Partial Replication of: Stronger CDA Strategies through Empirical Game-Theoretic Analysis and Reinforcement Learning



This study seeks to replicate the results of a study originally conducted by Julian L. Schvartzman and Michael P. Wellman [12, 13] on the relative strength of eight previously proposed principled strategies for automatically bidding in a continuous double auction (CDA). The strategies are compared by empirical game theoretic analysis [19], for which results are in terms of equilibria. The result of the original study study was that the GD and GDX strategies are superior to the other six strategies because they are the only strategies to appear in an equilibrium, and thus the hypothesis of this study is that the same results will be observed. Methods used during this replication follow closely the design of the auction and bidding strategies of the original study. The result of this study is that the GDX strategy is the only one of the eight strategies played in equilibrium, and thus GDX is superior to the other seven strategies tested, partially confirming the hypothesis.