Flash Talks

Abstract: Pacing equilibria model how platforms scale bidders’ values or bids so that agents respect hard budget constraints while competing across many auction mechanisms. In contrast to second-price pacing, first-price pacing is known to enjoy strong structural and computational properties in the single-item setting. It remains unclear, however, which of these guarantees persist in more general mechanisms with richer allocation and payment environments, such as multi-unit auctions and other realistic marketplace formats.

We develop a unified theory of pacing equilibria across increasingly general classes of mechanisms. We first study one of the simplest generalizations of the single-item case, first-price position auctions, and show that already in this setting, pacing equilibria need not coincide with market equilibria, even when bidders’ utilities respect the position-auction structure. We then identify a broad structured class, bid-maximizing pay-your-bid mechanisms, which subsumes both single-item and position first-price auctions. For this class, we establish that pacing equilibria exist, are uniquely determined, and admit an Eisenberg-Gale-type convex program characterization, implying efficient computation, Pareto efficiency of the induced outcome, and liquid-welfare guarantees. Finally, we study fully abstract mechanisms satisfying a monotonicity condition on payments, and prove existence of pacing equilibria through a smoothing framework that regularizes discontinuities. With additional conditions and appropriate tie-breaking, we obtain uniqueness, revenue maximality among budget-feasible pacing vectors, and shill-proof implementability. We also provide convergent budget-adjustment dynamics for computing approximate equilibria in the smoothed mechanism, supporting practical implementation in large markets.

Overall, our results demonstrate that many favorable properties of first-price pacing are not artifacts of single-item auctions, but instead follow from mechanism-level conditions that apply beyond traditional auction models.

Abstract: When should content be short or long? I answer this question with a model in which engaging with a creator reveals information at two levels: about the current video and about the creator’s underlying quality. This structure gives content length an informational role—it controls the speed of learning at each level. Under bad-news learning, users seek to avoid low-quality creators. When experimentation is costly, short content is optimal. Under good-news learning, users search for a creator worth following. Short content helps users identify a high-quality creator faster, while long content extracts value once a good match is found. The two formats can then be complements, rationalizing short-to-long funnels. A further extension studies within-video information structure and shows that front-loading information encourages entry into experimentation but can generate excessive switching.

Abstract: Consider the dispatch of ridesharing trips to drivers lining up at airports, or the allocation of deceased donor organs to patients in transplant wait lists. In such dynamic queueing games, heterogeneity in short-lived items combined with agents' discretion to decline induce substantial cherry picking, resulting in high rates of unfulfilled trips and discarded organs. Existing research typically focuses on easy-to-analyze dispatch policies, sometimes making fluid assumptions for tractability. In this work, we introduce a solver for general dispatch policies with closed-form updates, which computes agents' equilibrium acceptance, entry, and reneging strategies as functions of queue length and queue position. We prove that a Markov perfect equilibrium exists for any dispatch policy that never assigns the same item to an agent more than once. Moreover, our solver converges in finitely many iterations under policies where items are offered monotonically down the queue and dispatch probabilities depend only on queue position, not queue length. We further show that our solver naturally extends to parallel dispatching, where items may be offered to multiple agents simultaneously. Through extensive simulations, we show that the solver recovers known equilibria for theoretically analyzable policies, offers insights into more complex policies better suited for practice, evaluates the robustness of different policies under model misspecification, and elucidates the competitive dynamics between platforms in the presence of driver multihoming and rider dual-apping.

Abstract: This paper studies the dynamics of competitive equilibria in exchange economies where agents trade sequentially through a network. Agents with quasilinear HARA preferences engage in local Walrasian exchanges within cliques of the network, and the resulting allocations become endowments for subsequent trades. I establish a separation result: the allocation of every non-numeraire good converges to the competitive equilibrium of the socially unconstrained economy, regardless of the network topology or the sequence of trades. In contrast, the allocation of the numeraire is path-dependent---different trading sequences redistribute wealth differently relative to the global simultaneous benchmark. Consequently, the equilibrium is Pareto efficient but not unique: the network structure and trading sequence jointly determine distributional outcomes. I characterise the speed of convergence through spectral gap analysis and show that multilateral trade within network cliques yields strictly faster convergence than decomposing those cliques into bilateral exchanges. For non-periodic trading sequences, I introduce a pi-weighted operator norm that provides monotone step-by-step convergence guarantees. Numerical examples illustrate how different network topologies and trading sequences generate specific welfare redistributions.

Abstract: We study the computational complexity of strategic behaviour in primary elections. Unlike direct voting systems, primaries introduce a multi-stage process in which voters first influence intra-party nominees before a general election determines the final winner. While previous work has evaluated primaries via welfare distortion, we instead examine their game-theoretic properties. We formalise a model of primaries under first-past-the-post with fixed tie-breaking and analyse voters' strategic behaviour. We show that determining whether a pure Nash equilibrium exists is Σ₂ᴾ-complete, computing a best response is NP-complete, and deciding the existence of subgame-perfect equilibria in sequential primaries is PSPACE-complete. These results reveal that primaries fundamentally increase the computational difficulty of strategic reasoning, situating them as a rich source of complexity-theoretic challenges within computational social choice.

Abstract:  This paper investigates the effect of asynchrony in agents' updates in the emergence of algorithmic collusion. We present a continuous-time model for algorithmic collusion in which two firms use Q-learning algorithms to set  prices asynchronously in a Bertrand duopoly. The firms update their prices at times dictated by a Poisson clock. By controlling the extent of agents' asynchrony, we run extensive numerical experiments with three specifications of the algorithm to investigate the emergence of algorithmic collusion. The strength of collusion is measured by a standard collusion index, as well as by  automatically detecting the punishment-reward schemes. This is done by recording a large number of algorithms' reactions to unilateral price cuts and comparing them with the reactions of untrained algorithms. Our findings indicate that asynchrony hampers collusion, especially when the algorithms are stateless and when they condition on an average of the opponent's price since its own last update. However, when algorithms condition their price on the current price of their opponent, collusion is much more robust to asynchrony. The implications of these results for the regulation of algorithmic pricing are discussed.

Abstract: Mean-based algorithms are a class of online learning algorithms that assign low probability to actions with low average rewards. Recent work indicates these algorithms converge favorably to serially undominated actions, which approximate Nash equilibria in economic games. However, empirical studies also show slower convergence compared to established algorithms in bandit-feedback scenarios.

We study mean-based algorithms when the time horizon is unknown and only bandit feedback is available. In this setting, we provide the first lower bound on the algorithm-defining sequence γt that formally establishes a limit on how fast these algorithms can learn. Additionally, we propose two mean-based algorithms: one generalizes ϵ-greedy, and the other extends the mean-based Exp3 to unknown horizons. Our experiments show that mean-based algorithms, although slightly slower, can perform competitively with other bandit-feedback algorithms.

We further analyze the relationship to no-regret algorithms. Depending on the choice of γt, the intersection with no-regret algorithms is non-trivial, and we show that algorithms exist that are both mean-based and no-regret. This adds context to the "exploitability" of this class of algorithms that previous contributions suggest.

Abstract: We study ex-post equilibria (EPEs) in games with parameter uncertainty, compare them with alternative notions of robust equilibrium in the literature, and propose a foundation for EPEs as a robust solution concept. EPEs are characterized by two properties: monotonicity and set-consistency. Because EPEs may not always exist, we introduce the concept of optimal approximate EPE, where players adopt approximate best responses that minimize the overall degree of suboptimality. We then address the problem of finding EPEs when they exist, and finding optimal approximate EPEs when they do not. Our analysis focuses on two notable classes of games: zero-sum games and potential games. For these settings, we establish several hardness results and propose a class of general computational approaches based on auxiliary minimax formulations.

Abstract: We investigate learning dynamics in competitive newsvendor games, a class of continuous action games with strategic substitutes. Despite established equilibrium properties, convergence of independent learning algorithms in repeated general-sum play remains uncertain. We analyze structural properties under complete and incomplete information, deriving closed-form equilibria for a symmetric complete-information benchmark with perfect substitution. Our main theoretical contribution proves strict monotonicity in both complete-information and Bayesian models with private costs, ensuring equilibrium uniqueness. This provides convergence guarantees for variational-inequality-based algorithms. Numerical experiments using deep reinforcement learning agents with Proximal Policy Optimization empirically demonstrate convergence to Nash and Bayesian Nash equilibria, verified by equilibrium checks. These results establish a foundation for applying deep reinforcement learning in competi tive inventory management.

Abstract: We introduce the prophet inequality with delayed and uncertain acceptance, a variant of the classical prophet inequality in which a decision-maker sequentially evaluates options whose acceptance is uncertain and whose outcome is revealed only after a fixed delay. That is, at each time step, the decision-maker observes the realized value of the arriving option and must irrevocably decide whether to attempt to select it or to continue searching. If an option is attempted to be selected, the process is suspended for a fixed delay d, during which no other options can be considered. Once the delay expires, the selection succeeds with a known probability. If successful, the decision-maker receives the realized value and the process terminates; otherwise, the search resumes.

In addition to the online decision-maker, we consider two stronger benchmarks: the value-aware decision-maker, who knows all value realizations in advance but not the acceptance outcomes, and the prophet, who knows both the values and the acceptance realizations. We characterize the competitive ratios between the two decision-makers and the prophet, showing that each is lower bounded by 1/(d+2), and we construct instances demonstrating that these bounds are tight for two of the comparisons.

In the extreme case of no delay (d=0), where our result recovers the classical 1/2-competitive guarantee, we establish the tightness of the remaining competitive ratio and identify sufficient conditions under which the value-aware decision-maker can beat the 1/2 barrier against the prophet. In particular, we show that this occurs whenever all acceptance probabilities are strictly positive, by reducing the problem to a classical prophet inequality instance over appropriately scaled Bernoulli random variables.

Abstract: This paper develops a framework for dynamic matching markets without transfers. We study overlapping-generations environments in which, at each date, a finite set of institutions on one side of the market is matched to a continuum of agents on the other side, and a stable mechanism is operated in every period. We first characterize the stationary outcomes of these processes using stationary market-clearing cutoffs. We show that the stationary aggregate demand naturally violates the strict gross substitutes condition. As a result, structural properties that hold in static matching environments do not generally extend to the dynamic setting. Second, we provide foundations for stationary market-clearing cutoffs. Treating the distribution of agents’ priorities and its evolution over time as design instruments, we show that any equilibrium outcome in a broad class of mechanisms can be implemented by the Deferred Acceptance (DA) mechanism. We also establish convergence results that clarify when the continuum model provides a good approximation to large but finite markets. Finally, we study dynamic incentive schemes—common in practice, for example in the assignment of civil servants. We show that priority advantages that allow agents to retain desirable positions over time make non-assortative allocations fragile. However, if one is willing to relax such seniority-type requirements, dynamic incentives can be used to achieve a broad range of distributional objectives under a constrained-efficiency requirement. 

Abstract: I study a robust model of information disclosure and pricing in a monopoly market with buyer learning. A seller offers an indivisible good to a buyer who is uncertain about his valuation and can acquire additional costly information from outside sources. The seller chooses a price and a signal, but does not know which additional signals are available to the buyer and maximizes a robust objective. I characterize the seller's optimal price and information structure when the cost of outside signals is monotone in the informativeness of the signal. I then provide comparative statics and show that greater uncertainty with regard to the informational environment of the buyer leads to more disclosure from the seller. Since this also changes the optimal price the seller charges, the effect on consumer surplus is ambiguous.

Abstract: We study the optimal regulatory policy for a bayesian regulator who faces a monopolist with private information in cost and demand. Contrary to conventional wisdom, we find no genericity of exclusion: if demand is sufficiently high, the regulator allows all firm types to operate. We show that the optimal regulated price follows a straightforward generalization of Baron and Myerson (1982)'s adjusted marginal cost formula. Two novel and robust phenomena appear in this multidimensional environment. First, the regulator typically caps the firm's price at the maximum possible marginal cost inducing a ``bunching at the bottom" which arises independently of the distribution. Second, for some firms it'll be optimal to set the regulated price below marginal cost. Both of this phenomena arise from the fact that incentive compatibility in this bidimensional environment induces a strong conflict between social and private incentives for pricing. We perform comparative statics showing how the shape of the optimal regulatory policy responds to changes in the regulator's preference over efficiency and redistribution as well as changes on the relative size of uncertainty in cost and demand. We show that the bidimensional solution converges to Baron and Myerson (1982)'s solution as uncertainty in demand vanishes and to Lewis and Sappington (1988)'s solution as uncertainty in cost vanishes. We also make comparisons between the bayesian and the robust approaches to monopoly regulation.

Abstract: Trust management in peer‑to‑peer networks requires turning local, possibly asymmetric assessments between peers into global scores, which is the problem addressed by eigenvector‑based methods such as EigenTrust. We give a cooperative‑game analogue that reaches beyond trust management, valuing each peer by its marginal contribution to coalitional trust. On a weighted directed graph, the edges incident to a coalition split into internal and boundary (incoming/outgoing) edges, and an aggregation operator turns the boundary component into a transferable‑utility game. The framework is expressive: for some aggregators, computing the Shapley value is \#P‑hard. For the pessimistic‑minimum aggregator, however, we obtain a closed‑form, polynomial‑time Shapley value, because each per‑player boundary game admits a Möbius decomposition supported on a single chain of unanimity games. We generalize this through a gatekeeper property, where the coalitional value depends only on the first absent neighbor in a fixed ordering, a condition on the operator that suffices for a polynomial‑time Shapley formula across a broad, minimum‑like class.

Abstract: We study first-order learning dynamics in Bayesian signaling games, a minimal extensive-form model of strategic interaction under asymmetric information. In these games, a Sender observes a private type and chooses a signal, after which a Receiver forms a posterior belief and selects an action. This Bayesian belief update makes the induced learning field fundamentally different from the affine game-gradient fields of normal-form games. Empirically, we find a sharp dichotomy: standard online learning algorithms reliably converge to strict Perfect Bayesian Equilibria (PE), but not to non-strict ones. We show that this behavior cannot be explained by the global geometric conditions commonly used to prove last-iterate convergence, such as monotonicity or the Minty variational inequality. Indeed, these conditions can fail even in binary signaling games. Instead, we identify the relevant local geometry. Our main result characterizes variational stability in finite signaling games: a PE is variationally stable if and only if it is strict. This yields local last-iterate convergence guarantees for online mirror descent and regularized dual averaging near every strict PE. We then study global average-iterate convergence and identify a class of signaling games in which  coarse-trigger regret-minimizing dynamics converge to the unique PE. Finally, we complement the theory with experiments comparing projected gradient ascent, mirror-based methods, optimistic variants, and PPO across canonical signaling-game families. The results position signaling games as a compact testbed for understanding how Bayesian belief formation reshapes the convergence geometry of multi-agent learning.

Abstract: This paper investigates the existence of international environmental agreements (IEAs) that are both fair and stable in the sense that they belong to the γ-core. In a cooperative game setting with multilateral environmental externalities countries form agreements which allocate emissions reduction targets. I combine a fairness principle introduced by Athanasoglou (2022) with the γ-core and use the Bondareva-Shapley Balancedness Theorem to prove non-emptiness of the core. This paper contributes to the literature in two ways. First, by introducing a notion of fairness, it extends the models of Helm (2001) and Stamatopoulos (2020) which analyse core allocations in an economy with multilateral environmental externalities. Second, by allowing for deviations of coalitions of any size, it extends the result of Athanasoglou (2022), who shows that a fair IEA can be robust against a single country deviating. Results show that there exists a global agreement that is both fair and robust to deviations by coalitions of any size. This is a working paper.

Abstract: Imperfectly competitive electricity markets are susceptible to strategic bidding behaviors by dominant market participants, which can induce substantial systemic inefficiencies. Extant literature analyzes these market dynamics utilizing game-theoretic frameworks formulated as equilibrium problems with equilibrium constraints (EPECs). Within this literature the inefficiency is widely measured through the so called price of anarchy, a metric given by the social cost at the worst case nash equilibrium divided by the minimal system cost that can occur. 

However it is questionable if such a nash equilibrium is ever reached in realitywhen players use possibly imperfect policies to determine their bid. After all, it is a standard assumption for EPEC models to assume each player to have full knowledge about the state of the market and other players action. In this work we are therefore trying to search for the worst case outcome under more realistic condition. For that purpose instead of formulating an electricity market as an EPEC we are doing two things.

For our first approach we are training policies of different classes offline. Given such policies we then maximize the inefficiency over different exogenous state variables assuming that players strictly follow their learned policy. Our current results hint that with such a model using fairly simple policies extremely high inefficiencies can occur. Second, we are trying to instead of incorporating equilibrium constraints to include the learning dynamics of a no-regret algorithm. With such an approach one could measure the inefficiency not only in an equilibrium but over the whole, realistic trajectory of each players learning.

Abstract: An auctioneer is running a private auction on behalf of a principal, and has to publicly reveal the outcome. The auctioneer seeks to extract a rent from the auction by engaging some coalition of bidders in corruption. Given an auction, an auctioneer's plan is k-corrupt if for all type profiles, the auctioneer can find a coalition of size k such that his plan i) always weakly improves the sum of the utilities of the coalition members compared to the non-cooperative outcome, ii) that inequality holds strictly for a positive measure of the type space and iii) admits an innocent explanation for bidders that do not belong to the coalition. The corruption-robustness index of an auction is the minimum k such that a k-corrupt plan is feasible. The second-price sealed-bid auction with optimal reserve and its weak dominance equilibrium is maximally corruption-robust within the class of static, symmetric optimal auctions with an index of two. The Dutch auction (with optimal reserve) has an index equal to the number of bidders.