Influence of Star Power on Movie Revenue

Influence of Star Power on Movie Revenue Taewan Kim, Assistant Professor of Marketing, College of Business and Economics, Lehigh University, USA. E-mail: tak213@lehigh.edu Sang-Uk Jung, Assistant Professor of Marketing, College of Business, Hankuk University of Foreign Studies, South Korea. E-mail: sanguk.jung@hufs.ac.kr Dong Hyun Son, Assistant Professor of Accounting, College of Business, Hankuk University of Foreign Studies, South Korea. E-mail: dson@hufs.ac.kr Abstract Among many movies released every year only a few of them are successful and would achieve box office revenue. Weather to cast a movie star or not is a classical question for movie makers and the answers for this question are mixed. By using two-way cluster corrected 2SLS models and 3SLS models, our empirical models in a motion picture industry investigate the effect of star power on the early stage of movie revenue. Results show that number of screen and having an Oscar nominated superstar, Oscar winning superstar in the movie have a positive and significant effect on the box office revenue. Our findings provide a rational that the star power plays an important role for box office revenue. 433

1. Introduction Only a few movies are successful in a movie industry due to the complexity of market environment. One of the many factors that ensures that the movie can be a potential blockbuster is the star power (Elberse 2007, Lehmann and Weinberg 2000, Vogel 2007). Weather to cast a movie star or not is a classical question for movie makers; however, the answers for this question are mixed. We study empirical models in a motion picture industry to investigate the effects of star power on the early stage of movie revenue. Many academic papers investigate the effects of star power on box office revenue, however, the results seem unclear. Our goal of this paper is to contribute to the literature by identifying the effects of star power on the first week of box office revenue. 2. Related Literatures and Hypotheses Several related literature documents various factors influencing the box office revenues of movies. Elberse (2007) finds that a star participation indeed positively affects movies revenues; specifically, stars can be worth several millions of dollars in revenue. Moreover, the author shows that important determinants of the magnitude of that effect: stars prior performance in an economic and artistic sense and the number and prior performance of other star cast members. Eliasberg, Jonker, Sawhney and Wierenga (2000) shows that MOVIEMOD produces reasonably accurate forecasts of box office performance. However, this prerelease market evaluation model for a motion picture industry is generated after the movie has been produced but before it has been released. Therefore, even though it has great forecasting method for the early period of the movie release, it does not utilize the first week of box office revenue and other factors which shapes the revenue. Unlike this MOVIEMOD method, we use parsimonious linear models to estimate a box office revenue model including star power variable. Many scholars show that early stage of box office revenue determine the entire movie revenue. Specifically, Lehman and Weinberg (2000) show that the weekly box office revenue for a movie decays exponentially over time. De Vany and Walls (1999, 2004) also show that the box office revenue follows Pareto (heavy-tailed) distribution different from a general normal distribution. Therefore, we are focusing on the first opening week of the box office revenue for the new movie released instead of aggregated box office revenue. This is the unique difference between related literature and our paper. The main explanatory variable is a star power for our empirical research. Many diverse approaches are end up with multiple different results (Albert 1998; De Vany and Walls 1999; Elberse and Eliasberg 2003; Karniouchina 2011; Ravid 1999; Liu, Mazumdar, and Li 2014). Utilizing a switching model to account for endogenous assignment of stars and non-stars into respective movie samples, Liu, Mazumdar, and Li (2014) develop a complex and sophisticated 434

model and show that the star effects on revenue come indirectly through the theater allocations as well as from the characteristics of the movies in which they participate. De Vany and Walls (1999) suggest a model to investigate if the employment of movie star shifts the revenue distributions. They show that star involvement significantly increases the number of theaters allocated to a movie throughout the product life-cycle, and that this effect becomes even more noticeable in the later periods. However, De Vany and Walls (1999) and Ravid (1999) find that having a star does not increase the probability of the movie being a successful work, and the star beneficial star effect is only limited to a few stars. Elberse and Eliasberg (2003) shows that casting a superstar in a movie has no effect on the number of screens but has a significant positive effect on opening week revenue. Elberse (2007) shows that the announcements of a star cast increases HSX Movie Stock prices, a simulated stock market for movies, on the announcement day and the average cumulative abnormal returns are different from stars. This study is focusing on the investors reaction to the casting stars not on the revenue and cost of casting stars. Thus, we develop the following hypotheses: H1a: Casting a star actor/actress who will be nominated at Oscar Award in the movie produces greater box office revenue H1b: Casting a star actor/actress who will win at Oscar Award in the movie produces greater box office revenue Elberse and Eliasberg (2003) shows that stars are found to have significant positive effects on screen decisions in Germany, Spain, and France. We expect that the more number of screen showing the movie in the first opening week generate the greater possibility to have more number of ticket sales because of the greater availability (Davis 2006). The smaller traveling cost attenuate the disutility of going to movie theater (Hotelling 1929). The more number of screen showing the movie in the opening week provide the lower probability to have sold out, therefore, the higher ticket sales can be generated if the movie is successful. We therefore propose the following hypothesis: H2: Number of screen showing the movie in the first opening week has positive effect on box office revenue. Motion Picture Association of America (MPAA)-ratings categorize films according to its appropriates to several groups of audiences as follows; general audiences (G; All ages admitted. Nothing that would offend parents for viewing by children), parental guidance suggested (PG; Some material may not be suitable for children), parents strongly cautioned (PG13; Some material may be inappropriate for children under 13), restricted (R; Under 17 requires accompanying parent or adult guardian; Contains some adult material; Parents are urged to learn more about the film before taking their young children with them), adults only (NC-17; No one 17 and under admitted; Clearly adult; Children are not admitted). Medved (1992) insists that when the typical PG film generates nearly three times the revenue of the typical R 435

bloodbath or shocker, then the industry s insistence on cranking out more than four times as many R titles must be seen as an irrational and irresponsible habit. Ravid (1999) shows that G-rated movies earned higher average profit than R-rated movies. De Vany and Walls (2002) also argues that there are too many R-rated movies in Hollywood s portfolio. An executive seeking to trim the down-side risk and increase the upside possibilities in a studio s film portfolio could do so by shifting production dollars out of R-rated movies into G-, PG-, and PG13-rated movies. Thus, we posit the following hypothesis: H3: MPAA moderates the effect of the open screen showing the movie in the first opening week on the box office revenue. 3. Variable Definitions, Data and Descriptive Statistics 3.1 Variable Definitions We transform the three major variables such as Open.Box, Open.Screen, Budget by taking natural log. We have two similar measures for capturing star power in the movie; Oscar.N and Oscar.W. Both variables are dummy coded as 1 if the movie casted a movie star who either wins or is nominated Oscar Award in the movie release year, and as 0 otherwise. Genre has 19 different categories includes action, animation, adventure, biography, comedy, crime, documentary, drama, family, fantasy, horror, mystery, romance, Sci-Fi, short, sport, thriller, war, western showing 151 different cases. Distributors include 20 th Century Fox, Artisan, Buena Vista, Captured Light, Columbia, Dimension, DreamWorks SKG, Focus Features, Fox Searchlight, Lion s Gate, Luke Films, MGM/UA, Miramax, New Line, New Market, Palm Pictures, Paramount Pictures, Sony, Sony Classics, United Artists, Universal, Warner Bros. See Table 1 for the detail data description and variable names. 3.2 Data The main source of the data set is The Internet Movie Database also known as IMDB, (http://www.imdb.com/). Data periods starts at May 5, 2003 and ends at December, 29, 2004. We have 235 observations of data points. Data includes movie title, box office revenue in the opening week, Oscar nomination, Oscar winner, budget, number of screen in the opening week, genre, motion picture association of America (MPAA) rating, release date, year, and distributor. See Table 1 for data descriptions and operationalization for major independent and control variables. Especially we carefully collected the Oscar nominated movies and Oscar winning movies. Oscar awards include four relevant categories only (actor in a leading role, actor in a supporting role, actress in a leading role, actress in a supporting role). Each category has five nominees and one of them wins the Oscar award in the category. We code the Oscar.W as 1 for the movie title which casts a movie star who will win the Oscar, 0 otherwise. Similarly, Oscar.N is coded as 1 for the movie casting a movie star who will be nominated at the Oscar. 436

For MPAA ratings, we use the actual numbers e.g., 9, 11, 13, 21 which are the ages that are accepted to enter and enjoy the movie. For example, if MPAA rating is 13, it means this movie has PG-13. Therefore, the higher ratings means the narrower the audience for the movie. This might generate greater variation for the regression models than having a binary dummy coding scheme such that Rated-R and NC-17 are coded as 1, 0 otherwise (Basuroy et al. 2003) 3.3 Descriptive Statistics According to the correlation matrix of main variables, there are negative association for both between Oscar award winner and box office revenue, and between Oscar nomination and box office revenue in the first week, which are statistically significant (p<.05) respectively. Also see Table 2 Correlation for the details. Open screen has strong, positive, and highly significant relationship with the box office revenue in the first week. This supports H2; however, we need to take care of this in order to have unbiased and efficient regression estimates. Thus, we consider 2SLS and 3SLS for the estimation of the models. 4. Models 4.1 2-SLS First of all, there exists high and significant correlation between open box revenue and the number of open screen in the opening week. We utilize the budget variable as an instrumental variable for the second stage least square estimation. Since the budget variable is closely related to the dependent variable, open box revenue, and not directly associated to the open screen variable, we use budget as a valid instrumental variable. Moreover, since there is a possibility that the residuals in the open box revenue model are correlated across time, we consider a two-way genre and distributor cluster and test the statistical significance of the predictor variables using standard errors corrected for the correlated errors (see Cameron, Gelbach, and Miller 2011; Cameron and Miller 2015; Henderson 1990; Thompson 2011; Petersen 2008; Peterson 1989). We obtain the standard errors using Peterson s Stata code,5 robust cluster (cluster), as well as the cluster2 Stata command for estimating two-way standard errors. 4.2 3-SLS We conduct Breusch-Pagan / Cook-Weisberg test for heteroscedasticity. The test result shows that there is a heteroscedasticity issue in the OLS model. It means that at least one of the independent variables variance of residual is increasing with respect to the independent variables. Therefore, we use a 3SLS model to estimated consistent results. 5. Results 5.1 2-SLS By using the variable Budget as an instrumental variable, we control the strong and positive correlation between Open.Scrn and the dependent variable, the box office revenue in the first 437

week. First of all, both coefficient estimates of the main independent variables, Oscar.N and Oscar.W are positive and significant (.768, t=3.89 for the model with Oscar.N and 1.136, t=3.18 for the model with Oscar.W). These support H1a and H1b. In terms of size of these effects on the box office revenue, Oscar Winner has greater effect on the box office revenue than Oscar nomination (Oscar.W=1.136 > Oscar.N=.768). This result seems consistent to the common sense that casting a movie star who will win the Oscar award generates bigger impact on box office revenue than casting a movie star who will be nominated at the Oscar award (Ginsburg, Gutierrez-Navratil, and Prieto-Rodriguez 2016) Next, Open.Scrn variable also has positive and significant coefficient estimate for both models (.987, t=11.11 for the model with Oscar.N and 1.009, t=9.98 for the model with Oscar.W). As we expect in H2, the more number of movie theater showing the movie in the opening week generates the greater box office revenue of the first opening week. Therefore, this supports H2. Third, MPAA has positive but marginally significant relationship with the box office revenue (.024, t=1.57, p<.20 for the model with Oscar.N and.026, t=1.62, p<.20 for the model with Oscar.W). This seems reasonable for understanding that there is no direct relationship between MPAA ratings and the box office revenue. However, the overall trend also shows that the higher rating, the movie has more potential to achieve great buzz and popularity. This fact might be related to the Genre issue which is not clearly control in this paper. Therefore, this supports H3. Last, the overall fit seems very high for both model (Centered R square:.8 and.78 respectively; Uncentered R square:.998 and.998 respectively). This result show that even though we only implement a few important independent variables in the statistical model, those explanatory variables explain about 80% of the data. 5.2 3-SLS Table 4 shows two models each with the star power of Oscar Nomination and that of Oscar Winner. First, focusing on the main variables, casting a star who will win or nominated at the Oscar is positively and significantly associated with the dependent variable, the box office revenue in the opening week (.417, z=2.66 for Oscar.N and.527, z=2.17 for Oscar.W respectively). These results support H1a and H1b, respectively and these are consistent result to the results of 2SLS models. Next, the coefficient estimates of Open.Scrn are positive and significant (.949, z=18.20 for the model with Oscar.N and.956, z=17.14 for the model with Oscar.W). These results support H2 and are consistent to the estimates, however, these are far more significant. Third, budget variable has a positive and significant effect on the number of theater in the opening week (.778, z=7.32 in Model I and.783, z=7.39 in Model II, respectively). 438

MPAA has negative but not significant relationship with Open.Scrn variable which seems reasonable. Rated-R movie has systematically low demand because of the potential audience pool is narrower than other lower rating movies. The coefficient estimates of MPAA ratings are both negative but not significant (-.040, z=-1.49 in Model I and -,038, z=-1.43 in Model II, respectively). Therefore, these results partially support H3. See more the detail in Table 4. Overall model fits seem very well and even much greater than 2SLS models fits (Rsquare=.805 in Model I and R-square=.799 in Model II). Chi-square measure also seems to be very significant (61.87 for Open.Scrn model, 476.19 for the Box Office Revenue; 62.20 for Open.Scrn model, 472.71 for the Box Office Revenue, respectively). 6. Discussion 6.1 Managerial Implications There is no clear result showing whether the star power is beneficial to the box office revenue or not. This paper provides very specific results that the star power is meaningful and important strategic decision variable for movie makers focusing on the first opening week. See the Table 3 for the details. This is the main contribution of this paper. The second contribution is to suggest better model and estimation method for answering research questions including 2SLS with two-way cluster corrected standard errors taking Genre and Distributor as two clusters, and 3SLS models. 6.2 Study Limitations and Future Research This data set is too small to provide generalizable empirical results. However, by focusing on the very first week of the box office movie revenue, we extract clear managerial implications that movie maker has an incentive to cast movie stars who has a better probability to be nominated or to win the Oscar. For the future research opportunity, one might want to think about that correlation between award nomination & winning and Genre. Elberse (2007) finds that stars affect revenues and that some stars contribute more to revenues than others. Lastly, MPAA ratings effect on box office revenue is not clear. This fact might be related to the Genre issue which is not clearly control in this paper. One can run the two-way cluster models with Genre and MPAA ratings. Acknowledgement This work was partially supported by Hankuk University of Foreign Studies. 439

References Albert, S. (1998), Movie stars and the distribution of financially successful films in the motion picture industry, Journal of Cultural Economics, 22(4), 249 270. Basuroy, S., Chatterjee, S. and Ravid, S. A. (2003) How Critical are Critical Reviews? Box Office Effects of Film Critics, Star Power, and Budgets, Journal of Marketing, 67(4), 103-117. Cameron, A. Colin, Jonah B. Gelbach, and Douglas L. Miller (2011), Robust Inference with Multi-Way Clustering, Journal of Business & Economic Statistics, 29 (2), 238 49. Cameron, A. Colin and Douglas L. Miller (2015), A Practitioner s Guide to Cluster-Robust Inference, Journal of Human Resources, 50 (2), 317 72. Davis, P. (2006), Spatial Competition in Retail Markets: Movie Theaters, The RAND Journal of Economics, 37 (4), 964-982. De Vany, A. S. and Walls, W. D. (1999), Uncertainty in the Movie Industry: Does Star Power Reduce the Terror of the Box Office? Journal of Cultural Economics, 23(4), 285-318. De Vany, A. S. and Walls, W. D. (2004), Motion Picture Profit, the Stable Paretian Hypothesis, and the Curse of the Superstar, Journal of Economic Dynamics and Control, 28(6), 1035-1057. Elberse, Anita (2007), The Power of Stars: Do Star Actors Drive the Success of Movies? Journal of Marketing, 71, 102-120. Eliashberg, Jehoshua, Jedid-Hah Jonker, Mohanbir Sawhney and Berend Wierenga (2000), MOVIEMOD:An Implementable Decision-Support System for Prerelease Market Evaluation of Motion Pictures, Marketing Science,19(3), 226-243. Eliashberg, Jehoshua and Steven Shugan (1997), Film Critics: Influencers or Predictors? Journal of Marketing, 61(2), 68-78. Ginsburg, Victor, Fernanda Gutierrez-Navratil, and Juan Prieto-Rodriguez (2016), The Impact of Oscar Nominations and Wins on Box Office Revenue, working paper. Henderson, Glenn V. (1990), Problems and Solutions in Conducting Event Studies, Journal of Risk and Insurance, 57 (2), 282 306. Hotelling, H. (1929), Stability in Competition, Economic Journal, 39, 41-57. Karniouchina, E. V. (2011) Impact of star and movie buzz on motion picture distribution and box office revenue, International Journal of Research in Marketing, 28(1), 62 74. Klady, Leonard (1997), The Long Flat Summer, Variety, (June 23), 5. Kim, Taewan (2014), Star-Power and Movie Revenue, working paper, Lehigh University. Lehman, Donald and Charles Weinberg (2000), Sales through Sequential Distribution Channels: An Application to Movies and Videos, Journal of Marketing, 64 (2), 18-33. Liu, Yong (2006), Word-of-Mouth for Movies: Its Dynamics and Impact on Box Office Revenue, Journal of Marketing, 70 (3), 74-89. Petersen, Mitchell A. (2008), Estimating Standard Errors in Finance Panel Data Sets: Comparing Approaches, Review of Financial Studies, 22 (1), 435 80. Peterson, Pamela (1989), Event Studies: A Review of Issues and Methodology, Quarterly Journal of Business and Economics, 29 (3), 36 66. Ravid, S. A. (1999) Information, blockbusters, and stars: A study of the film industry, Journal of Business, 72(4), 463 492. Thompson, Samuel B. (2011), Simple Formulas for Standard Errors that Cluster by Both Firm and Time, Journal of Financial Economics, 99 (1), 1 10. 440

Vogel, H. (2007), Entertainment Industry Economics A Guide for Financial Analysis (7th Edition), Cambridge University Press, Cambridge. Table 1: Data Descriptions and Operationalizations Variable Description Operationalizations Open.Box Box office revenue in the opening week Natural log Open.Screen Number of theaters showing movie in the opening week Natural log Budget Budget for making and marketing film Natural log Oscar.N Oscar nomination Dummy variable Oscar.W Oscar winner Dummy variable Distributor film distributor Dummy variable MPAA Motion Picture Association of America (MPAA) ratings Categorical Variable Table 2: Correlation Box Office Oscar.N Oscar.W Open.Scrn Budget MPAA Box Office 1 Oscar.N -.31* 1 Oscar.W -.28*.62* 1 Open.Scrn.92* -.42* -.38* 1 Budget.53*.08.03.46* 1 MPAA -.15*.04.06 -.18* -.20* 1 *p=.05 Table 3: 2-SLS Estimation of Star Power on Box Office Revenue (1) (2) Independent Variables Estimates Estimates Oscar.N.768*** (3.89) Oscar.W 1.136*** (3.18) Open.Scrn.987*** 1.009*** (11.11) (9.98) MPAA.024.026 (1.57) (1.62) Constant 8.494*** 8.347*** (10.88) (9.17) F(3,19) 54.17 122.44 p.000.000 Centered R 2.800.788 Uncentered R 2.998.998 Root MSE.709.729 N 207 207 Notes: The dependent variable is a box office revenue. There exists a high correlation between number of open screen and box office revenue so we use the variable Budget as an instrument variable. In order to have the most conservative estimates we use clustering on both Genre and Distributor and the number of clusters for distributor are 151 and 20, respectively. This estimation results using Genre and Distributor as clusters 441

give us the best estimates among other cluster combinations. Other estimation results are available upon requests. We also tried one-way clustering either on Genre or on Distributor. The results are consistent so we only present two-way standard errors here. We use ivreg2 for the two-way cluster corrected standard errors (see Cameron, Gelback, and Miller 2011; Cameron and Miller 2015; Henderson 1990; Thompson 2011; Petersen 2008; Peterson 1989). Table 4: 3- SLS Estimation of Star Power on Box Office Revenue Model I Model II Open.Scrn Box Office Open.Scrn Box Office Independent Variables Estimates Estimates Estimates Estimates Oscar.N.417*** (2.66) Oscar.W.527*** (2.17) Open.Scrn.949***.956*** (18.20) (17.14) Budget.778***.783*** (7.32) (7.39) MPAA -.040 -.038 (-1.49) (-1.43) Constant -5.643*** 9.19*** -5.769*** 9.169*** (-2.88) (23.04) (-2.96) (21.87) X 2 61.87 476.19 62.20 472.71 p.000.000.000.000 R 2.229.805.229.799 N 207 207 207 207 Notes: The dependent variable is a box office revenue. There exists a high correlation between number of open screen and box office revenue so we use three stages least squares method to have consistent estimates while controlling heteroscedasticity. We use reg3 command for the three stage least square model in STATA. 442