Are traders on Kalshi being profiled?

And on Polymarket -- what did you think Copy-Trading was all about?

Disclosure: I run Kalshinomics.com, which may earn Kalshi referral fees. I may trade event contracts on Kalshi and securities on other platforms. Readers should consider this relationship when evaluating my analysis. For educational purposes only, not investment advice.

And to add a personal note: most of my career was involved with regulated, centralized exchanges, so I am naturally biased towards favoring that market structure vs decentralized exchanges and single dealer platforms. I try to be fair, but this is not completely unbiased analysis. I also have never traded on Polymarket.

Exchanges are not neutral. Market design choices determine who gets the best prices, who supplies liquidity, who gets picked off, and how much activity the platform attracts. The exchange attempts to create an environment that provides incentives to various user types to encourage more activity. More activity means more commissions. This differs from the sportsbook or single-dealer model where the exchange is taking the other side of the users’ bets.

The key point is not that one design is ‘fair’ and another ‘unfair’ in a moral sense. It is that each choice changes the balance between three groups: squares, sharps, and dealers. I’ll use that lens to examine three decisions: identifiable RFQ flow, public wallet visibility, and taker fees.

I’ll use a simple framework, adapted from how market practitioners think about these design choices. Let’s put users into three distinct buckets:

Squares: Traders whose orders contain little short-term price information. They may be large and move markets, but they are not consistently informed.

Sharps: Traders with superior information, models, or speed; their trades are, on average, more price-informed.

Dealers: Market-makers who provide liquidity on the orderbook and respond to quote requests. They tend to earn spread from squares and lose to sharps.

Note that sharps and dealers are not necessarily different people or groups. A dealer will often also have some sharp information from time to time and use that for both providing more liquidity and taking liquidity. A user might be sharp in one market category and square in another. You might be a winner trading election markets through sober analysis but also enjoy throwing some money down on the game.

But one thing a sharp has going for them is no obligation to quote. Imagine you are contractually obligated to provide two-sided quotes (continuously posting both buy and sell limit orders in the market), with spreads < $0.10, and on 500 different markets for 10 hours a day. Now imagine there’s a copy of you that has no obligations but sits there all day doing research and only trading when they believe they’ve found a mispricing. This is the source of well-founded paranoia amongst market-makers.

The money flows from the squares to the dealers, and from the dealers to the sharps. Each group’s activity creates flows in fees to the exchange.

This is different from the single dealer sportsbook model. This is by definition only one dealer, the book itself. Sharps and squares trade with this dealer, and sharps can be limited by the house to prevent them from winning further. Exchanges don’t have this limiting option, but that doesn’t mean it’s a “level playing field.” The sharp has more information than the square about the fair price. And most of the time there’s nothing untoward about this. Using public information, better models, and differentiated research is normal and desirable in markets.

There is a different kind of unfairness we have laws against. Insider trading, trading ahead of nonpublic customer orders, market manipulation. That’s not what this piece is about. I’m talking about the legal ways market design decisions affect participants, and those choices can still lead to massive differences in outcomes.

Let’s dig into a few examples from prediction markets where market structures are evolving rapidly:

Case 1: Should RFQs be anonymous?

Recently on X there was a debate on counterparty IDs on Kalshi, with a fair amount of confusion about how the design works. The choice they made has real consequences, but is not especially unusual in derivatives markets.

Note this is only for RFQs, NOT standard order books. The limit order books are anonymous.

A limit order book is an ordered set of all the buy and sell limit orders for that market. Whenever the prices cross, a trade is executed. The best (highest) buy order and the best (lowest) sell order are called the “top of book,” If the best buy limit order is 0.50 and the best sell limit order is 0.51, the top of book is 0.50 @ 0.51.

RFQ stands for “Request for Quote.” In this case it’s used for asking for a price on a combo (or parlay), where the user is making a bet on multiple outcomes all happening together. Either every prediction is right or the bet doesn’t pay out. RFQs are common in derivatives and often used for large or complex trades.

When the user builds the combo and sends the RFQ, a request message is publicly sent asking for a specific combination of markets, with direction and size, and includes a creator ID, which looks like this:

"creator_id": "079725096ca91f5aa4aa95809c39029cb315c28d4386f63d97a811f16403a597"

If the same individual makes another request, the responder can tell it’s the same counterparty, but not who it is. Creating a database of these requests and the results obviously allows for some degree of trader profiling. I would expect that any serious RFQ responder is already doing some version of this. A responder can choose not to show quotes to a specific counterparty, if they consistently make winning trades or are systematically hunting for pricing errors.

This is publicly available information via the API. It is not reserved for favored parties. Yes there’s a hurdle in accessing this information and building an effective profiling system. I put together a basic listener in about 15 minutes using Claude Code. Anyone can respond to RFQs, though this requires significant investment in pricing logic. This is different from the single-dealer sportsbook model, where only the house dealer itself is creating prices.

What’s the challenge in providing responses in an automated fashion to these requests? Toxic flow. Toxic flow is trades that are unusually bad for whoever takes the other side, often realized immediately post-trade, but not always. It’s often for technical reasons that are not obvious to a computer program but a human trader could smell and avoid, or else not have a long career in trading. In a combo trade, a common version would be a pricing system that misprices correlated events, if events A and B are independent P(A and B) = P(A)*P(B), if they’re correlated, you need to price out P(A conditional on B)*P(B). A team who scores more touchdowns is more likely to win the game. These two events are not independent and wouldn’t be priced as such.

Yes this is a favorable design decision for those responding to the quotes. Certain combinations can be tricky to price or mispriced, and clever requestors can assemble combos that get an automated system to respond at a price a thinking human would not, earning the requestor a profit. Responders obviously don’t want this, and can choose not to respond to specific IDs or show them wider markets. Most requestors are likely non-toxic, but if a flaw is discovered in the responder logic, it will be exploited repeatedly for as much size as possible.

This is not theoretical, in Dec 2025 some sharp (or you might say toxic) users on Novig found a bug in their parlay pricing, where correlation was not applied correctly for specific markets, and then the users made bets at advantageous prices. Following this was a back-and-forth X debate on whether the trades were voidable or not. The trades ended up standing.

Sportsbooks address a similar problem by limiting users directly. Kalshi’s RFQ model is different because there can be many responders to a request; one or several may decline to quote without preventing others from making prices.

This is not unique in the context of trading in general. On a bank sales and trading desk, the trader knows exactly the counterparty an order is coming from. All else equal an order from TOXIC_FUND is going to get a less favorable price than a long term buy and hold. In a trading pit you might not know the end user, but know who the broker is and what the average toxicity of their trades looks like.

To summarize: Versus complete anonymity, counterparty IDs tend to help dealers and hurt sharps. Assuming competition amongst dealers - they are also likely good for squares. If dealers are competing against other RFQ providers and can profile toxic vs non-toxic flow, they are going to provide tighter pricing to IDs identified as non-toxic. A dealer would have to be wider / more cautious with no counterparty ID to make the same expected profit. Note that solely from a trading perspective and assuming perfect competition, the sportsbook model, being friendliest to the dealer in terms of identifying and limiting smart counterparties, could afford to show the best prices to squares.

You could imagine other design decisions - Perhaps only the winner of the RFQ gets to see the requestor ID. Or maybe only the top N responses in price get to see it? Or make it anonymous.

What happens if we make the limit order books non-anonymous, but add in crypto wallets instead of KYC’d users?

Case 2: Kalshi vs Polymarket, anonymous vs on-chain:

Because Polymarket now has a U.S. exchange, I should be clear that here I am talking about the offshore crypto platform.

Trades on Kalshi’s limit order books are anonymous. No counterparty info is shown. On Polymarket’s offshore platform, you can see the wallet IDs for executed trades, but not the identity of resting orders.

This is similar to the “counterparty ID” above but with one major difference. On Kalshi, users are KYC’d and are limited to one account. It’s explicitly against their rules for an individual to have multiple accounts. On Polymarket, a user can control multiple wallets.

In theory a user could make a new wallet for every single trade, effectively making them anonymous.

It also leads to “copy-trading,” where users attempt to find patterns that indicate an account is sharp. There are many “whale-watching” or copy trading tools available. Users may profile wallets with large P&L in a group of markets, or new wallets making large bets on a geopolitical situation, on the theory that those traders may be informed. The thesis is that the price a sharp account is paying has positive expected value (+EV), and continuing to trade in the same direction would also be +EV. This increases the price impact of any trades the sharp account makes, depending on how much activity is driven by copy-traders.

Sharp traders could respond to this by fragmenting their trading across multiple accounts. They may have an account that has negative PNL on a certain market type. This account is unlikely to be copy-traded. When building a position, they would prefer to use this relatively anonymous account, rather than suffer the price impact of having their trades copied before they’ve built their position. If copy-traders are too aggressive following the sharp account, this creates an incentive to build the position on the anonymous account, and then trade in the followed account, generating further price impact and increasing profits. Is this manipulation or simply smart situational awareness of the impact of your trades? If the intent was to buy a large position anonymously, then buy on the main account to trigger copy-trading, and then sell at higher prices to those copy-traders in a third account….. that sounds like the kind of thing you eventually read about in an enforcement action, at least if it happened on a regulated market.

I would be cautious about using simple copy-trading strategies. The lesson is not to ignore all counterparty information, but to recognize that sophisticated traders are aware of it and can adapt.

Additionally if part of a sharp’s trading is more of a market-making strategy, copy-trading will fail to generate profits. The sharp’s profits are generated by their execution prices rather than taking mispriced liquidity.

Back to our framework, who is this good for? If no sharps ever used alternate accounts, this would be better for dealers, who would provide less liquidity to sharps and more to squares. If a sharp handles wallet management well, it may lead to increased profits for them, but it creates another technical challenge to manage.

It also creates a bad incentive for sharps who understand the impact of their own trades via copy-traders. This game of mimicking “smart” accounts without having information yourself is both gameable and possibly distorting to price discovery. This market design feature favors the most sophisticated sharps, at the expense of those who have expertise on pricing but not market microstructure.

Kalshi’s anonymous order books avoid these copy-trading dynamics, but also introduce an additional friction for sharps:

Case 3: The Taker Fee

Many exchanges choose to provide different fee structures for makers (those who post limit orders that get filled), vs takers (sending immediately executable orders). On Kalshi in particular the fee difference is stark. The goal is to incentivize providing liquidity on the orderbook vs removing it.

On Kalshi, the fee for takers on most markets is 0.07*p*(1-p), where p is the trade price. This is a downward opening parabola that peaks at 1.75 cents for 50 cent contracts, and declines toward 0 and 100 cents. There are some markets with no fees, and others with smaller fees for makers as well. But this fee sharply disincentivizes taking liquidity and is a large cost to sharp takers. Compared to having no fee, they need to disagree with the current price by as much as 1.75 cents more to still have edge. For example if a sharp is hoping to realize 1 cent of edge on a buy order, they now need to disagree with the offer price by 2.75 cents when the price is around 50.

For the most part, Polymarket offshore currently has no fees, though I would expect this to change in the future. No fees allow for more aggressive price correction, vs sharps attempting to get filled by resting their liquidity in the book and hoping to trade with a square.

Compared to sharps, squares are less likely to be price sensitive or compare the net price between making and taking. They may also favor features like UI and ease of use - one clue is significant trading volume on Kalshi through Robinhood, despite strictly higher fees. Sharps on the other hand are trying to make small edges repeatedly, and this fee will turn a trade into a no-trade for small edges.

Dials not switches

These choices about anonymity, identity, and fees do not affect all users equally. They shape the mix of squares, sharps, and dealers that a market attracts. Many of the design decisions in this article are primarily tradeoffs between sharps and dealers. Squares are affected too, but for that group distribution, product design, and ease of use are often even more important. That is a topic worth its own article; Adhi’s piece at 50c Dollars on Robinhood and distribution is a useful starting point.

Make a market too unfriendly to sharps and prices may become less efficient; make it too unfriendly to dealers and liquidity may dry up. In that sense, Kalshi’s anonymous limit order books are relatively sharp-friendly in terms of anonymity, even as taker fees create friction for sharps who primarily execute by taking liquidity. Its RFQ system, by contrast, gives dealers more protection on a more complex product.

There is no neutral setting of these dials. Identity, anonymity, fee schedules, and quoting rules all shape who shows up, who stays, and how information gets incorporated into price. Prediction markets are still deciding how to tune these dials for different users. Through competition, I expect we will see a variety of choices across venues, as well as aggressive fee competition on the sharper venues which will lead to more efficient prices.