Probabilistic Pricebots

Amy R. Greenwald and Jeffrey O. Kephart

Abstract

Past research has been concerned with the potential of embedding deterministic pricing algorithms into pricebots: software agents used by on-line sellers to automatically price Internet goods. In this work, probabilistic pricing algorithms based on no-regret learning are explored, in both high-information and low-information settings. It is shown via simulations that the long-run empirical frequencies of prices in a market of no-regret pricebots can converge to equilibria arbitrarily close to an asymmetric Nash equilibrium; however, instantaneous price distributions need not converge.