Rolling Volatility

Bitcoin Rolling Volatility plots three series on a shared linear axis capped at 250%. Each series is the annualised standard deviation of daily log returns over its window: 30 days (the tightest, most reactive line), 90 days (the quarterly lens), and 365 days (the yearly lens, the smoothest). The √365 annualisation factor converts a daily standard deviation to a yearly one under the standard independent-returns assumption. A muted reference line at 60% marks the S&P 500's crisis-peak realised volatility — where traditional equity vol stood in October 2008 and again in March 2020.

Today: 30-day 42.4%, 90-day 39.4%, 365-day 43.3% — regime Normal. The lifetime average 30-day reading across 5,780 defined days is 73.1% — that lifetime mean is the gravity the short window keeps falling back toward, and it sits far above where the long window now lives, which is itself the compression story in two numbers.

Three steps, applied to every rolling window:

rt = ln(pt / pt−1)
voln(t) = stdev(rt−n+1..t) × √365

Step one: take the daily log return as the natural log of today's close divided by yesterday's. Step two: compute the sample standard deviation across the trailing n-day window. Step three: multiply by the square root of 365 to annualise. The 30-day series becomes defined on the thirtieth daily close; the 90-day on the ninetieth; the 365-day on the first anniversary. Pre-warm-up rows are left blank on each series.

Calendar days, not trading days. Bitcoin trades every day — weekends, holidays, mid-year — so the trading-day vs calendar-day distinction that matters for equities collapses here. The √365 annualisation factor reflects that. A reader cross- referencing equity-style realised vol (which uses √252) should multiply our numbers by √(252/365) ≈ 0.83 for an apples-to-apples comparison. The full derivation, including the epoch boundaries and the 250% chart cap, is on the methodology page.

The three windows tell three stories, and the gaps between them are the tell. The 30-day line catches event regimes — cluster events around capitulation, post-event recoveries, flash crashes — and is the only one fast enough to time anything. The 90-day line filters most of that noise and reads regime transitions. The 365-day line is the maturation lens: it moves slowly, has compressed every halving epoch, and is the headline statistic this page contributes. When 30-day runs well above 365-day, the tape is in a transition window; when all three converge low, the market is coiled and quiet. The regime bands below bucket the 30-day window, because that is where the immediate signal — and most of the false signals — live.

Cycle anchors against all three windows surface two patterns. First: cycle tops and cycle troughs both produce elevated short-window vol, but tops produce it on the way up (parabolic acceleration) and troughs produce it on the way down (capitulation) — so a high 30-day reading is regime-agnostic by itself. Second: every cycle peak's 30-day vol has been lower than the previous cycle's, with one structural exception — the March 2020 Covid liquidity flush, whose 30-day spike, measured at its peak weeks after the crash, ran above the orderly 2017 blow-off top. (Note the Covid row's anchor date catches the onset of the flush, not the peak, so its tabled 30-day reading is far lower than the spike it eventually reached.) The 365-day column, immune to single-day events, compresses monotonically down the table.

Anecdote becomes evidence in the per-epoch table. Every halving period is a row, and each cell holds the mean and standard deviation of one window computed against the days inside that epoch. Read down any of the three window columns and the mean falls at every step — five epochs, no reversals. This is the page's significance test in plain sight: a decline that holds across all three windows, across thousands of daily observations per epoch, is not the kind of thing that survives by chance.

Three things are true at once. First, the 365-day mean has fallen every epoch from 93.1% in Era II to 47.3% in Era V. Deeper market depth, the launch of regulated derivatives in late 2017, and the 2024 spot-ETF channel all dampen what a given dollar of flow does to price — a mechanical story, not a coincidence. Second, the 30-day mean has fallen by less than the 365-day mean: cluster events still produce big short-window readings even when the long-window regime is calm, which is exactly why the short window is poor for timing. Third, the epoch-internal stdev — the second number in each cell — is itself shrinking, from 30.0% to 3.6% on the 365-day window. Each epoch's regime is not only calmer but more uniform than the one before.

Where the honesty has to land: the long-run compression is signal — it is monotonic across five epochs, replicated across three independent windows, and grounded in a real change in market structure, so it is not a sampling artefact. But the table says nothing about when the next regime shift comes or how far vol can still fall, and it offers no tradeable edge at the daily horizon. A maturing-asset trend is not a timing model.

The traditional-finance reference for scale: the CBOE VIX — implied vol on the S&P 500, which historically tracks realised vol with a small premium — averages near 19–20 over its 35-year history and peaked at 89.5 (October 2008) and 82.7 (March 2020). Bitcoin's recent epoch-mean 365-day realised vol of 47.3% is roughly 2−3× the long-run S&P benchmark. A decade ago the multiple was 5−10×. The compression is real, structural, and ongoing — but Bitcoin remains materially more volatile than any major equity index.

The implied-vol counterpart is Deribit's DVOL index, which applies the VIX methodology to BTC options and tracks the 30-day forward-looking implied vol. DVOL has tracked our 30-day realised vol with a typical premium in the 5−15-percentage-point range. Persistent gaps between implied and realised — either direction — have been useful regime indicators around cycle inflections; that analysis lives outside this chart's scope but is referenced on the methodology page.

The Covid print that broke the descending-peaks pattern. Read the cycle-anchor table and one row violates the otherwise-clean rule that each cycle's peak vol is lower than the last: 12 March 2020. The Covid liquidity flush sent the 30-day window to an event-driven spike that, at its peak weeks later, ran modestly above the orderly December 2017 blow-off top, even though 2017 was a far larger price cycle. The lesson is structural — a single cross-asset margin cascade can override the maturation trend in the short window for weeks. A reader who treated "peaks always descend" as a law would have been wrong precisely when it mattered.

Compressed does not mean safe — it often means coiled. The 30-day window read Compressed (≤ 30%) through stretches of mid-2023 and parts of 2024, and those quiet windows resolved upward. But low realised vol is direction-blind: the calm before the November 2022 FTX collapse and the lull ahead of the June 2022 deleveraging cascade both preceded violent downside. Treating a Compressed print as a green light is the most common misuse of this chart.

Daily closes hide the intraday experience. A flash crash that printed and recovered inside the same UTC day — the kind of wick-and-fill candle common around liquidation events — registers as a near-zero log return here. The chart understates the lived experience of holding Bitcoin through a stress hour. Five-minute or hourly candles on the same definition would print materially higher numbers, especially across weekend gaps when traditional-market hedging is absent.

The √365 independence assumption breaks during clusters. Annualising by √365 assumes daily returns are independent. They mostly are, except during cluster events: a multi-day capitulation (the FTX week, the March 2020 flush) produces autocorrelated negative returns, and a multi-day melt-up produces autocorrelated positive ones. In those windows annualised realised vol under-states the true multi-day dispersion; in fully mean-reverting whipsaw weeks it over-states it.

Survivorship and the limits of the trend. This series is Bitcoin alone — the survivor of a much wider universe of assets, many de-listed or dead, that would lift any basket's historical vol. And the compression itself is a description of the past, not a guarantee about the next epoch: vol could fall further with deeper institutional flow, or reset higher on a regulatory shock or a credit cycle that propagates into crypto. The decline is real and structural; it is not a promise.

If you accumulate on a schedule, the compression works in your favour and asks nothing of you. Falling vol means a steady buy program is exposed to smaller drawdowns relative to position size, on average, than the same program faced a decade ago. The one caveat: size against the cluster-event maximum the 30-day window still reaches, not the lifetime mean. Read peak drawdowns alongside this distribution on the drawdown page.

If you trade the cycle, use the gap between windows, never a single level. A persistent run of 30-day above 365-day flags a regime-transition window — early bull or early bear depending on direction, which this chart cannot give you. Pair it with rolling CAGR for the sign of the move and the halving-cycle overlay for where in the cycle you stand. And cross-check realised against Deribit's DVOL: when implied runs persistently rich to realised, the options market is paying up for protection the spot tape has not yet delivered.