This is the first of three posts this week on prediction markets. They can be read independently, but they build. Today’s question is about information. Wednesday’s is about jurisdiction. Friday’s is about ontology, which sounds worse than it is.
Start with a fact pattern. In the weeks surrounding the Iran nuclear negotiations, large futures positions were established in S&P 500 contracts shortly before President Trump announced a five-day pause on striking Iranian power plants. A similar pattern preceded an earlier social media post about “productive” talks with Iran. And separately — this is the part that makes it a Kalshi story rather than just a Wall Street story — an unknown trader reportedly cleared approximately $1 million on prediction-market contracts about Iran’s military situation, timed to the same sequence of events.
Senator Murphy went on the record calling this insider trading. That may be correct as a legal matter, and the legal question is worth watching. But there’s a distinct structural point that sits underneath the legal one, and I want to separate them — because the structural point is interesting regardless of whether any law was broken, and because it illuminates something specific about what prediction markets are and aren’t doing when they price political risk.
What cheap talk is, and why it sometimes works
In 1982, Vincent Crawford and Joel Sobel published a paper that has quietly become one of the most useful frameworks in political science.1 The question they asked was: when can a sender transmit genuine information to a receiver through a message that costs the sender nothing to send? This is the cheap talk problem, and the intuitive answer is “never” — if the message is free, the sender will always send whichever message serves their interests, so the receiver has no reason to update on it. The message carries no information because it is unverifiable and costless.
Crawford and Sobel showed that this intuition is wrong under one condition: when sender and receiver have partially aligned interests. If the sender wants the receiver to take an action that is approximately the action the receiver would take on full information — not identical, but close — then the sender has an incentive to transmit at least coarse information truthfully. The receiver, knowing this, updates rationally on the message. Communication happens, imperfectly, through words that cost nothing.
Presidential announcements are, in the standard reading, cheap talk. A president announces a policy direction, a negotiating posture, a decision. The announcement costs nothing to make. Markets move on it. And the standard complaint about this setup — from game theorists, from journalists, from anyone who has watched a president walk back a statement — is precisely that the costlessness of the announcement makes it unreliable. Presidents say things. Markets move. Sometimes the things turn out to be true.
The Iran pattern is not that story. And this is the point.
The signal that arrived before the message
In the standard model of price discovery, an announcement moves markets because it carries information that wasn’t already priced in. The gap between the pre-announcement price and the post-announcement price is a measure of how much information the announcement contained. When that gap is large, the announcement was news. When it is small, the market had already figured it out.
When the market moves before the announcement, something different is happening. The information has already diffused through some other channel. The announcement, when it comes, is not transmitting information — it is ratifying information that has already been incorporated. The presidential statement is the last thing to know.
Now here is the formal distinction that matters. The pre-announcement trades are not cheap talk. They are costly signals — in Spence’s original sense.2 A futures position requires capital. It bears risk. It is observable after the fact. Whether the trader had private information or was simply reading observable signals better than anyone else, the trade itself is a commitment of resources in a direction. It is costly. It is therefore credible in a way that a presidential tweet is not.
What the subsequent announcement does, in this framework, is not transmit information. The information has already been transmitted, through the price. The announcement transmits something else: it provides the sender with a plausible public account of why the market moved. It converts a costly signal — which is evidence of something — into an apparently sufficient cause. The trade looks like rational anticipation of a public announcement rather than the exploitation of private information. The second transmission doesn’t add information. It adds deniability.
Why this is a prediction market problem specifically
A futures position in the S&P 500 is a relatively blunt instrument. It moves in response to many things, and attributing a specific price move to a specific piece of private information requires work. A prediction market contract on a specific event — “Will the US strike Iranian power plants in the next 30 days?” — is not blunt at all. It is pointed directly at the question. A large, well-timed position on such a contract is not plausibly explained by general market sentiment or macroeconomic factors. It is, in the formal sense, a highly informative signal.
This creates a problem that is distinct from the insider trading question. Suppose no law was broken. Suppose the trader was simply a sophisticated reader of publicly available signals — the tone of statements, the body language of negotiators, the positioning of military assets that any attentive observer could track. The prediction market contract still functions as a public revelation of that private inference. Once the position is large enough to move the market price, the information is out. Anyone watching the Kalshi odds on Iranian military action in the days before Trump’s announcement was receiving a signal — not from Trump, but from whoever was trading.
The market is supposed to aggregate dispersed private information into a public price. In this case, it may have done exactly that. The question is whether “aggregating dispersed private information” is a description we are comfortable with when the private information in question was generated inside a government that also has trading-adjacent relationships with the platform doing the aggregating.3
Crawford and Sobel’s cheap talk model assumes that the sender and receiver are distinct — that the person with the information and the person transmitting the message are the same entity, with interests that mayor may not be aligned with the receiver. The Iran pattern scrambles this. The sender (whoever had the information) transmitted it through a price, not a message. The receiver (the market) incorporated it before the official message arrived. The official message — the presidential announcement — arrived last, carrying no information, conferring cover.
This isn’t cheap talk. It’s the opposite of cheap talk. The costly signal came first. The free message came after. And the free message’s function was precisely to make the costly signal look like a response rather than a cause.
Wednesday: who gets to decide whether any of this is legal, and what happens when the regulator is a player in the market it regulates.
With that, I leave you with this.
1 Crawford, V. and Sobel, J. (1982), “Strategic Information Transmission,” Econometrica, 50(6), 1431–1451. The result is that equilibria exist in which the sender partitions the state space into intervals and sends the same message for all states within an interval — “babbling” within bins, but transmitting genuine coarse information across bins. The coarser the misalignment of interests, the coarser the equilibrium partition, and the less information gets transmitted. At maximum divergence of interests, the unique equilibrium is the babbling equilibrium: the message carries no information at all.
2 Spence, M. (1973), “Job Market Signaling,” Quarterly Journal of Economics, 87(3), 355–374. The core insight: a signal is credible if and only if it is differentially costly — if it costs the high-type sender less (in utility terms) to send than it would cost the low-type sender. A presidential announcement costs the same regardless of whether the underlying information is accurate. A large futures position does not.
3 Donald Trump Jr. is a strategic adviser to Kalshi. The current CFTC chair — the sole sitting commissioner — is a Trump appointee who intervened publicly within hours of Arizona’s criminal charges to back the company. President Trump has reportedly discussed a crypto-based prediction market on Truth Social. These facts do not establish wrongdoing. They do establish that the boundary between the information-generating entity and the information-aggregating platform is not clean, which is precisely the condition under which the cheap talk / costly signal distinction becomes load-bearing.