A Comparison of Cursed Sequential Equilibrium and Sequential Cursed Equilibrium: Different Concepts of Cursedness in Dynamic Games
summary
The gist
Cursed Equilibrium of Eyster and Rabin (2005) has been a leading theory for explaining winner’s-curse-type behavior in static Bayesian games…
In short
This paper compares Cursed Sequential Equilibrium (CSE) and Sequential Cursed Equilibrium (SCE) to explain winner's-curse behavior in dynamic games. CSE captures neglect of how opponents' actions depend on private types, while SCE models a failure to recognize how unobserved actions depend on the game's history.
Key concepts
- Cursed Sequential Equilibrium (CSE)
- CSE describes a player neglecting how opponents' actions rely on their own private types. Players update beliefs using Bayes' rule but hold incorrect perceptions about other players' strategies, leading to dampened posterior beliefs.
- Sequential Cursed Equilibrium (SCE)
- SCE models a failure to recognize how unobserved actions depend on the game's history of play. Players always form correct Bayesian posteriors about types because their cursed-plausible conjecture matches the true Bayesian inference.
- Cursedness Distinction
- The paper identifies six key differences between CSE and SCE. The core distinction is that CSE reflects ignoring jointness with private information, whereas SCE reflects failing to infer dependence on historical actions.
Terminology used across episodes
This episode discusses
- A Comparison of Cursed Sequential Equilibrium and Sequential Cursed Equilibrium: Different Concepts of Cursedness in Dynamic Games · Paper Radio
- Cursed Sequential Equilibrium
- Cursed Job Market Signaling
The paper
A Comparison of Cursed Sequential Equilibrium and Sequential Cursed Equilibrium: Different Concepts of Cursedness in Dynamic Games · Read on arXiv
Meng-Jhang Fong, Po-Hsuan Lin, Thomas R. Palfrey
Max Planck Institute for Behavioral Economics · University of Virginia · California Institute of Technology
Cursed Equilibrium of Eyster and Rabin (2005) has been a leading theory for explaining winner's-curse-type behavior in static Bayesian games, but it faces conceptual limitations when applied to dynamic games. Two recent extensions, Cursed Sequential Equilibrium (CSE) by Fong, Lin and Palfrey (2025) and Sequential Cursed Equilibrium (SCE) by Cohen and Li (2026), address these limitations in fundamentally different ways. Complementing these two papers, this paper provides a systematic comparison of CSE and SCE, clarifying their conceptual foundations and technical implications, including their notions of cursedness, belief updating, treatment of public histories, and relationship to cursed equilibrium in one-stage Bayesian games.
Transcript
Introduction to the show: ident: Economics. The day in economics: theory, mechanisms and the evidence beneath them.
Ruth: We're looking at the paper: "A Comparison of Cursed Sequential Equilibrium and Sequential Cursed Equilibrium".
Warren: Here's the gist of the paper. Cursed Equilibrium of Eyster and Rabin (2005) has been a leading theory for explaining winner’s-curse-type behavior in static Bayesian games, but it faces conceptual limitations when applied to dynamic games.
Ruth: This paper provides a systematic comparison of Cursed Sequential Equilibrium (CSE) and Sequential Cursed Equilibrium (SCE), clarifying their conceptual foundations and technical implications, including their notions of cursedness, belief updating, and treatment of public histories.
Warren: Conceptual Foundations The paper introduces two extensions to the Cursed Equilibrium concept for dynamic games: Cursed Sequential Equilibrium (CSE) and Sequential Cursed Equilibrium (SCE), which address limitations in applying CE to dynamic settings.
Ruth: CSE extends the notion of cursed behavior to multistage games with publicly observed actions by assuming that, after each stage of the game, each player in the game updates their current beliefs about the profile of other players’ types in a fully-cursed way, i.e., based on the average behavioral strategy of the other players rather than the type conditional behavioral strategy of the other players.
Title and authors: Ruth: So we're looking at this paper titled "A Comparison of Cursed Sequential Equilibrium and Sequential Cursed Equilibrium: Different Concepts of Cursedness in Dynamic Games". It’s about taking the idea from Eyster and Rabin's static winner’s curse behavior and trying to apply it to games where things happen over time.
Warren: That makes sense, Ruth. The core issue with the original cursed equilibrium is that it only works for one-shot games with incomplete information, but dynamic games are way more realistic for things like signaling or repeated interactions.
Iris: Exactly. What this paper does is systematically compare two new concepts, Cursed Sequential Equilibrium and Sequential Cursed Equilibrium, to see how they handle those dynamic settings differently. It’s essentially mapping out the conceptual differences between them.
Ruth: Right. The authors are setting up a framework to clarify exactly where the old idea stops working and where these new sequential versions take over when we have multiple stages of play and public actions involved.
Warren: And they’re focusing on how they define that "cursedness" itself, because that's the part that seems to be changing between the two concepts. It’s not just a minor tweak; it’s a fundamental difference in what kind of strategic failure you are modeling here.
Iris: I think the most important thing is how they treat public histories and how belief updating actually evolves when you move from one concept to the other. That's where the technical meat of the comparison is happening.
The paper's summary: Ruth: So, what this paper summarizes is that there are fundamentally different ways Cursed Sequential Equilibrium and Sequential Cursed Equilibrium capture behavior in these dynamic games, particularly concerning belief updating and how they look at past actions.
Warren: They lay out six distinct differences between the two concepts when you look at multistage games with observed actions. It’s not just one difference; it’s a whole set of qualitative distinctions that show how the underlying assumptions about player types and their strategic reasoning diverge.
Iris: I think what they highlight is that under Cursed Sequential Equilibrium, cursedness comes from neglecting how opponents' actions depend on their private types, whereas Sequential Cursed Equilibrium frames cursedness as a failure to see how future or simultaneous actions rely on the history of play.
Ruth: That distinction is huge for me because it tells us whether we are modeling someone who is ignoring the inherent correlation between what they are and what others do, or someone who is misinterpreting how past moves constrain future possibilities.
Warren: And regarding belief updating, the paper shows that under CSE, players update their beliefs using Bayes rule but with a dampened posterior belief that’s essentially a weighted average of the previous stage’s belief and the standard Bayesian update.
Iris: But under SCE, they maintain correct Bayesian inferences about types from observed past actions because their cursed conjecture happens to line up perfectly with what the standard Bayesian inference predicts for those observed moves.
Ruth: That means in CSE, even if you have perfect information about past moves, your belief about what others are doing can still be skewed by this averaging process. It’s a subtle but important distinction for modeling real-world decision making under uncertainty.
Warren: And they also touch on the treatment of public histories and how re-labeling actions affects things, which is another area where the two concepts diverge significantly in their equilibrium predictions.
The paper's improvements: Ruth: Now, looking at what this paper suggests as improvements, it seems to be about making the comparison more useful by showing how these concepts can actually be tested in practice.
Warren: They suggest that these distinctions can be used to design experiments testing whether an AI system is exhibiting CSE or SCE behavior in a dynamic environment. It moves the discussion from just theoretical definitions into something observable.
Iris: I think this is where the paper gets really practical, because it’s not just about math on a page; it’s about creating a framework for empirical investigation. You can design tests based on those six qualitative differences they identify.
Ruth: And the authors are also pointing out that AI systems could be trained to identify the degree of cursedness, parameterized by some values like chi or phi, allowing researchers to pinpoint exactly when standard sequential equilibrium breaks down and when these more complex cursed behaviors actually emerge.
Warren: That’s valuable because it gives us a way to quantify the level of deviation from the expected behavior. Instead of just saying "it's cursed," you can measure how 'cursed' it is using those parameters.
Iris: And they also examine how structural changes, like re-labeling actions or imposing public history consistency, affect the equilibrium predictions of an AI system in these models. It shows that the structure of the game itself matters a lot to where you end up.
Conclusion: Ruth: So to wrap up, this paper is systematically comparing Cursed Sequential Equilibrium and Sequential Cursed Equilibrium across multistage games, clearly distinguishing between how they define cursedness, belief updating dynamics, and the treatment of public histories.
Warren: The main implication is that CSE captures a neglect of the jointness between other players’ actions and private information, while SCE models a failure to recognize how opponents’ unobserved actions depend on the history of play.
Iris: I think this comparison helps us understand that when we talk about strategic failures in dynamic settings, there are two distinct mechanisms at play that we need to model separately.
Ruth: And they provide tools for testing—by parameterizing the degree of cursedness and seeing how structural changes shift the predictions—giving researchers a clearer way to approach these complex games.
Warren: They’ve given us a clearer map of where these concepts live in game theory, which is really useful for anyone trying to apply these ideas to things like signaling games or crisis bargaining scenarios.
Iris: I agree, it gives us the necessary tools to move from just describing an anomaly to actually identifying the mechanism causing it.
Ruth: That’s what we have today with this comparison of Cursed Sequential Equilibrium and Sequential Cursed Equilibrium. It’s a deep dive into how different strategic failures manifest in games that evolve over time.
Warren: We'll be back after the break with more papers on the arXiv.
More episodes
- Daily Summary for 2026-11-19
- 2608.21274-Recommendation Quality and the Concentration of Consumption: Experimental Evidence from Netflix
- 2408.02757-A nonparametric test for diurnal variation in spot correlation processes
- 2306.16393-High-Dimensional Canonical Correlation Analysis
- 2605.17299-Geometric Brownian motion with intermittent entries and exits
- 2607.04753-Fooling Yourself: how narratives shape beliefs
- 2408.14671-Inference on Treatment Effects with Repeatedly Mismeasured High-Dimensional Controls
- 2201.06898-Difference-in-Differences Estimators for Treatments Continuously Distributed at Every Period
- 2501.15307-Influence Functions: A Unified Differentiation-Based Framework
- The day in conversation for 2026-11-19