From Boesky to Bots: Porting Hedge Fund Alpha to Prediction Markets

Classic hedge fund strategies map onto Polymarket and Kalshi with surprising fidelity, but the alpha is concentrated, the risks are different in kind, and academic evidence from 2024–2026 shows roughly 3% of traders capture nearly all gains. The most direct analogs — merger arbitrage as "tail-end" trading, fixed-income relative value as ceasefire/Fed term structures, and convertible arb as multi-outcome overround — generate documented eight-figure profits, but each carries a structural risk (UMA dispute risk, settlement divergence between Polymarket and Kalshi, adverse selection from informed traders) that has no clean equivalent in 1980s Wall Street. The newest empirical work (Akey et al. 2026; Gómez-Cram et al. 2026; Le 2026; Saguillo et al. 2025) confirms that prediction markets are roughly as efficient as a maturing equity venue circa the late 1990s — meaning the playbooks of Levy, Boesky, Buffett, Greenblatt, Tartaglia, Meriwether, Soros, and Taleb are not just metaphors. They are operational templates being executed today, mostly by quantitative shops, increasingly by Jump Trading and Susquehanna, and occasionally by individual traders deploying $30–80 million on a single conviction call.

The ten mappings below are organized from cleanest analog to most speculative, with documented cases, quantitative data, and named academic sources throughout.

Tail-end trading is merger arbitrage with a UMA twist

Gus Levy formalized merger arb at Goldman Sachs in the 1940s, building the desk that Robert Rubin would later run. Warren Buffett ran the same playbook at Graham-Newman from 1954 — his 1988 Berkshire letter disclosed $78M in arbitrage profits on $147M average invested capital, a 53% pre-tax return, anchored by the 1981 Arcata/KKR deal where the unresolved redwood claim eventually paid an extra $19.3M years after closing. Ivan Boesky industrialized the strategy with $200M by 1985, then collapsed it when SEC's 1986 case revealed he had been paying Drexel banker Dennis Levine for nonpublic merger tips — producing a $100M fine, a 22-month sentence, and the 1988 Insider Trading and Securities Fraud Enforcement Act.

The Polymarket equivalent is "tail-end trading" — buying YES contracts at $0.95–$0.99 after the underlying event has factually occurred but before on-chain settlement. A trader nicknamed "Fish" told BlockBeats that roughly 90% of large Polymarket trades over $10,000 execute above $0.95, with arbitrageurs supplying liquidity to retail sellers who want capital free for the next bet. The mechanic is identical to Levy's: capture a small absolute spread that, given short settlement windows, annualizes attractively. Polymarket's UMA Optimistic Oracle v2 normally resolves within a 2-hour challenge window with a $750 USDC bond; full DVM escalation takes 4–6 days. Kalshi, as a CFTC-regulated DCM, typically settles 1–12 hours after market close plus a ~3-hour settlement leg.

The "deal-break risk" analog is resolution dispute risk, and several documented cases have produced total wipeouts on positions held above $0.95. The Zelenskyy "suit" market ($237M volume) resolved NO in July 2025 despite Zelenskyy's June 24 NATO summit appearance in collared dress — UMA cited insufficient "credible reporting consensus." The TikTok ban market ($120M) resolved YES on January 20, 2025 hours before Trump's 75-day extension; the Cardi B Super Bowl halftime market resolved YES on Polymarket but NO on Kalshi using stricter "performing" criteria, breaking cross-platform arb legs. Per Dune analytics cited by MEXC, Polymarket's directional accuracy is 95.4% four hours pre-settlement, 89.4% at twelve hours, and 88.2% at one day — a measurable, tightening "merger spread."

Cross-contract repricing lag is Greenblatt's spinoff thesis on-chain

Joel Greenblatt's You Can Be a Stock Market Genius (1997) codified the post-1980s playbook: indiscriminate selling by parent shareholders after a spinoff creates predictable mispricing. The Marriott/Host Marriott "Project Chariot" split of October 1993 is the canonical case — Greenblatt bought October 1993 $25 calls at $3.125 in August 1993 and Host Marriott nearly tripled within four months. The academic Penn State finding (Cusatis, Miles, Woolridge) showed spinoffs outperformed the S&P by ~10% annually in their first three years. Greenblatt's Gotham Capital produced 50% gross / 34.4% net annualized through 1994 by extending this logic to merger securities, recapitalizations, stub stocks, and post-bankruptcy equities.

In prediction markets, the analog is slow repricing of "child" contracts after a "parent" event resolves. When Polymarket's Trump-wins-2024 market settled in November 2024, the Trump-administration sub-markets ($7.7M volume) and confirmation sub-markets ($23M volume) repriced unevenly, with leaks moving wire-service-tracked Polymarket cabinet odds minutes-to-hours behind Reuters/AP reports. The Matt Gaetz withdrawal on November 21 instantly moved the "3+ Trump cabinet picks fail" market — but only after a delay measurable in single-digit minutes. Tsang & Yang (arXiv 2603.03152, March 2026) studied transaction-level data on three 2024 shocks: the Biden-Trump debate price jump largely reversed, the Trump assassination repricing persisted, and the Biden dropout produced heavy two-sided trading with little net price change — meaning shock persistence depends on shock type, liquidity, and trader heterogeneity, not just speed.

The cleaner economic analog to Greenblatt is Saguillo et al. (arXiv 2508.03474, August 2025), which estimates that arbitrageurs have extracted roughly $40M from market-rebalancing and combinatorial mispricings on Polymarket. One trader documented turning $10,000 into $100,000 in six months by systematically buying the full set of options whenever a multi-outcome market summed below $1.00 — pure "stub stock" math. The "Anatomy of Polymarket" working paper (arXiv 2603.03136) explicitly invokes Shleifer-Vishny's limits-to-arbitrage framework, observing that deviations narrowed and Kyle's lambda fell by more than an order of magnitude as the 2024 election market matured — the prediction-market analog of deepening institutional coverage of newly spun-off equities.

Distressed debt does not port — the longshot pile is a value trap

Apollo (Leon Black, 1990) and Elliott (Paul Singer, 1977) built fortunes buying distressed bonds at 10–30¢ on the dollar. Apollo's first big trade was Crédit Lyonnais's Executive Life high-yield portfolio (~$3.25B purchase, ~$5B two years later); Elliott bought ~$20M face of Peruvian debt for $11.4M in 1995–96 and collected $58M in 2000 after winning the pari passu litigation, then repeated the playbook against Argentina via NML Capital (15-year saga, ultimate 2016 victory). Howard Marks/Oaktree's 2008 Opportunities Fund VII raised $10.9B and returned ~31.5% IRR. Moody's Ultimate Recovery Database documents 49% average recoveries across 629 bankruptcies — the structural source of distressed alpha.

The naive prediction-market analog is buying 1–5¢ contracts hoping for 5x–20x reversals. The empirical evidence rejects this strategy on retail-dominated venues. Bürgi, Deng & Whelan (CEPR DP 20631, 2026) analyzed over 300,000 Kalshi contracts and found a strong favorite-longshot bias: a 5¢ contract winning ~2% of the time loses 60% of capital invested, with pre-fee average returns on a typical contract around −20%. Le (2026, arXiv 2602.19520, 292M trades / 327K contracts on Kalshi and Polymarket) confirms favorite-longshot bias at long horizons across all comparable domains, with prices compressed toward 50% — favorites underpriced, longshots overpriced. Snowberg & Wolfers (JPE 2010, NBER WP 15923) traced this pattern back to misperception of probabilities in the Kahneman-Tversky tradition rather than risk-love.

The structural reason distressed-debt alpha doesn't port: Apollo and Elliott earn returns from information edges in bankruptcy proceedings, creditor-committee coalition-building, legal optionality (NML's pari passu litigation), and forced sellers (insurers and rating-mandated mutual funds) — none of which exist for a 3¢ Polymarket contract priced at 3¢ precisely because consensus probability is 3% or lower. The narrow exception is contracts in deeply illiquid markets where a known whale is unwinding, public information has not yet been priced, or cross-platform divergence persists. Polymarket's own data shows outcomes priced ≤10% occur 14% of the time (Fensory 2026) — meaning longshots are slightly underpriced on Polymarket specifically, possibly because crypto-native traders are more sophisticated than Kalshi retail.

Pairs trading lives, but the alpha sits inside the spread

Gerry Bamberger originated pairs trading at Morgan Stanley around 1982–83; Nunzio Tartaglia's APT group reportedly produced $50M in 1987 profits including through Black Monday. The diaspora — David Shaw founding D.E. Shaw, Two Sigma spinning out, Peter Muller resurrecting it as PDT, Renaissance growing in parallel — institutionalized cointegration-based mean reversion as a major strategy class through the 1990s. By the 2000s, easy single-stock pairs were arbitraged out and stat arb migrated to higher-dimensional factor models.

Three prediction-market analogs are documented. First, within-platform structural pairs: Polymarket markets for "Dem wins by 6–7%" and "GOP wins by 6–7%" sometimes moved in the same direction simultaneously during 2024 (Clinton & Huang, Vanderbilt 2025), violating mutual exclusivity. Second, cross-platform Polymarket↔Kalshi spreads of 1.5–4.5% on liquid Bitcoin and election contracts, executed by open-source bots. Third, mean reversion in PM price dynamics: QuantPedia's April 2026 study of three Polymarket contracts (Jesus's return, China invading Taiwan, US confirms aliens) found mean-reversion signals generate substantial alpha under maker-only execution but degrade significantly once realistic spreads and fees are imposed — exactly the late-1990s arc of single-stock pairs trading.

The killer risk is settlement specification divergence. The Cardi B halftime case settled YES on Polymarket and NO on Kalshi for the identical underlying event; a US government-shutdown contract did the same. This is the prediction-market analog of Royal Dutch and Shell decoupling because of a corporate-action change — and unlike equity pairs, the binary $0/$1 payoff means correlation breakdown is total rather than partial. No published academic paper as of April 2026 formally tests Engle-Granger or Johansen cointegration on PM pairs, but Le (2026) finds politics contracts are systematically underconfident (slopes 0.93–1.83) while weather is overconfident at short horizons — implying long politics-favorite / short weather-favorite as a calibration-spread trade.

Multi-outcome overround is the cleanest convertible arb analog

Ken Griffin started Citadel in November 1990 with $4.6M trading converts, returning 43% in 1991 and 40% in 1992 by writing software that priced embedded conversion options more accurately than dealers using "calculators and intuition." The structural insight — that traders systematically misprice components of a composite security relative to the whole — ports almost perfectly to multi-outcome prediction markets, where a complete set of mutually exclusive contracts should sum to exactly $1.00 but frequently doesn't.

Saguillo et al. (arXiv 2508.03474, August 2025) identifies two distinct overround/underround strategies on Polymarket and estimates realized arbitrageur profits near $40M: market-rebalancing arbitrage within single multi-outcome markets, and combinatorial arbitrage across logically related markets. A documented Fed-decision multi-outcome example: ">50 bps cut" at $0.001, "25 bps cut" at $0.008, "no change" at $0.985, "hike" at $0.001, summing to $0.995 — buying one share of each guarantees 50 bps on resolution. The historical analog is PredictIt's 2014–15 Democratic and Republican nomination markets, where sums of all candidates' YES prices reached over $1.55, producing "sell-all" arbitrage of up to 55% before PredictIt introduced linked markets and 10% profit fees.

Polymarket's 2024 election (17 candidates, $3.7B volume) traded a tight 99.9% binary sum on election eve, but earlier in 2024, with Biden, Newsom, and Michelle Obama all priced 3–5%, sums diverged meaningfully. The systematic edge is the favorite-longshot bias as a "structural overround" — Bürgi, Deng & Whelan confirm 5¢ Kalshi contracts win only ~2%, while 90¢+ contracts return small positive expected value. The 1990s converts-arb thesis (issuers underprice vega) maps directly: retail PM traders systematically overprice the long-tail option and underprice the favorite, and the disciplined strategy is to be a maker in high-price favorites — the equivalent of buying cheap convertibles and shorting expensive equity.

Term structure: ceasefire markets contain a tradable hazard rate curve

John Meriwether's Salomon arb group (founded 1977) reportedly produced 80–100% of Salomon's earnings through the 1980s on on-the-run/off-the-run convergence trades, swap spreads, and yield-curve flatteners. LTCM (founded February 1994 with $1.25B and Nobel laureates Scholes and Merton) ran the same playbook at 25:1 leverage, returning 21% / 43% / 41% in 1994–96 before the August 1998 Russian default flight-to-liquidity widened every convergence trade simultaneously. The intellectual core: a term structure of related instruments contains an embedded forward-rate or hazard-rate curve, and inconsistency is tradable.

This is the most underappreciated of the ten analogs. Polymarket runs nested ceasefire markets — by April 30 (0.5%), May 31 (5.2%), June 30 (7.5%), and December 31 2026 (25.5%) — that mathematically must satisfy P(T₁) ≤ P(T₂) or pure calendar arbitrage exists. From these prices the bootstrapped marginal hazard rates are: April→May ≈ 4.7%, May→June ≈ 2.4%, June→December ≈ 3.6%/month. The non-monotonic shape (high April, dip in May–June, gradual rise into year-end) is exactly the type of "kink" LTCM-era credit-arb desks would have analyzed for relative value. The same arithmetic applies to Polymarket's Fed rate-cut term structure: per-meeting "no change" probabilities of 99.7% / 93.5% / 85.5% across April / June / July 2026, "first cut by October" at 54%, "first cut by December" at 60%, and a "how many cuts in 2026" multi-outcome distribution all must reconcile by no-arbitrage.

A quant can check internal consistency among (a) per-meeting decision markets, (b) "first cut by date X" markets, and (c) cumulative-count markets — three-leg combinatorial arbitrage analogous to LTCM's swap-spread vs. on/off-the-run vs. Eurodollar relative-value triangles. The arXiv 2510.15205 paper ("Toward Black-Scholes for Prediction Markets," 2025) formally treats prices as risk-neutral martingales and proposes "calendar variance swaps" and "threshold notes" as derivatives on this curve. This remains the most theoretically rich and least empirically explored alpha in prediction markets as of April 2026.

Soros's 1992 pound trade has a $82M Polymarket twin

Stanley Druckenmiller and George Soros's September 16, 1992 short of the British pound — scaled to ~$10B in a day after Soros's "go for the jugular" instruction — netted Quantum Fund $1–1.5B as the Bank of England raised rates from 10% to 15% before withdrawing from the ERM. Julian Robertson's Tiger Management (peak $22B), Bruce Kovner's Caxton, and Paul Tudor Jones (200% gross / 125.9% net in 1987) institutionalized the playbook of cross-asset directional bets driven by macro thesis, asymmetric risk-reward, and reflexivity.

The anonymous French trader "Théo" / "Theo4" is the closest documented PM analog. Between October 1 and November 4, 2024, he deployed approximately $80M across 11 Polymarket accounts on Trump winning the presidential, popular-vote, and swing-state contracts — at peak holding ~25% of all Trump-Electoral-College YES contracts and over 40% of popular-vote YES contracts. His thesis was Druckenmiller-style: he commissioned proprietary YouGov "neighbor polling" to detect the shy-Trump-voter effect that consensus polls were missing, bought YES at ~50–55¢ when his estimated true probability was 80–90%, and realized roughly $82.3M in profit. Polymarket-priced Trump odds rising to 60–67% pre-election produced the macro spillover Soros would recognize — what equity strategists called the "Trump trade" in regional banks, small caps, BTC, and the dollar.

The 2025 tariff cycle produced a textbook macro round-trip. After Liberation Day on April 2, 2025 (10% baseline + reciprocal tariffs, 54% cumulative on China), Polymarket's "US recession in 2025" went from 38¢ to 66¢ — more bearish than JPMorgan's 60% and Goldman's 45%. By July 2025, after the so-called "TACO trade" (Trump Always Chickens Out), the contract collapsed to 22¢, a roughly 3x return for short sellers. The carry-trade analog in PMs is systematic shorting of dated longshot YES contracts on stable-status-quo questions — the favorite-longshot bias plus theta decay pays a structural premium until a Black Wednesday-style tail event (Russia-Ukraine 2022, October 7, 2023) wipes out years of accumulated theta, exactly as Soros's pound trade was the catastrophic mirror image of selling FX vol.

Volatility arbitrage works because political markets are systematically underconfident

Nassim Taleb founded Empirica with Mark Spitznagel in 1999; Spitznagel's Universa Investments returned 65–115% in October 2008, ~20% in August 2015, and famously 3,612% on invested capital in March 2020. The mechanic is comparing implied to realized volatility, buying convexity when implieds are too low, selling when rich. LTCM is the cautionary opposite — short vol/long convergence, blown up when correlated vol spiked.

In prediction markets, a binary contract at price p has variance p(1−p), maximized at 0.25 when p=0.5 — the binary analog to ATM implied volatility. If a trader's true probability estimate is 80%, buying YES at 50¢ is structurally a 30-vol-point edge, exactly Théo's 2024 trade. The empirical question is whether market prices systematically misprice this binary "implied vol":

• Iowa Electronic Markets (Berg-Nelson-Rietz 2008): 1.33 percentage point average absolute election-eve forecast error in U.S. presidential markets across 1988–2008; markets beat polls in 74% of head-to-head comparisons.
• Polymarket (platform data): aggregate Brier score 0.0843 across resolved markets; markets >$1M volume hit Brier 0.016–0.026 at 12–24h horizons — better than state-of-the-art weather forecasting and far better than sports betting lines.
• Le (2026, arXiv 2602.19520): political markets exhibit persistent underconfidence at nearly all horizons with calibration slopes 0.93–1.83 — meaning a 70¢ political contract one week before resolution corresponds to a true probability near 83%, a systematic exploitable mispricing. Sports markets are well-calibrated short-horizon but underconfident long-horizon.
• Clinton & Huang (Vanderbilt 2025): in matched 2024 election samples, PredictIt was ~93% accurate, Kalshi ~78%, Polymarket ~67% — a controversial finding Kalshi's PR team disputed.

The systematic vol-arb trades are: buy political favorites above 60¢ on long-dated markets (capture the underconfidence slope); sell longshots below 10¢ on dated geopolitical questions (harvest the favorite-longshot premium and theta); avoid mid-price entertainment markets (~62% accuracy, weakest calibration); and identify high-conviction 50¢ markets where private information is strong (Théo archetype — maximum implied variance to convert into directional edge).

The Iran ceasefire case is the modern Boesky moment

1990s electronic trading — Island ECN, Archipelago, decimalization in 2001 — destroyed dealer rents and produced a generation of latency-arb firms. Spread Networks's $300M Chicago–NJ dark fiber (2010) was leapfrogged by McKay Brothers/Quincy microwave towers; co-location and proprietary feeds (NYSE OpenBook arriving nanoseconds before the SIP) created a two-tiered information system Michael Lewis popularized in Flash Boys (2014).

The prediction-market analog has produced the first criminal indictment for prediction-market insider trading in U.S. history. On April 25, 2026, the DOJ charged U.S. Special Forces soldier Gannon Ken Van Dyke with five felonies for using classified intel about the January 2026 Maduro-capture operation in Venezuela to bet $33,000 on Polymarket and cash out roughly $400,000–$440,000. The April 2026 Iran ceasefire cycle saw at least 50 brand-new Polymarket accounts place concentrated YES bets in the hours and minutes before Trump's Truth Social ceasefire announcement; eight accounts bet ~$70,000 to win nearly $820,000, with Bubblemaps clustering $600K of profits to a coordinated wallet ring across the $170M total volume. Senators Schiff and Blumenthal demanded CFTC investigation; Reps. Torres and Moore filed bipartisan legislation. Earlier documented cases include the June 2025 first Iran ceasefire announcement, the July 2024 Trump assassination attempt (Polymarket Trump-victory contract jumped 60¢ → 70¢ within minutes), the April 2025 Pope Francis death market (33% → 99% in hours), Biden lame-duck pardons ($300K winning trade), and the Israeli Air Force officer prosecutions where one soldier told interrogators "the entire squadron is on Polymarket, the entire air force is betting."

Akey, Grégoire, Harvie & Martineau (Harvard Law Forum, March 2026) systematically screened 93,000+ Polymarket markets across ~50,000 wallets using five informed-trading signals; flagged traders had a 69.9% win rate, exceeding the null distribution by over 60 standard deviations under permutation test, with estimated $143M in aggregate anomalous profit. Gómez-Cram, Guo, Jensen & Kung (LBS/Yale, April 2026) found ~3% of traders consistently beat chance after 10,000 coin-flip permutation reruns, and 60% of "lucky winners" become losers in out-of-sample tests. The legal analog is exact: as in 1986 with Boesky, the practice will likely continue until prosecution, regulation, or both close the asymmetry between Polymarket's pseudonymous offshore CLOB and Kalshi's CFTC-regulated identity-verified DCM.

Market making is the highest-Sharpe strategy — and the hardest

NYSE specialists (LaBranche, Spear Leeds, Fleet, Van der Moolen) and NASDAQ dealers (Knight, Mayer Schweitzer, Herzog) earned eighth- and quarter-point spreads through the 1990s before Christie-Schultz's 1994 collusion paper, the 1997 order-handling reforms, and 2001 decimalization collapsed dealer rents. The successor electronic MM industry — Citadel Securities, Virtu, Jane Street, Hudson River, Susquehanna, Jump, Tower — earns spreads on millions of trades per day with tight inventory limits. The Glosten-Milgrom (1985) and Kyle (1985) framework: market makers earn from uninformed flow, lose to informed flow, and must set spreads wide enough to cover both adverse selection and inventory holding costs.

Polymarket runs an off-chain CLOB with on-chain Polygon settlement and a Liquidity Rewards Program modeled on dYdX: USDC payouts proportional to resting depth within max_incentive_spread of mid, weighted toward two-sided quotes via a quadratic spread function. As of April 2026, Polymarket allocated over $5M monthly to LP incentives (concentrated in sports/esports), plus a "Sponsor Rewards" feature letting users top up market reward pools. Early LPs reported $200–300/day on $10K capital, but rewards became "a thin bonus on top of real trading edge" because adverse selection from news jumps frequently exceeds spread capture absent automated quote-pull on news monitoring. Susquehanna disclosed a market-making relationship with Kalshi in April 2024; on February 9, 2026, Bloomberg reported Jump Trading taking equity stakes in both Kalshi and Polymarket in exchange for liquidity provision, with a team of over 20 traders for event contracts. Robinhood/Susquehanna's 2024 LedgerX acquisition completes the institutional picture.

The most economically significant finding from Akey et al. (2026) is that moving from pure taker to pure maker status reduces probability of losing money by roughly 36 percentage points on Polymarket — the single largest predictor of profitability. The top 1% of traders capture 84% of all gains; the top 0.1% capture 58.5%; 70.8% of users are net losers; 63% of trades execute at extreme prices below 10¢ or above 90¢ (the "lottery ticket" segment). The resolution risk premium is real and observable: cross-platform Kalshi-vs-Polymarket spreads on identical contracts persist at 1.5–4.5% even on liquid markets, and Cardi B / government-shutdown / Khamenei cases prove the spread cannot be fully arbitraged because settlement specifications can diverge. Tarek Mansour (Kalshi CEO) summarized why even Wall Street MMs struggle: "they can't spin up a new desk to price politics or culture in an hour."

What this means for hedge fund strategy in 2026

The five strategies that port cleanly are merger-arb (tail-end trading), event-driven (cross-contract repricing lag), convertible arb (multi-outcome overround), fixed-income relative value (term-structure hazard rates), and market making — these have documented eight-figure realized profits, peer-reviewed or working-paper academic backing, and active institutional participants. Vol arb and global macro port as conviction trades for sophisticated traders with private signals (Théo's $82M is the proof of concept). Distressed debt does not port — the 1–5¢ longshot pile is overpriced retail flow, not undervalued senior claims. Stat arb works in principle but transaction costs eat retail-size alpha. Latency arb is being exploited so aggressively it is producing the prediction-market era's first criminal prosecutions and bipartisan legislation.

The deepest insight from the 2024–2026 academic wave is that prediction markets are not the "wisdom of crowds" they are marketed as — they are venues where a small skilled minority extracts rents from a large unskilled majority, with information-edge arbitrage producing the largest wedge. This is the same structure as 1980s Wall Street. The strategies that worked then work now, the regulatory architecture is roughly fifteen years behind, and the next decade will likely repeat the 1986–2001 arc of insider-trading enforcement, decimalization-style spread compression, and the consolidation of liquidity provision into a small number of professional firms. The traders who will earn the alpha are the ones who recognize that Polymarket in 2026 is structurally closer to NASDAQ in 1996 than to a "wisdom of crowds" forecasting tool — and who execute Levy's, Buffett's, Greenblatt's, Tartaglia's, Meriwether's, and Spitznagel's playbooks accordingly.