Modeling Movie Choice - UCLA Anderson School of Management

[Pages:43]Modeling Movie Choice

July 2003

Andrew Ainslie Xavier Dr?ze Fred Zufryden*

* Andrew Ainslie is an assistant professor of marketing at the University of California, Los Angeles. Xavier Dr?ze is an assistant professor of marketing at the Wharton School of the University of Pennsylvania, Philadelphia. Fred Zufryden is the Ernest W. Hahn professor of marketing at the University of Southern California, Los Angeles. The authors would like to thank The UCLA Center for Communication Policy for financial support, and Olav Sorenson for his help with the data.

Modeling Movie Choice

Abstract In this study, we examine box office sales in the context of a choice model. Rather than studying movies in isolation of each other, as has been done traditionally, we account for the variety of movies available to moviegoers at any given time. This is accomplished by developing a combination of a sliding-window logit model and a Gamma diffusion pattern, in a hierarchicalBayes framework. Using our model, we show that accounting for the full choice set available every week not only increases the fit of weekly movie sales, but also leads to parameter estimates that depict a richer picture of the movie industry. We show that movie studios appear to have a good understanding of the products they produce, knowing when to support them and when not to. We also show that the large studios are correct in their product segmentation strategy of mainstream versus artistic movies and that many studios behave in accordance with Krider and Weinberg's (1998) model of movie release timing.

Our research also indicates that actors have a direct effect on consumer choice. This leads viewers to watch the movie earlier in its release. Directors have a more indirect effect on consumers. Finally, releasing a movie at the same time as other movies of the same genre adversely affects box office performance all around; releasing a movie against movies of the same MPAA rating hurts its sales in the beginning, but there is a displacement effect which means that in the long run sales loss is less severe.

1 Introduction

The American public has long had a fascination with movies. With 462 new movies released, 8.4 billion dollars in domestic box-office sales, and 1.5 billion tickets sold, 2001 was a record year (MPAA 2002). Although total movie production costs for the studios that are members of the MPAA1 seemed to have leveled around $50 million per movie in the last few years, marketing costs are still rising. These costs, which represent the lion's share of production costs, have now reached an average of $31 million per movie (again for the MPAA member companies). Recent films such as "The Matrix Reloaded" and "Harry Potter" have production costs well over $100 million (). The expenditures on production represent a significant investment that the studios seek to recoup, first through box-office release, and then through the sales of pre-recorded tapes and DVDs.

Given the size of the bets made by each studio when they produce and release a movie, the budgets available for marketing purposes, and the frequency with which they release new movies (9 new movies every week on average for 2001 (MPAA 2002)), one would assume that studios have become proficient at predicting movie success. Thus, it might be expected that studios would use tools similar to those of Procter and Gamble or 3M to predict sales before launching one of their products. However, surprisingly, Hollywood has not put much stock in sales prediction models, arguing that movies are artistic creations that cannot be modeled. Thus, the movie industry believes more in instinct and analysis by anecdote (Red Herring 1998).

Studios do, however, believe in the impact of competition. Many movies have had their release dates pushed back, or brought forward, to avoid coming out at the same time as a competing movie which might be a stronger player (see for instance Eonline 2002). This behavior has been shown to be optimal from a theoretical standpoint by Krider and Weinberg

2

(1998) in the presence of a strong seasonal pattern. However, there has been little empirical research incorporating the individual effects of competing movies in the modeling of box office receipts. This can be easily understood as the nature of the business (e.g., hundreds of movies entering and exiting the market every year and a rapid diffusion process over a very short period) makes such modeling efforts difficult. But given the emphasis of studios on competition, we believe that there is a need for a box office sales model that considers movie sales from a choice perspective. We provide such a model in this paper and show, using three years of movie sales data, that looking at movies in a choice context not only leads to a better fit of box office sales, but also to a better interpretation of the drivers of movie choice. We also show that studios seem to follow different release strategies depending on their sizes.

2 Literature review

Recent work on movie box-office modeling has been carried out by Sawhney and Eliashberg (1996). In their BOXMOD approach, they decompose the consumer's decision to see a movie in two steps: first the consumer makes the decision to see a movie, and second, the consumer acts on this decision. By modeling the time-to-decide and the time-to-act as exponential decays, they derive a three-parameter model that can take the form of a General Gamma, an Erlang-2 (timeto-act parameter is equal to time-to-decide parameter), or an Exponential distribution (time-todecide parameter is infinite). They then perform a meta-analysis on the three parameters that allows them to understand what factors drive movie sales (e.g., MPAA rating and movie genre).

Shugan (1998) looks at box office performance based on the team that participated in the creation of the movie (i.e., writers, directors, actors). His goal is to help studios predict box office sales early on during the production process--a time at which the finalized product is not available, but the track record of the production team is known. By looking at the past box office

3

performance of the movies in which the production team was involved Shugan is able to predict opening day box office with an R2 of 0.59 and total box office with an R2 of 0.34.

There has also been some theoretical work on the competitive aspect of movie release. Krider and Weinberg (1998) investigate film release strategies based on the assumption of exogeneity of the highly seasonal nature of the movie business (sales peak during the summer months and during the holiday season) coupled with the shortness of movies' lifecycles. In particular, they suggest that strong movies should compete head to head during peak weeks while weak movies might want to delay their release if they are facing strong competition. Radas and Shugan (1998) also study the timing game. They propose an ingenious approach for handling seasonality through acceleration and deceleration of time.

Finally, some researchers have concentrated on post launch profitability. Swami, Eliashberg, and Weinberg (1999) study the allocation of multiplex screens to movies so as to maximize distributor profits. Neelamegham and Chintagunta (1999) and Elberse and Eliashberg (2002) study the international diffusion of movies.

2.1 Challenges with Movie Sales Forecasting

There are various factors that make modeling movie box-office sales daunting. First, the lifecycle of movies is very short. Most movies last no more than ten weeks at the box office. Second, sales frequently peak in the first week, with rapid decays. Third, the diffusion pattern is not constant across movies. Indeed, there are two broad patterns of diffusion. The blockbuster type movies (64% of the movies in our database) have an exponential-decay type sales pattern with the opening week grossing the largest sales. The sleeper type movies (36% of movies) build sales gradually after launch and generally peak three to six weeks after launch, requiring a model that handles a wider variety of patterns than just exponential.

4

Furthermore, weekly total box office sales exhibit a highly seasonal pattern (e.g., Radas and Shugan 1998). The causes of this seasonality are not entirely clear. Common wisdom in the industry is that seasonality is exogenous. There are good weeks (e.g., 4th of July week-end) during which the American public sees more movies, and bad weeks (e.g., the second week of September) during which it does not. Movie studios vie for good week ownership and signal to each other which weeks they intend on `owning.' However, some industry insiders have questioned this practice (Red Herring 1998). They believe that there is no such thing as a bad week, that the seasonal patterns are endogenous, and that the belief in good and bad weeks is a self-fulfilling prophecy. For instance, the release of Spiderman in May of 2002 was purposely scheduled the week before Labor Day (traditionally viewed as a dead week) on a countermovement effort that claims that there are no bad weeks. The exogeneity of seasonality has also been challenged in an econometric model incorporating the timing decision (Einav 2002), which shows that it may be sub-optimal to compete for peak viewing weekends.

Finally, it is difficult to use data gathered on past movies to predict the success of future movies. This is because the movie industry is very much a `short-term contract' industry, with high-turn-over, where the combination of actors, directors, writers, and studios change frequently. For instance, in 38 movies, Harrison Ford has worked with 12 different studios, 29 directors, and 29 writers. In addition, movies, by design, are created to be different from previous movies. Whereas one often looks for continuity in a product (one buys Tide because it produces the same results every wash), one expects discontinuity from movies. Indeed, being unique is a crucial part of entertainment. Thus, the industry is purposely making it difficult for their audience to predict what is coming next.

5

2.2 Modeling Goals

The main purpose of our model is to study box office performance in an environment where moviegoers choose which movie to see among all the movies release at any one time. Framing the problem as a choice has important ramifications in terms of the model specifications. First, we do not limit ourselves to only studying the big blockbuster movies, but also want to consider smaller movies. Thus the model should be flexible enough to fit both sleeper and blockbuster type movies (e.g., as in Sawney and Eliashberg (1996)). Second, we frame the problem in terms of market share rather than box office sales (e.g., using a logit type framework) as we wish to account for the consumer's choice between all movies released at any point in time. Working with market share rather than sales means that the behavioral underpinnings of models such as BOXMOD or Bass do not continue to apply to a choice model. For example, the Bass model uses cumulative sales as a covariate; but it is unclear what cumulative market attractiveness is. There is no obvious candidate for a model with strong behavioral underpinnings when dealing with market shares. This has led some authors to use simpler alternatives. For example, Einav (2002) uses a simple exponential decay (which cannot account for sleepers). Thus, we will strive to develop a model that is flexible enough to incorporate both types of movies and that is easily interpretable from a managerial standpoint.

We develop a model that meets these requirements in the next section of this paper. We then calibrate the model using data assembled from public sources of 404 movies released between 1995 and 1998. This empirical analysis is followed by a discussion of the empirical findings and managerial implications.

6

3 A Flexible Bayesian Model for Predicting Movie Sales

We build our model by merging a random effects logit model with a Gamma function to model diffusion, adapting each part to fit our goal. To account for the short life of movies, the logit model incorporates an indicator variable ( Iit ) for each movie that is set to 1 during weeks for which a particular movie is at the box office, and 0 otherwise (i.e., before the movie is released or after it has been pulled out of the theaters). We describe this as a sliding window logit model, in that it is essentially a logit model that allows for a different choice set in each period. Finally, we include an outside good to account for those consumers who choose not to go to the movies (i.e., the no-purchase option; see for example Erdem and Keane 1996) at any given time. As we will show, an outside good allows us to incorporate seasonal effects directly into the market share model. The logit formulation for the market share of a movie is as follows:

M it

=

eUOt

e I Uit it

+

e I U jt jt

.

(1)

j

And for the outside good, it is:

M ot

=

eUOt

eUOt

+

e I U jt jt

.

(2)

j

Where:

Mit is the expected market share of movie i in week t,

Mot is the expected market share of the outside good in week t,

Uit is the market attractiveness of movie i in week t,

Iit indicates whether movie i is screened in week t (1 if it is, and 0 otherwise),

7

 ................
................

In order to avoid copyright disputes, this page is only a partial summary.

Google Online Preview   Download