Does SPY actually stay inside the expected move?

Every options seller leans on the same picture: implied volatility implies a one-standard-deviation "expected move," the stock should land inside it about 68.3% of the time, inside two of them about 95.5%, and you sell your strikes outside that band. It is the engine under the expected move, the iron condor and every strangle. So we checked it against 21 years of real history: for every trading day, we drew the VIX-implied band and asked whether SPY's realized move actually stayed inside it.

The short answer: the band is real, and it is a little too wide. SPY stayed inside the 1-SD expected move 83.3% of the time - not 68.3% - and inside the 2-SD move 98.6% of the time, not 95.5%. For a premium seller that sounds like free money. It isn't - and why it isn't is the whole point of this page.

How often SPY stayed inside the band, by horizon

The expected move is the same formula at any expiration - price × VIX/100 × √(days/365) - so we ran it across the horizons people actually trade. Every row is the share of overlapping daily-start windows whose realized move finished inside the band:

Every horizon lands in the same place: the band catches the move 82.3% to 83.6% of the time at 1 SD and 98.3% to 98.6% at 2 SD. At the 30-day horizon, that is 15.0 percentage points more than the textbook at 1 SD. At the band, it is not a coin flip - the move stays home far more often than it bolts.

Why it beats the textbook - and it is not what you think

The obvious explanation is "VIX runs scared, so the band is always too fat." That is wrong here. Over these 21 years, average implied volatility (VIX 19.2) and average realized volatility (19) were a rounding error apart - a ratio of 1.01. The band is not wide because VIX is cautious; it is sized about right.

The real reason is the shape of returns. The "68.3%" rule assumes a normal bell curve. Real markets are not normal - they are peaked and fat-tailed. Picture the bell curve pinched in the middle and the spare probability shoved out into the tails: most days go almost nowhere - so the band over-covers the quiet center, catching 83.3% instead of 68.3% - and the few days that move go far further than the curve allows, so the breaches are monsters. Same redistribution, both effects: the band's reliability and its danger come from the one fact.

It also holds up across regimes, because VIX scales the band to conditions - it is wide when fear is high and tight when it is low, and the hit rate barely moves:

The breaches: up more often, down much harder

When the band broke, direction told two different stories. Up-breaches were slightly more frequent (8.7% of 30-day windows versus 8.1% down, or 51.8% of breaches, and the gap widens the longer you hold) - the market drifts up over weeks, so a symmetric band gets nudged through the top. Harmless: a stock running past the upside of your expected move is a good problem - the worst upside overshoot in 21 years was just +7.2%, a 2.4-sigma nudge next to the craters below.

Down-breaches were slightly rarer, but the worst extremes were brutal. Every one of the 8 breaks in the worst-breach list was downward, and they did not just clip the edge - they detonated through it:

The Feb-Mar 2020 crash is the one to sit with: the band, drawn on 2020-02-19's VIX, was ±4.12% for the month; SPY fell -32.4% - a 7.9-sigma move, the kind a normal model says happens once in the life of the universe. It happened over this 30-day window. A bell curve is not just wrong about the tail - it is wrong by astronomical margins, and the 8 worst extremes in this sample were all downward. The expected move tells you where the stock usually lands; it tells you nothing about how far it can go when it doesn't.

What it means if you sell premium

The band had a 83.3% historical containment rate across 5,374 overlapping, non-independent 30-day price windows. That is what this study measured. It did not measure option premiums, expiry trades or profit and loss, so it does not establish a trading edge or show that a Seller outside the band made money.

"Too wide" describes historical price containment, not which strike you should sell. Moving closer to the money changes the premium, probability and loss profile, and this dataset cannot judge that trade-off. The width you give up is still the cushion you keep when the 7.9-sigma month hits.

But the 1.4% of windows that blew through even the 2-SD band is the entire risk, and it is not spread evenly - it pools in crashes and overshoots by multiples, on the downside, exactly when you are short and leveraged to it. The honest way to use the expected move is as a probability, not a guarantee: set strikes with it, then size the trade so the 7.9-sigma month - the one that will come - is survivable, not fatal. A defined-risk structure (a condor's long wings, a spread instead of a naked put) is how you cap that tail; the iron condor calculator shows the trade-off.

Two practical reads. First, the band is a sound default - reach for the expected move calculator to place strikes, and trust that it errs a little safe in calm markets. Second, sell that band when it pays you most: premium is richest when implied vol is high, so check it with the IV rank calculator first - high IV both fattens the premium and widens the band that has to hold.

Caveats - read these

• Overlapping windows. Every trading day with enough forward data is a start, so the windows overlap - a single crash shows up in many of them. The 30-day sample has 5,374 windows. That is the right way to estimate "for a random entry, will the band hold?", but it means the breaches are clustered events, not independent coin flips. The worst-breach table is de-duplicated to one row per event.
• Price, not total return. We measure SPY price moves against a price-based band, which is what an option actually pays on. Dividends add a small downward drift the band ignores (about 0.15% over a month) - immaterial next to the band itself.
• One 21-year sample. 2005-2026 includes major crashes but is still one path of history. The normal-model benchmark (68.3%/95.5%) is the textbook approximation the calculators use, not a law.
• VIX is the S&P's gauge. The expected move on a single stock uses that stock's own implied volatility, which is noisier; this study is the index case, where the data is cleanest.
• Educational, not advice. Past behavior is not a promise about the next crash.

Source: daily SPY closes and CBOE VIX closes, 2005-01-03 to 2026-06-12 (5,395 days). For each trading day and each horizon we draw the 1-SD band as VIX/100 × √(days/365) - the same formula the expected move calculator uses - and check the realized move at the first close on or after the horizon (so a 30-day window can run up to ~33 calendar days). The implied-vs-realized check compares average VIX with annualized daily-return volatility over the full sample. Every reported study result regenerates from the data. The full horizon-by-horizon dataset behind this page is published as JSON and CSV at /data/, licensed CC BY 4.0 for anyone who wants to check the containment math themselves.