SocialTraders: A practical examples of agents.
2 likes183 views
This document discusses using social networking strategies in the CAT trading agent competition platform. It describes how traders in CAT markets can be connected through different social network topologies like hierarchical, ring, or fully-connected networks. The document also examines how network structures, trader mentalities like normal, credulous or skeptical, and reputation systems can impact trader performance and the performance of specialists. It presents results of experiments comparing average trader profits and other statistics under different social networking conditions.
![... [A] new research tournament was launched in 2007 to promote
research into automated mechanism design: the Trading Agent
Competition Market Design (or CAT) Tournament.
The CAT Tournament comprises a series of artificial parallel markets,
designed to mirror the competition between global stock markets. These
parallel markets, called specialists , are created by entrants to the
Tournament, and they compete with one another to attract and retain
traders, who are potential buyers and sellers of some abstract commodity.
The traders are software agents created and operated by the organisers of
the CAT Tournament, in a reversal of the usual Trading Agent Competition
structure.
Extract taken from E. Robinson, P. McBurney and X. Yao. How specialised are specialists?
Generalisation properties of entries from the 2008 and 2009 TAC Market Design Competitions. 2010.](https://image.slidesharecdn.com/cx2zcpgaq8mbyxw7in8r-signature-1d71bc69af8e1ce5cfd06165f1692fb262b99bc72e1cbb4e076cc65ccd63e69f-poli-180215150739/85/SocialTraders-A-practical-examples-of-agents-11-320.jpg)
SocialTraders: A practical examples of agents.
- 3. “Catallactics” Cat`al`lac´tics n. 1. The science of exchanges, a branch of political economy.
- 4. “Catallactics” Cat`al`lac´tics n. 1. The science of exchanges, a branch of political economy.
- 10. Agent Research
- 11. ... [A] new research tournament was launched in 2007 to promote research into automated mechanism design: the Trading Agent Competition Market Design (or CAT) Tournament. The CAT Tournament comprises a series of artificial parallel markets, designed to mirror the competition between global stock markets. These parallel markets, called specialists , are created by entrants to the Tournament, and they compete with one another to attract and retain traders, who are potential buyers and sellers of some abstract commodity. The traders are software agents created and operated by the organisers of the CAT Tournament, in a reversal of the usual Trading Agent Competition structure. Extract taken from E. Robinson, P. McBurney and X. Yao. How specialised are specialists? Generalisation properties of entries from the 2008 and 2009 TAC Market Design Competitions. 2010.
- 12. Source: JCAT: The Software Platform for CAT Games, version 1.05 edition, May 1, 2009 JavaCAT.
- 13. Trader A Trader B
- 14. Trader A Trader B
- 15. Trader A Trader B BUY@ 1.234
- 16. Trader A Trader B BUY@ 1.234
- 17. Trader A Trader B BUY@ 1.234 BUY@ 4.321
- 25. “Networks ofinfluencebetweenmarket traders can significantlyimpactupontheirindividualperformanceandthe performanceofthespecialists theyinteract with.”
- 26. “Networks ofinfluencebetweenmarket traders can significantlyimpactupontheirindividualperformanceandthe performanceofthespecialists theyinteract with.” “Assumingthehypothesis is true,thentowhat quantifiable extentandunderwhatconditions donetworks ofinfluence betweenmarket traders impact upontheirindividual performanceandtheperformanceofthespecialists they interactwith?”
- 27. Contributions.
- 28. Contributions. Server-orientedcommunication;‘dumb’mimicking. GuanGui (c/o. Dr. MillerandDr. Kirley), TheUniversityofMelbourne.
- 29. Source: G. Gui. CatSocialNetwork: User Guide, version 1.05 edition, Summer 2010.
- 30. Contributions. Server-orientedcommunication;‘dumb’mimicking. GuanGui (c/o. Dr. MillerandDr. Kirley), TheUniversityofMelbourne.
- 31. Contributions. Server-orientedcommunication;‘dumb’mimicking. GuanGui (c/o. Dr. MillerandDr. Kirley), TheUniversityofMelbourne. Traderexperiences;theprecursortogivingadvice.
- 32. Contributions. Server-orientedcommunication;‘dumb’mimicking. ‘Giving’advice;Networktopologies. GuanGui (c/o. Dr. MillerandDr. Kirley), TheUniversityofMelbourne. Traderexperiences;theprecursortogivingadvice.
- 33. Hierarchical Ring Fully Connected Fully Connected
- 34. Contributions. Server-orientedcommunication;‘dumb’mimicking. ‘Giving’advice;Networktopologies. GuanGui (c/o. Dr. MillerandDr. Kirley), TheUniversityofMelbourne. Traderexperiences;theprecursortogivingadvice.
- 35. Contributions. Server-orientedcommunication;‘dumb’mimicking. ‘Giving’advice;Networktopologies. ReputationSystem;N-armedbandit. GuanGui (c/o. Dr. MillerandDr. Kirley), TheUniversityofMelbourne. Traderexperiences;theprecursortogivingadvice.
- 37. Contributions. Server-orientedcommunication;‘dumb’mimicking. ‘Giving’advice;Networktopologies. ReputationSystem;N-armedbandit. GuanGui (c/o. Dr. MillerandDr. Kirley), TheUniversityofMelbourne. Traderexperiences;theprecursortogivingadvice.
- 38. Contributions. Server-orientedcommunication;‘dumb’mimicking. ‘Giving’advice;Networktopologies. ReputationSystem;N-armedbandit. ‘Taking’Advice;Agentmentalities. GuanGui (c/o. Dr. MillerandDr. Kirley), TheUniversityofMelbourne. Traderexperiences;theprecursortogivingadvice.
- 40. SocialTraderGUI.
- 41. No social networking strategy A hierarchical network A ring network A small world network A random social networking strategy A fully-connected network (response variable: Average Trader Profit) Normal Credulous ScepticalTrader type / Social Network Experiments.
- 42. No social networking strategy 21.9597262766661 A hierarchical network 21.9635335736683 24.1223991979382 22.5708117023643 A ring network 24.4381454210531 31.9911033817235 21.2901718805965 A small world network 28.1729734360446 25.3014017341968 21.2901718805965 A random social networking strategy 31.746038347001 32.6369474322239 26.3634805824215 A fully-connected network 35.7998881478977 27.0993644179867 30.0296602534778 (response variable: Average Trader Profit) Normal Credulous ScepticalTrader type / Social Network Experiments.
- 43. TraderProfit.
- 44. 0 10 20 30 40 Profits Smallworld Ring Hierarchical Random Fully-conntected No network Normal Credulous Sceptical TraderProfit.
- 45. 0 10 20 30 40 Profits Smallworld Ring Hierarchical Random Fully-conntected No network Normal Credulous Sceptical TraderProfit.
- 46. T-Value. -4.07726925759703 Confidence. 0.95 Degrees of Freedom. 148 Standard Error of Difference. 1.23318153195583 Mean Difference. -5.02801314927993 Standard Deviation. 6.7385672298091 8.28534631117571 Variance. 45.4082883106571 68.6469634961129 Mean. 21.9597262766661 26.987739425946 YES No Social Network Informed Social StrategyPlatform state / Statistical Measure
- 51. Clustering&Hubs.