1. A Multi-Agent Prediction Market
Based on
Boolean Network Evolution
Janyl Jumadinova., Mihaela T. Matache and Prithviraj Dasgupta
WI-IAT.2011
Presenter: Yu Hsiang Huang
Date: 2011-12-09

2. Outline
• Introduction
• Related work
• Boolean Network based prediction market
• Experiment result
• Conclusion

3. Introduction
• Prediction market
• Multi-Agent system
• Boolean Network (BN)
– Boolean rule(0/1)
• BN vs. LMSR
– Eliminate the frequently fluctuating price

4. Related work
• Prediction Markets
• Outcome geopolitical events – US presidential elections
• Outcome of sporting events
• Predicting the box office performance of Hollywood movies
– Belief
• Opinion of individual trader about the outcome of a future event
• Market price

5. Boolean Network-Based
Prediction Market
• A. Prediction Market Preliminaries
– Major participants in agent-based prediction market
• A set of trading agents
• A market maker agent – central entity
– Outcome of an event is binary (will happen/won’t happen)
– Trading agent
• At time t
• Bet security (bought/sold/held)  discrete quantities
– Market maker agent
• Aggregate the price at which securities of event traded by agents
• Market price – probability of the outcome of the event
• Compare market price with actual decision in real world  cost

6. Trading period t Quantity=10
Security
bet
(bought)
Trading agents
Quantity=7
Security
bet
(sold)
Quantity=5
Security
bet
(bought)
Market maker agent
cost
Quantity=8 Quantity=10
Market Security Security
bet
(held) bet
(bought)
price cost
vs.
Actual
decision

7. Boolean Network-Based
Prediction Market(cont.)
• B. BN-based Prediction Market
– Major participants in BN-based prediction market
• Trading agent  buy and sell securities on behalf of human traders
• Market maker agent
• Information sources
– Based on traditional prediction market’s operation
• Belief
– Outcome of security corresponding to the event
– Used to calculate the market price
• State
– In BN, used to represent belief
– Updated using Boolean function
– 1 or ON : trading agent believes the event will happen
– 0 or OFF : trading agent believes the event won’t happen

8. Tradition PM
BN-based PM
1
2
2
4
3
4

10. Boolean Network-Based
Prediction Market (cont.)
• C. Trading Agents’ Boolean Belief Update
Market price State Information signal
Threshold
trust

11. Boolean Network-Based
Prediction Market (cont.)

12. (1
)

13. A
B
C
(5
D )
E
F

15. Boolean Network-Based
Prediction Market (cont.)
A F
D
C

16. Experimental results

18. Experimental results(cont.)
• B. Patterns and validation of the mean-field based
price aggregation mechanism
– Pattern formation plot
– Arranging nodes representing the trading agents in one dimensional array left to right
– State : 1 black plot
– State : 0 neutral plot

19. Experimental results(cont.)

20. Experimental results(cont.)
A F D C

21. Experimental results(cont.)
• C. Comparison to Conventional Prediction Markets
– LMSR vs. BN

22. Experimental results(cont.)

23. Experimental results(cont.)
stable

24. Experimental results(cont.)
• E. Scalability
– The number of trading agents v.s. model ‘s accuracy

25. Conclusion
• Boolean network
– Behavior of trading agent
• Aggregated market price
• Past beliefs
• Information flow
• BN v.s. LMSR
– Less fluctuation of the market price
– Analyze and predict the dynamics of prediction market
– Simpler
• Future
• Variation of weight and threshold parameters
• More possible state
• Limit untruthful belief