Automated Bidding Strategy Adaption using Learning Agents in Many-to-Many e-Markets

In this paper the issue of bidding strategy learning in electronicmarkets is addressed. The primary aim is to identify machine learningtechniques which are best suited to learn bidding behaviourin electronic markets. The developed methodologies are appliedwithin a structured market engineering process to improve the qualityof market designs. Market simulations are carried out based ona discriminatory price double auction market design.The simulations of the market behavior are modelled in a multiagentsystem. The market participants which are (i) market makeragents (ii) supplier agents (iii) consumer agents act as autonomousagents on a simulated market. Supplier agents are static, i.e. notequipped with learning techniques. Consumer agents are modelledusing different machine learning methods for price determination.We develop the market simulation library MELBOURNE (MarketEngineering Library for Bidding Objectives Using Realtime NegotiationEnvironments). In the simulation runs using the library, weshow that learning algorithms always outperform heuristic methodsin price discovery.