When implied volatility hits its floor, does it come back up?

When implied volatility sits in the bottom tenth of its own range on every horizon at once — one month, three months, six months and twelve — it rises within the next ten trading sessions in 74% of cases, against 57% from an ordinary starting point.

There is a piece of folklore in options trading that nobody bothers to check: buy volatility when it is cheap, because it always comes back.

It sounds obvious. Implied volatility cannot go below zero, and it rarely sits near its lows for long. So a position that profits from rising vol, opened when vol is at rock bottom, ought to be a reasonable bet.

I wanted to know whether that is true, and by how much. Not because it makes a strategy — it does not, at least not on its own, and I will come to why — but because every long-volatility idea rests on it. If implied vol at the bottom of its own range does not actually tend to rise, then a whole family of trades is built on sand and the search can stop there.

So I measured it.

What was measured

For every ticker in my options archive — 122 of them, US equities and ETFs, with end-of-day option chains going back to 2016 — I reconstructed the at-the-money implied volatility of the roughly 30-day expiration, one reading per session. About a million of them.

Then I asked, for each session, where that reading sat inside the stock's own recent range. Not against some absolute level: 20% implied vol is high for a utility and low for a biotech, so the only comparison that means anything is against the ticker's own history.

A ticker is "at its floor" when implied vol sits in the bottom tenth of its own trailing range on four horizons at once — one month, three months, six months and twelve months. All four, simultaneously. That is a demanding condition: it fires about three times a year on a given name, 6,600 times in total across the whole universe.

Then: where is implied vol 5, 10, 20 and 40 sessions later?

The trap that would have made this meaningless

Here is the thing that would have wrecked the answer if I had not controlled for it, and that most versions of this study skip.

Implied volatility rises before an earnings announcement. Always. It has nothing to do with mean reversion — it is the market pricing a known event on a known date. A stock sitting at its vol floor three weeks before it reports will show a handsome rise, and that rise is a calendar effect that anyone can see coming.

Count those episodes and the whole study turns into an expensive way of rediscovering that companies publish results four times a year.

So every episode is classified. Was there an announcement inside the window, or not? And separately, there are 34 ETFs in the universe that have no earnings calendar at all — no announcements, ever. They are the control group, and they cost nothing to obtain.

Only the episodes with no announcement in the window answer the question as asked.

The answer

Ten sessions after a vol floor, with no earnings anywhere in the window:

Implied volatility rose in nearly three cases out of four, against a bit better than half the time from an ordinary starting point. The typical rise was three times larger.

The control group agrees. The 34 ETFs with no earnings calendar — where the question of announcements cannot arise at all — came in at 75.4%, a shade higher than the main figure.

And it holds at every horizon tested:

Both columns exclude windows containing an announcement, so the comparison is like-for-like at every horizon. Note that the baseline climbs too — over 40 sessions implied vol is higher 79% of the time from any starting point, which is worth knowing before anyone quotes the 94% on its own.

One caveat on those 40-session figures I would rather give than have pointed out. Requiring no announcement within 40 sessions, on a company that reports quarterly, selects the weeks immediately after a report — when implied vol has just been crushed and has nowhere to go but up. Both columns are inflated by that. The 5 and 10-session rows are the clean ones, and they are what the headline figure uses.

So: yes. The folklore is right. Implied volatility at the floor of its own range does come back up, considerably more often than not, and the effect is large enough to see without squinting.

The part that makes sense

One result is worth more than the headline, because it has an obvious mechanism behind it.

The lower the starting point, the stronger the bounce. Twenty sessions out:

Perfectly monotonic, and it is not mysterious. Volatility has a hard floor at zero and no ceiling. A ticker whose implied vol has fallen to 15% has very little room left to fall and a great deal to rise. One at 50% has room to do either.

A result that behaves the way the mechanism says it should is a result you can lean on a little harder than one that merely appears in the data.

Now the part you will not like

Two findings cut against the obvious next move, and they matter more than the headline.

It is not worth hunting for the best tickers

The natural instinct is to ask which names respond best, trade those, and skip the rest. I tried. I split the nine years in two, ranked every ticker on the older half, and checked what those same tickers did on the newer half that the ranking had never seen.

The ranking barely survives. The correlation between the two halves is 0.29 — positive, so there is something, but weak. In practice: TSM led the first half at +17.4% and returned +4.3% in the second. Apple went from +14.9% to +2.5%.

The fifteen best names of the first half delivered a median of +8.8% in the second. The whole universe delivered +7.5%.

All that selection bought was one percentage point.

Because the effect is nearly everywhere

Which is the other half of the same finding, and the more encouraging one: 108 of 112 tickers were positive on both halves independently. Ninety-six per cent. If this were noise you would expect roughly a quarter.

So the effect is real and it is close to universal. That is exactly why picking names does not pay — there is no scarce edge to find, because almost every ticker has it.

And my own hypothesis was wrong

Before running the split I wrote down a prediction: broad ETFs should respond better than individual stocks. An index's volatility is driven by the market as a whole and has a level to return to, whereas a single company's vol floor can be broken by news belonging to that company alone. It seemed sound.

It is not there. ETFs 74.1% against stocks 73.1% on the first half; 76.8% against 74.1% on the second. One point, and not consistently in the same direction.

I am including this because a study that only reports the hypotheses that worked is not a study. This one was mine, it was stated in advance, and the data declined it.

What this does not tell you

Here is where I have to be blunt, because the gap between this result and a tradable strategy is wide and I would rather name it than let anyone fall into it.

A 7.9% rise in implied vol does not put 7.9% in your account.

Between the two sit theta, the structure you choose, and the bid/ask you cross to get in and out. Any one of them can eat the whole move.

I know, because I tested a structure designed precisely to capture this: a ratio double calendar, long vega by construction, opened on exactly these vol floors. Across 1,167 cycles over ten years, priced at the ask going in and the bid coming out, it lost 33% of the debit on average and had a losing year in every single year. Twenty-eight per cent of cycles reached a +10% target — the vol thesis worked often enough — but four strikes across two expirations means crossing eight spreads on a round trip, and that costs far more than an ordinary vol move pays.

The premise was right. The vehicle was wrong.

Which leaves the interesting question, and the one I am working on now: what is the cheapest structure that captures a rise in implied volatility? Two legs instead of four? A single expiration? No options at all?

I do not have that answer yet. When I do, it will be measured the same way as this one — split in two, checked on the half that was never looked at, and published whichever way it comes out.

For anyone who wants to check the working

Every figure here comes from end-of-day option chains, 2016 to 2026, 122 tickers. Implied vol is reconstructed at the money on the nearest-to-30-day expiration; a session's reading is the mean of the four contracts closest to the money.

Two details that changed the numbers materially, in case you run something similar.

Data-provider floor values. A contract with no usable quote carries an implied volatility of 0.015 in this feed. That is not a reading, it is a placeholder. Left in, it gets picked up on thin chains, the session is classified at its vol floor by definition, and a normal reading ten sessions later looks like an 800% rise. It put one ETF at a median of +847% before I caught it. Contracts below 3% implied vol, or with no bid or ask, are excluded.

Means are useless here. Before that fix, one group showed a mean of +113% against a median of +9%. Even after it, a handful of vol explosions pull every average upward. Every headline figure in this article is a median.

Consecutive sessions at the floor count as one episode, not fifteen — a condition that holds for three weeks is one opportunity.

Backtested and observational figures on historical data. Past behaviour does not predict future results. Nothing here is investment advice.