How to Read Implied Volatility

Every number in this series so far has been a record. Funding was paid, open interest was open, the taker crossed the spread, the book was resting. Implied volatility is different. It is a price for something that hasn't happened yet: how much the market thinks bitcoin will move, backed out of what people are paying for options today. That makes it the first number here you can mark against an answer.

Deribit, where most crypto options trade, rolls its bitcoin option quotes into one index the way the Cboe rolls S&P 500 options into the VIX. It is called DVOL, it has been published every day since 24 March 2021, and it is what "bitcoin's implied volatility" means in practice. This article takes every daily close it has ever printed, for bitcoin and for ether, and asks how the forecast did.

What the number actually measures

An option's price depends on how much the underlying is expected to move. Fix everything else — the strike, the time left, the interest rate — and one number is left over: the volatility the price implies. DVOL takes a whole strip of bitcoin options expiring around thirty days out and solves for that one number, quoted annualised. A reading of 50 means the market is pricing about 50% of movement over a year. Nobody trades a year of it. Divide by the square root of 365 for a day (2.6%) or multiply by the square root of thirty over 365 for a month (14%). Those are one-standard-deviation figures: the market's guess is that two days in three land inside the daily one, and one day in three lands outside it.

Two things it is not. It isn't a direction — an implied volatility of 50 says nothing about up or down, only how far. And it isn't a measurement. Realised volatility, the subject of the volatility article, is computed from prices that already printed. Implied volatility is what the market is willing to pay for the next thirty days of them. When the two are set side by side you get the one thing an options trader actually cares about: whether the insurance was priced right.

The level has halved. Bitcoin's median reading was 90 in 2021 and is 41 so far in 2026; the index has printed 53 days at or above 100, the last of them on 13 November 2022, and 288 days below 40, the first on 3 June 2023. Ether's runs higher — a median 70 against bitcoin's 56 over the whole sample — because ether moves more, and the market knows it.

Finding 1: the market overpays for insurance

On 72% of days, bitcoin's DVOL was higher than the volatility that followed. The median gap was 10 points, a ratio of 1.23×: for every 100 points of movement the market got, it had paid for 123. Ether's premium was smaller — 7 points, right 64% of the time. This is the volatility risk premium, and it is the same thing insurance companies live on: people pay more for protection than protection costs on average, because the average isn't what they are afraid of.

The premium has shrunk. Bitcoin's was 21 points in 2021 and is 5 this year. Ether's fell to 1.6 points in 2025, with the index right only 54% of the time — close to a coin flip. As the market has matured, the price of movement has come down to meet the movement. Whoever is selling bitcoin options is being paid less for it than they were.

And the premium is biggest exactly where you'd expect: when the index is high. In the top tenth of readings against the trailing year, DVOL exceeded what followed 94% of the time, by a median 15 points. In the bottom tenth it was right 59% of the time, by 5. Fear is when the insurance is dearest, and fear is when the insurance is most overpriced.

Finding 2: when it is wrong, it is wrong on a crash

Take the six thirty-day windows where realised volatility beat the forecast by the most. Four of the six were down months, and the biggest exception was a month bitcoin rose 32% — the misses are sudden moves in either direction, mostly down. None of them was preceded by a high reading: the highest of the six sat at the 59th percentile of its trailing year, and four of them were in the bottom fifth. The index was at 38 on 28 January 2026 — the 17th percentile of the year before it, an ordinary quiet day — and the month that followed realised 83 while bitcoin fell 26%.

Look at what the index did during the biggest one. On 28 January 2026 it closed at 38. A week later it was 54. Then, on 5 February 2026, bitcoin fell from $73,166 to $62,910 — −14% in a day — and DVOL jumped 29 points to 83, the biggest one-day rise since 2022. The next day bitcoin bounced +12% and the index gave back 24 of those points. The forecast caught up with the move after the move. That is not a criticism of the index; it is what a price does. But it means the reading you need is the one you won't have.

Finding 3: too wide in the middle, too narrow at the edges

Turn the index into a daily forecast — divide by the square root of 365 — and count how often the next day beat it. If the forecast were a well-calibrated bell curve, a day would exceed its one-sigma move 31.7% of the time and its three-sigma move 0.27% of the time. Bitcoin exceeded one sigma 19.5% of the time and three sigma 1.0% of the time. Ether: 22.0% and 1.4%.

Read those two numbers together. On the ordinary day, the index is too wide: the median bitcoin day used 0.45 of its implied move, where a bell curve says 0.67, and a day moved more than half a sigma 46% of the time against 62% expected. On the extraordinary day, it is too narrow: three-sigma days came 3.7× as often as the curve allows. The index buys too much protection against a Tuesday and too little against 5 February 2026. Two-sigma, the boundary between the two, is almost exactly right — 4.3% against 4.5%.

This is what "fat tails" means with a number on it, and it's the mechanism behind finding 1. The premium exists because a bell curve is the wrong shape for bitcoin. Sellers of options are paid, on most days, for a risk that lands rarely and lands hard; the 10-point median premium is the fee for the 1.0% of days that used more than three sigmas.

Finding 4: the fear index changed sides

The VIX is called the fear index because it rises when stocks fall. For a while, bitcoin's didn't. Rank each year's one-day changes in DVOL against the same day's return. In 2021–22 the correlation was −0.27 and −0.47 — the index rose on down days, VIX-style. In 2023 it flipped to +0.33: the index rose on UP days. It has since flipped back, to −0.43 this year.

The biggest jumps say it more plainly. Of the twenty largest one-day rises in bitcoin's DVOL during 2021–22, one came on an up day. During 2023–24, sixteen of twenty did — the biggest of them 14 January 2023, +13 points on a +5% day. During 2025–26, four of twenty. For two years the options market's fear was of missing the move up, and it paid up for calls on the days bitcoin ran. Now it pays up for puts on the days it falls, which is what an equity index does.

So "fear index" is a habit, not a property. The index measures the price of movement in either direction; which direction people are paying for depends on which one they are more afraid of, and that has changed sign twice in five years. If you read a rising DVOL as bearish, you would have been wrong for most of 2023 and 2024.

Finding 5: a high reading forecasts the reading, not the price

The last question is the one people actually want answered: does a high DVOL mean a big move is coming, and does a low one mean calm before a storm? Sort each day by its percentile against the trailing year and look at the next thirty days.

Three things in that table. First, a high reading does forecast a bigger month — the top decile realised a median 72 × 0.75 ≈ 54, the bottom decile 44 × 0.89 ≈ 39 — but by less than it says, every time. Second, it forecasts itself: from the top decile DVOL fell a mean 14 points over the next month; from the bottom decile it rose 3. Fear fades and calm ends, and the index is a bet on both. Third, it forecasts nothing about price. The next month was up 49% of the time from the top decile and 50% from the bottom; the middle of the range was 55%. There is no bin you would trade direction from.

How to read it yourself

• Turn it into a move. Divide DVOL by 19 (the square root of 365) for a one-day sigma and multiply by 0.29 (the square root of 30 ÷ 365) for a monthly one. At today's 39 that is 2.0% a day and 11% a month, either way.
• Read it against its own past year, not against a number you remember. 156 was a normal reading in 2021; today 39 is the 27th percentile of the trailing year.
• Expect it to be high. The index has exceeded what followed on 72% of days, by a median 10 points. If you are buying options, you are usually paying that; if you are selling them, you are usually collecting it — until finding 2.
• Don't wait for it to warn you. The biggest miss in the record started from a reading of 38, the 17th percentile of the year before it. The index catches up with a crash on the day of the crash.
• Treat the daily sigma as a floor for the ordinary day and a ceiling the tail ignores. Half the days use less than 0.45 of it; 1.0% of days use more than three of it, 3.7× what a bell curve allows.
• Check which way it's spiking before you call it fear. In 2023–24, 16 of the twenty biggest jumps came on up days. A rising index means people are paying for movement — and in a bull market the movement they pay for is up.
• A high reading is a forecast of the reading. Expect DVOL to be lower in a month (mean −14 points from the top decile) and expect the month to be calmer than it priced. Expect nothing about direction.

Method, and what this cannot tell you

Each day's DVOL is Deribit's daily close for the index, from the first day it was published. Realised volatility is the annualised standard deviation (√365) of the next thirty daily log returns of Binance spot, so every reading is scored against the exact window it was pricing; the premium is DVOL minus that figure, in points, and the ratio is DVOL over it. The one-day test divides each day's absolute return by the previous day's DVOL ÷ √365 and counts how often the ratio exceeded 0.5, 1, 1.5, 2, 3 and 4, against the shares a normal distribution would give. The sides test ranks each year's one-day changes in DVOL against the same day's return, and takes the twenty biggest jumps in each two-year span. A day's percentile is its DVOL against the trailing 365 days, and the five percentile bands are scored on the next thirty days' realised volatility, return, largest excursion, and the change in DVOL itself. Years are calendar years; 2021 starts on 24 March and 2026 ends on 14 September.

• One index, one venue. DVOL is built from Deribit's option quotes, which is most of the crypto options market but not all of it, and it is a 30-day figure. Shorter and longer expiries carry their own implied volatility and can disagree with it.
• Realised volatility is measured on close-to-close daily returns. A day that fell 12% and closed flat counts as a quiet day. Intraday ranges would give a higher realised figure and a smaller premium; the choice here matches how a 30-day variance swap would be settled.
• The premium is not free money. Selling volatility collects it on ordinary months and gives it back on the six months in the misses table; a median of 10 points says nothing about the size of the worst month. This article measures whether the forecast was right, not whether a trade on it paid.
• Five and a half years, one bear market and one bull. The 2023–24 'greed index' stretch is two years of a five-year sample. Whether the index rises on rallies again is not something this data can say.
• The percentile bands overlap. Consecutive days sit in the same band and share most of their forward window, so the effective number of independent observations in the top decile is closer to a dozen episodes than 110 days.
• DVOL's first year is its wildest and its shortest. 2021 starts on 24 March, after the index launched, and includes the May 2021 crash; the yearly rows are not like-for-like in length.