The Polymarket premium over Bitcoin options has closed
Polymarket lists daily contracts on whether Bitcoin and Ether close above a strike at noon Eastern, and Deribit's options price a near-identical event. On 2023 data, Victoria Portnaya's June 2026 study found Polymarket priced Bitcoin 5.6 to 12.5 points above the options.
Does Polymarket still price Bitcoin above the options market?
Every day Polymarket runs a ladder of contracts on where Bitcoin and Ether will close at noon Eastern. Each strike on the ladder is a contract that pays $1 if the coin’s price on Binance closes above that level at 12:00 ET, and nothing if it does not (Polymarket, 2026a; Polymarket, 2026c). Deribit sells options on the same two coins, settled on its own price indices, with an expiry every morning at 08:00 UTC (Deribit, 2026). A spread of two calls just either side of a strike, scaled by its width, pays about a dollar when the index finishes above that strike (Breeden and Litzenberger, 1978). Its price is the options market’s probability that the coin ends above the strike, the option-implied probability. The two venues therefore price a near-identical event, the same coin above the same strike, measured on Deribit’s index rather than on Binance and at an expiry the next morning or later. Between June 26 and September 21, 2026, Polymarket listed 84 daily events for each coin, with about eleven strikes to an event (Polymarket, 2026a).
A Polymarket contract pays $1, so a price of 62 cents is a 62% chance and a one-cent gap between the venues is one percentage point. A gap is worth trading only if it is wider than the cost of crossing the book, which is Polymarket’s taker fee plus the half-spread, the distance from the market price to the best offer (Polymarket, 2026b). The fee peaks at 1.75 cents a share at a price of 50 cents, and with an assumed one-cent half-spread on top, crossing the book there costs 2.75 cents. A premium has closed when it sits below the cost of crossing at every level of price. On 2023 data, Portnaya (2026) found Polymarket priced Bitcoin 5.58 points above Binance’s options on her main contract and 11.05 points above Deribit’s across three contracts.
By the summer of 2026 the premium had closed. Across 167 of those events, Polymarket priced Bitcoin 0.66 points above the options one day before settlement and 0.49 points five days before (Fig. 1). Both are less than a tenth of Portnaya’s 2023 gap on Deribit, and about a tenth of even her smallest figure, the 5.58 points on Binance. Ether had little premium to close: Portnaya measured 1.29 points on four Ether contracts in early 2023, and in 2026 Ether’s premium was under a point at both horizons (Portnaya, 2026). The two venues now price the same strikes within a point of each other on average, below the cost of crossing in every band, and neither forecasts the outcome better. The samples differ in kind: three monthly Bitcoin contracts in 2023, against daily ladders in 2026 whose 440 markets observed one day out traded $81.2 million over their lives.
Polymarket priced Bitcoin 11.05 points above Deribit's options in 2023 and less than one point above them in 2026, below the cost of trading the gap.
Polymarket's mean premium over the option-implied probability in 2026 (red dot), by asset and days before settlement, against Portnaya's 2023 figure for the same asset (ring), in percentage points; the label is the difference, and the dashed line is the 2.75-cent cost of crossing the book at a price of 50 cents.
The premium is about a point, even on the long shots
In 2023 the gap was widest on the long shots, the strikes the options gave the lowest chance. Portnaya’s regression finds the gap shrinking as the option-implied probability rises, and she reads the pattern as the favourite-longshot bias of betting markets, their habit of overpricing unlikely outcomes (Portnaya, 2026).
In 2026 the long shots carry about a point. Strikes the options gave less than a 20% chance traded 1.15 points above the options one day out and 0.62 points above them five days out (Fig. 2). The two middle bands carry premiums of 0.60 to 1.74 points. Above 80% the sign turns, and Polymarket priced the favourites a fraction of a point below the options, 0.35 points one day out and 0.21 points five days out.
Polymarket's premium is under two points in every band of probability, the long shots included.
Mean Polymarket price minus the option-implied probability, in percentage points, by band of the options' probability that the coin finishes above the strike, one day before settlement (red, 440 strikes) and five days before (grey, 963 strikes).
Single strikes stray little further. One day out, half of them sit within 1.46 points of the options and a third sit within one point. Five days out, half sit within 1.04 points and 48% within one point. Polymarket sat 5.6 points or more above the options, the smallest 2023 average, on about 5% of strikes one day out and 2% five days out.
What remains is a premium of under a point that still sits above zero. Over all strikes Polymarket averaged 0.79 points above the options one day out and 0.43 points five days out, and the 95% intervals on both, as on each coin alone, lie above zero (Table 4). In every band, at both horizons, that premium is smaller than the average cost of crossing the book at those prices, which runs from 1.42 to 2.56 cents. It stays smaller for any half-spread above 0.66 cents (Limitation 1). This force pushes the answer toward a closed premium: the wedge that 2023 found in the long shots has gone, and trading what is left one day out broke even (Table 2).
Neither venue forecasts better
A premium that carried information the options lacked would make Polymarket the better forecaster, and a premium that was only a bias would make it the worse one. On the same strikes the two cannot be told apart. The Brier score grades a forecast by the mean squared distance between its price and the outcome, 1 for YES and 0 for NO, so a lower score is better (Brier, 1950). One day out Polymarket scored 0.1202 and the options 0.1186. Five days out the scores were 0.0931 and 0.0933, and at both horizons the 95% interval on the difference includes zero (Table 1).
Table 1. Brier scores of Polymarket and the option-implied probability on the same strikes, by horizon and asset; lower is better.
| Horizon and asset | Strikes | Polymarket | Options | Difference, 95% interval |
|---|---|---|---|---|
| One day, both assets | 440 | 0.1202 | 0.1186 | +0.0016, from −0.0014 to +0.0046 |
| One day, Bitcoin | 240 | 0.1167 | 0.1149 | +0.0018, from −0.0022 to +0.0063 |
| One day, Ether | 200 | 0.1244 | 0.1230 | +0.0014, from −0.0024 to +0.0057 |
| Five days, both assets | 963 | 0.0931 | 0.0933 | −0.0002, from −0.0016 to +0.0013 |
| Five days, Bitcoin | 519 | 0.0934 | 0.0933 | +0.0001, from −0.0017 to +0.0022 |
| Five days, Ether | 444 | 0.0928 | 0.0934 | −0.0006, from −0.0023 to +0.0015 |
The two venues also missed in the same direction. Five days out, both priced every band below the share of strikes that settled YES, and their average prices sit within 1.21 points of each other in every band (Fig. 3). The miss was widest between 50% and 80%, where Polymarket sat 12.4 points below the share that settled YES and the options 13.7 points below it. Under 20% the two came closest, at 0.9 and 1.6 points below.
Five days before settlement, Polymarket and Deribit both priced every band below the share of strikes that settled YES, by almost the same amount.
Each venue's mean price minus the share of strikes that settled YES, in percentage points, by band of option-implied probability, over 963 strikes.
The miss came in a summer in which both coins rallied, Bitcoin by 36% and Ether by 67% on Deribit’s indices. Across all five-day strikes Polymarket’s average price was 49.9% and the options’ 49.5%, while 55.7% of strikes settled YES, so both venues missed the same rally by about six points. This force also pushes the answer toward a closed premium, because the gap that remains carries no measurable forecasting value in either direction.
What is left of the gap trades as a bet on direction
Trading what is left of the gap broke even one day out and lost money five days out. The rule takes the options as the fair price and holds every bet to settlement. It buys Polymarket’s YES when the option-implied probability sits above Polymarket’s price by more than the cost of crossing, and NO when it sits below by more than that cost. With the one-cent half-spread and Polymarket’s taker fee as the cost, it made 207 trades one day out and 285 five days out (Table 2).
Table 2. Returns from trading Polymarket toward the option-implied probability and holding to settlement, after the one-cent half-spread and the taker fee.
| One day | Five days | |
|---|---|---|
| Trades, on events and calendar weeks | 207, on 127 and 14 | 285, on 106 and 13 |
| YES and NO bets | 66 and 141 | 97 and 188 |
| Mean return per trade | +9.1% | −15.1% |
| 95% interval, events resampled | −13.6% to +36.9% | −28.4% to −1.3% |
| 95% interval, weeks resampled | −12.9% to +35.7% | −32.9% to +0.9% |
| Profit per dollar paid | +0.19% | −2.30% |
| YES bets: mean return and share won | +16.6% and 68.2% | +8.9% and 64.9% |
| NO bets: mean return and share won | +5.5% and 61.7% | −27.6% and 52.7% |
One day out the average trade returned 9.1%, with a 95% interval that runs from a loss of 13.6% to a gain of 36.9%. Weighted by what each trade cost, the rule made 0.19% per dollar paid, close to break-even. Five days out it lost 15.1% a trade and 2.30% per dollar paid. Resampling events puts that loss’s interval wholly below zero, at −28.4% to −1.3%, and the interval from resampling calendar weeks, which allows for overlapping windows, reaches +0.9%.
The five-day loss came from betting against the summer’s rally. Polymarket usually sat above the options, so the rule mostly bought NO, a bet that the coin finishes below the strike: five days out, two trades in three were NO bets. Those NO bets lost 27.6% on average, while the YES bets gained 8.9%. The trade is unhedged, so it carries the market’s direction along with the gap, and the option-implied probability missed that direction as well (Fig. 3).
Even with the market’s direction hedged away, the 2023 version of the trade was marginal. Portnaya’s delta-hedged proxy, which offsets each bet’s exposure to the market with spot Bitcoin, made 16 trades and earned a net alpha, the return left after the hedge, of 0.067 per unit of notional (Portnaya, 2026). With a p-value of 0.053, it just misses the usual 5% test of significance. This force pushes the answer toward a closed premium as well: one day out the gap pays close to nothing per dollar after costs, and five days out the result follows the trend of the market.
The premium may still live in contracts weeks from expiry
The strongest case that the premium survives rests on horizon. Portnaya’s 2023 Bitcoin contracts were monthly, settling at the end of August and the end of September 2023, and in her data the gap grew with time to expiry by about 1.9 points a day (Portnaya, 2026). She calls her sample intentionally narrow, at three Bitcoin contracts with an exact Binance option to match (Portnaya, 2026). Polymarket lists each daily event about a week before it settles, so five days out is close to the longest horizon these contracts allow (Polymarket, 2026a). The 2026 observations sit at the short end of the 2023 slope, where it puts the premium smallest. On this reading the premium lives at horizons beyond a week, which the daily contracts do not reach.
Portnaya measured that slope close to settlement, which lets the 2026 window test it. Her main contract was observed hourly over its last ten days, from September 19 to September 29, 2023 (Portnaya, 2026). At her slope, four more days to expiry would add about 7.7 points and put the five-day Bitcoin gap near 8.3 points. The measured five-day gap is 0.49 points, no larger than the one-day gap of 0.66 points, and the two intervals overlap (Table 4). The long shots, where the 2023 gap was widest, show the same: strikes the options put under 20% carry 1.15 points one day out and 0.62 five days out (Fig. 2).
A second reading traces the fall in the gap to a change of method rather than a change in the market. Swapping the option venue on her main contract, from Binance to Deribit, more than doubled Portnaya’s gap, from 5.58 to 12.48 points, and her paper does not give the horizon of the Deribit panel (Portnaya, 2026). In 2026 neither the yardstick nor the price it measures forecasts measurably better on the same strikes (Table 1), and five days out the two sit within 1.21 points of each other in every band (Fig. 3). A yardstick that forecasts as well as the price it measures leaves little room for a bias that makes Polymarket merely look close. Running the 2026 method on the 2023 contracts is the direct check (Limitation 7).
A premium that appears only beyond a week of horizon would fit both results. The gap on contracts listed weeks ahead is the first thing to watch.
What to watch
The premium would be back if it grew past the cost of trading it. From June to September the one-day gap averaged 0.79 points, under the 2.75 cents it costs to cross the book at 50 cents with the one-cent half-spread and the fee (Eq. 6). A calendar month of events from October 2026 on in which the one-day gap averages more than 2.75 points would reopen a trade near the middle of the book.
The other side’s case lives beyond a week of horizon, and Bitcoin contracts that trade more than a week before settlement would test it directly, if Polymarket lists them. A gap that grows by anything close to the 2023 slope of about 1.9 points a day would confirm it, and a gap that stays near one point would carry the one-day and five-day result to longer horizons.
Polymarket’s crypto taker fee rate stood at 0.07 on every market in the sample, and its fee page sets that rate by category (Polymarket, 2026b). Any change to the rate posted there moves the bar a gap has to clear: a cut would make a smaller premium tradable, and a rise would bury the one that remains.
The five-day loss came from NO bets during a rally (Table 2). A profit from the same rule over a month in which the Deribit Bitcoin index ends lower than it started would confirm the loss as a bet on direction. A loss in such a falling month as well would mean Polymarket’s premium carries information the options miss.
How we got the number
What the measurement used, and when
The sample is every Polymarket daily contract on Bitcoin or Ether above a strike that settled between June 26 and September 21, 2026 (Polymarket, 2026a; Polymarket, 2026c). Polymarket listed 84 such events for each coin in that window. Its public API returns no events for July 18, July 19, August 2 or August 3, queried on September 25, 2026, and REPRODUCIBILITY.md records the query. Polymarket’s public Gamma API supplied each market’s strike, fee schedule and settled outcome, and all market and trade data were retrieved on September 24, 2026. Each contract settles on the Binance one-minute candle that opens at 12:00 ET, 16:00 UTC in these months. Deribit’s options are European and cash-settled on Deribit’s own Bitcoin and Ether indices, with an expiry at 08:00 UTC every day (Deribit, 2026). Each strike is observed at 16:00 UTC one day and five days before settlement, one observation per event and horizon (Table 3).
The Polymarket price at an observation is the last point of the market’s public price history no more than 30 minutes before it, and the points used sit 9.4 to 9.9 minutes before the observation. The Deribit side is the coin’s option trades in the two hours up to the observation, or the most recent 1,000 of them when there were more, from Deribit’s public history API. That call returns at most 1,000 trades, newest first, and it hit the limit for 46 of 163 one-day events and 42 of 149 five-day events. In those capped windows the trades reached back as little as 15.9 minutes, with medians of 79.9 minutes one day out and 79.5 minutes five days out. Every trade used lies within 120 minutes of the observation.
Of those trades, only out-of-the-money trades on the first expiry after the Polymarket settlement that traded in the window are kept. The forward, F, is the Deribit index price at the latest trade of that expiry in the window. One day out, all 163 events used the 08:00 UTC expiry of the next morning, 16 hours past settlement. Five days out, 20 of the 149 events, with 116 strikes, used that next-morning expiry, and the other 129 used expiries 40 to 160 hours past settlement. The median lag at five days is 88 hours.
Table 3. The sample by horizon, from the strikes Polymarket listed to those kept; the three removal rows and the kept row sum to the 1,871 strikes listed at each horizon.
| One day | Five days | |
|---|---|---|
| Events listed | 168 | 168 |
| Strikes listed | 1,871 | 1,871 |
| Strikes with a price outside 0.02 to 0.98 | 1,382 | 669 |
| Strikes without a price within 30 minutes | 5 | 140 |
| Strikes in events without a usable smile, under four smile strikes | 44, in 4 events | 99, in 9 events |
| Strikes kept, Bitcoin and Ether | 440: 240 and 200 | 963: 519 and 444 |
| Events with no strike left after the price filters | 1 | 10 |
| Events kept | 163 | 149 |
| Settlement dates | June 26 to September 21, 2026 | June 30 to September 21, 2026 |
| Lifetime traded volume of the markets kept | $81.2 million | $127.2 million |
| Median smile strikes per event | 11 | 12 |
| Polymarket taker fee rate | 0.07 | 0.07 |
The two horizons together hold 1,403 strike rows on 167 events. The rally is measured on the Deribit index at the last option trade before the June 25 and September 20, 2026 observations: it went from 59,532.70 to 80,907.82 for Bitcoin, up 35.9%, and from 1,565.10 to 2,614.58 for Ether, up 67.1%. Portnaya’s five 2023 figures, four on Bitcoin and one on Ether, were transcribed by hand from arXiv 2606.19517 (Portnaya, 2026). Her regression of the gap has a coefficient of −0.398 on the option-implied probability and a slope of 0.0008 an hour on time to expiry, both with p < 0.001, and over 24 hours that slope comes to 1.92 points (Portnaya, 2026). Her delta-hedged proxy’s net alpha of 0.067 has an interval of −0.008 to 0.143 (Portnaya, 2026). Deribit’s contract specification, Polymarket’s contract rules and its fee page were retrieved on September 25, 2026 (Deribit, 2026; Polymarket, 2026a; Polymarket, 2026b; Polymarket, 2026c).
The calculation
Deribit options are priced with the formula of Black (1976) for a call on a forward, with no carry and no discounting (Eq. 1).
Each trade’s mark price, quoted in coin, is converted to dollars at the index price recorded with that trade. Inverting the call formula then gives the trade’s implied volatility, with that same trade index price as the forward and the time to the trade’s own expiry (Eq. 1). The smile for an event and horizon is the median implied volatility at each traded strike. Volatility at any strike K is linear between smile strikes and flat beyond the lowest and highest, and events with fewer than four smile strikes are dropped.
The option-implied probability D that the index finishes above strike K is a call spread of half-width h, $250 for Bitcoin and $20 for Ether (Eq. 2; Breeden and Litzenberger, 1978). Both calls in the spread use the forward F, the index at the latest trade of the chosen expiry in the window (Eq. 2). They are priced at the Polymarket horizon, with the smile’s volatility at each of their two strikes.
Here T is the time from the observation to the Polymarket settlement in years of 365.25 days. The probability D is clamped to the range 0.0005 to 0.9995, and 4 rows sit at the clamp. The gap on a strike is the Polymarket price q minus D (Eq. 3), in dollars per $1 contract, reported in cents, which equal percentage points. Strikes with a Polymarket price outside 0.02 to 0.98 are dropped.
The Brier score of N probabilities p against outcomes o, 1 for YES and 0 for NO, is their mean squared difference (Eq. 4). It is computed for Polymarket with p set to q and for the options with p set to D, on the same strikes (Brier, 1950).
Polymarket’s taker fee per share at price q scales with q times one minus q, at a rate r of 0.07 on every market in the sample (Eq. 5; Polymarket, 2026b).
The trade rule buys YES or NO when the option-implied probability clears the Polymarket price by the cost of crossing, c, made of an assumed half-spread s of one cent and the fee (Eq. 6).
At a price of 50 cents the fee is 1.75 cents and the cost of crossing is 2.75 cents (Polymarket, 2026b). The cost of crossing for a probability band is the mean of c over the band’s strikes, and the mean fee alone is the same average without s.
A YES bet pays the half-spread over q and the fee at that price, and a NO bet does the same on the other side of the book. Each bet is held to settlement, and its return R is per dollar paid, P (Eq. 7).
Profit per dollar paid sums each trade’s payout, o for a YES bet and one minus o for a NO bet, less the price paid, and divides by the total paid (Eq. 8).
Every interval resamples whole clusters with replacement 4,000 times, with seed 20260924, and reads the 2.5th and 97.5th percentiles of the pooled mean (Table 4). The cluster is the event, one asset on one date, because every strike of an event settles on the same price. The trade returns also carry an interval that resamples ISO calendar weeks of the event date. The probability bands are set on D: under 20% runs up to and including 0.2, 20% to 50% runs above 0.2 up to 0.5, 50% to 80% runs above 0.5 up to 0.8, and over 80% is everything above 0.8.
Table 4. Polymarket’s mean premium over the option-implied probability, with its 95% interval, by horizon and asset, in percentage points.
| Horizon and asset | Strikes | Mean premium | 95% interval |
|---|---|---|---|
| One day, both assets | 440 | 0.79 | 0.38 to 1.23 |
| One day, Bitcoin | 240 | 0.66 | 0.05 to 1.26 |
| One day, Ether | 200 | 0.95 | 0.35 to 1.67 |
| Five days, both assets | 963 | 0.43 | 0.23 to 0.63 |
| Five days, Bitcoin | 519 | 0.49 | 0.21 to 0.79 |
| Five days, Ether | 444 | 0.37 | 0.08 to 0.63 |
Limitations
-
An assumed half-spread. Polymarket prices are points of the public price history, so the one-cent half-spread is an assumption, and rebuilding the test on recorded bids and asks is the next step. The fee alone exceeds the mean gap in 5 of the 8 band-and-horizon cells, and in the other three the gap exceeds the fee by 0.17 to 0.66 cents. Any half-spread above 0.66 cents therefore keeps every band below the cost of crossing.
-
An approximate option-implied probability. The probability rests on the trades in the window, flat volatility beyond the traded strikes, the index as the forward, and the volatility of an expiry 16 hours past settlement at one day and up to 160 hours at five. The five-day result holds across expiries: 0.42 points (0.19 to 0.67) on the 20 events priced off the next-morning expiry, and 0.43 points (0.20 to 0.67) on the later ones. Pricing from quoted marks on several expiries, interpolated to the Polymarket settlement, is the next step.
-
Capped trade windows. One day out the gap is 0.53 points on the windows capped at 1,000 trades, with an interval of −0.54 to 1.76 over 149 strikes, and 0.92 on full windows, with 0.62 to 1.24 over 291. Five days out the two are 0.46 points (−0.07 to 0.99) and 0.42 (0.23 to 0.58), and the intervals overlap at both horizons. Paging the API for every trade in the two hours is the next step.
-
Two reference prices. Polymarket settles on the Binance BTC/USDT and ETH/USDT close, Deribit on its own index, and USDT is a separate asset from the dollar (Polymarket, 2026a; Polymarket, 2026c; Deribit, 2026). Recording the Binance close and the Deribit index at each settlement would measure that basis.
-
Scope in time. The measurement covers one and five days before settlement, one observation per event and horizon, from late June to September 2026, a window in which Bitcoin rose 35.9% and Ether 67.1%. Adding the horizons between, more observations per day and a window that includes a falling market is the next step.
-
Overlapping windows. Five-day windows of neighbouring events overlap, so their outcomes share one price path. The trade test carries the interval that resamples calendar weeks beside the one that resamples events for that reason.
-
A different method in 2023. Portnaya’s figures come from repeated observations of three Bitcoin and four Ether contracts, with the option-implied probability taken from single calls under Black-Scholes (Portnaya, 2026). The 2026 figures use call spreads on the Deribit smile at two horizons, and running the 2026 method on the 2023 contracts would put both years on one method.
This is an exploratory analysis, and no hypothesis was registered before it.
Reproducing it
The analysis runs offline in Python 3.12.3 and otherwise uses only the standard library. It rebuilds every option-implied probability from the smile and checks that it matches the stored value to within 10⁻¹². It then checks the extraction parameters against the data for every event, including the price staleness, the two-hour window and the 1,000-trade limit. On September 25, 2026 it ran top to bottom on a clean kernel with no errors.
The per-strike table was built from three public sources. Polymarket’s Gamma API supplied the markets, strikes, fee schedules and outcomes, and its CLOB price history, /prices-history, supplied the prices. The option trades come from Deribit’s history API, get_last_trades_by_currency_and_time on history.deribit.com, at up to 1,000 trades per call.
References
- Black, F. (1976). The pricing of commodity contracts. Journal of Financial Economics, 3(1-2), 167-179.
- Breeden, D. T., and Litzenberger, R. H. (1978). Prices of state-contingent claims implicit in option prices. Journal of Business, 51(4), 621-651.
- Brier, G. W. (1950). Verification of forecasts expressed in terms of probability. Monthly Weather Review, 78(1), 1-3.
- Deribit. (2026). Inverse Options. Deribit Support, contract specification; retrieved September 25.
- Polymarket. (2026a). Bitcoin above ___ on September 10? Event page and resolution rules, one of the daily Bitcoin events in the sample; retrieved September 25.
- Polymarket. (2026b). Fees. Polymarket Documentation; retrieved September 25.
- Polymarket. (2026c). Ethereum above ___ on September 10? Event page and resolution rules, one of the daily Ether events in the sample; retrieved September 25.
- Portnaya, V. (2026). Do prediction markets match option prices? Bitcoin threshold evidence from Binance and Polymarket. arXiv:2606.19517, June 17.