Research

Work in progress

First-price Sealed-bid Auctions with Smoothly Ambiguity Averse Bidders

Tianyu Ma and Frank Riedel

We study first-price sealed-bid auctions with two risk-neutral bidders who have independent private values but are smoothly ambiguity averse about the distribution of their opponents' valuations. Adopting the smooth ambiguity model of Klibanoff, Marinacci, and Mukerji (2005), we separate ambiguity attitudes from risk and establish existence and uniqueness of a symmetric, non-decreasing equilibrium. Although the primitives feature independent private values, we show that equilibrium bidding is observationally equivalent to that in a subjective expected utility auction with correlated private values: there exists a correlated-values environment whose interim beliefs reproduce the same bidding function, and under constant relative ambiguity aversion these effective beliefs become type independent. Greater ambiguity aversion induces more aggressive bidding, and as ambiguity aversion tends to infinity the equilibrium converges to the maxmin expected utility benchmark of Lo (1998). Finally, smooth ambiguity breaks the standard ranking between first-price and second-price auctions: the first-price format narrows the seller's revenue range across priors, while bidder welfare admits no uniform ranking and may favor either format depending on ambiguity attitudes.

Ambiguous Contracts with an \(\alpha\)-MEU Agent

Tianyu Ma

We study ambiguous contracts in a finite moral-hazard principal-agent model in which the agent has \(\alpha\)-MEU preferences (Ghirardato, Maccheroni, and Marinacci, 2004). Dütting, Feldman, Peretz, and Samuelson (2024) show that under maxmin expected utility (Gilboa and Schmeidler, 1989) ambiguity can enlarge the set of implementable actions and that, under consistency, optimal ambiguous contracts admit a single-outcome-payment (SOP) structure. We show that these conclusions are not robust to moderate ambiguity attitudes. When \(\alpha < 1\), consistency is no longer without loss of generality: an inconsistent set of payment functions can strictly improve the principal's payoff, so consistency becomes a substantive credibility restriction. Imposing consistency, we derive sharp thresholds for implementability. If the target action is dominated by a mixture of \(d\) actions, then no consistent contract can \(\alpha\)-implement it when \(\alpha \le 1/d\); under full dimensionality, this threshold is sharp. We also show that if \(\alpha \le 1/(n-1)\), every consistent \(\alpha\)-incentive-compatible set of payment functions is equivalent to a classic contract. Finally, the SOP characterization fails in general for \(\alpha \in (0,1)\), although any consistent incentive-compatible ambiguous contract can still be reduced to at most \(n-1\) payment functions.

Contracting Against Hidden AI Degradation

Tianyu Ma and Weichu Wang