The Index-Inclusion Trade: how Millennium's rebalance pods made $3.7B, and how to replicate it

In June 2026, two trading pods inside Millennium Management made a combined roughly $3.7 billion, more than half of the firm's approximately $6.6 billion of pre-fee profit for the month, and lifted the firm to +4.1% in June, +10.5% year to date. The teams are run by Glen Scheinberg in New York (the larger book, known internally as SRBL) and Pratik Madhvani in Dubai. Both specialize in index-rebalance arbitrage: positioning ahead of the forced buying and selling that index funds must do when index membership changes. [reported: Bloomberg 2026-07-06, corroborated by four secondary re-reports]

Five events made June unusually rich: the S&P 500 June quarterly rebalance, the Nasdaq-100 June rebalance (the first under new membership-changing rules), the Russell reconstitution, the fast-track inclusion of newly public SpaceX, and quarter-end multi-asset rebalancing. The single most-discussed winner was Marvell (MRVL) joining the S&P 500.

This note defends three claims. First, index-rebalance flow is forced and partly predictable. Second, the naive version of the trade has decayed toward zero since roughly 2015. Third, the professionals still clear billions through breadth, leverage, superior prediction, and disciplined timing, not through the textbook trade. One framing correction matters up front: the same strategy lost Millennium about $900 million in early 2025, with further losses in November 2025. The June 2026 windfall is the reversal of a drawdown, not a monthly annuity. [reported]

An index fund's job is to minimize tracking error, not price impact. When a name enters or exits the index, every fund tracking that index must transact, largely at or near the effective-date closing auction, regardless of price. The size of that order is knowable in advance: it is approximately the name's index weight multiplied by the assets tracking the index.

The pool is enormous. As of year-end 2024, roughly $13.0 trillion was indexed to the S&P 500, of which about $9.0 trillion is physical index-fund assets that mechanically buy shares (the remainder is derivatives notional and does not). A further ~$7.2 trillion is benchmarked against it. Roughly $12.2 trillion is benchmarked to the Russell US indexes. [P: S&P DJI Annual Survey of Assets; LSEG 2026]

Concentrated, calendar-dated, one-directional demand against a finite tradable float moves price. The right way to measure the pressure is not the dollar figure but days-of-ADV: the forced dollars divided by the stock's daily dollar volume. A small-float name facing several days of forced buying is where the effect is largest, and it is the primitive our predictor ranks on.

"Index rebalancing" is not one event but a portfolio of structurally different ones. Breadth across them is the professionals' first real edge. They range from the purely mechanical to the fully discretionary.

The spectrum, from rules to discretion

Russell is the most predictable. Membership is set by ranking every US stock by total market capitalization on a single rank day (April 30 for the June 2026 event): the largest 1,000 form the Russell 1000, the next 2,000 the Russell 2000. A banding rule (a name only switches index if it moves beyond ±2.5% of cumulative market cap around the breakpoint) makes the outcome computable weeks ahead. 2026 is the first year Russell reconstitutes semi-annually, adding a December cycle. [P: FTSE Russell ground rules, May 2026]

Nasdaq-100 is rules-plus-ranking. Full reconstitution is annual (December), but since May 1, 2026 the quarterly rebalances can also change membership: constituents outside the top 125 by full market cap are removed, and a top-40 new listing can enter via Fast Entry. That rule is exactly how SpaceX (SPCX) joined the Nasdaq-100 effective July 7, 2026, days after its June 12 IPO. [P: Nasdaq-100 methodology 2026]

S&P 500 is rules-plus-committee. A name must clear hard gates (unadjusted market cap ≥ $22.7B, float-adjusted cap ≥ ~$11.35B, four trailing quarters of positive GAAP earnings and a positive most-recent quarter, a liquidity ratio ≥ 0.75, US domicile, common stock only, no new multi-class shares, 12 months seasoned), but eligibility is necessary, not sufficient. A name is added only when a slot opens (a member is acquired, taken private, or falls out) and the committee judges it the best size and sector fit. More than 70% of additions now migrate up from the S&P MidCap 400. [P: June 2026 S&P U.S. Indices Methodology; Greenwood-Sammon]

The index effect is one of the most-studied anomalies in finance. Early work found a roughly 3% abnormal return around S&P 500 additions and argued about whether it was temporary price pressure or a permanent, downward-sloping demand curve. The premium grew as indexation grew, peaking around 2000 at roughly +8.8% for additions and −15.1% for deletions from announcement to effective date. [P: named papers]

Then it faded. Greenwood and Sammon's The Disappearing Index Effect (Journal of Finance, 2025) shows the average S&P 500 addition abnormal return fell from about 7.4% in the 1990s to roughly 1% (statistically indistinguishable from zero) in 2010-2020, and deletions from about −16% to −0.6%. The price-impact multiplier fell by a factor of about twenty. The drivers: arbitrageurs crowded in, more additions migrate from the MidCap index (netting off the demand shock), and the market simply became more liquid and efficient at absorbing the flow. [P]

What the data shows now

We reconstruct the effect from scratch: every S&P 500 change since 2010 (n=229 additions, 111 deletions with enough price history), market-model abnormal returns benchmarked to SPY, three significance tests. Two results stand out, and together they explain why the trade has changed shape.

The run-up is gone. Averaged across the full sample, an addition's abnormal return into its effective date is small and statistically insignificant (about +1% over the prior three weeks, indistinguishable from zero once we deflate for the fact that all changes at one rebalance share a date). The old announcement pop has, on our data too, faded to noise.

What remains is a reversal. After the effective date, additions give back −3.4% over the next month (t = −5.5) and −6.3% over two months (t = −6.0); the full round-trip is negative (−2.9%, t = −3.2). Deletions do the opposite: oversold into the event, they bounce +3.1% and +8.3% over the following one and two months (t = +2.1 and +4.2). These reversal effects are the most statistically robust in the entire study. The tradeable edge has migrated from "buy the addition" to fading the forced flow: the passive complex overpays into the rebalance, and price mean-reverts afterward. This is temporary price pressure (Harris-Gurel) outliving the permanent premium that decayed away.

Windows in trading days relative to the effective date (announcement is typically 5-12 trading days earlier, so the pre-effective windows blend the faded announcement move with the run into the rebalance). KP-t is the Kolari-Pynnonen statistic, which deflates for the calendar clustering of same-rebalance events; the reversal survives even that conservative bar in magnitude. [ENGINE]

Marvell (MRVL) → S&P 500. S&P announced on June 5, 2026 that Marvell and Flex would join the index effective before the open on June 22 (Marvell replacing Pool Corp, Flex replacing Campbell's; a two-in, two-out batch). Three days earlier, on June 2 at Computex, Nvidia's Jensen Huang had called Marvell "the next trillion-dollar company," and the stock had already jumped 25-33%. A rebalance arbitrageur would have accumulated ahead of, and into, the June 22 forced buy. [P: S&P DJI 2026-06-05; secondary on prices]

SpaceX (SPCX) → Russell 1000 and Nasdaq-100. SpaceX went public on June 12, 2026 at $135 (about $75 billion raised, the largest offering on record). It entered the Russell 1000 at the June reconstitution (with estimated forced buying of $22-27 billion, a market estimate) and the Nasdaq-100 via Fast Entry effective July 7. It was not fast-tracked into the S&P 500, which enforces a 12-month seasoning and GAAP-profitability test. [P: SEC 424B4; Nasdaq IR; secondary on the $ estimate]

Russell reconstitution and quarter-end. The Russell reconstitution (rank day April 30, effective June 26) forces two-sided flow across hundreds of names as they migrate between the 1000 and 2000, and quarter-end multi-asset rebalancing added a final layer. Breadth across all of these, at leverage, is how a pod turns a per-event edge of a few percent into billions.

The professional process reduces to four stages: predict membership, accumulate ahead of the announcement or effective date, hedge the position index-relative, and exit into the forced passive buy. The position is run market- and often sector-neutral, so the profit is the abnormal return of the event, not a directional bet. The exit is the crucial discipline: the run-up is real, but it reverses, so the edge belongs to whoever sells into the effective-date auction rather than holding through it.

The signal is public. The economics are not symmetric.

Individual investor

What is realistically playable: the mechanical events (Russell banding, Nasdaq-100 ranking) where the outcome is computable from public data, and S&P adds triggered by an already-announced acquisition of a current member (the highest-certainty slot). Position early, on the prediction rather than the press release, because the announcement move is largely gone by the time you read it. Express through options to get leverage and defined risk without a borrow franchise. Expect a much smaller edge, real capacity limits, and no ability to be the liquidity at the print.

Institutional

The full version needs probabilistic membership models across every index, sizing driven by the passive dollars that must transact, prime-broker financing and cheap borrow for the short legs, breadth across dozens of events per year, and execution that provides liquidity into the rebalance rather than paying for it. This is what converts a 2-3% per-event edge into a multi-billion-dollar month, and it is also what a retail account cannot replicate.

The remaining edge has migrated away from the textbook trade toward structural refinements. Our backtest of the legs, net of realistic cost, makes the point.

Per-event hedged (index-relative) return, net of 1bp commission and 15bp slippage per side, plus borrow on short legs. Deflated Sharpe accounts for the number of legs tried. [ENGINE]

The trade is capacity-constrained: past a few days of ADV you become the marginal flow and move the price against yourself. Yet the biggest, most-certain flows are also the most crowded, so a retail account is squeezed at both ends. The decisive risk reminder is the historical record: this same strategy cost Millennium about $900 million in early 2025. A per-event edge of a few percent, run at leverage across many events, is a real business, but it is a volatile one, not a monthly annuity.

The companion engine ranks current S&P MidCap 400 members (the migration ladder) against the S&P 500 hard gates and by forced-flow magnitude (days-of-ADV). CONFIRMED-eligible names clear every gate; WATCHLIST names miss exactly one. This is the transparent, reproducible version of "who's next."

[ENGINE] Market cap and float are yfinance estimates; true float-adjusted index weights require vendor data. Committee discretion caps S&P prediction accuracy; only Russell is deterministic. Not a recommendation.

Method and reproducibility

Event set from the public S&P 500 change history (effective dates); market-model abnormal returns estimated on [−260,−25] trading days; significance by plain-t, BMP, and Kolari-Pynnonen; decay by era and a CAR-on-year regression; cost-aware legs with a borrow sweep; anti-overfit battery (Deflated Sharpe, jackknife by era, sign test, Benjamini-Hochberg). Full code, config, and the primary-source verification ledger ship with the engine.

This document is research on market structure and positioning for educational purposes. It is not investment advice and contains no recommendation, price target, valuation, or allocation. Past relationships between index events and returns may not persist; the effect studied here has demonstrably decayed. Figures are drawn from primary sources (index-provider methodologies and press releases, SEC filings, the S&P DJI Annual Survey of Assets, and the original academic papers) and from a reproducible event-study engine built on public price data; reported hedge-fund figures originate with Bloomberg (2026-07-06) and are labeled as reported, not independently verified. Company names appear only as factual examples of index events, not as recommendations. Full primary-source citations are in the engine's verified-facts ledger.

WALL STREET PROMPT · MARKET-STRUCTURE RESEARCH · JULY 2026