Case Study 2: Forecasting Box Office Returns

For years, people in the motion picture industry critics, film historians, and others have eagerly awaited the second issue in January of Variety. Long considered the show business bible, Variety is a weekly trade newspaper that reports on all aspects of the entertainment industry; movies, television, recordings, concert tours, and so on. The second issue in January, called the Anniversary Edition, summarizes how the entertainment industry fared in the previous year, both artistically and commercially.

In this issue, Variety publishes its list of All Time Film Rental Champs. This list indicates, in descending order, motion pictures and the amount of money they returned to the studio. Because a movie theater rents a film from a studio for a limited time, the money paid for admission by ticket buyers is split between the studio and theater owner. For example, if a ticket buyer pays $12 to see a particular movie, the theater owner keeps about $6 and the studio receives the other $6. The longer a movie plays in a theater, the greater the percentage of the admission price returned to the studio. A film playing for an entire summer could eventually return as much as 90% of the $12 to the studio. The theater owner also benefits from such a success because although the owners percentage of the admission price is small, the sales of concessions (candy, soda and so on) provide greater profits. Thus, both the studio and the theater owner win when a film continues to draw audiences for a long time. Variety lists the rental figures (the actual dollar amounts returned to the studios) that the films have accrued in their domestic releases (United States and Canada).

In addition, Variety provides a monthly Box-Office Barometer of the film industry, which is a profile of the months domestic box-office returns. This profile is not measured in dollars, but scaled according to some standard. By the late 1980s, for example, the scale was based on numbers around 100, with 100 representing the average box-office return of 1990. The figures from 1997 to 2006 are given in the table below and in the file BoxOffice.xlsx in blackboard.

All the figures are scaled around the 1990s box-office returns, but instead of dollars, artificial numbers are used. Film executives can get a relative indication of the box-office figures compared to the arbitrary 1990 scale. For example, in January 1997 the box-office returns to the film industry were 95% of the average that year, whereas in January 1998 the returns were 104% of the average of 1990 (or, they were 4% above the average of 1990s figure).

Month	1997	1998	1999	2000	2001	2002	2003	2004	2005	2006
Jan	104	101	88	132	125	111	127	119	147	145
Feb	100	96	110	109	118	123	129	147	146	149
Mar	99	82	129	101	121	121	132	164	133	148
Apr	88	84	113	111	140	139	108	135	148	148
May	89	85	114	140	141	119	115	124	141	148
Jun	108	124	169	179	201	156	149	168	191	201
Jul	109	134	131	145	152	154	155	159	178	184
Aug	101	109	139	140	138	136	129	137	156	166
Sep	106	121	120	120	137	105	117	149	119	151
Oct	102	111	115	129	138	132	166	159	138	166
Nov	78	101	116	118	144	123	152	175	175	170
Dec	111	112	128	139	148	164	173	195	188	194

From the time series given in the above table, you will make a forecast for the 12 months of the next year, 2007.

Managerial Report is due on Thursday, 16 Sept (40 pts)

1. Produce a time series plot of the data. From this graph, do you see a pattern? Can you see any seasonality in the data?
2. Use exponential smoothing to fit the data. Select an appropriate constant a based on the variation you see in the data. Comment on the appropriateness of exponential smoothing on this data set. Plot the predictions from this model on the graph with the original data. How well does this technique fit the data? Make forecasts for each month in 2007.
3. Use regression to build a linear trend model. Comment on the goodness-of-fit of this model to the data (or, how well does R² explain the variance in the data?). Plot the predictions from this model on the graph with the original data.
4. Develop multiplicative seasonal indices for the linear trend model developed in question 3. Use these indices to adjust predictions from the linear trend model from question 3 above for seasonal effects. Plot the predictions from this model on the graph with the original data. How well does this technique fit the data? Make forecasts for the next 12 months of 2007 using this technique.
5. Which forecasting method of those that you tried do you have the most confidence for making accurate forecasts for 2007? Use MAPE (mean absolute percent error) as your criterion to justify your decision.

Enrichment (5 pts): Use Optimization (and Solver in Excel) to find the optimal smoothing constant in problem 2 above (by minimizing the Mean Squared Error or MSE).