Kalshi and Polymarket: October Fed hike odds, from 66.5% to 33.5%

The Kalshi contract that pays one dollar if the Federal Reserve raises rates on October 28, 2026, dropped from 0.67 to 0.46 in a single hour, between 18:00 and 19:00 UTC on September 29, 2026, with 310,742 contracts traded. On Polymarket, the same outcome went from 66.5% on September 25 to 33.5% on September 30. Of those 33 probability points, 30 were lost during two hours of trading separated by one night, and both started at the exact same minute on both platforms: at 11 and 35 seconds past 18:00:00 UTC on September 29, and at 8 and 14 seconds past 12:30:00 UTC on September 30. The first hour created 189,170 contracts of open interest (live, unclosed positions) on Kalshi; the hour following the second one destroyed 99,965. This article reconstructs the two bursts using second-by-second trade tape, measures what did not move (December), rules out four alternative explanations with data, distinguishes "delay" from "cancellation" on the Treasury curve, and explains why that crossover invalidated a claim we made on September 25 in a working note that was never published.

How did Kalshi and Polymarket strip 33 points from the October rally in two hours?

Between September 25 and 30, 2026, the Polymarket contract "Will the Fed increase interest rates by 25 bps after the October 2026 meeting?" dropped from 0.665 to 0.335, while the equivalent Kalshi contract (KXFEDDECISION-26OCT-H25) fell from 0.68 to 0.35. These are two distinct markets, with two different order books and two separate audiences, yet the movement in both was concentrated within two sixty-minute windows.

The first window is the hour from 18:00 to 19:00 UTC on September 29, 2026, which coincides with the scheduled start of the event where John C. Williams, President of the Federal Reserve Bank of New York, read a text titled "Unwavering Dedication" at the University of Buffalo. The second is the hour from 12:00 to 13:00 UTC on September 30, 2026, which contains the exact minute the U.S. Bureau of Economic Analysis (BEA) lifts the embargo on its personal income and outlays report. That report brought a core PCE price index —the inflation indicator monitored by the central bank, personal consumption expenditures excluding food and energy— of +0.2% monthly compared to +0.1% in July.

The price convention changes the calculation, so it must be stated: the 0.67 from the first burst is the last price printed on Kalshi before 18:00:00 UTC, at 17:57:22, while the candle from 18:00 to 19:00 opened at 0.66 — 21 probability points of a drop with the first and 20 with the second. This article uses the last previous price in both bursts: 21 points and 13.

Before reading the table, it is useful to know what each box measures, as both platforms use the same word for different things:

• Kalshi's volume is in contracts, not dollars. The literal field in its API is volume_fp and it showed 1,963,776.90 in the October contract on October 1, 2026: cumulative contracts since the market opened. To convert it to dollars, you must multiply by the price of each trade, and then the figure applies to that specific trade and not the market as a whole.
• Polymarket's volume is indeed in dollars. The volume field for the October event showed 20.25 million dollars on October 1, 2026. The two magnitudes cannot be added together across platforms.
• Open interest is not volume. Volume counts every buy and sell; open interest (open_interest_fp on Kalshi) counts the active contracts that no one has closed. An hour can see 213,908 contracts traded and leave 99,965 fewer active contracts than at the start, meaning most of that volume was people unwinding their positions.
• The price of a binary contract is not a clean probability. It is what someone paid for a ticket worth one dollar if the outcome occurs, factoring in the bid-ask spread, platform fees, and the cost of locking up capital until October 28. On October 1, 2026, at 13:30 UTC, that Kalshi contract had the best bid at 0.35 and the best ask at 0.37: the "probability" at the same instant changes by two points depending on which one you look at.

With that in mind, the table shows something specific: on September 29, 2026, the price of YES on Kalshi lost 21 probability points in sixty minutes with 310,742 contracts, and on September 30, it lost another 13 with 156,067. Dividing these figures —our own calculation— results in 14,797 contracts per point in the first burst and 12,005 in the second.

Why did Polymarket and Kalshi open the same burst 24 seconds apart?

The tape is the record of every trade with its timestamp, and both platforms publish it. Read second by second, the correlation is tight and symmetrical.

In the first burst, Polymarket printed the first trade at 18:00:11 UTC on September 29, 2026 (a purchase of 14 NO shares at 0.33 — YES at 0.67) and Kalshi at 18:00:35.36: 24 seconds later. Before that, there was silence on both. The last price printed on Kalshi was at 17:57:22.19 and the last on Polymarket at 17:56:30, leaving three minutes and thirteen seconds without a single trade on one platform and three minutes and forty-one seconds on the other. The three largest individual executions of that hour on Kalshi were three of 50,000 contracts each, all with the aggressor on NO: at 18:09:18 with YES at 0.51; at 18:12:07 at 0.53 and at 18:27:33 at 0.47.

In the second burst, the order is reversed. Kalshi printed at 12:30:07.99 UTC on September 30, 2026 —25 contracts, with the YES dropping from 0.45 to 0.44— and Polymarket at 12:30:14: six seconds later. Twenty-two seconds after the embargo, fourteen after the first Kalshi trade, at 12:30:21.99 a NO aggressor took 50,000 contracts with the YES at 0.40 and at 12:32:45 the contract printed 0.29.

The fact that the order is reversed rules out the most convenient explanation, which is that one platform watches the other and copies it: neither of the two is always in the lead. What remains is that both order books react to the same source with latencies of a handful of seconds, and that this source has a schedule.

The time of the first burst arrives with a limit that is best left intact. The Federal Reserve Bank of New York published a media advisory on September 18, 2026, framing the Buffalo event as follows: «Date: Tuesday, September 29, 2026 — Time: 2:00 PM EDT — Location: University at Buffalo (North Campus) Student Union Theater», and the host university repeats it: «When: 2 p.m. on Tuesday, Sept. 29». 14:00 EDT is 18:00:00 UTC. What neither publication discloses is the exact moment Williams uttered the phrase nor the moment the bank posted the text: the speech page lacks a timestamp, opens with «Good afternoon» and is marked «As prepared for delivery». The measured coincidence aligns with the time the event was scheduled eleven days prior, not with a delivery timestamp. The second burst does not carry such doubt: the 8:30 AM Eastern embargo is printed directly on the BEA report itself.

What happened on Polymarket to the figure we wrote on September 25 regarding "3 hikes in 2026"?

The Polymarket market "how many Fed rate hikes will there be in 2026?" no longer pays for the path of two more hikes at double the weight that path carries in the Federal Reserve committee: as of October 1, 2026, it pays 0.82 times that weight. The reading of 1.85 times that we measured on September 25, 2026, ceased to be valid on September 30 at 12:30 UTC, and the measurement that invalidates it is ours.

That reading came from a working note dated September 25, 2026, which was never published, and the note contained, in figures and with a date, the number that would leave it without a basis. The full condition, as drafted, is this: "if the '3 hikes' market falls below 25% without Fed communication in between (converging only to the dots), the divergence supporting the headline has disappeared and the piece is not published."

That condition was not met as written, and it must be said before anything else: the 25% threshold was not crossed through spontaneous convergence. It was crossed following a Federal Reserve speech on September 29 and the BEA price report on the 30th, which are precisely the "intervening Fed communication" and the data point that the clause excluded. The gap between what the market was paying and what the committee projected disappeared nonetheless —the "3 hikes in 2026" outcome stood at 41.0% on September 25 at 08:41 UTC and at 17.85% on the 30th at 13:28, with the Polymarket hourly series at 0.283 at 12:00 UTC and 0.180 at 13:00— but it disappeared via a path the condition did not contemplate. The figure was set to distinguish between "the market correcting itself" and "the market reacting to something that happened," and what occurred was the latter. On October 1, at 13:27 UTC, the same outcome marked 0.1815 — with a daily high of 0.2205 at 10:00 UTC.

The Fed's side did not move at all during those five days: Figure 2 of the economic projections from the rate-setting committee (FOMC, the Federal Open Market Committee), published on September 16, 2026, places 4 of the 18 participants at a rate of 4.375% at the close of 2026, which represents the path of two more hikes, and 12 of the 18 at 4.125%, the path of just one more. Those 4 out of 18 represent 22.2%, and the ratio between what the market was paying and that 22.2% (own calculation) went from 1.85 times on September 25 to 0.80 on the 30th and 0.82 on October 1. The "exactly one more hike" outcome was trading at 63.0% on October 1 against the 66.7% of the 12 out of 18: 0.94 times. In five days, the market stopped paying for the committee minority and aligned itself with the majority.

Why did the first burst create 189,170 open interest contracts on Kalshi and the hour following the second destroy 99,965?

In a binary market, every live contract has someone on the YES side and someone on the NO side. If two new participants are matched, open interest increases; if the person holding the YES sells to the person holding the NO, both positions are canceled and it decreases. Volume does not distinguish between these two cases; open interest does.

Between 18:00 and 19:00 UTC on September 29, 2026, the KXFEDDECISION-26OCT-H25 contract traded 310,742 contracts, and its open interest rose from 597,725 to 786,895. Those 189,170 new contracts represent 60.9% of the hourly volume (own calculation): capital that was not previously in the market and entered to take the side of the pause.

The second burst behaved in the opposite manner, and in two stages. During the price jump hour, from 12:00 to 13:00 UTC on September 30, 156,067 contracts were traded with open interest remaining virtually intact (−47). The following hour, from 13:00 to 14:00 UTC, saw 213,908 contracts traded with the price holding steady at 0.33 and open interest falling from 838,501 to 738,536. Ninety-nine thousand nine hundred and sixty-five active contracts disappeared without the price moving, which is the signature of positions being closed.

The first burst brought new positions betting that the Fed will not hike in October, and the second finished flushing out those who held the opposite bet from the market. Total open interest for the October contract ended at 758,468 on October 1, 2026, 52.4% above the 497,722 recorded on September 25: following the episode, there is more money committed to that market than before.

Why did Kalshi's December contract trade 531 contracts on September 29, while the October contract traded 362,890?

The December 9, 2026 meeting also has a market on both platforms, and nothing happened there. The Polymarket December contract was at 71.5% on September 25, 2026, at 74.5% on September 30, and at 74.5% on October 1. The hourly series during the first burst goes from 0.765 at 18:00 UTC to 0.745 at 19:00 and back to 0.755 at 20:00; during the second, it remains stuck at 0.745 before and after.

At Kalshi, the stillness is easier to measure because the API publishes volume in contracts. By summing the sixty-minute candles of the full series per UTC day —a single date criterion, proprietary calculation—, the KXFEDDECISION-26DEC-H25 contract traded 531.04 contracts on September 29, 2026 and 2,007.01 on the 30th, compared to 362,890.06 and 506,356.85 for the October contract. On September 29, the October contract traded 683 times more contracts than the December one. And the candle covering the embargo on September 30, from 12:00 to 13:00 UTC, shows zero volume in December: not a single contract during the hour that moved 13 probability points in October.

That order book is so thin that its "last price" is misleading. The only price printed by the Kalshi December contract during the hour of the first burst was 0.76 with 323.51 contracts, and the contract closed the episode at 0.72 from 0.70 on September 25. Both figures are true and do not contradict each other: in a market that trades three figures a day, the last price jumps four cents without any movement to measure, and it is the price from September 25 versus October 1 that provides the real information.

Reading the three markets together adds one more thing. With the October contract at 0.355 on October 1, 2026, the December contract at 0.745, and the "3 hikes in 2026" contract (equivalent to hiking in both October and December) at 0.1815 — the implied probability of a December hike is 51.1% if October hikes and 87.4% if October does not hike (own calculation). The market pays more for December when October does not hike, which is the structure of "there will be one more hike and the only thing in doubt is the date." Whether this is correctly or incorrectly priced is not addressed in this article. The two previous snapshots show the same: 61.7% versus 91.0% on September 25 and 53.3% versus 85.2% on the 30th.

Was it the JOLTS report, was it Barr, or did Kalshi and Polymarket copy each other?

There are four reasonable alternative explanations for a 33-point drop in probability over five days, and all four are debunked by the figures from the same series:

• The Job Openings and Labor Turnover Survey (JOLTS) for September 29, published by the U.S. Bureau of Labor Statistics (BLS), was released at 14:00 UTC, four hours before the first burst, showing 7.1 million vacancies and labeled as "little changed" by the bureau itself. During that hour, the Kalshi contract moved from 0.69 to 0.67 with 9,038.31 contracts, and Polymarket went from 0.685 to 0.645 before returning to 0.675 at 17:00 UTC. Two probability points with 9,038 contracts, recovered in three hours, compared to 21 points with 310,742 contracts four hours later.
• Federal Reserve Governor Michael Barr spoke on that same September 29 at the Detroit Economic Club, and his timing is public: the Federal Reserve Board's publication channel dates his text at 16:40 UTC (12:40 PM Eastern Time). In the hour from 16:00 to 17:00 UTC, the Kalshi October contract traded 296.57 contracts, and 2,453.75 in the following hour — with the price flat at 0.67 for both. Furthermore, the text repeats the line from September 23 — "further policy adjustments are likely to be needed" — which on the 23rd and 24th accompanied a rise in the price of that contract. Timed and without effect.
• The possibility of one platform copying the other is ruled out by the inverted order: Polymarket opens the first burst 24 seconds before Kalshi, and Kalshi opens the second one six seconds before Polymarket.
• The ADP private employment report was released at 12:15 UTC on September 30 with +90,000 jobs, above the forecast of 70,000, providing the best contrast in the piece: a stronger than expected figure, which logically should push the probability of a hike upward, moved nothing. In that fifteen-minute bucket, Polymarket executed 23 trades and 8,306.71 dollars with the YES price between 0.44 and 0.47; the Kalshi "no change" contract traded 4,178.76 contracts across seventeen instances with the price between 0.53 and 0.56. The October contract itself saw three trades and 408.94 contracts, with the price stuck at 0.45. Fifteen minutes later, the soft BEA data wiped out 13 points.

There is one hour that does not fit, and now we know which way the bet was placed. From 07:00 to 08:00 UTC on September 30, 2026, the Kalshi October contract traded 45,918.47 contracts and its open interest rose by 41,547.63 — with a high of 0.49 and the price closing where it opened, at 0.44. Reading the tape for that window, 43,862.45 of those contracts have the aggressor on the YES and only 2,056.24 on the NO: 95.5% of the volume was someone buying the upside. It arrived in two clusters, at 07:02:10 and between 07:36 and 07:38, with two consecutive executions of 10,000 contracts at 0.46, and no block trades. Someone tried to buy back the October upside less than five hours before the inflation data and the book absorbed it entirely. We do not know who it was, and knowing would change the reading of the 189,170 new contracts from the first burst: it would tell us if there was a trader with a large position on the other side or a tail of small participants.

What does the US Treasury curve say about whether the rally is delayed or canceled?

Prediction markets can be wrong on their own. The way to check if there is something behind the movement is to look at the instrument that actually charges or pays the official rate, and that is short-term U.S. Treasury debt.

Between September 28 and 30, 2026, the 3-month point fell by 8 basis points (hundredths of a percentage point) and the 2-year point by 4. The 10-year point rose by 5. The series runs through September 30 because, according to the Treasury's own methodology, the quotes that feed it are taken "at or near 3:30 PM" Eastern Time and the rates "are usually available by 6:00 PM" in that time zone: at 1:30 PM UTC on October 1, the time of the readings for this article, the data for the day had not yet been formed.

Those 3 months are not a specific security either: the same methodology warns that rates "are read at fixed points of constant maturity and may not coincide with the exact yield of any specific security." The following calculation is our own and refers to a hypothetical 91-day bill starting on September 30, 2026 and maturing on December 30, with a one-to-one pass-through of the increase to the average rate for that term.

A 25 basis point hike decided on October 28 takes effect the following day and covers 62 of those 91 days: 17.0 basis points. The same hike decided on December 9 covers 20 days and is worth 5.5. Moving it from October to December with total certainty is worth the difference —11.5 basis points— and with the observed price change (33 points less probability in October and 3 more in December) the expected value is 0.33 × 17.0 − 0.03 × 5.5 ≈ 5.5 basis points. Canceling it entirely would be worth a full 17.0 basis points.

The 3-month point moved 8, with 5 of them on September 30. It is on the side of delay and far from the side of cancellation, and aligns with what the three Polymarket markets are saying among themselves: one more hike, with the date in doubt. A separate piece from this one measures what happened with the hike the Fed already made on September 16, 2026, and how it reached on-chain dollar rates; this one measures how and at what time the two platforms priced in the next cut.

What did Kalshi and Polymarket do on October 1, the day without a burst?

No hour on October 1, 2026, resembles the two from the previous day. The hour containing 12:30:00 UTC —the same minute on the clock that opened the second burst— traded 1,097.99 contracts on Kalshi's October contract, with the price between 0.34 and 0.35. The minute alone moves nothing; what moved was what was published within it.

Throughout the entire day, the October contract traded 31,796.01 contracts during the thirteen hours from 00:00 to 13:00 UTC on October 1, 2026, compared to 362,890.06 for the full day of September 29: 11.4 times less, using the same counting method (sum of sixty-minute candles per UTC day). The price moved within a range of 0.32 to 0.38 and closed the 13:00 UTC candle at 0.34. On Polymarket, the hourly series remained flat at 0.335 from midnight until 08:00 UTC and rose to 0.355 — where it remained at 13:28. Neither platform approached 50% again.

The two-point rally has a single author. On Polymarket, on October 1, 2026, there were 120 trades and 69,000.73 dollars in the October contract until 13:25 UTC, and a single trade accounts for 76.7% of that figure: at 08:40:33 UTC, a purchase of 151,598.56 YES shares at an average price of 0.3492 — 52,934.62 dollars. On Kalshi, the largest trade of the day was of the opposite sign: at 11:57:46.66 UTC, a NO aggressor took 19,259.75 contracts with the YES at 0.37 and the next entry, six seconds later, printed 0.34.

The two largest operations of the day on the two platforms move in opposite directions and are separated by three hours, seventeen minutes, and thirteen seconds. In the two bursts on September 29 and 30, the separation was 24 and 6 seconds, with the same sign both times. Without a common publication to trigger them, the synchrony between the two platforms disappears.

The magnitude of that sale on Kalshi adds a liquidity data point: 19,259.75 contracts moved 3 probability points, about 6,420 contracts per point, compared to 14,797 per point during the first burst. The order book on a quiet day is thinner than during a news hour, and the same size moves the price more. Between the snapshot on September 30 at 19:48 UTC and that of October 1 at 13:30, the contract traded 57,257.94 contracts and its open interest rose by 4,673.94.

What remains to be verified on Kalshi and Polymarket before October 28?

What this article claims has specific expiration dates, and we publish them with numbers so they can be verified without waiting for the verdict:

• October 2, 2026, 12:30 UTC. The BLS has the September employment report scheduled for 8:30 AM ET. If the October contract returns above 50% on both platforms, the late September cut was premature and what remains standing is the mechanism, not the level. If the probability moves 20 points in a single day without a scheduled event or embargoed data, the description of the two bursts does not apply to this market.
• October 9, 2026. If the Polymarket December contract drops below 60% or the Kalshi KXFEDDECISION-26DEC-H25 falls below 0.60 without Fed communication or intervening data, what the two bursts did was not move the October hike to December: they removed it, and the Treasury curve's "delay" reading collapses with it.
• Before October 28, 2026. If "3 hikes in 2026" returns above 25% on Polymarket, the episode on September 29 and 30 described a day and a half, not a regime change, and it will have to be recorded as such.
• October 28, 2026, 17:59 UTC. This is the closing time written into the rules of the Kalshi contract itself. If the Fed hikes, Kalshi and Polymarket mispriced the meeting on September 29 and 30, and this article will have documented with second-by-second precision exactly when they did so.
• December 9, 2026, 18:59 UTC. Closing of the December contract. This is the one that did not move, and it is the one that decides if "one more hike, date in doubt" was the correct reading.

For the reader using these markets as a thermometer, there is a practical consequence beyond the Fed: the figure published by an aggregator in the mid-afternoon does not contain the information that the tape does. The Kalshi October contract hit 0.35 at 13:28 UTC on September 30, 2026; 0.40 at 17:00; 0.36 at 19:00 and 0.34 at 19:48. Anyone comparing two platforms, or the same platform on two different days, without noting the time of the reading is comparing figures that are not from the same moment, and during a news hour, that is worth 21 probability points.

Sources and links: NY Fed — Williams, “Unwavering Dedication” (29-sep-2026) · NY Fed — media advisory with event time (18-sep-2026) · University at Buffalo — event press release · Federal Reserve — Barr at the Detroit Economic Club (29-sep-2026) · BEA — personal income and outlays for August (30-sep-2026) · BLS — Job Openings and Labor Turnover Survey (JOLTS) · BLS — release schedule · Federal Reserve — FOMC projections from 16-sep-2026 · Kalshi — contract KXFEDDECISION-26OCT-H25 · Polymarket — October decision event · U.S. Treasury — yield curve methodology · U.S. Treasury — daily par curve for September 2026