WhitmanTrading

Sharpe Ratio: It Penalises Good Surprises

The Sharpe ratio divides a return above what cash pays by the standard deviation of those returns, giving a reward-per-unit-of-variability figure. Because it uses total variability, it penalises upside movement exactly as much as downside, and it assumes a distribution returns do not follow.

How it works

A candlestick chart of the site's shared price history. The headline on the chart reads: Return above what cash pays, per unit of deviation.
Return above what cash pays, per unit of deviation. Illustrative chart - not real market data.

The Sharpe ratio is a fraction. Return minus the rate paid on cash on top, the standard deviation of those returns underneath.

A gently rising stretch of the long price series with an account equity curve beneath it. The headline on the chart reads: Excess return on top, standard deviation underneath.
Excess return on top, standard deviation underneath. Illustrative chart - not real market data.

Subtracting what cash pays is the part people skip. A 6% return when cash pays 5% is not the same achievement as 6% when cash pays nothing, and the ratio is built to make that difference visible.

A calmly advancing stretch of the long price series with a slowly rising equity curve beneath it. The headline on the chart reads: It exists so two returns can be compared fairly.
It exists so two returns can be compared fairly. Illustrative chart - not real market data.

It exists to make returns comparable. A method returning 30% with wild swings and one returning 12% steadily cannot be ranked by return alone, and the ratio is one attempt at a common scale.

Two structural problems

A flat, quiet stretch of the long price series with an account curve breaching its limit. The headline on the chart reads: And it penalises upside movement exactly as much as downside.
And it penalises upside movement exactly as much as downside. Illustrative chart - not real market data.

Standard deviation counts every deviation, in both directions. A method that occasionally returns far more than usual is penalised for it, identically to one that occasionally loses far more.

A strongly rising stretch of the long price series with a gradually rising equity curve beneath it. The headline on the chart reads: This series put 8 moves beyond three deviations.
This series put 8 moves beyond three deviations. Illustrative chart - not real market data.

The second problem is the distribution. On this site’s shared 576-bar history, 8 of the 575 bar-to-bar moves exceeded three standard deviations.

A choppy, directionless stretch of the long price series. The headline on the chart reads: Where the bell curve allows fewer than two.
Where the bell curve allows fewer than two. Illustrative chart - not real market data.

A normal distribution across that many observations implies fewer than two. The measured shares within one, two and three deviations were 74.1%, 93.2% and 98.6% against the bell curve’s 68.3%, 95.4% and 99.7%. A ratio built on a distribution that understates its own tails understates the risk it is measuring, which is the substantive criticism of it.

A declining stretch of the long price series. The headline on the chart reads: A smooth curve can be manufactured, and has been.
A smooth curve can be manufactured, and has been. Illustrative chart - not real market data.

And a smooth curve can be produced deliberately. Strategies that make small amounts consistently and lose large amounts rarely report excellent ratios right up until the rare event, because the denominator only sees what has already happened.

In practice

A 72-bar candlestick section of the shared price history with an account curve shown with and without fees. The headline on the chart reads: Fees lower it through the numerator, quietly.
Fees lower it through the numerator, quietly. Illustrative chart - not real market data.

Costs come off the top line. They reduce the return without necessarily changing the variability, so a high-turnover method’s ratio falls further than its equity curve suggests.

A candlestick chart with a volume histogram beneath it, with the volume histogram emphasised. The headline on the chart reads: It is an equity measure and ignores the market.
It is an equity measure and ignores the market. Illustrative chart - not real market data.

The input is an equity curve. Price, volume and market conditions are all irrelevant to the calculation, which is both its generality and its blind spot.

A long-horizon candlestick view of the same price series. The headline on the chart reads: And it scales with the square root of time, roughly.
And it scales with the square root of time, roughly. Illustrative chart - not real market data.

Annualising uses the square root of time, which assumes periods are independent. A daily ratio multiplied by the root of 252 overstates the annual figure where returns are serially correlated.

A candlestick series containing several opening gaps, with the largest opening gap marked. The headline on the chart reads: One gap can move the denominator for a whole year.
One gap can move the denominator for a whole year. Illustrative chart - not real market data.

A single gap can dominate the denominator because deviations are squared, so one event can halve a reported ratio for the period containing it.

A declining stretch of the long price series, with the entry price and the level at which a stop would trigger drawn as horizontal lines. The headline on the chart reads: No stop affects it, because it measures the sequence.
No stop affects it, because it measures the sequence. Illustrative chart - not real market data.

No stop changes it directly. Stops cap individual trades; the ratio describes the distribution of period returns, and the lever that moves it is position size.

A candlestick chart of the site's shared price history, annotated with the round-trip cost. The headline on the chart reads: And every round trip comes off the top line.
And every round trip comes off the top line. Illustrative chart - not real market data.

Turnover is the quiet drag. At 2% of a median bar’s range per round trip on this history, an active method spends a meaningful share of its numerator before the ratio is computed.

One more figure belongs beside it, and it comes from this site’s own corpus. Of the 31,760 trading and investing videos studied, exactly 1 has “sharpe ratio” in its title, against 258 videos for “backtest” at a median of 3,630 views. The measure that decides whether a result is worth anything is almost absent from the material teaching people to produce results.

That gap is worth naming rather than complaining about. A backtest produces a curve, which is photogenic; a ratio produces a number that usually says the curve is less impressive than it looks. Compute it anyway - it takes one line beside any equity curve you already have.

What to read alongside it

Two numbers fix most of its blind spots. The maximum drawdown says how bad the worst moment was, which the ratio never reports; the ulcer index says how long the account spent below its peak, which the ratio also never reports.

And the sample period has to be stated with the figure. A ratio computed over three years of one regime is a description of that regime. A Sharpe ratio quoted without a period, a drawdown and a trade count is not a performance claim — it is one number from a table, chosen because it was the most flattering.

What the Sharpe ratio is not

It is not a measure of downside risk. It counts both directions.

It is not distribution-free. The bell curve is assumed.

It is not comparable across periods. Length and regime both matter.

And it is not gameable-proof. A smooth curve is easy to construct.

When it fails

A sideways, range-bound candlestick series. The headline on the chart reads: In a range a flat curve can score well and earn nothing.
In a range a flat curve can score well and earn nothing. Illustrative chart - not real market data.

A nearly flat equity curve produces a small denominator. In a range a method that grinds out almost nothing with almost no variability can report a respectable ratio while earning less than cash, which is the arithmetic working exactly as specified and telling you nothing useful.

The second failure is a short sample. Fewer than a hundred periods and the figure is mostly noise.

A third is comparing across strategies with different return shapes. An options seller and a trend follower are not on the same scale here.

A fourth is ignoring the cash rate. It changes, and a ratio computed against the wrong one is wrong.

And a fifth is treating a high figure as a recommendation. It says variability was low relative to return, on the sample measured.

The original data

Of the 31,760 videos in this site’s corpus, 1 has “sharpe ratio” in the title against 258 for “backtest” at a median of 3,630 views; the counts are in research/corpus-coverage.json. On this site’s shared 576-bar history the 575 bar-to-bar returns have a mean of 0.0068% and a standard deviation of 0.347%. Measured shares within one, two and three deviations were 74.1%, 93.2% and 98.6% against the normal distribution’s 68.3%, 95.4% and 99.7%, with 8 moves beyond three deviations. The maximum drawdown was 3.76% and the ulcer index 1.67%. The figures are in research/series-measurements.json, produced by site/measure_series.py.

A 72-bar window of the shared price history, cut short at the decision bar. The headline on the chart reads: Two point one on three years. Good or short sample?
Two point one on three years. Good or short sample? Illustrative chart - not real market data.

The tail count is the specific reason to distrust a high ratio. Eight three-deviation moves where the model allows fewer than two means the denominator was computed from a distribution that does not describe the data — and every reported figure inherits that error. Ask for the drawdown and the trade count whenever you are shown a Sharpe ratio; those two numbers answer the questions it cannot, and a claim that survives all three is worth considerably more than one that survives only the first.

Standard deviation is the denominator and its assumptions. Ulcer index penalises only downside persistence. And drawdown is the number to demand alongside any ratio.

What I actually do

I use it as a screening number and never as a verdict. A method with a high ratio and a shape I would not want to hold is still a method I would not want to hold. Where it genuinely helps is stopping me from being impressed by a return that came with variability I had not looked at.

— Michael Whitman

This page is educational, not financial advice. Test every idea on your own charts before risking money.