The sortino ratio is a risk-adjusted return measure calculated as a strategy’s excess return over a minimum acceptable return, divided by its downside deviation. The formula in one line is sortino = (mean return – MAR) / downside deviation. What sets it apart from every other Sharpe-family ratio is the denominator: it counts only the returns that fell short of the target. Positive months contribute nothing.
Downside deviation is the square root of the mean squared shortfall below MAR, where the mean is taken across every period, not just the losing ones. Positive returns are recorded as zero in the shortfall series before the average is computed. Two consequences follow. A strategy whose losses are shallow will have a much smaller denominator than its total volatility would suggest, so the sortino ratio will always be greater than or equal to the Sharpe. And a strategy with no returns below MAR has a downside deviation of zero and an undefined sortino.
That mathematical asymmetry is the whole reason the sortino ratio exists. Two systems land on my desk with identical Sharpe ratios of 1.40. One of them spends 18 months grinding out a quiet equity curve with shallow 4% pullbacks. The other had a 17% drawdown in March 2020 that took 9 months to recover, and a 12% drawdown in late 2022 it never fully recovered before its next loss. On Sharpe, they tie. On the lived experience of trading them, they are nothing alike. The sortino ratio is built to settle that disagreement.
How the sortino ratio is calculated
The mechanics are worth stepping through carefully because the convention matters. The numerator is the strategy’s average excess return over the minimum acceptable return, often written MAR (the risk-free rate, zero, or a hurdle rate, depending on the trader’s choice). The denominator is where the sortino ratio departs from Sharpe.
Concretely: take every return that fell below the MAR, square the shortfall, average those squared shortfalls (dividing by the total number of periods, not just the losing periods, which is the convention used by the original Sortino-Price papers), then take the square root. That number is the downside deviation.
The formula collapses to: sortino = (mean return – MAR) / downside deviation. The upside months contribute zero to the denominator. A trader who runs the same monthly returns through this calculation will always get a sortino that is greater than or equal to the Sharpe, because the denominator can only shrink or match.
What the Sharpe ratio actually measures
The Sharpe ratio takes a strategy’s excess return over a risk-free rate and divides it by the standard deviation of its returns. Both gains and losses sit in that standard deviation. A strategy that returns +8% in one month and -2% in the next contributes the same amount of “risk” from the +8% month as from a +8% month in a strategy with a -8% month sitting next to it.
That mathematical symmetry is the source of every quiet complaint about Sharpe. A momentum book that catches a +14% month in a runaway tech name gets punished in the denominator the same way a trend system that takes a -14% gap-down loss gets punished. The Sharpe ratio cannot tell those two events apart. I find that more than an academic problem when I am comparing screeners, and our AlphaSharpe screening note walks through how the Sharpe rank shuffles when you re-rank by sortino instead.
Why upside variance is not a loss
If a trader runs a strategy for a year and one month delivers a +22% return, that month does not need to be penalised. Capital is not destroyed by a good month. The trader did not have to sit through psychological pressure to hold the position, and the equity curve did not retrace. The only reason a +22% month is “noise” to the Sharpe ratio is that the formula was written symmetrically.
What the Sharpe ratio does NOT say is that upside variance is bad for the trader. It says upside variance is bad for the ratio. Those are different sentences. A common misread is to look at a high-variance momentum strategy with a Sharpe of 0.9 and call it “too volatile,” when in fact every one of its volatile months may have been a gain. Strip the upside out of the denominator and the same equity curve can produce a sortino above 2.
A worked example with two equal-Sharpe strategies
I will use a deliberately small dataset to make the arithmetic visible. Strategy A returns, in twelve monthly percentages: +3, +2, +4, -3, +2, -2, +3, +1, +2, -1, +4, +5. Strategy B returns: +6, +5, +1, -4, +7, -5, +1, +8, -2, -1, +2, +4. Both average +1.67% per month. The standard deviation of both rounds to 3.0%. With a 0% MAR, Sharpe is roughly 0.56 for each.
Now the downside numbers. Strategy A has three negative months: -3, -2, -1. Squared shortfalls are 9, 4, 1, summing to 14. Divide by 12 periods, take the square root, and downside deviation is about 1.08%. Sortino A is 1.67 / 1.08, which is 1.55.
Strategy B has four negative months: -4, -5, -2, -1. Squared shortfalls are 16, 25, 4, 1, summing to 46. Divide by 12, take the square root, and downside deviation is about 1.96%. Sortino B is 1.67 / 1.96, which is 0.85.
Same Sharpe. Sortino A is almost twice Sortino B. The trader who runs Strategy B is taking deeper losses to produce the same average, and the sortino ratio surfaces exactly that. I keep these two columns side by side whenever I am comparing strategies, and our equity-curve-trading guide shows how the same divergence shows up visually on the curve itself.
Picking the MAR (minimum acceptable return)
The MAR is where most of the disagreement between practitioners shows up. Set it to 0% and the sortino measures “how often, and how badly, the strategy lost capital.” Set it to the risk-free rate (call it 0.4% per month for a roughly 5% annualised T-bill) and the sortino measures “how often the strategy underperformed cash.” Set it to a hurdle rate like 1% per month and the sortino measures “how often the strategy missed its own internal target.”
I generally prefer MAR = 0% for technical strategies because the question I am answering is whether the system loses money, not whether it beats Treasuries. For a fund that has to clear a benchmark, the MAR moves to the benchmark’s return. The number itself is less important than picking one and holding it across every strategy you compare. Comparing a sortino calculated against 0% MAR to a sortino calculated against a 5% hurdle is a category error, not a comparison.
What counts as a good sortino ratio
Practitioners tend to read the sortino ratio in bands. Below 1.0 is weak: the strategy is generating downside deviation at roughly the same rate as excess return, and the trader is being paid poorly for the losses being absorbed. Between 1.0 and 2.0 is the range most long-only trend and momentum books live in over multi-year windows, and I treat it as an acceptable, workable band. Above 2.0 is strong, and above 3.0 is exceptional. A number above 3 is also often a signal to check the lookback and the strategy design before trusting the reading; that band is where curve-fitting and hidden leverage tend to show up.
Those bands are not universal. Three caveats change what “good” means before the ratio is even compared.
The first is lookback. A sortino computed over 12 monthly returns is unstable, because the denominator is dominated by two or three losing months, and a different starting date can move the reading by half. I want at least 36 months of monthly data, or two to three years of daily data, before I trust a sortino as a strategy characteristic rather than a snapshot. A 2.4 sortino over 18 months is not the same object as a 2.4 sortino over 60 months.
The second is strategy type. Mean-reversion strategies clip individual losses tightly and tend to produce sortinos in the 1.5 to 2.5 range naturally, until a tail event that the ratio cannot see arrives. Trend-following books take frequent small losses and hold rare large winners, which drags the mean down while keeping downside deviation moderate, so a 1.1 to 1.6 sortino is common and healthy for that style. Buy-and-hold equity over long windows usually sits between 0.4 and 0.8. Ranking a trend book against a mean-reversion book on a single sortino number treats different animals as the same species.
The third is the rate environment. Anchor MAR to the risk-free rate and a 2026 sortino has to clear a higher bar than a 2020 sortino, because cash pays more before the ratio starts building. A worked example: a swing trend system with a mean monthly return of 1.2% and a downside deviation of 1.5% has a sortino of 0.80 against a 0% MAR. Move the MAR to 0.4% per month, roughly a 5% annualised T-bill, and the same return series produces a sortino of 0.53. Nothing about the strategy changed. The band the number falls into moved from “acceptable” to “weak” purely because cash pays more. Any comparison across time has to hold MAR constant, or the comparison is measuring the rate cycle and not the strategy.
The rule I settle on is that a sortino above 1.5 over a 3-year window with a consistent MAR is a strategy worth investigating further. A sortino below 1.0 on the same window is a strategy whose downside profile has to be examined before it is put in front of live capital. Everything in between is where most real edges live, and the ratio alone will not sort them. The sortino ratio is a filter, not a verdict.
Where the sortino ratio misleads you
The sortino ratio does NOT measure tail risk. A strategy with a long run of small losses and one catastrophic -40% month can still produce a respectable sortino, because the downside deviation averages the squared shortfalls across all periods. A single fat-tail event gets smoothed into a number. The Taleb critique of these ratios is exactly that: any single-number summary of a return distribution misses the shape of the tail.
It also rewards strategies that quietly add leverage. If a trader doubles position size for a year, the negative months get bigger and the positive months get bigger in lock-step. Sortino can hold steady or even improve, because the mean return scales linearly while the downside deviation scales linearly too. The ratio looks clean. The capital at risk has just doubled.
A third misread is treating the sortino as a “loss” metric. It is a return-per-unit-of-downside-deviation metric, not a maximum drawdown. A strategy with a healthy sortino can still take a 25% intra-period drawdown the trader has to sit through, because drawdown depends on the sequence of losses and the sortino only sees their squared average. I cover the sequence question in our hard drawdown stops note.
How I use the sortino against Sharpe in practice
The single most useful pattern I have found is to run both ratios side by side and look at the gap. A strategy with a Sharpe of 1.2 and a sortino of 1.4 is roughly symmetric: its volatility is genuinely two-sided. A strategy with a Sharpe of 1.2 and a sortino of 2.8 is heavily right-skewed: most of the variance is on the upside, and the trader is being unfairly punished by the Sharpe denominator.
The reverse is the warning sign. A Sharpe of 1.2 paired with a sortino of 0.9 is unusual and bad. It says the strategy’s downside deviation is higher than its full-period standard deviation, which can only happen when the losing months are deeper than the gaining months and the calculation is being dominated by a small cluster of large losses. I treat that pattern as a flag to look at the drawdown chart directly and not to trust the Sharpe rank.
I also keep a sortino-to-Sharpe ratio (sortino divided by Sharpe) as a quick skew read. Above 1.5, the strategy is asymmetric to the upside. Below 1.0, it is asymmetric to the downside. Between 1.0 and 1.5, the two ratios largely agree.
What the sortino ratio does not tell you
The sortino ratio does NOT tell you about trade frequency. A strategy with two trades a year and a strategy with two hundred trades a year can produce identical sortinos. The trader has to know whether the number is built on a sample large enough to be stable.
It does NOT tell you about correlation to the rest of the book. A 1.8 sortino on a single strategy is meaningless if that strategy is 0.9 correlated to another strategy already in the portfolio. The combined book sortino is the only number that pays the bills.
It does NOT replace the equity curve. The number is a summary statistic. The curve is the actual record. Two strategies with the same sortino can have visibly different curves, one with the losses clustered and one with them spread, and the trader has to look at the curve before sizing. Paul Tudor Jones built a career on watching the actual record and not the ratio, and I think that order of operations is still right.
Compare the ratios, then look at the actual losses
The sortino ratio fixes a real defect in the Sharpe ratio. It stops punishing upside variance, surfaces strategies that look equivalent on Sharpe but lose money differently, and gives you a cleaner read on whether the volatility you see is the volatility you fear. None of that makes it a sufficient statistic. Use it as the second number in a pair, not as a single verdict, and always look at the drawdown sequence after the ratio.
Learn the pattern. Ride the trend. Keep the gains.
Educational content only. Not investment advice. Trading involves risk. You are responsible for your decisions.