Bid Determination in Simultaneous Auctions: A Case Study

ABSTRACT This paper introdu es RoxyBot, one of the top-s oring agents in the First International Trading Agent Competition. A TAC agent simulates one vision of future travel agents: it represents a set of lients in simultaneous au tions, trading omplementary (e.g., airline ti kets and hotel reservations) and substitutable (e.g., symphony and theater ti kets) goods. RoxyBot fa ed two key te hni al hallenges in TAC: (i) allo ation|assigning pur hased goods to lients at the end of a game instan e so as to maximize total lient utility, and (ii) ompletion|determining the optimal quantity of ea h resour e to buy and sell given lient preferen es, urrent holdings, and market pri es. For the dimensions of TAC, an optimal solution to the allo ation problem is tra table, and RoxyBot uses a sear h algorithm based on A to produ e optimal allo ations. An optimal solution to the ompletion problem is also tra table, but in the interest of minimizing bidding y le time, RoxyBot solves the ompletion problem using beam sear h, produ ing approximately optimal ompletions. RoxyBot's ompleter relies on an innovative data stru ture alled a pri eline.