Understanding volatility measures

Risk statistics are quoted routinely and understood rarely, and each has a specific meaning and a specific failure mode.

Standard deviation

The most commonly quoted.

Measures the dispersion of returns around their average, generally annualised.

Its problems: it treats upside and downside movement identically, when only one is unwelcome; it assumes a distribution that understates the frequency of extreme events; and it says nothing about the shape of the losses.

Two portfolios with identical standard deviation can have very different experiences if one produces occasional large losses and the other produces steady small ones.

Beta

A relative measure.

Describes how much an asset has moved relative to a benchmark: a beta of one means it moved in line, above one means it amplified moves, below one means it dampened them.

Its problems: it measures only the relationship with the chosen benchmark, so an inappropriate benchmark produces a meaningless figure; it is estimated over a period and changes; and it captures only linear relationships.

A low-beta holding is not necessarily low risk — it may simply be exposed to different risks.

Maximum drawdown

Arguably the most useful for individuals.

The largest peak-to-trough decline over a period.

It describes what actually happened rather than a statistical abstraction, which makes it considerably easier to relate to.

Its limitation: it depends entirely on the period examined, and a fund launched after the last crisis has an unimpressive drawdown record simply because it has not been tested.

Which means checking the period covered is essential.

Time to recovery is a useful companion figure and is rarely quoted.

Value at risk

Used institutionally.

Estimates the loss that will not be exceeded with a given probability over a given period.

Its problems are well known: it says nothing about how bad losses are beyond the threshold; it depends on the distributional assumptions; and it has performed poorly during exactly the extreme events it was designed to address.

Expected shortfall, which measures the average loss beyond the threshold, addresses part of this and is increasingly preferred.

Sharpe ratio and similar

Risk-adjusted return measures.

Sharpe divides excess return over a risk-free rate by standard deviation, producing a return per unit of volatility.

Its problems: it inherits the limitations of standard deviation; it is sensitive to the period; and it can be improved by strategies that produce steady returns punctuated by rare large losses, which is a known way of gaming it.

Sortino uses only downside deviation, which addresses one limitation.

Both are more useful for comparison than as absolute measures.

The general problems

Common to all of them.

They are backward-looking, and the period chosen determines the answer.

They assume the future resembles the past.

They are affected by valuation practices, which is why illiquid assets appear less volatile than they are.

They measure volatility rather than the risk of not meeting your objective.

And they are frequently presented over periods selected to flatter.

What is more useful for an individual

Practical alternatives.

The largest historical decline for a proposed allocation, expressed in the actual amount of money you would see disappear.

How long previous recoveries took.

Whether you could continue your plans if the portfolio fell by that amount.

Whether you would sell.

And what the shortfall risk is — the probability of not reaching the objective — which is the risk that actually matters and which volatility statistics do not measure.

Expressing risk in money

The single most useful change.

People respond quite differently to a percentage and to a currency amount.

A thirty per cent decline sounds tolerable; the equivalent figure in money frequently does not.

Which means converting any risk discussion into the actual amount is the most honest way to assess whether an allocation is appropriate.

And doing it before investing rather than after a decline is the point.

Where the statistics are genuinely useful

For balance.

Comparing similar products on a consistent basis.

Identifying when a fund's behaviour has changed.

Checking that a fund is doing what it claims — a supposedly low-volatility fund with high volatility is not.

And institutional risk management, where the limitations are understood and multiple measures are used together.

The problem is not the statistics but the use of a single one as a summary of something complex.

The practical approach

For a retail investor.

Look at the maximum drawdown over the longest available period, and check whether that period included a crisis.

Convert it into money.

Ask whether you would have held on.

Ignore composite risk scores and single-number ratings.

And treat any product marketed on the basis of favourable risk statistics with the scepticism that backtested and period-selected figures deserve.

General information only, not investment advice. Investments can fall in value and past performance does not indicate future returns. Consult a regulated financial adviser.