Strategic Cheap Talk in Negotiations

When Can Words Be Trusted?

At first glance, the idea of credible cheap talk seems paradoxical. If a potential Bull can say "I'm looking for an emotional connection" just as easily as "I want a one-night stand," with no cost difference, why should a couple believe him? Game theory's cynical answer is often that they shouldn't. An equilibrium where all statements are ignored—a "babbling equilibrium"—is always a theoretical possibility.

However, as the paper "Cheap Talk" by Joseph Farrell and Matthew Rabin argues, this ignores a fundamental aspect of communication: we use language with the assumption of shared meaning, and we assess credibility based on incentives.

The Sally/Rayco example demonstrates the fundamental condition for credible cheap talk: alignment of interests. Let's map this to a relationship scenario.

Scenario 1: Perfect Alignment

A Hotwife is discussing her desires for an upcoming encounter with her husband. The husband's primary goal is to facilitate an experience that brings his wife maximum pleasure and fulfillment (this includes his own compersion). He is the "Receiver" (Rayco), and she is the "Sender" (Sally).

• Hotwife's 'Type': Her true desire is for a more "Romantic" evening.
• Her Message: "I'm feeling like I want a more romantic, sensual connection this time."
• Husband's Action: He can arrange for a charming, suave Bull (Demanding Job) or a more direct, physical Bull (Undemanding Job).

In this case, the Hotwife has every incentive to tell the truth. If she lies and says she wants a direct, physical encounter when she doesn't, her husband will arrange for the "wrong" Bull, leading to a suboptimal experience for her. Since his utility is tied to hers, it's a worse outcome for him too. Because a lie would make them both worse off, her statement is credible. Farrell and Rabin call this self-signaling: she wants her husband to believe she is the "Romantic" type if and only if she truly is.

When Words Fail: Conflict of Interest

Now, let's consider the opposite case. What happens when the sender's and receiver's interests diverge?

This scenario is common in vetting new partners. Imagine the couple is speaking with a potential Bull. The Bull knows the couple strongly prefers partners who are interested in a recurring, emotionally intimate connection (Demanding Job) over those just seeking a one-time hookup (Undemanding Job). However, the Bull, regardless of his true intentions, gets a higher payoff from simply securing the encounter (he wants the Demanding Job, i.e., to be chosen, no matter what).

• Bull's 'Type': He is only interested in a one-night stand (Low Ability).
• Bull's Message: "I'm really looking to build a genuine, ongoing connection."
• Couple's Action: Choose to "hire" him or not.

Here, the Bull's message is not credible. The couple knows that both a "long-term" type and a "one-night" type have an incentive to say they are long-term. Since the message does not help distinguish between types, it is disregarded. It's just babble.

Partial Credibility: The Crawford-Sobel Model

Most real-life situations fall between the extremes of perfect alignment and total conflict. Interests are often partially aligned, with some bias. This is the focus of the classic Crawford-Sobel (1982) model.

The model imagines a sender (S) who knows the true state of the world, $t$ (a value on a continuum, say [0, 1]), and a receiver (R) who will take an action, $a$.

• The Receiver wants to choose an action that matches the state, $a=t$.
• The Sender wants the Receiver to choose a slightly different action, $a=t+b$, where $b$ is the sender's bias.

Scenario 2: A Biased Sender

A Hotwife (Sender) is communicating her level of sexual readiness or desire ($t$) to her husband (Receiver). Her desire is on a scale from 0 (not at all) to 1 (extremely). The husband wants to respond appropriately—initiating if she's ready, giving space if she's not. However, the Hotwife has a small bias ($b>0$). She consistently prefers her husband to believe she is slightly more ready than she actually is, perhaps to encourage more frequent initiation or attention.

She can't credibly reveal her exact state $t=0.6$, because her husband knows she has a bias to exaggerate. If he were to believe her precise statements, she would always claim her state is $t+b$. Knowing this, he can't trust the number.

What can she do? She can send a less precise, but still informative, message. Instead of saying "My desire is 0.6," she might say "I'm in the mood." This message might credibly communicate that her desire is in the upper half of the range, say $t\in [0.5,1]$, but not exactly where.

The Crawford-Sobel model shows that communication in this setting results in a partition equilibrium. The sender divides the continuum of states into a series of intervals and sends a message that only reveals which interval the true state lies in. The size of these intervals, and thus the amount of information conveyed, depends inversely on the size of the bias $b$. A larger conflict of interest leads to coarser, less informative partitions.

As you can see from the plots:

• When $N=1$, no information is transmitted. The sender sends the same message regardless of the state. This is a babbling equilibrium.
• When $N=2$, the sender sends one of two messages, effectively telling the receiver whether the state is "low" or "high." This is more informative than babbling.
• When $N=3$, the partitions become even finer ("low," "medium," "high"), and more information is transmitted.

Crucially, the plots of sender's and receiver's utility show that both players are better off as more information is transmitted (as N increases). This provides a powerful incentive for players to find ways to communicate credibly, even when their interests are not perfectly aligned.

Cheap Talk for Coordination and Negotiation

So far, we've discussed revealing private information about one's type. Cheap talk is also essential for coordinating on future intentions.

For a message about one's intentions to be credible, it must satisfy two conditions:

1. Self-Signaling: You want the receiver to believe the message if and only if it's your true intention.
2. Self-Committing: If the receiver believes your message, you then have an incentive to follow through on it.

Consider the Battle of the Sexes. A couple is deciding where to go on Saturday night. Pat prefers a poly-friendly social event, Chris prefers a quiet night in. They both prefer being together over being apart.

• Pat says, "I'm going to the social event on Saturday."
  • This is self-committing: If Pat believes Chris heard this and will go to the social to be with her, Pat's best response is to actually go to the social.
  • This is also self-signaling: Pat wants Chris to believe she's going to the social because that's her preferred outcome.
• Simultaneously, Chris says, "I'm stocking up on snacks for a movie night at home Saturday."
  • This is also a self-signaling, self-committing message for Chris's preferred outcome.

When both send conflicting, credible messages, they may end up in a "sidewalk shuffle," achieving no coordination at all. This highlights a key aspect of pre-negotiation talk: it's not just about sending a message, but about establishing a shared understanding, which can be difficult when preferences over coordinated outcomes diverge.

Long Cheap Talk: Revelation and Compromise

Real negotiations are rarely a single utterance. They are extended conversations. The paper "Long Cheap Talk" by Aumann and Hart provides a powerful, though technically dense, framework for thinking about this.

The key insight from Aumann and Hart is that a long negotiation isn't just more talk; it has a structure. It can be viewed as an alternating sequence of revelations and agreements (compromises).

Let's construct a scenario: A couple (Hotwife and Husband) is negotiating the rules for MFM (male-female-male) play for the first time.

1. Stage 1 (Revelation): The Husband is unsure of the Hotwife's true feelings. Is she genuinely enthusiastic, or is she hesitant but willing to try? The Hotwife sends a cheap talk signal: "I've been thinking about it a lot and I'm very excited" vs. "I'm open to it, but also a bit nervous." This is a partial revelation of her 'type'.
2. Stage 2 (Agreement/Compromise): Based on her message of excitement, they can now "compromise" on the scope. They could, for instance, agree to flip a coin (a "jointly controlled lottery"): heads, they agree to allow a very wide range of activities; tails, they agree to a more restricted set of activities for the first time. This isn't about one person dictating, but about jointly agreeing to a path.
3. Stage 3 (Revelation): Depending on the outcome of the "coin flip," another revelation can occur. If the outcome was "wide range of activities," the Hotwife might then offer a more specific, credible signal about a hard limit within that wide range. If the outcome was "restricted set," such a signal might not be necessary or credible.

This structured process allows the couple to explore preferences and possibilities in a way that a single "take-it-or-leave-it" statement could not. It allows them to zero in on a mutually acceptable outcome by breaking down a complex negotiation into manageable steps of information sharing and joint decision-making.

Conclusion

In this lesson, we have moved beyond costly signals to understand the power and limitations of mere words. We've established that the credibility of cheap talk is not an inherent property of the message but a function of the strategic environment, primarily the alignment of interests between players.

Key Takeaways:

• Cheap talk can be perfectly credible when sender and receiver interests are aligned, as lying would make both parties worse off.
• When interests are in direct conflict, cheap talk is often ignored and becomes "babble."
• With partial alignment of interests (a bias), cheap talk can still be informative, but it becomes coarse and imprecise, as shown by the Crawford-Sobel partition model.
• Cheap talk is vital for coordination, and its credibility depends on messages being both self-signaling and self-committing.
• Extended negotiations ("long cheap talk") can be understood as a structured process of alternating revelations (sharing information) and agreements (making compromises).

Preview of the Next Lesson:

So far, our focus has been on the informed player—the Sender—and their strategic decision of what to say. We've treated the uninformed player—the Receiver—as a passive listener who simply updates their beliefs. But the uninformed player can also be strategic. In the next lesson, we will flip the script and analyze screening models, where the uninformed partner actively designs a "test" or a menu of options with the specific goal of inducing the informed partner to reveal their private information.