysis focuses on pairs of lms. 3. Easter. This method creates a total of 16 seasons. but it is unlikely that all of these pairs are of equal interest. The US Motion Picture Industry: Data 3. Independence Day.D. This creates 20 windows of varying length over the course of the two years of data. Murphy’s historical index [reported in Vogel (1994)] of average lm attendance to divide the calendar year into distinct demand windows by two different criteria. and demonstrates the presence of signi cant variation in demand both between and across windows. Motion Picture Industry 515 the summer of 1995: “What went wrong? Executives blame too many expensive movies stacked too close together at the beginning of the season. Method 2 creates eight seasons per year by centering 5-week windows on each of the eight peaks identi ed by Murphy: President’s Day. The result was that one big movie was ’cannibalized’ by the next one” (Anon. I rely on A.S. For example. with marked demand spikes coinciding with holiday weekends and school vacations. Memorial Day. To analyze lm pairs. For each of these window de nitions. Table I shows the mean and standard deviation of weekly total US box of ce revenue. Films one week apart presumably cut into each other’s attendance. Midsummer.U. one must de ne the set of lms from which pairs will be created. 1995). covering a strict subset of the weeks included in method 1. by window.

8 27. dev.5 16.5 98.2 81.8 83.0 .3 52.5 79.4 74.1 71.3 12.5 75.5 91.8 46.0 72.8 51.5 10.5 7.2 4.6 18.6 16. Revenue ($ million) Memorial Day Fall 1 Fall 2 4th of July Midsummer Labor Day Thanksgiving Christmas– New Year Weekly Total Box-Of ce Revenue by Window Method Year Statistic Presidents’ Day Easter 1 1995 1996 Mean Std.4 4.5 22.8 46.2 13.4 18.5 53.7 18.9 20.3 3.3 74.0 95.5 10.3 3.5 53. Mean Std. Mean Std.5 81.0 26. 60.6 81.6 30.1 2 1995 Journal of Economics & Management Strategy 1996 Mean Std.7 82.5 48.3 50.0 9.0 6.5 7.4 5.2 4.1 26.2 3. dev.9 22.0 10.5 7.2 9.6 56.6 81. dev.0 1.2 76.5 49.1 71.2 55.8 10.2 46.516 TABLE I.7 95. dev.6 4.5 51.8 10.8 23.1 57.2 95.2 72.0 32. 60.

Because method 2 excludes some weeks altogether (mostly in the fall when demand is relatively steady). The 300 lms in the dataset were released by a total of 13 distributors.U. First. but the de nition employed here is that of copyright owner. This yields a set of 300 lms for analysis. distributor. I focus on lms that reach wide release at some point in their theatrical run. Wide release is understood in the industry to be achieved when a lm is playing on at least 600 screens simultaneously. some lms open earlier in the week. These are assigned to the rst weekend they are in wide release. Two complications arise. Release dates of these lms are typically straightforward to determine. for which release-timing issues are likely to be most salient. the number of lms they released. for example. some lms open on a limited number of screens and then ramp up to wide release. July 4. as most major lms open on a large number of screens on a Friday. which is appealing in that it is the entity that retains the net pro ts from the domestic distribution deal as well all rights associated with exhibition . Independence Day. 3. and producer of all major lms released in 1995 and 1996 (and the rst four weeks of 1997) were obtained from the box-of ce reports in The Hollywood Reporter. but is coded as opening on the following Friday. The use of the term “producer” in the lm industry varies widely. genre. especially around holidays. These are coded as if they had opened on the rst weekend they surpassed 600 screens. and so on. July 2. To focus the analysis on the lms with the broadest appeal. a pair of lms released at Memorial Day of 1995 and at Thanksgiving of 1996 will enter the analysis with the same implicit importance as a pair of lms released on the same weekend. but mitigates head-to-head competition with major holiday releases. Second. especially near the end of the year. I address this issue by de ning demand windows and then pairing movies only when both are released in the same window. when it abruptly broadened its showing to 1268 screens. for example. Ghosts of Mississippi . If one studies all possible pairs within the sample. This yields 2524 pairs of movies under method 1.2 Film Characteristics The title. and 1611 pairs under method 2. it generates fewer pairs than method 1. These distributors. and the numbers of producers responsible for these lms are listed in Table II. but is coded as opening on the rst weekend of 1997.S. played on 21 screens for the last two weeks of 1996. Motion Picture Industry 517 were released head to head. when a limited opening secures Academy Award candidacy in the previous year. opened on Wednesday. 1996.

for example.). Films by Distributor Distributor Buena Vista Sony Warner Brothers Universal Paramount MGM New Line Fox Miramax Gramercy Savoy Orion Goldwyn a An No. this de nition implies. that a lm made outside the studio system with private nancing.518 Journal of Economics & Management Strategy TABLE II. This can be an individual who coordinates a project (Don Simpson and Jerry Bruckheimer coproduced Crimson Tide). The ultimate corporate parents of the production companies were determined by reference to Who Owns Whom and other corporate directories. Inc. but subsequently acquired outright by a major studio/distributor after production. in which lms are categorized as produced by the parent company of the lead producer.. a major studio (Warner Brothers produced Space Jam). (3) number of parent companies of lead producers.. in foreign territories and other media. This will be referred to hereafter as narrow producer de nitions. of Films In-Housea 39 32 21 31 28 19 23 20 9 2 2 1 1 No. b De nitions of distinct producers: (1) number of distinct combinations of production companies. or an entity established to fund a speci c lm (Last of the Dogmen. Counted by these broader de nitions. is not independently produced. then this number falls to 58. There is often joint ownership of a lm’s copyright by several rms or individuals. it will prove fruitful to employ broad producer de nitions. produced The Last of the Dogmen). but rather is the property of the studio. In addition. of Films 50 48 35 31 30 26 25 23 15 7 7 2 1 No. (2) number of distinct lead producers. being economic and not operational in its derivation. of Producersb (1) 15 9 13 4 7 9 6 4 8 3 4 2 1 (2) 10 9 7 1 5 7 5 4 7 3 4 2 1 (3) 6 8 6 1 3 7 4 4 7 3 4 2 1 in-house lm is de ned here as one for which the lead producer’s parent company is also the parent company of the distributor. If one counts only distinct lead producers (often a studio that is coproducing a lm with an independent entity) as different entities. Also. 42 . Treating every distinct combination of such rms as a different entity yields a count of 80 producers. an independent studio like Steven Spielberg’s Amblin Entertainment (which coproduced To Wong Foo.

however. Arnold Schwarzenegger. if big-budget lms tend to be released on the same peak weekends. stars is set equal to one if both lms featured stars as de ned above. Motion Picture Industry 519 production entities are responsible for the 300 lms being studied. action. . horror. The Hollywood Reporter assigns each lm to one of nine genres. gap 1). sg is set equal to one if both lms in a pair are classi ed as the same genre. 3. The models were also estimated using a different classi cation system of genres. Mel Gibson. Table II also shows the number of lms produced in-house by each distributor’s production companies. gap is the number of weeks between the two releases (e.g.U. Demi Moore.. Tom Hanks. drama. 2. Two control variables account for the extent of competition between the lms. animation. For the present purposes. In addition. Michael Douglas. for two lms released on consecutive weekends. The number of production companies responsible for each distributor’s lms. Robin Williams. then stars would be expected to have a positive coef cient when gap is the dependent variable. Only once during this period did a rm with a distribution arm produce a movie but distribute it through another rm: Mallrats was produced by Universal but distributed by Gramercy. family/animation. thriller. Harrison Ford.1 Two variables control for other lm characteristics. romance.S. by each of these de nitions. 2 The cast of each lm was determined by reference to the Internet database IMDb:. that is. If blockbusters are spaced further apart to mitigate competition. is given in Table II. drama/romance. Julia Roberts. one of the 17 highestpaid actors (all those whose asking price in late 1995 was at least $12 million) according to Entertainment Weekly (1996). 3. and comedy. Sylvester Stallone. family. These actors were: Jim Carrey. Kevin Costner. I determined whether each lm featured a major star . Sean Connery. and Bruce Willis. timing issues aside. John Travolta. then stars would be expected to have a negative coef cient. the most interesting aspect of a lm pair is the relationship between the companies involved in producing and distributing each lm. Tom Cruise. Eddie Murphy. Clint Eastwood. The results were qualitativel y similar to those reported here. and western. Dummy variables for the nature of the 1. I constructed a measure of the temporal proximity of two lms’ release. in which they were pooled to form only four: action/horror/thriller/western. in order of decreasing prevalence they are: comedy. the coef cient on sg is expected to be positive when gap is the dependent variable.3 Since lms in the same genre are likely to be in closer competition.3 Characteristics of Film Pairs Having gathered the above data on lms released in 1995 and 1996 and having paired all possible lm combinations within demand windows.

are used separately. dummies for each demand window. which is consistent with the hypothesis of joint scheduling of those lms. . so that a pair falls into sdsp or ddsp if the lms’ lead producers have a common corporate parent. or of the mitigation of scheduling competition. same producer) and a value of 0 for the other variables: dddp. sdsp2 equals one if the lms have the same producer according to the wide de nition given above. The Tobit models account for the fact that gap is constrained to be between zero and the maximum number of weeks in that particular window. and whether they both feature stars. 4. whether they are in the same genre. the broad producer de nition is used. To account both for heteroskedasticity that might arise from differences in window characteristics and for the correlation of errors induced by the inclusion of lms in more than one pair. but not the narrow de nition. two additional variables. or that the vertical structure has come closer to achieving the jointly ef cient outcome.)” option. In the rst set of regressions. sdsp1 and sdsp2.1 Empirical Model I model the gap between two lms as a linear function of sg.520 Journal of Economics & Management Strategy relationship are labeled according to whether the pair has the same (s) or different (d) distributor (d) and producer (p). Two basic models are presented: one investigates the role of joint distribution and joint production. sdsp1 equals one if the lms have the same producer according to the narrow de nitions of production companies given above. Table III shows the distribution of movie pairs by window according to the relationships between their distributors and producers. where clustering by demand window permits correlation of the errors within windows but constrains errors to be independent across windows. Thus. Each model is estimated four different ways: both as OLS and as a two-sided Tobit. as well as the same distributor. and sddp. Huber-White robust standard errors are presented in all cases (White. for each of the two window de nitions.4 4. stars. and some measure of the relationship between the rms involved with those two lms. and one investigates the role of divisionalization. ddsp. 1980). Empirical Analysis 4. a pair of lms that share both a distributor and a production company has a value of 1 for sdsp (same distributor. In the second set of regressions. as well as the same distributor (thus sdsp sdsp1 sdsp2). Robust standard errors were calculated by Stata’s “robust cluster(. A positive coef cient on the relationship variable indicates that relationship leads to less clustering.

TABLE III.36 1.32 0. Films Narrow 3 12 0 2 3 2 0 4 1 6 2 23 2 2 4 4 0 3 0 6 209 79 182 9 48 2 5 10 16 1 9 0 19 28 90 7 10 31 20 2 33 1 42 444 3 0 1 21 1 0 1 1 0 6 65 Continued.S.50 1.81 1.67 1.92 0.66 1. sg Gap Method 5 9 4 6 5 4 3 5 3 8 5 9 4 6 5 4 3 5 3 8 Year Window No.82 2.11 2.52 1.32 2.18 0. Pairs No.92 Broad stars Mean Std.84 1..15 1.08 0.94 1.89 2.97 0. SP No.93 1.90 1.52 1.67 1. Films by Window No.58 3.77 5 19 3 3 5 5 2 4 1 16 13 42 13 7 14 10 10 21 2 48 0 0 6 10 1 0 0 10 3 1 1.37 1.01 0.41 0. 14 25 13 11 14 14 11 15 9 24 15 32 8 11 18 15 7 16 5 23 300 2524 105 496 28 55 153 105 21 120 10 253 14 41 2 6 14 12 0 11 0 22 91 300 78 55 91 91 55 105 36 276 10 22 3 4 8 6 2 7 2 23 No.81 1.39 1. Weeks U.25 1.82 1.43 0. sd No.05 1.14 1. 1.. Motion Picture Industry 1 1995 President’s Day Easter Memorial Day Independence Day Midsummer Labor Day Fall1 Fall2 Thanksgiving Christmas–New Year’s 1996 President’s Day Easter Memorial Day Independence Day Midsummer Labor Day Fall1 Fall2 Thanksgiving Christmas–New Year’s Total 104 521 .74 1.00 2. Dev.01 0.28 1.05 1.

Pairs 2 1995 1996 President’s Day Easter Memorial Day Independence Day Midsummer Labor Day Thanksgiving Christmas President’s Day Easter Memorial Day Independence Day Midsummer Labor Day Thanksgiving Christmas 230 1611 143 44 111 284 45 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 14 16 14 10 14 16 11 18 15 17 10 12 18 20 10 15 91 120 91 45 91 120 55 153 105 136 45 66 153 190 45 105 10 10 4 3 8 11 4 11 14 7 5 6 14 24 3 9 3 3 0 1 3 3 1 4 2 4 5 2 4 6 2 1 5 7 3 2 5 8 2 6 9 10 5 3 10 24 3 9 13 14 16 4 14 15 5 28 28 31 16 10 31 34 8 17 0 0 6 10 1 0 6 1 3 0 3 10 1 0 3 1 1. Dev.82 1.62 1.. No. sg Gap Std.42 1.07 1.58 1.25 1.06 1.67 1.24 1.76 1.25 1.94 1.47 1.82 1.17 0.87 1.91 1.11 1.32 1.75 1.52 1.58 1.34 Journal of Economics & Management Strategy Total 80 .12 1.20 1. sd Narrow Broad stars Mean No.31 1.10 1.05 1..81 1.82 1.28 1. Method Year Window No. Films No. Continued.522 TABLE III. SP No.58 1. Weeks No.

3022# 0. 5%.1362 0.0653 0.1577 0.2122 0.2996 0.0628 0.15 0.0987 0.0000 5.0156 5.0223 10%.1016 0.S. This is consistent with the expectation that.1540 0. indicating that lms that are in the same genre are released about 0.2353 0.2556 0.0233 11. and in particular whether it has . *. This is consistent with the scheduling of blockbuster lms near holiday weekends.2 The Effect of Joint Distribution and Joint Production Table IV presents the basic results on the effect of vertical market structure on scheduling competition.2836 1611 Yes Test of joint signi cance F Pr > F Wald Â2 Pr > Â2 F Pr > F Â2 Pr > Â2 a Robust 23. star-laden lms. rms try to mitigate the clustering of lms with close rivals.18 0.0826 0.7750 0. 1%.3337 0. 523 The Effect of Joint Control Method 1 Variable ddsp sddp sdsp sg stars Number of obs. which would tend to cluster big-budget.5981 0.U. other things equal.2048 0.33 0.2510 0. The control variable sg is signi cant at 5% in all four speci cations.1044 0. standard errors below coef cient estimates. Motion Picture Industry TABLE IV. its point estimate is relatively stable and negative.94 0.1451 0.2296 0.1801 0. the point estimates are positive. Window dummies OLS 0.7359 0.92 0. #.0279 5.1427 0.2306 1611 Yes Method 2 Tobit 0.2469 0.1185 0.1517 0. Recall that the fundamental question is whether vertical market structure has an effect on competition. 4.0089 0. The control variable stars is not signi cant in any speci cation.22 0.0875 0.1305 0.3073 2524 Yes OLS 0.2259 0.0000 351.0787 0. Signi cance levels: **.0001 60.06 0.2368 2524 Yes Tobit 0.75 0.0507 0.5819 0.0000 Tests of sddp sdsp 7.1542 0.2 weeks further apart.1588 0. Dependent variable: gap .

since competition is not mitigated. Thus. since sddp = sdsp can be rejected at 5% in all four speci cations. dddp is the excluded dummy variable. consider common distribution. consider joint production. coef cients on the other rmrelationship dummies measure the extent to which those relationships increase the gap between lms (reduce clustering) relative to this maximal-clustering benchmark. or at least more ef cient outcomes than purely competitive structures? Is one common agent enough to achieve ef cient outcomes. While the full joint control of two lms (sdsp 1) should allow the vertical structure to achieve an ef cient outcome. Second. compared to a mean gap value of 1. Now the real question of interest arises: Do intermediate vertical structures—structures that are neither purely competitive nor purely bilateral combinations of one upstream and one downstream rm— achieve ef cient outcomes.75 weeks in the Tobit models. These regressions separate the lms . Second. as in the rst Bernheim-Whinston model? The results in Table IV address this in several ways.524 Journal of Economics & Management Strategy any predictive power for the relative closeness of two lms’ release dates. which is signi cant at 1% for all four speci cations. First. which ranges from about 0. since the coef cient on sddp is positive and sometimes signi cant. its coef cient can con rm two statements that have thus far been only assertions. assuming that antitrust laws do in fact preclude horizontal contracts that would implement joint-pro tmaximizing outcomes. common distribution alone does not match the ef ciency of outcomes achieved by bilateral structures. First.6 weeks in the OLS estimates to about 0. depending on the severity of the various contracting problems discussed in the introduction. Since ddsp is never signi cantly different from zero. In the models presented in Table IV. lms with no joint control (dddp 1) should exhibit maximal clustering. Since the fully jointly controlled category (sdsp 1) proxies in a sense for the ef cient outcome. However. The results suggest that common distribution alone may help achieve ef cient outcomes somewhat. 4. the claim that ef cient outcomes involve less clustering is con rmed by the positive coef cient on SDSP. At the other extreme.3 The Effect of Divisionalization Table V presents results from slightly different regressions that emphasize the role of divisionalization. other structures may or may not. the assertion that this is important in this industry can be supported by the magnitude of the effect. there is no evidence that common production alone aids in implementing ef cient outcomes. for both window de nitions.7 for both window de nitions.

2481 0. and stars is insigni cant with a negative point estimate.40 0.2549 0. but rather how one should think of what constitutes common production.28 0.2433 2524 Yes Tobit 0.0343 0.1579 0.3565 0.2501 0. The divisionalization literature discussed in the introduction maintains that divisions with a common parent behave like .1781 0.6895 13.4426 0.2449 0. 5%.2139 0. Both control variables’ coef cients have about the same magnitude as in the previous regressions.3032# 0.2911 1611 Yes Test of joint signi cance F Pr > F Wald Â2 Pr > Â2 F Pr > F Â2 Pr > Â2 a Robust 9.1108 0.3761 0. 10%.63 0. the question is not whether a vertical structure with only common distribution or common production achieves ef cient outcomes.1362 0.2420 0.0821 0. to highlight this distinction and to facilitate the appropriate statistical tests.0796 0.1789 0.3137 2524 Yes OLS 0.1766 0. As before.1003 0.2048 0.2053 0.8174 0.0813 #.0000 2.2823 0.1451 0. the control variable sg is positive and signi cant.2237 0.5154 0.2996 0. Here.04 0.2789 0.2525 0.96 0.0000 79. sddp is the excluded dummy variable.1256 3. Window dummies OLS 0. standard errors below coef cient estimates. that are jointly distributed and jointly produced by the same division (sdsp1 1) from lms that are jointly distributed but jointly produced by different divisions of the same studio (sdsp2 1).0000 Tests of sdsp1 sdsp2 0.1950 0.01 0.61 0. The Effect of Divisionalizationa Method 1 Variable dddp ddsp sdsp1 sdsp2 sg stars Number of obs.6040 0.0001 355.2556 0.1615 0.2356 1611 Yes Method 2 Tobit 0.2022 0. *.1187 0. Signi cance levels: **. 1%.1541 0.0933 0.2147 0. Motion Picture Industry 525 TABLE V.2828 0.2118 0.6093 0.1908 0.S.U.2056 0. In addition.16 0.0621 0. Dependent variable: gap .

. One is capacity constraints. suggesting that these divisions behave essentially the same as an integrated rm or a single division. 5. However. 1986. the hypothesis that such vertical structures act just like vertical structures that include only a single division (sdsp2 sdsp1) cannot be rejected. distribute. they tend to reduce the clustering of their lms’ releases. However. 1992. except at 10% in one speci cation. in contrast to the assumption of much of the divisionalization literature (Schwartz and Thompson. If a distributor can only effectively schedule. not due to the mitigation of competitive externalities. with point estimates suggesting that common corporate parenthood increases the gap between such lms by at least half a week. 1994. since such structures’ lms are more clustered than those that are fully controlled by a single upstream and a single downstream rm. The evidence suggests that when vertical structures internalize competitive externalities across a set of products. One can imagine a number of alternative explanations for the observation of reduced clustering among sets of lms. . this is rejected at 5% in all four speci cations. 1996). 1999). Segal. in this context that implies the equality of the coefcients on sdsp2 and sddp. then jointly distributed lms will tend to be less clustered due simply to these cost considerations. this can be tested by simply looking at the signi cance of the sdsp2 coef cient. 1986. If this coef cient were zero. In addition. multiple divisions of the large studios seem to internalize the externalities across their products. since the latter is the excluded category. due to numerous and diverse contracting problems (Bernheim and Whinston. Conclusion This paper describes the in uence of vertical market structure on one facet of competition—release-date scheduling—in the US motion picture industry. This is consistent with various theoretical papers that demonstrate that in many circumstances bilateral contracting in multilateral vertical relationships fails to implement the vertical structure’s ef cient outcome. and promote a certain number of lms at a time. it would imply that vertical structures with joint distribution and joint production only by the broad de nition (only because they share a corporate parent) are no more able to achieve ef cient outcomes than vertical structures that truly share only common distribution. O’Brien and Shaffer. neither joint production nor joint distribution alone appears to be effective in achieving ef cient outcomes for the vertical structure. Baye et al. This suggests that they behave more like an integrated rm than like independent competing rms. McAfee and Schwartz. In addition.526 Journal of Economics & Management Strategy independent rms.

“Bose-Einstein Dynamics and Adaptive Contracting in the Motion Picture Industry. and D. 1996. neither seems likely to explain these results fully. Walls.” American Economic Review. raises the question of why a studio’s independent production companies’ lm are less clustered than other jointly distributed lms. aside from any concerns about competitive externalities. While both of these explanations are dif cult to refute in view of the data limitations. 2000. since such capacity constraints should affect all jointly distributed lms. 86. then that distributor’s lms will tend to be less clustered. 1985. Ju. 13. IL: Southern Illinois University Press. leaving substantial questions about competition and contracting between rms in this industry. The Feature Film Distribution Deal. Bernheim. Anon.” Harvard Business School Working Paper. further work is needed both to integrate the various strategic models of vertical market structure discussed here and to further develop an understanding of this particular industry. this seems unlikely. 1996. 16. 923–942. and Divestiture Incentives in Oligopoly. “Credible Delegation. just as observably similar lms (in the same genre) were shown to be less clustered. Carbondale. “Divisionalization . J. and Organization.. In addition. 1996a. p. given the full set of ndings... 1986. “Value Judgments: We Crunch Hollywood’s Numbers to Tell You Who’s Really Worth What. such considerations seem more likely to affect broader scheduling considerations—how many lms to release this summer. when those production companies are typically set up precisely in order to increase the breadth of lms produced in-house. Franchising.” Journal of Law. Whinston. Corts.S. Chisholm. A second alternative explanation is unobserved correlation in lm characteristics. however. 1493–1514. DeVany. D.. If a particular distributor tends to distribute similar types of lms. Crocker. Economic research on the motion picture industry is generally limited to the analysis of demand (DeVany and Walls.” Economic Journal. References Baye. and M. and J. for example. “Common Agency. and D. April 12. rather than how many to release this weekend. 223–236. 1997. Economics. “Pro t-Sharing versus Fixed-Payment Contracts: Evidence from the Motion Picture Industry. not just those that are both jointly distributed and jointly produced. This explanation. 269–281. 1997). and . “Common Marketing Agency as a Device for Facilitating Collusion. Motion Picture Industry 527 However. D. 30.U.” Econometrica .” Entertainment Weekly. Cones. 1996) and the analysis of contracts with stars (Chisholm. 54. Many questions remain unanswered. K. which is where the effect is empirically most pronounced.” RAND Journal of Economics. M. A. 169–201. 1997. K. . Neher. 106.

p. 817–830. 23.” Wall Street Journal.. Thompson. 1999. G. Entertainment Industry Economics. 210–230. I. 1991. . Cambridge: Cambridge University Press. Bannon. 1997. 1996b... D. King. August 11. Klady. 22. B1. 1994.” Variety.. 1. R. Leonard. 48. Shaffer.” Variety .” RAND Journal of Economics. “Vertical Control with Bilateral Contracts.” Quarterly Journal of Economics. H. Vogel.” RAND Journal of Economics.. “A Heteroskedasticity-Consistent Covariance Matrix Estimator and a Direct Test for Heteroskedasticity. Lisa. 299–308.” Wall Street Journal. 531–543. “For Movies. O’Brien. 101. 307–321. 64. Schwartz.” Quarterly Journal of Economics. 1992. 1991. 1994. p. “Divisionalizatio n and Entry Deterrence. Thomas R. and G. Exclusivity.” Econometrica . M. Hindes. “Contracting with Externalities. B1. 1980. 84. and E. “Opportunism in Multilateral Vertical Contracting: Nondiscrimination. H. White. “Trying to Enhance New Films’ Prospects. “Credible Spatial Preemption through Franchising. Schwartz. April 7–13.” American Economic Review. It’s the Dog Days of Summer. 40. and Uniformity. September 23–29. Studios Increasingly Shift Release Dates. 337–388. “Hollywood’s Holiday Havoc. McAfee. and M. Segal. p. ”Pic Jam Leads to Hit and Run. Andrew. p. October 15. 1995.528 Journal of Economics & Management Strategy Had eld. 1986.