Analytic Forensics Logo 






I spent a large part of the last 12 years doing models for the big movie studios.  The result was movie models which were the most accurate movie prediction models in the market (as far as we could determine).

To put that in context, the hardest part of movie modeling is predicting the opening weekend gross before the opening weekend.  Once you know the opening weekend gross, the other weekends and total gross is easy to predict.

My models predicted a movie’s opening weekend gross with +/- 15% accuracy 42% of the time. 

Doesn’t sound that accurate.  I know.

But comparing it to other movie tracking company models, and the various experts and websites that publish movie predictions; Their predictions average about 29% accuracy (+/- 15% on opening weekend gross).  The best forecaster we could find had 35% accuracy.  While 42% doesn’t seem that good, in comparison it is very good.

Now the methods and models I used are proprietary, so I can’t share the details here.  Just trust me they took a lot of data and a lot of effort just to get to 42% accuracy.  But I can show you somethings about movie predictions that don’t use proprietary data or methods.

To start the best prediction of how well a movie will do is the site count.  In other words, how many theaters it will be shown in on its opening weekend.  Part of this obvious, the more theaters a movie runs in the easier it is for people to see the movie and so the total gross increases.  But the site count also reflects the fact that the theater owners know movies and want to make money.  Every theater that runs a movie is likely an owner who is convinced it will do well for them and attract an audience; if a movie is in a lot of theaters than a lot of movie owners think it will do well (and likely they are correct).

If you look at the graph below, you can see a very clear relationship between opening site count and opening weekend gross.  If you were to base a prediction on the opening site count alone you would be correct 16% of the time (+/- 15%). 



On top of the opening weekend site count, there are other factors that are easily available that effect gross.  One big one is seasonality.  Movies do better on some weekends then others.  Look at the graph below (also based on movies opening 2002-2010).



If you compare what you know about the opening site count, the seasonality, and add other information like the genre of the movie or what the competition is that weekend, you can get a reasonably good model for movie opening week end gross.  Putting all that together into a “base model” is how we got the graph below:



So, all that standard information gets you to 24% accuracy.  To push the accuracy up to 35% or higher takes some very “fancy” modeling and other data fields that are not as easy to find.  It is not an easy task, but if you are a “data geek” like me, it is a very fun assignment.


Home Page


Analytic Forensics Logo




Text Box:                                                                                                                                                To Contact Us:







                                                                                                                      Phone:                                                     212-529-5337

                                                                                                                      Voice Mail:                                               917-838-7966


                                                                                                                      Address:                              23 East Tenth Street #304

                                                                                                                                                                New York City, NY     10003