Two bets on the VIX, one for each extreme. When the VIX crosses above 25, fear is high and the bet is that it overshoots: go long volatility through UVXY. When it crosses below 20, markets are calm and futures sit in contango: short volatility through VXX and collect the premium that hedgers pay to insure against a crash. The short leg earns a real, published premium; the long leg loses to roll decay most of the time and needs the timing to be right.
volatility regime / volatility risk premium
VIX regime switch
The idea
How it was tested
The rules were locked on 2026-08-27 and first measured on 2026-10-17, on a frozen data snapshot. Entry at the next open, one position at a time, $2,000 per trade, VIXY standing in for VXX over the whole period. Exit when the VIX returns to the neutral zone, after 20 days, or at −15% on the product, filled at the open if it gaps through. Costs: 15 bp per side plus 5% a year to borrow the short.
Two checks were added before the run. Jensen alpha with a separate beta for each leg, because the two products move very differently. And a timing placebo: 500 versions with the VIX series rotated against prices, keeping products, exits and costs, to see whether the signal adds anything beyond being short volatility.
Where it died
Before any of that, on 2026-08-04, it had been recorded as validated: 318.3% a year in-sample and 327.7% out-of-sample, with no alpha gate. It ran in the demo on that record. On 2026-10-17 the demo executor was found to compute its own entry price and P&L from the VIX level. It never sent an order.
The first real measurement failed every condition. Jensen t was +0.17 on 105 trades: +0.48 for the long leg and −0.78 for the short. The placebo put the signal at the 54th percentile, where the bar was the 95th. Alpha per trade was 0.24%, below the 0.30% friction line, and t was negative in both halves of the sample. Sharpe was 0.16.
The record
Thesis
VIX regimes revert at the extremes. When VIX crosses above 25, go long volatility (UVXY); when it crosses below 20, short volatility (VXX), which collects the volatility risk premium that hedgers pay through futures contango. Exit on return to the neutral zone, after 20 days, or at a −15% stop.
Specification
- signal
- VIX close crosses above 25 (long UVXY) or below 20 (short VXX) — rules locked 2026-08-27
- instruments
- UVXY for the long leg; VIXY as a stand-in for VXX over the whole period (same index)
- entry
- next open, one position at a time
- exit
- VIX back below 22 / above 23, 20 calendar days, or −15% on the product (filled at the open if it gaps through)
- costs
- 15 bp per side plus 5%/yr borrow on the short
- sizing
- fixed $2,000 notional per trade
Gates
Pre-registrationpass
Was the hypothesis, its parameters and its pass/fail rule written down before any data was looked at?
First real measurement of rules written in August; ambiguities resolved in writing before the run.
Gate 7 — Jensen alphafail
After regressing out market exposure, is there alpha left?
t ≥ 1.65 (Jensen with separate betas for the long and short legs)
t (HAC)+0.17trades105
Per leg: long t +0.48, short t −0.78. Timing placebo: daily alpha −0.75 bp, percentile 54 of 500 versions with the VIX series rotated against prices (needed > p95).
Gate F — Frictionfail
Is the alpha per trade large enough to pay for its own execution?
alpha/trade ≥ 0.30%
alpha / trade %+0.24
Stabilityfail
Is the alpha present in both halves of history?
t > 0 in both halves
t 2015 2019-1.15n 2015 201931t 2020 2026-0.36n 2020 202674
Why it died
No alpha and no timing: the VIX regime signal did no better than the same strategy with the VIX series misaligned (54th percentile — exactly chance). The positive average P&L is a single trade: the UVXY long of February-March 2020 made +429%, and without it the long leg has nothing. The sleeve's maximum drawdown was −97.6%, and 7 of 31 stops gapped through. The August "validation" (318.3% / 327.7% a year) was not a backtest on real prices, and the demo executor invented its own prices and P&L from the VIX level without sending orders — nothing it showed was evidence.
What it taught
A short-volatility leg looks like alpha in a linear regression because it is paid for tail risk the regression cannot see. Testing timing against a placebo that keeps products, exits and costs but breaks the signal-market alignment is what separated "earns the premium" from "times the regime" — and here it did neither.
Verdict history
- 2026-08-04Recorded as validated: 318.3% in-sample / 327.7% out-of-sample annualized; Sharpe listed as expected, not measured; no Gate 7/8
- 2026-08-27Proper pre-registration of the rules; results sections left empty
- 2026-10-17Found that the demo executor invented prices from the VIX and sent no orders
- 2026-10-17First real measurement: fails all four conditions (Jensen t +0.17, placebo percentile 54, Sharpe 0.16, drawdown −97.6%) No alpha