Decision Making, Strategy dynamics, and Crowd Formation in Agent-based models of Competing Populations

TL;DR

This paper develops a generalized theory for the Minority Game, analyzing strategy rankings and decision distributions in the efficient phase, improving agreement with numerical results by accounting for tied strategies.

Contribution

It introduces a new theoretical framework based on strategy ranking patterns that better models agent decision dynamics, especially in the presence of tied strategies.

Findings

The theory accurately predicts the distribution of agents' decisions.

It improves agreement with numerical simulations over previous crowd-anticrowd models.

The model accounts for the effects of tied strategies on decision dynamics.

Abstract

The Minority Game (MG) is a basic multi-agent model representing a simplified and binary form of the bar attendance model of Arthur. The model has an informationally efficient phase in which the agents lack the capability of exploiting any information in the winning action time series. We develop a theory based on the ranking patterns of the strategies and the number of agents using a particular rank of strategies as the game proceeds. The theory is applied to calculate the distribution or probability density function in the number of agents making a particular decision. From the distribution, the standard deviation in the number of agents making a particular choice (e.g., the bar attendance) can be calculated in the efficient phase as a function of the parameter m specifying the agent's memory size. Since situations with tied cumulative performance of the strategies often occur in the efficient phase and they are critical in the decision making dynamics, the present theory is developed to take into account the effects of tied strategies. The analytic results are found to be in better agreement with numerical results, when compared with the simplest forms of the crowd-anticrowd theory in which cases of tied strategies are ignored. Hence the present theory represents a generalization of the crowd-anticrowd theory of multi-agent models, by focusing on the evolution of strategy ranking patterns instead of the performance of individual strategies. Paper to be presented in the 10th Annual Workshop on Economic Heterogeneous Interacting Agents (WEHIA 2005), 13-15 June 2005, University of Essex, UK.