SERVING THE QUANTITATIVE FINANCE COMMUNITY

 
User avatar
albertmills
Topic Author
Posts: 380
Joined: March 13th, 2007, 1:09 pm

Monte Carlo simulation to predict movie success

January 5th, 2015, 10:29 pm

Do you guys believe that this producer actually uses monte carlo simulation to predict the success of a movie (or if he does that it actually works and he's not just deluding himself or others)? See link below for entire article "Before Relativity commits to financing a particular movie ? either through its slate deals with Sony and Universal or on its own ? it's fed into an elaborate Monte Carlo simulation, a risk-assessment algorithm normally used to evaluate financial instruments based on the past performance of similar products. Enough variables are included in the Monte Carlo for Wilson and his team to have reached the limits of their Excel's sixty-five thousand rows of data: principal actor, director, genre, budget, release date, rating, and so on. After running the movie through ten thousand combinations of variables (in marathon overnight sessions), the computers will churn out a few hundred pages that culminate in two critical numbers: the percentage of time the movie will be profitable, and the average profit for each profitable run. The computers will also calculate the best weekend for the movie to be released, whether Russell Crowe will earn his salary or Sam Worthington will be good enough, and the box-office effect of an R rating versus PG-13. But for Kavanaugh, those are secondary considerations: Unless the movie shows the distinct probability of a return ? no one at Relativity will reveal the precise green-light figure, but it's something like 70 percent ? the script gets shredded. "Everything has to run on the principle of profit," Kavanaugh says. "We'll never let creative decisions rule our business decisions. If it doesn't fit the model, it doesn't get done."http://www.esquire.com/features/best-an ... ugh-1209-2
 
User avatar
Traden4Alpha
Posts: 23951
Joined: September 20th, 2002, 8:30 pm

Monte Carlo simulation to predict movie success

January 5th, 2015, 10:59 pm

This sort of forecasting probably does work for a wide range of movies, but not movies that break new ground in cinematography, plot devices, special effects technologies and the like.If it's yet another cops vs. mob thriller with Russell Crowe vs. Bruce Willis, then the movie's performance probably does follow the distribution of outcomes for movies of that type with those actors. For many of the mainstream movies, there's a sufficient sample size of previous similar movies to have statistical data.The only part that is unclear to me is the aggregation function. Are the effects of particular actors, directors, genre, etc. additive, multiplicative, or something in between? If a movie has two star actors that each tend to double the ceteris paribus box office revenues of a movie, then will their combined effect add 200% to the revenues or will revenues be 4X greater?
 
User avatar
katastrofa
Posts: 8946
Joined: August 16th, 2007, 5:36 am
Location: Alpha Centauri

Monte Carlo simulation to predict movie success

January 5th, 2015, 11:17 pm

QuoteOriginally posted by: Traden4AlphaThe only part that is unclear to me is the aggregation function. Are the effects of particular actors, directors, genre, etc. additive, multiplicative, or something in between? If a movie has two star actors that each tend to double the ceteris paribus box office revenues of a movie, then will their combined effect add 200% to the revenues or will revenues be 4X greater?Copulas, babe :-)Usually, when there are complex interrelations between variables in the model, causality or memory, simple regression is not enough and one goes for MC simulations. I'm not surprised movie industry uses them too since they have so much data.
 
User avatar
Traden4Alpha
Posts: 23951
Joined: September 20th, 2002, 8:30 pm

Monte Carlo simulation to predict movie success

January 6th, 2015, 12:49 am

QuoteOriginally posted by: katastrofaQuoteOriginally posted by: Traden4AlphaThe only part that is unclear to me is the aggregation function. Are the effects of particular actors, directors, genre, etc. additive, multiplicative, or something in between? If a movie has two star actors that each tend to double the ceteris paribus box office revenues of a movie, then will their combined effect add 200% to the revenues or will revenues be 4X greater?Copulas, babe :-)Usually, when there are complex interrelations between variables in the model, causality or memory, simple regression is not enough and one goes for MC simulations. I'm not surprised movie industry uses them too since they have so much data.Good point. Copulas seems like a good way to go but how do copulas deal with, for example, four dimensions of discrete independent variables that are very sparsely combined to reveal sample size of near-1 values in the one dependent variable (revenue or profit)?
 
User avatar
Paul
Posts: 10421
Joined: July 20th, 2001, 3:28 pm

Monte Carlo simulation to predict movie success

January 6th, 2015, 8:34 am

Monte Carlo is just a numerical method. You may as well say that the programmer was wearing green pants.The question is what does the simulation simulate?P
 
User avatar
savr
Posts: 48
Joined: January 21st, 2013, 3:28 pm

Monte Carlo simulation to predict movie success

January 6th, 2015, 10:26 am

It's been done before with much finer algorithms such as AWESOM-O.Question: is there a way (like there is for television) the movie industry uses to find out actual demographics of movie-goers? I can imagine (I've seen in movies) they have focus groups BEFORE a given movie's release, that might point to a particular subgroup of people being over-positive about it, but that doesn't mean the actual turnout at the movie isn't different from expected. Although my impression is that they are pretty well designed to hit their target demographic, and do so with predictable accuracy, the only way to make sure the latest robot-driven-by-humanbrain-traveling-in-time flick isn't mostly seen by pensioners is to measure who's actually sitting in the theatres. How do they do that?Because that's where a lot of the important data for this software would come from, to answer questions like "Would I see a n+1st timetravelingrobotwithhumanconsciousness flick after having seen n=1000"? And if such fine modelling of histories of cohorts (eg people old enough to remember when this was already done 45 years ago) turns out to be unnecessary because the population is so large, can this be explained with a limit theorem? I hope they work hard at it, they might make progress towards psychohistory.
 
User avatar
katastrofa
Posts: 8946
Joined: August 16th, 2007, 5:36 am
Location: Alpha Centauri

Monte Carlo simulation to predict movie success

January 6th, 2015, 11:28 am

QuoteOriginally posted by: Traden4AlphaQuoteOriginally posted by: katastrofaQuoteOriginally posted by: Traden4AlphaThe only part that is unclear to me is the aggregation function. Are the effects of particular actors, directors, genre, etc. additive, multiplicative, or something in between? If a movie has two star actors that each tend to double the ceteris paribus box office revenues of a movie, then will their combined effect add 200% to the revenues or will revenues be 4X greater?Copulas, babe :-)Usually, when there are complex interrelations between variables in the model, causality or memory, simple regression is not enough and one goes for MC simulations. I'm not surprised movie industry uses them too since they have so much data.Good point. Copulas seems like a good way to go but how do copulas deal with, for example, four dimensions of discrete independent variables that are very sparsely combined to reveal sample size of near-1 values in the one dependent variable (revenue or profit)?Discrete variables are not a problem. What does " very sparsely combined to reveal sample size of near-1 values" mean? Sorry :-)@PaulIf that was an actual question, dynamical simulations are usually used to calculate various utility functions (be it profit or revenue) - and this is typically done for different scenarios (e.g. investment strategies, spendings structure, logistics, etc.). The next step is sensitivity analysis or shock tests, which is supposed to provide risk assessment. It's no rocket science and such simulations are quite straightforward. It's nothing like in physics when they show you experimental data and ask about the known or new physical mechanism behind them.
 
User avatar
Traden4Alpha
Posts: 23951
Joined: September 20th, 2002, 8:30 pm

Monte Carlo simulation to predict movie success

January 6th, 2015, 1:11 pm

QuoteOriginally posted by: PaulMonte Carlo is just a numerical method. You may as well say that the programmer was wearing green pants.The question is what does the simulation simulate?PMy guess: the simulation draws a random movie that had director, D, and combines that's movie's financial outcome with financial outcomes from random movies that had actor, A, genre, G, budget, B, etc. I would hope it uses some combined effects (using copulas?) such as genre+opening date or genre+actor.
 
User avatar
Traden4Alpha
Posts: 23951
Joined: September 20th, 2002, 8:30 pm

Monte Carlo simulation to predict movie success

January 6th, 2015, 1:23 pm

QuoteOriginally posted by: katastrofaQuoteOriginally posted by: Traden4AlphaQuoteOriginally posted by: katastrofaQuoteOriginally posted by: Traden4AlphaThe only part that is unclear to me is the aggregation function. Are the effects of particular actors, directors, genre, etc. additive, multiplicative, or something in between? If a movie has two star actors that each tend to double the ceteris paribus box office revenues of a movie, then will their combined effect add 200% to the revenues or will revenues be 4X greater?Copulas, babe :-)Usually, when there are complex interrelations between variables in the model, causality or memory, simple regression is not enough and one goes for MC simulations. I'm not surprised movie industry uses them too since they have so much data.Good point. Copulas seems like a good way to go but how do copulas deal with, for example, four dimensions of discrete independent variables that are very sparsely combined to reveal sample size of near-1 values in the one dependent variable (revenue or profit)?Discrete variables are not a problem. What does " very sparsely combined to reveal sample size of near-1 values" mean? Sorry :-)What I meant is that most of the combinations of director x actor x genre x date x etc. have never occurred or have occurred only once or twice. The statistical distribution of financial outcomes for a given combination might be unknown or have only one or a few samples values.I also thought copulas were used if one had two or more outcome variables (e.g., returns on two related assets) which is not the case here.
Last edited by Traden4Alpha on January 5th, 2015, 11:00 pm, edited 1 time in total.
 
User avatar
Paul
Posts: 10421
Joined: July 20th, 2001, 3:28 pm

Monte Carlo simulation to predict movie success

January 6th, 2015, 2:34 pm

QuoteOriginally posted by: Traden4Alpha QuoteOriginally posted by: PaulMonte Carlo is just a numerical method. You may as well say that the programmer was wearing green pants.The question is what does the simulation simulate?PMy guess: the simulation draws a random movie that had director, D, and combines that's movie's financial outcome with financial outcomes from random movies that had actor, A, genre, G, budget, B, etc. I would hope it uses some combined effects (using copulas?) such as genre+opening date or genre+actor.But this is the classical problem. Even talking about the numerical method without mentioning the model assumptions is nonsense. (Of course, I could well be talking nonsense myself because the assumptions might be well known and generally accepted...but I suspect not.)Dur!I could easily imagine the scenario of a producer who didn't have any skin in the game or any direct financial interest in the venture using "Monte Carlo" to persuade dumb Angels that he had a winning project based on his sophisticated scientific analysis. The Angels invest and he gets to make his first Hollywood movie. He wouldn't mind if it was a flop. Or he could just use that other Monte Carlo method called, I believe, "tossing a coin."There's one born every minute.Do I think that some scientific analysis will help make a good decision? Yes. Will it be based on probabilistic ideas? Yes. Do I think that the assumptions will make a big difference to the results. Of course.P
 
User avatar
Cuchulainn
Posts: 61518
Joined: July 16th, 2004, 7:38 am
Location: Amsterdam
Contact:

Monte Carlo simulation to predict movie success

January 6th, 2015, 2:57 pm

QuoteBut this is the classical problem. Even talking about the numerical method without mentioning the model assumptions is nonsense. (Of course, I could well be talking nonsense myself because the assumptions might be well known and generally accepted...but I suspect not.)(Implicit) assumptions should always be flagged and made explicit.
http://www.datasimfinancial.com
http://www.datasim.nl

Every Time We Teach a Child Something, We Keep Him from Inventing It Himself
Jean Piaget
 
User avatar
katastrofa
Posts: 8946
Joined: August 16th, 2007, 5:36 am
Location: Alpha Centauri

Monte Carlo simulation to predict movie success

January 6th, 2015, 3:11 pm

QuoteOriginally posted by: Traden4AlphaQuoteOriginally posted by: katastrofaQuoteOriginally posted by: Traden4AlphaQuoteOriginally posted by: katastrofaQuoteOriginally posted by: Traden4AlphaThe only part that is unclear to me is the aggregation function. Are the effects of particular actors, directors, genre, etc. additive, multiplicative, or something in between? If a movie has two star actors that each tend to double the ceteris paribus box office revenues of a movie, then will their combined effect add 200% to the revenues or will revenues be 4X greater?Copulas, babe :-)Usually, when there are complex interrelations between variables in the model, causality or memory, simple regression is not enough and one goes for MC simulations. I'm not surprised movie industry uses them too since they have so much data.Good point. Copulas seems like a good way to go but how do copulas deal with, for example, four dimensions of discrete independent variables that are very sparsely combined to reveal sample size of near-1 values in the one dependent variable (revenue or profit)?Discrete variables are not a problem. What does " very sparsely combined to reveal sample size of near-1 values" mean? Sorry :-)What I meant is that most of the combinations of director x actor x genre x date x etc. have never occurred or have occurred only once or twice. The statistical distribution of financial outcomes for a given combination might be unknown or have only one or a few samples values.Scarce data are usually treated with Bayesian methods or some simple regression methods (sometimes very crude), resulting in large confidence intervals (or whatever scores or indicators one uses). When you have nothing or just a few points, it's sometimes justified to use proxy data. This is what one feeds to the MC simulation.QuoteI also thought copulas were used if one had two or more outcome variables (e.g., returns on two related assets) which is not the case here.I personally prefer to call them independent and dependent variables and they can be both described by copulas: e.g. correlations of times of default and correlations of cash flows from CDSs. Of course, one needs to understand the data first to avoid including same effects twice.
Last edited by katastrofa on January 5th, 2015, 11:00 pm, edited 1 time in total.
 
User avatar
albertmills
Topic Author
Posts: 380
Joined: March 13th, 2007, 1:09 pm

Monte Carlo simulation to predict movie success

January 6th, 2015, 3:19 pm

When I read the article I was suspecting something like the scenario Paul describes below...using "sophisticated scientific analysis" to persuade investors to part with their money, rather than for any real predictive purpose, and mostly just posted to see if people thought there might be more to it than this. QuoteOriginally posted by: PaulQuoteOriginally posted by: Traden4Alpha QuoteOriginally posted by: PaulI could easily imagine the scenario of a producer who didn't have any skin in the game or any direct financial interest in the venture using "Monte Carlo" to persuade dumb Angels that he had a winning project based on his sophisticated scientific analysis. The Angels invest and he gets to make his first Hollywood movie. He wouldn't mind if it was a flop. Or he could just use that other Monte Carlo method called, I believe, "tossing a coin."There's one born every minute.P
 
User avatar
Paul
Posts: 10421
Joined: July 20th, 2001, 3:28 pm

Monte Carlo simulation to predict movie success

January 6th, 2015, 3:38 pm

There's a lot of it about. They might have good intentions though. P
 
User avatar
daveangel
Posts: 17031
Joined: October 20th, 2003, 4:05 pm

Monte Carlo simulation to predict movie success

January 6th, 2015, 4:02 pm

QuoteOriginally posted by: PaulThere's a lot of it about. They might have good intentions though. Pthe road to hell and all that
knowledge comes, wisdom lingers
ABOUT WILMOTT

PW by JB

Wilmott.com has been "Serving the Quantitative Finance Community" since 2001. Continued...


Twitter LinkedIn Instagram

JOBS BOARD

JOBS BOARD

Looking for a quant job, risk, algo trading,...? Browse jobs here...


GZIP: On