Key takeaways
- Calibration in aggregate is not a usable probability for one position. Kalshi's study of 2.2 million resolved markets found that prices behave broadly like probabilities: contracts priced at 70¢ resolved YES about 70% of the time. A working paper on more than 300,000 Kalshi contracts finds a persistent favorite–longshot pattern, and that liquidity takers lose more than makers. Both findings describe averages. An institution sizing one position needs that contract's price at its own size, after costs, under that contract's terms.
- The headline describes the first few thousand dollars. At 08:23 UTC on 31 October 2026, Kalshi's December FOMC "no change" contract showed an ask of 30¢. It had about $1,000 of contracts within a point of that price. $10,000 would have filled at an average of 38.2¢, and $100,000 at 68.0¢. On Polymarket International, the equivalent contract filled $100,000 at 30.0¢ but $250,000 at 45.9¢. Depth also moves: thirteen minutes earlier, $1 million of the Senate contract on Polymarket would have cost 73.5¢ on average. At 08:23 it would have cost 80.8¢.
- Fees bite hardest at the extremes. Kalshi and Polymarket International both charge takers a fee of rate × price × (1 − price) per contract. As a share of the premium paid, that is rate × (1 − price): 6.3% on a 10¢ contract and 0.7% on a 90¢ contract at a 0.07 rate. Whether a fee applies is set market by market. The four Polymarket contracts in our snapshot were flagged as fee-free.
- A contract is a zero-coupon claim, and its coupon date can move. At a 4% cost of capital, a contract certain to pay $1 in two years is worth about 92.5¢, so a price of 88¢ implies about 95%. Venue interest programs offset part of this for eligible holders: Kalshi pays a variable 3.50% APY, and Polymarket pays a 3.25% holding reward on eligible markets. The settlement date itself can be uncertain. It can be pulled forward by early-settlement clauses, or pushed back by disputes or runoffs.
- Put together, a 70.5% headline becomes a breakeven of roughly 72% to 74% for a $250,000 buyer of Senate control, depending on venue, fees and carry. That is before allowing for differences in terms or resolution risk. We propose that every price an institution relies on carries its provenance: venue, contract and terms version, side, size, time, fee basis and expected settlement date.
1. Introduction: a 20% chance is not a 0% chance, and a 70¢ price is not 70%
On 25 October 2026, Luiz Inácio Lula da Silva won Brazil's presidential runoff. Before the vote, prediction markets had priced his opponent as a 75–80% favorite. Critics called it a miss. Galaxy Research replied that a 20% outcome is expected to happen about one time in five. It pointed to Kalshi's calibration study as evidence that, across many markets, prices behave like probabilities.1
We agree with that reply, and this paper is about the next question. Suppose prices are well calibrated in aggregate. Can an institution take the price of one contract as the probability it needs for a hedge, a risk model or a research note? Usually it cannot, at least not without six adjustments. Each is small on a quiet contract at small size. Together they can be large, and they are largest where institutions most want to use these markets: macro and political contracts, at size, far from expiry.
The adjustments are:
- which number counts as the price (the quote);
- how much can be traded near it (size);
- what it costs to trade (fees);
- when the claim pays (time);
- what the claim pays on (terms);
- how it gets decided (resolution).
A note on who is on the other side follows them.
2. What calibration studies do and don't tell you
Two findings frame the evidence:
- Kalshi's calibration study. It covers more than 2.2 million resolved markets. Accuracy improves as resolution approaches: Brier scores fall from roughly 0.08–0.09 three months out to about 0.02 at close. Even in well-calibrated election markets, favorites won about 90% of the time across most horizons, so roughly one favorite in ten lost.12
- Bürgi, Deng and Whelan. Their working paper uses transaction-level data on more than 300,000 Kalshi contracts. They find that makers, who post resting orders, earn higher returns than takers, who cross the spread. Both groups' returns show a favorite–longshot pattern: low-priced contracts win less often than their prices imply. Prices become more accurate as contracts approach close.3
These results are consistent with an older literature:
- Under some conditions, prediction-market prices approximate the mean belief of traders, but not in general.45
- Mispriced longshots are a durable feature of markets for contingent claims.6
- Prediction markets are reasonably well calibrated close to expiry but significantly biased for events further out, with a favorite–longshot pattern that grows with time to expiration.7 In our reading, the carry effect described in Section 6 is one plausible contributor.
Calibration describes a population of contracts at their quoted prices. A position is one contract, entered at a price the market actually offers, for a size the book can bear, under the contract's own terms. Each step below moves from the population toward the position.
3. Adjustment 1: the quote
"The price" of an event contract can mean the last trade, the best bid, the best ask or the midpoint. These differ, sometimes by a lot. Venues display different ones. Polymarket's documentation says its interface shows the midpoint of the best bid and best ask as the market's implied probability. When the spread is wider than 10¢, it shows the last traded price instead.8 A buyer pays the ask. A seller receives the bid. A model fed the midpoint assumes a trade nobody is offering.
Predict Here's methodology uses the midpoint only for a fresh, valid, two-sided book, and calls it a market price rather than a forecast.9 The first rule for an institutional reader is simple: name the price you are using, and for a trade, use the side you would actually have to take.
4. Adjustment 2: size
A quote describes the top of the book. Behind it are further price levels, each with a finite quantity. To estimate what a given size actually costs, we walked the visible ask side of four contracts on two venues. The contracts were Senate and House control for the 3 November midterms, plus two outcomes of the December FOMC decision. We bought the YES side with fixed amounts of cash.
| Contract | Venue | Best bid / ask | $10k | $100k | $250k | $1m |
|---|---|---|---|---|---|---|
| Senate 2026: Democratic control | Kalshi | 70¢ / 71¢ | 71.0¢ | 71.1¢ | 71.8¢ | 82.1¢ |
| Polymarket Intl. | 70¢ / 71¢ | 71.0¢ | 71.9¢ | 72.0¢ | 80.8¢ | |
| House 2026: Democratic control | Kalshi | 93.7¢ / 93.8¢ | 93.8¢ | 94.0¢ | 94.3¢ | 95.8¢ |
| Polymarket Intl. | 94¢ / 95¢ | 95.0¢ | 95.0¢ | 95.0¢ | 95.0¢ | |
| FOMC Dec 2026: 25 bp hike | Kalshi | 69¢ / 70¢ | 70.6¢ | 86.4¢ | 91.8¢ | 96.3¢ |
| Polymarket Intl. | 69¢ / 70¢ | 71.0¢ | 71.0¢ | 71.1¢ | 80.2¢ | |
| FOMC Dec 2026: no change | Kalshi | 29¢ / 30¢ | 38.2¢ | 68.0¢ | 81.1¢ | 91.7¢ |
| Polymarket Intl. | 27¢ / 29¢ | 29.0¢ | 30.0¢ | 45.9¢ | 76.6¢ |
Snapshot of public order books at 08:23:45 UTC on Saturday 31 October 2026. Kalshi YES asks are derived from NO bids in the Trade API v2 order book; Polymarket asks come from its CLOB order-book endpoint. Cash is spent across visible ask levels in price order. The figures exclude fees, hidden or replenishing liquidity, routes through other outcomes in the same event, and any market movement. The snapshot was taken outside US trading hours, when market makers may quote less depth. It illustrates the method and is not a measure of either venue's typical liquidity.
Four features stand out.
The headline can describe very little money. Kalshi's December "no change" contract showed a 30¢ ask, but only about $1,000 of contracts sat within a point of it. Spending $10,000 would have cost 38.2¢ a contract on average, and $100,000 would have cost 68.0¢. In other words, the 30¢ headline was a price for a few thousand contracts. At $100,000, the trade breaks even only if the outcome is more than 68% likely.
Depth near the touch matters more than total depth. Kalshi's Senate contract showed $13 million of visible asks in total. Most of it sat far from the best price, so $1 million would have averaged 82.1¢ against a 71¢ ask. Predict Here's methodology measures capacity within fixed bands of the touch, for this reason.9
| Contract | Kalshi: within 1 / 2 / 5 points | Polymarket Intl.: within 1 / 2 / 5 points |
|---|---|---|
| Senate 2026: Democratic control | $211k / $235k / $309k | $384k / $497k / $524k |
| House 2026: Democratic control | $289k / $418k / $1.02m | $6.29m / $6.33m / $8.05m |
| FOMC Dec 2026: 25 bp hike | $12k / $13k / $18k | $221k / $264k / $573k |
| FOMC Dec 2026: no change | $1k / $1k / $2k | $94k / $97k / $103k |
Same snapshot and method as Exhibit 1. The bands follow the ±1, ±2 and ±5 point capacity bands in Predict Here's published methodology, applied to the ask side.
Neither venue is uniformly cheaper. In this snapshot, Polymarket International offered more depth near the touch on all four contracts. Yet for House control, Kalshi was the cheaper venue at every size up to $500,000, because its ask started at 93.8¢ against Polymarket's 95¢. Polymarket became cheaper only at $1 million, at 95.0¢ against 95.8¢. Depth and cost depend on the contract, the size, the hour and the day. Neither is a fixed property of a venue.
Depth moves. Thirteen minutes earlier, at 08:10 UTC, $1 million of Polymarket's Senate contract would have averaged 73.5¢. At 08:23 the same purchase averaged 80.8¢, because a large offer had been taken or withdrawn. A single snapshot describes one moment. A policy for trading at size needs repeated measurement.
5. Adjustment 3: fees
Both major venues now charge takers a fee proportional to price × (1 − price):
- Kalshi. The general schedule charges takers 0.07 × contracts × price × (1 − price), rounded up to the next cent per order.10 Fee type and multiplier are set by series. On 31 October 2026, 163 of Kalshi's 15,579 series also charged makers, including the FOMC decision series. Nineteen carried a half-rate multiplier and fourteen were fee-free.11
- Polymarket International. Its documentation gives the same functional form, with category rates of 0.04 to 0.07. Geopolitical and world-event markets are fee-free, combos have their own rate, and makers pay nothing.12 Category tables are only a guide: the four Polymarket contracts in our snapshot were flagged fee-free in Polymarket's API.
Because the fee per contract is rate × p × (1 − p), the fee as a share of the premium paid is simply rate × (1 − p). Cheap contracts carry far heavier fees than expensive ones.
| Price | Fee per contract at rate 0.07 | Rate 0.07 (Kalshi general; Polymarket crypto) | Rate 0.05 | Rate 0.04 | Fee-free markets |
|---|---|---|---|---|---|
| 5¢ | 0.33¢ | 6.65% | 4.75% | 3.80% | 0% |
| 10¢ | 0.63¢ | 6.30% | 4.50% | 3.60% | 0% |
| 25¢ | 1.31¢ | 5.25% | 3.75% | 3.00% | 0% |
| 50¢ | 1.75¢ | 3.50% | 2.50% | 2.00% | 0% |
| 75¢ | 1.31¢ | 1.75% | 1.25% | 1.00% | 0% |
| 90¢ | 0.63¢ | 0.70% | 0.50% | 0.40% | 0% |
| 95¢ | 0.33¢ | 0.35% | 0.25% | 0.20% | 0% |
Fee as a share of premium = rate × (1 − price). Kalshi rounds fees up to the cent per order, which raises the effective rate on small orders. Check each market's current fee flag before use.
The effect on breakeven is direct. A taker buying a 10¢ contract at a 0.07 rate needs the event to be about 10.6% likely to break even, not 10%. That is a 6% uplift in the required probability, before any other adjustment. At 90¢, the uplift is under 1%. Fee drag is one mechanical reason longshots look expensive to takers.
6. Adjustment 4: time
An event contract pays at settlement and nothing before. Economically it is a zero-coupon claim, so its price reflects both the probability of the event and the cost of capital tied up until settlement. Suppose a holder's opportunity cost is c per year, the contract settles in t years and no interest is paid on the position. Ignoring risk premia, the breakeven probability is roughly price × (1 + c)^t.
| Price | 3 months | 1 year | 2 years |
|---|---|---|---|
| 10¢ | 10.1% | 10.4% | 10.8% |
| 50¢ | 50.5% | 52.0% | 54.1% |
| 88¢ | 88.9% | 91.5% | 95.2% |
| 95¢ | 95.9% | 98.8% | above 100% |
The 4% rate is an illustrative assumption, not a forecast. At 4%, a certain $1 due in two years is worth about 92.5¢. A price above that on a two-year contract is consistent only with a lower cost of capital, or with interest earned on the position.
Two features of the current market change the arithmetic:
- Interest programs. Kalshi pays a variable interest rate, 3.50% APY at the time of writing, on cash and open positions for eligible US users with an average balance of at least $250.13 Polymarket pays a variable 3.25% annualized holding reward on eligible markets, funded from its treasury.14 For an eligible holder these programs offset most of the carry. A holder who is ineligible, or whose capital costs more, does not get that offset. A price compressed by carry still looks like a lower probability than the market actually assigns.
- Uncertain settlement dates. Kalshi's House and Senate control contracts expire at 10:00 a.m. ET on 1 February 2027, but they can settle much earlier: once at least four of eight named media desks call the result and none contradicts it. Polymarket's can wait for runoffs, and for independents to announce their caucus until 4 January 2027.15 Section 4 of our companion paper shows that a disputed Polymarket market takes a median of about four days to settle, against hours for an undisputed one.16 For risk purposes, time-to-cash is a distribution, not a date.
7. Adjustment 5: terms
A price is the market's valuation of a specific contract, not of the question in its title. Three kinds of difference matter most:
- Definitions. Kalshi's Senate control contract pays on the party of the President pro tempore, as recorded on congress.gov at 10:00 a.m. ET on 1 February 2027. Polymarket's pays on the party that wins a majority of voting seats, under its own counting rules.15 The two usually agree. They can split in edge cases, such as a 50–50 chamber, a member who switches parties, or a delayed runoff.
- Carve-outs. When Iran's Supreme Leader died on 28 February 2026, Kalshi's "out before March 1" contract settled at $0.02 under a death carve-out. Polymarket's "out by February 28" paid $1.00.16 The price of a contract with a carve-out is the price of a payoff that is not purely binary.
- Fallbacks. About one in five resolved markets in our Polymarket sample carried an equal-split fallback, under which both sides receive $0.50 if the market cannot be resolved.16
None of this makes a price wrong. It makes the price conditional. A probability read from one venue's contract is a probability of that contract's payout condition, which may not be the event the institution cares about.
8. Adjustment 6: resolution
Contracts are also priced for the chance that their resolution is contested, delayed or decided on a reading the buyer did not expect. In our companion study of 96,414 resolved Polymarket International markets, disputes touched 0.36% of markets by count. They touched 20.3% of non-sports volume, and 41% of non-sports markets that traded more than $50 million.16 At regulated venues, outcomes are determined by the exchange. Its rulebook can allow terms to be modified, sources to be substituted, or a committee to allocate value in contingent cases.16
We do not offer a universal haircut for resolution risk. Any honest estimate depends on the wording of the contract and the history of its adjudicator. But an institution should ask the question for every contract, and it should give the answer more weight the more value a contract carries and the more its wording depends on interpretation.
9. A note on who is on the other side
Bürgi, Deng and Whelan find that makers, the participants who post resting orders, earn higher returns than takers on Kalshi.3 Venue programs that pay makers widen that edge: Polymarket rebates fees to makers, and Kalshi runs incentive programs for liquidity providers. An institution that crosses the spread to build a position is a taker by definition. Its realized price will on average sit on the unfavorable side of whatever fair value the market eventually reveals. That is not a reason to avoid these markets. It is a reason to measure the full cost of entry, which is the subject of everything above.
10. Putting it together
Here is the full adjustment stack for one contract: a $250,000 purchase of Democratic Senate control on each venue, using the snapshot in Exhibit 1. This is an illustration of method, not a view on the outcome or a recommendation.
| Step | Kalshi | Polymarket International |
|---|---|---|
| Headline (midpoint) | 70.5% | 70.5% |
| Best ask | 71.0¢ | 71.0¢ |
| Average price for $250,000 (Exhibit 1) | 71.8¢ | 72.0¢ |
| Taker fee per contract | 1.4¢ (0.07 × p × (1 − p)) | 0¢ (market flagged fee-free) |
| All-in cost per contract | 73.2¢ | 72.0¢ |
| Latest settlement | 2 Feb 2027; earlier if media desks call the result | When the seat count is determinable; can wait until 4 Jan 2027 |
| Carry at 4%, no interest, to the latest date | +0.7 points | +0.5 points |
| Carry with venue interest offsetting most of it | +0.1 points (3.50% APY, eligible users) | Depends on market eligibility for holding rewards |
| Breakeven probability | about 73.3–74.0% | about 72.0–72.5% |
| What the contract measures | Party of the President pro tempore on congress.gov, 1 Feb 2027 | Party winning a majority of voting seats, under Polymarket's rules |
| Who decides | Exchange under its CFTC-filed rules | Proposal and challenge process, then a UMA token-holder vote if disputed twice |
Figures are rounded. Carry runs from 31 October 2026 to the latest stated settlement point: 94 days to 2 February 2027 for Kalshi, and 65 days to 4 January 2027 for Polymarket. Resolution risk is reported qualitatively; see the companion paper.
At $250,000, a 70.5% headline becomes a breakeven of about 72% on one venue and about 73–74% on the other. The gap between the venues comes mostly from fees, not from the order book. The two contracts also pay on different definitions, and different adjudicators decide them. A risk model that records "Senate D = 70.5%" has lost every one of those facts.
11. Our take
Use prices as evidence, not as answers. Event-contract prices aggregate information well, and the evidence on calibration is encouraging. But the number on the screen is a quote for a small size, before costs, under one contract's terms and one adjudicator's process. Institutions should convert it into a breakeven at their own size before treating it as a probability.
Report every price with its provenance. We suggest a minimum record for any event-contract price used in a decision, a risk report or published research:
- venue, and contract ID with its terms version and capture time;
- side (bid, ask or midpoint), and the time the book was observed;
- size, the average price at that size, and any shortfall;
- fee basis (taker or maker, rate, and any market-specific flag);
- the expected settlement window, including any early-settlement and dispute provisions;
- known carve-outs and fallbacks;
- the adjudication route.
This is what Predict Here's methodology calls provenance. It is also why our platform reports price at fixed sizes from $10,000 to $1,000,000. When the visible book cannot fill a size, it says "insufficient visible depth" rather than extending the last price.
Expect the adjustments to matter more, not less. As institutional demand pushes toward larger positions in macro, policy and political contracts, the gap between the headline and the breakeven at size widens. Those contracts are also where disputes cluster and terms diverge. The markets are becoming more useful, and their prices are becoming easier to misuse.
Methodology, data and limitations
Order books. We took snapshots through Kalshi's Trade API v2 order-book endpoint and Polymarket's CLOB order-book endpoint at 08:10:43, 08:12:20 and 08:23:45 UTC on 31 October 2026. Exhibits 1, 2 and 5 use the 08:23:45 snapshot. The 08:10 snapshot is cited in Section 4 for comparison.
Contracts. On Kalshi: CONTROLS-2026-D, CONTROLH-2026-D, KXFEDDECISION-26DEC-H25 and KXFEDDECISION-26DEC-H0. On Polymarket International: the YES tokens of "Which party will win the Senate in 2026? (Democratic Party)", "Which party will win the House in 2026? (Democratic Party)" and "Fed Decision in December?" (25 bps increase; no change).
Walk. We derived Kalshi YES asks as 1 − NO bid. Polymarket's order book for each outcome token already includes complementary orders. Cash was applied level by level in price order. Fees, latency, queue position, hidden liquidity, replenishment, and synthetic routes through other outcomes in the same event are all excluded.
Fees. From each venue's published fee documentation, and each market's fee flag where available. Carry. Computed as price × (1.04)^t, an illustration only. Calibration. Cited from published research, not re-estimated.
Limitations.
- The snapshot was taken early on a Saturday, outside US trading hours. Market makers may quote less depth then.
- Results for four contracts at one moment do not generalize to venues or categories.
- Breakeven figures ignore risk premia, taxes and the value of optionality.
Reproducibility. The scripts and derived results are available on request.
Notes and references
- Galaxy Research, "What's the Difference Between a 20% Chance and a 0% Chance?" (30 October 2026). https://www.galaxy.com/insights/research/whats-the-difference-between-a-20-chance-and-a-0-chance ↩
- Kalshi Research, calibration study of resolved markets. https://kalshi.com/research/kalshi-research-calibration.pdf (figures as summarized by Galaxy Research). ↩
- Constantin Bürgi, Wanying Deng and Karl Whelan, "Makers and Takers: The Economics of the Kalshi Prediction Market," CESifo Working Paper 12122 (2025); CEPR Discussion Paper 20631; GWU Working Paper 2026-001. ↩
- Charles F. Manski, "Interpreting the Predictions of Prediction Markets," Economics Letters 91(3), 2006, 425–429. ↩
- Justin Wolfers and Eric Zitzewitz, "Interpreting Prediction Market Prices as Probabilities," NBER Working Paper 12200 (2006); see also "Prediction Markets," Journal of Economic Perspectives 18(2), 2004, 107–126. ↩
- Erik Snowberg and Justin Wolfers, "Explaining the Favorite–Long Shot Bias: Is It Risk-Love or Misperceptions?" Journal of Political Economy 118(4), 2010, 723–746. ↩
- Lionel Page and Robert T. Clemen, "Do Prediction Markets Produce Well-Calibrated Probability Forecasts?" The Economic Journal 123(568), 2013, 491–513. ↩
- Polymarket documentation, "Prices and order book." https://docs.polymarket.com/concepts/prices-orderbook ↩
- Predict Here, Methodology (market-implied probability; executable liquidity and capacity bands). https://predicthere.com/methodology ↩
- Kalshi fee schedule. https://kalshi.com/docs/kalshi-fee-schedule.pdf ↩
- Kalshi Trade API v2,
/series(fee_type,fee_multiplier), retrieved 31 October 2026. ↩ - Polymarket documentation, "Fees." https://docs.polymarket.com/trading/fees ↩
- Kalshi Help Center, "APY on Kalshi." https://help.kalshi.com/en/articles/13823847-apy-on-kalshi ↩
- Polymarket Help Center, "Holding Rewards." https://docs.polymarket.com/polymarket-learn/trading/holding-rewards ↩
- Predict Here, "Midterms 2026: what 'control' of the Senate and House means on each venue" (terms read 27 October 2026). https://predicthere.com/insights/midterms-2026-what-control-means ↩
- Predict Here Research, "Same Question, Different Contract" (companion paper, 31 October 2026). https://predicthere.com/insights/same-question-different-contract ↩
Disclosures
This paper is general research published for information only. It is not investment, legal or tax advice. It does not recommend any position, venue or strategy, and is not tailored to any reader's circumstances. The worked example in Exhibit 5 illustrates a method and expresses no view on any outcome. Predict Here is operated by Odyssea Research LLC. Predict Here does not list markets, hold client funds, take the other side of trades, or execute or route orders.
Order-book figures are derived calculations from public interfaces at the times stated. They describe those moments only.
Persons associated with Predict Here may hold positions in event contracts, including on venues discussed in this report.