Archive / Equities & volatility / E828
Leveraged volatility fund rebalancing leaves no next-day reversal
Next-day VIX futures returns show no detectable reversal in any of four eras of fund leverage, and the 2024-2026 holdout rule's +16 bp average per trade rests on 3 days, one of them a stress day.
Leveraged and inverse volatility funds must rebalance at the close in proportion to the day's move in VIX futures. That forced buying after rises and selling after falls should push futures back the next day, most strongly when fund leverage was highest. Across 264 monthly contracts and four eras of fund leverage, the next-day slope was not distinguishable from zero in any era (+0.008 in the era with no such funds, -0.060 to -0.080 in the others, all intervals including 0). A rule trading against the biggest daily moves earned +16 bp per holdout trade; the top 3 days were 371% of its profit.
Leveraged and inverse volatility funds have to rebalance at the close, so VIX futures reverse the next day.
Why people believe it
A fund that returns 2x (or -1x) the daily move of a volatility futures index must hold exposure proportional to its assets times its leverage. After a rise in volatility futures, a long fund has to buy more at the close and an inverse fund has to sell; the funds and their hedgers trade at the close whatever the price, and the size is known from the day's move. The resulting price pressure should fade the next day. The test also uses history as a natural experiment: leveraged and inverse products did not exist in the first era, were large in the second, were smaller in the third and returned in the fourth, so the reversal should be stronger where the funds were larger.
Why it would persist The argument, as stated before testing
Capacity is small (about $135-450 thousand a year for an ordinary futures account, as estimated before the test), so large funds would have no reason to compete it away.
For every trading day we measured how much a 30-day constant-maturity blend of the first two VIX futures moved. We then asked what the front contract did over the following day, and estimated that relationship separately in four eras that differ in how large the leveraged and inverse funds were. If forced rebalancing mattered, the next-day slope should be negative in the two high-leverage eras and flat in the other two. We then turned it into a rule: after the biggest up days, short the front contract to the next settle; after the biggest down days, go long. The rule was judged on a holdout from 2024-01 on.
Holdout net mean above 0 with the interval above 0 on the short and the long side separately, alpha against the unconditional short carry above 0, and the result survives dropping the stress days.
The first era (no funds) shows the same reversal, or holdout net at or below 0, or more than 50% of holdout profit from 3 days or fewer, or a worst day below -40% of notional.
Per trade, 1 tick (0.05 index points x $1,000 multiplier = $50) plus $2.50 per contract per side ($5), divided by the contract's notional (futures price x $1,000). Settlement prices are not tradeable fills.
Era comparison: the first era (2004-03..2010-11) has no leveraged or inverse volatility funds, so the same reversal there would be generic volatility mean reversion, not fund flow. Alpha of each side versus the unconditional same-side carry on holdout days. Dropping stress days: the rule named February 2018, March 2020 and one day in August 2024, but the holdout starts in 2024-01, so only the August 2024 day is removed.
The rule failed two pre-registered tests: the top 3 holdout days equal 371% of profit (limit 50%) and the worst trade lost 47.1% of notional (limit 40%). Holdout net was +16 bp and the first-era slope +0.008, so it did not fail on those two.
- Next-day slope of the front contract on r(t): R0 (no funds) +0.008 with 95% interval -0.04 to +0.08; R1 -0.060 (-0.16 to +0.12); R2 -0.080 (-0.16 to +0.01); R3 -0.066 (-0.22 to +0.12). Every interval includes 0.
- Differences between eras (means of bootstrap differences): R1 minus R0 -0.058, R3 minus R0 -0.075, R1 minus R2 +0.031, R3 minus R2 +0.014. All intervals include 0, so no slope difference is detectable; the most negative point estimate is R2, the low-leverage era.
- Holdout, short after big up days: 68 trades, +148 bp on average, interval -152 to +384 bp, which includes 0. Alpha against the unconditional short carry: +166 bp. Across all 118 holdout trades the gain depends on one stress day (August 2024).
- Holdout, long after big down days: 50 trades, -162 bp on average, interval -301 to +7 bp.
- All 118 holdout signal trades average +16 bp. Without one stress day (August 2024) they average -0.3 bp, and the top 3 days equal 371% of total holdout profit, so the other trades lost money in total.
- The worst holdout trade lost 4,714 bp (47.1% of notional).
No era shows a reliable reversal
Slope of next-day front-contract return on the day's index return, with week-block bootstrap 95% intervals. A reversal from fund rebalancing would be negative in R1 and R3 and near 0 in R0 and R2. Every interval includes 0, and the era with the most negative slope is the low-leverage one.
Chart data (slope)
| Mean | 95% low | 95% high | |
|---|---|---|---|
| R0 2004-2010, no funds | 0.01 | -0.04 | 0.08 |
| R1 2010-2018, high leverage | -0.06 | -0.16 | 0.12 |
| R2 2018-2022, low leverage | -0.08 | -0.16 | 0.01 |
| R3 2022-2026, high leverage | -0.07 | -0.22 | 0.12 |
Slope does not follow fund leverage
Means of the bootstrap differences between era slopes, so they can differ slightly from differences of the listed slopes. The mechanism needs R1 and R3 below R0 and R2; they are below R0 but about the same as R2 (+0.031 and +0.014). The intervals of all four differences include 0.
Chart data (slope difference)
| Slope difference | |
|---|---|
| R1 minus R0 | -0.06 |
| R3 minus R0 | -0.08 |
| R1 minus R2 | 0.03 |
| R3 minus R2 | 0.01 |
The holdout: the two sides disagree
Net result per trade after costs in the 2024-2026 holdout, with week-block bootstrap 95% intervals. Shorting after up days earned +148 bp (interval -152 to +384, which includes 0); going long after down days lost 162 bp. All trades together average +16 bp, and -0.3 bp without the one stress day.
Chart data (bp per trade)
| Mean | 95% low | 95% high | |
|---|---|---|---|
| Short after up days (68 trades) | 148 | -152 | 384 |
| Long after down days (50 trades) | -162 | -301 | 7 |
| All 118 signal trades | 16 | Not reported | Not reported |
| All trades without stress days | -0.3 | Not reported | Not reported |
Why it failed Interpretation · not established
The short-side gain looks like short-volatility carry on up-spike days. The mandated rebalancing is likely front-run intraday, which would leave no close-to-close reversal. The test used settlement prices only and did not measure intraday trading, so this is a reading, not a result.
Lesson
A flow that is mandated and known in advance is not necessarily one that can still be traded profitably after the close. The era comparison is the control: the era without funds shows no reversal (+0.008) and no era with funds differs from it detectably. And on a series dominated by a few volatility spikes, a rule's average is mostly a statement about those days: the top 3 days were 371% of holdout profit, so the remaining trades lost money in total.
Scope of this result
Covers monthly VIX futures settlements from 2004-03 to 2026-10, a close-to-close horizon of one trading day, and a 30-day constant-maturity trigger. It does not cover intraday or last-hour execution, weekly VIX futures, or the actual rebalancing notional of individual funds, which was not estimated.
A rejected experiment applies to the tested conditions and does not prove an idea can never work. bp = basis point, 0.01%.
1. Test design
Pre-registeredDataUniverseSignalExecutionCostsControlsSample splitPass criterionKill criterionPower2. Result
ObservationsInstrumentsNet mean95% intervalMedianWorstFinding 1Finding 2Finding 3Finding 4Finding 5Finding 63. Interpretation
Why it failedLessonScope4. Reproducibility
StepsGrade