Alpha from the Box Office?

When I recently watched some trailers of upcoming cinema movies I was wondering what the number of Youtube views (or also other properties, i.e. reviews) can tell about the success of the movie. For example, the official trailer of the Marvel movie “Black Panther” has about 33 million views at the moment. Black Panther was released on 2018/02/15 in the US and is as of yet the by far most successful movie of 2018 with more than 1.1 billion USD gross worldwide. By way of comparison the official upload of the first trailer of upcoming “Avengers: Infinity War” counts now 156 million views. So, it seems reasonable to expect an even higher turnover for this one (but surely not 4.7 times as big) and thus a pretty large amount of money for Disney. Therefore, it is interesting to investigate (1) what drives the Box Office of a movie, (2) what methods are available to get a forecast and (3) whether there is a obvious connection to the performance of the producing company.

Indeed, if issues (1) and (2) are forwarded to Google Scholar some papers about the Box Office including forecasting procedures are quickly found, see Sawhney and Eliashberg (1996) or Ghiassi et al. (2015). The approaches are quite diverse and can be differentiated by several aspects. These include what kind of input data is used, at which production stage (Pre-Production, Pre-Release, Post-Release) the forecast is made or which modeling technique is used (counting processes, regressions, neural networks). For example, in Derrick et al. (2014) a two step procedure utilizing regressions is proposed. First, the outcome of the starting week of a particular movie is predicted (using variables as Number of Reviews, Percentage of Positive Reviews, Star Power, Genre, …) and second, the difference between the forecasted and realized results for the first week is used to make a statement about the expected remaining Box Office. They show that this difference can be informative.

Having in mind that forecasts are possible, it seems natural to suspect that expected Box Office results are also in some way contained in the valuation of the producing company (question 3). Then, if the uncertainty of the revenue is resolved and the actual numbers differ substantially from the expectations a shift in the valuation could follow. If there are any persistent patterns this could be of potential use in a strategy to generate alpha, that is excess return over a benchmark model.

Luckily, some of the major American film studios which are sometimes called “Big Six” are part of a larger stock exchange listed conglomerate. For example, both above mentioned movies are made by Marvel Studios which belongs to Walt Disney Studios which belongs to The Walt Disney Company whose shares are mainly traded at the New York Stock Exchange (NYSE). This possibly enables to investigate how stock returns react when a new movie comes to the cinema. The following plot shows a first intuitive approach to this question.

The stock price of Disney is shown. Additionally, the red vertical lines depict the release dates of several rule-of-thumb selected Disney movies which were on the one hand really successful and on the other hand perhaps ex-post more successful than ex-ante expected. The blue and green lines correspond to a 15 trading day period directly before and after the release, respectively. Since I conjecture that these six movies (Marvel’s The Avengers, Iron Man 3, Frozen, Guardians of the Galaxy, Star Wars: The Force Awakens, and Beauty and the Beast) did better than expected the mean return in the green Post-Release periods should be positive and perhaps also larger than in den Pre-Release period. However, the mean of the Pre-Release returns could be positive as well for several reasons: promising forecasts, good reviews, many trailer views, good feedback from tests crowds and suchlike. However, there are of course aspects which are able to blur the results: too few movies and therefore little statistical power, too small influence of the movie studios contribution in view of the turnover of the whole company, a bad choice of period length or also that the returns are possibly driven by other factors. To tackle some of these issues the following table reports estimates of mean returns (scaled to an annual basis; t statistics, and p values in brackets) not only for the particular case shown in the plot. Instead, a second larger film set consisting of 29 Disney movies since 1999 is used (a movie is included if it was in the Top 5 of the yearly US Box Office ranking), several variations of the period length (2 to 15 trading days) and finally results for residual returns of regressions using the three factors of Fama and French.

Selected Movies (6 Movies)
Successful Movies since 1999 (29 Movies)
Period Length251015251015
Raw Returns:
Pre-Release Mean0.380 (t=0.5172, p=0.6153)0.381 (t=0.7536, p=0.4572)0.687 (t=2.0079, p=0.0492)0.168 (t=0.5337, p=0.5949)0.525 (t=1.0508, p=0.2978)0.709 (t=2.4750, p=0.0145)0.390 (t=1.6482, p=0.1004)0.186 (t=0.9822, p=0.3266)
Post-Release Mean1.487 (t=1.7172, p=0.1139)1.017 (t=2.2665, p=0.0311)0.225 (t=0.6896, p=0.4932)0.157 (t=0.6491, p=0.5180)0.463 (t=1.1436, p=0.2576)0.199 (t=0.6635, p=0.5081)0.046 (t=0.2287, p=0.8193)0.220 (t=1.3879, p=0.1659)
Post Minus Pre Mean1.107 (t=1.1402, p=0.2784)0.635 (t=0.8844, p=0.3837)-0.462 (t=-0.9655, p=0.3382)-0.011 (t=-0.0271, p=0.9784)-0.062 (t=-0.1039, p=0.9176)-0.510 (t=-1.3386, p=0.1828)-0.344 (t=-1.1320, p=0.2586)0.034 (t=0.1368, p=0.8913)
Pre/Post Together Mean0.934 (t=1.6456, p=0.1134)0.699 (t=2.0698, p=0.0429)0.456 (t=1.9294, p=0.0561)0.162 (t=0.8210, p=0.4127)0.494 (t=1.5429, p=0.1256)0.454 (t=2.1880, p=0.0295)0.218 (t=1.4069, p=0.1600)0.203 (t=1.6447, p=0.1004)
Pre-Release Mean0.807 (t=1.4360, p=0.1788)0.588 (t=1.8102, p=0.0806)0.448 (t=2.1500, p=0.0357)0.189 (t=0.9584, p=0.3405)0.526 (t=1.3239, p=0.1908)0.565 (t=2.6437, p=0.0091)0.181 (t=0.8992, p=0.3693)0.067 (t=0.4314, p=0.6664)
Post-Release Mean1.454 (t=1.7009, p=0.1170)0.886 (t=1.7525, p=0.0903)0.227 (t=0.7304, p=0.4680)0.238 (t=1.0604, p=0.2918)0.275 (t=0.7874, p=0.4343)0.165 (t=0.6663, p=0.5063)0.057 (t=0.3471, p=0.7287)0.091 (t=0.7250, p=0.4689)
Post Minus Pre Mean0.647 (t=0.6330, p=0.5397)0.298 (t=0.4957, p=0.6239)-0.222 (t=-0.5723, p=0.5693)0.049 (t=0.1805, p=0.8572)-0.251 (t=-0.5682, p=0.5721)-0.400 (t=-1.2553, p=0.2114)-0.124 (t=-0.4996, p=0.6177)0.024 (t=0.1227, p=0.9024)
Pre/Post Together Mean1.130 (t=2.2395, p=0.0351)0.737 (t=2.4690, p=0.0165)0.338 (t=1.8107, p=0.0727)0.214 (t=1.4331, p=0.1536)0.400 (t=1.5193, p=0.1314)0.365 (t=2.2300, p=0.0265)0.119 (t=0.9175, p=0.3593)0.079 (t=0.7923, p=0.4284)

The results suggest that the period lengths 10 and 15 are likely too long. Here, noteworthy results (significant at 5 % level) can be found for the Pre-Release period of length 10 in the 6 movies case (raw returns as well as residuals) which is contradictory to the initial conjecture. Instead, focusing on the 6 movies case for period lengths 2 and 5 the crude mean estimates show the suspected behavior. However, since only 6 movies are considered the sample sizes seem too small (2*6=12 returns and 5*6=30 returns) to be able to state statistically covered conclusions (“Post Minus Pre Means” not significant). This is supported by the fact that merging Pre- and Post-Release Periods (doubled number of returns) leads to more significant results for the residuals returns.

In the 29 movies case considerable results are achieved only for the period length 5. For raw as well as residuals returns the Pre-Release period shows significantly positive mean returns, for residuals even at the 1 % level. This is carried over to the “Pre/Post Together” row where slightly less convincing values are reported. Note, that the 29 movies here are simply all Disney movies which were overall successful. That is, there are likely films included which did good but no as good as expected. How robust these Disney results are could be assessed using one of the other major studios. Besides, a more objective way to select unexpected blockbusters would be useful to increase the 6 movies sample and validate its members. For example, one of the available Box Office forecasting approaches could be useful to compute a historical pseudo Pre-Release prediction and then compare this to the realized turnover to justify the magnitude of surprise. Of course, real historical forecasts would be useful, too.



  • Derrick, F. W., N. A. Williams, and C. E. Scott (2014). A two-stage proxy variable approach to estimating movie box office receipts. Journal of Cultural Economics 38(2), 173–189.
  • Ghiassi, M., D. Lio, and B. Moon (2015). Pre-production forecasting of movie revenues with a dynamic artificial neural network. Expert Systems with Applications 42(6), 3176–3193.
  • Sawhney, M. S. and J. Eliashberg (1996). A Parsimonious Model for Forecasting Gross Box-Office Revenues of Motion Pictures. Marketing Science 15(2), 113–131.


Print Friendly, PDF & Email