Persuasion with Ambiguous Communication

summary

The gist

The paper explores whether ambiguous communication can be beneficial for a sender in a persuasion problem, specifically when both the receiver and possibly the sender are ambiguity averse.

In short

The paper investigates if ambiguous communication helps a sender persuade a receiver, especially when both are risk-averse regarding uncertainty. It shows that ambiguity can be beneficial if an optimal experiment can be split into two Pareto-ranked versions. The economic gain comes from using the better ambiguous experiment more often, balanced against the cost of the sender's own aversion to ambiguity.

Key concepts

Pareto-ranked experiments
These are different experimental setups that are compared based on whether one is strictly better than another across all outcomes. The paper uses these rankings to determine if an optimal experiment can be split into two versions, which is a key condition for ambiguous communication to provide a benefit to the sender.
Ambiguity aversion
This refers to a player's dislike or aversion to uncertainty or ambiguity in the information they receive. The analysis shows that this aversion affects how much gain is realized; for example, an ambiguity-neutral sender facing an ambiguous receiver benefits if a certain experimental split exists.
Bayesian persuasion experiment
This is the optimal experiment designed by the sender based on what they know and what they believe the receiver will do. The paper examines how this optimal setup can be broken down into simpler, Pareto-ranked experiments to analyze the impact of ambiguity.

Terminology used across episodes

This episode discusses

The paper

Persuasion with Ambiguous Communication · Read on arXiv

Florida State University · Kellogg School of Management at Northwestern University · Queen Mary University of London

Transcript

Introduction to the show: ident: Economics. The day in economics: theory, mechanisms and the evidence beneath them.

Ruth: I'm Ruth.

Warren: And I'm Warren, here with Iris.

Ruth: We're looking at the paper: "Persuasion with Ambiguous Communication".

Warren: Here's the gist of the paper. The paper explores whether ambiguous communication can be beneficial for a sender in a persuasion problem, specifically when both the receiver and possibly the sender are ambiguity averse.

Ruth: The analysis hinges on demonstrating that an optimal Bayesian persuasion experiment can be split into two Pareto-ranked experiments, which is shown to be necessary for ambiguous communication to benefit the sender (Theorem 2).

Warren: The source of economic gain from this ambiguous communication, for both parties, is quantified by the greater use (mu sigma - lambda) of the Pareto-better experiment sigma, which must then be offset by a probability premium representing the cost due to a player's own ambiguity aversion regarding exposure to ambiguity.

Ruth: Theorem 3 establishes that ambiguous communication benefits the sender only if phi r is not affine and there exists a Pareto-ranked splitting (sigma, sigma, lambda) of an obedient experiment sigma hat such that u s(sigma, tau*) > u BP s.

Paper summary: Ruth: We just touched on the core idea of "Persuasion with Ambiguous Communication." The authors are exploring how adding uncertainty to the communication strategy changes things for both players in a persuasion problem >

Warren: Specifically, they look at when that benefit actually materializes. They argue that if a specific mathematical structure exists—a way to split an obedient experiment into two parts where both players agree on which is better—then this ambiguity can help anyone who isn't completely averse to uncertainty >

Iris: That splitting is crucial because it shows the necessity of using a collection of experiments that form a splitting of an obedient experiment, and some of those experiments must be Pareto-ranked for both players, as mentioned in the paper >

Ruth: Right. The source of any economic gain from this ambiguous communication comes from using a Pareto better experiment, which they call sigma, more often than a less-better one, lambda >

Warren: And that gain has to be balanced because the sender has to pay a premium for being exposed to this ambiguity themselves—that cost comes from their own aversion to uncertainty >

Iris: The analysis shows that ambiguous communication only benefits the sender if some specific conditions are met, like when the receiver's payoff function is not affine and there exists that Pareto-ranked splitting of an obedient experiment sigma, tau star, lambda >

Ruth: And the paper also points out a limitation: they show that if you only have two possible actions for your receiver, then ambiguous communication can’t actually help you at all >

Warren: So we see these structural requirements early on before diving into the actual results of this work. That means the benefit isn't automatic; it has to fit a certain shape for the game structure.

Conclusion: Ruth: So, to wrap up "Persuasion with Ambiguous Communication," Cheng, Klibanoff, Mukerji, Renou, Guo, Jakobsen, Li and Pavan are showing that adding uncertainty to communication is not always a waste of time >

Warren: It seems they suggest that for a sender who isn't totally neutral about uncertainty—meaning they have some level of ambiguity aversion—ambiguous communication can actually be an advantage over the standard Bayesian approach in certain setups >

Iris: And the implications are pretty concrete for how we model these interactions. If you’re looking at a scenario where the receiver has only two actions, the paper tells us straight up that ambiguous communication doesn't help because you can't exploit that structure to get a benefit >

Ruth: That’s right. But if you look at threshold environments with safe and risky actions, the paper shows that an ambiguity-neutral sender can still find a benefit by using this ambiguous communication, provided those experiments line up correctly >

Warren: And they also found some robustness in the findings. The same ambiguous experiment works for any sender who isn't overly averse to ambiguity, even if they underestimate how much the receiver is worried about uncertainty >

Iris: So, to summarize what we’ve heard from "Persuasion with Ambiguous Communication," it hinges on maximizing that use of the Pareto better experiment, mu sigma minus lambda > while making sure you account for the probability premium your own aversion to ambiguity imposes on you >

Ruth: And they contrast this approach with other models like BLL, showing where their specific mechanism—the Pareto-ranking of experiments within a sender's set sigma—offers a different kind of gain that other approaches might miss >

More episodes

← Home