Kalshi Flight Cancellation Odds: The Hedge That Drew A Lawsuit

Updated August 27, 2026 · 16 min read · by Sam Smith

In July, a conference organizer walked onto a federally regulated exchange and bought something that does not exist in the insurance market: a contract that pays seven figures if half or more of the flights at John F. Kennedy International Airport are cancelled on a single named day. The premium was five figures. The counterparty was a Wall Street trading firm. The day was October 21, 2026, the day before their conference opens, when everyone flies in.

Two weeks later a flight-tracking company sued the exchange over it, and withdrew the case the following day.

The contract is still open. It is the only flight-cancellation market Kalshi lists, which we confirmed by querying the exchange's own market feed for every related series. No other airport, no other date, nothing at LaGuardia or Newark, on an exchange that mostly prices weather, elections and economic data. And the price on it right now sits several times above the number seven AI models produced when we handed them fifteen years of federal cancellation records for the airport, fourteen consecutive Octobers of them, and never showed them the market. The reason for that gap is not what most readers will guess, and it is buried in the one thing nobody writing about this market has actually gone and counted.

Why This Contract Exists At All

The buyer is NEXTPredict, which runs a business-to-business prediction markets conference. Its summit is scheduled for October 22 and 23 at Convene in Hudson Yards, and the risk that keeps an events business awake is not the venue burning down. The nightmare is a travel meltdown that empties the room while every deposit stays spent.

"Over 13 years and more than 800 events, we have seen firsthand how quickly circumstances outside an organizer's control can undo years of work," co-founder and managing director Pierre Lindh said in the company's announcement of the trade. On why an event contract instead of a policy: "Traditional event insurance remains essential and unchanged. It is well suited to established venue-related risks, while force majeure coverage may depend on the event being formally canceled. Our challenge is that a major travel disruption could severely damage attendance while leaving us obligated to hold the event and pay existing costs."

The trade itself is simple arithmetic once you know how these contracts work. A Kalshi contract is a one-dollar promise: if the thing happens, the holder collects a dollar, and if it does not, the holder collects nothing. So the price in cents is the market's read on the chance, and the number of contracts is the maximum payout in dollars. NEXTPredict paid $12,000 for three million contracts at four tenths of a cent each. If the contract settles Yes they collect $3 million, which is 250 times the premium. Susquehanna, a market-making firm, took the other side, which is how Kalshi makes its money without ever taking a position itself.

That block is still sitting there. When we read the market feed on August 27, lifetime volume stood at 3,049,813 contracts against open interest of 3,023,704. The feed does not name who holds what, but a block-sized position that size still being open, against total lifetime trading barely above it, means everything that has changed hands since the hedge amounts to a rounding error beside it. If the mechanics of that dollar promise are new to you, our guide to how Kalshi settlement works is the place to start.

The Lawsuit That Lasted A Day

On August 10, FlightAware sued Kalshi in the U.S. District Court for the Southern District of New York. The complaint is a data and trademark case on its face. What made it worth reading is the argument stacked on top of that one.

"Kalshi never informed FlightAware that it would rely on FlightAware's data to determine the outcome of these betting markets," the complaint said, per TechCrunch, which also reported the company's position that it learned of the markets from press coverage. FlightAware argued these markets "could similarly be abused to interfere with flights, posing a safety risk for travelers and airport employees," and told the court that "as a company whose business depends on the safety and integrity of aviation, FlightAware is facing imminent and irreparable harm."

Kalshi's public answer was a find-and-replace. Every reference to FlightAware came out of the market page, replaced by the phrase "Primary Source Agency," alongside new language stating that the market "has not been endorsed by the Primary Source Agency or its affiliates." FlightAware withdrew the suit without prejudice the next day, and the market stayed open for trading.

One detail has gone unreported, and we found it by reading the exchange's data rather than the exchange's web page. The scrub was cosmetic. Kalshi's public market feed still lists the settlement source for this series, by name, as FlightAware, with a link to flightaware.com. The anonymised phrase appears in the rules text a reader sees; the machine-readable record underneath it never changed. Both things are true at once, and only one of them is visible on the page.

The safety objection deserves to be taken on its own terms rather than waved at, so read what the contract actually says about it. The rules define a category of "Excluded Events" that void the market outright:

• sabotage and unlawful interference with civil aviation
• unauthorised drone operations and laser illumination
• trespass on the airfield, or tampering with airport, carrier, air-traffic or utility infrastructure
• false reports or threats causing an evacuation, ground stop or suspension of operations
• malicious cyber incidents affecting the airport, a carrier or an air navigation service provider
• any security closure a government authority orders in response to the above

Under those rules, cancellations flowing from any of those acts do not count toward the percentage. FlightAware's filing argued the markets could be abused anyway. Both of those things are on the record, and the exchange and the flight tracker never argued them out in front of a judge.

What Actually Has To Happen At JFK

Most of the coverage described this as a market on arriving flights. The contract is broader than that, and the difference matters.

The denominator is frozen. Kalshi takes the total number of commercial passenger and cargo flights scheduled at JFK for October 21, arrivals and departures combined, as that schedule stands at 11:59:59 p.m. Eastern on October 19. Flights added after that moment count for nothing. Flights removed after that moment still sit in the denominator. The numerator is how many of those frozen flights end up classified as cancelled, and cancellations announced pre-emptively after the freeze still count, which means a storm forecast three days out does not save the seller. The freeze cuts the other way too: flights an airline drops before 11:59 on the 19th were never in the fixed total, so an early pre-cancellation wave shrinks the denominator rather than filling the numerator.

What does not count: delays of any length, diversions, and gate returns that eventually depart. A flight cancelled and later reinstated drops back out. If the airport closes entirely for the day, the result is deemed 100.00%. The percentage is truncated at two decimals rather than rounded, so 49.999% settles No.

A Worked Example: How The Percentage Is Actually Computed

Take a real October 21 rather than a made-up one. On October 21, 2025, the U.S. carriers we can measure operated 564 flights into and out of JFK and cancelled none of them. Run that through the contract's arithmetic and the denominator is 564, the numerator is 0, and the result is 0.00%, truncated at two decimals. Nothing settles Yes.

Now hold that same denominator and ask what the strike would have required. Half of 564 is 282, so 282 cancellations clears it and 281 does not. For scale, JFK's single worst capacity-coded day in fifteen years of records cancelled 193 flights out of 696, which is a bigger raw number against a bigger schedule and still only 27.7%. The numbers here are a walkthrough of the mechanics on a past day, not an estimate of this October's schedule, which will not be fixed until the night of the 19th.

One more piece of trivia that turns out to be the whole story. Kalshi quotes most of its markets in whole cents. This one is quoted in tenths of a cent, which is the exchange telling you plainly what kind of question it thinks this is.

Fourteen Octobers Of JFK Cancellations

Nobody writing about this contract has published what JFK actually does on an ordinary day, or on its worst ones. So we built it. Using the Bureau of Transportation Statistics Reporting Carrier On-Time Performance ledger, which records every domestic nonstop flight operated by the large U.S. carriers, we pulled every flight with JFK as origin or destination across fourteen consecutive Octobers, 2012 through 2025. The sample runs to 434 days and 249,692 flight records, with no month skipped and no day dropped. The method is one line: take every record whose origin or destination is JFK, count it once, and divide the ones flagged cancelled by the total for that date. A leg has only one JFK end, so nothing is double-counted.

The pooled October cancellation rate is 1.22%. Those 434 days distribute like this.

A Day At JFK In October	Days	Share of all October days
Half The Schedule Or More Cancelled	3	0.69%
A Quarter To Half Cancelled	0	0.00%
A Tenth To A Quarter Cancelled	6	1.38%
Five To Ten Percent Cancelled	7	1.61%
One To Five Percent Cancelled	28	6.45%
Under One Percent Cancelled	390	89.86%

Read the second row again. In fourteen Octobers there is not one single day where JFK cancelled between a quarter and half its flights. The airport does not drift toward this contract's threshold. It sits under a tenth for nine days in ten, and then, very rarely, it stops.

All three of the qualifying days came from one storm. Hurricane Sandy cancelled 100.00% of JFK's schedule on October 29, 2012, 98.60% the next day and 69.01% the day after. Two days before that, on October 27, the airport cancelled two flights out of 478. The next day, as the forecast hardened, it cancelled 103 of 522, or 19.73%. Then the airport stopped. That is the shape of this thing: 0.42%, then 19.73%, then everything.

This market has a cliff where most markets have a slope. In fourteen Octobers JFK has never once cancelled between a quarter and half its flights. It stays under a tenth, or it stops.

And the specific date on the contract has its own record. Across all fourteen October 21sts in the sample, JFK cancelled five flights out of 8,085, a rate of 0.062%. Eleven of the fourteen were perfect. The worst was 0.71%.

Two honest limits on that ledger. It covers domestic nonstop flights on the large U.S. reporting carriers, which leaves out foreign carriers, cargo-only operators and the U.S. carriers' own international departures, all of which Kalshi's contract counts and all of which JFK has in quantity. Which is why the daily counts above run in the hundreds rather than the roughly one thousand movements JFK actually handles. This is a well-measured proxy for the contract's universe, not the universe itself. Which way it leans is our reasoning rather than our measurement: long-haul flights are cancelled less readily than short domestic hops, so the full-universe rate on an ordinary day is probably a shade below what we measured. On a day the airport closes, everything cancels and the two converge. And the ledger runs through June 2026, because that is the newest month the government has published, which we verified by probing the archive rather than assuming it.

Hottest Prediction Markets Right Now

• 2028 Democratic presidential nominee · $195M traded
• 2027 Pro Football Champion · $72M traded
• Pro Baseball Champion · $61M traded
• 2028 Republican presidential nominee · $60M traded
• Bitcoin price at the end of 2026 · $33M traded
• Will the U.S. confirm that aliens exist? · $32M traded

Every market above links to our full AI model verdict; browse them all on the OddsShopper prediction markets hub, and see how every settled call actually scored on the full graded scoreboard.

The Path That Isn't A Storm

There is an obvious second route to a cancelled airport in 2026, and it is worth pricing properly rather than gesturing at. The FAA has kept reduced flight limits at New York's airports on air-traffic-controller staffing, and staffing-driven ground stops hit JFK repeatedly this summer.

So we measured that path too. Every cancellation in the ledger carries a cause code, and one of them, National Aviation System, is the bucket holding air-traffic control, airport operations, heavy traffic and non-extreme weather. It is the closest thing the data has to a capacity signal, and it is not proof of a staffing cause on any given day. Filtering for days where most cancellations sat in that bucket rather than the weather bucket, the highest day in fifteen years of sampled data is January 12, 2018, at 27.73%. Next is October 30, 2025, at 23.62%, and then October 17, 2019, at 20.42%. Our percentages here are rounded to two decimals; the contract's own truncation rule applies only at settlement.

Now the strongest test available. In November 2025, during the government shutdown, the FAA issued an emergency order cutting scheduled flights at 40 major airports including JFK. It began at 4% on November 7 and was scheduled to climb to 10% by the 14th. It never got there: the cuts peaked at 6%, were lowered to 3% on November 15 as controller staffing recovered, and the order was terminated on November 17. Even so, it is the only time outside a national emergency that the FAA has ordered scheduled airlines to fly less at 40 major airports at once. JFK's worst day under it was November 8, at 15.29%, with 72 of the 85 cancellations coded to the aviation system.

A systemwide federal order to fly less moved JFK less than a third of the way to this contract's threshold, and no capacity-coded day in fifteen years of records has got past 27.73%. The finding holds, and it makes a second government shutdown a story about the Kalshi government shutdown odds rather than about this market.

Where Seven Models Landed

We asked nine AI models to price this contract without showing them the market. Each one received the settlement rules verbatim, the full October ledger, the catalogue of every day in the sample that cleared the threshold, and the highest days on the non-weather path. Seven returned an answer; two seats were unavailable at run time. Then every seat read the other six anonymised answers and could revise.

Here is what the exchange is quoting as of August 27, 2026, read off the order book at 12:10 p.m. Eastern.

Live Quote On Kalshi, August 27	Yes price
What Sellers Are Asking	2.9¢
What Buyers Are Bidding	1.6¢
Last Trade	1.6¢

And here is where the panel landed, after the revision round. We blend a panel by taking the middle number rather than the average, so one seat at an extreme cannot drag the answer on its own.

Seat	Round 1	Round 2
Claude Sonnet	0.18%	0.16%
ChatGPT	0.25%	0.25%
Gemini	0.25%	0.25%
Claude Fable	0.3%	0.3%
Claude Opus	0.4%	0.35%
Grok	0.4%	0.4%
Kimi	0.4%	0.4%
Panel Blend, The Median Of The Seven	0.30%	0.30%

Every seat on this panel is graded against real market settlements — records to date: Claude Fable 83% on 779 graded calls · Claude Opus 84% on 849 graded calls · Claude Sonnet 82% on 824 graded calls · GPT 84% on 6,310 graded calls · Gemini 86% on 5,009 graded calls · Grok 84% on 4,307 graded calls · Kimi 82% on 2,550 graded calls. Recomputed daily; the full scoreboard is public.

These are model estimates, not predictions of fact and not financial advice. Kalshi lists CFTC-regulated event contracts, open to traders 18+ where the platform is available; check its own eligibility page for your state. Every number here gets graded in public once the market settles, on the full graded scoreboard.

More live boards from the same panel: Kalshi government shutdown odds, the October 1 contract, 8¢ bid and 9¢ ask · the Eastern Pacific storm-count board, where more than 18 named storms trades at 65¢ bid and 66¢ ask. Prices fetched August 27, 2026.

The panel's median is 0.30% and the market is quoted between 1.6¢ and 2.9¢. Before anyone reads that as a signal, two things. First, the record: when this panel disagrees with a market by a wide margin, the market has been right roughly two thirds of the time, which is why we publish the gap as context and never as a call. Second, and more interesting here, a contract bought as insurance is supposed to cost more than fair value. A premium is exactly that gap. NEXTPredict paid four tenths of a cent against a panel number of three tenths, which is remarkably close to fair for a bespoke seven-figure hedge. The wider gap belongs to the quote today, where the difference between the bid and the ask is larger than the entire value of the contract. Buying at the ask and selling at the bid costs 1.3¢ on a position the panel prices at 0.3¢, so the round trip is worth more than the thing being traded.

Where The Panel Changed Its Mind

Only one seat moved, and it moved because of an argument nobody else had made.

Claude Sonnet, the lowest number on the board, made a point about the calendar. Sandy's shutdown days were October 29 through 31. The other worst October days in the sample cluster in the same late-month window. October 21 is eight days earlier, before the season's hybrid-storm peak. Sonnet held its number and explained why: "That's a genuine date-specific discount none of the seven addressed, since none conditioned on where in the month Oct 21 sits relative to the historical landfall cluster."

Claude Opus took the point and moved. "I had spread the Sandy hazard roughly uniformly across October, but October 21 sits before the late-October hybrid/nor'easter peak that produced both Sandy (10/29) and the sample's other worst October days, so a flat 1/31 allocation overstates the 21st specifically. That seasonal-timing point is evidence I had not weighted, not group agreement, so I move 0.4 to 0.35 rather than to the panel's 0.25." [Editor's note: 0.25% was where two seats sat, not the panel's middle number, which was 0.30% in both rounds.]

Kimi held, and its reason is the best argument against the low end of the board. "The clearing threshold is weaker than Sandy itself — the winter-storm catalogue shows ordinary major storms, not record-breakers, routinely close JFK, so the capable-storm set is larger than the Sandy-class set and aggressive climatological shrinkage overcorrects." The ledger gives that objection real support, short of the word "routinely." January and February 2026 alone produced five days at or above the threshold: January 25 at 92.19% and January 26 at 53.17%, then February 22, 23 and 24 at 66.55%, 99.11% and 50.56%. Every one was weather-coded, and none came from a storm anyone will remember in a decade. What saves the seller is the season, not the severity.

What Would Change The Panel's Mind

Four things, each of them checkable rather than atmospheric.

A named Atlantic system tracking toward the Northeast in the third week of October. This is the only trigger with real weight, and it is the one that cannot be forecast today. Meaningful skill on a specific day's weather runs out somewhere around ten days, so the panel's number will not move on weather until roughly October 11. Worth knowing: when we queried the exchange's storm series on August 27, every open contract covered the Eastern Pacific, so there was no Atlantic market to read as a leading indicator. That could change before October.

A federal funding lapse on October 1 that runs three weeks. Direction: up, but less than instinct suggests. The highest capacity-coded day ever observed in this data is 27.73%, and the November 2025 emergency order produced 15.29% at JFK. It would need to stack on top of a storm to matter.

An FAA order cutting New York capacity further before the schedule freeze. Direction: down, counterintuitively. Cuts announced before 11:59 p.m. on October 19 shrink the frozen denominator rather than the numerator, which makes the percentage harder to hit, not easier.

A change to the schedule baseline or the source agency. The exchange has already rewritten this market's public rules once under legal pressure. Any further amendment to how the denominator is fixed changes the contract, not the weather.

Settlement Timeline

Date	What happens
October 1, 2026	Federal fiscal year begins; the shutdown contract resolves
October 11, 2026	Roughly the first date a ten-day forecast reaches October 21
October 19, 2026, 11:59:59 P.m. ET	Schedule baseline freezes; the denominator is fixed
October 21, 2026	The measured day
October 22, 2026, 10:00 A.m. ET	Expected expiration and settlement

This page is re-scored when the story moves, and every number on it carries the stamp it was read at. Prices as of 12:10 p.m. Eastern, August 27, 2026; panel run the same morning.

More on this:

• Kalshi Tour Championship Round 1 Scores: Morikawa 4-For-4
• Kalshi Tour Championship Top 10 Odds: 11 Cashed Last Year
• Best AI Coding Model Odds: One Column Decides The Whole Race
• 2026 Governor Races: Every Kalshi Board We Track
• NBA East 1 Seed Odds: Kalshi Doubts The 60-Win Pistons

The Bottom Line

The interesting thing about this contract was never the probability. It is that an events business found a way to buy a risk transfer that the insurance industry does not sell, at a price a forecaster would call close to fair, on a public exchange, and that the transaction was novel enough to draw a lawsuit within two weeks and see it withdrawn a day later.

What the record says is narrower and sturdier. In fourteen consecutive Octobers, JFK has cleared half its schedule in cancellations exactly three times, all of them inside one hurricane, and the specific date on this contract has gone 0.062% across those same fourteen years. The route that does not require a storm has never in fifteen years of records got past 27.73%, and the November 2025 federal flight-reduction order moved JFK to less than a third of the way to the strike at its worst. Seven models read all of that and landed at three tenths of one percent.

The buyer, for what it is worth, does not need to be right. They need the room to be full.

Hero illustration: OddsShopper, in the house collage style. John F. Kennedy International Airport photo by DimiTalen, licensed CC0; photos cropped, toned, and composited.