Abstract
We investigate the problem of revenue optimal mechanism design under the context of the partial rationality model, where buyers randomize between two modes: rational and irrational. When a buyer is irrational, he acts according to certain fixed strategies, such as bidding his true valuation. The seller cannot observe the buyer's valuation or his rationality mode but treats them as random variables from known distributions. The seller's goal is to design a single-shot auction that maximizes her expected revenue. Our findings show that, even for the simplest setting with one buyer, direct value revelation loses generality, and the posted-price mechanism is no longer optimal. In fact, more complex mechanisms yield higher revenue. For multiple bidders with IID uniform valuations, we show that first-price auctions outperform second-price auctions when irrational buyers are truthful.
Introduction
Since Roger Myerson’s seminal work in 1981, the gold standard for optimal auction design has assumed that every bidder is a perfectly rational agent. However, human behavior in real-world auctions is often "messy." Bidders may be "lazy," repeat previous bids regardless of context, or simply bid their true valuation because they lack strategic knowledge. This paper, presented at IJCAI 2015, challenges the assumption of perfect rationality by introducing a model where bidders are only "partially rational."
In this model, a bidder acts rationally with probability $p$ and follows a fixed, irrational strategy (such as truthful bidding) with probability $1-p$. The seller knows the distributions of valuations and the probability of rationality but cannot observe the specific mode of a bidder in a given instance. This shift reveals that classic rules—such as the optimality of simple posted prices—no longer hold.
Methodology
The authors employ a combination of mechanism design theory and novel geometric arguments to analyze optimal revenue. Instead of looking at the auction rules directly, they represent the mechanism through the bidder's utility function $u(x)$, where $x$ is the bidder's private valuation.
Modeling Behavioral Modes
A key step in the methodology is distinguishing between the paths taken by rational and irrational bidders. While a rational bidder will strategically report a value $s(v)$ to maximize utility, an irrational bidder follows a fixed strategy $c(v)$. The seller must design a mechanism $M$ that accounts for both possibilities simultaneously.
Geometric Variation
To prove the sub-optimality of standard auctions, the authors use a "variation" technique. They demonstrate that if a utility function has a "breaking point" or "kink" (characteristic of posted-price auctions), it can be slightly modified to a smoother curve. This modification increases revenue from irrational bidders more than it loses from rational ones. This is a purely theoretical study; no empirical datasets were used. The findings are derived from mathematical proofs using standard theoretical distributions, such as IID uniform valuations.
Key Insights and Results
The Death of the Posted Price
In standard Myersonian theory, a simple "take-it-or-leave-it" posted price is often optimal for a single bidder. However, the authors prove the "Smoothness Theorem": the optimal utility function for the rational mode must be $C^1$ (continuously differentiable). Because a posted price creates a kink in the utility curve, it is never optimal for partially rational bidders.
Complexity and the Revelation Principle
Contrary to the "simplicity is best" mantra, this model suggests that complexity is a feature. The more "menu items" (bid/price/probability combinations) the seller provides, the higher the revenue. While an infinite menu is theoretically optimal, the authors show that a $k$-piecewise linear mechanism can approximate the optimal revenue with an error of at most $1/k$.
Furthermore, the "Revelation Principle" is partially challenged. While asking only for bids (value revelation) loses the generality of the principle, the maximum achievable revenue remains the same as when asking for both valuation and rationality mode (type revelation). The difference lies in how the rational bidder is forced to misreport their value in the bid-only version.
First Price vs. Second Price Auctions
The paper provides a concrete counter-example to the standard revenue equivalence theorem. In a multi-bidder setting with identical uniform distributions where irrational bidders are truthful, the First Price Auction (FPA) extracts strictly more revenue than the Second Price Auction (SPA). The FPA is better suited to exploit the truthfulness of irrational players.
Conclusion and Future Work
This research highlights the significant impact of behavioral assumptions on mechanism design. However, several challenges remain for future exploration:
- Parameter Estimation: The model assumes the seller knows the exact irrational strategy $c(x)$ and the probability $p$, which are difficult to estimate in practice.
- Dynamic Settings: In repeated auctions, rational bidders might strategically "mask" their rationality to secure better future deals.
- Distributional Generalization: Some results are currently tied to uniform distributions; extending these to a broader class of "regular" distributions remains an open challenge.
Summary updated