Spiteful Bidding in the Dollar Auction

Marcin Waniek, Agata Niescieruk, Tomasz P. Michalak, Talal RahwanIJCAI 2015

Abstract

The "Dollar Auction" is a classic game theory paradox where rational players often engage in ruinous bidding wars. While traditional models suggest escalation should not occur under finite budgets, this paper introduces the "spite coefficient" to explain why it does. By modeling agents who value their opponent's loss, the authors demonstrate that malicious strategies can force rational players into catastrophic budget depletion, particularly when budgets are unequal or knowledge is asymmetric.

Introduction: The Paradox of Escalation

The Dollar Auction, first proposed by Martin Shubik in 1971, is a deceptively simple game: two players bid for a dollar. The unique constraint is that it is an "all-pay" auction—both the winner and the loser must pay their final bids to the auctioneer. In experimental settings, players frequently bid far beyond the value of the prize (e.g., bidding $1.10 to win $1.00 to avoid losing a $1.05 bid), leading to a "paradox of escalation."

While landmark research by O’Neill (1986) proved that rational players with finite budgets should avoid such wars by placing a specific "golden bid" or folding immediately, real-world behavior remains stubbornly irrational. This paper addresses the gap by asking: What if players aren't just trying to win money, but are also motivated by the desire to see their opponent lose?

Methodology: Modeling Spite

The authors introduce a formal variable, the spite coefficient ($\alpha$), to quantify a player's motivation:

  • Non-spiteful ($\alpha = 0$): Maximizes only their own profit.
  • Spiteful ($0 < \alpha < 1$): Cares about both their profit and the opponent's loss.
  • Malicious ($\alpha = 1$): Cares only about maximizing the opponent's loss, regardless of their own cost.

Using these definitions, the game is modeled as a state-space graph of bid pairs $(x_1, x_2)$. The authors use backtracking to identify Subgame Perfect Equilibria across various scenarios, including equal and unequal budgets, and varying states of knowledge regarding an opponent's spitefulness.

Figure 6: Partitioning of the bid space into strategic regions (A, B, E) and equilibrium points (X, Y) for different bid states.

Key Insights: The Trap of the Rational Player

The research reveals that a non-spiteful player who assumes their opponent is also rational is highly vulnerable. A malicious player can "lure" a rational opponent into bidding by making the next step appear to be a winning move, only to force them into a total budget drain.

One of the most counter-intuitive findings is the Power of the Small Budget. A malicious player with a smaller budget can be more dangerous because they appear "easy to beat." This perception encourages the non-spiteful player to engage, only to find themselves forced to pay a high price ($b_{small} + 1$) to finally win the stake, resulting in a net loss.

Results: Escalation as a Rational Strategy

The paper proves that for a malicious agent, the optimal strategy is to outbid the opponent by the minimum possible increment ($+1$) until the very end. This keeps the opponent "in the game" and maximizes their eventual loss, as illustrated by the "staircase" of bids in the state-space analysis.

Figure 5: The "staircase" of bids representing the escalation path in a malicious bidding scenario, where players are forced toward high-bid states.

Key results include:

  • Theorem 1 (The Malicious Win): If a non-spiteful player follows standard rational strategies against a malicious player, the malicious player will always win the stake and force the opponent to spend nearly their entire budget.
  • Spite Thresholds: Players with a spite coefficient $\alpha > 0.5$ behave almost identically to purely malicious agents, while those below $0.5$ behave more like rational actors.

Conclusion and Future Work

Spite completely breaks the traditional O'Neill solution. If an opponent is even slightly spiteful, the truly "rational" move is often to not participate at all. Future work may explore the opposite end of the spectrum—altruistic bidders—and how bounded rationality or communication (deals and threats) might further alter these dynamics in real-world conflicts like corporate R&D wars or lobbying.

Summary updated

BibTeX
@inproceedings{waniek2015spiteful,
  author = {Marcin Waniek and Agata Niescieruk and Tomasz P. Michalak and Talal Rahwan},
  title = {{Spiteful Bidding in the Dollar Auction}},
  booktitle = {International Joint Conference on Artificial Intelligence},
  year = {2015},
}
Figure