Notes from the lecture of Prof. Jagmohan Raju of Wharton on “Forecasting Demand for Movie Purchase” at ISB by Bhat Dittakavi on 13th August 2022
Title: Forecasting Demand for Movie Purchase -Developed for Warner Brothers
Speaker: Prof. Jagmohan S Raju, The Wharton School
I have just happened to land into Hyderabad this morning and my classmate Bindu’s message led me to attend this lecture. It has been 9 years since I attended Prof. Jagmohan Raju’s course on “Competitive Marketing Strategy” at Wharton Business School back in 2013 as part of my Executive MBA program known as PGPMAX at ISB. Prof. Raju is just the same with his wits and simplicity. Nothing changed about him. I still remember the hands-on MarkStrat simulations from his class. These simulations are great way to learn how effective our business decisions are without having to burn real money. It is like flight simulations without burning the fuel.
Forecasting using linear regression is very fundamental for an engineer in me. However, the business school reminds me that many times the answer to a business challenge lies not Just in the Math of Regression but in the Unseen of the Known. The study in this lecture is an epitome of that.
Raju specialises on pricing and marketing strategy, coupon strategy and sales force programs. He is an alumnus of IITD, IIMA, Stanford.
Prof. Jagmohan Raju:
I was here in Delhi for a meeting and I was supposed to be meeting ISB Dean Madan and before I realised, I am roped into this.
Two things I would discuss today: Network neutrality which is a policy issue across countries. Then the forecasting demand. Let me start with the second one.
I taught at ISB in 2001 the first time and I did it till 2015 and I haven’t physically taught here after that. I also kicked off Mohali campus. I represent Wharton on ISB board as marketing area chair.
Back to classroom is typically all about leadership and all and I have nothing to say about it. My work is all about model building for better decision making. I convince the folks that better decision making can be achieved even in unconventional industries like medical decides and entertainment.
Entertainment industry has different meaning to the word model! I am into business models! My colleague Joseph Aresty did a lot of work with movie industry, not me.
From Hollywood to Tollywood in 39 steps
This is a revised title for the talk.
This work is about 10 years old but can be easily related. On an average it costs $59 million to produce a movie in Hollywood. Then $31 million for marketing. Then $16 million for distribution.
It means the numbers don’t add up! For many movies, international box office is a market. Titanic is taken over by Maverick movie with highest grossing movie ever. There are some interesting failures also. It looks like movie business is going to lose lots of money. We call this business a business of blockbusters. Pharmaceutical is also in the business of blockbusters. So is music too. Make a lot of them and only one takes off.
Over time, film studios have learnt that they must make money from post box office revenues. At the time of study, 50% of revenues of a movie come from post-box office sales like DVD or OTT.
What would be the rental revenue of a movie?
Our prediction error for rental business is around 15% either way. However prediction of sales, the error is off by 100-200% which is not acceptable.
Rental Demand is very highly correlated with box office performance with correlation R2 = 0.9
If you know box office, you know the rental demand. It takes three weeks to go to rental after the box office release as most movies do box office business in three weeks as there is no shortage of screens. 85% of the box office happens within three weeks. We know the box office numbers in three weeks and we can make the prediction of rentals. Then I can plan my activities and finances accordingly for the rental demand.
How do we better predict the DVD sales business? Our R2 here with box office is only 0.4.
We can’t put a structure that is too complicated. Idea is that we can do something quick, accurate and scalable. Warner launches 30-40 titles a year and even higher with OTT now.
Building blocks have to be easy to put together and also to understand.
If you launch a new car or phone, you already have cars and phones in the market where you know their characteristics. Existing is there and the new and we can see the difference between.
In movie: The oldest movie is three weeks old. When your movie is available for rental or sales, you have new titles in the market. This is “new against new” challenge.
Second challenge: This professor from Israel told me “even though I made good predictions, studios don’t listen to my models and they still go ahead and do what they want to do despite the past where my predictions worked”.
In pharmaceutical world, they listen to us and our model advice! They paid us good. In entertainment world, it is different.
Both of us brainstormed.
Any pattern between the rows above?
The pattern is blockbuster, a failure and a blockbuster.
Whatever the wood Hollywood or Bollywood, certain stars work with certain studios on an ongoing basis as the relationships are durable. Viacom18 and Akshay Kumar work together and on an average Akshay does good. Studio likes to keep the relation with the star. To keep this long term relationships, the studio knowingly pushes the movie even though it won’t fare well.
Sales of the movies say DVDs happen at walmart that studios don’t go and spend time.
Idea of the model: How big is the pie? How much will we get? Predict the demand for say DVD and then predict how much do I get. The problem is new against new.
Predict the sales in week t, t+1 and t+2. What is the install base of people who can buy. Say existing DVD buyers or Netflix as the platform installed base. I know which titles are coming and the calendar is announced. Industry knows it. Box office and rental calendars are known. These are known variables. We can use them to predict what will be total sales for the week.
Box office data is known variable. What else? Seasonality, star power, relative advertisement (let me bust this: in a class room it is fine to our distribution of my product versus someone else’s as a predictor, in real world it is not wise to use it as a predictor. This means the unknown part of other company’s efforts. Unknown from unknown is guess work), IMDB rating. Do these play any additional role than what they already do to the box office? One variable that makes the difference is that genre. Musicals as a genre have a higher impact on say sales than box office.
As far as box office is concerned say there are two equally good movies as they gave the same total box office. Movie 2 is likely to sell more DVDs. This is the same and existing box office data but with more granularly. Movie 2 may have had multiple visits with families. It means DVD sales could get more as it can be watched again and again.
Attractiveness: cumulative box office, MPAA rating and cinema score
Rethinking the box office data gave us R2 from .4 to .8. Add genre and it only goes up a little bit.
The outlier is a spoiler. Answer from the studio is keep the outlier. This is what they said: We see something taking off as we put more effort on good movie to make it a blockbuster.
We were told that the forecast error went down by 30%. They started using our models.
Why is ISB a pioneer?
1 year MBA. Not really. Create knowledge first, then secondly disseminate knowledge. Both are important. Strong links between research, teaching and practice. Good reach impacts business practice.
Put one more little variable to make it practice like in the study and don’t make it too complex.
Some people in industry are the reason why others join them as there are torchbearers from whom others can replicate or learn.