The claim: The CFTC deleted the licence condition that made a shortfall at Kalshi’s clearinghouse arithmetically impossible, and the rulebook Kalshi wrote to replace it seven days later exempts the company from ever replenishing its own default capital while requiring members to restore theirs on demand.

The numbers: The book opened in June. The affiliated FCM’s customer segregated funds went $0 to $17,765,306 in one month while its capital rose 28.5 times, per the CFTC’s own FCM reports. Members are assessable to 550% of their deposit and must replenish it on demand; the company’s contribution replenishes never. Seven days separated the amended order from Kalshi’s own rulebook filing.

The catalyst: Two emergency orders under CEA section 8a(9), on 14 July and 11 August 2026, are what currently keep the exchange open. Kalshi told its regulator the emergency risked its compliance with Core Principle 21, Financial Resources, against a New York claim of $36bn at a company valued at $22bn.

Wrong if: Kalshi Klear amends Rule 12.5C to replenish the company contribution, or the Commission requires its futures margin methodology to arrive through a Regulation 40.5 approval instead of a 40.6 self-certification. Either would mean the constraint moved back to the regulator.

The half of the Kalshi story nobody has opened

Every piece I’ve read on the New York fight is about jurisdiction: federal preemption against state gambling law, one state reaching a national market, a regulator reviving a power untouched since the Carter grain embargo. I won’t retell it.

I went looking for the other half: what happens to money and open positions if the claim lands. The interesting change happened months before New York filed, in documents carrying no press coverage.

Full collateralization was a licence condition, and a regulatory exemption

I will start with what Kalshi Klear was. Its own application says it plainly.

When Kalshi applied to run its own clearinghouse, it told the Commission every product would be pre-funded and fully collateralized. Then it drew the consequence itself. That design was “negating the need to calculate variation margin levels or maintain a default fund, and making it easier to offer clearing services directly to participants” (Exhibit A-3).

Full collateralization is not a risk management style. It is an arithmetic property. A fully collateralized position, in the Commission’s definition, requires the clearinghouse to hold “at all times, funds in the form of the required payment sufficient to cover the maximum possible loss that a party or counterparty could incur upon liquidation or expiration of the contract.” Hold the maximum possible loss and a shortfall cannot arise. There is nothing left to mutualise.

Here is the part I had not appreciated before reading the regulations, and it makes the April change consequential. Full collateralization also switches off two of the Commission’s own requirements. Under Regulation 39.11, a clearinghouse must run monthly stress tests to size the resources it needs against its largest member default, and must hold liquid resources against a one-day settlement cycle. Both carve out the same thing: “The requirements of this paragraph (c) do not apply to fully collateralized positions,” and “A derivatives clearing organization is not subject to paragraph (e)(1)(ii) of this section for fully collateralized positions.”

Those words did two jobs: they described the book, and they exempted the clearinghouse from sizing a default package at all.

Both jobs were written into the licence. The original order of registration, 28 August 2024, condition (1): “Kalshi is permitted to clear, in its capacity as a registered DCO, fully collateralized positions in swaps.”

On 14 July 2025 Kalshi asked to widen that. On 13 April 2026 the Commission did. The amended order supersedes the original, and its condition (1) now reads, in full: “Kalshi is permitted to clear, in its capacity as a registered DCO, futures, options on futures, and swaps.”

I checked whether that is loose drafting, the obvious way for this to be nothing. It is not. Of the nineteen registered clearinghouses on the Commission’s filings registry, twelve carry the “fully collateralized” qualifier, and those twelve include every prediction market venue on the list: Polymarket Clearing, ForecastEx, ProphetX, Aristotle, Electron, Gemini Olympus. The six clearing without it are Eurex, LCH SA, ICE Clear Credit, ICE Clear Europe, ICE NGX and Nodal Clear.

One entry settles whether the drafting is careless. Nadex, now Crypto.com Derivatives North America, is the other retail event venue permitted to run margin, and the Commission wrote its remark out longhand: “permitted to clear margined futures, and fully collateralized futures, options on futures, and swaps.” When it means a split book it says which half is which. Kalshi Klear’s line splits nothing. Its permission now reads like Eurex’s, and not like Polymarket’s.

Seven days later, Kalshi wrote the replacement itself

This part matters more than the order.

On 20 April 2026, seven days after the amended order, Kalshi Klear self-certified a new rulebook under Regulation 40.6(a), effective after close of business on 27 April. Self-certification is the route that needs no approval. You file, the clock runs, the rule takes effect. Kalshi also requested confidential treatment of the substantive amendments until the effective date, so the redline was not public while the review period ran.

I would not call the changes housekeeping. Rulebook v1.3 contains the phrase “Guaranty Fund” zero times. Version 1.4 contains it seventy-five. The amendment introduces Margined Contracts as a defined class, Initial and Variation Margin, Futures Customer Collateral in its own segregated account, an expanded Risk Management Committee, and an entire new Chapter 12 establishing a default management framework where there had been none.

Note what those two documents are, because I think that swap is the whole finding. What constrained Kalshi Klear used to be an instrument the Commission controls, amendable only by the Commission, on the record, after a request that took nine months. What constrains it now is a rulebook Kalshi controls, amendable by Kalshi, effective seven days later, with confidentiality on request. The protection didn’t weaken in April. It changed custodian.

The obvious answer is the one I’d make myself. Self-certification is not a loophole. It is the ordinary route every clearinghouse uses, and the Commission keeps a stay power over it: Regulation 40.6(c)(1) lets it halt a certification for “novel or complex issues,” take 90 days to review, and run a 30-day comment period. That power is not theoretical here. The Commission used it against this exact registrant six weeks later, staying Kalshi’s Michigan emergency rule in the July order.

So the custodian point needs stating precisely. The Commission can stop a self-certified rule. What it no longer does is size the package in advance. An order condition is tested before anything happens and changeable only by the body that wrote it. A stay is reactive and needs somebody to notice. A state court unwinding trades is a loud trigger. A guaranty fund calibration is quiet.

The waterfall they built is good, and I want to say so plainly

I read Chapter 12 as a conventional, competently drafted loss waterfall of the kind a serious futures CCP runs. Calling it shoddy would be wrong.

Defaulter’s margin goes first, then its guaranty fund deposit, then the clearinghouse’s own capital, then the mutualised fund, then more of that capital, then assessments on survivors. Deposits are calibrated on stress loss, open interest and volume, with a $1,000,000 floor for a new member.

Its own layer is generous, too, and here I had to correct myself. Rule 12.5C defines a First and a Second Tranche Company Contribution, each “not to exceed the greater of $20 million or 10% of the total aggregate value of the Guaranty Fund.” That’s twenty percent across both tranches. Published comparisons put OCC near 0.5% of member contributions, LCH around 0.9%, CME about 2.8%, ICE Clear US about 10%. On the percentage, Kalshi Klear is not skimping.

The one layer that never replenishes

What I could not find in Chapter 12 is a replenishment obligation on the clearinghouse. I went looking because the peers I checked have one.

Rule 12.5C, immediately after defining the two tranches: “If the Company Contributions are so applied, the Company will have no obligation to provide additional funds to replenish such contribution or otherwise provide additional funds in respect thereof.”

Now hold that against what the same rulebook demands of everyone else. Rule 12.4H-2 says a member whose guaranty fund deposit is drawn down “shall restore the deficiency … on demand (a ‘Replenishment’).” Members are separately assessable up to 200% of their deposit for a single default and 550% across multiple defaults inside six months.

So after one bad day the mutualised layer refills and the owner’s layer does not. That is the asymmetry, it sits in one document, and it runs opposite to where the industry has moved: the OCC proposed rules for exactly how it replenishes its minimum corporate contribution after each chargeable loss.

Put a number on the member side, because the rulebook supplies one. A new member’s guaranty fund deposit may not be less than $1,000,000, and assessments run to 550% of it across multiple defaults inside six months. A member at the floor is exposed to its $1m deposit plus $5.5m of assessments, $6.5m against $1m posted, and then it replenishes and the meter restarts. The owner’s layer is capped at the greater of $40m or twenty percent of the fund, and spent once. Both numbers sit in the same chapter.

I ran that default through all six steps at the rulebook’s own floors in the trade note on Patreon.

There’s a second consequence I’d want priced before clearing anything there. Regulation 39.11(b) lets a clearinghouse count guaranty fund deposits and potential assessments toward the resources it must hold. The regulation sizes the package. The rulebook decides who provides it. Kalshi Klear can satisfy a growing requirement substantially with member money and member assessment capacity, while the parent’s exposure stays fixed and refills never.

The margined book is no longer hypothetical

I’d assumed the April filing was scaffolding for something distant. Kalshi’s certification letter encourages that: the changes “do not alter the economics, risk, or operation of any currently cleared Contract” and merely “create the definitional and structural framework” for “new products in future filings.”

Six weeks later the product arrived. On 29 May 2026 the Commission approved KalshiEX’s BTCPERP contract, a perpetual future on the spot price of bitcoin, cash settled, no expiry, with a periodic funding mechanism. A perpetual exists to be held on margin, and Kalshi Klear clears it.

Three weeks before the amended order, Kalshi’s affiliate Kinetic Markets LLC was registered as a futures commission merchant. That matters because of a representation inside the amended order: Kalshi represented that “only clearing members that are either futures commission merchants or eligible contract participants are permitted to clear margined contracts.” Retail traders aren’t eligible contract participants. An affiliated FCM is the piece that lets customer margin be carried at all.

Here the paper trail becomes a measurement, because the CFTC publishes every FCM’s capital monthly. Kinetic appears in no FCM financial report before the one dated 31 March 2026. Then it sits still for three months and moves at once:

Three months at the regulatory minimum, no customer money at all. Then in June the capital goes up 28.5 times and $17.8m arrives against a requirement of $8,370,057. Cleared swaps segregation is still zero: these are futures customers.

I read that as the margined book opening.

End to end: ask to widen the licence (July 2025), register the affiliated broker (March 2026), get it widened (13 April), self-certify the margined rulebook seven days later, list the perpetual future (29 May). Then Michigan (29 June), then New York (31 July). The reconstruction finished a month before the state actions started. Kalshi owns the exchange, the clearinghouse and the broker.

Where the money sits, and why that part is fine

Here I part company with the alarmed reading, and the protective case deserves its strongest statement.

An ordinary Kalshi user’s event contract collateral is in good shape. Those contracts are still Fully Collateralized Contracts under Rule 7.5, and the clearinghouse “will not clear a position in a Fully Collateralized Contract that is not fully collateralized.” Cleared swaps collateral is segregated under Part 22, futures collateral under Regulation 1.20.

If Kalshi Klear itself ever failed, 11 U.S.C. 766(i) splits the estate into two pools, one for ordinary customers and one for clearing members, each distributed “in priority to all other claims.” I’d assumed Kalshi’s direct users, who are self-clearing members and not a broker’s customers, would sit in a junior class. They don’t. The Senate Report says section 766(i) gives member property “the same priority.” The real subtlety is that the pools are separate, each distributed only within itself, so a shortfall in one drops to a section 726 general claim instead of reaching the other.

Rule 12.6 then walls the old book off from the new one. On a margined default, “Members and the holders of Contracts shall have no recourse to any other funds or any other entity, including without limitation the Margin or Guaranty Funds that support clearing of other products (including, without limitation, the Margin held by the Company for Fully-Collateralized Contracts), the Company, Kalshi Exchange, or any of its affiliates.”

Read one way that’s a genuine protection: a blow up in bitcoin perpetuals can’t be paid out of the collateral behind somebody’s Fed decision contract. My read is that this one is right.

The exposure is the discretionary layer, and the rulebook disclaims it

Read the same sentence the other way, because it also disclaims recourse to the Company, Kalshi Exchange, or any of its affiliates.

We’ve got exactly one live test of what protects a Kalshi user when a court interferes with their trades, and it wasn’t the waterfall. When a Michigan court ordered trades “voided, cancelled and refunded” in June, Kalshi’s proposed remedy was to force liquidate the affected positions at market and then, in its own words to the Commission, “pay the difference to the affected user from Kalshi’s operational funds,” absorbing “the entire shortfall” (July order).

Operational funds. Unsegregated corporate money, paid at the company’s discretion, and the rulebook tells contract holders they have no right to it.

Kalshi also told the Commission why it could afford the gesture: it could “bear the cost of reimbursement” because “the total number of positions impacted by the specific order is limited.” That qualifier is the whole thing, and I’d underline it. My read: the remedy that made users whole in the only real test was scoped to a small population, funded from the balance sheet a $36bn claim attaches, at a company valued at $22bn.

One more note on how that exposure has been described. When the 11 August order relays a warning that a TRO “would expose traders to substantial losses exceeding the collateralized value of the contracts,” I read that as describing the value of open positions. It doesn’t describe a collateral shortfall. Note also where the sentence sits. It’s in the Background section, reciting what Kalshi told the Commission, and it is not among the Commission’s findings. The most alarming line in the public record about customer exposure is the company’s own characterisation of its own risk.

What is holding the doors open

Both emergency orders rest on CEA section 8a(9) and both assert near total insulation from review. The July order says “the merits of the Commission’s emergency determination are precluded from judicial review by operation of 5 U.S.C. 701(a)(2).” The August order notes review lies only in a court of appeals.

The power had been dormant since 1980. Reported counts of its total historical use differ, and no Commission source enumerates them, so treat the number as unconfirmed. The gap isn’t in doubt: nothing between 1980 and this summer.

The Commission that issued them “By the Commission” has one member. Its own page says the Commission “consists of five Commissioners” and lists Chairman Michael Selig, sworn in 22 December 2025, alone. Regulation 140.11 covers acting without colleagues and presumes they exist. Neither order was issued that way.

I’m not writing the one commissioner story, which is taken and overrated. I raise it for one reason. This venue stays open on a power nobody used in the forty six years before this summer, exercised by a Commission of one, asserted to be unreviewable.

Four things I cannot rule out

I can’t tell how much of that $17.8m is margined rather than fully collateralized. The FCM report gives segregated funds, not the collateralization basis of the positions behind them, and reporting says the margin feature is institutional-only with CFTC sign off outstanding. June proves the futures complex took real customer money. It doesn’t prove that money sits behind leverage today.

I couldn’t read the v1.4 redline as filed, because Kalshi requested confidential treatment. I’m working from the clean rulebook and the certification letter’s description of what changed.

The guaranty fund’s balance isn’t public, and neither are Kalshi Klear’s Regulation 39.11 filings. I can give you the formula and the floor, not the balance. At 30 June the customer money in the affiliated FCM’s segregation was $17.8m, so the $40m floor alone was about 2.25 times it. That ratio’s comfortable now, and it decays as the book grows, because the numerator is fixed and the denominator isn’t.

And the strongest objection: a judgment against KalshiEX LLC isn’t a judgment against Kalshi Klear LLC, and segregated collateral isn’t available to the exchange’s creditors. I think that’s correct, and it’s why I’ve put the exposure at the discretionary layer instead of claiming customer funds are at risk. They do share a parent and a balance sheet posture.

What would change my view

Three dated, public, resolvable things.

An amendment to Rule 12.5C obliging the company to replenish would remove the asymmetry this piece is built on, and I’d say so. Second, if the futures margin methodology arrives through a Regulation 40.5 approval instead of a 40.6 self-certification, the constraint has moved back to the regulator. I’d note that Kalshi’s representation to seek 40.5 approval names “swaps or options on futures“ and does not name futures, while BTCPERP is a future. I won’t assert that’s a loophole.

Third, if New York’s claim is dismissed or cut to a number that fits inside the balance sheet, the discretionary layer stops being interesting.

What I’d do with this

If you hold size on Kalshi as a trader, the question isn’t whether your collateral’s segregated, because it is. It’s which protections are structural and which are discretionary. The Michigan file answers that: what made users whole was a payment the rulebook says they couldn’t have demanded.

If you’re considering clearing there as a member, my question is sharper. You replenish on demand and you’re assessable to 550%. The owner’s layer is spent once. Price that before you price the fees.

I’d watch three filings, not the docket. The next Kalshi Klear rulebook and what it does to Rule 12.5C. Any filing carrying a futures margin methodology. And the first margined product open beyond the affiliated FCM’s institutional book, which is when Chapter 12 starts backing something real.

My transferable point: a clearinghouse’s safety can live in a licence condition or in a rulebook, and those degrade differently. A condition takes a Commission and nine months. A rulebook takes seven days. Kalshi’s protection didn’t disappear in April. It moved from the regulator to the regulated, and the layer that moved furthest is the one that never has to be put back.

So the question I’d put to anyone underwriting this venue: when a clearinghouse’s owner is exempt from replenishing its own default capital and its members are not, what is the right premium for clearing there, and is any level of bitcoin perpetual volume worth paying it?

📊 The Decision-Grade Version

This piece is complete on its own. The thesis, the evidence, the confounds and the filings that would kill the view are all above. Nothing was held back to sell a next step.

The Patreon note is a separate piece of work. It takes one decision from inside this story: what a clearing member at Kalshi Klear’s own minimum is exposed to, run through all six steps of the Chapter 12 waterfall with the arithmetic shown and the layers that refill separated from the one that does not. Written for people who put capital behind a view.

→ Read the Kalshi Klear waterfall trade note

→ Or join the Patreon community for every note

Elsewhere: YouTube and LinkedIn.

Cover: kalshi.com, screenshot taken 16 September 2026.

The claim: Kalshi runs two liquidity subsidies whose eligibility clauses are exact complements. One is open to every member except its affiliate and its Market Maker Agreement holders, and its arithmetic is published to the cent. The other is open only to Market Maker Agreement holders, pays up to seven times as much, and files its selection process as Confidential Appendix B. The tier you can read is the tier without a contract.

The numbers: the open tier caps at $1,000 per market per calendar day; the Market Maker tier caps at $50,000 per series per week (Liquidity Provider Program, 4 May 2026). Inside the open tier, a flat $0.005 reward against a taker fee of $0.00204 on a three cent contract is 245% of the whole fee. Buyers of sub-10¢ contracts “lose over 60 percent of their money” across 46,282 contracts (Bürgi, Deng and Whelan, 2026).

The catalyst: DMO staff asked every DCM to review and amend its incentive filings by 14 September 2026 (CFTC Letter 26-23). Kalshi refiled its Volume Incentive Program on 4 August, eight days earlier, and extended it to September 2027. Comments on the affiliate rulemaking close 5 October.

Wrong if: the September amendments publish reward amounts, payout caps and the Designated Liquidity Provider selection process in open text. Then the record becomes readable and my thesis dies.

The consensus, stated the way its holders would state it

On 12 August 2026 the CFTC’s Division of Market Oversight told designated contract markets their incentive filings have been sloppy: staff “have observed an increase in incentive program rule filings relating to event contract products under Rule 40.6(a), including filings that are procedurally or substantively deficient.” Volume based rewards “with steep tiers or threshold bonuses” can push participants “to trade solely to reach volume targets, heightening risks of wash-trading, pre-arranged trading, or other... disruptive trading practices.”

The coverage read this as a wash trading warning, and I think that reading is defensible. It is what the document emphasises. The implied fix is procedural: file better, disclose what Appendix A lists, and the Commission recovers the visibility it lost.

The advisory says what it is: it “represents only the views of DMO staff.” No rule changed on 12 August.

My view is that the reading is right about the advisory and wrong about the market. The deficiency staff describe is real and it is being addressed. It is also not the reason a macro desk cannot tell how much of a prediction market price is bought.

The filings are enumerable, and almost nobody enumerates them

Rule 40.6(a)(2) makes a DCM publish every submission on its own website concurrently with filing, and that obligation is load bearing. The CFTC’s own site surfaced nine Kalshi incentive filings; the bucket behind Kalshi’s regulatory notices page holds 456 documents, 52 of them incentive or fee filings, and two of my nine were superseded by documents that search never returned. The exchange’s own site is the complete index. Its regulator’s is not.

One word needs pinning down first. These programs are self certified: under Rule 40.6(a) an exchange files its own certification that a program complies with the Act, and a ten business day clock runs unless the Commission intervenes. Nobody approves them. “Certified” here means the exchange said so, in writing, to its regulator.

Two programs, complementary by construction

Kalshi’s Volume Incentive Program and its sibling Liquidity Incentive Program are drawn fully in public. Eligible volume is trades “executed on the central limit order book at prices between $.03 and $.97”; rewards come pro rata from a fixed per market pot and “shall be capped at, and shall never exceed, $.005 per contract traded for each participant”; the liquidity program publishes its scoring algebra down to the geometric discount on a resting bid. I have read a lot of exchange filings, and this is more disclosure than most DCMs offer.

I read the same clause on both: Eligible Participants are all Kalshi members “except the following: (i) affiliates of Kalshi; (ii) members who have executed a Market Maker Agreement with Kalshi,” plus brokers and their customers.

Then I found the May 2026 Liquidity Provider Program, running to December 2027. Its eligibility clause, which I read twice, is one sentence: “Eligible Participants are all Kalshi members who have executed a Market Maker Agreement with Kalshi.” How much it pays any one of them is set by a reverse auction: Kalshi’s own program page asks a provider to “share the minimum amount you’re willing to receive as an Incentive Period Reward.”

$1,000 a day is $7,000 a week, so that is seven times the ceiling per market, and a series spans many markets while the $1,000 does not. Kalshi states the Market Maker tier’s targeting itself: it “incentivizes liquidity provision in markets that are new or would otherwise have more limited trading activity.”

And the two compensation disclosures, from one exchange, sit like this. The open tier: “capped at, and shall never exceed, $.005 per contract.” The Market Maker Program, in its entirety, on what it pays: “Predetermined incentives will be available upon satisfying all Program obligations as determined by the Exchange.” No amount. No rate. No cap.

That filing’s enclosure list names “Confidential Appendix C - Market Maker Agreement,” and its cover letter says members “must execute a separate Agreement with Kalshi that will describe the relevant obligations, requirements, and benefits.”

The geometry: a flat cap on a curved fee

Kalshi charges takers round up(0.07 × C × P × (1-P)), P being the price in dollars, with index markets at half that (fee schedule). That is a parabola peaking at 50¢ and collapsing toward both ends, which is the correct shape, because P(1-P) is also the payoff variance of a binary contract. A contract at 50¢ costs $0.0175 to take and carries eight and a half times the risk of a contract at 3¢, which costs $0.00204. The fee tracks the risk.

The reward does not. The cap is flat. Half a cent per contract, everywhere inside the band.

Solve 0.07 × P(1-P) = 0.005 and I get roots at 7.74¢ and 92.26¢. Below and above those, the certified maximum reward exceeds the whole fee the exchange collects. That region runs 9.48 cents wide inside a 94 cent band, so 10.1% of the eligible range pays out more per contract than it takes in. At the inner edge the ratio hits 245%, or 8.59 times the ratio at the midpoint, which is exactly the ratio of the variances. Same curve, twice.

Here is why I read that as a design choice and not an artifact. The filing gives a rationale for the cap: it exists “in order to help avoid price distortion caused by these Volume Incentives.” It gives none for the 3¢–97¢ band, which on a book quoting 1¢ to 99¢ excludes exactly two ticks at each end. A cap that actually neutralised distortion would have to scale with P(1-P), which is to say it would have to look like the fee schedule. Kalshi knows how to write a per contract amount that scales with price, because it wrote one for what it charges.

A second gradient stacks on the first, and I nearly read it backwards. Rewards come pro rata from a fixed pot, so the realised reward per contract is that pot divided by total eligible volume, and the cap binds only when volume is low. A $500 pot pays more than $0.005 per contract on any market trading under 100,000 eligible contracts. The subsidy peaks precisely in the thin markets, which is where the Market Maker tier says it is aimed too.

What actually happens at 4¢, measured

Somebody has measured this zone. Bürgi, Deng and Whelan took data on 46,282 Kalshi contracts across 12,403 events, 2021 to April 2025, into a George Washington University working paper. Their result is a favourite longshot bias: cheap contracts price above their worth. The magnitude isn’t textbook. Buyers of contracts under 10¢ “lose over 60 percent of their money.” Contracts above 50¢ earn a small positive return, and the average across everything is about minus 20%. Separate work finds the bias is domain specific, strongest in political markets, so I wouldn’t carry it uniformly across every category.

Two details in that paper matter more to me than the headline.

The first is who wins. Kalshi records which side initiated every trade, so the authors could split returns without inferring anything, and “Makers earn higher returns than Takers.” Part of that gap is simply that takers pay the fee. The rest is price. Either way, the participants Kalshi’s open tier excludes are the ones already on the better side of it.

The second is timing. The sample ends April 2025, and the current Volume Incentive Program text was filed 4 August 2026. So this is a clean pre period baseline for the exact zone the geometry points at, and nobody has run the post period. The authors note “some evidence that the bias in prices is diminishing over time,” which cuts awkwardly: as the natural mispricing shrinks, a subsidy on top of it becomes a larger share of what remains.

Then the capacity question a principal asks first.

Kalshi’s own filing answers it. The Liquidity Incentive Program sets a “Target Size” that “will be greater than 100 contracts and less than 20,000 contracts”, the depth above which resting size stops earning. Twenty thousand contracts at 4¢ is $800 of premium. Real books can be deeper; that is my ceiling on the depth the program pays for.

A share threshold is a headcount cap wearing a percentage

A third structure I nearly missed, and the one I’d push hardest on in a comment letter.

Kalshi’s 2023 Maker Order Protections Program gives qualifying members a safety net: if a group of resting orders moves your position by a set delta inside 15 seconds, the rest cancel.

Qualification is where I stopped. A member must account for “at least a minimum percentage of all maker volume on Kalshi over each of the prior two calendar months,” and the exchange sets that percentage on its website, not in the filing. The stated range is “between 1% and 7%, with the goal to have at least more than one qualifying member.”

I did the arithmetic the filing does not. A threshold of s percent of total maker volume caps the number of qualifying participants at 1/s, by construction. At 7%, no more than fourteen members can hold the benefit at once. At 3%, the figure the filing uses in its own worked example, thirty three. The exchange’s stated ambition is “more than one.” That is a bounded club whose size the exchange sets unilaterally, on a webpage, between filings.

The same shape sits in Kalshi’s June 2026 perpetual futures fee schedule, whose top maker rungs are reachable by holding a minimum share of the exchange’s own maker volume, up to 10.0%. At the top rung, ten participants can qualify. Not ten thousand. Ten.

Appendix A asks a DCM to “discuss any limitations on program enrollment or the maximum number of program participants.” A share threshold is exactly that limitation, in a unit that hides it from me. No filing I read does the division.

The same shape at the other venues

Polymarket US, which is QCX LLC, filed a Market Incentive Program on 5 March 2026 and asked that Appendix B stay confidential. Its request letter enumerates what is inside: “the specific reward amounts, volume multipliers, payout caps, and program parameters that the Exchange found necessary to incentivize volume,” with confidential treatment requested “in perpetuity.”

Set that beside Appendix A, which requires a filing to “specify the economic terms and amounts of each available form of compensation” and to identify all incentives “and whether they are capped or unlimited.” The four items named as trade secrets are the four named as mandatory public content. FOIA Exemption 4 is real, and both statements are lawful.

The public half says rebate rates and caps are “communicated in writing to Program Participants,” adjustable “on a prospective basis with seven (7) days’ notice,” and that applicants may be waitlisted “indefinitely.” Section III says programs letting a DCM amend terms without a fresh 40.6 filing “may conflict with Part 40.” I expect that to matter in September.

ForecastEx files the programs whole under seal. Its 2 March 2026 submission states that “the incentive rates for the Market Maker Program are being amended to account for the new transaction fee structure.” The attachment list reads: “Exhibit C (Confidential) – Volume Incentive Program. Exhibit D (Confidential) – Market Maker Program.” The rates moved. Where they moved to, I cannot tell you.

Kalshi has done a version of this too, and I found it the most striking line in the set. In June 2024 it amended the Volume Incentive Program and wrote that the amendment complies with the core principles “as discussed in confidential appendix C.” The compliance analysis itself went in the sealed envelope.

Three objections, and what I think of each

“Market makers are a small share of volume, so calling them the price setters is wrong.” The strongest objection, and it measures the wrong quantity. Share of volume is not share of quote. The Liquidity Incentive Program pays for resting orders, and the Market Maker tier pays for maintaining “continuous two-sided markets” at a “maximum bid/offer spread” and a “minimum quote size.” A modest share of volume can still be most of the depth at the touch. Volume is what happened; the quote is what a desk copies.

“These venues are just young and under lawyered.” My control refutes it. Bitnomial is a crypto derivatives DCM with no event contracts, and its June 2026 incentive filing reads: “Exhibit 1 sets forth the updated terms of the Program (Confidentiality Requested).” Coinbase Derivatives files its equivalent as 2026-27 Crypto Market Maker Program_REDACTED. I read that as ordinary DCM practice, inherited and not invented, and I had been half expecting the opposite.

“Manufactured volume is not actually large, so where is the harm?” On the one venue anybody has measured, it is not large. A May 2026 microstructure study of the Polymarket order book ran a self counterparty wash detector across 600 markets and 6.4 million trades and found a median wash share of 0.97%, with 22.2% at the maximum. Unregulated token exchanges run 25% to 70%, which its author calls “a sanity bound rather than an apples-to-apples reference.” Roughly one percent, not seventy. It is a preprint, it measures Polymarket and not Kalshi, and its author calls the figure a lower bound.

The harm I see does not run through wash volume. It runs through concentration: the subsidy is largest in the thinnest markets and at the edges of the band, where the measured return is worst, on books the exchange pays to keep a few hundred dollars deep. And the output of these venues is a probability consumed as data by people who never trade on it. Nobody reads a Bitnomial bitcoin future as a forecast. A great many people read Kalshi’s 4¢ as a 4% chance.

What I can’t rule out

I still can’t see the realised numbers. I have both ceilings now, $1,000 a day and $50,000 a week, and no venue publishes a single realised payout. A ceiling tells you what an exchange has permitted itself, never what it spent. So everything I’m saying about relative size rests on the filed caps, and that’s a real limit.

Confidentiality may be doing exactly what it claims. Reward parameters are real competitive information, and an exchange that published its market maker economics would hand rivals a recruiting sheet. Every request here is lawful and I’m not alleging otherwise. My problem is narrower: the record can’t answer what a price consumer needs, and September won’t change that, because procedural compliance and economic legibility are different properties.

The favourite longshot bias predates all of this. Bürgi and his coauthors reproduce it with a model needing only modest disagreement and a mild tendency to overstate small probabilities, so the sub-10¢ losses aren’t evidence anyone engineered them. My claim stays narrow: the certified subsidy per unit of risk is largest where the measured return is worst.

What would change my view

Three dated, public, resolvable tests. I’ll hold myself to them. Diary them.

14 September 2026 is the staff amendment deadline, and Kalshi refiled its Volume Incentive Program on 4 August, eight days before the advisory landed. If an amended filing publishes the Market Maker Program’s incentive amounts, or the Designated Liquidity Provider selection process that’s now sitting in Confidential Appendix B, then the asymmetry I’ve described is closing and I’m wrong about its persistence.

If Polymarket US or ForecastEx withdraws a confidential treatment request and refiles in public, same conclusion. I’d take either as a clean refutation. No hedging.

And comments on the Conflicts and Affiliations proposal close 5 October 2026. Proposed Regulation 38.852(c)(1)(ii) would make any incentive program covering an affiliate market maker enumerate its obligations “on terms no less favorable to the DCM than those offered to unaffiliated” firms, with 38.852(c)(1)(i) forcing the affiliate to be “filled last at every price level.” Kalshi’s 2023 filing already says “Kalshi Trading LLC, an affiliate of the Kalshi Exchange” participates, so it’d bite on a structure that’s live today.

What I’d actually do with this

If you’re pricing off event contract quotes, my adjustment is narrow and cheap. Discount the 3¢–8¢ and 92¢–97¢ zones on thin Kalshi markets specifically. That’s where the two gradients compound and where the measured return is minus sixty. Everything from roughly 20¢ to 80¢ on a deep contract looks fine to me. I’d keep using the prices. I’d stop using their tails as though they were the same instrument as the middle. They aren’t.

Name the cost of being wrong about this, because it is not zero. Discounting a tail print that was in fact correct means leaving a real risk unhedged, and the 3¢–8¢ band is exactly where a cheap hedge lives. The adjustment is a haircut on confidence, never a reason to skip the hedge.

Size accordingly. At $800 of incentivised depth, a tail print on Kalshi is a data point, never a position.

I set out to measure what share of headline volume is purchased. It isn’t answerable from public sources. What is answerable is which tier gets to see the terms, and the answer sits on the exchange’s own website: the subsidy published to the cent is the one Kalshi’s market makers are barred from claiming, and the subsidy set by auction is the one only they can. One test would settle the rest, and nobody has run it. Re-run that wash detector on Kalshi’s post-September-2025 tape and on Polymarket US since January, and see whether a certified volume tied rebate moves 0.97% at all. If you’ve got the data, I’d like to see it. I tried. I couldn’t.

📊 The Decision-Grade Version

Two of mine on the same beat: Chainlink Gave Polymarket 30 Seconds, on settlement manipulation at the other end of a contract’s life, and The Math of Prediction Markets.

This piece is complete on its own: the thesis, the evidence, and what would kill the view are all above, and nothing was held back to sell you a next step.

The Patreon note is a separate piece of work rather than a deeper cut of this one. It takes the pricing haircut from inside this story and writes it the way a desk would act on it: a reconstruction in six steps you can run on any event contract DCM in an hour, the crossover solved in closed form, and the capacity bound. Written for people who put capital behind a view.

→ Read the Kalshi subsidy trade note

→ Or join the Patreon community for every note

One institutional trade a week on video: The Mathematical Trader. Shorter reads and the filings I am working through go on LinkedIn.

Cover photograph: Web Summit, Diarmuid Greene, CC BY 2.0, via Wikimedia Commons.

The numbers: $8.22m moved from retail to 821 wallets across 1,613 manipulated cycles in 56 days (Dai, Jia and Yu, arXiv 2606.31675). The five-minute contract gets a 30-second Chainlink averaging window; the fifteen-minute contract, which the same paper found largely clean, gets 60 seconds (that pairing comes from Polymarket’s launch announcement, not its documentation; I treat it as the weakest link below). Polymarket is spending $550,000 of a $1m incentive programme keeping the five-minute contract liquid.

The catalyst: The new settlement rule went live at 00:00 UTC on August 7, 2026. The first clean post-change sample exists now.

Wrong if: Post-settlement price reversal in near-the-money five-minute cycles falls to zero after August 7. A collapse in the order-flow spike alone will not settle it.

I made this piece as a film. It shows the things the text can only assert: the paper’s own remedy sentence on the page, Polymarket’s developer documentation saying the settlement reference cannot be independently reproduced, and the window arithmetic worked on screen.

The consensus is nearly right, which is what makes it expensive

State the prevailing view at its strongest, because it deserves that. A binary settling on a single instantaneous print is trivially attackable: whoever holds it buys the underlying in the closing seconds and drags the print across the strike. Averaging the reference over a window forces the attacker to move the whole average instead of one tick, and the cost of doing so rises linearly with the length of that window (Aspembitova and Bentley, Entropy 2023). Polymarket replaced the snapshot with an average. Manipulation therefore gets more expensive. Every link in that chain holds, and I’ll say plainly that the change is an improvement.

My disagreement is about what got fixed, and it turns on a distinction the coverage collapsed. Dai, Jia and Yu identify two design levers and have evidence for exactly one. Polymarket pulled the other.

What the paper actually measured

Dai, Jia and Yu had an unusually clean natural experiment. Polymarket’s fifteen-minute Bitcoin up/down contract began trading on October 9, 2025 and the four-hour on October 15. The five-minute contract was created on-chain in December 2025, then sat dormant until its public launch on February 12, 2026 (rollout dates and the three regimes). Three regimes: no short contract, the longer two only, all three live. I haven’t seen a cleaner staggered rollout in a live market.

Manipulation leaves a footprint, and it appears exactly when the five-minute contract does. In the final ten seconds before each close, Binance spot order flow jumps to about 50% above the pre-launch level, concentrating exactly where a push can change the answer. In the roughly 6% of cycles still priced near even at the close, that jump runs about 3.9 times ordinary cycles. Then it reverts: within ten seconds the price gives back about a quarter of the move in near-even cycles and a tenth elsewhere (order-flow and reversal estimates). Real information persists. A push reverts.

Pushes decide outcomes, and the mechanism is not new to crypto: Société Générale ran the same trade on an equity close, pushing the print in the window that set the settlement. In near-even cycles a push against the favoured side flipped the winner 65% of the time against 41% in ordinary trading. Where the market had already given one side a 90-to-100% chance, a push reversed the outcome 34% of the time against 1%. I find that second pair the harder one to explain away. A bet the market treated as settled was overturned one time in three.

I recomputed the sample, because the cycle count carries the PnL table. A five-minute contract generates 288 cycles a day, the window runs 56 days, and the top decile of settlement-window flow is classified as manipulated. That gives 1,613. Their own count: 1,613 manipulated against 14,460 normal. It reconciles.

The number that rules out the innocent explanation

A sophisticated reader reaches for the benign story immediately, and it’s a good one. A Polymarket maker sells Up tokens to retail through the cycle, accumulates a short, and hedges it by buying spot Bitcoin. The hedge moves the close. No intent is required anywhere in that chain.

Three facts kill it, and the third is the one I’d put to a skeptic.

There is nothing to hedge in the cycles the push flips. A binary’s delta is large only near the strike. Once one side is almost certain, the contract barely moves with spot and the maker’s exposure has vanished. Yet those are precisely the cycles a push overturns one time in three. A trade placed when there is nothing to hedge cannot be a hedge.

The timing is wrong. Dynamic hedging tracks the short as it builds, so spot buying should ramp in step. It does not. Cumulative Binance flow sits near zero deep into the cycle, if anything slightly against the eventual push direction, and the final ten seconds alone carry roughly $1.7m, close to the cycle’s entire net order flow in one step.

The magnitudes are absurd. One variant survives the timing test: a maker holding naked delta through the body who hedges only as gamma surges would also trade in one late burst. So compare sizes. A short binary position can lose at most its notional, and the market-maker cohort’s short puts about $1,400 per cycle at risk. The channel would have that $1,400 exposure hedged with a $1.7m spot trade, a mismatch of three orders of magnitude. No desk hedges a four-figure maximum loss with a seven-figure position.

One line of the PnL table closes it. Manipulators earn $5,096 per manipulated cycle and $6 per normal cycle. A cohort with a real edge earns it everywhere. This one earns it only where the pushing happens.

One caveat belongs in this section, because it qualifies all of it: none of this rests on anyone having watched a single account trade both venues. I take that apart under the confounds.

Netting those two legs against each other gives a number neither the paper nor any of the coverage states, because the paper’s spot data is venue-level and cannot be attributed to wallets. I work it in full on Patreon: $5,096 of prediction-market gain against $1.7m of pushed notional is 30.0 basis points, which is the round-trip cost the push has to beat before it clears. That note carries the derivation line by line, the three ways it can fail, and what the 30-second window does to the breakeven. Binance, Polymarket, and the 30 Basis Points a Settlement Push Has to Beat.

Two levers, and only one has evidence behind it

Here is the passage the reporting skipped, and it is the whole argument:

“A robust settlement reference is a second, equally powerful lever... a time-weighted oracle, for one, raises the manipulation cost roughly in proportion to the averaging window. The two levers address manipulation from opposite sides: a longer horizon adds price discovery before settlement, while an aggregating reference spreads the settlement price across orders or time.”

Lever one is the contract horizon, and it is measured: the signature is sharp at five minutes and “much attenuated” at fifteen, across contracts trading simultaneously on the same asset through the same oracle. As controlled as field evidence gets.

Lever two is the settlement reference, and here it is theory. The proportionality claim is imported from Aspembitova and Bentley, whose linear-cost result is derived for automated market maker pools used as oracles. A Chainlink feed aggregating centralised spot exchanges is a different animal. I think the direction transfers. I wouldn’t assume the coefficient does.

Polymarket kept the five-minute horizon and changed the reference, adopting the unmeasured lever and leaving the measured one alone.

That choice wasn’t careless, and the reason matters more than the criticism. Within months of launch the five- and fifteen-minute crypto up/down markets traded over $4bn cumulatively and roughly tripled Polymarket’s daily volume. Measured lever and growth engine are the same object here. Lengthening the horizon means retiring the product that tripled the platform, and no venue does that on one preprint. Read the incentive programme in that light: $550,000 of $1m aimed at five-minute markets, $300,000 at Bitcoin alone, against $350,000 for fifteen-minute and $100,000 for four-hour (Cryptobriefing). Money defends the horizon; engineering patches the reference.

It’s a coherent commercial position and I’d probably take it too. It’s also why I don’t expect the horizon lever pulled later if the reference underdelivers. That constraint is structural, which is what makes the next sixty days of data worth reading.

Read more

The academic record now answers that precisely. The answer is yes, but the mechanisms are narrower than the headlines suggest.

🎬 Prefer to watch rather than read? A NotebookLM-generated video overview of this article is available here: Watch the video overview → Full analysis, citations, and data remain in the article below.

The full mechanism breakdown — infrastructure cost rankings for each of the three edges, the rebalancing arbitrage carried through to a verified net profit number, the three failure modes, and the precise misappropriation test SDNY applies — is in the extended note on Patreon.

The Statistical Portrait: An Informed Minority Runs the Market

The most important piece of research on prediction market alpha is a working paper by Roberto Gómez-Cram, Yunhan Guo, and Howard Kung of London Business School and Theis Ingerslev Jensen of Yale, published to SSRN on April 20, 2026 and revised April 25. The authors analyzed every Polymarket transaction from 2023 through 2025: 1.72 million accounts, 210,322 markets, approximately $13.76 billion in trading volume. They used a sign-randomization test, flipping each trader’s trade sign thousands of times to distinguish persistent directional skill from luck variance.

The result: 3.14% of accounts qualify as “skilled winners.” These roughly 54,000 accounts consistently positioned in the direction of final outcomes, carried that classification to out-of-sample markets at 44% persistence, compared to just 10% for skilled mutual funds, and traded across an average of 79 markets each. Together with market makers, they captured more than 30% of all platform gains while comprising under 3.5% of accounts. The remaining 96% either broke even by luck or lost money, effectively financing the informed minority.

The quantitative implication: a one-percentage-point increase in skilled-trader net buying corresponded to an 8 basis point increase in the probability of the correct final outcome. Skilled order flow is a leading indicator of both the next-period price move and the final resolution. The researchers separately flagged 1,950 accounts that met their timing and conviction criteria for suspected insider trading, including three accounts that took positions on the Maduro contract hours before Operation Absolute Resolve, collectively clearing more than $630,000. The DOJ criminal indictment of Van Dyke was unsealed three days after the paper’s original publication date.

Layer 1: The Structural Arbitrage That Requires No Prediction

Gómez-Cram et al. establish who wins and why. A separate paper by Saguillo, Ghafouri, Kiffer, and Suarez-Tangil at IMDEA Networks and the Oxford Internet Institute establishes how much systematic mispricing exists independently of any information edge.

Their paper, “Unravelling the Probabilistic Forest: Arbitrage in Prediction Markets,” presented at the 7th Conference on Advances in Financial Technologies (AFT 2025), analyzed 86 million bets across 17,218 Polymarket conditions using on-chain historical order book data covering April 2024 to April 2025. They found two distinct arbitrage forms.

Market Rebalancing Arbitrage (intra-market): Polymarket guarantees that the prices of all outcomes in a condition must sum to $1.00 because exactly one outcome resolves true and pays $1.00. When the combined cost of YES and NO dips below $1.00, a trader buying both sides locks in guaranteed profit regardless of resolution. The realized profit formula is:

Net Profit = (1 - Price_YES - Price_NO) - (Taker Fees + Network Gas)

Any net percentage below approximately 2% is consumed by Polygon gas and platform fees, making execution infrastructure the primary bottleneck, not identification of the opportunity.

Combinatorial Arbitrage (inter-market): Prediction markets with a logical dependency, for example “Team A wins by 2+ goals” implies “Team A wins,” create constraints on the joint state space. When markets price these correlated outcomes inconsistently, buying the logically implied position against the overpriced one generates profit. Saguillo et al. address the identification problem via a heuristic reduction from a naïve O(2^{n+m}) comparison space using topical similarity and combinatorial relationships.

The realized profit estimate across both types: approximately $40 million extracted from Polymarket across the one-year measurement period. The advantage was execution speed rather than predictive accuracy. On Polymarket’s Polygon network, effective execution requires bypassing the web UI entirely via the py-clob-client API and managing gas priority to front-run competing bots in the mempool. 14 of the 20 most profitable wallets on Polymarket’s public leaderboard are estimated to be bots. The structural alpha game is already algorithmic. Funds expecting edge without bespoke infrastructure will be on the losing side of these same mispricings.

Layer 2: Market Making as Delta-Neutral Options Replaying

Susquehanna Government Products, LLLP, a market-making affiliate of Susquehanna International Group, became the first dedicated institutional market maker on Kalshi on April 3, 2024, per the official Business Wire announcement. Kalshi CEO Tarek Mansour described the move as unlocking “institutional grade liquidity” that “changes everything.” The strategic logic is identical to Susquehanna’s options business: post continuous two-sided quotes, remain delta-neutral to event outcomes, extract the bid-ask spread per resolved contract. SIG, which traded more than $1.5 trillion in ETFs globally on an annual basis as of 2018, according to public records, and which Kalshi’s own announcement states trades approximately $2 trillion yearly, set up a dedicated trading division for event contracts, the first time a first-tier Wall Street market maker committed entirely to prediction markets.

By February 2026, Bloomberg reported that Jump Trading, which had assembled a team of approximately 20 traders dedicated to event contracts, agreed to receive equity stakes in both Kalshi and Polymarket in exchange for providing liquidity. The Kalshi deal grants a fixed equity stake while the Polymarket stake scales with the volume of trading capacity Jump provides to the platform’s U.S. operation. This is not a speculative bet. It is the proprietary trading firm’s standard playbook: take equity exposure in the exchange infrastructure you are making liquid, aligning incentives for platform volume growth with returns from liquidity provision.

In March 2026, BitGo Prime and Susquehanna Crypto launched an institutional OTC offering giving eligible hedge funds, family offices, and ultra-high-net-worth clients access to listed prediction markets using crypto or stablecoin collateral already held on BitGo’s platform. Susquehanna Crypto CEO Chase Lax called prediction markets “a genuine institutional asset class” in the official announcement.

Per the Reuters report from May 27, 2026, AQR Capital Management, Susquehanna, and OKX had all recently advertised specialist prediction market trader roles, with AQR declining to comment and Susquehanna and OKX not responding. Kalshi’s annualized trading volumes had more than tripled over the prior six months to $178 billion, with institutional trading volumes growing 800% over the same period according to the company’s head of institutional business, Andy Ross, as quoted to Reuters.

Layer 3: Cross-Asset Signals and Discrete Outcome Hedging

The core distinction between prediction market instruments and conventional derivatives is precise: options hedge price risk (how far something moves), while event contracts hedge outcome risk (whether something happens at all). They are not substitutes but complementary layers of the factor stack, addressing a risk dimension that derivatives cannot isolate directly.

Three execution-level applications are documented in institutional practice:

Cross-asset calibration: Prediction market odds for FOMC decisions, CPI prints, or nonfarm payrolls can be compared against implied probabilities extracted from Fed funds futures or rates options. Discrepancies between the two can inform position sizing around discrete events. A macro fund long duration ahead of a Fed meeting can use the Kalshi FOMC contract as a second source on market-implied probability, particularly when options skew may be distorted by hedging flows from other participants.

Event-driven pairs: A desk trading semiconductor equities can take a position on relative regulatory outcomes and delta-hedge against the correlated equity basket, isolating the regulatory risk factor in a way not achievable with conventional instruments.

Structured product packaging: In April 2026, Marex Group issued the world’s first bond-like note linked to a prediction market outcome, per Bloomberg. The $10 million issuance, sold to a Swiss institutional client and restricted to non-U.S. jurisdictions due to regulatory constraints, pays a 7% coupon if Nvidia remains the world’s largest company by market cap in one year and returns principal otherwise. Marex hedges its exposure by taking positions directly on Kalshi as odds shift, capturing the spread between the coupon offered to investors and the cost of hedging. Marex Solutions CEO Nilesh Jethwa stated the firm will “effectively build our own prediction market structured products, and then leverage Kalshi and other exchanges to replicate that”. Robert Romano, head of structured products Americas at TP ICAP, called it a potential first of many.

Separately, Tradeweb Markets took a minority stake in Kalshi to embed prediction markets into its institutional client workflows, and Clear Street partnered with Kalshi to give hedge funds direct access to event contracts.

The Surveillance Layer: Academic Detection Is Already Formalized

Three methodologically distinct approaches to identifying informed trading appeared in April-May 2026, as documented in the comparative paper “Per-Market Information Leakage and Order-Flow Skill: Two Methodological Lenses on Informed Trading in Decentralized Prediction Markets.”

The first is the Mitts-Ofir screen, from the paper “From Iran to Taylor Swift: Informed Trading in Prediction Markets“ by Joshua Mitts of Columbia Law School and Moran Ofir of the University of Haifa, published to SSRN in March 2026. Their composite screen, combining bet size anomalies, profitability, pre-event timing, and directional concentration, analyzed more than 210,000 suspicious wallet-market pairs across Polymarket from February 2024 through February 2026. Flagged traders achieved a 69.9% win rate well in excess of chance, and the authors estimated approximately $143 million in aggregate anomalous profit. The second is the Gómez-Cram sign-randomization classifier, which separately flagged 1,950 accounts as probable insiders via a lifecycle-and-conviction heuristic.

Most technically developed for market-level surveillance is Nechepurenko’s ForesightFlow framework (arXiv:2605.00493), which builds an Information Leakage Score (ILS) for binary event markets. The ILS quantifies the fraction of the terminal information move priced in before the public news event, using a Murphy decomposition connecting the score to the proper scoring rule literature. A critical structural finding: all 24 documented insider cases in the ForesightFlow Insider Cases inventory are “deadline-resolved” contracts, not news-resolved ones. The companion paper (arXiv:2605.02286) extends the framework to this class specifically, fitting exponential hazard baselines for military-geopolitics markets with a half-life of 2.9 days (KS p = 0.426), and applying them to the 2026 Iran conflict cluster.

The implication for compliance: the Mitts-Ofir dataset, the FFIC inventory, and the detection methodologies are public and citable. A regulator or sophisticated platform can now run near-real-time scoring against live order books.

The Legal Mechanism, Precisely Stated

The criminal charges against Spagnuolo and Van Dyke run on a three-element theory clearly articulated in Norton Rose Fulbright’s post-Spagnuolo analysis and the Sidley Austin analysis of the Van Dyke case: CEA Section 6(c)(1) and CFTC Rule 180.1 are modeled directly on Securities Exchange Act Section 10(b) and SEC Rule 10b-5. Because event contracts are swaps under CFTC jurisdiction, these provisions apply to prediction markets without ambiguity.

The applicable doctrine is the misappropriation theory. Per Norton Rose Fulbright’s analysis, liability attaches when an individual: (i) possesses material nonpublic information; (ii) misappropriates that information by trading on it in breach of a duty of trust and confidence owed to the source; and (iii) acts with scienter. The DOJ identifies Spagnuolo as “a software engineer at Google” in its official press release; the charge holds because the misappropriation theory does not require that the information relate to an “issuer” in the conventional securities law sense. The CFTC’s February 25, 2026 enforcement advisory confirmed this framework explicitly, and CFTC Division of Enforcement Director David Miller stated at NYU Law on March 31, 2026 that “those who hold MNPI are often subject to a web of legal and confidentiality obligations... chances are that trading on information you learn from work” crosses the line.

The money laundering counts in the Spagnuolo case are triggered by his post-settlement crypto mechanics. Per the criminal complaint as reported by BleepingComputer, on or about December 10, 2025, the AlphaRaccoon Polymarket account sent approximately 5.045 million USDC.e, representing the full account balance including the original stake and not only the $1.2M profit, to an external wallet. Blockchain data then showed subsequent transactions with a decentralized swapping service and a cryptocurrency privacy service designed to remove wallet addresses from the blockchain, before funds reached an Italian payment processor account opened with Spagnuolo’s government ID. Maximum exposure: 10 years on the commodities count and 20 years each on wire fraud and money laundering, per the DOJ press release.

The New Surface: Private-Company Contracts Expand the MNPI Population by Orders of Magnitude

Polymarket’s May 19, 2026 launch of prediction markets on private-company valuations, in partnership with Nasdaq Private Market as the exclusive resolution data provider, expands the insider-trading surface structurally. Contracts now price whether OpenAI exceeds a $1 trillion valuation by end-2026, whether Anthropic crosses $500 billion, and IPO timing for SpaceX, Stripe, Databricks, Anduril, and Kraken. Polymarket, which has done nearly $39 billion in U.S. volume in 2026 year-to-date, is explicitly positioning these as a real-time pricing signal for institutional investors.

The MNPI population for a Google Year in Search contract was the small team that produced one internal report. The MNPI population for a private-company valuation contract is everyone who touches the secondary transaction pricing, the term sheet, the resolution methodology at Nasdaq Private Market, the law firms papering the funding round, and the investors who see the cap table. Amanda Fischer, a financial policy expert at Better Markets and formerly chief of staff to SEC Chair Gary Gensler, put the concern bluntly on X: “Why is Nasdaq partnering with an offshore war gambling website to offer betting on illiquid, hard to value private companies?” Under the misappropriation theory applied by SDNY and the CFTC, an analyst who sees a secondary price print on a portfolio company and takes a position in the corresponding valuation contract has the same exposure as Spagnuolo.

What a Desk Should Actually Do With This

The alpha in prediction markets is real, academically documented, and concentrated in three mechanisms: structural arbitrage (Saguillo et al.’s $40M extracted from pure mispricing in one year), market-making spread capture (Susquehanna’s model, now extended by Jump), and cross-asset signal use and structured product packaging (Marex’s Nvidia note as the first executed proof of concept). None of these three require MNPI. All three are execution-intensive and currently capacity-constrained.

The academic surveillance infrastructure, including Mitts-Ofir’s $143 million anomalous-profit estimate across 210,000 wallet-market pairs, ForesightFlow’s open-source ILS scoring, and the Gómez-Cram sign-randomization classifier, means behavioral detection of informed trading is moving toward near-real-time flagging. The on-chain ledger has always been transparent. The detection methodology is now catching up to what that transparency makes possible.

The synthesis: the wallet-level per-trade signature is permanently visible on a public ledger, the academic detection is now formalized at more than 60 standard deviations of statistical significance, and two federal prosecutions in two months have made the enforcement priority clear. The legitimate edge is real but execution-constrained and narrow. The felony, per every DOJ and CFTC filing in the documented record, is visible from the moment the position is placed.

The Patreon note carries these three edges through their exact infrastructure cost rankings, works the rebalancing arbitrage to a verified net number with gas and capacity constraints accounted for, states the three failure modes precisely, and incorporates the peer-review critique of the exchange-reported volume figures that the overview above does not fully address.

📊 Want Deeper Quantitative Analysis?

This research took a very long time of data collection, verification, and analysis. If you found value in this deep-dive, I publish exclusive quantitative research, trading strategies, and institutional-grade analysis on Patreon.

By joining, you’ll be supporting my work and motivating me to publish more content like this.

→ Join the Patreon community here

Follow the research: YouTube — The Mathematical Trader · LinkedIn · Patreon

Cover: polymarket.com, screenshot taken 15 September 2026.

This is a detailed research piece. If you find value in institutional-quality hedge fund analysis, support this work on Patreon.

Founders Fund’s multi-layer Polymarket thesis delivers substantial returns in 18 months through converging vectors: venture appreciation ($45M Series B → ~$9B valuation), platform data monetization, and concentrated alpha capture favoring sophisticated participants. The ICE partnership transforms prediction market infrastructure into institutional information flow.

Capital Deployment & Return Profile

Series B (May 2024): Founders Fund led $45M round (BusinessWire reports $70M total disclosed across related rounds; valuation undisclosed)
ICE Strategic (October 2025): ICE invested up to $2B, reflecting ~$8B pre-money (~$9B post-money) valuation
Estimated Return Multiple: Undisclosed Series B to $9B represents significant appreciation; exact multiple depends on undisclosed entry valuation
Market Context: Industry sources estimated pre-Series B valuations in $350–600M range, implying potential 15–25x return if accurate

Founders Fund partner Joey Krug: “We developed a habit of checking Polymarket at times of breaking news.” The conviction play marks the firm’s first event-betting contract investment.

Sources: Series B Announcement, ICE Investment

Platform Metrics (2024–2025)

Election Volume: $3.3B+ presidential market (November 2024)
Total 2025 Volume: $21.5B across all markets
Monthly Active Peak: 477,850 (October 2025)
Current Open Interest: $326M (+170% YoY)
Unique Addresses: 1,733,785

Academic analyses (Columbia University/SSRN) flagged significant wash trading in 2024–25, so gross volume figures should be read with that caveat.

Market Composition (December 2025):

• Sports: 46.6% (NHL, UFC partnerships)

• Politics: 21.1%

• Crypto: 12.4%

Methodology Notes: Volume figures are industry aggregates from platform data and analytics providers. Election market volume reported between $3.0–3.6B across sources; $3.3B represents conservative estimate from multiple outlets. Academic research (Columbia University, published in Fortune) identified significant wash trading activity in 2024–2025, suggesting reported volumes may include artificial inflation. Figures presented here are gross volumes as reported by platform and industry sources.

Sources: Election Volume, ICE Partnership Details

Regulatory Timeline & Risk Mitigation

January 2022: CFTC settlement $1.4M, cease US operations
November 2024: FBI raids CEO Shayne Coplan
July 2025: DOJ/CFTC investigations closed, no charges
December 2025: Announced plans toward US relaunch; pursuing DCM licensing and regulatory compliance (final approval/filing dates should be verified from CFTC notices)

The regulatory resolution eliminates existential risk while creating barriers to entry. Polymarket’s compliance infrastructure now creates competitive moat against unauthorized platforms.

Source: Regulatory Closure

Alpha Concentration: Zero-Sum Market Structure

Profit Distribution Analysis (1.73M addresses, DeFi Oasis data):

• Top 0.04%: Captured 70%+ of $3.7B total realized profits

• Bottom 70%: Realized net losses

• Median profitable trader: $0-$1,000 range (24.56% of addresses, 0.86% of profits)

Théo Case Study (French trader, 11 accounts verified by Chainalysis):

• Capital deployed: $30–45M

• Realized profit: $48–85M range across sources (Bloomberg reports $85M; Chainalysis ~$79M; other analyses $48–65M depending on realized vs. unrealized gains and account aggregation methodology)

• Method: Proprietary “neighbor polling” (commissioned YouGov surveys in swing states)

• Win rate: 88.9% (Theo4 account)

• Known accounts: Fredi9999, Theo4, PrincessCaro, Michie + 7 others

Information arbitrage ROI: <$100K polling investment → $48–85M return per various analyses (480–850x)

Sources: Profit Distribution, Théo Analysis

Cross-Platform Arbitrage (Measured Returns)

April 2024-April 2025 Aggregate: $40M+ extracted through Polymarket-Kalshi spreads (based on IMDEA Networks Institute research analyzing 86M+ on-chain transactions)

Fee Formula Differential:

• Polymarket US: Taker = (C × P) × 0.0001

• Kalshi: Taker = roundup(0.07 × C × P × (1-P))

Example Trade:

• Polymarket YES $0.45 × 1,000 shares

• Kalshi NO $0.48 × 1,000 shares

• Net cost: $947.53 → Payout: $1,000 → Risk-free profit: $52.48 (5.5%)

Fee arbitrage exploits regulatory-driven market fragmentation.

Source: Arbitrage Strategies Analysis

ICE Partnership: Institutional Data Monetization

Strategic Value Creation:

1. Global Distribution: ICE becomes exclusive distributor of Polymarket event-driven data to institutional clients

2. Sentiment Indicators: Real-time probability signals integrated into trading infrastructure

3. Tokenization Initiatives: Joint development roadmap (unspecified)

Bloomberg Terminal integration (August 2024) preceded ICE deal, establishing institutional legitimacy. Google Finance integration (November 2025) surfaces market probabilities in search results.

The data licensing creates recurring revenue independent of trading fees, transforming prediction markets from speculative venue to information infrastructure.

Source: ICE Strategic Investment

Technical Infrastructure (Polygon L2)

Smart Contracts: Gnosis Conditional Token Framework (CTF)
Collateral: USDC.e
Order Matching: Hybrid CLOB (off-chain matching, on-chain settlement)
Oracle: UMA DVM (98% resolution without disputes)

Known Oracle Vulnerabilities:

• Ukraine Mineral Deal ($7M): UMA whale override (5M tokens, 25% voting power)

• Zelenskyy market ($200M): Resolution dispute despite evidence

• Mitigation: EigenLayer integration under exploration

CTF Mechanics:

splitPosition(): 1 USDC → 1 YES + 1 NO
mergePositions(): 1 YES + 1 NO → 1 USDC  
redeemPositions(): Winner → $1.00, Loser → $0.00

Sources: CTF Documentation, Oracle Analysis

Liquidity Provider Economics

Q-Score Formula:

Order_Score = ((max_spread - order_spread) / max_spread)² × order_size
Daily_Reward = (Your_Q_min / Total_Q_min) × Pool

Conservative MM Strategy: 40–80% APY
Aggressive MM Strategy: 80–200% APY

LP rewards subsidize tight spreads, improving platform liquidity while creating additional yield layer for sophisticated participants.

Source: Liquidity Rewards

Industry Context & Competitive Dynamics

Polymarket vs. Kalshi (2025):

• Polymarket total volume: $21.5B

• Kalshi total volume: $17.1B

• Regulatory advantage: Kalshi (US-licensed), Polymarket (offshore until December 2025)

Market TAM Projections (Certuity Research):

• 2024: $1.4B

• 2025: $44B combined volume across platforms

• 2035: $95.5B projected (46.8% CAGR)

Kalshi raised $185M Series C (June 2025) at $2B valuation. Both platforms benefit from regulatory clarity following CFTC policy shifts under Trump administration.

Sources: Market Share Analysis, Industry Projections

Value Capture Layers

Layer 1 — Venture Equity: $45M Series B (May 2024) → ~$9B valuation (October 2025) in 18 months; exact multiple undisclosed due to unannounced entry valuation
Layer 2 — Platform Economics: Fee revenue (0.01% taker US, 2% global net winnings) + ICE data licensing
Layer 3 — Ecosystem Alpha: Sophisticated participants extract $3.7B from retail flow
Layer 4 — Information Infrastructure: Network effects + regulatory moat + institutional integration

Key Takeaway

Founders Fund executes the “own the venue” thesis: capture equity appreciation while sophisticated participants extract alpha from retail order flow, with the platform layer monetizing the information spread. The ICE partnership validates prediction markets as institutional information infrastructure, not speculative betting venues.

Validated by:

• $45M Series B → ~$9B valuation in 18 months (entry valuation undisclosed)

• 70%/0.04% profit distribution confirming adverse selection

• ICE $2B investment at ~$8B pre-money (~$9B post-money)

• Regulatory investigations closed with no charges

• Institutional data integration (Bloomberg, Google, ICE)

Complete Source List

Funding & Valuation:

1. Series B — Business Wire

2. ICE Investment — ICE IR

3. Fortune Coverage

Platform Metrics: 4. Election Volume — The Block 5. Polymarket Resolution — CoinDesk 6. Wikipedia Entry

Regulatory & Compliance: 7. CFTC 2022 Settlement 8. Investigations Closure — Unchained

Profit Distribution: 9. 70% Loss Rate — Crypto News 10. Profit Concentration — Yellow

Théo Case Study: 11. Bloomberg Analysis 12. WSJ Coverage — Entrepreneur 13. Chainalysis Research

Technical Infrastructure: 14. CTF Exchange — GitHub 15. LP Rewards — Polymarket Learn

Competitive Landscape: 16. Kalshi vs Polymarket — CCN 17. Industry TAM — AInvest

Key Personnel: 18. Joey Krug Profile — Founders Fund 19. Founders Fund Portfolio

Additional Coverage: 20. The Block Series B 21. CoinDesk Series B 22. Wash Trading Research — Fortune 23. Network-Based Detection of Wash Trading — SSRN

For quantitative readers: Polymarket operates as an Arrow-Debreu complete market with on-chain verification. The 70%/0.04% distribution validates adverse selection theory. Founders Fund’s venue-layer position captures information rent through equity appreciation (Series B to $9B in 18 months) plus data monetization via ICE partnership. Platform economics favor sophisticated flow traders with information advantages or arbitrage infrastructure. Volume figures subject to wash trading caveats per Columbia/SSRN research.

📊 Support this research: https://www.patreon.com/c/NavnoorBawa

Cover photograph: Gage Skidmore, CC BY-SA 3.0, via Wikimedia Commons.

Cover photograph: Gage Skidmore, CC BY-SA 3.0, via Wikimedia Commons.

This is a detailed research piece. If you find value in institutional-quality hedge fund analysis, support this work on Patreon.

TL;DR: What Changed

Original Brier score had data leakage from using future prices in features

1. The Critical Problem with v1: Data Leakage

The original system reported 93–95% cross-validation accuracy and a Brier score of 0.022. These numbers were artificially inflated.

What Went Wrong

# v1 FLAWED: Used recent prices to predict... recent price direction
past_avg = np.mean(prices_arr[:mid_point])
recent_avg = np.mean(prices_arr[-5:])  # ← This IS the target
outcome = 1 if recent_avg > past_avg else 0

The model was trained on price momentum to predict price momentum. It learned “if price went up recently, predict UP” — a trivial pattern with no predictive edge.

v2 Fix: True Out-of-Sample Testing

# v2 CORRECT: Use ONLY pre-resolution data to predict resolution
# Target: Final outcome (YES=1, NO=0) from outcomePrices
if yp > 0.95:  # Resolved YES
    m[’resolved_yes’] = True
elif yp < 0.05:  # Resolved NO
    m[’resolved_yes’] = False

# Features: Volume, text, duration — NO price information

Key insight: On efficient prediction markets, 55–60% real accuracy is statistically significant. The 93–95% accuracy in v1 was a bug, not an edge.

2. Data Pipeline: From 100 to 9,862 Markets

API Maximization Strategy

def fetch_all_markets():
    “”“Paginate through entire Polymarket history”“”
    markets = []
    for offset in range(0, 10000, 100):
        r = requests.get(
            “https://gamma-api.polymarket.com/markets”,
            params={’closed’: ‘true’, ‘limit’: 100, ‘offset’: offset}
        )
        
        # Filter for clearly resolved markets (>95% or <5%)
        for m in r.json():
            prices = json.loads(m[’outcomePrices’])
            if prices[0] > 0.95 or prices[0] < 0.05:
                markets.append(m)
    
    return markets  # 9,862 resolved markets

Dataset Statistics

• Total resolved: 9,862 markets

• Training set: 7,889 (80%)

• Test set: 1,973 (20%)

• Class balance: 43.5% YES / 56.5% NO

• API cost: $0.00 (Gamma API is free)

3. Feature Engineering: 10 → 54 Features

v1 Features (Price-Dependent — Problematic)

# These leak future information
current_price, rsi, momentum, volatility, 
one_day_change, one_week_change

v2 Features (Resolution-Predictive — Valid)

Volume Signals (12 features)

log_vol, log_liq, vol_ratio, liq_ratio,
activity, momentum, vol_ultra, vol_high, 
vol_med, vol_low

Text Structure (17 features)

q_len, q_chars, word_diversity, avg_word_len,
has_number, has_year, has_percent, has_dollar,
has_date, starts_will, has_by, has_before, 
has_above, is_binary, has_or

Sentiment Analysis (5 features)

# Positive: win, pass, above, exceed, achieve, surge...
# Negative: lose, fail, below, crash, reject, decline...
pos_count, neg_count, sentiment, 
sentiment_abs, sentiment_total

Category Classification (12 features)

# 6 categories × 2 (binary + strength)
sports, crypto, politics, world, tech, finance

Temporal (5 features)

log_duration, dur_short, dur_med, 
dur_long, vol_per_day

Interactions (3 features)

vol_x_sentiment, activity_x_category, 
sentiment_x_duration

4. Model Architecture: Addressing Ensemble Diversity

The Valid Criticism

A reader asked: “Why build an ensemble of all tree-based models?”

Fair point. XGBoost, LightGBM, HistGradientBoosting, ExtraTrees, and RandomForest share:

• Same inductive bias (axis-aligned splits)

• Similar failure modes on extrapolation

• Correlated predictions on edge cases

v2 Response: Different Regularization = Different Models

# GB: L1/L2 regularization, slow learning
GradientBoostingClassifier(
    n_estimators=1200, learning_rate=0.012,
    min_samples_split=50, subsample=0.7
)

# RF: Bagging + feature subsampling
RandomForestClassifier(
    n_estimators=1200, max_depth=8,
    class_weight=’balanced_subsample’
)

# ET: Extreme randomization in splits
ExtraTreesClassifier(
    n_estimators=1200, max_depth=8,
    class_weight=’balanced_subsample’
)

# HGB: Early stopping + histogram binning
HistGradientBoostingClassifier(
    max_iter=1000, learning_rate=0.015,
    l2_regularization=0.25
)

New: Double Calibration

# First pass: Sigmoid (Platt scaling)
cal1 = CalibratedClassifierCV(ensemble, cv=3, method=’sigmoid’)
cal1.fit(X_train, y_train)

# Second pass: Isotonic regression
calibrated = CalibratedClassifierCV(cal1, cv=2, method=’isotonic’)
calibrated.fit(X_train, y_train)

Result: Brier score improved from 0.237 → 0.229

5. Results: High-Confidence Accuracy

The Key Metric Shift

Instead of optimizing overall accuracy (capped at ~60% on efficient markets), we target accuracy at high confidence levels.

Category Performance

Key insight: Politics and world events have clearer resolution signals (confirmable outcomes). Sports markets are noisier (game variance).

6. Live Market Analysis

Current Predictions (400 Active Markets)

Top High-Conviction Predictions:

1. “Tether insolvent in 2025?” → NO (92% confidence) | Edge: +8¢

2. “Venezuela invade Guyana in 2025?” → NO (82% confidence) | Edge: +6¢

3. “Fed abolished in 2025?” → NO (81% confidence) | Edge: +7¢

4. “North Korea invade South Korea in 2025?” → NO (80% confidence) | Edge: +5¢

Signal Distribution

• Strong NO: 65% of predictions

• Moderate NO: 20%

• Hold: 5%

• Moderate YES: 7%

• Strong YES: 3%

Interpretation: Model is bearish on “unlikely event” markets — these tend to be overpriced due to retail interest.

7. Technical Lessons Learned

1. Real Accuracy < Reported Accuracy (Usually)

Lesson: If your prediction market model exceeds 65% accuracy, check for leakage.

2. Confidence Calibration > Overall Accuracy

A model that’s 60% accurate overall but 97.8% accurate when confident is more valuable than a model that’s 75% accurate but poorly calibrated.

3. Volume + Text > Price History

For predicting resolution (not price movement):

• Volume patterns indicate informed trading

• Question structure predicts difficulty

• Category affects resolution clarity

• Price momentum predicts nothing about final outcomes

4. Tree Diversity via Regularization

Different regularization strategies create sufficient ensemble diversity:

• GB: Slow learning + high shrinkage

• RF: Bagging + feature subsampling

• ET: Random split thresholds

• HGB: Histogram binning + L2

Adding LogisticRegression or MLP improved calibration ~1% but not accuracy.

8. Code Changes Summary

- # v1: 10 features including prices
+ # v2: 54 features, no price leakage

- # v1: 100 training samples
+ # v2: 7,889 training samples

- # v1: No backtesting
+ # v2: 1,973 market backtest

- # v1: Single sigmoid calibration
+ # v2: Double calibration (sigmoid + isotonic)

- # v1: Fixed ensemble weights
+ # v2: Performance-weighted voting

- # v1: Accuracy only
+ # v2: Confidence-tiered accuracy reporting

9. Limitations & Next Steps

Current Limitations

1. No execution layer — System generates signals, doesn’t trade

2. No sentiment integration — Twitter/news could add 5–10%

3. Single-point predictions — No confidence intervals

4. Static retraining — Model doesn’t update with new resolutions

Roadmap

Priority Enhancement Expected Impact 1 Historical P&L simulation Validate edge persistence 2 Sentiment from social APIs +5–10% on political markets 3 Cross-market arbitrage Detect mispriced correlations 4 Real-time WebSocket feed Sub-second signal generation

10. Repository

GitHub: github.com/NavnoorBawa/polymarket-prediction-system

Key Files Updated

polymarket-predictor/
├── train_model_v2.py           # NEW: Production model
├── train_model_feature_select.py  # NEW: Feature selection variant
├── fetch_all_markets.py        # NEW: Maximum data collection
├── ensemble.py                 # UPDATED: Double calibration
├── data_collection.py          # UPDATED: Pagination to 10k
└── requirements.txt            # No new dependencies

Conclusion

The v2 system addresses the fundamental flaws in v1:

The key insight: On efficient markets, the edge isn’t in raw accuracy — it’s in knowing when you’re right. A model that’s 97.8% accurate when it says “≥90% confidence” is tradeable. A model that’s 60% accurate with no confidence calibration is noise.

All data from Polymarket Gamma API (free). No proprietary data sources.

This is educational content. Prediction markets involve real money and significant risk.

📊 Support this research: https://www.patreon.com/c/NavnoorBawa

Cover: polymarket.com, screenshot taken 15 September 2026.

This is a detailed research piece. If you find value in institutional-quality hedge fund analysis, support this work on Patreon.

TL;DR

I built an end-to-end ML prediction system for Polymarket that:

• Fetches 100% real data from Polymarket’s CLOB and Gamma APIs

• Trains a 5-model stacking ensemble (XGBoost, LightGBM, HistGradientBoosting, ExtraTrees, RandomForest)

• Uses probability calibration for reliable confidence estimates

• Implements fractional Kelly criterion for position sizing

• Achieved 93–95% cross-validation accuracy with Brier score of 0.022

• Signal distribution on 30 markets: 6 Strong YES, 8 Buy YES, 9 Buy NO, 5 Strong NO

1. The Problem: Prediction Markets Have Exploitable Inefficiencies

Polymarket is a decentralized prediction market where YES/NO contracts trade between $0.00 and $1.00. If a market resolves YES, YES contracts pay $1.00; otherwise, NO contracts pay $1.00.

The edge: Markets are often mispriced. Our goal is to detect when the market price diverges from the “true” probability, then bet accordingly:

Scenario Current Price Predicted Price Action Edge Underpriced YES 12.5¢ 31.8¢ BUY YES +19.3¢ Overpriced YES 85.5% 70.1% BUY NO +15.4¢

2. Architecture Overview

┌─────────────────────────────────────────────────────────────────┐
│ POLYMARKET PREDICTION SYSTEM                                    │
├─────────────────────────────────────────────────────────────────┤
│                                                                 │
│ ┌──────────────┐ ┌──────────────┐ ┌──────────────┐              │
│ │ CLOB API     │ │ GAMMA API    │ │ DATA API     │              │
│ │ (Prices,     │ │ (Markets,    │ │ (Historical  │              │
│ │ Trades,      │ │ Metadata)    │ │ Prices)      │              │
│ │ Orderbook)   │ │              │ │              │              │
│ └──────┬───────┘ └──────┬───────┘ └──────┬───────┘              │
│        │                │                │                      │
│        └────────────────┴────────────────┘                      │
│                         │                                       │
│              ┌──────────▼──────────┐                            │
│              │ PolymarketFetcher   │                            │
│              │ • Rate limiting     │                            │
│              │ • Retry logic       │                            │ 
│              │ • Caching           │                            │
│              └──────────┬──────────┘                            │
│                         │                                       │
│              ┌──────────▼──────────┐                            │
│              │ Feature Extraction  │                            │
│              │ • 10 features       │                            │
│              │ • RSI, Momentum     │                            │
│              │ • Order Imbalance   │                            │
│              └──────────┬──────────┘                            │
│                         │                                       │
│        ┌────────────────┼────────────────┐                      │
│        ▼                ▼                ▼                      │
│ ┌─────────────┐ ┌─────────────┐ ┌─────────────┐                 │
│ │ Direction   │ │ Price       │ │ Confidence  │                 │
│ │ Model       │ │ Model       │ │ Model       │                 │
│ │ (Stacking   │ │ (Stacking   │ │ (Logistic   │                 │
│ │ Ensemble)   │ │ Regressor)  │ │ Regression) │                 │
│ └──────┬──────┘ └──────┬──────┘ └──────┬──────┘                 │
│        │                │                │                      │
│        └────────────────┼────────────────┘                      │
│                         ▼                                       │
│              ┌───────────────────┐                              │
│              │ Kelly Criterion   │                              │
│              │ Position Sizing   │                              │
│              └─────────┬─────────┘                              │
│                        ▼                                        │
│              ┌───────────────────┐                              │
│              │ BUY YES/NO        │                              │
│              │ or HOLD           │                              │ 
│              └───────────────────┘                              │
└─────────────────────────────────────────────────────────────────┘

3. Data Pipeline: 100% Real Market Data

API Endpoints Used

CLOB_URL = “https://clob.polymarket.com”     # Prices, trades, orderbook
GAMMA_URL = “https://gamma-api.polymarket.com”  # Market metadata
DATA_URL = “https://data-api.polymarket.com”    # Historical prices

Fetching Strategy

I fetch training data from 100+ high-volume markets with historical price data:

def fetch_real_training_data(self, n_markets: int = 100):
    “”“
    NO SYNTHETIC DATA - uses actual historical prices and outcomes.
    “”“
    # Fetch markets ordered by 24h volume
    markets = self.get_markets(limit=n_markets * 3, order=’volume24hr’)
    
    for market in markets:
        # Get 7-day price history (hourly resolution)
        price_history = self.get_price_history(
            token_id, interval=’1h’, fidelity=170
        )
        
        # Create training sample using historical price movement
        if len(price_history) >= 20:
            prices_arr = price_history[’price’].values
            mid_point = len(prices_arr) // 2
            
            # Features: first half of data
            # Target: did price go UP or DOWN in second half?
            past_avg = np.mean(prices_arr[:mid_point])
            recent_avg = np.mean(prices_arr[-5:])
            outcome = 1 if recent_avg > past_avg else 0

Feature Vector (10 Features)

features = np.array([
    current_price,      # Current YES price (0–1)
    volume_24h,         # 24-hour trading volume ($)
    liquidity,          # Market liquidity ($)
    rsi,                # Relative Strength Index (0–1)
    momentum,           # Price momentum (change rate)
    order_imbalance,    # (buy_vol - sell_vol) / total_vol
    volatility,         # Price standard deviation
    one_day_change,     # 24h price change
    one_week_change,    # 7d price change
    spread,             # Bid-ask spread
])

4. Model Architecture: 5-Model Stacking Ensemble

Base Estimators

# XGBoost - optimized for tabular data
xgb_clf = xgb.XGBClassifier(
    n_estimators=100,
    max_depth=4,
    learning_rate=0.1,
    subsample=0.8,
    colsample_bytree=0.8,
    reg_alpha=0.1,
    reg_lambda=1.0,
    use_label_encoder=False,
    eval_metric=’logloss’
)

# LightGBM - fast gradient boosting
lgb_clf = lgb.LGBMClassifier(
    n_estimators=100,
    max_depth=4,
    learning_rate=0.1,
    num_leaves=15,
    subsample=0.8,
    colsample_bytree=0.8,
    reg_alpha=0.1,
    reg_lambda=1.0,
    verbose=-1
)

# HistGradientBoosting (sklearn native)
hist_clf = HistGradientBoostingClassifier(
    max_iter=100,
    max_depth=4,
    learning_rate=0.1
)

# ExtraTrees & RandomForest for diversity
extra_clf = ExtraTreesClassifier(n_estimators=100, max_depth=6)
rf_clf = RandomForestClassifier(n_estimators=100, max_depth=6)

Stacking Architecture

# Meta-learner combines base predictions
stacking_clf = StackingClassifier(
    estimators=[
        (’xgb’, xgb_clf),
        (’lgb’, lgb_clf),
        (’hist’, hist_clf),
        (’extra’, extra_clf),
        (’rf’, rf_clf)
    ],
    final_estimator=LogisticRegression(C=1.0, max_iter=1000),
    cv=5,
    passthrough=False
)

# Probability calibration for reliable confidence
self.direction_model = CalibratedClassifierCV(
    stacking_clf,
    method=’sigmoid’,  # Platt scaling
    cv=2
)# Probability calibration for reliable confidence
self.direction_model = CalibratedClassifierCV(
    stacking_clf,
    method=’sigmoid’,  # Platt scaling
    cv=2
)

5. The Critical Bug I Fixed: Feature Dimension Mismatch

The Problem

The model was predicting P(UP) = 0.72 for every market, regardless of input. All signals were BUY YES with no BUY NO.

Root Cause

Training used 10 features (from polymarket_fetcher.py), but prediction generated 33 features (from FeatureExtractor.combine_features()). The scaler was fitted on 33D data but received 10D at prediction time.

Training:   X.shape = (100, 10) → scaler.fit(X)
Prediction: features.shape = (1, 33) → scaler.transform(features)  # ERROR!

The Fix

I aligned both paths to use exactly 10 features:

# In predict() method:
features = np.array([
    current_price,                          # From market data
    market_features.get(’volume_24h’, 0),
    market_features.get(’liquidity’, 0),
    trade_features.get(’rsi’, 0.5),
    trade_features.get(’momentum’, 0),
    trade_features.get(’order_imbalance’, 0),
    trade_features.get(’volatility’, 0),
    trade_features.get(’momentum_5’, 0),    # 1d change proxy
    trade_features.get(’momentum_20’, 0),   # 1w change proxy
    market_features.get(’spread’, 0),
]).reshape(1, -1)

6. Prediction Logic: Direction + Magnitude

Core Algorithm

# 1. Direction Model: P(price goes UP)
direction_proba = self.direction_model.predict_proba(features_scaled)[0]
prob_up = direction_proba[1]

# 2. Price Model: Direct price prediction
raw_predicted_price = self.price_model.predict(features_scaled)[0]

# 3. Use price model as PRIMARY signal (trained on actual future prices)
price_model_move = raw_predicted_price - current_price
predicted_up = price_model_move > 0

# 4. Confidence scaling: boost if both models agree
direction_agrees = (prob_up > 0.5 and price_model_move > 0) or \
                   (prob_up < 0.5 and price_model_move < 0)

if direction_agrees:
    confidence_scale = 0.6 + direction_confidence * 0.4  # 0.6 to 1.0
else:
    confidence_scale = 0.3  # Models disagree → conservative

# 5. Cap magnitude to realistic bounds
max_move = min((1 - current_price) * 0.5, 0.20)  # Max 20%
move_magnitude = abs(price_model_move) * confidence_scale
predicted_price = current_price + (move_magnitude if predicted_up else -move_magnitude)

Action Determination

# Kelly Criterion for position sizing
if predicted_price > current_price:
    action = ‘BUY_YES’  # Price going UP → market resolves YES
else:
    action = ‘BUY_NO’   # Price going DOWN → market resolves NO

# Minimum 2% edge required
if abs(predicted_price - current_price) < 0.02:
    action = ‘HOLD’

7. Position Sizing: Fractional Kelly Criterion

class KellyCriterion:
    @staticmethod
    def calculate_full_kelly(p_true: float, p_market: float) -> float:
        “”“
        Full Kelly: f* = (P_true - P_market) / (1 - P_market)
        
        Example:
        - Model estimates 55% true probability
        - Market price is 48¢
        - f* = (0.55–0.48) / (1–0.48) = 13.4% of bankroll
        “”“
        if p_true <= p_market:
            return 0.0
        return (p_true - p_market) / (1 - p_market)
    
    @staticmethod
    def position_size(predicted_price, current_price, confidence, 
                     bankroll=1000, kelly_fraction=0.25):
        “”“
        I use QUARTER KELLY (25% of full Kelly).
        
        Why? Risk management:
        - Full Kelly: 33% chance of halving bankroll before doubling
        - Half Kelly: 11% chance
        - Quarter Kelly: 4% chance (recommended)
        “”“
        p_true = predicted_price * confidence + current_price * (1 - confidence)
        full_kelly = calculate_full_kelly(p_true, current_price)
        position = full_kelly * kelly_fraction * confidence * bankroll
        return min(position, bankroll * 0.25)  # Max 25% per trade

8. Results: Performance Metrics

Training Metrics

Metric Value Interpretation Direction Accuracy 95.0% Model predicts UP/DOWN correctly Cross-Val Accuracy 93.0% ± 4.0% Generalization check (5-fold) Brier Score 0.022 Probability quality (0 = perfect) Log Loss 0.115 Calibration metric (lower = better) Calibration Error 0.183 Predicted vs actual probabilities Class Balance 45% UP / 55% DOWN Training data distribution

Live Prediction Distribution (30 Markets)

• 🟢🟢 Strong YES: 6 (20%)

• 🟢 Buy YES: 8 (27%)

• 🔴 Buy NO: 9 (30%)

• 🔴🔴 Strong NO: 5 (17%)

• ⚪ Hold: 2 (6%)

Key observation: Balanced distribution (47% YES vs 47% NO signals) indicates the model isn’t biased toward one direction.

Top Opportunities Detected

These predictions represent markets where our model detected significant mispricings, with edges ranging from +19.3¢ to +20.0¢ per contract.

9. Key Technical Learnings

1. Feature Alignment is Critical

The most insidious bug was the feature dimension mismatch. Training on 10D features but predicting with 33D features caused the model to output near-0.5 probabilities for all inputs (maximum uncertainty).

Lesson: Always verify X_train.shape[1] == X_predict.shape[1].

2. Use Multiple Models with Different Failure Modes

When the direction model predicted all UP, the price model (trained on absolute prices) provided contrarian signals. By requiring agreement between both models for high confidence, I reduced false positives.

3. Probability Calibration Matters for Betting

Raw model probabilities are often overconfident. Using CalibratedClassifierCV with Platt scaling ensures that when the model says “70% confidence,” it’s actually correct ~70% of the time.

4. Real Data Distribution ≠ Neutral

Training data had 45% UP / 55% DOWN balance. Features like RSI and momentum had positive correlations with UP outcomes. Test markets with neutral features (RSI=0.5, momentum=0) were extrapolations — the model hadn’t seen these patterns.

10. Limitations & Future Work

Current Limitations

• No backtesting: I haven’t validated predictions against actual market resolutions

• Small training set: 100 samples may not capture all market dynamics

• Feature engineering: Only 10 features; could add sentiment, news, cross-market correlations

• No execution: System generates signals but doesn’t place trades

Future Enhancements

• Historical backtesting against resolved markets

• Cross-market arbitrage detection (mutually exclusive events)

• Sentiment integration from Twitter/news APIs

• Real-time streaming with WebSocket connections

• Portfolio optimization across multiple markets

11. Full Code Repository Structure

polymarket-predictor/
├── main.py                  # Entry point, runs predictions
├── prediction_model.py      # ML models, Kelly criterion, risk management
├── polymarket_fetcher.py    # API client with rate limiting, caching
├── requirements.txt         # Dependencies
└── .venv/                   # Python virtual environment

💻 Access the Full Code

GitHub Repository: https://github.com/NavnoorBawa/polymarket-prediction-system

The complete implementation is open-source and available on GitHub. Feel free to clone, fork, and experiment with the code!

Dependencies

• numpy

• pandas

• scikit-learn

• xgboost

• lightgbm

• shap

• requests

Conclusion

I built a functional prediction system for Polymarket that:

• Fetches real data from Polymarket’s CLOB and Gamma APIs

• Trains an ensemble of 5 gradient boosting models with probability calibration

• Generates actionable signals (BUY YES, BUY NO, HOLD) with confidence scores

• Sizes positions using fractional Kelly criterion

The key insight: prediction markets are inefficient, but exploiting them requires careful feature engineering, proper probability calibration, and rigorous testing.

The biggest lesson learned: feature dimension mismatches are silent killers. The model appeared to work but produced garbage predictions until I aligned the training and prediction pipelines.

Next step: Backtest against 6 months of resolved markets to estimate actual P&L before risking capital.

🔗 Links & Resources

Full Code on GitHub: https://github.com/NavnoorBawa/polymarket-prediction-system

Questions or want to collaborate? Feel free to reach out or open an issue on the repository!

Disclaimer: This is educational content. Prediction markets involve real money and significant risk. Past model performance does not guarantee future results. Always do your own research.

📊 Support this research: https://www.patreon.com/c/NavnoorBawa

Cover image: Polymarket market page (screenshot), public domain, via Wikimedia Commons.

This is a detailed research piece. If you find value in institutional-quality hedge fund analysis, support this work on Patreon.

Prediction market contracts are fully-collateralized binary options trading on central limit order books where the core invariant YES + NO = $1.00 creates deterministic P&L mechanics and persistent arbitrage opportunities. Edge extraction comes primarily from superior probability forecasting combined with Kelly-optimal position sizing. These are vanilla digital options priced by the market, not Black-Scholes.

The Setup: Arrow-Debreu Securities on CLOBs

Prediction markets issue binary contracts with terminal payoffs of $1.00 or $0.00. On Polymarket, each market creates two ERC-1155 tokens (YES and NO) via Gnosis Conditional Tokens Framework. On Kalshi, the first CFTC-regulated Designated Contract Market for event contracts, the same binary structure applies but with fiat settlement.

The foundational invariant is contract-enforced: P_YES + P_NO = $1.00. Violation creates immediate arbitrage. If YES trades at $0.55 and NO at $0.40 (total = $0.95), buying both locks $0.05 risk-free profit at settlement. If the sum exceeds $1.00, selling both captures the excess.

Shares are atomically minted when opposing orders match. Trader A submits limit buy YES @ $0.65, Trader B submits limit buy NO @ $0.35. The exchange collects $1.00 total, mints 1 YES + 1 NO token simultaneously, distributes to each party. Zero fractional reserves, zero counterparty credit risk, 100% collateralization.

Price discovery occurs through order book mechanics. Polymarket displays the midpoint of best bid/ask spread (or last traded price if spread exceeds $0.10). Kalshi operates similarly. Market Price ≈ Implied Probability under risk-neutral pricing (subject to risk preferences, liquidity, and platform frictions). YES @ $0.72 equals 72% market-assigned probability.

Trade Structure: CLOB Architecture and Settlement

Polymarket uses hybrid-decentralized infrastructure: off-chain order matching (gasless), EIP-712 signed orders, on-chain settlement via Polygon PoS with USDC collateral. Kalshi operates as fully-regulated exchange under CFTC oversight with fiat USD settlement.

Resolution mechanisms differ:

• Polymarket: UMA Optimistic Oracle with $750 USDC bond, 2-hour dispute window, Schelling-point arbitration by UMA token holders if contested

• Kalshi: Event contracts self-certify outcomes based on authoritative data sources (AP, NBC, government statistics)

The October 2, 2024 D.C. Circuit decision allowed Kalshi to list political event contracts and materially changed the regulatory landscape for event contracts in the United States.

Risk & P&L Mechanics: Deterministic Payoffs

Pre-resolution trading P&L follows standard equity mechanics:

P&L = (Exit_Price - Entry_Price) × Position_Size

Example: Buy 5,000 YES @ $0.28, sell @ $0.61 → $1,650 profit [($0.61 — $0.28) × 5,000].

Holding to resolution creates asymmetric convexity tied to entry price:

Low-probability tails ($0.05-$0.20) offer lottery-like payoff structures. Buy 10,000 YES @ $0.12, pay $1,200. If event occurs, receive $10,000 → $8,800 net profit (+733%). If event fails, lose entire $1,200 stake (-100%).

The zero-sum structure means every dollar won equals a dollar lost by counterparty. This is pure information transfer. Polymarket charges no trading fees on its primary platform. Kalshi charges variable fees calculated as ceil(0.07 × contracts × price × (1-price)) (fee schedule as of December 2024), ranging from ~0.6% for tail events to 1.75% at mid-market ($0.50) prices.

Key Quant Insight: Kelly Criterion for Binary Options

Edge extraction requires two components: probability estimation superiority and Kelly-optimal position sizing.

Expected value formula:

EV = p(true) × $1.00 - Market_Price

Kelly criterion determines optimal bankroll fraction:

f* = (bp - q) / b
where:
  p = true probability (your forecast)
  q = 1 - p
  b = (1 - Market_Price) / Market_Price  [net odds]

Worked Example: Market prices “Fed Rate Cut in March” at $0.60 (60% implied probability). Your econometric model forecasts 75% true probability.

b = (1 - 0.60) / 0.60 = 0.667
f* = (0.667 × 0.75 - 0.25) / 0.667 = (0.500 - 0.25) / 0.667 = 0.375

Kelly recommends 37.5% of bankroll in YES shares. EV per $1 risked = $0.75 — $0.60 = $0.15 edge (25% ROI if forecast correct).

Fractional Kelly (typically 0.25x to 0.5x full Kelly) reduces volatility and protects against probability estimation errors. Academic research by Wolfers and Zitzewitz (2006) shows prediction market prices typically approximate mean beliefs but can deviate for extreme probabilities near $0.00 or $1.00, where risk preferences distort pricing.

Arbitrage Mechanics: Cross-Platform and Intra-Market

Cross-platform arbitrage exploits price disparities between Polymarket, Kalshi, PredictIt. Research on 2024 US Presidential election markets documented persistent opportunities:

• Clinton and Huang (2025) found prices for identical contracts diverged significantly across platforms

• Accuracy rates in their sample: PredictIt 93% correctly predicted outcomes, Kalshi 78%, Polymarket 67%

• Arbitrage opportunities peaked in final two weeks before Election Day (contrary to efficiency hypothesis)

Example arbitrage:

• Polymarket YES: $0.42

• Kalshi NO: $0.56

• Total cost: $0.98

• Guaranteed payout: $1.00

• Net profit: $0.02 per contract (2.04% return)

These opportunities persist because:

1. Low institutional participation (retail-dominated flow)

2. Platform fragmentation (no unified orderbook)

3. Thin liquidity in niche markets

4. High latency between price discovery sources

Intra-platform arbitrage occurs when dependent markets misprice probability distributions. Example: If individual swing state markets sum to 87% probability while national “Trump wins” market trades at 65%, buying the underpriced distribution captures guaranteed profit when probabilities converge.

However, arbitrage profitability faces constraints:

• Platform fees: Polymarket charges no fees (main platform), Kalshi fees range 0.6–1.75% depending on price

• Slippage in thin markets

• Withdrawal/settlement delays

• Millisecond HFT competition for obvious opportunities

Market Inefficiencies: Longshot Bias and Behavioral Distortions

Prediction markets exhibit systematic mispricing patterns exploitable by quantitative traders:

Longshot Bias: Retail traders systematically overpay for low-probability outcomes ($0.01-$0.15) seeking lottery-like payoffs, while underpricing favorites. Professional traders exploit by systematically selling tail events and buying high-probability outcomes.

Recency Bias: Prices overreact to recent news then mean-revert, creating short-term reversals. Markets show negative autocorrelation in daily price changes. Momentum doesn’t persist.

Volume Distortions: National presidential markets on Polymarket/Kalshi showed highest trading volume but greater inefficiency than lower-liquidity state-level markets, suggesting attention-driven mispricing overwhelms information aggregation.

From Ng, Peng, Tao, Zhou (2025): “Using unique dataset from Polymarket, Kalshi, PredictIt, and Robinhood during 2024 U.S. election, we find significant price disparities across platforms creating economically meaningful arbitrage opportunities.”

Comparing CLOB vs AMM Architectures

Polymarket originally used Automated Market Maker (AMM) with Constant Product formula but migrated to CLOB in 2023 for superior price discovery and capital efficiency.

AMM Mechanics (legacy):

• Constant Product: x × y = k (YES shares × NO shares = constant)

• Logarithmic slippage for large orders

• Liquidity providers earn bid-ask spread

• Vulnerable to impermanent loss

CLOB Mechanics (current):

• Linear slippage based on orderbook depth: Δp ≈ Q / (2 × depth)

• Discrete price jumps from sparse orders

• Market makers post limit orders at specific prices

• No impermanent loss risk

CLOB architecture dominates for binary outcomes. Polymarket’s 2023 migration to CLOB enabled significant liquidity growth and improved price discovery versus AMM predecessors. Hybrid concentrated liquidity designs (Uniswap v3-style) attempted but suffer gap risk outside specified ranges.

Lesson: Probability Edge + Kelly Sizing = Systematic Alpha

Prediction markets are pure information bets where systematic profitability requires:

1. Probability Forecasting Superiority: Build proprietary models aggregating polling data, fundamental analysis, regression forecasts. Even 3–5% edge over market consensus compounds significantly via Kelly criterion.

2. Kelly-Optimal Position Sizing: Full Kelly maximizes long-run growth rate but creates 33% probability of halving bankroll before doubling. Fractional Kelly (0.25x-0.5x) recommended for real-world risk management.

3. Exploit Behavioral Patterns: Retail-dominated markets systematically misprice tail events. Target $0.05-$0.20 contracts where small probability improvements = large price movements.

4. Platform Arbitrage: Cross-platform price monitoring captures 1–3% risk-free returns on liquid markets. Automated systems scan for YES(A) + NO(B) < $1.00 violations.

5. Avoid Adverse Selection: Thin markets (<$50K volume) often hide informed trader activity. Large sudden moves signal private information. Fade overreactions, don’t chase momentum.

The deterministic payoff structure eliminates model risk present in vanilla options. If your probability forecast is calibrated, binary options offer the cleanest expression of edge. No volatility surface to model, no Greeks to hedge, just pure probability times payout.

Conclusion

Prediction market contracts are fully-collateralized digital options where YES + NO = $1.00 invariant enforces zero-sum structure. Alpha comes primarily from superior probability estimation combined with Kelly-optimal sizing. Retail behavioral biases and platform fragmentation create persistent inefficiencies for quantitative exploitation. These are information markets in purest form: bet better probabilities, size correctly, capture edge systematically.

Sources (All Links Verified December 2024)

Primary Platform Documentation:

1. Conditional Tokens Framework Documentation —  https://conditional-tokens.readthedocs.io/en/latest/

2. Polymarket CLOB Documentation —  https://docs.polymarket.com/developers/CLOB/introduction

3. Polymarket Price Calculation —  https://docs.polymarket.com/polymarket-learn/trading/how-are-prices-calculated

4. Polymarket Trading Fees (Zero Fees) —  https://docs.polymarket.com/polymarket-learn/trading/fees

5. Polymarket Market Resolution (UMA Oracle) —  https://docs.polymarket.com/polymarket-learn/markets/how-are-markets-resolved

6. Polymarket UMA Resolution Technical Docs —  https://docs.polymarket.com/developers/resolution/UMA

7. Kalshi Official Site —  https://kalshi.com/about

8. Kalshi Regulation Info —  https://help.kalshi.com/kalshi-101/how-is-kalshi-regulated

9. Kalshi Fee Structure —  https://help.kalshi.com/trading/fees

Academic Research:

10. Wolfers & Zitzewitz (2006), “Interpreting Prediction Market Prices as Probabilities” —  https://www.iza.org/publications/dp/2092/interpreting-prediction-market-prices-as-probabilities

11. Wolfers & Zitzewitz (2006), Full Paper PDF —  https://rodneywhitecenter.wharton.upenn.edu/wp-content/uploads/2014/04/0611.pdf

12. Meister (2024), “Application of Kelly Criterion to Prediction Markets” —  https://arxiv.org/abs/2412.14144

13. Lu & Song (2024), “Pricing Binary Options Under Mixed Exponential Jump Diffusion” —  https://doi.org/10.3390/math12203233

Market Analysis & Arbitrage:

14. Clinton & Huang (2025), “The Perils of Election Prediction Markets” —  https://goodauthority.org/news/the-perils-of-election-prediction-markets/

15. Clinton & Huang (2025), Research Paper —  https://www.researchgate.net/publication/398449904_Prediction_Markets_The_Accuracy_and_Efficiency_of_24_Billion_in_the_2024_Presidential_Election

16. Ng, Peng, Tao, Zhou (2025), “Price Discovery and Trading in Prediction Markets” —  https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5331995

17. QuantPedia, “Systematic Edges in Prediction Markets” —  https://quantpedia.com/systematic-edges-in-prediction-markets/

18. Bet Metrics Lab, “Prediction Market Arbitrage” —  https://betmetricslab.com/arbitrage-betting/prediction-market-arbitrage/

Regulatory & Industry:

19. DC Circuit Court Ruling (Oct 2, 2024): KalshiEx LLC v. CFTC —  https://media.cadc.uscourts.gov/opinions/docs/2024/10/24-5205-2077790.pdf

20. Reuters, “US Appeals Court Clears Kalshi Election Contracts” —  https://www.reuters.com/legal/us-federal-court-upholds-ruling-letting-kalshiex-list-election-betting-contracts-2024-10-02/

21. Kalshi Wikipedia (Regulatory Timeline) —  https://en.wikipedia.org/wiki/Kalshi

22. Prediction Market Wikipedia —  https://en.wikipedia.org/wiki/Prediction_market

23. KPMG, “Current State of Prediction Markets” —  https://kpmg.com/us/en/articles/2025/current-state-of-prediction-markets.html

24. CF Benchmarks, “Kalshi Leads Crypto Event Contract Market” —  https://www.cfbenchmarks.com/blog/kalshi-leads-surging-crypto-event-contract-market-powered-by-cf-benchmarks

Mathematical Foundations:

25. Kelly Criterion Wikipedia —  https://en.wikipedia.org/wiki/Kelly_criterion

26. Kelly Criterion Calculator (With Probability Verification) —  https://www.albionresearch.com/tools/kelly

27. Black-Scholes Digital Options Formulas —  https://quantpie.co.uk/bsm_bin_c_formula/bs_bin_c_summary.php

28. Black-Scholes Model Wikipedia —  https://en.wikipedia.org/wiki/Black–Scholes_model

29. Stanford, “Probability: The Kelly Criterion” —  https://crypto.stanford.edu/~blynn/pr/kelly.html

Technical Implementation:

30. Polymarket GitHub (CTF Exchange) —  https://github.com/Polymarket/ctf-exchange

31. Polymarket GitHub (UMA CTF Adapter) —  https://github.com/Polymarket/uma-ctf-adapter

32. Polymarket GitHub (CLOB Client SDK) —  https://github.com/Polymarket/clob-client

33. Gnosis Conditional Tokens Framework —  https://github.com/gnosis/conditional-tokens-contracts

Data & Analytics:

34. DeFiLlama (Polymarket TVL Data) —  https://defillama.com/protocol/polymarket

📊 Support this research: https://www.patreon.com/c/NavnoorBawa

Cover: polymarket.com, screenshot taken 15 September 2026.

This is a detailed research piece. If you find value in institutional-quality hedge fund analysis, support this work on Patreon.

Event contracts on Polymarket and Kalshi are cash-or-nothing binary options trading with retail liquidity and institutional information asymmetry. The mathematical framework is identical to derivatives pricing, but execution exploits prediction market-specific inefficiencies: wider spreads, slower information incorporation, and structural arbitrage from probability constraint violations.

Binary Options Pricing: The Foundation

Event contracts pay $1 if the outcome occurs, $0 otherwise. Under Black-Scholes framework for cash-or-nothing options:

For prediction markets with no underlying volatility (binary settlement), this simplifies to:

P = Implied_Probability
E[Payoff] = P_true - P_market

Critical Difference: Traditional options have continuous underlying price movements; event contracts resolve discretely. Edge comes from estimating P_true more accurately than crowd consensus, not from delta-hedging volatility.

Greeks for Binary Options:

Implication: Binary options exhibit discontinuous payoff functions. Gamma is negative, meaning delta hedging requires increasing position sizes as settlement approaches — creating terminal risk concentration professional traders manage through fractional Kelly position sizing.

Kelly Criterion: Optimal Position Sizing

Kelly’s 1956 formula determines optimal bet fraction maximizing long-run capital growth:

For Event Contracts:

f* = (P_true - P_market) / (1 - P_market)

Example: Model estimates 55% probability on contract at $0.48:

f* = (0.55 - 0.48) / (1 - 0.48) = 0.134 (13.4% of capital)

Fractional Kelly Implementation:

Professional traders use 25–50% of full Kelly to account for model uncertainty:

Actual_Position = Kelly_Fraction * Confidence_Factor * Capital
Typical: Position = 0.134 * 0.25 * $100,000 = $3,350

Empirical Performance:

• Full Kelly: 33% probability of halving bankroll before doubling

• Half Kelly: 11% probability of halving bankroll

• Quarter Kelly: < 3% probability (industry standard)

Note: These are illustrative probabilities under specific win/loss distributions and edge assumptions. Actual ruin probabilities vary by payout structure and trade frequency.

Mathematical Justification:

Expected growth rate under Kelly criterion:

G = E[log(1 + f*R)] ≈ p*log(1 + f*b) + q*log(1 - f)

Maximizing G yields optimal long-run wealth accumulation, but assumes perfect probability estimates. Overconfidence in P_true destroys capital exponentially — hence fractional implementation.

Cross-Contract Arbitrage: No-Arbitrage Constraint Violations

For mutually exclusive outcomes, probability sum must equal 1:

Σ P_i = 1.00 (no-arbitrage condition)

Market Rebalancing Arbitrage:

When Σ P_i ≠ 1, construct synthetic portfolio guaranteeing profit:

Example: 3-way election

Candidate A: $0.38
Candidate B: $0.33  
Candidate C: $0.27
Total: $0.98 < $1.00

Execution: Buy all three outcomes:

Cost: $0.98
Guaranteed payout: $1.00 (exactly one wins)
Profit: $0.02 (2.04% gross)
Net profit after fees (1.5%): 0.54%

Combinatorial Arbitrage:

Across related markets with logical dependencies:

Example: Presidential winner + Popular vote margin

Market 1: Candidate A wins → $0.52
Market 2: Winner by 3-5% margin → $0.28

If conditional probabilities imply:
P(A wins AND 3-5% margin) > min(P(A wins), P(3-5% margin))

Construct correlated position:
Long A wins, Long 3-5% margin
Expected value: Exploit correlation mispricing

Research Findings:

Saguillo et al. (arXiv submission Aug 2025) documented $40M realized arbitrage profit on Polymarket through:

• Market rebalancing: 60% of total

• Combinatorial arbitrage: 40% of total

Execution Constraints:

• Transaction fees (1–2%) eliminate sub-0.5% edges

• Kalshi position accountability: 25,000 contracts per strike, per member

• Simultaneous execution required (partial fills destroy arbitrage)

Order Book Microstructure: Predicting Short-Term Price Movement

Order flow imbalance (OFI) predicts near-term mid-price changes:

Static Order Book Imbalance:

OBI = (Q_bid - Q_ask) / (Q_bid + Q_ask)

Where Q represents aggregate depth at best bid/ask.

Volume-Adjusted Mid Price (VAMP):

VAMP = (P_bid * Q_ask + P_ask * Q_bid) / (Q_bid + Q_ask)

Cross-multiplying volumes weights price by liquidity on opposite side.

Empirical Predictive Power:

Cont et al. (2014) findings on order flow dynamics (applicable to prediction markets):

• OBI explains ~65% of short-interval price variance (R² = 0.65)

• Trade imbalance alone: R² = 0.32

• OBI subsumes trade imbalance information entirely

Prediction Market Application:

Imbalance ratio:

IR = Bid_volume / (Bid_volume + Ask_volume)

IR > 0.65 predicts price increase within 15–30 minutes (58% accuracy vs. 50% random).

Strategy:

if IR > 0.65:
    buy_contract()
    hold_duration = 15-30 minutes
    exit_on_momentum_reversal()

Multi-Level Order Flow:

Deep book information improves predictive accuracy. Empirical studies (sample-dependent outcomes):

• Single-level OBI: baseline predictive power

• 5-level OBI: significant improvement in R²

• 10-level OBI: further marginal gains

Each additional depth level adds information about informed trader positioning deeper in the book.

Probability Estimation: The Alpha Source

Professional market makers deploy three parallel estimation frameworks:

1. Bayesian Model Aggregation:

P_posterior = (w₁*P_polls + w₂*P_fundamentals + w₃*P_market) / Σw_i

Weight optimization via historical calibration (minimize Brier score).

2. Real-Time Information Processing:

Latency advantage comes from faster incorporation of public information. Prediction markets update on minute-to-hour timescales vs. milliseconds in equity markets.

Information Incorporation Rate:

Polymarket Nobel Peace Prize (Oct 2025): Unusual betting activity flagged winner hours before official announcement. Market-implied probability surged from ~3–4% → ~70–73% in final hours preceding the announcement, prompting regulatory investigation — demonstrating extreme alpha capture for informed participants.

3. Cross-Market Correlation Extraction:

For political contracts:

Corr(Election_outcome, VIX) ≈ -0.35
Corr(Election_outcome, USD/MXN) ≈ 0.42

Hedge political exposure using correlated financial instruments while capturing pure event risk premium.

Dynamic Hedging and Terminal Risk

Event contracts exhibit increasing gamma as settlement approaches:

Time-to-Settlement Effect:

Gamma(T) ∝ 1/√(T_remaining)

Small probability shifts cause exponential price volatility near resolution.

Risk Management Protocol:

Position sizing as function of time-to-settlement:

Position(t) = Initial_Position * √(T_remaining / T_initial)

Reduce exposure ~65% in final week before settlement to avoid terminal volatility.

Example:

Initial position: $10,000 (30 days out)
7 days remaining: $10,000 * √(7/30) = $4,830
1 day remaining: $10,000 * √(1/30) = $1,826

Market Efficiency Analysis

Accuracy Metrics:

McCullough analysis (March 2025, Dune dashboard) of Polymarket historical data:

• 90% accurate 30 days before event

• 94% accurate hours before event

• Methodology: Excludes markets with probabilities >90% or <10% to avoid skewed results

• Calibration: Slight overestimation bias (2–3%)

Comparison to Polling:

2024 U.S. Presidential Election forecasting:

• Polymarket: ~67% Trump implied probability in final days

• Polling aggregates: ~51% Trump (statistical tie)

• Outcome: Trump victory

Prediction markets demonstrated superior aggregation of non-public information signals vs. traditional survey methods.

Inefficiency Sources:

1. Noise Trader Participation: Uninformed betting dilutes information signal

2. Capital Constraints: Position limits prevent efficient arbitrage

3. Wash Trading: Columbia University research (Nov 2025) flagged ~14% of wallets with activity consistent with wash trading, accounting for 20–60% of trading volume across different periods — significantly reducing price discovery efficiency

Information Efficiency Timeline:

Sports markets show faster convergence than political markets:

• Sports (in-game): Information incorporation <5 minutes

• Politics: 15–60 minutes

• Economics: 30–180 minutes

Faster convergence = smaller alpha window for informed traders.

Systematic Return Profile

Target Performance Metrics:

Expected Annual Return: 15-25%
Sharpe Ratio: 2.0-2.8
Win Rate: 52-58%
Average Edge per Trade: 2-4%
Maximum Drawdown: 12-18%
Capital Utilization: 40-60%

Edge Composition:

Revenue sources:

• Probability estimation advantage: 60%

• Arbitrage capture: 25%

• Microstructure momentum: 15%

Portfolio Construction:

Diversification across uncorrelated event categories:

• Politics: 30%

• Sports: 30%

• Economics: 25%

• Entertainment/Pop culture: 15%

Zero correlation to traditional assets provides pure alpha.

Execution Infrastructure

Technical Requirements:

1. Real-Time Data Pipeline:

• Order book depth monitoring (100ms latency)

• Cross-platform price aggregation

• News feed integration

2. Probability Model:

• Bayesian updating framework

• Historical calibration database

• Multi-source aggregation

3. Position Management:

• Kelly calculator with confidence adjustment

• Dynamic hedging triggers

• Terminal risk protocol

4. Arbitrage Detection:

• Cross-contract probability sum monitoring

• Combinatorial relationship mapping

• Simultaneous execution capability

The Quantitative Advantage

Why Financial Quants Win:

1. Process Over Prediction: Edge accumulation across hundreds of trades, not individual accuracy

2. Transferable Infrastructure: Order flow analysis, microstructure modeling, and risk frameworks map directly from equities/derivatives

3. Scale Discipline: Markets can’t support $100M+ AUM — prevents institutional arbitrage, maintaining inefficiencies

4. Mathematical Rigor: Kelly position sizing, Greeks analysis, and correlation hedging create systematic edge

Capacity Constraints:

Individual event liquidity limits scale:

• Major political events: $50–100M

• Sports finals: $20–50M

• Economics releases: $5–15M

Total addressable market ~$500M AUM per strategy. Small enough to preserve edge, large enough for institutional-grade infrastructure.

Conclusion

Prediction market alpha derives from three mathematical foundations: binary options pricing theory determines contract value, Kelly criterion optimizes position sizing under uncertainty, and order flow microstructure predicts short-term dynamics. The edge exists not in superior forecasting but in systematic exploitation of structural inefficiencies — slower information incorporation than financial markets, wider spreads from retail liquidity, and persistent arbitrage from probability constraint violations. Execution requires quantitative infrastructure, risk discipline, and scale constraints that prevent commoditization of the strategy.

Sources

Binary Options Pricing

1. Black-Scholes Binary Options Formula — QuantPie: https://quantpie.co.uk/bsm_bin_c_formula/bs_bin_c_summary.php

2. Cash-or-Nothing Option Pricing — MATLAB Documentation: https://www.mathworks.com/help/fininst/cashbybls.html

3. Binary Option Theory — Wikipedia: https://en.wikipedia.org/wiki/Binary_option

4. Binary Options in Uncertain Markets — Yang & Gao (2023), AIMS Mathematics: https://www.aimspress.com/article/doi/10.3934/math.20231186

Kelly Criterion

1. Kelly Betting System for Favorable Games — Ferguson, UCLA: https://www.math.ucla.edu/~tom/papers/unpublished/Kelly.pdf

2. Kelly Criterion Theory — Wikipedia: https://en.wikipedia.org/wiki/Kelly_criterion

3. Kelly Criterion in Blackjack and Securities — Thorp (2007): https://web.williams.edu/Mathematics/sjmiller/public_html/341/handouts/Thorpe_KellyCriterion2007.pdf

4. Kelly Criterion Calculator — Albion Research: https://www.albionresearch.com/tools/kelly

Prediction Market Mechanics

1. What Are Prediction Markets? — Polymarket Documentation: https://docs.polymarket.com/polymarket-learn/FAQ/what-are-prediction-markets

2. Prediction Market Theory — Wikipedia: https://en.wikipedia.org/wiki/Prediction_market

3. Primer on Prediction Markets — Wharton IFPR: https://wifpr.wharton.upenn.edu/blog/a-primer-on-prediction-markets/

4. Polymarket Accuracy Analysis — McCullough (2025), Dune Dashboard: https://www.dune.com/alexmccullough/how-accurate-is-polymarket

5. Polymarket 94% Accuracy Study — Yahoo Finance: https://finance.yahoo.com/news/polymarket-achieves-94-accuracy-predicting-010322785.html

6. Polymarket 90% Accuracy Report — CoinDesk (March 2025): https://www.coindesk.com/markets/2025/03/21/polymarket-is-94-accurate-in-predicting-world-events-research

Order Book Microstructure

1. Order Flow Imbalance Price Impact — Cont, Kukanov & Stoikov (2014): https://arxiv.org/abs/1011.6402

2. Order Book Imbalance Research — Emergent Mind: https://www.emergentmind.com/topics/order-book-imbalance-obi

3. Order Flow Imbalance Overview — Emergent Mind: https://www.emergentmind.com/topics/order-flow-imbalance

4. Queue Imbalance as Price Predictor — Gould et al., Market Microstructure and Liquidity: https://www.worldscientific.com/doi/10.1142/S2382626616500064

5. Multi-Level Order-Flow Imbalance — Xu et al., Oxford Research: https://ora.ox.ac.uk/objects/uuid%3A9b7d0422-4ef1-48e7-a2d4-4eaa8a0a7ec1

6. Deep Order Flow Imbalance — Kolm, Turiel & Westray (2023), Mathematical Finance: https://onlinelibrary.wiley.com/doi/10.1111/mafi.12413

Arbitrage in Prediction Markets

1. Unravelling the Probabilistic Forest — Saguillo et al. (Aug 2025), arXiv: https://arxiv.org/abs/2508.03474

2. Arbitrage Dynamics Research — Emergent Mind: https://www.emergentmind.com/topics/arbitrage-dynamics-in-decentralized-prediction-markets

Market Structure & Events

1. ICE Investment in Polymarket — Financial Times (Oct 2025): https://www.ft.com/content/9437b5b5-59c0-44a3-8b25-58b3b4aa1b0c

2. ICE Polymarket Deal — Investopedia: https://www.investopedia.com/nyse-owner-ice-commits-dollar2-billion-to-polymarket-8741854

3. Polymarket Deep Dive — BUVCG Research: https://medium.com/buvcg-research/polymarket-deep-dive-06afa8c9a02b

4. Nobel Prize Betting Probe — Bloomberg (Oct 2025): https://www.bloomberg.com/news/articles/2025-10-10/unusual-bets-on-nobel-peace-prize-winner-prompt-probe-into-leaks

5. Prediction Market Accuracy Concerns — DL News: https://www.dlnews.com/articles/markets/polymarket-kalshi-prediction-markets-not-so-reliable-says-study/

Regulatory & Compliance

1. Kalshi Position Accountability Levels — CFTC Filing (Nov 2024): https://www.cftc.gov/sites/default/files/filings/orgrules/24/11/rules1114248729.pdf

2. Polymarket Wash Trading Investigation — Fortune (Nov 2025): https://fortune.com/2025/11/07/polymarket-wash-trading-inflated-prediction-markets-columbia-research/

All citations verified and links tested December 2025.

📊 Support this research: https://www.patreon.com/c/NavnoorBawa

Cover: polymarket.com, screenshot taken 15 September 2026.

This is a detailed research piece. If you find value in institutional-quality hedge fund analysis, support this work on Patreon.

Susquehanna International Group operates Nellie Analytics, a Dublin-based quantitative sports betting unit established in 2017. The operation focuses primarily on in-game wagering, applying the same statistical modeling techniques SIG uses in options trading to sports markets. Jane Street has also reportedly built specialized teams in this space. The trade isn’t gambling — it’s market making and statistical arbitrage executed on exchange-style order books where information efficiency creates exploitable microstructure.

The Exchange Architecture

Betfair pioneered the betting exchange model in 2000, operating as an order-driven market where bettors trade directly through continuous double auction. Unlike traditional bookmakers who take counterparty risk, the exchange provides only the platform infrastructure, charging 5% commission on net profits (reducible to 2% through reward programs) while offering complete order book transparency.

The market structure mirrors equity exchanges: “back” prices function as ask prices (buying a bet on an outcome), while “lay” prices function as bid prices (selling the bet). This creates genuine price discovery — participants can act as both liquidity providers and takers, with order books displaying depth and volume at each price level.

Market Scale: Horse racing generates the highest volumes on Betfair, with billions in annual matched bets driven by daily event frequency and pre-race market volatility. Total matched volume across all sports approaches significant institutional scale, though individual event liquidity varies dramatically — popular Premier League matches generate millions in matched bets, while smaller markets may see only tens of thousands.

Three Core Trading Strategies

1. Market Making

Traders simultaneously place back and lay orders across all possible outcomes, creating synthetic bookmaking positions. The strategy targets 105–110% total book percentage (overround) by positioning orders 1–2 ticks from mid-market prices.

P&L Mechanics:

• Revenue = (Total Matched Volume × Book Percentage Above 100%) — Exchange Commission

• On a £50,000 market with 108% book: £4,000 gross, £3,800 net after 5% commission (or £3,920 with 2% discount rate)

• Execution risk: unmatched orders leave single-sided exposure

• Requires continuous rebalancing as market moves

Technical Implementation: Automated algorithms monitor order flow, adjust quotes based on probability updates, and hedge residual risk across correlated markets. Speed matters — liquidity providers must react within milliseconds to information.

2. Cross-Book Arbitrage

Systematic identification of discrepancies where implied probability sums to less than 100% across different bookmakers or exchanges. Returns are minimal — 98% of opportunities yield below 1.2% profit — requiring substantial capital deployment.

Execution Challenges:

• Odds shift within seconds; one leg fills while the other moves

• Account limits imposed on consistent winners

• Requires simultaneous API access to multiple platforms

• Transaction costs and commission can eliminate edge

Risk Management: Position sizing must account for partial fill risk and account closure probability. Most operators identify and restrict arbitrageurs within weeks.

3. In-Play Order Flow Trading

Monitoring order book depth, bet size asymmetry, and price impact patterns to predict short-term odds movements. Similar to high-frequency trading in equities — capturing spread by providing liquidity, then hedging as momentum shifts.

Key Signals:

• Large back orders accumulating 3–5 ticks below market indicate institutional positioning

• Sudden volume spikes on specific outcomes suggest information flow

• Order book imbalances predict imminent price movement

• Correlation with live event dynamics (score changes, injuries, momentum)

Third-party analysis of Susquehanna’s approach suggested their real-time models could adjust win probabilities on microsecond timescales — for example, calculating how a single play decision impacts outcome likelihood. This capability demonstrates the sophistication of quantitative modeling in live betting markets.

Event Contracts: The New Instrument

Polymarket and Kalshi introduced prediction market event contracts — binary outcome tokens priced $0-$1 where price represents implied probability. Unlike traditional betting, these platforms use peer-to-peer exchange models where users set prices and trade with each other.

Contract Mechanics:

• Buy 100 shares at $0.40 → if outcome occurs, receive $100 (profit: $60)

• Exit pre-settlement by selling shares as probability updates

• Enables momentum trading and dynamic hedging

• Transaction fees replace traditional bookmaker margins

On October 7, 2025, Intercontinental Exchange (ICE) — owner of the New York Stock Exchange — invested up to $2B in Polymarket at an approximately $8–9B valuation, signaling institutional validation of prediction markets as a legitimate asset class. The platform trades binary contracts on sports, politics, economics — any verifiable future event. Unlike sportsbooks, profit comes from transaction fees (typically 1–2%) rather than taking the opposite side of bets.

Revenue Drivers and Risk Factors

Primary Revenue Sources:

1. Bid-ask spread capture (2–8 ticks depending on liquidity)

2. Book percentage premium above 100% in market making

3. Arbitrage edge (0.5–1.2% per opportunity when available)

4. Momentum scalping on in-play volatility

5. Transaction rebates on exchanges for liquidity provision

Critical Risks:

Account Restrictions: Bookmakers and exchanges systematically identify and limit winning traders. Susquehanna’s investment in PointsBet (12.8% stake for $65.2M USD / A$94M in June 2022) was partially strategic — securing a partner willing to accept sharp money and stand by pricing rather than limiting accounts.

Liquidity Fragmentation: Unlike financial markets with consolidated liquidity, sports betting spreads across thousands of independent events. Even horse racing — the largest category — averages only ~£500k matched per race. This prevents institutional-scale capital deployment beyond £10–20M per strategy.

Execution Risk: Centaur Galileo, an early sports betting hedge fund, collapsed in 2012 after losing $2.5M investor capital. Likely cause: over-leveraging position sizes beyond Kelly criterion constraints and inadequate risk management during losing streaks.

Market Efficiency: Research shows betting exchanges exhibit higher informational efficiency than traditional bookmakers. Unlike financial assets with volatility clustering, sports markets demonstrate light-tailed return distributions with rapidly decaying autocorrelations — information incorporates within minutes, not hours. This creates opportunity for microsecond latency arbitrage but limits slower strategy alpha decay.

The Successful Model: Priomha Capital

Priomha Capital, established in 2009 as The Cloney Multi-Sport Investment Fund, exemplifies successful institutionalization. The Australian fund achieved 118% returns by end of 2011 (versus S&P 500/ASX 200 losing 17.4% over the same period), then delivered consistent 17% average annual ROI from 2010–2015 after fees.

Operational Structure:

• Computer models for Premier League soccer, cricket, horse racing, golf, tennis

• Bets entire capital base up to twice monthly

• Quarterly audits and stringent risk management

• 30% performance fee + management fees

• Relocated to Gibraltar to expand beyond Australian regulatory limits

The fund’s success demonstrates that systematic, quantitatively-driven sports betting can generate uncorrelated alpha. However, performance fees (30%) and bet size limitations remain structural constraints — even Priomha managed only A$5M AUM initially, expanding slowly.

Why This Matters for Quant Finance

True Portfolio Diversification: Sports betting returns show zero correlation to traditional asset classes, providing genuine beta-neutral exposure. Market crashes don’t affect sports outcomes.

Transferable Microstructure Skills: Order book dynamics, market making, and statistical arbitrage concepts transfer directly from financial markets. The alpha exists in microstructure, not prediction.

Scale Constraints Define Opportunity: Fragmented liquidity across thousands of events creates persistent inefficiencies that institutional capital cannot arbitrage away. Individual UK horse races average ~£500k matched volume; even major Premier League matches rarely exceed single-digit millions in pre-match liquidity. Small, specialized teams maintain edge specifically because markets can’t support large-scale capital deployment — the very constraint that caused hedge fund structures to fail while proprietary trading shops succeeded.

Technology Convergence: As event contract platforms (Polymarket, Kalshi) scale and regulatory frameworks mature, sports betting increasingly resembles financial market infrastructure. Susquehanna’s SIG Sports division explicitly states they “provide liquidity for sports markets” using “advanced statistical forecasting models” — identical language to options market making.

The Bottom Line

Hedge funds profit from sports betting markets through the same mechanisms that drive financial market trading: market making spread capture, arbitrage execution, and order flow analysis. Susquehanna didn’t hire sports analysts — they deployed options traders and quant developers to apply existing statistical models to a new asset class.

The constraint isn’t strategy sophistication but market capacity. Liquidity fragmentation prevents whale-scale capital deployment, maintaining opportunity for specialized teams. As PointsBet executive stated when explaining Susquehanna partnership: “Our real strength is in in-play. That is the most mathematical part of sports betting, and it’s very complicated.”

Translation: the edge comes from processing large amounts of real-time data and computing confident pricing — exactly what quantitative trading firms do in options markets. Sports betting is options trading with different underlying assets.

Sources

Primary Research Sources:

1. Susquehanna/Nellie Analytics:

• eFinancialCareers: Sports betting: the new niche for quants at Jane Street, SIG and Jump Trading (Oct 2024)

• Bloomberg: A Well-Known Quant Firm Is Looking for Traders Who Want to Bet on Sports (Oct 2017)

• Susquehanna Official: SIGSports | Sports Analytics

• Sports Business Journal: Susquehanna Invests $65.2 Million in PointsBet (Jun 2022)

• Legal Sports Report: PointsBet Stock Jumps After Susquehanna Partnership (Jun 2022)

• Sportico: Susquehanna Bets $65M That Superior Pricing Will Win (Jun 2022)

• Sports Handle: PointsBet Stock Jumps After Susquehanna Investment (Jun 2022)

• Wikipedia: Susquehanna International Group

2. Betting Exchange Mechanics:

• LUISS University Thesis: Betting Exchanges: A Market Maker Process (PDF)

• Betfair Official: Betfair Charges

• Wikipedia: Betfair

• Matched Betting Blog: Betfair 2% Commission Details

• Caan Berry: Using Matched Volumes on Betfair (Dec 2018)

3. Priomha Capital Performance:

• Bloomberg: Hedge Fund Returning 17% on Sports Bets Moving to Europe (Apr 2015)

• Ivey Business Review: Over / Under — Sports Betting Funds Analysis (Aug 2021)

• Medium: Priomha Capital: A Sports Betting Hedge Fund (Jan 2023)

4. Market Failures:

• Financial Times: Investors Offered a Gamble on Sports Betting Fund (May 2017)

• Business Insider: Sports Betting Hedge Fund Collapses (Jan 2012)

• HuffPost: Centaur Galileo Sports Betting Hedge Fund Failure (Feb 2012)

5. Arbitrage & Market Structure:

• Wikipedia: Arbitrage Betting

• Sumsub: Arbitrage in Sports Betting: Detection Methods

• Arbitrage Calculator: How Arbitrage Betting Works

• OddsJam: Arbitrage & Hedge Calculator

6. Polymarket/Event Contracts:

• ICE Press Release: ICE Announces Strategic Investment in Polymarket (Oct 7, 2025)

• Wikipedia: Polymarket

• Sportico: What Are Sports Prediction Markets? (Nov 2024)

• Covers: Polymarket Launches U.S. Prediction Markets (Dec 3, 2025)

• Investopedia: NYSE Owner ICE Commits $2B to Polymarket (Oct 2025)

• Gaming Today: Polymarket Relaunches in the US (Oct 2025)

7. Academic Research:

• SSRN: Algorithmic Trading in Financial and Sports Exchanges (Aug 2024) (PDF)

• ResearchGate: Betfair Order Book Analysis

8. Additional Context:

• Sports Betting Dime: Why Are Hedge Funds Betting on Sports (Jul 2024)

• Finsmes: Hedge Funds in the Sports Sector (Feb 2024)

• Betting Websites UK: Sports Betting Hedge Funds Overview (Feb 2025)

• eFinancialCareers: Are Quant Firms Too Snobby About Sports Betting? (Dec 2024)

Statistics and performance figures verified through multiple independent sources. ICE/Polymarket data reflects October 2025 developments; other market data current through December 2024.

📊 Support this research: https://www.patreon.com/c/NavnoorBawa

Cover photograph: Alex Kinney, CC BY-SA 2.0, via Wikimedia Commons.

Cover photograph: Alex Kinney, CC BY-SA 2.0, via Wikimedia Commons.

This is a detailed research piece. If you find value in institutional-quality hedge fund analysis, support this work on Patreon.

Susquehanna Government Products, Jump Trading, and Founders Fund are actively deploying capital into Kalshi and Polymarket. With Intercontinental Exchange (NYSE parent) investing up to $2 billion and Kalshi reaching $11 billion valuation, prediction markets have transitioned from retail novelty to institutional infrastructure. Here’s how the smart money is positioning.

The Institutional Players: Who’s In

Susquehanna Government Products (SGP)

Role: First dedicated institutional market maker on Kalshi (April 3, 2024)

Susquehanna Government Products, LLLP — a market-making affiliate of the Susquehanna group — operates a dedicated trading desk exclusively focused on event contracts. The firm provides continuous two-sided liquidity across Kalshi’s political, economic, and sports markets.

Strategic Rationale: SGP’s market-making model extracts bid-ask spread while remaining delta-neutral to event outcomes. As part of one of Wall Street’s largest options market-making groups, the firm is replicating its equities playbook in a new asset class.

In late November 2025, Robinhood and Susquehanna announced a joint venture to acquire a CFTC-licensed derivatives exchange (reported variously as MIAXdx/LedgerX), with closing expected Q1 2026 — signaling deeper commitment to prediction market infrastructure.

Sources:

• BusinessWire — Kalshi Onboards First Institutional Market Maker (April 2024)

• Reuters — Robinhood, Susquehanna Take Over Exchange LedgerX (November 2025)

Jump Trading

Role: Active market maker on Kalshi (confirmed November 2025)

Bloomberg reported that Jump Trading has “quietly become one of the earliest proprietary trading firms to become active on prediction markets like Kalshi,” making markets on event contracts across elections, sports, and economic indicators.

Strategic Rationale: Jump’s crypto division (Jump Crypto) previously dominated DeFi liquidity provision before regulatory pressure in 2023. Kalshi’s CFTC-regulated structure provides a compliant venue for event-driven market making.

Source: Bloomberg — Jump Trading Joins Event-Betting Craze

Founders Fund (Peter Thiel)

Role: Lead investor in Polymarket’s $45M Series B (May 2024); in June 2025, multiple outlets reported Polymarket was nearing a $200M round led by Founders Fund at ~$1B valuation

Founders Fund has positioned Polymarket as a core portfolio company alongside Palantir, SpaceX, and OpenAI. The firm views prediction markets as a new data infrastructure layer.

Strategic Rationale: Thiel’s thesis centers on prediction markets as “truth engines” — mechanisms that aggregate dispersed information more efficiently than polls or expert forecasts.

Source: CoinDesk — Peter Thiel’s Founders Fund Backs $45M Investment in Polymarket

Citadel Securities (Peng Zhao)

Role: Personal investment by CEO Peng Zhao in Kalshi’s $185M Series C (June 2025)

While Citadel Securities hasn’t announced direct market-making on prediction platforms, CEO Peng Zhao participated in Kalshi’s funding round alongside Paradigm, Sequoia, and Multicoin.

Source: Yogonet — Kalshi Secures $185M Funding Round

Intercontinental Exchange (ICE)

Role: Up to $2 billion strategic investment in Polymarket (October 2025)

ICE announced a major strategic investment in a DeFi-adjacent platform, valuing Polymarket at approximately $8 billion pre-investment. ICE will distribute Polymarket’s event-driven data to institutional clients globally.

Strategic Rationale: ICE views prediction market data as a new sentiment indicator asset class. The partnership includes collaboration on tokenization initiatives.

Source: ICE Press Release — Strategic Investment in Polymarket

The Strategies: How They’re Making Money

1. Market Making (SGP, Jump Trading)

Mechanism: Quote continuous bid/ask prices on event contracts, capturing spread while hedging directional exposure.

P&L Drivers:

• Spread capture (typically 1–5% on illiquid contracts)

• Rebates from exchanges for liquidity provision

• Inventory management across correlated contracts

Risk Profile: Market makers are exposed to adverse selection (informed traders picking them off) and inventory risk during volatile events.

Scale: Robinhood said its product attracted ~1 million customers and ~9 billion event contracts traded in the prior 12 months. SGP provides liquidity across multiple Kalshi markets; exact flow-through volumes are not publicly disclosed.

Source: Robinhood Newsroom — Prediction Markets Joint Venture

2. Directional Information Trades (Case Study: “Théo”)

The Trade: A French trader using the pseudonym “Théo” placed more than $30 million across multiple Polymarket accounts betting Trump would win the 2024 election, popular vote, and key swing states.

The Edge: Théo commissioned proprietary “neighbor polls” asking respondents who their neighbors would vote for — a methodology designed to capture “shy Trump voters” that traditional polls missed.

Execution:

• Started August 2024 under account “Fredi9999”

• Created additional accounts (Theo4, PrincessCaro, Michie) in September–October to avoid price impact

• Over 450 bets placed during a 10-hour period from the Theo4 account alone

P&L (Disputed Across Sources): Estimates vary — mainstream outlets (WSJ/Bloomberg/Fortune) place profit nearer to ~$48–50M on ~$30–35M staked; Chainalysis (reported by The Block) estimated ~$78.7M; some outlets rounded higher to ~$85M.

Note: The disparity likely reflects different account attribution methodologies and timing of blockchain analysis.

Key Insight: The trade exploited information asymmetry between prediction market participants (sentiment-driven pricing) and rigorous polling methodology. Théo’s edge was better processing of publicly available polling artifacts, not proprietary data.

Sources:

• Wall Street Journal — How the Trump Whale Correctly Called the Election

• The Block — French Polymarket Whale ~$79M Profit Estimate (November 2024)

• Fortune — French Whale’s ~$30M Trump Bet (November 2024)

3. Tail-End Trading (Yield Extraction)

Mechanism: Buy contracts priced $0.95–$0.99 on near-certain outcomes, wait for settlement, capture the “certainty premium.”

Example:

• Contract: “Will [resolved event] happen?” trading at $0.97 with 7 days to settlement

• Return: 3.09% gross (~161% annualized, assuming 52 similar 7-day trades per year)

• Risk: Resolution delays, edge-case outcomes, platform risk

Anecdotal Evidence: Experienced Polymarket traders interviewed by BlockBeats claim ~90% of large orders ($10K+) execute at prices above $0.95. This is participant-sourced data, not an audited on-chain statistic.

Quant Insight: Tail-end trading effectively prices the time value of money plus resolution uncertainty. The implied discount rate (10–50% annualized) reflects counterparty/platform risk rather than outcome uncertainty — making this a credit spread trade disguised as a prediction bet.

Source: BlockBeats — A Silent Arbiter Profiting on Polymarket

4. Macro Overlay Hedging

Use Case: Hedge portfolio exposures to discrete policy outcomes that traditional derivatives cannot efficiently price.

Examples:

• Rate Hedging: Fund with duration exposure shorts “YES” contracts on Fed rate cuts when positioning for hikes

• Political Risk: Long defense stocks + buy “NO” contracts on candidates proposing military budget cuts

• Geopolitical: Energy-exposed portfolio hedging via event contracts on regional conflict escalation

Cost Structure: Event contracts typically cost 1–5% of notional exposure vs. options-based hedges with complex vol surface modeling. (Rule-of-thumb reported in analyst commentary; varies by market and maturity.)

Institutional Framework: Kalshi’s CFTC-regulated structure allows compliance teams to approve event contract hedging. Susquehanna’s market-making provides institutional-grade liquidity depth.

Platform Comparison: Where Institutions Trade

Volume Context:

• Kalshi: ~$50 billion annualized trading volume (mid-October 2025, per NYT/TechCrunch)

• Polymarket: ~$3.7 billion on 2024 presidential election market

Sources:

• TechCrunch — Kalshi Valuation Jumps to $11B

• CNBC — ICE Takes $2B Polymarket Stake

Key Takeaways

1. Market making is the dominant institutional strategy. SGP and Jump Trading are replicating their equities/options playbooks in event contracts — capturing spread while remaining directionally neutral.

2. Information edges compound at scale. Théo’s estimated $48–79M profit required superior polling methodology, not proprietary data. The edge was better processing of publicly available information.

3. Tail-end trading extracts credit-like returns. Near-certain contracts ($0.95+) offer annualized yields exceeding Treasury rates — reflecting settlement timing and platform risk rather than outcome uncertainty.

4. Event contracts complete the hedging market. Discrete policy outcomes (rate decisions, elections, geopolitical events) that defy continuous derivatives find natural expression in prediction markets.

5. Traditional finance is integrating fast. ICE’s up to $2B investment and Robinhood/Susquehanna’s exchange acquisition signal prediction markets are becoming core financial infrastructure.

Verified Sources

Institutional Activity:

• BusinessWire — SGP Becomes Kalshi’s First Institutional Market Maker (April 2024)

• Bloomberg — Jump Trading Joins Prediction Markets (November 2025)

• ICE Press Release — Up to $2B Polymarket Investment (October 2025)

• Reuters — Robinhood/Susquehanna Exchange Acquisition (November 2025)

Funding & Valuations:

• TechCrunch — Kalshi $11B Valuation (November 2025)

• CoinDesk — Founders Fund Leads Polymarket $45M Round (May 2024)

• Yogonet — Kalshi $185M Series C (June 2025)

Trade Case Studies:

• Wall Street Journal — How the Trump Whale Correctly Called the Election

• The Block — French Polymarket Whale ~$79M Profit Estimate (November 2024)

• Fortune — French Whale’s ~$30M Trump Bet (November 2024)

Prediction markets have crossed the institutional threshold. The question is no longer whether hedge funds will trade event contracts — it’s which strategies will generate durable alpha as liquidity deepens and spreads compress.

📊 Support this research: https://www.patreon.com/c/NavnoorBawa

Cover photograph: OtoPhuploader, CC0, via Wikimedia Commons.

This is a detailed research piece. If you find value in institutional-quality hedge fund analysis, support this work on Patreon.

Intercontinental Exchange’s October 2025 investment in Polymarket will compress prediction market spreads from 10¢ to sub-1¢ within 12–18 months, mirroring cryptocurrency exchange arbitrage elimination timelines. IMDEA research documents $39.59M retail extraction from structural mispricings — but institutional market maker infrastructure will destroy profitability for undercapitalized participants. Deploy capital aggressively in Q1-Q2 2026 or accept opportunity closure.

Verified Arbitrage Opportunity (April 2024–April 2025)

IMDEA Networks analyzed 86 million Polymarket bets, documenting $39,587,585 total arbitrage extraction across three strategies:[¹]

Single-condition arbitrage (YES + NO ≠ $1.00): $10.58M from 7,051 exploitable conditions

• Long positions (sum < $1.00): $5.90M

• Short positions (sum > $1.00): $4.68M

Multi-condition rebalancing (Σ(prices) ≠ 1.00): $28.40M extracted ($11.09M buying YES + $17.31M buying NO), demonstrating superior capital efficiency versus single-condition strategies[¹]

Combinatorial arbitrage: $95K from 5 of 13 detected pairs, with 62% failure rate due to liquidity asymmetry

Top performer: Wallet 0xd218e4… extracted $2,009,631.76 across 4,049 transactions ($496 average per trade), executing 11+ trades daily with systematic bot-like behavior[¹]

Distribution: Top 3 wallets captured $4.2M (10.6% of total); top 10 captured $8.18M (20.7%)[¹]

ICE Infrastructure Deployment

Investment structure: October 7, 2025 — ICE investing up to $2 billion in Polymarket at $8 billion pre-money valuation[²]

Strategic initiatives include global distribution of Polymarket event-driven data to institutional clients, tokenization partnerships, and integration with ICE’s regulated exchange infrastructure.

Historical precedent: ICE’s NYSE Euronext acquisition closed November 2013, with Pillar trading system rollout requiring 24–36 months for full platform integration[³]

Projected timeline (analogy-based, not confirmed):

• Q1-Q2 2026: Data feed integration, surveillance infrastructure beta

• Q3-Q4 2026: Market maker programs, institutional custody, regulatory framework

Critical uncertainty: Deployment speed depends on regulatory approval, blockchain integration complexity, and competitive pressure from Kalshi, which captured 62% market share in September 2025 versus Polymarket’s 37%[⁴]

Professional Market Maker Economics

Institutional MM cost structure (HFT industry estimates):

• Technology infrastructure: $2M–5M annually

• Quantitative personnel: $500K–1M per team

• Regulatory/compliance: $1M+ annually

• Total overhead: $3.5M–7M for dedicated prediction market desk[⁵]

Profitability threshold at $100M capital:

• Required daily P&L = $4M annual costs / 252 days = $15,873

• At 1¢ spread: 1,587,300 shares matched daily

• At 0.5¢ spread: 3,174,600 shares matched daily

Market capacity: September 2025 data shows weekly prediction market volumes exceeding $500M for Kalshi alone during September 11–17. Sufficient depth exists for institutional MMs to deploy profitably at sub-1¢ spreads.[⁴]

Retail arbitrageur costs:

• Polygon gas fees: ~$0.007 per transaction[⁶]

• Polymarket trading fees: $0[⁷]

• Minimum viable spread: ~2¢ to cover execution costs and opportunity cost

Compression mechanism: Institutional MMs can profitably operate at 0.5–1¢ spreads due to automated execution and scale. Retail participants cannot compete below 2¢ threshold due to manual execution overhead.

Oracle Risk: March 2025 UMA Governance Attack

March 24–25, 2025: $7M market “Ukraine agrees to Trump mineral deal before April?” resolved YES despite no official agreement after whale deployed ~5M UMA tokens (≈25% voting power) to influence resolution vote[⁸]

Polymarket declined refunds, calling it “unprecedented situation.”

Cross-platform risk: Polymarket uses UMA governance; Kalshi uses CFTC-regulated settlement. Oracle divergence destroys hedges if resolution mechanisms disagree.

Mitigation: Exit positions 24–48 hours pre-resolution; avoid markets with UMA token concentration >10% single holder; prioritize objective resolution criteria.

Edge Degradation Timeline

Cryptocurrency Arbitrage Precedent

Kimchi Premium compression (South Korea vs. global exchanges):

• 2016: 4.8% average spread[⁹]

• Late 2017: 30% spreads[⁹]

• January 2018: Peak 54.48%[¹⁰]

• 2019: Volatile, with periods of $1,000+ spreads returning[¹¹]

Critical limitation: Crypto arbitrage compression was constrained by capital controls (South Korea limited foreign remittances to $100K annually), not institutional MM deployment. Transfer time risk and regulatory friction prevented efficient arbitrage.[¹²]

Prediction market differences:

• No capital controls between Polymarket/Kalshi

• ICE brings institutional MM expertise directly

• Regulatory framework exists (Kalshi CFTC-approved; Polymarket negotiating)

• Oracle risk unique to prediction markets

Projected Compression (Analytical Framework)

Phase 1 (Months 0–6): Current state

• Spreads: 10–15¢ cross-platform; 2–5¢ single-market

• Retail opportunity: Maximum extraction window

• Action: Deploy 100% capital; prioritize multi-condition rebalancing

• Target: 15–25% monthly ROI (extrapolated from top performer)[¹]

Phase 2 (Months 7–12): Institutional beta

• Spreads: 3–5¢ flagship markets; 5–8¢ mid-tier; 8–12¢ long-tail

• Retail opportunity: 50–70% degradation

• Action: Pivot to N≥5 condition markets, ML whale-tracking, MM infrastructure

• Target: 5–10% monthly ROI

Phase 3 (Months 13–18): Full professionalization

• Spreads: 0.5–1¢ liquid markets; 2–4¢ illiquid

• Retail opportunity: Extinct on major markets

• Action: Institutional partnership or exit to emerging platforms

• Target: 1–3% monthly

Caveat: Timeline assumes ICE deploys at pace comparable to NYSE Euronext integration (24–36 months). Acceleration scenarios (Robinhood/Coinbase entry, regulatory fast-track) could compress to 6–12 months. Delays (regulatory obstacles, oracle incidents) could extend to 24+ months.

Regulatory Risk

September 12, 2025: Massachusetts Attorney General Andrea Joy Campbell sued Kalshi, alleging illegal sports betting operations without state gaming license. Litigation ongoing; Kalshi argues federal CFTC designation supersedes state regulation.[¹³]

Polymarket settled with CFTC in January 2022 ($1.4M penalty), agreeing to restrict U.S. access. ICE partnership may enable re-entry under regulated framework, but state-level challenges remain unresolved.[¹⁴]

Trading impact: Platform risk requires diversification across international venues (Smarkets, Betfair).

Capital Allocation Framework

Strategy 1: Aggressive Extraction (Months 0–6)

Thesis: 50–75% of 12-month profit potential concentrated in Q1-Q2 2026.

Execution:

• 100% capital deployed

• Prioritize multi-condition rebalancing (demonstrated capital efficiency advantage)[¹]

• Accept 1.5–2× leverage via stablecoin lending

• Monthly profit withdrawal to reduce platform/oracle exposure

Risk management: CVaR₉₅ <15% portfolio limit; correlation regime shift monitoring.

Strategy 2: Transition Infrastructure (Months 6–12)

Capital allocation:

• 50% arbitrage (illiquid/complex markets)

• 30% market making (dynamic spread + ML directional edge)

• 20% international platforms (regulatory arbitrage)

Infrastructure:

• WebSocket monitoring across 100+ markets (sub-second latency)

• LightGBM order flow imbalance models

• CVaR₉₅ risk system with correlation-regime detection

Network: Establish institutional MM relationships (Susquehanna, Jump Trading, Jane Street) for employment/partnership optionality.

Strategy 3: Post-Compression (Months 12+)

Option A — Employment: $150K-300K base + 20–40% bonus (tier-2 MM firms); competitive advantage from track record and infrastructure code.

Option B — Proprietary MM: $500K-1M capital requirement; 10–15% annual ROI post-compression; institutional custody and clearing partnerships required.

Option C — Geographic arbitrage: Deploy to platforms with lagged institutional entry (Smarkets UK, Azuro Protocol); accept higher regulatory/counterparty risk; edge persistence +12–18 months versus U.S. markets.

Risk Factors

Compression accelerators (6–12 month window):

• Robinhood/Coinbase prediction market launch

• CFTC expedited approval for Polymarket DCM status

• Multiple institutions deploy simultaneously (ICE + Citadel + Susquehanna)

Compression delays (18–24+ month window):

• Massachusetts lawsuit expands to Polymarket

• Additional UMA governance attacks

• Post-2024 election volume contraction (elections represented 90% of Polymarket open interest)[¹⁵]

• Blockchain/regulatory integration complexity

Quantitative Takeaway

Highest-probability scenario (60% confidence):

• Months 0–6: Current spreads persist (10–15¢)

• Months 7–12: Compression begins (3–5¢ flagship, 5–8¢ mid-tier)

• Months 13–18: Mature state (0.5–2¢ liquid markets)

Optimal retail strategy:

1. Extract 50–75% of total 12-month profit potential in Q1-Q2 2026

2. Build transition infrastructure (MM systems, ML models, institutional relationships) in Q2-Q3 2026

3. Complete pivot or exit by Q4 2026

Critical decision point: Month 6–9. Exit arbitrage too early = sacrifice 30–40% extractable profit. Exit too late = 60–80% profit decline + stranded capital.

The IMDEA top performer’s $2.01M extraction demonstrates feasibility: systematic execution at 11+ trades daily, $496 average per trade, prioritizing multi-condition rebalancing. Replication requires capital ($50K-100K minimum), automated monitoring infrastructure, and discipline (correlation-aware position sizing, oracle risk management, pre-resolution exit windows).

Window is compressing. Deploy infrastructure now or accept that institutional capital will eliminate structural inefficiencies before you build capacity to exploit them.

References

[¹]: Saguillo, O., Ghafouri, V., Kiffer, L., & Suarez-Tangil, G. (2025). “Unravelling the Probabilistic Forest: Arbitrage in Prediction Markets.” 7th Conference on Advances in Financial Technologies (AFT 2025). Available at: https://suarez-tangil.networks.imdea.org/papers/2025aft-arbitrage.pdf

[²]: Intercontinental Exchange (October 7, 2025). “ICE Announces Strategic Investment in Polymarket.” Available at: https://ir.theice.com/press/news-details/2025/ICE-Announces-Strategic-Investment-in-Polymarket/

[³]: Intercontinental Exchange (November 13, 2013). “IntercontinentalExchange Completes Acquisition of NYSE Euronext.” Available at: https://ir.theice.com/press/news-details/2013/IntercontinentalExchange-Completes-Acquisition-of-NYSE-Euronext/

[⁴]: Finance Magnates (September 20, 2025). “Kalshi Captures 60% Share, Ending Polymarket’s Prediction Market Dominance.” Available at: https://www.financemagnates.com/forex/analysis/kalshi-captures-60-market-share-ending-polymarkets-prediction-market-dominance/

[⁵]: Shallbd.com (2024). “The costs of starting a High-Frequency Trading (HFT) business.” Industry cost estimates for dedicated trading desks. Available at: https://shallbd.com/the-costs-of-starting-a-high-frequency-trading-hft-business-a-comprehensive-guide/

[⁶]: Polygon (POL) Blockchain Explorer. Average Transaction Fee statistics. Available at: https://polygonscan.com/chart/avg-txfee-usd

[⁷]: Polymarket Documentation (2025). “Trading Fees.” Available at: https://docs.polymarket.com/polymarket-learn/trading/fees

[⁸]: CoinDesk (March 27, 2025). “Polymarket, UMA Communities Lock Horns After $7M Ukraine Bet Resolves.” Available at: https://www.coindesk.com/markets/2025/03/27/polymarket-uma-communities-lock-horns-after-usd7m-ukraine-bet-resolves

[⁹]: Corporate Finance Institute. “Kimchi Premium — Overview, History, Example.” Available at: https://corporatefinanceinstitute.com/resources/cryptocurrency/kimchi-premium/

[¹⁰]: CNBC (April 3, 2024). “South Korea’s ‘kimchi premium’ in the spotlight after BTC’s record highs.” Available at: https://www.cnbc.com/2024/04/03/south-koreas-kimchi-premium-in-the-spotlight-after-btcs-record-highs.html

[¹¹]: CoinDesk (July 1, 2019). “Bitcoin’s ‘Kimchi Premium’ Returns With $1K Price Spreads on Crypto Exchanges.” Available at: https://www.coindesk.com/markets/2019/07/01/bitcoins-kimchi-premium-returns-with-1k-price-spreads-on-crypto-exchanges

[¹²]: Kim & Chang Law Firm (2023). “Revised Foreign Exchange Transactions Act.” South Korea capital control framework. Available at: https://www.kimchang.com/en/insights/detail.kc?sch_section=4&idx=26756

[¹³]: WBUR News (September 16, 2025). “Campbell sues prediction exchange platform Kalshi.” Available at: https://www.wbur.org/news/2025/09/16/massachusetts-kalshi-lawsuit-predictions-sports-wagering

[¹⁴]: CFTC (January 3, 2022). “CFTC Orders Polymarket to Cease Unauthorized Trading and Pay $1.4 Million.” Available at: https://www.cftc.gov/PressRoom/PressReleases/8478-22

[¹⁵]: DL News (November 2024). “Polymarket volumes plummet 84% after US elections.” Available at: https://www.dlnews.com/articles/markets/polymarket-volumes-and-users-plummet-after-trump-wins-election/

Methodology Note: Timeline projections based on ICE’s NYSE Euronext integration precedent (24–36 months) and professional MM economics. ROI estimates extrapolate from IMDEA top-performer data with downward adjustment for realistic capital deployment and risk management. Compression phases calibrated to institutional deployment capacity, not guaranteed outcomes.

Disclaimer: Prediction markets involve significant risk including oracle manipulation, regulatory uncertainty, and execution risk. ICE deployment timeline is estimated; actual integration may differ materially. Past performance does not guarantee future returns. Regulatory changes may accelerate or eliminate opportunities. The author may hold positions in prediction markets.

📊 Support this research: https://www.patreon.com/c/NavnoorBawa

Cover photograph: 颐园居, CC BY-SA 4.0, via Wikimedia Commons.

This article is dedicated to my maternal grandfather, whose wisdom and love have shaped my journey. His memory will continue to inspire me and guide me in all my endeavors.

IMDEA Networks analyzed 86M Polymarket bets (April 2024–April 2025), documenting $39,587,585 total arbitrage extraction. Top performer: $2,009,631.76 across 4,049 transactions averaging $496 per trade. Strategy: systematic NegRisk rebalancing. Combinatorial arbitrage captured 0.24% of profits; probability-sum deviations in multi-condition markets delivered 73%.

Market Structure: Three Arbitrage Classes

Single-condition (YES + NO ≠ $1.00): 7,051 conditions with exploitable opportunities generated $10,580,362 total extraction ($5.90M from buying below $1.00; $4.68M from selling above $1.00).

NegRisk rebalancing (Σ(prices) ≠ 1.00 across mutually exclusive outcomes): 662 markets with opportunities. Total extraction: $28,990,000 from buying YES ($11.09M), buying NO ($17.31M), and selling strategies ($612K YES, $4K NO). Average profit per opportunity: $43,800 vs $1,500 for single-condition — 29× capital efficiency advantage.

Combinatorial (cross-market dependency exploitation): 13 dependent pairs detected during 2024 election using LLM semantic analysis. 5 pairs generated realized extraction: Pair 2 ($60,237), Pair 4 ($18,472), Pair 1 ($15,819), Pair 3 ($629). Total: $95,157. Failure rate: 62% due to liquidity asymmetry and execution risk.

Note: Category sums differ from reported grand total by ~$105K due to rounding/aggregation conventions in paper appendices.

Execution Infrastructure

Transaction timing: Study grouped related bids using 950-block windows. 75% of Polymarket orders executed within this window. The paper reports “950 blocks (approx. 1 hour)”; with Polygon’s ~2s/block average (2025), this equals ≈32 minutes. Timing methodology optimized for capturing non-atomic multi-leg execution patterns.

Cost structure:

• Polygon gas fees: Operational estimate <$0.01-$0.02 per transaction based on 2024–2025 network conditions (PolygonScan average transaction fees, accessed November 2025). Fees fluctuate with network congestion; paper does not specify fixed USD gas cost.

• Polymarket platform fees: $0 (waived during study period per research methodology)

• Minimum profitable deviation: |Σ(prices) — 1.0| > $0.02 to exceed transaction costs

Position sizing constraint: Capital deployment limited by min(liquidity across all N conditions). Research documents severe disparities — top outcomes showing deep liquidity while tail conditions often constrained to low-thousand token volumes.

Strategy Distribution: Buy NO Dominance

Why Buy NO dominated: Retail systematically overprices favorites and underprices tail outcomes in multi-condition markets. Professional market makers concentrated capital in simpler binary markets, leaving NegRisk orderbooks structurally inefficient despite billion-dollar platform valuations.

Top Performer Profile (Table 1, AFT 2025)

Top 10 total: $8,184,861.26 (20.7% of ecosystem extraction). Distribution confirms power law: frequency over position size.

Top performer execution pattern:

• 4,049 trades / 365 days = 11.08 transactions daily

• Bot-like consistency suggests automated monitoring across 100+ markets

• Transaction frequency and diversity indicates systematic approach rather than discretionary trading

• Zero documented exposure to combinatorial strategies

Oracle Risk: March 2025 Case Study

UMA governance attack on $7M Ukraine mineral deal market. Whale deployed ~5M UMA tokens (~25% voting power) to influence resolution. Market resolved YES despite no official agreement. Polymarket stated resolution occurred “against the expectations of our users” but declined refunds. Documented outcomes: maximum winner $55K, maximum loser $73K.

Risk management inference: Top performer’s $2.01M extraction occurred without documented governance losses. Pattern suggests risk avoidance through:

• Market selection (objective vs subjective criteria)

• Exit timing discipline (avoiding resolution clustering)

• Platform selection (single-platform vs cross-platform exposure)

Explicit pre-resolution exit timing cannot be verified from transaction data but aligns with documented risk-management best practices for oracle-dependent markets.

Reverse-Engineered Detection Algorithm

def detect_negrisk_opportunities(markets, capital_pool):
    “”“
    Derived from documented opportunity characteristics:
    - 662 NegRisk markets with arbitrage (AFT Section 6.2)
    - Deviation threshold > $0.02 after costs
    - Liquidity constraint by minimum across conditions
    “”“
    opportunities = []
    
    for market in markets:
        # NegRisk markets only (N≥3 conditions)
        if len(market.conditions) < 3:
            continue
            
        # Probability sum deviation
        prob_sum = sum(c.price for c in market.conditions)
        deviation = abs(1.0 - prob_sum)
        
        # Filter: deviation must exceed transaction costs
        if deviation < 0.02:
            continue
            
        # Liquidity constraint
        min_liquidity = min(
            min(c.yes_liquidity, c.no_liquidity) 
            for c in market.conditions
        )
        
        if min_liquidity < 100:  # Minimum deployment threshold
            continue
            
        # Risk filters (inferred from distribution patterns)
        if market.resolution_type == ‘subjective’:
            continue
            
        # Capital efficiency scoring
        max_profit = deviation * min_liquidity
        complexity_penalty = len(market.conditions) * 2  # Orders per condition
        
        opportunities.append({
            ‘market_id’: market.id,
            ‘expected_profit’: max_profit,
            ‘efficiency_score’: max_profit / complexity_penalty,
            ‘position_size’: min(min_liquidity, capital_pool * 0.1)
        })
    
    return sorted(opportunities, 
                  key=lambda x: x[’efficiency_score’], 
                  reverse=True)

Implementation note: 11 trades/day consistency suggests automated monitoring with strict filtering. Research documents 75% of orders executing within 950 blocks (paper reports ≈1 hour; Polygon 2s/block implies ≈32 minutes), requiring rapid placement to minimize leg risk in multi-order positions.

Why Combinatorial Strategies Failed

Theoretical sophistication vs practical execution:

• LLM-based dependency detection identified 13 pairs

• Required 10+ orders per opportunity

• Cross-market coordination complexity

• Result: $95,157 (0.24% of total profits)

Failure modes documented in research:

1. Liquidity asymmetry: Primary market $500K+ depth, dependent market <$10K. Position sizing constrained to thinner venue eliminated profit potential in 62% of detected pairs.

2. Execution timing risk: Non-atomic cross-market fills. First leg moves market before second leg executes. 950-block window captures execution delays but cannot eliminate directional exposure.

3. Oracle divergence amplification: Single-market arbitrage faces one resolution event. Cross-market positions compound oracle risk. March 2025 incident demonstrated governance votes can deviate from factual outcomes.

Top performers concentrated on simple sum-price deviations exploitable through frequency rather than pursuing sophisticated cross-market strategies.

Capital Allocation Framework

Theoretical extraction potential:

• Single-condition: 7,051 opportunities

• NegRisk: 662 opportunities with 29× capital efficiency advantage

Actual extraction:

• Single-condition: $10,580,362

• NegRisk: $28,990,000

• Combinatorial: $95,157

• Total: $39,587,585*

*Category sums differ from total by ~$105K due to rounding/aggregation in paper appendices.

Top performer capture rate: 5.1% ($2.01M / $39.59M)

Success factors:

1. Automated detection across market universe

2. Rapid execution minimizing leg risk (sub-minute window)

3. Strict liquidity filters (minimum deployment thresholds)

4. Oracle risk avoidance (market selection, timing discipline)

5. Transaction cost optimization (Polygon gas <$0.02 operational estimate)

Market Evolution Timeline

October 7, 2025: Intercontinental Exchange announced up to $2B investment in Polymarket at $8B pre-money valuation. ICE (NYSE parent) becomes global distributor of Polymarket event-driven data to institutional clients. Partnership includes tokenization initiatives.

Compression trajectory — historical parallel: Cryptocurrency exchange arbitrage (2016–2018) compressed from 1,000%+ returns to <2% within 18 months of institutional market maker deployment. Kimchi Premium (Korean exchange vs global markets) peaked >50%, compressed to institutional basis as professional capital entered.

Research timing significance: Study captured peak retail inefficiency during 2024 U.S. election cycle ($3.7B total Polymarket volume). Politics markets dominated extraction: November election plus August Democratic VP pick generated largest absolute profits.

Projection (12–24 months): Institutional infrastructure deployment compresses spreads. Simple sum-price arbitrage eliminates first. NegRisk rebalancing persists longer but faces automated market maker competition. Future returns require:

• Sub-second execution infrastructure

• $100K+ capital deployment per opportunity

• Professional risk management systems

• Sophisticated cross-venue coordination

Strategic Takeaway

Top performer’s $2.01M extraction demonstrates frequency-optimized simplicity defeats sophisticated complexity during retail-to-institutional transitions.

Core framework:

• Monitor comprehensively (100+ markets)

• Execute systematically (11 trades/day consistency)

• Avoid tail risk (oracle governance, subjective criteria, resolution clustering)

• Optimize transaction costs (Polygon gas <$0.02 operational, zero platform fees)

• Scale through frequency over position size

Pattern transferability: Framework applies to any microstructure inefficiency during market structure professionalization. As institutional capital enters (ICE $2B investment signals inflection point), edges compress predictably following historical trajectories observed in cryptocurrency exchange arbitrage evolution.

Window compression analysis: Research captured optimal extraction period. Post-October 2025, institutional market maker deployment accelerates spread compression. High-frequency retail arbitrage window closes as professional infrastructure stabilizes orderbooks and automated strategies eliminate simple mispricings.

Sources

[1] Saguillo, O., Ghafouri, V., Kiffer, L., & Suarez-Tangil, G. (2025). “Unravelling the Probabilistic Forest: Arbitrage in Prediction Markets.” 7th Conference on Advances in Financial Technologies (AFT 2025). Leibniz International Proceedings in Informatics, Vol. 354, pp. 27:1–27:24. DOI: 10.4230/LIPIcs.AFT.2025.27. https://drops.dagstuhl.de/entities/document/10.4230/LIPIcs.AFT.2025.27

[2] DL News. (August 25, 2025). “Polymarket users lost millions of dollars to ‘bot-like’ bettors over the past year, study finds.” https://www.dlnews.com/articles/markets/polymarket-users-lost-millions-of-dollars-to-bot-like-bettors-over-the-past-year/

[3] CoinDesk. (March 27, 2025). “Polymarket, UMA Communities Lock Horns After $7M Ukraine Bet Resolves.” https://www.coindesk.com/markets/2025/03/27/polymarket-uma-communities-lock-horns-after-usd7m-ukraine-bet-resolves

[4] Yahoo Finance. (March 27, 2025). “Polymarket Suffers UMA Governance Attack After Rogue Actor Becomes Top-5 Token Staker.” https://finance.yahoo.com/news/polymarket-suffers-uma-governance-attack-101646076.html

[5] Intercontinental Exchange. (October 7, 2025). “ICE Announces Strategic Investment in Polymarket.” https://ir.theice.com/press/news-details/2025/ICE-Announces-Strategic-Investment-in-Polymarket/

[6] PolygonScan. (November 2025). “Average Daily Transaction Fee Chart.” Network statistics for Polygon PoS. https://polygonscan.com/chart/avg-txfee-usd

Verification Notes

Data sourcing: All numerical claims traced to IMDEA Networks AFT 2025 conference paper [1]:

• Table 1: Top 10 performer data (Section 7.4)

• Section 6.1: Single-condition opportunities (7,051 conditions)

• Section 6.2: NegRisk opportunities (662 markets)

• Section 6.3: Combinatorial pairs (13 detected, 5 profitable)

• Section 7: Extraction breakdowns by strategy

Timing conversions: Paper states “950 blocks (approx. 1 hour)” for execution window. Polygon block time ~2 seconds (2025 network average per PolygonScan [6]) yields ≈32 minutes mathematical conversion. Both figures provided for transparency.

Cost estimates: Gas fees presented as operational estimates based on 2024–2025 network conditions. Paper does not specify fixed USD gas costs; values fluctuate with network congestion.

Category reconciliation: Sum of per-category extractions ($10.58M + $28.99M + $0.095M = $39.665M) differs from reported total ($39.588M) by ~$105K, attributed to rounding/aggregation conventions in paper appendices. This minor variance does not affect strategic conclusions.

Oracle risk analysis: March 2025 UMA governance incident details from contemporaneous reporting [3,4]. Behavioral inferences about top performer risk management derived from absence of documented governance losses rather than explicit transaction timing data.

Cover photograph: Thehinse, CC BY-SA 4.0, via Wikimedia Commons.

IMDEA Networks analyzed 86 million Polymarket bets, documenting systematic arbitrage extraction across 17,218 market conditions.[1] Market rebalancing within multi-condition NegRisk markets generated 73% of total profits despite representing 8.6% of opportunities — revealing a 29× capital efficiency advantage over binary arbitrage that most participants missed entirely.

The Structural Inefficiency

NegRisk markets: N≥3 mutually exclusive conditions where Σ(prices) must equal 1.0.

Mispricing mechanism: Liquidity fragments across outcomes. Retail flow concentrates on 1–2 favorites. Complementary probability space trades thin. Result: Σ(prices) ≠ 1.0.

Example:

• Trump wins: 0.48

• Harris wins: 0.46

• Other: 0.03

• Sum: 0.97 (3¢ arbitrage per dollar deployed)

When Σ(prices) <1.0 → buy underpriced conditions
When Σ(prices) >1.0 → sell overpriced conditions

At resolution, exactly one condition pays $1.00. Profit = |1.00 — Σ(purchase prices)|.

P&L Distribution: The Buy NO Asymmetry

Market rebalancing (multi-condition): $28.99M

• Buy NO positions: $17.31M (60%)

• Buy YES positions: $11.09M (38%)

• Sell strategies: $4.88M (2%)

Single-condition arbitrage: $10.58M

• Buy both (sum <$1.00): $5.90M

• Sell both (sum >$1.00): $4.68M

Combinatorial arbitrage: $95K

• 13 dependent pairs detected via LLM semantic analysis

• 5 pairs profitable

• 62% failure rate due to liquidity asymmetry, execution timing risk

Key insight: Buying NO dominated because retail systematically overprices favorites and underprices tail outcomes. Market makers absent in fragmented multi-condition orderbooks where capital deployment constrained by min(liquidity across all N conditions).

Capital Efficiency Analysis

Execution complexity:

• Binary: 2 orders

• NegRisk: 4–8 orders (scales with N conditions)

• Combinatorial: 10+ orders across multiple markets

Despite 3× higher implementation complexity, NegRisk rebalancing delivered 29× better capital efficiency per opportunity.

Top arbitrageur performance:
$2.01M across 4,049 transactions. Average: $496 per trade. Strategy: frequency over position size. Bot-like execution prioritizing rebalancing and single-condition opportunities over combinatorial strategies.

Execution Constraints

Liquidity fragmentation:
Research documents large disparities in orderbook depth. Top outcomes exhibit deep liquidity while many tail conditions show only modest token volume — often low-thousand token volumes during opportunistic windows. Position sizing capped by minimum liquidity across conditions.[1]

Timing risk:
Non-atomic execution. Research documents 75% of Polymarket orders execute within 950 blocks (~1 hour on Polygon).[1] Multi-leg positions face compounded price movement risk between fills.

Transaction costs:
Polygon PoS gas costs during study period were typically fractions of a cent up to a few cents per transaction. Polymarket did not charge per-trade fees during measurement window; AFT authors therefore do not model platform fees in P&L calculations.[1][4] Minimum profitable deviation: |Σ(prices) — 1.0| > 0.02 to maintain positive expected value.

Oracle Risk: March 2025 Governance Attack

Market: “Ukraine agrees to Trump mineral deal before April?”
Volume: $7M
Attack vector: Whale holder with 1.3M UMA tokens (top-5 staker, ~25% voting power) influenced resolution despite no official agreement.

Outcome:
Market resolved to YES when probability was 9%.
Largest winner: $55,000
Largest loser: $73,000[5]

Risk profile:
Single-market arbitrage: one oracle event
Multi-condition positions: single oracle determines entire portfolio
Cross-platform hedges: compounded risk if Polymarket (UMA governance) and Kalshi (CFTC-regulated) resolve differently

Mitigation: Avoid cross-platform dependent positions unless spread exceeds 15¢ to buffer oracle divergence.

Market Microstructure: Why Inefficiency Persists

Retail behavior:
Anchors on favorites. Ignores complementary probability space. Buys YES on high-probability outcomes at inflated prices. Underweights tail conditions.

Market maker absence:
Professional MM capital concentrates in binary markets: higher volume, tighter spreads, simpler hedging. Multi-condition markets with N>4 outcomes remain structurally inefficient despite billion-dollar platform valuations.

Detection signal:
Markets with N≥4 conditions where retail attention concentrates on top 2 outcomes systematically misprice probability distributions. Monitor for |Σ(prices) — 1.0| deviations >2¢.

Institutional Capital Timeline

October 7, 2025: Intercontinental Exchange invested up to $2B in Polymarket at $8B pre-money valuation.[2] NYSE parent becomes global distributor of prediction market event-driven data. (Post-study context; AFT measurement window ended April 1, 2025.)

September 11–17, 2025: Kalshi captured 62% prediction market share, processing $500M+ weekly volume with $189M average open interest. Polymarket: 37% share, $430M weekly volume.[3] Kalshi operates as CFTC-designated contract market.

Compression trajectory:
Historical parallel: Early crypto exchange arbitrage (2016–2018) generated 1,000%+ returns. Kimchi Premium compressed from 50%+ to <2% within 18 months of institutional market maker deployment. Prediction market arbitrage follows identical evolution.

Research timing:
Study captured peak retail inefficiency during 2024 U.S. election cycle ($3.7B Polymarket volume). Post-ICE investment, spread compression inevitable as institutional infrastructure professionalizes market structure.

Quantitative Framework

Detection algorithm:

def scan_negrisk_opportunities(markets):
    opportunities = []
    for market in markets:
        if len(market.conditions) < 3:
            continue  # Binary markets use standard arbitrage
        
        prob_sum = sum(c.price for c in market.conditions)
        deviation = abs(1.0 - prob_sum)
        
        if deviation > 0.02:  # 2¢ minimum after costs
            total_liquidity = sum(
                min(c.yes_liquidity, c.no_liquidity) 
                for c in market.conditions
            )
            max_profit = deviation * total_liquidity
            
            if max_profit > 100:  # $100 minimum threshold
                opportunities.append({
                    ‘market’: market.id,
                    ‘deviation’: deviation,
                    ‘direction’: ‘buy’ if prob_sum < 1.0 else ‘sell’,
                    ‘max_profit’: max_profit,
                    ‘complexity’: len(market.conditions) * 2
                })
    
    return sorted(opportunities, key=lambda x: x[’max_profit’], reverse=True)

Position sizing:

Capital allocation proportional to liquidity across all conditions. Maximum extractable profit = min(liquidity₁, …, liquidityₙ) × |Σ(prices) — 1.0|.

Execution requirements:

1. Monitor 100+ markets simultaneously

2. Accept 1–5% per-trade margins

3. Scale through frequency (4,000+ transactions/year)

4. Deploy sub-5-second execution to minimize leg risk

Strategic Takeaway

Market rebalancing captured 73% of arbitrage profits ($29M) despite 10.6× fewer opportunities than single-condition arbitrage (662 vs 7,051). Capital efficiency: 29× superior per opportunity.

Why the edge persists:

• Retail flow systematically misprices probability distributions across N>2 outcomes

• Professional market makers concentrate capital in simpler binary markets

• Multi-condition orderbooks lack institutional liquidity provision

• Execution complexity deters automated strategies optimized for binary arbitrage

Window compression:
ICE’s $2B investment (October 2025) marks institutional inflection point. Research captured peak inefficiency during 2024 election cycle. Future returns require: faster execution (sub-second), larger capital ($100K+ per trade), sophisticated cross-venue risk management.

Actionable framework:

• Prioritize markets with N≥4 conditions where attention concentrates on top 2

• Deploy when |Σ(prices) — 1.0| > 0.02 after accounting for execution costs

• Size positions by minimum liquidity across all conditions

• Accept 4–8 orders per opportunity versus 2 for binary arbitrage

• Exit before resolution clustering when oracle risk compounds

The documented extraction demonstrates that while retail obsesses over YES+NO≠$1.00 in binary markets, the quantitative edge exists where probability must distribute across many outcomes. Liquidity fragmentation creates structural deviations that persist for hours despite platform scale. Multi-dimensional arbitrage dominates binary strategies when market makers remain absent and retail flow concentrates on favorites.

References

[1] Saguillo, O., Ghafouri, V., Kiffer, L., & Suarez-Tangil, G. (2025). “Unravelling the Probabilistic Forest: Arbitrage in Prediction Markets.” 7th Conference on Advances in Financial Technologies (AFT 2025). Leibniz International Proceedings in Informatics, Volume 354, pp. 27:1–27:24. https://doi.org/10.4230/LIPIcs.AFT.2025.27

[2] Intercontinental Exchange. (2025, October 7). “ICE Announces Strategic Investment in Polymarket.” https://ir.theice.com/press/news-details/2025/ICE-Announces-Strategic-Investment-in-Polymarket/

[3] Rodriguez, F. (2025, September 20). “Kalshi Outpaces Polymarket in Prediction Market Volume Amid Surge in U.S. Trading.” CoinDesk. https://www.coindesk.com/markets/2025/09/20/kalshi-outpaces-polymarket-in-prediction-market-volume

[4] PolygonScan. (2025). “Average Daily Transaction Fee Chart.” https://polygonscan.com/chart/avg-txfee-usd

[5] Khatri, Y. (2025, March 27). “Polymarket Suffers UMA Governance Attack After Rogue Actor Becomes Top-5 Token Staker.” Yahoo Finance. https://finance.yahoo.com/news/polymarket-suffers-uma-governance-attack-101646076.html

Source notes: Empirical claims based on AFT 2025 conference paper (measurement period: April 1, 2024 → April 1, 2025; dataset: ~86M bids, 17.2K conditions). Key verified figures: total extraction $39,587,585.02; NegRisk rebalancing $28.99M (662 opportunities) vs single-condition $10.58M (7,051 opportunities); 13 dependent pairs detected (5 profitable, ~$95K); top account $2,009,631.76 across 4,049 transactions; 75% of bids within 950-block window; detection threshold |1 − VWAP Sum| > $0.02. ICE/Kalshi timeline items (Sept-Oct 2025) are post-study context from press reporting.

Cover: polymarket.com, screenshot taken 15 September 2026.

IMDEA Networks researchers analyzed 86 million Polymarket bids (OrderFilled events), identifying 13 logically dependent market pairs during the 2024 U.S. election using LLM-based semantic analysis. Only 5 pairs generated realized arbitrage profits totaling $95,157. The 62% failure rate exposes structural execution barriers that make combinatorial strategies inferior to single-market arbitrage despite theoretical sophistication. Total realized arbitrage across all strategies: $39,587,585. Combinatorial arbitrage: 0.24% of total profits.

The Dependency Detection Problem

Prediction markets price logically dependent events independently. When Market A (“Trump wins presidency”) and Market B (“Republican Senate majority”) trade at probabilities violating joint probability constraints, arbitrage theoretically exists.

Researchers analyzed the 2024 U.S. election markets (November 5, 2024): 128 NegRisk multi-condition markets and 177 single-condition markets. LLM-based semantic analysis using Linq-Embed-Mistral embeddings processed 46,360 market pairs. Initial LLM detection flagged 1,576 pairs as potentially dependent. Probabilistic constraint validation reduced candidates to 374. Manual expert verification confirmed 13 pairs with true logical dependencies satisfying formal combinatorial arbitrage definitions.

Critical finding: Detection ≠ profitability. Only 5 of 13 dependent pairs (38%) generated realized extraction. Documented profits: Pair 2 ($60,237), Pair 4 ($18,472), Pair 1 ($15,819), Pair 3 ($629). Total combinatorial arbitrage: $95,157 over 12-month measurement period (April 1, 2024–April 1, 2025).

Performance comparison across strategies:

• Buying “YES” positions: $11,092,286

• Buying “NO” positions: $17,307,114

• Selling “YES” positions: $612,189

• Selling “NO” positions: $4,264

• Combinatorial arbitrage: $95,157

Combinatorial strategies captured 0.24% of total $39,587,585 arbitrage profits despite requiring 10× implementation complexity relative to single-condition arbitrage.

LLM Implementation: Semantic Dependency Analysis

Traditional keyword matching fails. “Presidential Election Winner” and “Senate Control” share little lexical overlap but exhibit strong conditional probability relationships requiring semantic analysis.

Technical approach: Transformer-based embeddings (Linq-Embed-Mistral) generate vector representations of market questions. GPT-class LLMs perform logical reasoning over condition relationships, outputting structured JSON with dependency scores and dependent subset identification.

from openai import OpenAI
from typing import List, Dict, Tuple
import json

class DependencyDetector:
    def __init__(self, api_key: str):
        self.client = OpenAI(api_key=api_key)
        
    def analyze_dependency(
        self, 
        market1: Dict, 
        market2: Dict
    ) -> Tuple[float, List[Dict]]:
        “”“
        LLM-based logical dependency detection
        Returns: (dependency_score, dependent_subsets)
        “”“
        prompt = f”“”Analyze logical dependency between prediction markets:

Market 1 conditions: {market1[’conditions’]}
Market 2 conditions: {market2[’conditions’]}

Task: Determine if resolution of conditions in one market constrains 
possible resolutions in the other market.

Output JSON format:
{{
  “dependency_score”: 0.0-1.0,
  “dependent_subsets”: [
    {{”market1_conditions”: [indices], “market2_conditions”: [indices]}}
  ],
  “reasoning”: “Explanation of logical constraints”
}}

Dependency exists when: P(A ∩ B) ≠ P(A) × P(B)
Examples:
- “Team A wins semifinal” → “Team A in finals” (temporal constraint)
- “Trump wins popular vote” + “Trump wins electoral college” (dependent but not deterministic)
- “Bitcoin > $100K” + “Tech stocks rally” (correlated, NOT dependent)

Return ONLY valid JSON.”“”

        response = self.client.chat.completions.create(
            model=”gpt-4-turbo”,
            messages=[
                {”role”: “system”, “content”: “You are a probabilistic reasoning expert. Analyze logical dependencies between prediction market conditions using formal probability theory.”},
                {”role”: “user”, “content”: prompt}
            ],
            temperature=0.1
        )
        
        result = json.loads(response.choices[0].message.content)
        return result[’dependency_score’], result[’dependent_subsets’]

Prompt engineering critical. Research tested 128 NegRisk markets on election day. Results: 101 of 124 tests (81.45%) produced valid JSON with correct logical inference. LLM performance degraded with markets containing >4 conditions. Solution: Pre-process markets to top 4 conditions by traded volume, aggregate remaining conditions into “Other” category to maintain tractable prompt length.

False positive rate: Initial LLM detection returned 1,576 dependent pairs from 46,360 total pairs analyzed. Probabilistic constraint validation (verifying joint probability space satisfies mutual exclusivity and exhaustiveness requirements) reduced candidates to 374. Manual expert verification eliminated false positives from: (1) conflating electoral college vs. popular vote outcomes, (2) identifying correlation without causation, (3) detecting temporal proximity without logical constraint. Final result: 13 confirmed dependency pairs with valid combinatorial arbitrage structure.

Failure Modes: Why 62% of Dependencies Generate Zero Profit

Failure Mode 1: Liquidity Asymmetry

Dependent markets exhibit severe liquidity imbalance. Documented example from research:

• High-profile market (“Presidential winner”): $500K+ available liquidity

• Related dependent market (“Specific cabinet appointment”): $5K available liquidity

Position size constrained by min(liquidity_A, liquidity_B). Cannot deploy meaningful capital when constraining market has <$10K volume.

Empirical evidence: Maximum profit per combinatorial opportunity = min(liquidity_market_A, liquidity_market_B) × |price_divergence|. Median maximum profit across 13 detected dependent pairs: ~$100 per opportunity (estimated from total $95K extraction / 5 profitable pairs / estimated opportunities per pair). Single-condition arbitrage median: significantly higher due to concentrated liquidity in individual conditions.

Failure Mode 2: Execution Timing Risk

Non-atomic execution across dependent markets. Price moves between leg placements destroy arbitrage. Research documents 75% of matched Polymarket orders execute within 950 blocks (~1 hour on Polygon). Cross-market arbitrage requires simultaneous fills; second leg often executes at degraded price after first leg moves market.

from decimal import Decimal
import asyncio

class CombinatorialExecutor:
    def __init__(self):
        self.execution_window = 5  # seconds
        
    async def execute_dependent_arbitrage(
        self,
        market_a_id: str,
        market_b_id: str,
        dependent_conditions: Dict
    ) -> bool:
        “”“
        Atomic execution across dependent markets with timeout protection
        “”“
        try:
            order_a = asyncio.create_task(
                self.place_order(
                    market_a_id, 
                    dependent_conditions[’market_a’][’side’],
                    dependent_conditions[’market_a’][’size’],
                    dependent_conditions[’market_a’][’price’]
                )
            )
            order_b = asyncio.create_task(
                self.place_order(
                    market_b_id,
                    dependent_conditions[’market_b’][’side’],
                    dependent_conditions[’market_b’][’size’],
                    dependent_conditions[’market_b’][’price’]
                )
            )
            
            results = await asyncio.gather(
                order_a, 
                order_b,
                timeout=self.execution_window,
                return_exceptions=True
            )
            
            # Verify both fills completed
            if all(r.get(’filled’, False) for r in results):
                return True
            
            # Partial fill: Emergency exit unhedged position
            await self.emergency_exit(results)
            return False
                
        except asyncio.TimeoutError:
            await self.cancel_all_pending()
            return False

Failure Mode 3: Oracle Divergence Amplified

Single-market arbitrage faces one resolution event. Cross-market positions compound oracle risk. Case study: March 2025 UMA governance attack. Whale holder with 5 million UMA tokens (25% voting power) manipulated $7M market resolution through governance vote. Largest winner: $55K. Largest loser: $73K. Cross-platform hedges (Polymarket position vs. Kalshi position with different resolution oracles) face complete strategy failure if oracles resolve inconsistently.

Research mitigation: Avoid cross-platform dependent positions unless spread exceeds 15 cents (sufficient buffer for oracle manipulation risk). Same-platform dependent markets reduce but don’t eliminate oracle risk.

Failure Mode 4: Correlation Mistaken for Dependency

LLM semantic analysis detects relationships. Not all relationships create arbitrage structures. Example: “Trump wins election” and “Stock market hits all-time high” may exhibit correlation (estimated ρ ≈ 0.6) but both can independently resolve TRUE without probability violation.

True dependency requires: ∃ condition set where P(A ∩ B) = 0 (mutual exclusivity) or P(B|A) = 1 (deterministic implication). Validation method: Compute all possible truth-value assignments for condition pairs. Verify assignments satisfy market exhaustiveness (exactly one condition per market resolves TRUE). Only vectors satisfying this constraint constitute exploitable dependencies.

Capital Deployment Analysis: Risk-Adjusted Returns

Opportunity frequency comparison (April 2024–April 2025):

• Single-condition arbitrage: 7,051 conditions with exploitable deviations (4,423 single-market + 2,628 NegRisk)

• Market rebalancing within NegRisk markets: 662 markets with opportunities

• Combinatorial arbitrage: 5 profitable pairs from 13 detected dependencies

Execution complexity ratio:

• Single-condition: 2 orders per opportunity (YES + NO within one market)

• Combinatorial: 4+ orders across 2 markets + LLM analysis + cross-market coordination

Risk-adjusted ROI calculation:

from decimal import Decimal

def risk_adjusted_roi(
    expected_profit: Decimal,
    execution_risk: Decimal,  # probability of partial fill
    oracle_risk: Decimal,      # probability of resolution failure
    capital_deployed: Decimal
) -> Decimal:
    “”“
    Calculate risk-adjusted ROI for arbitrage strategies
    “”“
    ev_after_execution = expected_profit * (1 - execution_risk)
    ev_after_oracle = ev_after_execution * (1 - oracle_risk)
    return ev_after_oracle / capital_deployed

# Single-condition arbitrage profile:
single_roi = risk_adjusted_roi(
    expected_profit=Decimal(’450’),     # estimated per opportunity
    execution_risk=Decimal(’0.15’),     # 15% partial fill rate
    oracle_risk=Decimal(’0.02’),        # 2% resolution dispute
    capital_deployed=Decimal(’1000’)
)
# Result: ≈ 37.2% risk-adjusted ROI

# Combinatorial arbitrage profile:
combinatorial_roi = risk_adjusted_roi(
    expected_profit=Decimal(’19000’),   # $95K / 5 pairs
    execution_risk=Decimal(’0.35’),     # 35% partial fill (multi-market)
    oracle_risk=Decimal(’0.08’),        # 8% oracle risk (compounded)
    capital_deployed=Decimal(’10000’)
)
# Result: ≈ 113.7% risk-adjusted ROI per opportunity

# Frequency adjustment:
# Single-condition: 7,051 opportunities
# Combinatorial: 5 opportunities (1,410× less frequent)
# Frequency-adjusted expected value: single-condition dominates by 11×

Top arbitrageur profile: Research documented top 3 addresses executed 10,558 bids generating $4.38M profit. Execution pattern: high-frequency bot-like behavior prioritizing single-condition and market rebalancing opportunities. Combinatorial arbitrage represented <1% of sophisticated actors’ total profit.

Market Structure Evolution: Institutional Capital Entry

ICE investment signals professionalization: Intercontinental Exchange (NYSE parent) announced an investment of up to $2 billion in Polymarket, reflecting a valuation of approximately $8 billion pre-investment (October 7, 2025). ICE becomes global distributor of Polymarket event-driven data, bringing institutional infrastructure to retail-dominated orderbooks.

Kalshi’s regulatory advantage: Week of September 11–17, 2025: Kalshi captured 62% prediction market share, processing $500+ million weekly volume with $189 million average open interest. Polymarket: 37% market share, $430 million weekly volume. Kalshi operates as CFTC-designated contract market (DCM), providing institutional compliance framework absent in decentralized venues.

Arbitrage compression timeline: Research period (April 2024–April 2025) captured peak retail participation during 2024 U.S. election cycle. Polymarket processed $3.7 billion total volume during election cycle. Post-election volumes fluctuated; by mid-September through October 2025, industry-level data showed renewed surges with weeks registering hundreds of millions in volume, driven by platform share shifts such as Kalshi’s 62% capture during the September 11–17 period and institutional market maker entry beginning to compress spreads.

Historical parallel: Early cryptocurrency exchange arbitrage (2016–2018) generated 1,000%+ returns before institutional market makers professionalized infrastructure. Kimchi Premium (Korean exchange vs. global exchanges) reached 50%+ at peak, compressed to <2% within 18 months of institutional capital deployment. Prediction market arbitrage follows identical trajectory.

Quantitative Takeaway: Frequency Dominates Sophistication

LLM-based dependency detection produces research-grade analysis, not production trading infrastructure. Computational cost, false positive rate, and opportunity scarcity make combinatorial strategies inferior to simpler approaches.

Documented evidence from $39,587,585 total arbitrage:

• Buying “YES” positions: $11,092,286 (28.0%)

• Buying “NO” positions: $17,307,114 (43.7%)

• Selling “YES” positions: $612,189 (1.5%)

• Selling “NO” positions: $4,264 (0.01%)

• Combinatorial arbitrage: $95,157 (0.24%)

Practical strategy for quants entering prediction markets:

1. Deploy capital into single-condition arbitrage (7,051 opportunities documented, median sum-price ≈ $0.60 implies 40 cents profit per dollar)

2. Scale to market rebalancing within NegRisk markets (662 markets with opportunities, average max profit ≈ $400 per market)

3. Reserve combinatorial detection for manual research only — not algorithmic execution

4. Use LLM dependency analysis to identify high-conviction trades but execute manually with full risk assessment

Timeline compression: ICE’s up to $2 billion investment (October 2025) marks institutional capital inflection point. Research captured arbitrage during peak inefficiency (2024 election cycle). Future returns require: faster execution (sub-second), larger capital ($100K+ per trade), sophisticated cross-venue risk management as market structure professionalizes.

The $95K extracted from combinatorial arbitrage over 12 months demonstrates theoretical viability but practical inferiority. Sophisticated dependency detection generates academic citations. Simple sum-price deviations (YES + NO ≠ $1.00) generate profit. Top arbitrageur: $2.01 million across 4,049 transactions. Average per trade: $496. Strategy: frequency over sophistication.

References

Saguillo, O., Ghafouri, V., Kiffer, L., & Suarez-Tangil, G. (2025). “Unravelling the Probabilistic Forest: Arbitrage in Prediction Markets.” arXiv:2508.03474. IMDEA Networks Institute. https://arxiv.org/abs/2508.03474

Intercontinental Exchange. (2025, October 7). “ICE Announces Strategic Investment in Polymarket.” https://ir.theice.com/press/news-details/2025/ICE-Announces-Strategic-Investment-in-Polymarket/

Rodriguez, F. (2025, September 20). “Kalshi Outpaces Polymarket in Prediction Market Volume Amid Surge in U.S. Trading.” CoinDesk. https://www.coindesk.com/markets/2025/09/20/kalshi-outpaces-polymarket-in-prediction-market-volume

Cover: polymarket.com, screenshot taken 15 September 2026.

Academic research documented $40 million in arbitrage profits extracted from Polymarket between April 2024–April 2025.[1] The opportunity persists because retail-dominated orderbooks lack institutional market-making infrastructure. Here’s the technical implementation for capturing structural mispricings.

The Structural Inefficiency

Prediction markets exhibit chronic pricing failures exploitable through systematic strategies.

Single-condition arbitrage: When YES + NO token prices deviate from $1.00, profit is guaranteed. Research analyzed 86 million bets across 17,218 conditions.[1] When sum-price falls below $1.00, arbitrageurs buy both positions and realize guaranteed profit at resolution.

Cross-platform fragmentation: Kalshi captured 62% of prediction market volume during September 11–17, 2025, processing $500+ million weekly with $189 million average open interest.[2] Polymarket accounted for 37% during the same period.[2] Material price divergences between platforms create cross-venue arbitrage opportunities.

Mispricings cluster during volatility events — polling releases, debate nights, economic data. Price discovery lags real-world information by minutes to hours. Top arbitrageur generated $2.01 million across 4,049 transactions.[1] Strategy requires execution frequency over position size.

System Architecture

API Integration

Polymarket operates a hybrid-decentralized Central Limit Order Book (CLOB) on Polygon (Chain ID 137).[3] Off-chain matching, on-chain settlement via conditional token framework.

from py_clob_client.client import ClobClient
from web3 import Web3
import asyncio
import websockets
import json

class ArbitrageMonitor:
    def __init__(self, private_key: str):
        self.host = “https://clob.polymarket.com”
        self.chain_id = 137
        self.w3 = Web3(Web3.HTTPProvider(’https://polygon-rpc.com’))
        
        # Initialize CLOB client
        self.client = ClobClient(
            self.host,
            key=private_key,
            chain_id=self.chain_id
        )
        
    async def monitor_markets(self, market_ids: list):
        “”“Real-time WebSocket monitoring”“”
        ws_url = f”{self.host}/ws/market”
        
        async with websockets.connect(ws_url) as ws:
            subscribe_msg = json.dumps({
                “type”: “subscribe”,
                “markets”: market_ids
            })
            await ws.send(subscribe_msg)
            
            async for message in ws:
                data = json.loads(message)
                await self.process_update(data)

Kalshi integration: REST API with tiered rate limits. Basic tier: 20 requests/second (read), 10 requests/second (write). Premier tier: up to 100 requests/second.[4] Implement per-second throttling with token bucket algorithm.

Mispricing Detection

from decimal import Decimal
from typing import Optional, Dict

def detect_single_condition_arbitrage(
    yes_price: Decimal,
    no_price: Decimal,
    liquidity: Decimal,
    min_profit: Decimal = Decimal(’0.05’),
    max_capital: Decimal = Decimal(’10000’)
) -> Optional[Dict]:
    “”“
    Identify YES+NO≠$1 opportunities
    Returns None if below profitability threshold
    “”“
    sum_price = yes_price + no_price
    
    # Long arbitrage: Buy both if sum < $1
    if sum_price < (Decimal(’1.00’) - min_profit):
        profit_per_dollar = Decimal(’1.00’) - sum_price
        position_size = min(liquidity, max_capital)
        
        return {
            ‘type’: ‘buy_both’,
            ‘yes_price’: yes_price,
            ‘no_price’: no_price,
            ‘position_size’: position_size,
            ‘expected_profit’: profit_per_dollar * position_size,
            ‘roi’: profit_per_dollar / sum_price
        }
    
    # Short arbitrage: Sell both if sum > $1
    elif sum_price > (Decimal(’1.00’) + min_profit):
        profit_per_dollar = sum_price - Decimal(’1.00’)
        position_size = min(liquidity, max_capital)
        
        return {
            ‘type’: ‘sell_both’,
            ‘yes_price’: yes_price,
            ‘no_price’: no_price,
            ‘position_size’: position_size,
            ‘expected_profit’: profit_per_dollar * position_size,
            ‘roi’: profit_per_dollar / sum_price
        }
    
    return None

Research shows extreme outliers exist: one trader converted $0.02 into $58,983.36 exploiting severe mispricing where both YES/NO traded below $0.02.[1]

Order Execution: Managing Leg Risk

Critical constraint: Non-atomic execution. One position may fill while hedge fails, creating directional exposure.

class OrderExecutor:
    def __init__(self, client: ClobClient):
        self.client = client
        self.timeout = 5  # seconds
        
    async def execute_atomic_arbitrage(
        self,
        market_id: str,
        opportunity: Dict
    ) -> bool:
        “”“
        Execute both legs with timeout protection
        Cancel unfilled orders if partial fill occurs
        “”“
        try:
            # Place orders simultaneously
            yes_task = self.place_order(
                market_id, 
                ‘YES’,
                opportunity[’position_size’],
                opportunity[’yes_price’]
            )
            no_task = self.place_order(
                market_id,
                ‘NO’,
                opportunity[’position_size’],
                opportunity[’no_price’]
            )
            
            yes_order, no_order = await asyncio.gather(
                yes_task,
                no_task,
                timeout=self.timeout
            )
            
            # Verify both fills
            yes_filled = await self.check_fill_status(yes_order[’id’])
            no_filled = await self.check_fill_status(no_order[’id’])
            
            if yes_filled and no_filled:
                return True
            
            # Cancel unfilled leg, exit partial position
            await self.handle_partial_fill(yes_order, no_order)
            return False
                
        except asyncio.TimeoutError:
            await self.cancel_pending_orders(market_id)
            return False

Research data: 75% of matched orders execute within 950 blocks (~1 hour on Polygon).[1] Sub-5-second timeouts minimize leg risk.

Gas Optimization: Polygon Transaction Costs

Polymarket executes on Polygon. Gas fees compress net returns on small arbitrages.

from decimal import Decimal
from web3 import Web3

class GasOptimizer:
    def __init__(self, w3: Web3):
        self.w3 = w3
        self.matic_usd_price = self.fetch_matic_price()
        
    def estimate_transaction_cost(self, order_size: Decimal) -> Decimal:
        “”“
        Calculate total transaction cost
        Typical Polygon gas: 30-100 gwei
        “”“
        gas_price = self.w3.eth.gas_price
        gas_limit = 150000  # Standard CLOB order
        
        gas_cost_wei = gas_price * gas_limit
        gas_cost_matic = Decimal(gas_cost_wei) / Decimal(10**18)
        gas_cost_usd = gas_cost_matic * Decimal(str(self.matic_usd_price))
        
        # Polymarket: 0% platform fee (current)
        return gas_cost_usd
    
    def is_profitable_after_costs(
        self,
        expected_profit: Decimal,
        order_size: Decimal,
        min_roi: Decimal = Decimal(’0.02’)
    ) -> bool:
        “”“Minimum threshold: 2% net ROI after costs”“”
        total_cost = self.estimate_transaction_cost(order_size)
        net_profit = expected_profit - total_cost
        net_roi = net_profit / order_size
        
        return net_roi > min_roi

Position Tracking: Oracle Risk Management

Critical risk factor: Cross-platform arbitrage faces resolution divergence.

Polymarket uses UMA’s Optimistic Oracle with token-holder governance. March 2025 incident: whale with 5 million UMA tokens (25% of voting power) manipulated $7M market resolution on Ukraine mineral deal despite no formal agreement.[5] Largest winner netted $55,000; biggest loser forfeited $73,000.[5]

Kalshi operates under CFTC-regulated settlement mechanisms. Oracle mismatch can destroy cross-platform hedges.

from typing import Dict, List
from decimal import Decimal

class PositionTracker:
    def __init__(self):
        self.positions = {
            ‘polymarket’: {},
            ‘kalshi’: {}
        }
        
    def calculate_oracle_risk_exposure(self) -> Dict:
        “”“
        Flag positions with cross-platform oracle exposure
        “”“
        exposure = {
            ‘capital_deployed’: Decimal(’0’),
            ‘oracle_risk_positions’: []
        }
        
        for platform, positions in self.positions.items():
            for market_id, pos in positions.items():
                exposure[’capital_deployed’] += pos[’capital’]
                
                # Identify cross-platform hedges
                if self.has_cross_platform_hedge(market_id):
                    exposure[’oracle_risk_positions’].append({
                        ‘market’: market_id,
                        ‘exposure’: pos[’capital’],
                        ‘platforms’: self.get_hedge_platforms(market_id),
                        ‘risk_type’: ‘oracle_divergence’
                    })
        
        return exposure
    
    def has_cross_platform_hedge(self, market_id: str) -> bool:
        “”“Check if position exists across multiple platforms”“”
        platforms_with_position = [
            platform for platform, positions in self.positions.items()
            if market_id in positions
        ]
        return len(platforms_with_position) > 1

Mitigation strategy: Avoid cross-platform positions unless spread exceeds 15 cents — sufficient buffer to absorb potential oracle manipulation.

Alert Infrastructure

Top arbitrageurs executed 4,000+ transactions over 12 months.[1] Frequency matters more than per-trade size.

from dataclasses import dataclass
from decimal import Decimal
import asyncio
import time

@dataclass
class ArbitrageAlert:
    market_id: str
    platform: str
    opportunity_type: str
    expected_profit: Decimal
    roi: Decimal
    timestamp: float
    urgency: str

class AlertSystem:
    def __init__(self, notification_handler):
        self.handler = notification_handler
        self.min_profit = Decimal(’10.00’)  # $10 minimum
        
    async def process_opportunity(self, opp: Dict):
        “”“
        Filter and prioritize by profit potential
        High urgency: ROI > 10% or profit > $100
        “”“
        if not self.meets_criteria(opp):
            return
            
        alert = ArbitrageAlert(
            market_id=opp[’market_id’],
            platform=opp[’platform’],
            opportunity_type=opp[’type’],
            expected_profit=opp[’expected_profit’],
            roi=opp[’roi’],
            timestamp=time.time(),
            urgency=self.calculate_urgency(opp)
        )
        
        await self.handler.send(alert)
    
    def calculate_urgency(self, opp: Dict) -> str:
        “”“
        High: ROI > 10% or profit > $100
        Medium: ROI > 5% or profit > $50
        Low: Below thresholds but above minimum
        “”“
        if (opp[’roi’] > Decimal(’0.10’) or 
            opp[’expected_profit’] > Decimal(’100’)):
            return ‘high’
        elif (opp[’roi’] > Decimal(’0.05’) or 
              opp[’expected_profit’] > Decimal(’50’)):
            return ‘medium’
        return ‘low’

Deploy multi-threaded monitoring across 100+ active markets. Mispricings cluster during polling releases, debate nights, economic data prints.

P&L Analysis: Historical Extraction

Research findings from April 2024–April 2025 measurement period:[1]

Single-condition arbitrage:

• Buy both positions (sum < $1.00): $5.90M extracted

• Sell both positions (sum > $1.00): $4.68M extracted

• Total single-condition: $10.58M

Market rebalancing (multi-condition):

• Buy YES across conditions: $11.09M

• Buy NO across conditions: $17.31M

• Sell strategies combined: $4.88M

• Total market rebalancing: $28.99M

Combinatorial arbitrage (cross-market):

• 13 dependent market pairs identified during 2024 election cycle

• 5 pairs showed realized extraction: ~$95K total

• Cross-market opportunities significantly rarer than single-market

Total extracted: $39,587,585.02

Top 10 arbitrageurs captured $8.18M (21% of total).[1] Distribution confirms power law: sophisticated actors dominate.

Market Microstructure Analysis

Trading volume growth:

• Polymarket processed $3.7B total volume during 2024 U.S. election cycle[1]

• Weekly prediction market volume reached $2B+ by October 2025[6]

• Kalshi processed $500M+ weekly during peak periods[2]

Institutional capital influx:

• ICE invested $2B in Polymarket at $8B pre-money valuation (October 2025)[7]

• Signals professionalization of market infrastructure

• Parallel to early crypto exchange arbitrage before institutional market makers

Competitive dynamics:

• September 11–17, 2025: Kalshi 62% market share, Polymarket 37%[2]

• Politics markets: largest absolute profits

• Sports markets: higher opportunity frequency, lower per-trade margins

The window compresses as institutional capital enters. Early crypto arbitrage generated 1,000%+ returns before market makers professionalized infrastructure. Prediction markets follow identical trajectory.

Execution Constraints

Capital efficiency: Most conditions offer $50-$500 maximum profit at available liquidity. Scaling requires:

• Monitoring 100+ markets simultaneously

• Accepting 1–5% per-trade margins

• Deploying capital through frequency, not position size

Latency requirements: Information propagation lag creates 5–10 minute windows during volatility events. Sub-second execution required for high-frequency opportunities.

Regulatory risk: Massachusetts sued Kalshi September 2025 for allegedly operating unlicensed sports betting.[8] Over $1B in wagers January-June 2025, 75% sports-related volume.[8] Polymarket settled with CFTC in 2022, restructured operations.

Key Takeaway

Prediction market arbitrage is market microstructure exploitation — not outcome forecasting. The strategy requires:

Infrastructure:

• Real-time monitoring: WebSocket feeds across 100+ markets

• Execution speed: Sub-5-second order placement with atomic timeout protection

• Capital allocation: $10K-$50K per condition based on liquidity

• Risk management: Avoid oracle exposure; implement leg-risk controls

Profitability: Accept 1–5% returns per trade. Scale through frequency. Top performer: 4,049 transactions generating $2.01M ($496 average per trade).

Timeline: ICE’s $2B investment signals professionalization. Spread compression inevitable as institutional infrastructure deploys. Research captured peak inefficiency during 2024 election cycle — future returns require faster execution, larger capital, and more sophisticated cross-venue strategies.

Deploy now or wait for compressed spreads. The opportunity window follows crypto’s 2016–2018 trajectory: early arbitrageurs extract outsized returns before market makers eliminate structural inefficiency.

References

[1] Saguillo, O., Ghafouri, V., Kiffer, L., & Suarez-Tangil, G. (2025). “Unravelling the Probabilistic Forest: Arbitrage in Prediction Markets.” arXiv:2508.03474. IMDEA Networks Institute. https://arxiv.org/abs/2508.03474

[2] Rodriguez, F. (2025, September 20). “Kalshi Outpaces Polymarket in Prediction Market Volume Amid Surge in U.S. Trading.” CoinDesk. https://www.coindesk.com/markets/2025/09/20/kalshi-outpaces-polymarket-in-prediction-market-volume

[3] Polymarket Documentation. (2025). “CLOB Introduction.” https://docs.polymarket.com/developers/CLOB/introduction

[4] Kalshi API Documentation. (2025). “Rate Limits and Tiers.” https://docs.kalshi.com/getting_started/rate_limits

[5] Yahoo Finance. (2025, March 27). “Polymarket Suffers UMA Governance Attack After Rogue Actor Becomes Top-5 Token Staker.” https://finance.yahoo.com/news/polymarket-suffers-uma-governance-attack-101646076.html

[6] Yahoo Finance. (2025, October 21). “Prediction Markets Hit All-Time High of $2 Billion in Weekly Volume.” https://finance.yahoo.com/news/prediction-markets-hit-time-high-200142742.html

[7] Intercontinental Exchange. (2025, October 7). “ICE Announces Strategic Investment in Polymarket.” https://ir.theice.com/press/news-details/2025/ICE-Announces-Strategic-Investment-in-Polymarket/

[8] Campbell, A.J. (2025, September 12). “Massachusetts Attorney General Files Lawsuit Against Kalshi.” WBUR News. https://www.wbur.org/news/2025/09/16/massachusetts-kalshi-lawsuit-predictions-sports-wagering

Cover photograph: Brynn Who Likes Editing, public domain, via Wikimedia Commons.

Prediction markets reached $2 billion in weekly volume by late October 2025. Academic research documented $39.59 million in arbitrage profits extracted between April 2024 and April 2025 — a direct analog to early crypto exchange arbitrage before institutional capital professionalized market structure.

Market Structure

Intercontinental Exchange invested $2 billion in Polymarket at an $8 billion pre-money valuation in October 2025, bringing institutional capital to crypto-native prediction markets. Kalshi processed $500+ million weekly volume with $189 million open interest during the week of September 11–17, 2025, capturing 62% market share in that period. Despite billion-dollar valuations and institutional backing, these venues exhibit chronic structural inefficiencies.

Cross-platform fragmentation. Identical events trade at materially different implied probabilities. Snapshot examples show spreads exceeding 10 cents persisting for hours — Bitcoin Reserve markets showed 51% on Polymarket versus 37% on Kalshi in documented instances, creating 14-cent arbitrage opportunities.

Single-market inefficiencies. IMDEA Networks Institute researchers analyzed 86 million bets across 17,200 conditions and found 7,051 conditions with exploitable Yes + No price deviations from $1.00. The paper finds the median sum-price of YES+NO is approximately $0.60, which implies a median profit of approximately $0.40 per $1; the average maximum single-opportunity across topics is also about $0.40 on the dollar.

Resolution oracle divergence. Polymarket uses UMA’s Optimistic Oracle with governance-vote resolution; Kalshi operates under CFTC-regulated settlement mechanisms. Different resolution criteria create execution risk — positions hedged across platforms face oracle-mismatch exposure that can destroy seemingly risk-free arbitrage.

Trade Execution

The academic study identified two primary arbitrage structures:

Market Rebalancing Arbitrage. When Yes ($0.55) + No ($0.40) < $1.00, buy both positions and realize $0.05 profit at resolution. Single-condition arbitrage generated $10.58 million in extracted profits: $5.90 million from buying below $1.00, $4.68 million from selling above $1.00.

Combinatorial Arbitrage. Build hedged positions across logically dependent markets. LLM-based semantic analysis detected dependencies among 305 markets, identifying 13 logically dependent pairs during the 2024 U.S. election cycle. Only 5 dependent pairs showed realized extraction (the listed pair amounts sum to approximately $95,157) — cross-market opportunities were significantly rarer than single-market inefficiencies.

Top 3 arbitrageur addresses executed 10,558 bets generating $4.38 million profit, displaying algorithmic execution frequency. Most trades yielded 1–5% returns. Extreme outliers existed: one trader converted $0.02 into $58,983 exploiting a severe mispricing where both Yes/No tokens traded below $0.02.

P&L Mechanics

Total extracted arbitrage across all strategies: $39,587,585 over the April 2024–April 2025 measurement period.

Politics markets contributed the largest absolute profits. Polymarket recorded over $3.7 billion in total trading volume during the 2024 U.S. election cycle. Sports markets exhibited higher opportunity frequency but lower per-trade profit margins.

Top arbitrageur generated $2.01 million across 4,049 transactions. Top 10 accounts collectively extracted $8.18 million, accounting for 21% of total arbitrage profits.

Execution constraints bind capital deployment. Most conditions offered limited absolute profit even at maximum available liquidity. Deploying serious capital requires monitoring hundreds of markets simultaneously, accepting small per-trade margins, and scaling through execution frequency. Non-atomic orderbook execution introduces leg risk — one position may fill while the hedge fails.

Transaction costs matter. Polymarket operates on Polygon with gas fees. Cross-platform arbitrage requires liquidity on multiple venues plus withdrawal/deposit latency. Platform fee structures (where they exist) further compress net returns.

Quant Insight

Order book depth analysis revealed absence of sophisticated market-making infrastructure. Traditional crypto market makers would eliminate most observed mispricings within seconds. The persistence of 40-cent deviations from fair value indicates retail-dominated flow without institutional MM capital.

Arbitrage opportunities clustered during volatility events: polling releases, debate nights, economic data prints. Price discovery lagged real-world information by minutes to hours — standard latency arbitrage dynamics apply.

The parallel to 2016 crypto markets is structural: fragmented exchanges, retail-dominated orderbooks, primitive APIs, information asymmetry. Early crypto quants generated 1,000%+ returns before institutions professionalized market structure. Prediction markets are replicating this evolution.

Risk Factors

Resolution risk is non-zero. UMA governance relies on token-holder voting for disputed outcomes. Documented governance attacks exist: March 2025 incidents involved concentrated voting power forcing favorable settlements on multi-million-dollar markets. Oracle manipulation is a real P&L exposure.

Regulatory risk persists. Massachusetts sued Kalshi in September 2025 for allegedly operating unlicensed sports betting. Polymarket settled with CFTC in 2022 for unregistered activity and restructured operations. Regulatory action can freeze positions or force platform exits.

Competition intensifies. As arbitrageurs deploy specialized strategies similar to DeFi AMM bots, mispricings compress. The academic study characterizes arbitrage as “generally positive-sum MEV that promotes price alignment” — meaning the opportunity self-destructs as markets mature. Early-stage inefficiency windows close quickly once capital and sophistication arrive.

Actionable Takeaway

Prediction markets are mispriced binary derivatives trading in fragmented venues without institutional market-making infrastructure. The opportunity exists because:

1. Retail-dominated orderbooks lack sophisticated arbitrage mechanisms

2. Cross-platform liquidity fragmentation prevents price convergence

3. Resolution oracle divergence creates fundamental hedge risk

4. Volatility clustering around news creates predictable mispricing windows

For quants, the trade is pure market microstructure — not outcome forecasting. Buy when Yes + No < $1.00. Execute cross-platform arbitrage when spreads exceed transaction costs. Deploy automated monitoring across hundreds of markets. Accept small per-trade margins; scale through frequency.

The window is compressing. ICE’s $2 billion investment signals institutional interest. Kalshi’s $500+ million weekly volume and Polymarket’s $8 billion valuation demonstrate market maturation. As capital and execution sophistication increase, spreads will compress — replicating crypto’s evolution from retail arbitrage playground to institutional derivatives market.

The documented $40 million extraction over 12 months occurred during peak retail participation (2024 U.S. election cycle). Future arbitrage returns will likely require faster execution, larger capital deployment, and more sophisticated cross-venue hedging as market structure professionalizes.

Sources

1. Saguillo, O., Ghafouri, V., Kiffer, L., & Suarez-Tangil, G. (2025). “Unravelling the Probabilistic Forest: Arbitrage in Prediction Markets.” arXiv:2508.03474. https://arxiv.org/abs/2508.03474

2. Intercontinental Exchange. (2025, October 7). “ICE Announces Strategic Investment in Polymarket.” IR Release. https://ir.theice.com/press/news-details/2025/ICE-Announces-Strategic-Investment-in-Polymarket/

3. Yahoo Finance. (2025, October 21). “Prediction Markets Hit All-Time High of $2 Billion in Weekly Volume.” https://finance.yahoo.com/news/prediction-markets-hit-time-high-200142742.html

4. Rodriguez, F. (2025, September 20). “Kalshi Outpaces Polymarket in Prediction Market Volume Amid Surge in U.S. Trading.” CoinDesk. https://www.coindesk.com/markets/2025/09/20/kalshi-outpaces-polymarket-in-prediction-market-volume

5. Financial Times. (2025, October 7). “NYSE parent to invest up to $2bn in prediction platform Polymarket.” https://www.ft.com/content/9437b5b5-59c0-44a3-8b25-58b3b4aa1b0c

6. Reuters. (2025, September 12). “Massachusetts accuses Kalshi of running illegal sports wagering operation.” https://www.reuters.com/world/massachusetts-accuses-kalshi-running-illegal-sports-wagering-operation-2025-09-12/

7. Yahoo Finance. (2025, March 27). “Polymarket Suffers UMA Governance Attack After Rogue Vote.” https://finance.yahoo.com/news/polymarket-suffers-uma-governance-attack-101646076.html

Cover: polymarket.com, screenshot taken 15 September 2026.