Counterfactual Decomposition Of Movie Star Effects With Star Selection

Counterfactual Decomposition of Movie Star Effects with Star Selection 汇报人:凡志明

01 Introduction 02 03 Models with Star Selection 04 05 Literature Review and Conceptual Freamwork Descriptive Statistics Results 06 Contributions and Limitation

01 Introduction

0 1 Introduction Whether stars indeed contribute to the success of a movie? There have positive relationship between star and movie's finncial outcome. Stars do not increase a movie's ROI or the value of the movie company. To what extent do stars actually contribute to the success of movies? the number of theaters allocated in opening week opening week revenue

01 Introduction Two difficulties Star-movie matching process is unrandom and the star variable has endogenous director producer production companies shooting location MPAA ratings star movie suitability to the role mitigate the risks of movie outcome Account for the endogenous by employing an endogenous switching model,producing a selection correction term

01 Introduction A second difficulty in the assessment of star effects arises from potentially systematic differences between the star and nonstar movie characteristics characteristics effects counterfactual decomposition QR star return effects Authors use a quantile regression framework to estimate the star effects on theaters and revenue. A QR approach is less susceptible to outliers and heavy tails. The counterfactual decompositions using a QR approach provide a more complete picture of how the contributions of the star and the movie characteristics might differ at different levels of the theater and revenue distributions.

01 Introduction control for the star selection bias as well as other endogeneities commonly present in the movie research when estimating the theater and revenue models summary take a distributional perspective and employ QR to investigate the star effectsacross the entire movie outcome distributions counterfactually decompose the star versus nonstarmovie outcome differences into movie characteristics effects and star return effects, after controlling for other endogeneities

02 Literature Review and Conceptual Freamwork

02 Literature Review and Conceptual Freamwork There are two themes in researching star effects Why studios appoint a star and why stars agree to appear in a movie Signaling explanation Rent captue Other research investigates the star effects on different outcomes of a movie in the U.S. as well as in international markets The star’s involvement increases the number of theaters allocated to amovie throughout its theatrical life. The positive effect of stars on profitability is limited. The presence of stars has a positive and significant impact on the viewership in the U.S. market, but less so in other countries.

02 Literature Review and Conceptual Freamwork

03 Models with Star Selection

0 3 Models with Star Selection Theater Model i Movie, 1,2,.Nj; j 1 if the movie contains a star, and 0 otherwise; Xij denotes a vector of characteristics of movieij ; α1j and α2j denote respective returns on budget and movie characteristics that nonstar and star movies generate Independent variables: sequel Friday genre seasonally major studio budget Log(Budget) endogenous

03 Models with Star Selection Revenue Model β1j and β2j denote respective returns on opening week theaters and movie characteristics that nonstar and star movies generate; Wij is similar to Xij except that this vector additionally includes critics's reviews Independent variables: sequel Friday genre seasonally major studio critics' reviews open theater Log(Open Theater) endogenous

0 3 Models with Star Selection Star Selection Model Two-step Heckman correction construct the star selection model Independent variables: PowerProducer PowerDirector PowerProduction Ek is the latent utility that drives an actor to participate in Foreign movie project k ij , Zk is the movie-related information Rated R available; ηk includes the unobservables IV-Star include higher-order terms of the inverse Mills ratios,λi1(γ) ϕ(Zi1γ)/Φ(Zi1γ) and λi0(γ) -ϕ(Zi0γ)/[1Φ(Zi0γ)] ,then specify the conditional θth quantile of the error term in Equations (1) and (2), respectively, as

0 3 Models with Star Selection construct the terms in Equations (4) and (5) with the estimated first-stage parameters and append them to Equations (1) and (2)

0 3 Models with Star Selection An Alternative Approach(CF approach) probit model D 1 if the movie contains a star, 0 otherwise;ϕ(·)is the cumulative distribution function of a standard normal distribution second stage Y is the second-stage dependent variable, X denotes movie characteristics, V denotes the error in the first-stage star selection equation, and V is replaced by a residual to construct the control function,

0 3 Models with Star Selection Counterfactual Decomposition The difference in mean revenues of the two groups can be decomposed as follows characteristic effect star return effect let YS and YN be the revenues of a star movie and a nonstar movie; XS and XN denotes the characteristics of star versus nonstar movies ; βS and βN denotes the return that a star (versus a nonstar) can generate on this characteristic

0 3 Models with Star Selection Decompositions at different levels of the outcome distribution star return effect characteristic effect Qθ(XS,βθS ) denotes the unconditional θth quantile of the distribution of the outcomes of the movies that have the star movies’ characteristics profile (distribution) and the star movies’ returns on the characteristics Qθ(XN,βθN ) is defined for the nonstar movies and nonstar returns Qθ(XS,βθN ) represents the th quantile of the countefactual distribution of the revenues of the movies that have the star movies’ characteristics profile but have returns resembling those of nonstar movies

0 3 Models with Star Selection Operationalization of “Star” In this study, authors use the annual trade magazine Premiere's “Power 100 list”, published typically in the May or June issue and listing Hollywood's most powerful people.Any actor/actress who appears on that year's list is regarded as a “star” for the rest of that year after the list's publication and the year after that. There are 71 movie actors/actresses in total who make up our final Premiere star lists and 233 star movies and 777 nonstar movies between year 2001 and 2005.

04 Descriptive Statistics

0 4 Descriptive Statistics More star movies are directed by award-winning directors, produced by award-winning producers, and produced by big six production companies. There are fewer foreign movies with stars than with nonstars. The average budget of star movies is twice as largeas that of nonstar movies. Major studios tend to employ more stars. There are greater proportions of movies in the action and adventure genres for the star movie subsample, and a smaller proportion of drama and thrillers.

0 4 Descriptive Statistics A large number of movies (about 20% of the sample), mostly made up of nonstar movies, are allocated 10 theaters. More than 70% of star movies were allocated 2,000 or more theaters. 32% of all movies, mostly nonstar movies, having less than 1 million in revenue in the opening week.

0 4 Descriptive Statistics Star movies have higher values than nonstar movies across whole distributions for both DVs, suggesting that star movies outperform nonstar movies on both metrics. Moreover, generally there are wider gaps between the star movies and the nonstar movies at the lower tail than those at the upper tail of the distributions.

05 Results

0 5 Results Star Selection Model All the regressors have significant coefficients with expected signs. The likelihood of a star's accepting a movie decreases if a movie is a foreign movie or if it is expected to receive an R rating by the MPAA.

0 5 Results Star and Movie Characteristics Effects on Theater The star selection parameters The effect of production by at (inverse Mills) are significant major studiossuggesting on openingthe week all quantiles, For the nonstar subsample, the theater allocations of nonstar presence of significant selection effects budget onand open theater moviesof isthe positive significant bias.For pooled sample, theall are positive and significant at at all quantiles under the pooled The effects of Comedy, star control functions are levels of the theater distribution analysis. For the separate Romance, and Thriller aresample significant at the 25th and the under both methods. Rated R at movies are less widely analysis, again the major studio positive nearly all quantiles of 50th quantiles. distributed thantheater otherfor MPAA effect on open isnonstar stronger theater distributions rated below the median. at themovies lower quantiles. movies.

0 5 Results The coefficients for the star sample (Table 4) are generally StarsThe generally star effects do notare generate negative greater or nonsignificant returns the for In the nonsignificant romance, thriller at allon quantiles. Stars do notforadd to the comedy, effects ofand major studio on moviemovies budgetbelow andstar other the median movie characteristics but become positive than do and and pooled analysis, the interactions with the comedy, romance, opening week theatrical releases. nonstars. significant at the median andand upper quantiles. thriller genres are generally negative significant.

0 5 Results Star and Movie Characteristics Effects on Revenue The Theinverse Sequels openhave theater Mills a significant ratio elasticity (andand to therevenue positive star control remains effect function) on at the about parameters are The effect of Metascore is positive and significant and is nearly mostly 1 at allnonsignificant, revenues quantiles of the nonstar of revenue suggesting movies, distribution. but thatonly staratselection the 15thbias and is not a concern in the same (0.02–0.03) at all levels of the revenue distribution. the 75th revenue percentiles. analysis.

0 5 Results Theremethods Both is also no suggest evidence thatofthe stars starproducing effect is nonexistent, any incremental except returns at on the lowesttheaters allocated quantile. and movies characteristics.The presence of a star actually depletes the revenues of movies in the drama genre and those rated R.

0 5 Results Counterfactual Decomposition The star return versuseffect nonstar gap significant in opening and characteristics is not The total difference in the star versus week theater allocations is wider for at effects any of arethe statistically quantile significant suggesting in all The estimated star andlevels, nonstar theater nonstar effects on opening week movies thenot bottom that the quantiles, starsindo the gap quantiles, is mostly allocation gap is impact positive and and the revenues is positive and significant difference declines the upper significant across theinentire distribution. Log(Open Revenue) explained by the star return directly. effect. between the 20th and 30th percentiles. quantiles.

0 5 Results

06 Contributions and Limitations

0 6 Contributiions and Limitations Estimated a star selection model to account for star endogeneity. Contributions QR approach provides a more complete picture of how the contributions of the star and the movie characteristics might differ at different levels of the theater and revenue distributions The counterfactual methodology combining with QR with the starselection correction utilized in this study is the first to be used in the marketing literature.

0 6 Contributions and Limitations a more complete analysis of star effect should consider a movie’s performance during its entire theatrical run need an expgenous variable in the Star Selection Model limitations not distinguish between movies with a single star and multiple stars stars may have different effects on international markets and DVD releases than on the U.S. opening week performance

Thanks

Star-movie matching process is unrandom and the star variable has endogenous star movie director producer production companies shooting location MPAA ratings suitability to the role mitigate the risks of movie outcome Two difficulties Account for the endogenous by employing an endogenous switching model,producing a selection correction term .