This annotated reading list surveys emerging works at the intersection of information design and large language models (LLMs). While classical information design theory studies signaling in abstract mathematical models, real-world communication often occurs in natural language. We highlight papers that use LLMs to elicit and communicate information through natural-language messages, cover techniques used in "information design + LLM" research such as language-space optimization and LLM proxies, and discuss papers on LLM persuasion. We aim to illustrate the potential of LLMs to bridge the gap between the theory and practice of information design.
Algorithms increasingly mediate repeated strategic interactions in marketplaces, from automated pricing to auction bidding. When one party commits to a learning algorithm, the other party can respond strategically over time by steering the algorithm's internal state toward a favorable long-run outcome. This note surveys a line of work that studies this "learning-as-commitment" perspective via a geometric object we call a menu : the convex set of long-run outcomes an opponent can induce against a fixed learning rule. Menus provide a common language for (i) comparing learning algorithms against strategic opponents, (ii) optimizing over learning rules under uncertainty about opponent objectives, and (iii) characterizing when an opponent can manipulate learning dynamics beyond what they could achieve with a static strategy. Using this machinery, we converge upon no-swap-regret algorithms as an "optimal" commitment strategy for robust learning against a strategic opponent. We also identify principled generalizations of no-swap-regret beyond normal-form games that preserve the same strategic guarantees while remaining computationally tractable.
This is the eleventh annual collection of profiles of the junior faculty job market candidates of the SIGecom community. The forty one candidates for 2026 are listed alphabetically and indexed by research areas that define the interests of the community. The candidates can be contacted individually, or collectively via the moderated mailing list ecom-candidates2026@acm.org.
ML model evaluation often takes one of two main approaches: risk minimization , associated with "high accuracy" or calibration , meaning that predictions are "trustworthy" and can be interpreted from a probabilistic lens. There is an extensive line of work which has studied the relationship between risk minimization and calibration, mostly focusing on the binary outcome setting. Even in the binary setting, there are a variety of proposed calibration metrics which non-trivially interact. In the multiclass label setting, the choices to be made are even more complex and particularly there are different semantics for different notions. Here, we briefly present an annotated reading list reviewing some of the proposed definitions and their relationships.
Von Neumann's minimax theorem asserts that the ability to defend against any opponent strategy implies the existence of an offensive strategy that guarantees the same value. This note revisits that symmetry from a constructive, oracle-based point of view. Given access to a defense oracle that, for any opponent strategy x , returns a response y guaranteeing payoff at least v , we ask how efficiently one can compute an offense strategy y * that guarantees value at least v – ε against all x. A classical construction via no-regret learning yields such a y * after O ((1/ε) 2 ) calls to the defense oracle. In this note, I describe a different construction that uses only O (log(1/ε)) calls to the oracle (up to polynomial factors in the dimension and encoding size). I then illustrate this primitive through three applications: computing Φ-equilibria in convex and extensive-form games beyond polynomial type, computing expected solutions to variational inequalities, and computing expected fixed points of possibly discontinuous maps.
Distributional constraints arise naturally in many matching markets, requiring the number of matches of specific types to satisfy predetermined bounds. This article reviews recent developments in the design and analysis of matching markets under such constraints. We discuss existing theoretical and empirical approaches. We then describe the results of [Ikegami et al. 2025], which develops a new framework for matching markets with distributional constraints and applies it to the Japan Residency Matching Program. The analysis illustrates how data can be used to evaluate regulatory instruments and to construct subsidy schemes that implement constrained-efficient outcomes.
This letter provides an overview of our recent work on "Learning Treatment Effects While Treating Those in Need" (published at the 2025 ACM Conference on Economics and Computation) as well as a more general perspective on design goals for algorithmic systems that are used to allocate limited resources in policy settings. Our motivation is the kind of algorithms that are used widely at present to prioritize candidates for various kinds of social interventions: public housing assistance, drop-out prevention programs in education, unconditional cash transfers in development, or a variety of other social services. By far the most common way of constructing such systems is the lens of predictive allocation : the algorithm designer identifies an outcome that the program seeks to alter (long-term home-lessness, dropping out of school, etc) and constructs a predictive model for that outcome [Vaithianathan and Kithulgoda 2020; Aiken et al. 2022; Pan et al. 2017; Toros and Flaming 2017]. Candidates are ranked by predictions of risk so that, e.g., limited spots in a housing program might be offered to those at greatest predicted risk of long-term homelessness.
It's an honor for us to serve as the SIGecom Executive Committee. As the incoming leadership team, we began our term with a deep appreciation for the strong and diverse community that has grown around the intersections of economics and computation. Our goals are to support the SIG's continued growth while emphasizing interdisciplinary engagement, recognition of impact, community involvement, and encouraging diversity across a range of dimensions.
The first invited talk of the session, by Haifeng Xu from the University of Chicago, highlighted a new research agenda: studying the wide range of problems in online content ecosystems through the formalisms of computational economics. Online content recommendation engines—core to platforms like YouTube, Instagram, and TikTok—serve personalized content to billions of users daily. The classic model considers both the users and the content library to be static, with the recommendation engine responsible for generating a mapping between the two. Xu's talk envisions a richer model that incorporates the incentives of content creators (e.g., YouTube rewarding videos based on length and views), the myopic and dynamic behavior of consumers, and the increasingly prominent role of AI in both generating content and being trained on it. This is a rich, dynamic multi-agent environment and the remainder of the talk considers two distinct directions within this framework: (1) Diagnosing and optimizing existing content ecosystems (2) How AI-generated content can transform future content ecosystems
This is the ninth annual collection of profiles of the junior faculty job market candidates of the SIGecom community. The forty candidates for 2025 are listed alphabetically and indexed by research areas that define the interests of the community. The candidates can be contacted individually, or collectively via the moderated mailing list ecom-candidates2025@acm.org.
Apportionment is the problem of allocating seats to political parties in a parliament in proportion to their deserved representation or to allocate the number of representatives to states in proportion to their size. Throughout history, most of the focus is on deterministic methods to apportion seats among the groups which may favour bigger or smaller parties or may have some inherent mathematical limitations. We survey various randomized rules that achieve exact apportionment in expectation.
In this letter, we summarize our recent work on the welfare impact of recommendation algorithms and propose questions for further study. We model recommendation algorithms as an information structure, which shapes how a third party takes actions that affect the welfare of different individuals in a population. Each recommendation algorithm thus induces a welfare profile, describing the expected payoffs of different individuals when the third party takes actions following the algorithm. Our framework allows us to characterize and compute the set of all such profiles, which we dub the Bayes welfare set. The Bayes welfare set allows us to reduce society's choice of an algorithm to the choice of a Bayes welfare profile. Our framework complements that of the algorithmic fairness literature which remains agnostic about the population's payoffs, focusing instead on statistical properties of algorithms, such as accuracy, parity, or fairness.
In this note, we survey automated mechanism design (AMD): the use of computational techniques to solve mechanism design problems. We describe three distinct but overlapping threads of research: an optimization-based paradigm that formulates mechanism design as linear programming, a line of work on sample complexity and learning theory, and the recent trend of differentiable economics, which has produced state-of-the-art results on a range of problems by borrowing tools and techniques from modern deep learning.
Liquid democracy is a democratic paradigm that introduces new challenges for researchers in fields around collective decision-making and, hence, it prompts a variety of compelling questions well-suited to the EC community. In this overview, we present a selection of papers that capture the breadth of research directions in this area.
Choice architecture is widely used to nudge consumers into sharing data in consent-based data exchanges. We present experiment evidence from Lin and Strulov-Shlain [2023] demonstrating that conventional choice architecture design could lead to biases in sample data. We illustrate how the tension between maximizing data volume and minimizing data bias depends on both supply and demand factors. We also highlight the need for organizations to consider both the volume and representativeness of sample data when optimizing their choice architecture for data collection. Categories and Subject Descriptors: [Applied computing]: Law, social and behavioral sciences-Economics; [Security and privacy]: Human and societal aspects of security and privacy-Economics of security and privacy
In our years as applied scientists and managers at Go ogle, Amazon, and Meta, we have seen both the strengths that EconCS researchers can leverage in industry, as well as common challenges that these researchers face. Most EconCS PhD programs do not emphasize exploratory data analysis, applied machine learning and statistics, or a coding mindset, even though these are valuable skills to have in industry. In this article we share how these skills are leveraged, and how you can invest in building these skills now. In doing so, we hope to make it easier for people from the EconCS community to be successful in industry, be it during an internship, a sabbatical, as a part-time consultant, or as a full time applied scientist!
This letter shows how Tullock contests-a class of all-pay auctions with proportional allocation rules-can be used to model and reason about several blockchain settings. We review the fundamentals of Tullock contests and their connections to potential games. We discuss why certain properties of Tullock contests, such as sybil-proofness and compatibility with "decentralization," have made them common in blockchain applications. We illustrate how Tullock contests naturally arise in proof-of-work and proof-of-stake blockchain protocols, and are an attractive design for emerging marketplaces for blockspace and succinct proofs.
Calibration is a classical notion from the forecasting literature which aims to address the question: how should predicted probabilities be interpreted? In a world where we only get to observe (discrete) outcomes, how should we evaluate a predictor that hypothesizes (continuous) probabilities over possible outcomes? The study of calibration has seen a surge of recent interest, given the ubiquity of probabilistic predictions in machine learning. This survey describes recent work on the foundational questions of how to define and measure calibration error, and what these measures mean for downstream decision makers who wish to use the predictions to make decisions. A unifying viewpoint that emerges is that of calibration as a form of indistinguishability, between the world hypothesized by the predictor and the real world (governed by nature or the Bayes optimal predictor). In this view, various calibration measures quantify the extent to which the two worlds can be told apart by certain classes of distinguishers or statistical measures.
The problem of delegated choice has been of long interest in economics and recently on computer science. We overview a list of papers on delegated choice problem, from classic works to recent papers with algorithmic perspectives.
