Prediction market trading fees: the entry price sets your cost, not the size of your trade (August 2026)

The fee you pay is a fixed percentage of what you staked, and that percentage is decided almost entirely by the price you bought at. On both venues the cost of a taker fill, as a share of your stake, is rate x (1 - P). The contract count cancels out. At a 0.07 coefficient that is 0.7% of stake at an entry of $0.90 and 6.3% of stake at an entry of $0.10, a factor of nine, for identical size and an identical trade. Then you cross the spread going in, pay a second fee going out, and cross the spread again on the way out. On a worked round trip below, 500 contracts bought at $0.30 and sold at $0.33, a 10% price move, the position finishes at minus $0.09. The same 500 contracts bought at $0.85 and sold at $0.88, a 3.5% price move, finish at plus $6.83. Same venue, same coefficient, same three cent move, opposite outcomes.

Everything on this page is computed from the coefficients the two venues publish, at fee schedule version 2026-08-04.a. Nothing here is an estimate of a fee.

The published coefficients, and the one number that does all the work

Polymarket publishes eleven category coefficients and zero-rates one of them. The maker figure is zero in all eleven, which is zero rather than a discount off a headline rate. Kalshi's two figures come from different Kalshi documents and its own publications disagree about whether resting orders are charged at all: the newsroom says they are fee-exempt, the help centre publishes 0.0175. Nobody has reconciled that publicly, including us.

The fee on a taker fill is rate x C x P x (1 - P), where C is contracts and P is the price in dollars. Your stake is C x P. Divide one by the other and the contracts and the price both cancel:

Fee as a share of stake = rate x (1 - P).

That is the whole page in one line. Size does not appear in it. Price does, and it dominates.

Read down any column. The longshot corner of the board is the expensive corner, and it is the corner most people start in because the tickets look cheap. A $0.10 contract does not cost less to trade. It costs nine times more, measured against the money you actually put up.

The 1.75% everybody quotes is not the number that comes out of your account

Every fee page in this niche quotes Kalshi's peak as 1.75%. That figure is correct and it is measured against the wrong thing for a trader's purposes.

At an entry of $0.50 the Kalshi taker fee is 0.07 x 0.50 x 0.50 = 1.75 cents per contract. A contract pays $1.00 if it resolves in your favour, so 1.75 cents is 1.75% of the payout notional. But you did not put up $1.00. You put up 50 cents. Against your stake, the peak fee is 3.5%, and that is the denominator that matters when you are deciding whether a position is worth taking.

Both numbers describe the same fee. Neither is wrong. But a page that prints 1.75% without naming the denominator halves the apparent cost of the worst-priced entry on the board, and none of the pages we measured names it. When you see a fee expressed as a percentage anywhere, including here, check what it is a percentage of before you use it.

A worked round trip, in full, at two entry prices

Kalshi, taker on both legs, 500 contracts each time, and a three cent favourable move each time. These are illustrative books chosen to show the arithmetic, not markets we traded.

Trade A, the cheap ticket. Book is $0.29 bid, $0.30 ask. You lift the ask at $0.30.

• Stake: 500 x $0.30 = $150.00
• Entry fee: 0.07 x 500 x 0.30 x 0.70 = $7.35
• The market moves. Book is now $0.33 bid, $0.34 ask. You hit the bid at $0.33.
• Exit fee: 0.07 x 500 x 0.33 x 0.67 = $7.7385, rounded up to $7.74
• Gross gain: 500 x $0.03 = $15.00
• Fees: $15.09
• Net: minus $0.09

The price rose 10% and the round trip lost money. Fees alone were 10.06% of the stake.

Now price it against the midpoint, because the midpoint is the only neutral reference. The mid went from $0.295 to $0.335, a four cent move worth $20.00 on 500 contracts. You captured $15.00 of it. The missing $5.00 is the spread, half of it crossed on entry and half on exit. Add the $15.09 of fees and the total cost of the round trip is $20.09 against a $20.00 move.

Trade B, the expensive ticket. Book is $0.84 bid, $0.85 ask. Same 500 contracts, same three cent move, same venue, same 0.07 coefficient.

• Stake: 500 x $0.85 = $425.00
• Entry fee: 0.07 x 500 x 0.85 x 0.15 = $4.4625, rounded up to $4.47
• Exit at $0.88 against a $0.88 bid, $0.89 ask book.
• Exit fee: 0.07 x 500 x 0.88 x 0.12 = $3.696, rounded up to $3.70
• Gross gain: $15.00
• Fees: $8.17
• Net: plus $6.83

The cheap trade got a price move three times larger in percentage terms and still finished behind. That is the entire argument of this page, done with real coefficients.

Note the line that never appears on a fee schedule: the $5.00 spread cost is identical in both trades and it is the largest single line in Trade B. A one cent spread on 500 contracts is $5.00 whether the contract costs 30 cents or 85 cents. Fee pages price the fee. Order books charge you the half-spread on the way in and the half-spread on the way out, and on a well-priced position that is the bigger number.

What the round trip actually needs to move before you are level

Combine both legs of the fee with a one cent spread and you get the move a position has to make before it is worth having. Fees below are shown unrounded, at 0.07, per contract.

Two readings, pointing in opposite directions, and both are true. In cents, the hardest place on the board is the middle, where you need four and a half cents of movement to break even. In percentage of stake, the hardest place is the cheap end, where you need a 22.6% move at $0.10 against a 1.8% move at $0.95. Which reading applies to you depends on whether your edge is expressed in cents of mispricing or in percentage terms. Most people carry a percentage in their head and trade the cent, which is how the longshot corner quietly eats a bankroll.

Two caveats on that table, because they are ours to own. Both fee legs are evaluated at the entry price, so the figure slightly understates below $0.50 and slightly overstates above it, because P x (1 - P) rises as a cheap contract appreciates and falls as an expensive one does. And a one cent spread is an assumption, not a measurement. On a thin book it will be wider, and every cent of extra spread adds a full cent to the "total needed" column at every price.

Rounding turns small longshot orders into the worst orders on the board

Kalshi rounds each fee up to the next whole cent. On any position of reasonable size that is a rounding error. On a small order at a low price it is the dominant cost.

One contract at $0.03 generates a raw fee of 0.07 x 1 x 0.03 x 0.97 = $0.002037. It rounds up to a full cent. You pay 4.9 times the unrounded rate, and that penny is 33% of your $0.03 stake. The same single contract at $0.50 rounds 1.75 cents up to 2 cents, a multiplier of 1.14 and 4% of stake.

So the rounding rule and the rate x (1 - P) rule push in the same direction, and they compound. The cheap end of the board is worse in percentage terms before rounding and worse again after it. Kalshi documents both the sub-penny rounding and a whole-cent rebate accumulator on its developer site, written for engineers rather than traders, and its own HTML fee page is a stub that redirects to a rate-limited PDF. There is no fee table, no per-price grid and no worked example anywhere on the crawlable web. That is the only reason a grid this basic is worth publishing, and it is why a 404 page held position 9 on kalshi fees at DR 26.

We do not know whether Polymarket rounds fractional fees, or in which direction. We have never fetched a human-readable Polymarket fee schedule; docs.polymarket.com/programs/builders/fees has never been read by this project. Every Polymarket figure on this page is carried unrounded and flagged as such. If Polymarket rounds up the way Kalshi does, small orders at low prices are worse than shown here, not better.

Our own book, priced with these same coefficients

Our paper trade log is public and it is not flattering. As of 4 August 2026 it holds 10 trades, 6 closed, net minus $28.30 on $500 deployed. Every entry carries paper_trade=true.

Separately, our scanner note records five positions opened on 28 July 2026 by the Collector B paper trader, run through the verified per-category rates. Break-even below is rate x (1 - P), the same formula as the tables above.

Two things in that table are worth more than the totals.

The last row was a guaranteed loss at the moment of entry. It was opened on a claimed 1.20% edge at a price where the fee alone is 2.81% of stake. That is not a bad outcome that happened to arrive. The arithmetic was negative before the market moved at all.

The fourth and fifth rows carry an identical 1.20% claimed edge and opposite economics, because break-even is a function of price and the two entries sit on opposite sides of it: 0.91% at $0.8693 against 2.81% at $0.2965. A single percentage edge threshold cannot tell them apart. Ours could not, and did not.

That is our own defect, not a finding about either venue. Our scanner still publishes a flat 1.0% minimum edge threshold on /scanner/ while our own notes record fee-adjusted break-evens of 3.63% at a 0.07 coefficient, 2.56% at 0.05 and 2.04% at 0.04. Every one of those is above the threshold we are still showing. The fix is a one-line change to reject any signal where the claimed edge is at or below the break-even for that entry price. It has not shipped. Until it does, the scanner keeps opening positions that cannot win.

The two-leg column is the number we would least like to publish: minus $3.48 across the book before any market moves. As a cross-venue strategy those five entries were negative on arrival.

Read none of these as rates. Five positions support no percentage. They are five worked examples with the arithmetic shown, which is the honest form at this sample size.

One stale coefficient, and what it does to a dollar figure

Polymarket documents eleven category coefficients ranging from 0.00 to 0.07. A competitor payout calculator, laikalabs.ai, has applied a single flat coefficient of 0.0625 to all of Polymarket for five months. That value is not one of the eleven, and a single flat number cannot be right for a venue that prices eleven categories differently and zero-rates one of them.

Here is what that costs a reader, on 100 shares at $0.50.

The geopolitical row is the one that should end the argument. On a category where Polymarket charges nothing at all, a flat coefficient prints $1.5625 of cost per hundred shares out of nothing, and it does so in a calculator, which is the format readers trust most because it looks like a machine did it.

This is not a claim that the people who wrote that page are careless. It is a claim about mechanism. Every fee figure on every page we measured is hand-typed, and nothing on any of them is reconciled against the venue's live schedule. A number typed once stays typed. Ours come from the feeSchedule object on the Gamma API and carry a version stamp, 2026-08-04.a, so that when it changes, the change is visible rather than silent. If our figures are stale, you can tell, because the version says when they were read.

We should be straight about our own read on this. The laikalabs.ai domain serves HTTP 403 to our fetchers, so the 0.0625 claim above is held in our research notes rather than in a live read we can re-run today. We are describing a claim we recorded, not asserting the current state of a page we cannot open.

The cost lines nobody prices, including us

An honest end-to-end number would include more than the fee and the spread. Here is what is missing and why.

Deposit and withdrawal charges. We have not verified a single deposit or withdrawal figure against either platform's own document. Competitor fee pages enumerate a Kalshi debit card deposit percentage, PredictIt withdrawal charges and bank transfer fees on other venues. We are not repeating figures we have not read at source, and neither should any page that cannot link the schedule it took them from.

The cost of collateral. These are fully collateralized, cash-settled contracts. Your stake sits locked from entry until settlement, earning nothing, and on a market that resolves in nine months that is a real cost that no fee schedule contains. It is raised constantly in trader forums and quantified nowhere. We are not going to attach a rate to it, because doing so means picking a risk-free rate and a holding period and presenting the product as a measurement.

Special-event fee tiers. Kalshi acknowledges these exist and names elections, awards ceremonies and large sporting championships. The rates are not quantified anywhere public. If you are trading an election contract, the coefficient on this page may not be the coefficient you pay, and there is no document that will tell you.

Adverse selection on the maker side. Posting a resting order is cheaper, at 0.00 on Polymarket and 0.0175 on Kalshi if the help centre is the current document. That gap is not free money. It is payment for writing an option to every better-informed trader in the book, and it gets exercised when you are wrong. The fee saving is measurable. The adverse selection is not, and we have not measured it.

What we could not verify

• Whether Polymarket rounds fees, and in which direction. docs.polymarket.com/programs/builders/fees has never been fetched by this project. Every Polymarket figure here is unrounded.
• Kalshi's maker fee. The newsroom says resting orders are fee-exempt; the help centre publishes 0.0175. Both are Kalshi documents. Unresolved.
• The entry price behind our own 3.63%, 2.56% and 2.04% break-evens. Our scanner note records those three figures without stating the price they were computed at. All three are consistent with an entry near $0.49, which would be a mid-board assumption the note does not make explicit. We are quoting them as recorded rather than recomputing them silently.
• Whether our open book holds four positions or five. Our site profile records 10 trades, 6 closed, 4 open. Our scanner note describes five open positions on $500 deployed. The two records come from different files and we are not netting figures across them. The settled log figure is 10 trades, 6 closed, net minus $28.30 on $500 deployed.
• Any deposit, withdrawal or transfer charge on either venue. Not read at source, therefore not published.
• Special-event fee tiers. Acknowledged by Kalshi, never quantified.
• Typical spread width. Every spread figure on this page is an assumption at one cent, not a measurement. Our collectors record book data and we have not yet published a distribution.
• Historical accuracy of one of our own break-even figures. An earlier version of our scanner table computed the Alaska position's break-even as 3.52% by applying the 0.05 sports rate to a market that carries 0.04. A cross-agent review pulled each market's live feeSchedule.rate and corrected it to 2.81%. The conclusion did not move, since the position was entered on 1.20% either way, but the number did. We are recording the correction rather than quietly shipping the fixed figure.

Nobody pays us for this page

PredictionEdge currently takes no referral revenue, and whether it ever will is undecided. There is no signup link on this page. A Polymarket code exists on the scanner page and has earned nothing.

That matters here more than on most pages, because the incentive on a fee page runs one way. A page whose revenue depends on you opening an account has a structural reason to quote 1.75% rather than 3.5%, to price the entry rather than the round trip, and to leave the half-spread out entirely. We have a structural reason to do the opposite, which you should discount for exactly as much as it is worth.

Our own arithmetic killed one of our own positions and it is still in the log.

Questions people actually type about this

How much are prediction market trading fees, really? As a share of what you staked, rate x (1 - P). At Kalshi's 0.07 coefficient that is 3.5% of stake at a $0.50 entry, 0.7% at $0.90 and 6.3% at $0.10. Polymarket runs 0.07 for crypto, 0.05 for sports, economics, culture, weather and other, 0.04 for finance, politics, mentions and tech, and 0.00 for geopolitical.

Does trading bigger size cost more in fees? Not as a percentage. The contract count cancels out of rate x C x P x (1 - P) divided by C x P, so the fee is a fixed share of stake at any size. The exception is very small orders on Kalshi, where rounding up to the whole cent bites: one contract at $0.03 pays a full penny, 4.9 times the unrounded fee and 33% of stake.

Are cheap contracts a cheap way to start? No. They are the most expensive place on the board in percentage terms. At a 0.07 coefficient a $0.10 entry costs 6.3% of stake against 0.7% at $0.90, a factor of nine. Add a one cent spread and a $0.10 round trip needs a 22.6% price move before it is level, against 1.8% at $0.95.

Why does my winning trade still lose money? Because you paid four costs and only counted one. On a real example above, 500 contracts bought at $0.30 and sold at $0.33 pay $7.35 entry fee, $7.74 exit fee and $5.00 of crossed spread against a $15.00 gross gain, finishing at minus $0.09 despite a 10% price move in the right direction.

Is Kalshi or Polymarket cheaper? It depends entirely on the category. Kalshi is 0.07 across the board. Polymarket is 0.07 for crypto, so identical, but 0.04 for politics and 0.00 for geopolitical. On a $0.50 entry, 100 shares, that is $1.75 on Kalshi against $1.00 on Polymarket politics and nothing at all on Polymarket geopolitical.

What is the fee at a $0.50 entry price? 1.75 cents per contract at a 0.07 coefficient. Stated as a percentage it is 1.75% of the $1.00 payout notional and 3.5% of your 50 cent stake. Both describe the same fee. Check which denominator a page is using before you compare its number to anything, because the two differ by 2x at exactly the price where fees peak.

Do makers pay anything? Polymarket charges makers 0.00 in all eleven categories, which is zero rather than a discount. Kalshi's help centre publishes 0.0175 while its newsroom says resting orders are fee-exempt, and the two have never been reconciled publicly. Either way the saving is payment for adverse selection, not free money: your resting order fills when you are wrong.

How do I check a fee figure I read somewhere? Ask what document it came from and when it was read. One competitor calculator has applied a flat 0.0625 coefficient for five months against Polymarket's documented 0.04 to 0.07 per category, which overstates a politics fee by 56.25% and invents $1.5625 per hundred shares in a category charged at zero. A fee number with no source and no date is a typo waiting to be found.