The "top-1% winners are patient limit-order liquidity providers, not insiders" finding is interesting, and I'd love to see it extended cross-venue.
I work on tooling that normalizes orderbooks across Polymarket, Kalshi, Limitless, and Smarkets. From that angle, a lot of what looks locally like skilled Polymarket market-making turns out to be cross-venue arbitrage that happens to land on Polymarket. The same underlying question routinely trades 3-8% apart across venues for hours at meaningful depth, and a fast multi-venue stack rests limits on the lagging book at the exact moments the leading book moves. Locally that's indistinguishable from disciplined liquidity provision; cross-venue it's closer to FX triangular arb on the consensus price.
If your timestamps are fine-grained enough, a clean follow-up: for the top 1% of Polymarket profit-takers, what fraction of fills land within N seconds of a same-question move on Kalshi or Limitless? If it's materially above baseline, some of "skill" resolves into "cross-venue infrastructure" — which is also a more durable edge than within-venue alpha, so it could partly explain the weak monthly persistence you observe (the cross-venue gap closes when too many players run the same stack).
This might also be consistent with your insider-trading conclusion rather than against it: an insider on a real-world event has every reason to hit the lowest-friction venue with aggressive market orders (Polymarket: permissionless wallets, no KYC, no withdrawal limits). That's a fundamentally different profile from the patient limit-posting strategy your top bucket runs, so the two populations cleanly separate in the data even if both are present.
I don't think that's surprising because the alternative would be that some people are able to predict the future. Whatever strategy one might figure out that works is long term destined to fail, as other people start using them. The only real way to make money there is by providing liquidity since it's a zero sum game. For the stock market this is not true because it's not zero sum, it grows over time.
We study trading gains and losses on Polymarket, the largest prediction market. Using 588 million trades ($67 billion in volume), we show that the gains are highly concentrated: the top 1% of users capture 76.5% of profits. Successful traders provide liquidity using limit orders that resolve favorably relative to realized outcomes while unsuccessful traders take liquidity using market orders. Monthly performance is weakly persistent, however, this may represent sample selection rather than skill. A detailed analysis of the trading behavior of the most successful accounts suggests that "insider'' trading is unlikely to explain the performance of the largest winners.
insider trading on events probably wouldn't show any trends, right? These are point in time events (they call them markets), but they are finite and short lived. An insider would be a one and done thing, so it would be pretty hard to spot them or trend any sort of month over month insider scheming imo.
Also...
> We study trading gains and losses on Polymarket, the largest prediction market
This is not a natural thing to say and I fucking hate that it's impossible to know anymore if I'm wasting time replying to an AI/bot or not
Not meant to sound like AI, but most academic journals limit abstracts to 100 words, so they rarely feel natural...
I agree: insiders are hard to study because they are finite and short-lived. We're pretty confident there are insiders out there trading on Polymarket; however, our conclusion is that they don't account for a significant fraction of the total trading gains on the platform.
I agree - you're not going to be an insider on a significant proportion of trades and it would be stupid to use the same account for more than a couple.
Insiders are going to be earning large amounts in single trades, either by betting a lot when it's odds-on or a small amount when it's out the odds (for a large return).
I think it's just bad tense, which I think makes it not AI amusingly.
Yes, power laws are everywhere. The exact shape of each distribution varies, however, and little is known empirically about the distribution of trading profits in financial markets.
Because it's not required and not common practice in our field at this stage. But none of us (I'm one of the authors) is affiliated with or has a financial interest in any prediction market platform.
What's the baseline here - in a world where every person is betting randomly X times a month, what would the distribution look like? There'd still be a small percentage that wins most of it, right?
We don't know the exact benchmark, but your insight is correct. We provide a simulation similar to what you have in mind towards the end of the paper, but you can generate almost any distribution you want by fine-tuning a simulation...
Nice paper, and thanks for releasing the dataset.
The "top-1% winners are patient limit-order liquidity providers, not insiders" finding is interesting, and I'd love to see it extended cross-venue.
I work on tooling that normalizes orderbooks across Polymarket, Kalshi, Limitless, and Smarkets. From that angle, a lot of what looks locally like skilled Polymarket market-making turns out to be cross-venue arbitrage that happens to land on Polymarket. The same underlying question routinely trades 3-8% apart across venues for hours at meaningful depth, and a fast multi-venue stack rests limits on the lagging book at the exact moments the leading book moves. Locally that's indistinguishable from disciplined liquidity provision; cross-venue it's closer to FX triangular arb on the consensus price.
If your timestamps are fine-grained enough, a clean follow-up: for the top 1% of Polymarket profit-takers, what fraction of fills land within N seconds of a same-question move on Kalshi or Limitless? If it's materially above baseline, some of "skill" resolves into "cross-venue infrastructure" — which is also a more durable edge than within-venue alpha, so it could partly explain the weak monthly persistence you observe (the cross-venue gap closes when too many players run the same stack).
This might also be consistent with your insider-trading conclusion rather than against it: an insider on a real-world event has every reason to hit the lowest-friction venue with aggressive market orders (Polymarket: permissionless wallets, no KYC, no withdrawal limits). That's a fundamentally different profile from the patient limit-posting strategy your top bucket runs, so the two populations cleanly separate in the data even if both are present.
We have a grad student working on matching markets across venues. Not a trivial task at scale, but we hope to look at that eventually.
I don't think that's surprising because the alternative would be that some people are able to predict the future. Whatever strategy one might figure out that works is long term destined to fail, as other people start using them. The only real way to make money there is by providing liquidity since it's a zero sum game. For the stock market this is not true because it's not zero sum, it grows over time.
There's probably also some hedging going on across accounts that look like directional bets.
[flagged]
We study trading gains and losses on Polymarket, the largest prediction market. Using 588 million trades ($67 billion in volume), we show that the gains are highly concentrated: the top 1% of users capture 76.5% of profits. Successful traders provide liquidity using limit orders that resolve favorably relative to realized outcomes while unsuccessful traders take liquidity using market orders. Monthly performance is weakly persistent, however, this may represent sample selection rather than skill. A detailed analysis of the trading behavior of the most successful accounts suggests that "insider'' trading is unlikely to explain the performance of the largest winners.
Full dataset available at https://huggingface.co/datasets/vgregoire/polymarket-users
insider trading on events probably wouldn't show any trends, right? These are point in time events (they call them markets), but they are finite and short lived. An insider would be a one and done thing, so it would be pretty hard to spot them or trend any sort of month over month insider scheming imo.
Also...
> We study trading gains and losses on Polymarket, the largest prediction market
This is not a natural thing to say and I fucking hate that it's impossible to know anymore if I'm wasting time replying to an AI/bot or not
Not meant to sound like AI, but most academic journals limit abstracts to 100 words, so they rarely feel natural...
I agree: insiders are hard to study because they are finite and short-lived. We're pretty confident there are insiders out there trading on Polymarket; however, our conclusion is that they don't account for a significant fraction of the total trading gains on the platform.
I agree - you're not going to be an insider on a significant proportion of trades and it would be stupid to use the same account for more than a couple.
Insiders are going to be earning large amounts in single trades, either by betting a lot when it's odds-on or a small amount when it's out the odds (for a large return).
I think it's just bad tense, which I think makes it not AI amusingly.
> the top 1% of users capture 76.5% of profits
This seems to be similar to OnlyFans, and the economy at large...
Yes, power laws are everywhere. The exact shape of each distribution varies, however, and little is known empirically about the distribution of trading profits in financial markets.
Wait- why isn't there any conflict of interest statement provided in this paper?
Because it's not required and not common practice in our field at this stage. But none of us (I'm one of the authors) is affiliated with or has a financial interest in any prediction market platform.
Isn't it common practice and required to disclose a conflict of interest? Just not to explicitly say there are none.
Thanks for the clarification. Given the scrutiny on these platforms, this is timely done. Thanks.
What's the baseline here - in a world where every person is betting randomly X times a month, what would the distribution look like? There'd still be a small percentage that wins most of it, right?
We don't know the exact benchmark, but your insight is correct. We provide a simulation similar to what you have in mind towards the end of the paper, but you can generate almost any distribution you want by fine-tuning a simulation...
Just curious but how are bets arbritated on these website?
Meaning who decides if an outcome was yes or no? Answers to things like "Who will win the next Best Picture Oscar?" are fairly obvious and binary.
Can we make bets whose answers are not binary yes/no?
What about "Will celebraty X and Y break up?"? Does Polymarket go to X and Y to confirm if they broke up or something :D
https://docs.polymarket.com/concepts/resolution