Evolution of unplanned coordination in a market selection game

Abstract

This paper examines the evolution of unplanned coordination among independent agents in a market selection game, which is a non-cooperative repeated game with many agents (e.g., 100 agents) and several markets (e.g., five markets). Every agent is supposed to simultaneously choose a single market for maximizing its own payoff obtained by selling its product at the selected market. It is assumed that the market price is determined by a linear decreasing function of the total supply at each market. For example, if many agents choose a particular market, the market price at that market is low. On the contrary, the market price is high if only a small number of agents choose that market. In this manner, the market prices are determined by the actions of all agents. The point of our market selection game is to choose a market with a high market price, i.e., a market that is not chosen by many other agents. In this paper, the evolution of game strategies is performed by localized selection and mutation. A new strategy of an agent is probabilistically selected from its neighbors' strategies by the selection operation or randomly updated by the mutation operation. It is shown that the maximization of each agent's payoff through the genetic operations leads to the unplanned coordination of the market selection where the undesired concentration of agents is avoided. Through computer simulations, the unplanned coordination is compared with the planned global coordination realized by the maximization of the total payoff over all agents. Simulation results show that almost the same average payoff is obtained by these two schemes. That is, the selfish maximization of each agent's payoff through the genetic operations leads to a near-optimal result with respect to the total payoff over all agents. The unplanned coordination of the market selection realized by the evolution is also compared with some simple market selection methods such as a minimum transportation cost strategy, a random selection strategy, and an optimal strategy for the previous actions. Simulation results show that much higher payoff is obtained by the unplanned coordination than those methods.

Evolutionary models in economics: a survey of methods and building blocks

“…This includes: agent learning (Arthur 1991;Ishibuchi et al 2001;Klos and Nooteboom 2001), the evolution of norms, and conventions (Axelrod 1997;Thebaud and Locatelli 2001;Hodgson and Knudsen 2004), financial markets (Arthur et al 1996;Caldarelli et al 1998;LeBaron 2001;Levy et al 2000), diffusion of innovations and industry dynamics (Aversi et al 1997;Cantner and Pyka 1998a;Cantner et al 1998Cantner et al , 2000Gilbert et al 2001;Birchenhall 1998, 2005;Saint-Jean 2006;Schwoon 2006), land use and environmental management (Paker et al 2003), labor economics (Tassier and Menczer 2001;Gabriele 2002;Fagiolo et al 2004), and environmental policies (Janssen and Jager 2002;Carrillo-Hermosilla 2006). Multi-agent models have been also applied to various specific markets, including the textile market (Brannon et al 1997), fish market (Kirman and Vriend 2001), wholesale electricity market (Bower and Bunn 2001), and agricultural practices in a developing country (Lansing and Miller 2004).…”

Co-evolution of agents in self-organized criticality

“…Evolutionary computation has been applied successfully to various research fields from optimization to machine learning [2]. One interesting application area of evolutionary computation is evolutionary multi-agent system (MAS) in electronic commerce (EC).…”

The Evolution of Cooperative Behaviours in Physically Heterogeneous Multi-Robot Systems

“…For instance, a group of robots are evolved to display aggregation and coordinated movements [9][10][11], perform multi-objective tasks [12,13], and learn to cooperate to play competitive games [14][15][16][17], etc. Evolution of unplanned coordination among independent agents in market selection games is also examined [18]. Moreover, embodied evolution (EE), which is a new methodology for ER, constitutes a fully distributed evolutionary algorithm embodied in physical robots [19].…”