Sortino Ratio: Sharpe, But Only Counting the Losses That Actually Hurt

The Sharpe ratio has one flaw that everybody notices within about ten seconds of understanding it: it treats your best month as a problem.

Standard deviation counts distance from the average in both directions. A +11% month and a −11% month contribute identically to the denominator. So a strategy that occasionally delivers an enormous winner gets marked down for it, and a strategy that grinds out small steady gains gets marked up, even when the grinder’s worst month was twice as bad.

The Sortino ratio is the fix. It is the same idea with one change: only the downside counts.

The formula

Frank Sortino developed it in the early 1980s as part of what became known as post-modern portfolio theory. The structure is identical to Sharpe. Only the denominator changes.

Two things are different from Sharpe, and both matter.

The benchmark is your target, not the risk-free rate. You choose T. Set it to zero and you are asking “how much return per unit of losing.” Set it to the risk-free rate and you are asking a Sharpe-like question. Set it to the 8% a year you actually need to justify the screen time and you are asking a business question. The number changes with the choice, so the choice has to be stated.

Only shortfalls enter the denominator. Months above target contribute exactly nothing. This is the whole point.

Which bars count

Here is a realistic twelve-month return series. Gold bars are months above target, blue bars are months below it.

Monthly returns: +3.2, −1.8, +2.1, +4.0, −2.6, +1.1, −0.4, +5.3, +2.2, −3.9, +0.8, +1.6 (percent). Target set to zero.

Sharpe puts all twelve bars into the denominator. Sortino puts four of them in and ignores the rest. Same trader, two different verdicts:

The denominator error that reverses rankings

This is the part almost every spreadsheet gets wrong, and it is worth the price of admission on its own.

Downside deviation is not the standard deviation of the losing months. It is the square root of the average squared shortfall across every period in the sample, where months above target contribute a shortfall of zero.

Why does it matter? Because the correct version penalises frequency of losing as well as depth. The incorrect version does not, and that flips results.

Take the two traders from the Sharpe ratio article. Trader A grinds: ten months of +1.0%, two months of −0.5%. Trader B waits: eleven months of −0.25%, then one month of +11.75%. Both finish the year at +9%.

Target return set to zero. Annualised by multiplying the monthly ratio by the square root of twelve. Ratios this high are an artefact of a deliberately clean toy series; real records rarely exceed 3.

Read the last two rows. Computed correctly, Sortino says these two traders are close, which is the sane answer given they made the same money and Trader B’s worst month was smaller. Computed the common wrong way, Sortino says Trader B is twice as good, because dividing by eleven losing months instead of twelve total months rewards him for losing often.

If a platform hands you a Sortino ratio, this is the first thing to check. Reproduce it on a series you built yourself before you trust it on the one you did not.

Choosing your target return

Three defensible choices, and the number moves a lot between them:

The third one is the most honest and the least used. It is also the version Sortino originally intended, since the minimum acceptable return was meant to be exactly that: the return below which the investor’s objective fails.

Where Sortino is better than Sharpe

Three situations where the switch genuinely changes your conclusion.

Positively skewed strategies. Breakouts, trend following, long-volatility structures, anything where you take many small losses and a few large wins. Sharpe systematically undersells these because the large wins inflate the denominator. Sortino stops doing that. If you run a system where the fat right tail is the entire edge, Sharpe is scoring your edge as a defect.

Comparing a grinder to a hunter. Two traders with identical returns and very different shapes are not comparable on Sharpe. Sortino gets much closer to an apples-to-apples read.

When you have already decided you do not care about upside variance. Which is almost every trader. Nobody has ever called their broker to complain about an unexpectedly large winner. Risk, as traders actually experience it, is one-directional. Sortino is the metric that agrees.

Where Sortino is worse than Sharpe

Two failure modes, and they are serious.

It flatters short-tail strategies even more than Sharpe does. Sell far out-of-the-money options and you produce a long run of small positive months and almost no shortfalls at all. Sharpe already rewards this pattern lavishly. Sortino, by discarding the upside variance and finding almost nothing on the downside, rewards it even more. Both metrics are silent about the loss that has not happened yet. As with Sharpe, a spectacular Sortino on a short-gamma book is not a measurement of skill; it is a measurement of how long the tail has stayed quiet. That is a question for risk of ruin, not for a ratio.

The sample-size problem is worse, not better. Sharpe uses every observation. Sortino throws most of them away. In our twelve-month example the entire denominator rests on four data points. Four. Any statistic built on four observations moves violently when a fifth arrives, which means Sortino needs a longer track record than Sharpe to be equally trustworthy, not a shorter one.

Run your own numbers

Also available on its own page: the full trading performance metrics calculator.

Frequently asked questions about the Sortino ratio

Is the Sortino ratio better than the Sharpe ratio?

It is better at one specific job: judging a strategy whose returns are not symmetric. It is not a universal upgrade. Sharpe is more widely reported, more comparable across sources, and built on a larger effective sample. The professional answer is to compute both and pay attention when they disagree, because the disagreement is the information.

What is a good Sortino ratio?

Over a multi-year record with a zero target, above 2.0 is good and above 3.0 is strong. Below 1.0 means your losing months are large relative to what you earn for enduring them. Treat any figure computed on less than three years of monthly data as indicative rather than conclusive, and be more sceptical than you would be with Sharpe, because the denominator uses fewer observations.

Can the Sortino ratio be infinite?

Yes, and it is a warning rather than an achievement. If no period in your sample fell below target, the denominator is zero. Some software reports this as infinity, some as a blank, some as an absurdly large number. All three mean the same thing: the sample contains no evidence about your downside, so the metric has nothing to measure.

What target return should I use?

Zero if you want a figure comparable to published ones. Your own required monthly return if you want a figure that tells you something about your business. Always state which you used, because a Sortino ratio without a stated target is an unlabelled number.

Can I compute Sortino on individual trades rather than monthly returns?

You can, but it stops being comparable to anything published, and per-trade risk-adjusted quality is better measured with expectancy and profit factor. Sortino, like Sharpe, is an equity-curve metric. Feed it periodic returns, not tickets.