Movies and related products sales prediction
"My company wants to sell movie-related products. We need to choose the target movies ASAP!"
How should I choose the target movies?
Blu-ray
DVD
Should be estimated from the sales of Bluray/DVD
Data Source
-
Box office sales data (web scraping)
- Boxofficemojo.com
-
DVD/Bluray sales (SQL)
- ~2000 titles from year 2006-2015
- Opusdata.com
Opening week gross?
Blu-ray/DVD total sales
Total
sold
units
(10 million)
Opening week gross (100 million $)
R^2 = 0.44
R2=0.44
BDDVD = \beta_0 + \beta_1x_1
BDDVD=β0+β1x1
Not good enough.
How about first month gross?
BDDVD = \beta_0 + \beta_1x_1
BDDVD=β0+β1x1
Wait.
It's not over yet.
Popularity Holding Index (PHI)
Popularity Holding Index (PHI) =
4th week gross per theater / average of first 2-week gross per theater
Weekly Gross ($)
Week
1.5
4
# of theaters
20%
100%
Most movie fades quickly
Weekly Gross ($)
Week
# of theaters
R^2 = 0.64
R2=0.64
BDDVD = \beta_0 + \beta_1x_1 + \beta_2x_2
BDDVD=β0+β1x1+β2x2
PHI
4week total gross (100 million $)
BD/DVD
(10million
units)
1. Four weeks total gross ($)
2. Popularity holding index (PHI)
3. Movie genre
Variables
Model
Total Blu-ray/DVD sales =
- 254700 + 0.348 x (4wk gross) + 530500 x (PHI) + (genre)
genre example: Animation = 1420000
R^2 = 0.65
R2=0.65
BD DVD = \beta_0 + \beta_1x_1 + \beta_2x_2 + \beta_3x_3
BDDVD=β0+β1x1+β2x2+β3x3
model = smf.ols('bddvdsale ~ total_4w_gross + w4_2w_phi + genres', df).fit()
"Bluray_DVD = beta0 + beta1*total_4w_gross + beta2*w4_2w_phi + beta3*genres"
PHI and 4week_gross are independent
Check, check
Music and History genre does not have enough sample cases
The final model predicted value vs true value
in 400 movie title from separate test data set
My suggestions
Bet on these movies now.
6108065
6008853
5937283
3706274
* Furious 7 and Home are not included as they have not pass week 4
2015 Box Office
DVD Prediction
root mean squared error(training_data) = 1669312
root mean squared error(testing_data) = 1779348