VIX Term-Structure Capitulation — 1d

Status: Regime entry: a record only — worked in one test window, not the other · Regime #8 · Listed 2026‑10‑14 Instruments: SPY (it cleared on QQQ and IWM too) · Timeframe: 1d · Direction: long only Frequency: ~5 trades/yr per ETF · Fees: 0.03%/side

Why it is in the regime section, not the library

It failed the six‑check bar. It is listed under the regime bar, which asks four things of one of the two test windows: a net profit after costs, still a profit without the best trade, a result above 95% of 400 random‑entry runs with the same exit, and at least 20 trades. It clears all four from January 2010 to January 2018 on SPY (38 trades), and on QQQ and IWM too: on all three, none of the 400 random runs beat it. From January 2018 to August 2026 it kept making money but fails the third. Everything below the setup is disclosure, not a pass.

Where it fails the full bar:

• Random entry, 2018–26: it made +26.9% on SPY, +44.0% on QQQ and +23.7% on IWM, but no more than random entries with the same exit did: the 75.3rd, 79.5th and 72.3rd percentiles, against random medians of +14.4%, +24.0% and +9.5%.
• That is the only check it fails on these three. In both windows it was profitable, still profitable without its best trade, drew down 8.1 to 14.5 summed points, still passed at 0.1% a side, and did better filled at the next day's open. It was one commit from shipping as library entry #7, and it is the strategy that made beating random entries the sixth check of the vetting bar.
• Cross‑asset: EFA, the fourth ETF tested, made +17.2% in 2010–18 at the 94th percentile, just under the bar, and lost money without its best trade in 2018–26.

It came out of sweep #11, pre‑registered on 2026‑09‑18: does the volatility term structure, spot fear (VIX) against three‑month fear (VIX3M), time equity index ETFs the way positioning times crypto? Three readings of it on SPY, QQQ, IWM and EFA in two windows make 24 window tests; the random‑entry null ran on this one's eight. With a 95th‑percentile bar, about one window test in twenty clears by luck alone. Three of the eight cleared, all in 2010–18, all at the 100th percentile.

Why the app can't run it

Its entry reads two series the app does not trade: CBOE's daily closes of the VIX and VIX3M indices, divided one by the other. The strategy language reads an instrument's own candles (and funding, on crypto perpetuals); it has no atom for an outside index and no way to divide one series by another, so the crossing of 1.0 can't be written. The exit, a 2×ATR(14) trailing stop, is ordinary.

The setup

Market: US index ETF SPY, daily bars (the same rule on QQQ and IWM). Longs only. Signal: CBOE's VIX close divided by its VIX3M close, the same day. Entry: the ratio crosses back below 1.0, spot fear falling back under three‑month fear after an inversion. Filled at the day's close. Exit: trailing stop 2×ATR(14) below the highest price since entry. Costs: 0.03%/side.

Evidence: two windows on daily ETF candles and CBOE's VIX and VIX3M closes

QQQ:

IWM:

Returns are per‑trade percentages summed, like every return in the research log. Holding SPY made +135.5% and +177.9% over the same windows, with drawdowns of 19.4% and 34.1%; QQQ made +235.6% and +334.1%, IWM +138.3% and +89.2%.

• Random entries, 2010–18: 400 runs of the same exit on random days had medians of +9.5% (SPY), +18.2% (QQQ) and +11.7% (IWM), and 95th percentiles of +33.0%, +46.9% and +47.1%. None beat the real result on any of the three.
• Random entries, 2018–26: medians of +14.4%, +24.0% and +9.5%, and 95th percentiles of +50.4%, +65.9% and +48.5%. SPY's +26.9% sits inside a random distribution whose own 95th percentile is +50.4%.
• Fees: at 0.1%/side, about 3.3 times the cost assumed, all three still passed in both windows.
• Fills: filled at the next day's open instead of the signal day's close, it did better. It also held up with the VIX reading lagged a full day. With both at once, only QQQ still passed.

The record by year

*2026 through July. Each trade counts in the year it was entered. The trade counts are SPY's; QQQ traded 77 times in all and IWM 79, never more than one a year away from SPY. No trade on any of the three in 2023.

• Spread out, not concentrated: SPY lost money in two years, 2011 (−5.6%) and 2018 (−5.7%). 2018 was the worst year on all three: −5.7%, −9.0% and −11.7%.
• Last high: May 2026, at +95.3% summed since January 2010. It ended July 2026 at +95.3%, 0 points below: SPY's record was at its high. QQQ's last high was also May 2026 (+135.6%, where it ended); IWM's was December 2025 (+108.1%), and it ended 0.7 points below.
• Last twelve months (August 2025 to July 2026): 6 trades, +5.2%. QQQ made +12.3% and IWM +6.3%, on six trades each.

What the result depends on

Sweep #11 ran its neighbors because the result showed genuine signal (research log, sweep #11, “V2 — the same entry with a trailing exit”):

The log calls that gradient monotone and mechanical rather than a knife‑edge. No neighbor was run against the random‑entry null. Fee stress and fill timing are above (sweep #11, “Results” and “The sixth test”).

Before you rely on it

• In 2018–26 it kept making money on all three — +26.9% on SPY, +44.0% on QQQ, +23.7% on IWM — but no more than random entries with the same exit did (72nd–80th percentile).
• It passed every other check in both windows and was one commit from shipping. It is the strategy that made beating random entries a gate.
• It is a dip‑buy: it buys after a panic and sells into the recovery, the same family as library entries #2 and #4. Only 1.3–5.2% of its entry days were also entry #2 signal days, but the exposure is the same kind.

Every number here can be re‑run: npx tsx scripts/research/sweep11-vol-term-structure.ts for the windows, the fee stress, the fills, the neighbors and the null. The record by year, the last twelve months and the last high are the same runs' trades, grouped by the date they were entered.