1. The dog and the tail
A fund promising L times a day’s index move must rebalance at every close, buying after up days and selling after down days. That is arithmetic, not strategy, and inverse funds obey it too, so all of them trade one way at once. The question here is whether a trader holding daily prices can be paid for the reversal that flow should cause.
The flow, per Morgan Stanley QDS, August 2026.
| Measure | Value |
|---|---|
| Leveraged and inverse ETF assets | over $200 billion |
| Global equities sold, June 5 to July 29 | over $100 billion |
| Bought on July 30 alone | nearly $30 billion |
| Share of US equity volume, peak days | 2 to 4% |
| Rebalance owed at each close | (L squared minus L) x assets x return |
| Desk conclusion | structural flows now matter as much as fundamentals |
| Mechanism, prior work | Cheng and Madhavan, 2009; Tuzun, Federal Reserve, 2013 |
2. Two tests, one rising bar
The first test faded a pressure signal scaled by a fund-volume proxy. Its gross effect was real and stable out of sample, and costs consumed it to the basis point, the third cost-death verdict in this pipeline’s records. The second used fund assets filed at the SEC, at a higher bar for being a second look at one idea. It made money on paper and failed anyway, because almost all of that money came from one regime.
Two registered tests, judged 14 August 2026.
| Measure | Test 1, volume proxy | Test 2, filed assets |
|---|---|---|
| Sample | 4,141 trading days | thirteen funds, quarterly SEC N-PORT filings, late 2019 to mid 2026 |
| Gross return | 7.5% a year | 6% a year |
| Trading costs | 7.53% a year | 5 basis points assumed |
| Net return | minus 0.11% a year | positive on paper |
| Breakeven commission | 4.96 basis points against the 5 charged | n/a |
| t-statistic | n/a | 0.89 |
| Bar to clear | 2.50 | 2.74 |
| Concentration | n/a | 2020 supplied 113% of lifetime profit; first active day March 3, 2020; no edge after late 2024 |
| Verdict | Rejected, cost death | Rejected, one regime |
| Family status | closed at daily frequency | |
3. What the reviewer caught
An independent automated reviewer attacks every result here, mine included. Pressure’s direction equals the day’s move by construction, so every daily-bar version of this idea is a bet against big moves behind some gate.
The reviewer’s findings.
| Claim | What the record says |
|---|---|
| The volume proxy carried flow information | it was a volatility filter in disguise |
| The second test used real assets | partly a price series in costume: scaling filings by fund prices discards investor flows, the one thing that separates assets from price |
| Fund data is what pays | a filter using none of it, only that the index moved a lot that day, beat both versions on the same window |
| The author’s own work was clean | two errors now on the permanent record: a wrong provenance label, and asset anchors applied before their filings were public |
4. What stands
Daily closing prices cannot harvest this flow at realistic costs, which does not refute the mechanism. Neither test ever observed an actual flow. The capture window is the close auction and the following open, minutes daily bars cannot see, a boundary an earlier paper here mapped. The market has meanwhile voted on what manufactured flow is worth. The next version runs on options data and roll calendars, at the bar a third look costs.
Mechanical flow as a product, August 2026.
| Item | Record |
|---|---|
| Buyer | Goldman Sachs |
| Target | NEOS, founded 2022 |
| Price | up to $2.3 billion |
| Assets acquired | $30 billion of option-income ETFs |
| Sequence | its second such acquisition in months |
5. What this does not establish
Nothing here measures the close auction, actual creations and redemptions, or intraday reversion; those need finer data than daily bars, and buying it is a separate decision. The cost assumption is generous, if anything, to a strategy trading into the very auction the flow crowds. The two rejections say nothing about whether the flow moves prices, only that its echo in daily closes is too small and too era-bound to pay for itself.
Provenance, bs-prov/1.0
- Method
- Two registered tests of one declared idea, judged 14 August 2026 at family-deflated bars of 2.50 and 2.74. Verdicts, diagnostics, the cost-death arithmetic (gross 7.5% a year against 7.53% of costs, breakeven 4.96 basis points), the one-regime finding (2020 supplied 113% of lifetime profit), and both reviewer interventions are recorded in the pipeline’s cycle report and hypothesis database, including two errors of the author’s the reviewer caught.
- Sources
- Fund assets from SEC N-PORT filings, thirteen funds, quarterly, late 2019 through mid 2026. Mechanism: Cheng and Madhavan, 2009; Tuzun, Federal Reserve, 2013. Flow figures: Morgan Stanley QDS, August 2026. Acquisition figures: Goldman Sachs press release and Bloomberg, August 2026.
- Honesty
- Neither test observed an actual flow; both inferred pressure from prices and assets. The close auction, creations and redemptions, and intraday reversion need finer data than daily bars. The rejections close the daily-bar version, not the mechanism.
- Related
- Working Paper No. 25 (what daily bars cannot see), No. 29 (the ninety-second quant), No. 30 (what remains of pairs trading), and the pairs family’s parallel cost-death verdicts.