A Multi-Agent Prediction Market based on Raj Dasgupta Partially - PowerPoint PPT Presentation

A Multi-Agent Prediction Market based on Partially Observable Stochastic Game Janyl Jumadinova, A Multi-Agent Prediction Market based on Raj Dasgupta Partially Observable Stochastic Game Outline Introduction Research Problem POSGI Janyl Jumadinova, Raj Dasgupta Trading Agents’ Strategy Experimental Results C-MANTIC Research Group Computer Science Department Future Work University of Nebraska at Omaha, USA ICEC 2011 1 / 37

Outline A Multi-Agent Prediction Market based on Partially Observable Stochastic Game Problem: Traders’ behavior in a prediction market and Janyl Jumadinova, its impact on the prediction market’s behavior Raj Dasgupta Outline Solution: A multi-agent system that formalizes the Introduction strategic behavior and decision making by market’s Research Problem participants based on a partially observable stochastic POSGI Trading Agents’ game Strategy Experimental Results Experimental validation: Comparison with other Future Work trading approaches 1 / 37

Prediction Market A Multi-Agent Prediction Market based on Partially Observable Stochastic Game Janyl Jumadinova, A Prediction market is Raj Dasgupta a market-based mechanism used to Outline Introduction - combine the opinions on a future event from different Research Problem people and POSGI - forecast the possible outcome of the event based on the Trading Agents’ Strategy aggregated opinion Experimental Results Future Work 2 / 37

Prediction Market A Multi-Agent Prediction Market based on Partially Observable Stochastic Game Janyl Jumadinova, Raj Dasgupta Outline Introduction Research Problem POSGI Trading Agents’ Strategy Experimental Results Future Work 3 / 37

Prediction Market A Multi-Agent Prediction Market based on Partially Observable Stochastic Game Janyl Jumadinova, Raj Dasgupta Outline Introduction Research Problem POSGI Trading Agents’ Strategy Experimental Results Future Work 4 / 37

Prediction Market A Multi-Agent Prediction Market based on Partially Observable Stochastic Game Janyl Jumadinova, Raj Dasgupta Outline Introduction Research Problem POSGI Trading Agents’ Strategy Experimental Results Future Work 5 / 37

Prediction Market A Multi-Agent Prediction Market based on Partially Observable Stochastic Game Janyl Jumadinova, Raj Dasgupta Outline Introduction Research Problem POSGI Trading Agents’ Strategy Experimental Results Future Work 6 / 37

Prediction Market A Multi-Agent Prediction Market based on Partially Observable Stochastic Game Janyl Jumadinova, Raj Dasgupta Outline Introduction Research Problem POSGI Trading Agents’ Strategy Experimental Results Future Work 7 / 37

Prediction Market A Multi-Agent Prediction Market based on Partially Observable Stochastic Game Janyl Jumadinova, Raj Dasgupta Outline Introduction Research Problem POSGI Trading Agents’ Strategy Experimental Results Future Work 8 / 37

Prediction Market A Multi-Agent Prediction Market based on Partially Observable Stochastic Game Janyl Jumadinova, Raj Dasgupta Outline Introduction Research Problem POSGI Trading Agents’ Strategy Experimental Results Future Work 9 / 37

Prediction Market A Multi-Agent Prediction Market based on Partially Observable Stochastic Game Janyl Jumadinova, Raj Dasgupta Outline Introduction Research Problem POSGI Trading Agents’ Strategy Experimental Results Future Work 10 / 37

Prediction Market A Multi-Agent Prediction Market based on Partially Observable Stochastic Game Janyl Jumadinova, Raj Dasgupta Outline Introduction Research Problem POSGI Trading Agents’ Strategy Experimental Results Future Work 11 / 37

Prediction Market A Multi-Agent Prediction Market based on Partially Observable Stochastic Game Janyl Jumadinova, Raj Dasgupta Outline Introduction Research Problem POSGI Trading Agents’ Strategy Experimental Results Future Work 12 / 37

Prediction Market A Multi-Agent Prediction Market based on Partially Observable Stochastic Game Janyl Jumadinova, Raj Dasgupta Outline Introduction Research Problem POSGI Trading Agents’ Strategy Experimental Results Future Work 13 / 37

Prediction Market Main Features A Multi-Agent Prediction Market based on Partially A prediction market is run for a real-life unknown event Observable Stochastic Game Each event has a finite duration Janyl Jumadinova, Raj Dasgupta Each event’s outcome has a security associated with it Traders buy and sell the securities based on their beliefs Outline about the outcome of the event Introduction Research Problem Traders’ beliefs are expressed as probabilities POSGI Market maker aggregates the probabilities from all the Trading Agents’ Strategy traders into a single probability, market price Experimental Market price of a security represents the probability of Results the outcome of an event associated with that security Future Work happening Traders get paid according to their reported beliefs 14 / 37

A Multi-Agent Prediction Market A Multi-Agent Prediction Market based on Partially Observable Stochastic Game Janyl Jumadinova, Raj Dasgupta Outline Introduction Research Problem POSGI Trading Agents’ Strategy Experimental Results Software trading agents perform calculations and trade Future Work on behalf of human traders Provides testbed for modeling different strategic behaviors of traders through simulations with trading agents 15 / 37

Do Prediction Markets Work? Yes, evidence from real markets, laboratory experiments, and theory A Multi-Agent Prediction Market based on Partially Observable I.E.M. beat political polls 451/596 [Forsythe 1999, Berg Stochastic Game 2001, Pennock 2002] Janyl Jumadinova, Raj Dasgupta HP market beat sales forecast 6/8 [Plott 2000] Outline Sports betting markets provide accurate forecasts of Introduction game outcomes [Debnath 2003, Schmidt 2002] Research Problem Market games work [Pennock 2001] POSGI Laboratory experiments confirm information Trading Agents’ Strategy aggregation [Forsythe 1990, Plott 1997, Chen 2001] Experimental Results Theory of Rational Expectations [Lucas 1972, Grossman Future Work 1981] and more... 16 / 37

Research Problem Addressed A Multi-Agent Develop a formal, game-theoretic model of the trading Prediction Market based on Partially agent behavior in prediction markets including Observable Stochastic Game - impact of information from external sources on trading Janyl Jumadinova, agent decisions/behavior, Raj Dasgupta - a solution concept for calculating the equilibrium Outline strategies of the trading agents Introduction Research Problem Research Questions: POSGI How does different traders’ behaviors affect market Trading Agents’ prices? Strategy Experimental What trading strategies give the highest utilities to the Results traders? Future Work How can prediction markets incentivize traders to participate and report their beliefs truthfully in order to achieve a higher prediction accuracy? 17 / 37

Our Solution A Multi-Agent Prediction Market based on Partially Observable Stochastic Game Janyl Jumadinova, Raj Dasgupta A partially observable stochastic game with information Outline (POSGI)-based model of the trading agent behavior Introduction A correlated equilibrium (CE)-based solution to Research Problem determine equilibrium strategy in the POSGI POSGI Trading Agents’ representation Strategy Experimental Results Future Work 18 / 37

Partially Observable Stochastic Game with Information (POSGI) A Multi-Agent Prediction Market based on Partially Observable Stochastic Game Janyl Jumadinova, Raj Dasgupta Outline Introduction Research Problem POSGI Trading Agents’ Strategy Experimental Results Future Work 19 / 37

Partially Observable Stochastic Game with Information (POSGI) A Multi-Agent Prediction Market based on Partially Observable Stochastic Game Janyl Jumadinova, Raj Dasgupta Outline Introduction Research Problem POSGI Trading Agents’ Strategy Experimental Results Future Work - Logarithmic Market Scoring Rule (LMSR) [Hanson 2007] gives formula to calculate aggregate market price from the outstanding quantity of a security 19 / 37

Partially Observable Stochastic Game with Information (POSGI) A Multi-Agent Prediction Market based on Partially Observable Stochastic Game Janyl Jumadinova, Raj Dasgupta Outline Introduction Research Problem POSGI Trading Agents’ Strategy Experimental Results Future Work - Belief state is updated using a Bayesian model of past beliefs, past actions and current observation 20 / 37

Partially Observable Stochastic Game with Information (POSGI) A Multi-Agent Prediction Market based on Partially Observable Stochastic Game Janyl Jumadinova, Raj Dasgupta Outline Introduction Research Problem POSGI Trading Agents’ Strategy Experimental Results Future Work - Information signal can be {− 1 , 0 , 1 } representing positive, neutral, or negative information 21 / 37

Partially Observable Stochastic Game with Information (POSGI) A Multi-Agent Prediction Market based on Partially Observable Stochastic Game Janyl Jumadinova, Raj Dasgupta Outline Introduction Research Problem POSGI Trading Agents’ Strategy Experimental Results Future Work 22 / 37