The Best Trader to Copy in Every Sport on Kalshi and Polymarket

Abstract

Public leaderboards on prediction markets rank traders by their overall profit, and copy trading tools invite users to follow the names at the top. This study asks whether a trader's record should instead be read one sport at a time. Using 90 days of settled sports bets by identified traders on Kalshi and Polymarket, each trader's bets are evaluated as a flat $100 copy at the price the trader paid, so that results reflect selection rather than position sizing. Among 5,408 trader-sport records on Kalshi and 4,359 trader-sport records on Polymarket with at least 30 settled bets, fewer than half came out ahead (46.9% and 45.1%). The leading trader in each sport returned between +7.2% and +127.2% on stake. The central finding is that this performance does not travel. The top 10 traders in each sport returned +26.8% inside their sport and +1.4% outside it on Kalshi, and +15.0% inside their sport and -4.2% outside it on Polymarket. The results are consistent with skill that is specific to a sport, and suggest that a copier should follow a trader only in the sport where the record was earned.

1. Introduction

A prediction market prices the probability of an event, and a trader profits only by buying outcomes the market has underpriced. On Kalshi and Polymarket a large share of volume is sports, and both venues expose enough of their trade history to attribute bets to individual traders: Kalshi through public profiles, Polymarket through on-chain wallets. That makes it possible to ask a question that matters to anyone who copies another trader's bets. Is a good trader good everywhere, or only in the markets they know?

The question has a practical edge. Leaderboards and copy tools treat a trader as a single record. If knowledge is specific to a sport, for example a feel for tennis in-play pricing or for injury news in the NFL, then an overall record mixes the sport where a trader has an edge with sports where they do not, and copying the whole record dilutes the part worth copying.

The study makes three measurements. Section 4.1 describes how many traders in each sport beat a flat copy at all. Section 4.2 identifies the leading trader in each sport on each venue. Section 4.3, the main result, takes the leading traders in every sport and evaluates them in every other kind of market they traded over the same window. Section 5 discusses what the results do and do not show, and Section 6 sets out the limitations.

2. Data

The data are every buy recorded on Kalshi and Polymarket between July 18, 2026 and October 16, 2026 (UTC) that carries an identified trader and whose market had settled as a win or a loss when the export was taken. Bets on markets still open, voided markets and sells are excluded. Kalshi identifies a trader only when their profile is public, so the Kalshi sample is the set of traders who choose to show their trades. Polymarket identifies every trader by wallet.

Each market is assigned to a sport from the venue's own classification (Kalshi's series categories and tags, Polymarket's event tags), mapped onto one shared list of sports. No sport is inferred from a market title. A trader enters the sample for a sport once they hold at least 30 settled bets in it, and a sport is reported on a venue only when at least 40 traders qualify there. Table 1 gives the resulting sample. The largest field is Kalshi tennis, with 1,883 qualifying traders.

3. Method

3.1 From fills to bets

A venue records a position as many fills, and a single order can produce dozens. A copier enters a position once. All of a trader's buys of one market and one side are therefore folded into a single bet, priced at the trader's average entry (dollars spent over shares bought). On Polymarket a side is identified by its outcome token, because both teams of a game are recorded as purchases on the same market. Bets with an average price below 5c or above 95c are excluded: at the extremes a one cent recording error changes the payout of a winning bet by an order of magnitude and would dominate a trader's result.

3.2 Flat-copy return

Each bet is evaluated as a fixed stake of $100, bought at the trader's average price and held to settlement. For a bet bought at price p, the profit per dollar staked is

A flat stake removes position sizing from the measurement. A trader who sizes up on their worst ideas and a trader who sizes evenly receive the same score for the same picks. The flat-copy return therefore measures selection, which is the part of a record a copier inherits when copying at a fixed amount.

3.3 Leaders and the out-of-sport test

Within each sport and venue, traders are ranked by total flat-copy profit, and the trader at the top is that sport's leader. Because the leaders are selected on the same bets they are scored on, their in-sport returns are the maximum of a large field and overstate what the same traders would earn going forward. The out-of-sport test addresses this in part. For the top 10 traders in every sport, the same flat copy is applied to all of their bets in other kinds of market over the same window. Nothing about those bets was used to select the traders, so the out-of-sport return is not inflated by the selection and shows how their picks perform away from the sport that put them at the top. A trader is scored outside their sport only with at least 30 bets there.

3.4 Cost case

Returns are reported gross of fees, spread and slippage, none of which can be reconstructed exactly for a hypothetical copier. As a stand-in, a cost case subtracts 4 percentage points from every return on stake.

4. Results

4.1 Most traders do not beat a flat copy

On Kalshi, 2,537 of 5,408 trader-sport records (46.9%) returned more than zero. On Polymarket, 1,965 of 4,359 trader-sport records (45.1%) returned more than zero. Pooling every qualifying trader's bets within a sport gives returns within a few points of zero on both venues (Table 2), which is what a market that prices sports reasonably well should produce for the average participant. The lowest share ahead on Kalshi is in Esports (44.8%), and The lowest share ahead on Polymarket is in NFL (39.3%).

4.2 The leading trader in each sport

Tables 3a and 3b list the leader in each sport on each venue. Every leader returned a positive flat-copy result in their sport, from +7.2% (nigiri99, Polymarket NHL) to +127.2% (gross.wound, Kalshi Cricket). The two venues' leaders trade differently. 8 of the 9 Kalshi leaders paid an average of 40c or less, so they win less often than not and earn their return from the size of each win. 7 of the 9 Polymarket leaders paid between 45c and 53c on average, near even money, and most won more often than they lost.

4.3 Performance outside the sport

Table 4 is the main result. Taking the top 10 traders in every sport and applying the same flat copy to their bets elsewhere, the return falls from +26.8% to +1.4% on Kalshi and from +15.0% to -4.2% on Polymarket. 34 of the 69 Kalshi traders with enough bets to score outside their sport lost money there, and 43 of the 66 Polymarket traders with enough bets to score outside their sport lost money there. Outside their sport these traders perform about as well as the average trader inside it.

4.4 Robustness

Under the 4 point cost case the top 10 traders still return +22.8% inside their sport on Kalshi and +11.0% inside their sport on Polymarket, and -2.6% outside it and -8.2% outside it. Every leader in Tables 3a and 3b remains positive in their sport after costs; the lowest, nigiri99, falls from +7.2% to +3.2%.

The pattern is not universal at the level of individual traders. 808kona805, the Kalshi MLB leader, returned +46.2% over 216 bets outside it; hellonewday, the Polymarket Tennis leader, returned +36.7% over 177 bets outside it. effective.minnow336, ranked 2 in Kalshi NFL at +32.6%, returned -21.2% over 117 bets elsewhere; zilong888, ranked 2 in Polymarket Esports at +32.7%, returned -11.1% over 323 bets elsewhere; 2themooooon, ranked 3 in Polymarket Tennis at +12.1%, returned -16.7% over 55 bets elsewhere.

5. Discussion

The results are consistent with skill that is specific to a sport. The traders who top each sport earn their returns there and roughly break even, or lose, elsewhere. Read naively, an overall record blends these together. A trader who is excellent in one sport and ordinary in five others can look ordinary on an overall leaderboard, and a copier who follows them everywhere receives mostly the ordinary part.

The in-sport returns should not be read as forecasts. Each leader is the best of a field of up to 1,883 traders, and the best of a large field will show a high return even when no one in it has any skill. The out-of-sport figures are free of that selection effect, and they are the more conservative evidence: on the information available, the leaders' edge is not visible outside the sport that selected them. This study does not separate how much of the in-sport return is skill and how much is the luck of being selected; a test on a later window would.

The venue difference in Section 4.2 suggests two styles. The Kalshi leaders win a minority of their bets at long odds, which produces large returns with wide variance; a copier would face long losing runs. The Polymarket leaders trade near even money with higher win rates and smaller returns.

Finally, a flat copy can diverge sharply from a trader's own profit. 0x2c33…0563 made $2,667,827 of their own money in Polymarket MLB, while a flat copy of the same bets returned +10.9%. The gap between the two is the effect of sizing. nigiri99 lost $12,604 of their own money in MMA while a flat copy of the same 85 bets returned +32.2%, because their largest bets lost. Leaderboards ranked by profit reward bankroll and sizing; a copier at a fixed amount inherits only selection.

6. Limitations

• Selection on the same data. Leaders are chosen and scored on the same 90 days. In-sport returns are upper bounds.
• Identified traders only. Kalshi's sample is limited to traders with public profiles, who may differ from those who hide their trades.
• No execution costs. Fees, spread, slippage and the delay between a trader's fill and a copier's are not modelled beyond the flat cost case.
• One window. 90 days covers part of several seasons. Sports out of season, or early in one, have thinner samples.
• Settled bets only. Bets still open at the export are excluded, which can understate traders who favour long-dated markets.
• Price band. Excluding bets under 5c and over 95c removes recording errors and also removes some genuine longshots and near-certainties.

7. Conclusion

On Kalshi and Polymarket, fewer than half of the traders active in a sport beat a flat copy there, and the pooled result for each sport is close to zero. The traders who lead each sport return substantially more inside it, and that advantage largely disappears outside it. For anyone copying another trader, the evidence favours copying a trader in the sport where their record was earned rather than copying the trader as a whole.

Appendix. Data and reproducibility

The manifest carries every figure on this page along with the window, the source revision and a SHA-256 hash over its contents. The page will not render a manifest whose hash does not match. Both files may be reused with attribution.