Workshop

This lecture will propose taking a mechanism design approach to the classroom. The classroom is appealing as a domain for mechanism design as many researchers are also instructors in the classroom.

A central concept in mechanism design for the classroom is scoring rules. We will review scoring rules as mechanisms for eliciting beliefs, and propose an optimization framework for identifying good scoring rules. This optimization framework admits many the research directions that have been studied for traditional mechanism design, such as simple versus optimal, robustness to prior beliefs, and sample complexity.

Some results will be taken from joint work with Yingkai Li, Liren Shan, and Yifan Wu.

Incentivizing the acquisition and aggregation of information is a key task of the modern economy (e.g., financial markets, prediction markets, auctions, organizations). The talk will discuss the design of optimal mechanisms for this task. I assume that agents are rationally inattentive (RI), i.e., engage in costly flexible information acquisition subject to uniformly posterior separable (UPS) information costs. A principal aims to procure a given information structure from one or many the agents at a minimal cost by designing design general dynamic mechanisms with report- and state-contingent payments. If the agents are risk-neutral, prediction markets implement the first-best. If the agents are risk-averse, no mechanism can approximate the first-best cost—not even those that harness the “wisdom of the crowd” by employing a large number of “informationally small” agents. This inefficiency derives from the combination of agents’ moral hazard and adverse selection. Our characterization of incentive compatibility, which exploits an equivalence between proper scoring rules and UPS information costs, is tractable and portable to other design settings with RI agents (e.g., principal-expert and screening problems).

Based on joint work with Alex Bloedel.

We consider a model of matching in trading networks in which firms can enter into bilateral contracts. In trading networks, stable outcomes, which are immune to deviations of arbitrary sets of firms, may not exist. We define a solution concept called trail-stability. Trail-stable outcomes are immune to consecutive, pairwise deviations between linked firms. We show that any trading network with bilateral contracts has a trail-stable outcome whenever firms’ choice functions satisfy the full substitutability condition.

Most of the algorithmic mechanism design literature assumes that a buyer's value for an item is private and independent from other buyers.  In contrast, the interdependent values model [Milgrom Weber '82] captures when one buyer's private information about an item impacts how much another buyer is willing to pay for it.  

In order to allocate a single-item efficiently and subject to incentive compatibility, valuations must satisfy a restrictive "single-crossing" condition, and prior to 2018, all work made this assumption.  However, since then, a series of work has broadened the domain of study by aiming for approximation and leveraging more natural structural assumptions.  

I will introduce the model, develop intuition for how it differs from the standard independent private value setting, and survey approximation results from recent years.

Based on joint works with Alon Eden, Michal Feldman, Amos Fiat, Anna Karlin, Simon Mauras, Divyarthi Mohan, and Shuran Zheng.

In a combinatorial auction, there are m items for sale to n bidders, each with a valuation function over sets of items (that is, fully describing a bidder's valuation function requires 2^m numbers). The designer's goal is to allocate the items in a way that optimizes the total welfare: the sum over all bidders of their value for the set they receive.

In this talk, I'll provide a brief overview of seminal results for this problem, and a deeper overview of several recent results. This talk will focus on the communication complexity of resolving this problem -- the amount of communication necessary among the bidders to find an (approximately) optimal allocation.

I'll highlight one recent result in further detail, which settles the communication complexity of achieving any approximation guarantee using a maximal-in-range mechanism. This result is joint work with Frederick Qiu.

It is common to assume that players in a game will adopt Nash equilibrium strategies. However, experimental studies have demonstrated that Nash equilibrium is often a poor description of human players' behavior, even in unrepeated normal-form games. Nevertheless, human behavior in such settings is far from random. Drawing on data from real human play, the field of behavioral game theory has developed a variety of models that aim to capture these patterns.

This talk will survey over a decade of work on this topic, built around the core idea of treating behavioral game theory as a machine learning problem. It will touch on questions such as:

- Which human biases are most important to model in single-shot game theoretic settings?

- What loss function should be used to evaluate and fit behavioral models?

- What can be learned about examining the parameters of these models?

- How can richer models of nonstrategic play be leveraged to improve models of strategic agents?

- When does a description of nonstrategic behavior "cross the line" and deserve to be called strategic?

- How can advances in deep learning be used to yield stronger--albeit harder to interpret--models?

Finally, there has been much recent excitement about large language models such as GPT-4. The talk will conclude by describing how the economic rationality of such models can be assessed and presenting some initial experimental findings showing the extent to which these models replicate human-like cognitive biases.