The volatility that is quoted, and the risk that is avoided

CoinShares' quarterly fund manager survey, published on 19 August 2026, records two things in the same document. Digital asset allocations rose to 1.2% of portfolios, the first increase since the sell-off that began in October 2025, and the move was driven entirely by institutional investors. At the same time, volatility rose to the top of the list of reasons preventing investors from buying digital assets at all.

The survey then notes what makes those two findings sit oddly together. Thirty-day volatility was, at the time of writing, close to its record low of 22%. The number most often given as the reason for staying out was at its least alarming reading on record.

CoinShares draws the conclusion itself, and it is worth quoting rather than paraphrasing: investors are pricing drawdown risk, not realised volatility. That single sentence is the subject of this note.

One caveat belongs here rather than in a footnote. The survey drew 30 responses from investors covering roughly US$1.16 trillion of assets under management. That is a small sample of large allocators. It is enough to make an observation worth examining and not enough to establish a market-wide fact, and it should be read that way.

Realised volatility is the standard deviation of returns over a window, annualised. It describes how widely returns have been dispersed around their own average during that window. It is a single number, it is cheap to compute, it is comparable across assets, and those three properties are why it appears in almost every risk report ever written.

What it does not describe is the shape of the path. Standard deviation is indifferent to order. A series of returns rearranged into a different sequence produces exactly the same volatility figure and an entirely different experience for whoever held it. Volatility does not know whether the down days arrived scattered through a year or consecutively in a fortnight.

It is also indifferent to depth. A market that oscillates briskly around a flat level can print a higher volatility number than one that declines steadily and does not recover. The first is uncomfortable; the second is what ends allocations.

And it is indifferent to when. An investor who allocated at a peak and one who allocated a quarter earlier hold the same asset with the same volatility and face different questions at their next investment committee.

Maximum drawdown measures the largest peak-to-trough decline over a period. It is a worse statistic than volatility in most respects: it is a single historical realisation rather than a distribution, it depends heavily on the window chosen, and it cannot be annualised or combined across assets in any clean way.

It has one advantage that outweighs those defects for an allocator. It answers the question a committee actually asks, which is not how dispersed the returns were but how bad it got and for how long.

That is the question behind a redemption request, behind a margin call, behind the decision to cut a position at the worst possible moment, and behind whether an allocation survives long enough to be judged on its merits. None of those events is triggered by a standard deviation.

So when a survey reports that volatility is the stated barrier while volatility is at a record low, the most economical explanation is not that respondents are confused. It is that they are using the word volatility to name a fear that the volatility statistic does not measure.

There is a further reason a low realised-volatility reading deserves care, and it is a matter of what the statistic can and cannot see rather than a prediction about what comes next.

Realised volatility is computed from observations that have already occurred. A thirty-day window contains thirty days. If the distribution of returns has tails — and in this asset class the historical record is not ambiguous on that point — then a quiet window is a window in which the tail did not appear. It is evidence about the centre of the distribution and close to silent about the edges.

This note makes no claim about what follows a low reading. The honest statement is narrower and more useful: a low realised-volatility number is a description of a period that has ended. It carries no information about the tails that were not sampled during it, and an allocator who reads it as reassurance about those tails has read more into the number than the number contains.

That is the same discipline this firm applies to a track record. A series describes what happened over the period it covers. What it does not cover is not evidence of anything.

If the risk being avoided is drawdown rather than dispersion, then several ordinary parts of an allocation decision are being sized against the wrong number.

Position sizing derived from a volatility target implicitly assumes the volatility observed is representative. Where the concern is a fall of a given depth, the constraint that binds is the depth the portfolio can absorb before something else breaks — a covenant, a liquidity requirement, a mandate limit, or a committee's tolerance.

Leverage compounds the difference. A leveraged position is not sensitive to the average size of daily moves; it is sensitive to the largest adverse move before a margin requirement is met. A strategy can run at a modest volatility for a long period and still be one path away from a forced reduction.

Liquidity is where the two measures diverge most sharply. Dispersion is measured on prices that were struck. Whether those prices could have absorbed the position being held is a different question, and it is the one that matters when a fall is under way and everyone else is asking it at the same time.

None of this argues against measuring volatility. It argues against letting one number stand in for a risk it does not describe.

In an operational review the more informative sequence is straightforward. Ask what the worst peak-to-trough decline was, over what window, and how long recovery took. Then ask what ended it — whether the strategy behaved as designed and the market moved, or whether something in the implementation gave way first.

That second question is the one that separates a strategy which experienced a bad market from one which discovered a flaw. Both produce a drawdown. Only one of them tells you something about the manager.

And ask what the position size would have been under the same rules at the start of the fall rather than today. A risk framework that only ever sizes against recent calm is a framework that has not been tested by the thing it is meant to protect against.