A risk-adjusted return metric built like the Sharpe ratio but dividing by downside deviation only, so a strategy is not penalised for the volatility of its winning periods.
also: Sortino
Updated · Reviewed
The same idea as the Sharpe ratio, but it only counts the bumps that went against you. Big winning months stop being treated as a fault, so a strategy with rare large gains is no longer marked down for having them.
The Sharpe ratio measures how spread out returns are, in both directions, which means an unusually good month lowers the score exactly as much as an unusually bad one. Sortino removes that asymmetry — and in doing so it makes the gap between the two ratios a readable description of a strategy's shape.
Sortino ratio = (Mean return − Target return) ÷ Downside deviation Downside deviation is the root-mean-square of the returns that fell below the target, and the choice of divisor inside that mean is not standardised: dividing by the count of all periods and dividing by the count of losing periods produce different numbers from the same data. Two Sortino figures are only comparable when the same tool, the same target and the same sampling frequency produced both.
No single number describes a strategy. These metrics change how this one should be read.
Two EAs post the same mean monthly return and the same overall volatility, but the volatility comes from opposite directions. Sharpe cannot tell them apart; Sortino can.
A Sharpe ratio computed on total deviation would rank these two much closer together, because it counts EA A's good months as risk.
Calculation 2.0 ÷ 1.0 = 2.00 · 2.0 ÷ 2.5 = 0.80
Result Sortino 2.00 against 0.80 — identical averages, opposite risk profiles
Read Sortino next to Sharpe rather than instead of it. The pair carries information neither figure holds alone, and a Sortino quoted by itself cannot be checked.
| Range | What it means |
|---|---|
| Sortino much higher than Sharpe | Positive skew — the large moves are mostly wins. Common in trend-following EAs, and generally the shape a trader wants. |
| Sortino close to Sharpe | Roughly symmetric returns. Neither a warning nor a compliment; typical of strategies with a fixed stop and a fixed target. |
| Sortino below Sharpe | Arithmetically unusual and worth investigating — it points to a target return set above zero, or to a computation that is not doing what its label says. |
| Quoted with no target return or sampling frequency stated | Unverifiable. Both choices move the number materially, and neither is implied by the word 'Sortino'. |
mt5depot EA pages state which tester run each published figure came from and over what window, so two figures on this site can be read against each other rather than against a number from somewhere else.
The divisor inside downside deviation and the default target return are both implementation choices, and they are rarely documented in EA marketing. Two tools can report ratios that differ by half from the same trade list without either being wrong.
Downside deviation is an average of the losing periods, so many small losses and one catastrophic one can average out to a comfortable figure. The worst single decline is a separate measurement and has to be read separately.
It does not. Both ratios read the realised return series, and a grid or martingale EA holds its losers open. The risk is real, unrealised and invisible to every metric computed from closed trades.
Downside deviation is estimated from the losing periods only, so a sample of a few dozen returns may contain a handful of them. The shorter the window, the more the figure describes which losses happened to land inside it.
| Across the 14 EAs published here | Figure |
|---|---|
| Listings that publish a Sortino or a Sharpe figure | 0 |
| Records in the 1.5–3.0 band, all-period divisor | 7 |
| The same records, losing-period divisor | 4 |
| Records that change band when only the divisor changes | 11 |
| Records where Sortino beats Sharpe, all-period divisor | 14 |
| The same comparison, losing-period divisor | 11 |
| Rank agreement between Sortino and Sharpe | 0.96 |
We ran each listing’s own closed-trade file through one convention: monthly returns off the balance curve with flat months counted as zero, target zero, annualised by the root of twelve. No listing states one, so these belong to us rather than to the tester.
Rows two through four carry the problem. Divide the squared shortfalls by every month and Cairn scores 6.04; divide them by the losing months alone and the same record scores 2.86. The second convention costs between 34% and 53% of the figure, and it walks eleven of the fourteen across a line this page draws.
Rows five and six correct the FAQ below. Sortino clears Sharpe on all fourteen under the first divisor, and Beacon, Peregrine and Windrose each fall under their own Sharpe under the second — a case the table above blames on a non-zero target or a mislabelled calculation, and it is neither. Our methodology records the run behind each ledger.
Row seven deflates the pitch. Seven of the fourteen swap place between the two ratios and none moves more than three, so the disagreement this page hunts for shows up and stays too small to trade on. Both do disagree with the profit factor the listings actually publish, and they name a different record first.