The formula
Sortino = (annualised return − risk-free rate) ÷ annualised downside deviation
Identical to Sharpe with one substitution: downside deviation replaces standard deviation. It is computed from the losing periods only, so a strong month raises the return and does not raise the denominator.
"Losing periods only" is doing more work in that sentence than it appears to. There are two accepted ways to average the shortfalls, they produce ratios that can differ by a third from identical data, and most tools state neither — which one this calculator expects, and which one our own journal reports, is set out on the Sortino ratio entry.
The consequence is that Sortino is normally higher than Sharpe for the same data, and the gap between them is itself informative. It can land below Sharpe when the losses are few and large — see the FAQ for why that is a finding rather than a mistake.
Reading the gap
| Gap | What it suggests |
|---|---|
| Sortino ≈ Sharpe | Returns are roughly symmetric — gains and losses similar in size |
| Sortino well above Sharpe | Upside is larger and lumpier than downside — a trend-following shape |
| Sortino barely above Sharpe | Most of the volatility is on the losing side |
| Sortino below Sharpe | Few losing periods, but large ones — a concentrated left tail |
A trend follower with several large winning months and many small losing ones will show a much better Sortino than Sharpe, and the Sortino is the more accurate description of what holding it felt like — the volatility being penalised by Sharpe was the part you wanted.
Where Sortino is genuinely better
For any strategy with an asymmetric return distribution, which is most of them:
- Trend following — a few enormous winners among many small losses. Sharpe penalises exactly the months that justify the strategy.
- Long options — small, frequent decay against occasional large gains.
- Anything with a positive skew, where the point is that the upside is uneven.
Where it does not help
Sortino fixes one flaw and inherits the rest, which is worth saying plainly because it is often presented as simply the better metric.
Sample size. Downside deviation is computed from the losing periods only, so it uses fewer observations than standard deviation. On a short record it is the less stable of the two, not the more reliable.
Distribution shape. Neither ratio sees the tails properly. A strategy with many small gains and one catastrophic loss can post a fine Sortino for years — the loss has not happened yet, so it is not in the deviation.
Path. Neither sees drawdown. Two records with the same Sortino can have entirely different worst periods, and the worst period is what decides whether the strategy was still held at the end. Read either ratio next to maximum drawdown, never instead of it.