SwarmFest 2004 May 11, 2004

11

Business Application Business Application of Agent-Based of Agent-Based

Simulation Simulation Complex and Dynamic Interactions of Motion Picture Complex and Dynamic Interactions of Motion Picture

MarketMarket

SwarmFest 2004SwarmFest 2004May 11, 2004May 11, 2004

이 승규이 승규 Seung-Kyu RheeSeung-Kyu Rhee이 원희이 원희 Wonhee LeeWonhee Lee

22Rhee and Lee: ABS for Movie IndustryRhee and Lee: ABS for Movie Industry

Movie: The Product and Movie: The Product and the Marketthe Market Movie

Is naturally a new product and Has short life-cycle from one week to several months

The Product With huge initial investment and High uncertainty of the market performance Highly risky

business The Market

Constituents of the movie supply chain From a writer with an idea To theater managers with screens to allocate and Everybody in between

Consumers in complex social network Local and central information Preference and constraints

Competing movies and substitutes


Focus of this paper

Feedback

Movie: The DecisionsMovie: The Decisions Given a movie to sell

A distributor has to decide How much marketing budget to spend, When to release it, How many screens to secure, etc.

The decisions should be based on The projected market performance, Which, in turn, would be influenced by the

decisions themselves and Many other uncontrollable factors, notably the early

performance of the movie itself.

Feed-forward


The ProblemThe Problem How is the market going to respond to

Various supplier’s decision alternatives under Various market conditions with competing movies and The communication dynamics about the movie quality

Adaptive reactions of competitors and myself What-if analysis is critical, but It is only possible with detailed knowledge of the dynamic

process

?


Existing ResearchExisting Research Ranges from simple statistical forecasting

models to a complex dynamic Markov chain model with behavioral parameter estimation

Some agent-based models have been proposed to describe the near-chaotic market behavior in terms of market share change

To our knowledge, no existing model is comprehensive enough to be useful for decision makers in motion picture industry


A Sample of Existing A Sample of Existing ModelsModelsResearch Objective

Method Authors Characteristics Limits

Decision support

system and Forecasting

Interactive Markov Chain

Eliashberg et al. (2000)

•Forecasting before the release by estimating parameters with audience survey

•Empirical test applied to real cases

•Competition•Dynamics

ForecastingQueuing model

Sawhney & Eliashberg

(1996)

•Estimating function and parameters

•Competition•Dynamics•Lack of explanatory variables

Understanding system behavior

Agent-based modeling

De Vany & Lee (2001)

•Reliability of product quality and market performance feedback by Information cascading perspective

•Marketing variables

•Low reliability of results

•Too simple decision rule

Finding major

variables

Empirical study

Bagella & Becchetti

(1999), De Vany & Walls

(1999)

•Finding important variables

•Comparison of coefficient between variables

•Competition•Dynamics


Issues Covered in Issues Covered in LiteratureLiterature

Competition

Movie characteris

tics

Marketing

(advertising,

distribution)

Critique

review

Quality or

WOM

Market performance feedback

Eliashberg et al. (2000) ○ ○ ○

Jedidi et al., (1998) ○ ○ ○

Lampel et al., (2000) ○

Zufryden (1996) ○ ○ ○

Mahajan et al. (1984) ○ ○ ○

Prag and Casavant (1994)

○ ○ ○

Linton & Petrovich (1988)

○

Litman & Kohl (1989) ○ ○ ○ ○

Sochay (1994) ○ ○ ○

Lampel and Shamsie (2000)

○ ○ ○

De Vany and Lee (2001) ○ ○ ○


Challenges to ABMChallenges to ABM KISS? Reality?

In agent-based simulation community, there is a tendency to prefer simple models

From practitioners’ viewpoint, however, it does not help much to confirm the fact that the market is too complex and anything is possible (e.g., De Vany and Lee, 2001)

Big question: How real is real enough? In this paper we expand the scope of the movie

market model by including diverse sources of movie quality information and competition effect.


Consumer State Consumer State Transition ModelTransition Model

Promotion: advertising effectivenessPlace: distribution effectiveness (number of screens)Product: theme acceptability and audience consensus to qualityPrice: irrelevant

PositiveSpreader

NegativeSpreader

Undecided Inactive

MarketingStrategy

MovieSelection

MovieQuality

PreviewPerformance

Competitive and composite quality evaluation: critique, preview audience, and audience consensus to quality

Box Office

WoM in Neighborhood

NeutralSpreader

Modified based onEliashberg et al. (2000)


Number of Neighborhood = 6Number of contacts = 2Number of WoM communication = 1

Incommunicable: Undecided or Inactive (post-WoM)

Communicable: positive/neutral/negative spreader

AgentToday’sContacts

ExampleExample

Agent in Social Agent in Social NetworkNetwork ABM v. EBM

Daily update of movie-going probability for each agent

Eliashberg et al. (2000) used aggregated market transition equations


Rich Microstructure in Rich Microstructure in Agent ModelAgent Model Modeling objective

Heuristic approach for better understanding of the market Gross and Strand (2000): Predictive, Explanatory, and Heuristic

Initial exploration of diverse variables and parameters Sensitivity analyses under diverse scenarios Part of bigger model: Production-Distribution-Competition Toward a commercially useful “Decision Support System”

Choice of rich microstructure The most salient characteristic of “culture products”

Experience goods: performance seriously affected by social interaction and human intervention

Model saturation can be determined by diverse experiments and sensitivity analyses


The Simulation The Simulation ProcessProcess

Pre-release Period Release PeriodSet-up

•Number of movies•Quality of movies •Marketing effectiveness•Frequency of marketing•Audience of preview

DIstributor

•Number of neighbors•Number of contacts•Frequency of WoM•Duration of WoM•Consensus to quality

Audience

•Preview a movie

DIstributor

•Review of critique

Critique

•WoM of preview audience

Audience

•Market performance feedback (Box office report)

Media

•WoM update•Movie- going decision•Spreading WoM

Audience

•Releasing a movie•Ending a movie(No audience)

DIstributor

•Frequency of critique review

•Consensus to quality

Critique

•Frequency of market performance feedback

Media


General ParametersGeneral ParametersReference Range Baseline Model

Number of movies De Vany & Lee (2001) 2~10 movies 5 movies

Quality of film Korean movie industry High, medium, low qualityHigh: 1, medium: 3, low:

1

Number of audience N/A 10,000~20,000 10,000 persons

Number of preview audience

Korean movie industry 1 – 20 (0.0001~0.002%) 10 persons

Preview period Korean movie industry 1~21 days 7 days

Marketing impacts Eliashberg et al. (2000) 0.0 – 1.0 0.5

Critique preferenceN/A

positive, negative, neutralDependent on critique

consistency

Critique consistency 0.0 – 1.0 1.0

movie-going probability

Korean movie industry 0.01~0.05 0.02

Maximum number of movie selection

N/A 1~5 movies 1 movie

WoM preference Mahajan et al. (1984) positive, negative, neutralDependent on WOM

consistency

WoM consistency De Vany & Lee (2001) 0.0 – 1.0 0.7

WoM: neighborhood N/A 0 – 10 persons 10 persons

WoM duration Eliashberg et al. (2000) 0-32 days 21 days

WoM frequency Eliashberg et al. (2000) 1 – 10 per week 2 per week


Signal ParametersSignal Parameters

Range Baseline Model

Characteristics

Marketing signals

1~7 times in pre-release

period2 times Performance

independent centralized informationCritique

signals

1~7 times in pre-release

period0.2 times

WoM signals Depend on WoM structure

Performance dependent

(box office and showing period) decentralized

information

Market feedback signals

1~3 times per week

(depends on release period)

Once a week

Performance dependent (showing

period)centralized information


Model Test Using Real Model Test Using Real DataData Test movies

Two week brackets in January, February, and July of 2000

Movies with more than 100,000 viewers Test with opening market share and final market

share Chi-square test (Chung and Cox, 1994)2

1

(Actual Predicted )

Predicted

N

i ii

i

Q

21 ( 1)N ?


Movie titleOpening

dayCritique quality

Audience quality

Marketing

impacts

Opening box office

Total box office

Jan. (Set 1)

Peppermint Candy (Korea) 1. 1 H H 0.2 6,206 290,276

A Happy Funeral Parlor (Korea) 1. 8 L M 0.3 6,725 111,837

Fly me to Polaris (Hong Kong) 1. 15 L M 0.3 10,120 202,840

The Bone Collector (USA) 1. 1 L M 0.4 13,372 212,564

Stuart Little (USA-Germany) 1. 8 L M 0.4 16,331 392,933

Happy End (Korea) 1. 1 M L 0.4 13,690 132,029

Lies (Korea) 1. 11 M L 0.8 19,035 307,702

Feb. (Set 2)

The Foul King (Korea) 2. 4 M H 0.4 22,741 787,412

Samurai Fiction (Japan) 2. 19 M M 0.4 14,232 224,256

The Beach (USA) 2. 3 M M 0.3 14,231 187,460

The Messenger: The Story of Joan of Arc (France)

2. 19 M M 0.4 13,084 220,986

Three Kings (USA) 2. 12 L L 0.2 10,060 134,376

Early July

(Set 3)

Dinosaur (USA) 7. 15 M M 0.8 27,859 554,169

Gone in 60 Seconds (USA) 7. 1 L M 0.5 21,272 348,710

Bichunmoo (Korea) 7. 1 L L 0.8 23,835 631,913

Late July

(Set 4)

Bayside Shakedown (Japan) 7. 22 M H 0.3 13,496 234,155

The Perfect Storm (USA) 7. 29 M M 0.9 35,184 508,913

The Patriot (USA) 7. 22 L M 0.4 18,229 149,415

Nightmare (Korea) 7. 29 L M 0.4 12,801 279,174

Ring 2 (Japan) 7. 29 M L 0.2 7,164 106,652

Test Data Set: Actual Test Data Set: Actual Movies in Korean Movies in Korean MarketMarket


0

5000

10000

15000

20000

25000

30000

1 9

17

25

33

41

49

57

65

73

81

89

반칙왕비치쓰리킹즈잔다르크사무라이픽션

쓰리킹즈비치

사무라이픽션잔다르크

반칙왕

Test Application to Test Application to Market Data: ShapesMarket Data: Shapes

Actual

Simulation


Test Application to Test Application to Market Data: FitnessMarket Data: Fitness


Result: Baseline Result: Baseline ModelModel

High Quality

Medium Quality

Low Quality

MediumMedium

LowMedium

High

MediumMedium

LowMedium

High


High quality

MediumMedium

LowMedium

High

MediumMedium

LowMedium

High

High Quality

Low Quality

Baseline Result: WoM Baseline Result: WoM DepletionDepletion WoM intensity gets weaker along the show

duration Initial audience size and signal accuracy (viewer

consensus) intervene

Signal accuracy = 0.7


Analysis: Marketing Analysis: Marketing ImpactsImpacts

Total Market size

72007400760078008000820084008600880090009200

Mkt

g=0.

1

Mkt

g=0.

2

Mkt

g=0.

3

Mkt

g=0.

4

Mkt

g=0.

5

Mkt

g=0.

6

Mkt

g=0.

7

Mkt

g=0.

8

Mkt

g=0.

9

Mkt

g=1.

0

0500

10001500200025003000350040004500

Mkt

g=0.

1

Mkt

g=0.

3

Mkt

g=0.

5

Mkt

g=0.

7

Mkt

g=0.

9

High QualityMedium QualityMedium QualityMedium QualityLow Quality

Marketing impacts positively affect the performance of good movies, and increase the total market size


Analysis: Marketing Analysis: Marketing ImpactsImpacts Bad movie’s increased marketing impacts

Bad movies only take the market away from other movies

Total Market Size

85508600865087008750880088508900

Mkt

g=0.

1

Mkt

g=0.

2

Mkt

g=0.

3

Mkt

g=0.

4

Mkt

g=0.

5

Mkt

g=0.

6

Mkt

g=0.

7

Mkt

g=0.

8

Mkt

g=0.

9

Mkt

g=1.

0

Unstable0

500

1000

1500

2000

2500

3000

3500

4000

Mkt

g=0.

1

Mkt

g=0.

3

Mkt

g=0.

5

Mkt

g=0.

7

Mkt

g=0.

9



Analysis: Marketing Analysis: Marketing ImpactsImpacts Decreasing returns to scale for the marketing

impact increase (inducing initial viewer increase) are confirmed for both good and bad movies with some irregularities

But if you have a good movie, then excessive marketing do not help much due to market information spreadsIncrease in the box office of high quality movie

0100200300400500600700800900

Mkt

g=0.

1

Mkt

g=0.

2

Mkt

g=0.

3

Mkt

g=0.

4

Mkt

g=0.

5

Mkt

g=0.

6

Mkt

g=0.

7

Mkt

g=0.

8

Mkt

g=0.

9

Increase in the box office of low quality movie

050

100150200250300350

Mkt

g=0.

1

Mkt

g=0.

2

Mkt

g=0.

3

Mkt

g=0.

4

Mkt

g=0.

5

Mkt

g=0.

6

Mkt

g=0.

7

Mkt

g=0.

8

Mkt

g=0.

9


Analysis: Marketing Analysis: Marketing SignalsSignals If consumers take central marketing

information more seriously (than other quality information), the market growth potential is seriously impaired

0500

1000150020002500300035004000

Mkt

g sig

nal=

1

Mkt

g sig

nal=

3

Mkt

g sig

nal=

5

Mkt

g sig

nal=

7

Mkt

g sig

nal=

9

Th

e n

um

ber

of

au

die

nce

High qualityMedium qualityMedium qualityMedium qualityLow quality


0500

100015002000250030003500400045005000

Wom

Rang

e=0

Wom

Rang

e=2

Wom

Rang

e=4

Wom

Rang

e=6

Wom

Rang

e=8

High QualityMedium QualityMedium QualityMedium QualityLow Quality Total Market Size

82008300840085008600870088008900900091009200

Wom

Rang

e=0

Wom

Rang

e=1

Wom

Rang

e=2

Wom

Rang

e=3

Wom

Rang

e=4

Wom

Rang

e=5

Wom

Rang

e=6

Wom

Rang

e=7

Wom

Rang

e=8

Wom

Rang

e=9

Analysis: WoM Range Analysis: WoM Range and Intensityand Intensity Increasing WoM signals positively affect the

performance of good movies, and increase the total market size


Total market size

7000

7500

8000

8500

9000

9500

1 2 3 4 5 6 7 8

Total market size

RandomWoM

consensus=1

RandomWoM

consensus=1

0

500

1000

1500

2000

2500

3000

3500

4000

4500

1 2 3 4 5 6 7 8


Analysis: WoM Analysis: WoM ConsensusConsensus Increasing WoM consensus positively affect

the performance of good movies, and increase the total market size


Analysis: WoM Analysis: WoM ImpactsImpacts By the “action-based WoM” assumption, good

WoM spreads widely, but bad WoM does not

Movie High Positive

WoM Listener

No WoMNegative

WoM Listener

High quality movie

Number of audience 3539 (35%) 5834 (58%) 609 (6%)

Number of movie-goer 1,275 (46%) 1,436 (52%) 42 (2%)

Movie-goer ratio 41% 22% 7%

Low quality movie

Number of listener 497 (5%) 7393 (74%) 2094 (21%)

Number of movie-goer 71 (7%) 954 (92%) 15 (1%)

Movie-goer ratio 14% 13% 1%


Analysis: Show Analysis: Show Duration and WoM Duration and WoM AccumulationAccumulation The accumulated impact of WoM shows the

“inverted U shape,” for the movie-going rate per WoM (probability) decreases after the peak

Good Movie-WoM & Movie going behavior

0

200

400

600

800

1000

1200

1400

1600

1800

0이상 2이상 4이상 6이상 8이상10이상

0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

0.45

Positive WoMListener

Movie goer

Movie going ratio

Longer show and more WoM


Analysis: CompetitionAnalysis: Competition

The number of movies

Good movies

Ordinary movies

Bad movies

Evenly distribute

d

Good Movie when evenly distributed

3 movies 3,027 2,668 1,977 2,956 4,924

6 movies 1,492 1,324 1,084 1,446 2,141

9 movies 991 913 794 956 1,328

Average number of viewers per movie according to competition scenarios

Movie mix in the market affects the total market size

Good and bad mix is better than all-average movies If you have a good movie, then release timing

strategy is critical


Discussion: Market Discussion: Market GrowthGrowth Effects of demand growth

Results from increased population (width) and increased frequency (depth) scenarios show that diminishing returns to scale

The width shows bigger effect in simultaneous release competition

Effects of movie supply and mix Total market size is positively related to

The number, quality and right mix of movies Marketing impacts and communication effects

interact in different fashion according to the movie quality and mix


Discussion: Critique Discussion: Critique DebatesDebates Debates

Critique influence (Handel, 1950; Litman, 1983) Critique influence timing: influencer vs. predictor

(Burzynski and Bayer, 1977; Eliashberg and Shugan, 1997)

Critique and consumer taste correlation and independence (Wanderer, 1970; Eliashberg and Shugan, 1997)

The model can incorporate the different assumptions and their consequences

What if critiques are ‘influencer,’ ‘predictor’ or both? It can be shown that the same results can be obtained

by changing parameters of initial marketing impact and WoM intensity


Hypothesis for Hypothesis for ReleaseReleaseCompetition

(Number of movies)

Market Size

Quality distribution (Number of good movies)

Competitor’s marketing

Marketing signals

WoM range/probability/Duration/consensus

Release Attractiveness

My Marketing

My Quality

—

+

—

—

+

+

Competition

AudienceCharacteristics


Discussion: Discussion: Competitive StrategyCompetitive Strategy Actual competition data

5 4

2, 3 11

LOW MEDIUM HIGH

LOW

MEDIUM

HIGH

Quality

Marketing

Feb. 2000Feb. 2000

5, 7 3

6 1, 4

2

LOW MEDIUM HIGH

LOW

MEDIUM

HIGH

Quality

Marketing

Jan. 2000Jan. 2000

3

1 22

LOW MEDIUM HIGH

LOW

MEDIUM

HIGH

Quality

Marketing

Early July. 2000Early July. 2000

5 3

2, 4

11

LOW MEDIUM HIGH

LOW

MEDIUM

HIGH

Quality

Marketing

Late July. 2000Late July. 2000


Discussion: Discussion: Competitive StrategyCompetitive Strategy Proposed taxonomy of movie quality and

marketing strategy

Weakest

Strongest

LOW MEDIUM HIGH

LOW

MEDIUM

HIGH

Quality

Marketing

Weak

Strong

Focused


Discussion: Modeling Discussion: Modeling IssuesIssues The model discussed in this paper is one focusing on the

complex consumer dynamics The concept of model “Saturation”

When applying agent-based simulation to a real and complex decision situation, it is more important that every additional variable and agent should be justified by increased insights and relevance

Heuristic approach Simplified analysis for central v. local communications

On consumer choice More empirical evidence is necessary for the model

improvement Acceleration phenomenon (e.g., The Passion of Christ,

Taegukgi in Korea)


Discussion: Modeling Discussion: Modeling IssuesIssues Model extension directions for practitioners

Market segmentation and competition Better consumer choice theory is necessary Overlapping release strategy

Theater objects Constraints and theater screen mix strategy

Producer objects Positive and negative feedback of innovation and imitation Resource-based theory of accumulating intangible assets

Combining the models for practical applications


Final ThoughtsFinal Thoughts ABM as a research method

Naturally lead researchers to think more about the “dynamics” and “adaptive behaviors” than traditionally thought to be adequate or acceptable

Implications We need more theoretical models, and Empirical data based on new models Especially in practical application purposes

Documents

SwarmFest 2004 May 11, 2004