How Noisy Is the 25-Delta Risk Reversal?

Front-week skew gets quoted as a level. Desk notes, dashboards, weekly wraps: a number stated to two decimal places, treated as something the market has made up its mind about. Look at what is underneath that number and the picture changes.

Between 13 January and 8 September 2026, Bitcoin's front-week 25-delta risk reversal (call implied volatility minus put implied volatility at the 25-delta point) moved an average of 2.07 volatility points from one ten-minute cycle to the next, across 33,777 consecutive steps. Its own average absolute level over the same window was 9.74. The 30-day risk reversal on the same coin moved 0.194 points per cycle. A ratio of 10.7 to 1. A single reading of front-week skew carries about a fifth of its own magnitude in movement that arrives and leaves before you finish writing the number down.

What gets quoted, and what sits underneath

In Major Repricing in Front-End Skew, published on Deribit Insights on 13 December 2023, Imran Lakha notes under "Big Repricing of Skew" that the weekly skew had shifted to a 3 to 4 volatility put premium. Good observation, correctly read at the time. But it is a number without an hour and without a window, and that is how front-end skew gets reported everywhere: one snapshot, one moment, stated as a level.

Cayø Largo samples the full Deribit option chain every ten minutes and keeps every cycle. That opens a question a daily snapshot cannot answer: over the life of one of these quotes, how far does the quantity itself actually travel?

Three things sit underneath that number. Each one kills the explanation you would reach for first.

The front week moves ten times faster

Figure 1: Coral bars are the front-week bucket, silver the 30-day. Bar length is how far the risk reversal moves from one cycle to the next. The gap between the pairs tells the story.

Coin	Front week, per 10 min	30-day, per 10 min	Ratio	Steps measured
ETH	2.303	0.323	7.1x	33,807
SOL	1.761	0.612	2.9x	33,596
XRP	1.554	0.467	3.3x	33,019
BTC	2.070	0.194	10.7x	33,777

Same direction on all four coins. Values in volatility points.

Your first instinct is probably the one everyone has: fewer options in a front-week smile means a wider interpolation, which means more noise. Worth killing before going further, because the data says the opposite.

Sort BTC's front-week cycles by how many options priced in each one:

Options in the cycle	Cycles	Mean options	Move per 10 min
Under 120	2,496	103	2.039
120 to 159	8,774	142	1.772
160 to 199	11,502	179	1.844
200 to 249	8,345	221	2.378
250 and over	2,660	270	3.090

The fullest books move the most, half again as much as the thinnest. Whatever drives this, it is not a sparse smile.

That distinction matters because it separates this finding from a related one. Our note on deep out-of-the-money implied volatility momentum found that strike-level implied volatility flicker in the wings is thin-strike microstructure. Two different quantities, two different mechanisms, and the tests point in opposite directions.

None of it accumulates

Figure 2: Coral marks how far the reading actually travels in 24 hours. Silver marks how far it would go if each step were independent. The connecting bar is the distance it fails to cover.

Here is where the number gets interesting. If each ten-minute move were independent of the last, 144 of them would carry it twelve times as far as a single step. BTC's front-week risk reversal moves 2.07 points per cycle. An accumulating process would put it 24.8 points from where it started a day earlier. The actual distance: 6.44. A factor of 3.1 where 12 was the prediction.

Over one hour the gap is already there: 3.24 observed against 5.07 predicted. It opens as the horizon lengthens, which is the signature of a series pulled back toward a level rather than wandering away from one. ETH reads 6.02 against 27.6, SOL 4.62 against 21.1, XRP 4.35 against 18.6.

Most of what front-week skew does between two glances is oscillation around a slower-moving level, and a single sample cannot tell you which one you are looking at.

The clock moves it as much as the day does

Figure 3: Each point averages about 1,410 cycles at that UTC hour, across 239 days. More negative means puts are bid over calls. This shape is what survives after everything else has averaged out.

Average 239 days of ten-minute cycles by hour of the UTC day and a pattern survives that does not average out. ETH's front-week risk reversal reads -9.19 at 04:00 and -5.77 at 15:00, a 3.42 point swing from nothing but the clock. BTC traces the same arc, shallower: -9.96 at 23:00 against -8.10 at 15:00, a 1.86 point swing.

Put that next to how much the series genuinely moves from one day to the next. ETH's front-week daily mean changes by 3.34 points on average, across 235 consecutive day pairs. The hour-of-day bias and a full day of real movement are the same size.

Not the option count, either. ETH prices 154 front-week options at 15:00 and 139 at 04:00, a 10% difference, against a 3.42 point swing in the reading.

How this was measured

Every figure comes from deribit_options_oria_cohort_iv_skew, the table behind the 25-delta skew endpoint, read at Cayø Largo's native ten-minute cadence. The field is risk_reversal_25d_pct, the 25-delta call implied volatility minus the 25-delta put implied volatility, taken from the DTE_7 and DTE_30 expiry buckets. The window runs 13 January to 8 September 2026, 239 days.

Only cycles with data_quality_tier = 'COMPLETE' are included, and only consecutive pairs separated by 8 to 12 minutes, so a gap in collection cannot masquerade as a jump in the market. That leaves between 33,019 and 33,807 usable steps per coin per bucket. The daily figures require 100 or more cycles in the day. The same surface is drawn live, without the history, on the volatility landscape skew page.

The 24-hour comparison uses the table's own risk_reversal_change_24h_pct field. For a process with independent increments the typical absolute move scales as the square root of the horizon, which is where the factor of twelve comes from.

What this does not say

The front-week risk reversal is not meaningless. Its daily mean is a usable quantity, and ETH's day-over-day movement of 3.34 points in that mean is real market information. A single reading of it, without context, is something else.

The mechanism is still open. Ruling out the thin book kills the obvious answer without supplying a better one. The most likely candidates are the extreme strike sensitivity of a 25-delta point as expiry approaches zero days, and the changing composition of the bucket as contracts roll through it. Neither has been tested here, and the honest position is that the effect is measured and unexplained.

None of this extends past crypto. In le Roux's model of the S&P 500 implied volatility surface, published in October 2007, almost all variation out to two years is captured by the volatility level and one or two further factors. Whether an equity front week behaves the way a crypto one does is a question this data cannot reach.

And it is one window. Nine months of one exchange's book, in a year that had its own character.

What a desk would watch next

Quote the window with the number. A front-week skew reading needs the hour it was taken and the period it was averaged over, the same way an implied volatility print needs its tenor. Two desks reading the same day eight hours apart are looking at books that differ by more than the day itself moved.

There is a version of this that people already do without naming it. When Amberdata's Samneet Chepal backtested the risk reversal on Deribit in Harvesting BTC's Risk-Reversal Premium, in the section of that name, the model entered weekly at a 30-day maturity and hedged on end-of-day exposure. The 30-day maturity was chosen, in the author's words, discretionarily, for liquidity. The numbers above give that choice a second reason the study did not need at the time: at 30 days the quantity being traded moves 0.194 points between observations. At the front week, 2.07.

The wider question of how much to trust an implied volatility number is alive enough that OptionMetrics devoted a July 2026 podcast to it with Brett Friedman, Accuracy of Implied Volatility, comparing implied against realised on the S&P 500. That work asks whether the number forecasts well. This one asks something earlier: whether the number holds still long enough to be read.

References

M. le Roux, A Long-Term Model of the Dynamics of the S&P500 Implied Volatility Surface, OptionMetrics research library, October 2007. Argues that almost all variation in the S&P 500 implied volatility surface, out to two years, is captured by the volatility index and one or two uncorrelated factors.

Samneet Chepal (Amberdata), Harvesting BTC's Risk-Reversal Premium, Deribit Insights, 24 January 2023. Backtests the risk reversal on Deribit Bitcoin options from April 2019 to December 2022, entering weekly at a 30-day maturity and delta-hedging on end-of-day exposure.

Imran Lakha, Major Repricing in Front-End Skew, Deribit Insights, 13 December 2023. Reports the weekly skew shifting to a 3 to 4 volatility put premium as short-dated downside protection was bid, in the section headed "Big Repricing of Skew".

Brett Friedman, Accuracy of Implied Volatility, OptionMetrics, 1 July 2026. Examines S&P 500 and volatility index history to ask how well implied volatility forecasts what the market goes on to do.