Eve is here. Below, Rajiv Sethi analyzes Polymarket’s larger position in betting that the US will attack Iran. That didn’t happen, and President Trump decided to go head-to-head at the last minute, making it look like a case where he acted on inside information that turned out to be false.
Or was it that? Sethi argues that the trader could still have made a profit by encouraging others to trade based on the assumption that the sudden move was based on special information. But it may have been more of a thorn in the side with the original trader than recouping his apparent losses.
Cheating other traders has long been considered acceptable in the western regions of foreign exchange and commodity markets, where there is no concept of insider trading and other games are played. The biggest one is front-running customer orders, where the size of the final trade (which can be spread across the trading venue) or simply who makes the trade can have a market-moving influence. It is also said to be illegal in FX and Commodities, but it is widely believed that it happens. One legendary trader was skilled at placing orders that made other market participants believe that he must be acting in anticipation of, or actually placing a portion of, a large order from a customer. Andy Kreiger, a foreign exchange trader at Salomon (then known as Bankers Trust), was considered by his colleagues at O’Connor & Associates, a top independent trader in foreign exchange options, to have an absolutely uncanny ability to repeatedly outsmart other traders.
Written by Rajiv Sethi, Professor of Economics, Barnard College, Columbia University. External professor at Santa Fe Institute. Originally published on Imperfect Information
In a previous post, we described the case of a polymarket trader who made a series of well-timed bets on the ouster of President Nicolas Maduro and made over $400,000 in profits (on a $32,000 investment) in just a few days. What was unusual was not the quick bucks or high rate of return, but the fact that the account was opened just weeks before the attack, bet only on Venezuela-related events, and stopped trading once the Maduro contract was settled. This pattern of activity suggested to many observers that this trader had access to inside information, and the coverage was extensive.
Trading on Polymarket has an unusual combination of public and private features. Although all transactions are visible on the blockchain, the real-world identity of most account holders is unknown even to the exchange. This not only makes it easier for insiders to trade undetected, but also allows the general public to look for activity that suggests insider trading. Tools such as Insider Finder were developed for exactly this purpose. Using such tools can be profitable, as insider trading can be identified early and reliably, allowing outsiders to copy the trades before they have a significant impact on prices.
But let’s think about the implications of this. If market participants are looking for patterns suggesting insider trading in order to replicate these trades, they can benefit from generating such patterns in the hopes that their trades will be actively copied. This imitation affects prices, and traders who are (falsely) suspected of insider activity may quickly take a profit and close out their positions. This can be done very carefully using separate wallets on Polymarket, so that the account that made the first move will record a loss and the profit will accrue undetected elsewhere. This is an example of spoofing. 1
The possibility of something like this happening occurred to me a few days ago when an account was flagged as a possible insider on social media. The activity here was even more obvious than in the Maduro incident. The account was opened on Jan. 14 and spent more than $40,000 buying 250,000 contracts at an average price of 15.6 cents, betting that the United States would attack Iran on the East Coast by midnight that day. All trades took place between 8:29 PM and 11:53 PM (UTC). Had the attack occurred, this trader would have made more than $215,000 in profits in just a few hours. However, no attack occurred and all stocks expired worthless.
I think it’s no exaggeration to say that if this trader’s bet had been successful, it would have received a lot of media attention. However, because they did not do so, this activity received little attention and was quickly forgotten. This is unfortunate. There is something to be learned from these trades, but there are (at least) two very different explanations for them.
One possibility is that a planned attack was postponed at the last minute and traders learned of the plan but not of the reversal. This seems very likely to happen to me. But there is another possibility worth considering. Let’s look at the price change between a trader’s first contract purchase (8:29 PM UTC) and last contract purchase (11:53 PM UTC):2
The horizontal dashed line is the average price paid by this trader. The total number of contracts traded during this period exceeded 5.58 million, and the price per contract rose to 55 cents. If this individual had used a different wallet, it would have been easy to bet on the attack at a price that would have resulted in a significant overall profit. This means that while losses were recorded in the flagged accounts, there were much larger gains elsewhere.
Importantly, impersonation can be profitable. If you trick others into believing that you are trading using inside information, which causes them to imitate your actions, you can amplify price movements in a way that can result in significant profits. This does not require any possession of inside information. This strategy relies solely on the assumption by other market participants that insider trading is popular on the platform.
Financial markets are complex adaptive systems where prices are determined by the interaction of disparate trading strategies. Moreover, strategy ecology is never static. It evolves under the pressure of payoff differentials. Some strategies (including insider trading) feed information into the market, others attempt to extract information from market data, and still others seek to deceive the extraction strategy and profit from it. 3 This can lead to bubbles, momentum effects, excessive volatility, and even flash crashes.
One of my favorite experimental papers in economics is Rosemary Nagel’s work on guessing games. This paper has made it very clear that there are conditions under which certain standard equilibrium concepts are insufficiently predictive. A typical guessing game is simple. Imagine you have a large group of players and each player chooses numbers privately in closed intervals. [0, 100]. The winner is the player whose selection is closest to half of the overall group average, and if there are multiple winners, the prize money is divided equally among everyone. 4
This game has a unique Nash equilibrium in which all players choose exactly zero, but this is neither descriptively accurate nor normatively useful. Numbers close to zero are rarely selected and will never win. There is usually a distribution of responses that peaks at certain values, and winning players often choose values in the high teens. If you’re playing with a group known to be familiar with Nagel’s findings (or the post you’re reading), you’ll probably choose something smaller. Still, choosing a value at or near zero will likely result in failure.
Such guessing games are sometimes called Keynesian beauty contests, after a memorable and insightful passage in The General Theory. Keynes favored professional investment in newspaper competitions in which “a prize is awarded to the contestant whose choice most closely approximates the average preference of the competitors as a whole.” As a result, according to Keynes, “we invest our wisdom in predicting what the average opinion expects us to be; and I believe that some of us practice the fourth, fifth, and higher levels.”
Financial markets in general, and prediction markets in particular, can be very effective mechanisms for aggregating information. New sources of information can be considered to generate forecasts, even under historically unprecedented circumstances. But they are far from perfect, and the evangelical zeal with which some have promoted them is not really justified. The problem of predictive accuracy for statistical models can only be solved empirically. You can’t think logically about how to get to the answer. I don’t think it will take decades, as some have claimed, but it will take time, and we are just getting started.5
__________
1 Insider trading, wash trading, and spoofing are all prohibited on Kalshi and other regulated exchanges. Although the overall prevalence of insider trading and spoofing at Polymarket is unknown, there are steps in place to detect wash trading, which were very important during certain periods.
2 Thanks to Allen Sirolly for providing the data for this article.
3 My post about spoofing ended up being published as an opinion piece in the New York Times.
4 I first encountered this game in a chapter by Mario Enrique Simonsen in a book edited by Arrow, Anderson, and Pines. I highly recommend this book to all graduate students majoring in economics. Guessing games (and similar games) have opened the door to exploration of the concept of alternative solutions that enable bounded rationality.
5 I spoke about some of these issues yesterday on Michael Smerconish’s CNN show, and a few days ago on his radio show in more detail. It covered a considerable amount of land in a relatively short period of time. This is a short but very important clip:
