Title:General-sum Bidding Games

Abstract:Bidding games are extensive form games, where in each turn players bid in order to determine who will play next. Zero-sum bidding games (also known as Richman games) have been thoroughly studied, focusing on the fraction of the initial budget that can guaranty a victory for each player [Lazarus et al.'99, Develin and Payne '10].
We extend the theory of bidding games to general-sum two player games, analyzing pure subgame-perfect Nash equilibria (PSPE). Although the game has simultaneous steps, we prove that a PSPE exists for any initial budget partition. Curiously, while any bidding game has at least one PSPE, in the general case having more budget may hurt a player. Moreover, the PSPEs of the game may be arbitrarily bad for both players compared to other outcomes.
We show that if the underlying game has the form of a binary tree (only two actions available to the players in each node), then a there exists a natural PSPE with the following highly desirable properties: (a) players' utility is weakly monotone in their budget; (b) a Pareto-efficient outcome is reached for any initial budget; and (c) for any Pareto-efficient outcome there is an initial budget s.t. this outcome is attained. In particular, we can assign the budget so as to implement the outcome with maximum social welfare, maximum Egalitarian welfare, etc.
We discuss implications of this result for various games and mechanism design problems, including Centipede games, lobbying, voting games, and combinatorial auctions.