Markov Decision Approach for Time-Constrained Trading in Electronic Marketplace

The recent advent in Internet technology has made electronic trading increasingly popular. Nevertheless, existing electronic trading systems still heavily rely on human decision making. In order to facilitate the new trading environment, one of the main trends is to employ autonomous agents as representatives of human buyers and sellers. In particular, given a user requirement and an imposed deadline, an autonomous agent has to search for possible deals from an electronic marketplace on behalf of its owner. One problem such an agent must face is: given a collection of offers and the remaining time, should I accept the current best offer, or continue to search for a better one, at a risk of losing the current offers? Similar time-constrained trading problems have been studied long time ago by the Markov decision process community. This paper alters the formulation by adopting assumptions suitable for electronic trading environments, and then derive the optimal trading strategy in terms of the agent's expected utility. Three optimization methods are proposed in order to reduce the infinite state space into the one with a manageable size. Experimental results verify the effectiveness of the methods for practical use.