In August 1993, Marriott Corporation traded at $27.75. Joel Greenblatt bought the October 15 call options struck at $25 for $3.125 a contract. Three months earlier, Marriott CFO Stephen Bollenbach had unveiled "Project Chariot," a recapitalization that would split the company in two: Marriott International, the asset-light hotel-management franchise, and Host Marriott, the rump entity into which roughly $3 billion of debt, distressed real estate, and airport-concession contracts would be poured. Bondholders sued. The Maryland litigation dragged on for years. Bollenbach himself moved to run the wreckage. By the close of trading on October 15, 1993, the two stocks Greenblatt's calls now controlled together fetched roughly $32.75 — Marriott International near $26, Host Marriott near $6.75. The $25 strike paid roughly $7.75 against the $3.125 premium: a 148 percent gain in two months. Within four months of the spinoff itself, Greenblatt's underlying Host Marriott position had nearly tripled.
Greenblatt told the story in You Can Be a Stock Market Genius (Simon & Schuster, 1997), the book that reduced his Gotham Capital franchise to a teachable system. Gotham's audited record from inception in 1985 through 1994 was a 50 percent gross / 34.4 percent net annualized return — a number that looks like a typo until you read the book and realize the engine was structural rather than analytical. Greenblatt was not betting on Bollenbach's competence or hotel REIT cap rates. He was betting on the predictable behavior of the index funds, pension plans, large-cap mandates, and bond-fund covenants that owned old Marriott Corporation and were required to sell the smaller, levered, real-estate-heavy "child" the moment it landed in their accounts. Most institutional mandates barred holding a $200–300 million market-cap stock with $3 billion of debt. The selling was forced, calendar-driven, and price-insensitive. The resulting underpricing was the alpha.
The academic case was already in print before the book. Cusatis, Miles, and Woolridge (Penn State), publishing in the Journal of Financial Economics in June 1993 ("Restructuring through spinoffs: The stock market evidence"), examined every U.S. corporate spinoff from 1965 to 1988. Spinoffs and their parents both beat the S&P 500 in the three years following separation — spinoffs by an average of 30 percent cumulative, parents by 18 percent — with the abnormal returns concentrated in firms subsequently involved in takeover activity. The mechanism the paper identifies is the same one Greenblatt monetized: indiscriminate selling at the moment of distribution, followed by gradual price discovery as the standalone entity finds its natural shareholder base.
Greenblatt then generalized the principle. Anywhere a corporate event hands shareholders a security they did not choose, did not analyze, and frequently could not legally hold — merger consideration paid in CVRs and warrants, recapitalization stub stocks, post-bankruptcy equity issued to former creditors, rights offerings — there is a window during which the price reflects the constraint, not the value. Special situations are a micro-structure trade dressed in fundamental clothes.
That micro-structure is now the dominant feature of resolved-event prediction markets.
The on-chain version of forced selling
Consider what happens when a Polymarket "parent" market resolves. The Trump-wins-2024 contract paid out in November 2024 against $3.7 billion of cumulative election-cycle volume. The resolution did not extinguish exposure; it transferred it. The traders who held YES on Trump now had stale, partially-hedged exposure to dozens of child markets keyed to downstream outcomes: which cabinet picks would be confirmed, which would withdraw, how many would fail, when each Senate vote would land, what sequence of executive orders would be signed in the first hundred days. Polymarket today hosts 128 active markets in the broader "Cabinet For Trump" category against more than $33 million of cumulative volume, with the narrower "Trump Cabinet" sub-cluster running ten markets and roughly $1.5 million of live exposure. Earlier in the cycle the Trump-administration sub-markets traded $7.7 million in volume, the confirmation sub-markets $23 million.
Those sub-markets did not reprice efficiently when the parent settled. They repriced piecewise, in the order in which news arrived, with measurable lag against wire reports. The sharpest documented case was Matt Gaetz's withdrawal as attorney general nominee on November 21, 2024. Pete Hegseth's confirmation contract had been trading near 89 percent the day Gaetz was first announced; within hours of the Gaetz withdrawal it collapsed toward 47 percent before recovering to 56 percent by late Friday. The "3+ Trump cabinet picks fail" market jumped on the same headline. Other contested nominees — Tulsi Gabbard for DNI, Robert F. Kennedy Jr. for HHS — repriced sequentially over the following days as Senate whip counts leaked. The forced-seller analog is precise: traders who had treated the Trump victory as a single bet were now exposed to a slate of correlated child contracts, and those without the operational capacity to monitor each Senate vote unwound at non-clearing prices.
The same pattern recurred the moment Maduro was apprehended. The U.S. operation in Caracas resolved on January 3, 2026; the parent "Maduro out by..." market settled, but the surrounding cluster — which date specifically, which successor scenario, what U.S. recognition follows — repriced on the schedule of news, not on the schedule of trading. The April 2026 indictment of Special Forces Master Sergeant Gannon Ken Van Dyke, who allegedly turned roughly $33,000 into approximately $409,881 in the week before the operation, is the law-enforcement footnote; the structural story is that the surrounding child markets continued mispricing for hours and days after the parent had factually resolved. Wire-service-tracked Polymarket cabinet odds in November 2024 likewise lagged Reuters and AP by minutes-to-hours on more than one occasion — a non-trivial window for an automated trader.
What the academic data says about repricing speed
Tsang and Yang's "Political Shocks and Price Discovery in Prediction Markets" (arXiv 2603.03152, March 2026) is the cleanest empirical decomposition. Using transaction-level on-chain data from the 2024 election market, they study three discrete shocks: the June 2024 Biden-Trump debate, the July 2024 Trump assassination attempt, and Biden's July 2024 dropout. The three shocks produce qualitatively different repricing dynamics. The debate-induced jump in Trump probability largely reverses over subsequent weeks. The assassination repricing persists. The Biden dropout produces heavy two-sided trading with little net price change — a market that absorbs information without a clear directional consensus. Trading volume rises after each event, especially among incumbent traders with pre-event exposure against Trump, who are also the most likely to flip positions. The takeaway is that shock persistence depends on shock type, liquidity, and trader heterogeneity rather than a single "price discovery speed." A Greenblatt-style operator does not assume the parent shock has fully repriced the child; the operator measures which kind of shock occurred and trades accordingly.
The companion paper from the same group — "The Anatomy of Polymarket: Evidence from the 2024 Presidential Election" (arXiv 2603.03136) — places the entire story explicitly inside the Shleifer-Vishny "limits to arbitrage" framework that Greenblatt's special-situations practice operationalized in equities. As the 2024 market matured, arbitrage deviations narrowed and Kyle's lambda — the inverse of market depth, the per-dollar price impact of an order — fell by more than an order of magnitude. That is the prediction-market analog of newly spun-off equity gradually attracting institutional analyst coverage. The limits relax over the lifetime of the contract, which means the largest mispricings live in the window immediately after the parent resolves, before specialist liquidity providers re-anchor the child.
Saguillo et al.'s "Unravelling the Probabilistic Forest" (arXiv 2508.03474, August 2025) measures the extracted rents directly. Using on-chain order-book data, the paper finds two distinct arbitrage families on Polymarket: market rebalancing (within a single multi-outcome market whose component prices fail to sum to one) and combinatorial arbitrage (across logically related markets whose joint probabilities violate basic constraints). Realized arbitrageur profits across the sample exceed approximately $40 million. The single most-cited case from the broader trading press: a Polymarket address turning $10,000 into $100,000 in six months by participating in over 10,000 markets, mechanically buying complete sets whenever the sum of mutually exclusive outcomes fell below $1.00. A worked example from the same coverage — a Fed-decision multi-outcome priced at $0.001 + $0.008 + $0.985 + $0.001 = $0.995 — is pure stub-stock math. Buy one share of each, collect a guaranteed $1.00 on resolution, pocket half a percent risk-free. The strategy scales linearly with bankroll until liquidity binds.
The patterns worth monitoring
The "parent → child" relationships that consistently produce exploitable lag share three features. First, the parent has a discrete, public resolution moment — a presidential call, a Senate confirmation vote, a Truth Social post (Polymarket's Iran-ceasefire markets explicitly accept Trump's personal posts as resolution evidence), a court ruling, a military operation. Second, the child markets are sequentially correlated rather than jointly resolved — Senate confirmations land one at a time, cabinet withdrawals arrive on a stochastic schedule, ceasefire extensions reprice on each presidential statement. Third, the holders of the parent are not the natural holders of the child — the trader who took the $80 million Trump-wins position is not, almost by definition, the trader most edged on the precise sequence of HHS Senate votes.
The actionable monitoring list, in current Polymarket terms, runs roughly as follows. Around any administration-formation parent: confirmation contracts on the contested nominees, the "X+ cabinet picks fail" aggregator, the day-one executive-order count markets, and the cross-references between cabinet identity and downstream policy markets (a particular Treasury pick reprices the tariff schedule, a particular HHS pick reprices vaccine-mandate markets). Around any military or capture parent: successor-government contracts, U.S. recognition contracts, sanctions-relief timing markets, and the corresponding regional ceasefire ladder (April 30 / May 31 / June 30 / December 31 dated contracts that must satisfy P(T₁) ≤ P(T₂) by no-arbitrage). Around any judicial parent: appeal-timing markets, related-case markets in adjacent circuits, and the implementation-deadline cluster.
The structural risk that breaks the analogy
Greenblatt's spinoff thesis assumed that the child equity, once distributed, was a real claim on real cash flows. Mispricing was about who held the claim, not whether the claim existed. Prediction markets break the analogy precisely on this point. Settlement-spec divergence between parent and child is the binding risk: a parent contract may resolve YES on a definition the child contract does not honor. The Cardi B halftime case settled YES on Polymarket and NO on Kalshi for the identical underlying event because the two venues used different definitions of "performing." A U.S. government-shutdown contract did the same. Polymarket's Iran-ceasefire cluster includes "ceasefire announced" markets that resolve on Truth Social posts but "permanent peace deal" markets that require formal documentation; the parent settling YES does not transfer cleanly into the child. UMA's Optimistic Oracle has, in documented cases — the Zelenskyy-suit market, parts of the TikTok-ban cycle — produced parent resolutions that contradict the spec a child trader was implicitly pricing.
The discipline is the same one Greenblatt enforced when he separated post-bankruptcy equities he could underwrite from the ones he could not: read the spec on every child before the parent resolves, not after. The spinoff trade was never a bet on the spinoff. It was a bet on the difference between what a forced seller had to do and what a careful reader had already understood. The on-chain version is identical. The forced sellers on Polymarket are the traders whose parent positions just settled, who hold partial exposure to a slate of correlated children, and who do not have the operational bandwidth to read every spec and every wire in the first hour after resolution. The alpha sits in the gap between their exit and the child's natural clearing price. It compounds for as long as Polymarket, Kalshi, and their successors keep adding multi-leg event ladders faster than specialists can underwrite them — which, on the current trajectory of prediction-market market-cap and contract count, is a long time.
Greenblatt's edge in 1993 lasted roughly a decade before institutional spinoff coverage compressed it. The Anatomy of Polymarket paper documents Kyle's lambda falling by an order of magnitude inside a single election cycle. The window is shorter on-chain. It is also wider, because there are more parents resolving more often.