Introduction

The aim of this paper is to explore how well four different pricebot algorithms do in competition. Pricebots are agents that determine the best price at which to sell an item [1]. The algorithms we are comparing are game-theoretic, myopically-optimal, derivative-following, and Q-learning [1,2,3]. We use a simulator which has the four algorithms implemented in it and attempt this comparison as follows. Essentially, we have a starting population of n pricebots of which the proportions of this population are made up of the different pricing algorithms. Those algorithms that earn greater profits than other algorithms during a generation will increase in number, while the algorithms making less profit will decrease in number, always maintaining a constant population size of n.

Section 2 presents our model. Section 3 explains the game-theoretic algorithm, Section 4 presents the myopically-optimal algorithm, Section 5 looks at the derivative-following algorithm, and Section 6 presents the Q-learning algorithm. Section 7 examines the results of when game-theoretic plays against myopically-optimal. Section 8 looks at the results of when game-theoretic plays against derivative-following. Section 9 explores the myopically-optimal versus the derivative-following results. Section 10 looks at the results when game-theoretic, myopically-optimal, and derivative-following play against each other. Section 11 presents the results when all four algorithms compete against each other. Section 12 provides a conclusion and Section 13 gives references and acknowledgements.