📈 economics

Strong Dominance for Dynamic Signals

This paper demonstrates that Gentzkow and Kamenica's (2017) signal representation framework can be effectively extended to dynamic decision problems to characterize when one information structure robustly dominates another, a condition equivalent to a dynamic version of the "reveal-or-refine" criterion.

Original authors: Mark Whitmeyer, Cole Williams

Published 2026-05-19

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Mark Whitmeyer, Cole Williams

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are an investor, a startup founder, or even a deer trying to decide where to forage. Every day, you have to make a choice. Sometimes you know exactly what's going on, but often, you are guessing. To make better guesses, you rely on information: news reports, phone calls, weather forecasts, or tips from friends.

This paper asks a simple but tricky question: How do we decide which source of information is truly "better" than another, especially when you already have other sources of information mixed in?

The authors, Mark Whitmeyer and Cole Williams, propose a new way to rank information that is "robust." This means it holds true no matter what other news you are reading or who you are talking to.

Here is the breakdown of their findings using simple analogies.

1. The "Strong Dominance" Rule

In the past, economists had a way to compare two information sources, but it only worked if that source was the only thing the person knew. The authors wanted to know: "If I have a subscription to the Financial Times AND the Wall Street Journal, plus random texts from my mom, can I still say one newspaper is objectively better than the other?"

They found a condition called Strong Dominance.

The Metaphor: Imagine two detectives, Detective A and Detective B. Detective A "strongly dominates" Detective B if, for every single clue B finds, Detective A either:
1. Finds the exact same clue (or a more detailed version of it); OR
2. Already knows the identity of the criminal (the "state of the world").

If Detective A always knows at least as much as Detective B (or knows the answer outright), then Detective A is always the better choice, regardless of what other clues the investigator has from other sources.

2. The "Reveal-or-Refine" Condition

The paper proves that this "Strong Dominance" is exactly the same as a concept they call "Reveal-or-Refine."

Refine: Think of a map. If Detective B has a map showing a whole city, and Detective A has a map showing just one specific street within that city, Detective A has "refined" the information. They know more specific details.
Reveal: If Detective A looks at a clue and immediately knows, "The criminal is in the bank," they have "revealed" the state.

The authors' main result is that for information to be "strongly dominant," it must reveal-or-refine the other information every single day (or time period) of the decision process. It's not enough to be better on average; you have to be better (or know the answer) at every specific moment in time.

3. Why This is Surprising (The "More, Sooner" Trap)

The paper highlights a famous counter-intuitive problem from previous research. Usually, we think getting more information earlier is always better. But in complex, multi-step decisions, getting a very detailed report first can sometimes make you worse off than getting a vague report first.

The Analogy: Imagine you are playing a game where you can wait for a second clue to make a final move.
- Scenario A: You get a huge, detailed clue first. It confuses you slightly, making you hesitate to wait for the second clue because you think you have enough info. You end up paying a "waiting fee" unnecessarily.
- Scenario B: You get a vague clue first. It tells you to wait. You wait, get the second clue, and make the perfect move.

In this specific "non-robust" scenario, the "less informative" clue was actually more valuable because of when it arrived.

However, the authors show that their new "Strong Dominance" rule avoids this trap. If one information source "Reveal-or-Refines" the other, it is always better, no matter the timing or the specific game you are playing. The "pathology" where less info is better disappears when you use this strict ranking.

4. The "Signal" Tool

To prove this, the authors used a mathematical tool called a "signal" (borrowed from previous work by Gentzkow and Kamenica).

The Metaphor: Think of information not as a list of facts, but as a set of partitions or slices of a pie.
- A "coarse" signal slices the pie into big chunks (e.g., "It's either raining or not").
- A "fine" signal slices the pie into tiny pieces (e.g., "It's raining in the north, but sunny in the south").
The paper shows that if you can look at the "slices" of the pie for Information Source A and see that every slice fits perfectly inside a slice of Information Source B (or reveals the whole pie), then Source A is the winner.

Summary

The paper argues that to compare dynamic information (information that comes over time), we shouldn't just look at the final outcome. Instead, we should check if one source always gives you everything the other source gives you, plus maybe more, or tells you the answer outright.

If this "Reveal-or-Refine" condition holds true for every single day of the process, then that information source is universally superior. It doesn't matter what other news you have, or what specific decisions you are making; having that source is always the best move.

Technical Summary: Strong Dominance for Dynamic Signals

Problem Statement
The paper addresses the challenge of comparing dynamic information structures in decision problems where an agent faces a sequence of actions over time (finite or infinite horizon) under uncertainty about the state of the world. While classical literature (e.g., Greenshtein, 1996) characterizes how to rank dynamic information sources when they are the agent's sole source of information, it does not provide a robust framework for ranking them in the presence of additional auxiliary information sources. The authors ask: Can we determine if one dynamic information structure is inherently more valuable than another, regardless of what other information the agent possesses?

Methodology
The authors extend the "signal" representation of information introduced by Gentzkow and Kamenica (2017) and Brooks, Frankel, and Kamenica (2022a) (BFK) from static to dynamic settings.

Model: The environment consists of a finite set of states $\Theta$ , a prior $\mu$ , and a discrete time horizon $T$ . A dynamic signal is defined as a sequence of signals $\eta = (\eta_t)_{t=1}^T$ where each subsequent signal refines the previous one.
Robust Dominance: The authors define "strong dominance" for dynamic signals. A dynamic signal $\eta$ strongly dominates $\hat{\eta}$ if, for any extended dynamic decision problem (including any auxiliary information sources $\rho$ ), the agent achieves a weakly higher expected utility using $\eta$ than using $\hat{\eta}$ .
Signal Representation: Utilizing the partition-based signal representation (subsets of $\Theta \times [0,1]$ ), the authors define a dynamic "reveal-or-refine" condition. A signal $\alpha$ reveal-or-refines $\hat{\alpha}$ if every element of $\alpha$ either reveals the state (is a singleton in $\Theta$ ) or is a subset of some element in $\hat{\alpha}$ .

Key Contributions and Results
The paper's central result, Theorem 2.5, establishes a necessary and sufficient condition for strong sequential dominance:

A dynamic signal $\eta$ strongly dominates a dynamic signal $\hat{\eta}$ if and only if $\eta$ dynamically reveal-or-refines $\hat{\eta}$ .

This means $\eta$ dominates $\hat{\eta}$ if and only if, at every period $t$ , the signal $\eta_t$ reveal-or-refines $\hat{\eta}_t$ .

Proof Logic:
- Sufficiency: If the condition holds, the agent with the dominant signal can mimic the strategy of the agent with the dominated signal at every history (since they know what the other saw, or know the state), thereby achieving at least the same payoff.
- Necessity: If the condition fails at any period $t$ , the authors construct a decision problem where only the action at period $t$ matters. By invoking the static result from BFK, they show that the dominated signal would yield a strictly lower value in this specific problem, violating strong dominance.
Additively Separable Problems (Corollary 2.8): The authors show that this equivalence holds even when restricting attention to the class of additively separable decision problems (where utility is the sum of period utilities). This contrasts with non-robust sequential dominance (Greenshtein, 1996), where such equivalence does not hold.
Contrast with Non-Robust Dominance (Section 2.2 & 2.3): The paper highlights a critical distinction between "strong dominance" (robust to auxiliary info) and standard "sequential dominance" (Blackwell dominance in dynamic settings).
- Standard sequential dominance does not require period-by-period reveal-or-refine.
- The authors present a counterintuitive example (adapted from Greenshtein, 1996) where a decision-maker prefers to observe a less informative experiment first, followed by a more informative one, due to the option value of deferring action.
- Under strong dominance, this "pathology" disappears. If $\alpha$ is more informative than $\beta$ in the Blackwell sense but does not reveal-or-refine $\beta$ (e.g., due to conditional independence), $\alpha$ does not strongly dominate $\beta$ . Strong dominance requires a history-by-history relation that standard Blackwell dominance lacks.
Interim vs. Ex Ante (Proposition 2.12): The paper establishes that strong dominance is equivalent to an interim notion of dominance. If $\eta$ strongly dominates $\hat{\eta}$ , then at any point in time, conditional on any realization of the signal and auxiliary information, the continuation value of $\eta$ is at least as high as that of $\hat{\eta}$ .

Significance and Claims
The authors claim their work provides a "robust notion of sequential dominance" that is:

Intrinsically Valuable: It addresses the realistic scenario where agents possess multiple information sources, allowing for a ranking of specific sources (e.g., newspapers) independent of other data.
Intuitive and Simple: The equivalent condition (dynamic reveal-or-refine) aggregates simple static conditions, making the ranking of dynamic structures "easy to comprehend" compared to the "impenetrable" conditions required for non-robust sequential dominance.
Modest in Novelty: The authors explicitly state that the "crucial novelty" is the observation that the signal apparatus of Gentzkow and Kamenica (2017) and BFK is "extremely useful in dynamic decision problems." They characterize their main result as "falling out essentially without effort" and being "almost an instantiation of Theorem 1 in BFK," rather than a complex new theoretical construction.

The paper concludes that strong sequential dominance is simply the aggregation of static strong dominance conditions across time, preserving the intuitive "reveal-or-refine" logic of the static world within a dynamic framework.