[Week 37 of 2026] The Price of Predictability

Welcome back to Price and Prejudice with a few musings from Week 37 of 2026. Today we talk about the often-discussed index rebalancing trade and the costs of having a system that is largely predictable.

Last Friday, S&P announced its quarterly index rebalance: Bloom Energy, Illumina, and Everpure will join the S&P 500, replacing Molson Coors, Trade Desk, and Builders FirstSource. The changes are effective before the open on Monday, September 21, which in practice means tracking funds implement them at this Friday's close. About $13 trillion is indexed to the S&P 500, and everyone has known since the announcement which stocks those funds will need to buy and sell, and roughly when. That two-week window is where one of the best-known trades in modern markets lives.

Traders have built strategies around index changes for four decades. This is the rare corner of markets where a press release tells you the constituents and the date, which is enough to have a rough estimate of the coming fund flows. The scale of the opportunity has been brought prominent this summer, when Bloomberg reported that two index-rebalancing teams at Millennium generated about $3.7 billion of gross trading profit in June, more than half of the firm's profit that month before fees. Among the contributors was SpaceX, which entered the Nasdaq 100 just 15 trading days after its IPO under a newly adopted fast-entry rule; JPMorgan estimated that index trackers needed to buy about $4.3 billion of the stock.

So what exactly is the trade? I find it easiest to think in terms of three actors and two dates. The index committee announces on a Monday that a stock will join the index in two weeks. The index fund buys at the effective-date close. The arbitrageur buys after the announcement and sells to the index fund at that close. The trade makes money when the price appreciation over those two weeks exceeds the costs of carrying and hedging the position. The economic interpretation of this trade is that the arbitrageur effectively charges the index fund for immediacy, since someone has to hold the inventory during the days the fund will not. As Matt Levine put it in his coverage of the SpaceX episode, the index funds have to buy the stock at whatever the price is, and everybody else gets to trade around that obligation.

One question is why the fund actually waits for the close. Well it's not because a law requires it: the prospectus typically says the fund seeks to track the index, and managers retain real discretion over execution. One reason is that the incorporating the new stock at the official closing price (so trading near the close, often in the closing auction itself) minimizes the main source of measured tracking difference. Any price pressure is still a real cost to the fund's investors, who end up buying the stock at a temporarily inflated price. But the index adds the stock at that same closing price, so when the pressure later unwinds, the fund and its benchmark absorb the decline together. So at least the cost does not appear as tracking error.

This footprint shows up in the volume data. Greenwood and Sammon compute that nearly 30% of all trading volume in the two months around an index addition now occurs on the effective date itself, up from about 15% in the early 1990s, and index trackers buy 7 to 8% of a company's shares outstanding upon addition. In a separate paper, Bogousslavsky and Muravyev document that closing auctions grew from 3.1% of daily volume in 2010 to 7.5% in 2018, growth they link to the rise of indexing and ETFs.

The surprising fact in the data is that the classic version of this trade has mostly faded. The same Greenwood and Sammon paper shows that the average price pop from joining the S&P 500 fell from 7.4% in the 1990s to 0.3% over the past decade, even as the assets tracking the index multiplied. Their evidence points to more institutions supplying shares around the events, with mixed evidence that some of the return also moved ahead of the announcements. Either way, the average addition effect that launched this literature is close to zero.

So what explains the Millennium numbers? Part of the answer is that the trade has migrated. In a recent paper, Harvey, Mazzoleni, and Melone show that the same logic now applies one level up, to the rebalancing of entire portfolios rather than index membership. Specifically, think a 60/40 fund that rebalances on a calendar sells equities after they rise and buys after they fall, often in predictable size and at predictable month-ends (following policies described in public documents). The authors estimate these flows cost investors about $16 billion a year, roughly $200 per American household.

The other part of the answer is that the surviving version of the trade carries real risk. A membership bet can fail when the committee surprises you, and a crowded trade can fail when other traders anticipate your positions and the unwind turns violent. The June profits are the right tail of that distribution; the left tail arrived fifteen months earlier, when Millennium lost about $900 million as volatility forced crowded index-rebalancing positions to unwind. Indeed, a quant researcher at a competing fund told Bloomberg that so much risk tolerance and craft goes into this conceptually simple trade that very few teams can do it well.

Who ultimately pays for all this? The reflexive story is that fast traders profit at index investors' expense. In a new paper, Pegoraro, Sammon, and Shim offer a more careful account. In their model, anticipatory traders accumulate shares before the event and supply them to trackers at the close. As a result, index investors pay the price impact but receive the liquidity, and the measured net cost of standard reconstitutions turns out to be modest. The costly exception is fast-track additions, where the compressed timeline works against the index funds, which is a reasonable description of the SpaceX event. (In companion work, Sammon and Shim estimate that the mechanical response of indexes to composition changes, additions, deletions, issuance, and buybacks, costs index investors 46 to 69 basis points a year.)

Some of these costs are avoidable. For example, one can implement wider trading windows, staggered or randomized execution, and index rules with lower turnover. Sammon and Shim estimate the savings from lower-turnover designs at around 50 basis points a year. In the end, investors value funds that follow their benchmark without exercising judgment, and the papers here are an accounting of the costs of that arrangement.