# What is at stake?

## Why trying to forecast the box-office ?

Having estimates for the box-office before the release of a movie can be useful for several players of the movie industry:

• marketing agencies: how efficient is their marketing campaign? Can they measure its impact?
• studios and distribution companies: how many copies should they try to distribute? How to choose between two movies?
• movie theaters: how many copies should they accept to show?
However, making those estimates is quite difficult, which makes many consider the movie industry like a casino. The statistical analysis confirms this idea. We focus our anaylsis on a data set of movies released in 2010 in France, and which appeared at least once in the top 20 (avoiding anonymous movies).

## The movie revenues are very uneven

The following graph shows the distribution of the ticket sales, drawing the cumulative proportion of ticket sales on the vertical axis and the sorted movies on the horizontal axis (this is a Lorentz curve ). We can see, for example, that 40% of the movies (the less popular ones) amount for about 8% of total ticket sales.

Half of the movies (the least popular) amounts for only 11% of total ticket sales during the first week, whereas, on the contrary, 13% of the movies (the most popular) amounts for more than half of the ticket sales.

Not only is the box office very uneven, but also the number of ticket sold per movie theater. This metrics shows the success relatively to the expectation of the professionals who decided how many copies should be shown.

## A more serious issue: success and failure are difficult to forecast

To measure the expectations of movie industry experts and their accuracy, we can study the number of tickets sold per week per movie theather, during the first week. We can see a very large dispersion:

The average ticket sales is 807 per week per movie theater during the first week, but the standard deviation is 481, which is quite large. Statistics from the French CNC shows that ticket are sold around 6.5 euros, which translates into a standard deviation of . Assuming no correlation between different movies and different weeks, the standard deviation for revenues during a year per theater is: .

# How Cinequant monitors the buzz for movies and forecasts the box-office

## Our idea: there is a lot of valuable information on consumers on the Internet

Cinequant first implementation of this idea is a comprehensive system forecasting the box office of a movie before its release, given only a measure of the buzz about it on the Internet. To achieve this goal, we created a set of tools aimed at:

• harvesting, in an automated fashion, data on the Internet (from social networks, twitter, public websites) that gives relevant information on customer behavior and tastes ;
• integrating large amounts of those data in our statistical model to produce a stream of highly relevant indices and low-error forecasting ;
• helping our client to integrate these predictions in its decision processes.
To achieve this goal, we calibrate our statistical model with a historical dataset of hundreds of movies, and then ask our model to predict the box office of a new movie.

# Our solution

Cinequant offer different solutions for each client's needs:
• a turn-key product : box-office forecasting for movies to be released, using only publicly available data
• a taylor-made solution : forecasting using private data. Then, it is necessary, as a first phase, to have a preliminary dialogue to fully detail the needs and expectations of our clients. The addition of private data can considerably improve the accuracy of our forecasting, bringing new information in the model. In that case, Cinequant offers an interface to exchange data in an automated fashion, using the usual standarts (XML).

## Nothing to install: the software runs on our own servers (SaaS)

Cinequant is a solution oriented towards services (Software as a Service) : you don't have to install any software nor database, we run everything on our own system. This architecture allows us to upgrade in real time the database and to constantly improve our statistical models. To implement such service, we use only well-known open-source technologies (linux, python, jQuery, AJAX, MySQL...).

Our client have a private secured access to our servers. In a dashboard, they can explore the data, visualize it, request automatic forecasting. They can also export the results to integrate them in their own documents.

This service is sold on a subscription basis.