Buying long-dated calls when volatility is cheap: what 38,570 entries show

The short answer

The idea is half right, and the wrong half is the half that pays. Volatility really does mean-revert: when a stock's own volatility sat in the bottom tenth of its three-year range, volatility rose over the following year 88% of the time. But the buyer of a one-year call was not paid for it. That same bottom decile produced the lowest win rate and the worst median result of any bucket we measured.

Over twelve months, what a call is worth depends far more on where the stock went than on what its volatility did. Quiet stocks stayed quiet. Stocks in the calmest decile returned 6.7% over the following year, against 27.4% for the noisiest. The volatility signal worked as a filter that selected against the stocks that went up.

This is a study of market history. It is not advice, and it is not a strategy we sell.

What we tested

We took 419 US stocks with at least fifteen years of price history and sampled an entry roughly every month for each one. At every entry we recorded the stock's trailing 63-day realized volatility as a percentile of its own prior three years, priced a one-year at-the-money call with Black-Scholes, and then settled that call against the price the stock actually reached a year later. That produced 38,570 entries.

Two groups ran through the identical pipeline: 164 stocks our own screens rated favourably at the time of the study, and 300 others drawn from the rest of our tracked universe as a control. The control exists because the first group is made of survivors, every one of which is still listed and still well rated, and a result that appears only there is a property of the selection rather than of the market.

Nothing in the test looks forward. The entry percentile reads only the trailing window, and the outcome reads only what happened after.

Volatility does mean-revert. That part holds up.

Group the entries by how cheap the stock's own volatility was at the time, and the next year of volatility behaves the way mean reversion predicts.

The pattern runs cleanly across all ten buckets and the control group reproduces it almost exactly, from 1.32 at the cheap end to 0.76 at the dear end. If the question were only whether volatility mean-reverts, the answer would be a confident yes.

The call buyer was not paid for it

Score the same buckets as a one-year at-the-money call and the ranking turns over.

The signal picked the worst bucket available. In the control group the same two ends read 36% and a 81% median loss at the cheap end, against 52% and a small median gain at the dear end, so this is not an artifact of which stocks we happen to rate well.

Split by calendar year, entries in the cheapest third of a stock's own volatility range lost money in nine of the twelve years with data, and the median position lost its entire premium in 2015, 2019 and 2021.

Flipping the rule does not fix it

The obvious response is to do the opposite and buy after volatility spikes. The year-by-year record says that is a bet on the regime rather than an edge.

The gains sit in a crash rebound and a momentum run. The losses sit in the two years when high volatility marked a top instead of a bottom. High volatility means either a crash or a melt-up, those two point opposite ways for a call buyer, and a percentile cannot tell them apart.

Cheap options were not better options

We also sorted by what the call cost, as a share of the share price, on the theory that cheaper contracts leave more room to be right. They did not. The share of calls that paid stayed between 41% and 46% across every cost bucket, from contracts costing under a tenth of the share price to contracts costing more than a third of it.

That result is less surprising than it looks. The price of an option already reflects the volatility you are buying it on, so a cheap contract is usually cheap for a reason that shows up later in the stock.

The market for these contracts is smaller than it looks

Separately from the historical test, we measured what is actually quotable today. Of 164 stocks in the study's primary group, only 66 listed any option expiring a year or more out. Among those that did, half had fewer than 39 contracts of open interest at the money, and a quarter had seven or fewer. The typical gap between the bid and the ask was about 10%, and the widest ran past 100%.

Applying nothing but a cost ceiling and a basic liquidity floor to those 66 names left three. Whatever a screener of this kind picked, it would be choosing from a very short list and paying a wide spread to get in and out.

What this study cannot tell you

Every number above carries limits, and they are worth more to a reader than another table.

• No delisted companies: both groups are drawn from stocks still tracked today, so companies that went to zero never appear. Every forward return here is flattered by that. Comparisons between buckets survive it; absolute levels do not.
• We had to estimate what the options cost: no long-dated implied volatility history exists to buy or download, so each contract was priced off the stock's own realized volatility. We checked that estimate against live quotes and found it close, at a median ratio of 1.14 against the 1.15 assumed, with the result turning negative only past about 1.55. It is still an estimate, and it is the one number that could move these conclusions.
• Overlapping windows: entries a month apart share most of the following year, so the effective sample is far smaller than 38,570. That is why this study reports medians and year-by-year splits rather than significance tests. No p-value here would be honest.
• One long bull market, and what a longer sample shows: 2011 to 2026 contains two brief crashes and a strong upward drift, which flatters any strategy that buys calls. We later ran the identical method over sixty years of history, 95,954 entries, to see which parts held. The ranking held: the cheapest decile was still the worse bucket on every measure we checked. The size of the gap did not. The spread in the share of calls that paid falls from eighteen points to under three, the dearest decile's median result is a loss rather than the 36% gain shown above, and the distance between the two next-year stock returns drops from about 21 points to about 7. That longer sample is also a harder survivor filter than the main study, because only a small set of companies has sixty years of history and all of them are still tracked today, which is the direction that would compress a gap between buckets. Treat the figures in this study as belonging to their window.
• The simplest possible contract: at the money, one expiry, held to the end, no adjustment along the way. A real position would be managed, and management changes outcomes in both directions.
• Dividends cut the same way: the price series adds dividends back while a real call does not, which slightly flatters every call in the test. Calm stocks tend to pay more of them, so this bias favours the bucket the study rejects.

What we did with the answer

We shelved the feature. The study cost an afternoon; building the screener first and learning the same thing from live results would have cost a year, because a one-year contract takes a year to tell you anything.

One piece of it we kept. Because no long-dated implied volatility history exists to test against, we now record one reading a week for every stock in the group, so that in twelve months the next version of this study can measure the number this one had to estimate.

We publish the negative result for the same reason we publish the misses in our other studies. A research process that only reports the ideas that worked is not showing you its work. You can see what we do publish, scored against what happened, on the public track record.