Adaptive strategies for price markdown in a multi-unit descending price auction: A comparative study

In a descending price multi-unit Dutch auction over the Internet, auctioneer gradually decrements per unit price of the item during the course of the auction. We investigate the problem of finding a decrementing price sequence that maximizes auctioneer’s total expected revenue using single-agent Reinforcement Learning. We contrast actual-return (Monte-Carlo based) learning methods with one step Q-learning and also with other adaptive strategies and report extensive comparative performance study. In our experimental design, we model bidders’ strategies in a unique way using various bid functions that capture realistic strategic behavior of bidders in such auction games. Monte-Carlo control algorithm developed here offers consistent performance and yields high average returns in all the experiments.