I got into an argument once about whether a strategy with a great month was actually risky. The equity curve looked jagged, sure, but almost all the jaggedness was upside. A few days it ripped 15 percent, a couple days it gave back 3 or 4. Its Sharpe ratio was mediocre because standard deviation counted those 15 percent days as risk. Which, if you say it out loud, sounds absurd. Nobody has ever called their broker to complain that the position went up too fast.
That is the problem the Sortino ratio was built to fix. It is a small change to a familiar formula, but the small change is the whole point, so it is worth walking through slowly with real numbers.
Why standard deviation gets it wrong
Sharpe divides your excess return by the standard deviation of your returns. Standard deviation measures how far each return sits from the average, in both directions. A day that beat the average by 10 points and a day that missed it by 10 points contribute exactly the same amount to the number. The math cannot tell delight from pain.
For a symmetric thing, that is fine. Most assets that matter to traders are not symmetric. Crypto in particular tends to grind sideways, chop, then jump. The jumps are usually why you held it. Penalizing the jumps as if they were losses is measuring the wrong thing, and it quietly makes your best strategies look worse than your boring ones.
Downside deviation asks a narrower question. How much do returns fall below a line you actually care about, and only those returns. Everything above the line counts as zero risk, because from a risk standpoint it is not a problem you need to solve.
Computing downside deviation step by step
You need a minimum acceptable return first, usually written MAR. This is the return you would be unhappy to fall short of. Zero is a common choice, meaning you only care about losing money. Some people use a risk-free rate, some use a target like 1 percent per period. Pick one and be honest about it, because the whole calculation hangs on it.
Say you have six periods of returns for a crypto position, in percent: +8, +12, -5, +3, -9, +6. Use a MAR of 0, so any negative period is a shortfall.
- For each period, compute the shortfall below MAR. If the return is at or above MAR, the shortfall is 0. So the +8, +12, +3, and +6 periods all contribute 0.
- The -5 period has a shortfall of -5, the -9 period has a shortfall of -9.
- Square each shortfall: 0, 0, 25, 0, 81, 0.
- Average those squares across all periods, not just the negative ones. That is the part people get wrong. Sum is 106, divided by 6 is roughly 17.67.
- Take the square root. Downside deviation is about 4.2 percent.
Dividing by all six periods, including the winners, is deliberate. It keeps the denominator comparable to standard deviation and stops a strategy from looking artificially safe just because it rarely loses. A rare, huge loss should still show up.
From downside deviation to the Sortino ratio
The Sortino ratio is your average excess return divided by that downside deviation. Excess return here means average return minus the MAR. Our returns average out to +2.5 percent per period. With a MAR of 0, excess return is 2.5. Divide 2.5 by 4.2 and you get roughly 0.60 for the period.
Now compare that to what Sharpe would have said. The standard deviation of that same series, counting the +12 and +8 days as risk, comes out around 7.3 percent. Sharpe would be about 2.5 divided by 7.3, or roughly 0.34. Same returns, same series, and Sortino is nearly double. The gap is entirely those upside days that Sharpe treated as danger and Sortino ignored.
That gap is the useful signal. When Sortino sits well above Sharpe, the strategy's volatility is mostly upside. When they sit close together, volatility is symmetric, or the losses are doing most of the moving, and you should worry more.
Which one to trust, and when it lies too
My rule of thumb is to read them as a pair, not to pick a winner. Sharpe tells you how bumpy the ride is overall. Sortino tells you how bumpy the bad part is. If a strategy has a strong Sharpe and an even stronger Sortino, the bumps are working for you. If Sharpe looks fine but Sortino is much worse, the drawdowns are concentrated and ugly even though the average smooths them over.
Sortino has its own failure mode, so do not oversell it to yourself. It gets unstable and unreliable when there are very few periods below the MAR. Divide by a downside deviation built from two data points and the number is basically noise. A strategy that has not lost much yet will flash a gorgeous Sortino right up until the first real drawdown, and then the denominator finally has something to chew on. Short backtests flatter it badly.
A few practical checks before you lean on the number:
- Confirm you have enough downside periods for the deviation to mean anything. A handful of losers over hundreds of periods is not a distribution, it is a coincidence.
- State your MAR out loud. A Sortino computed against a 0 percent bar and one computed against a risk-free rate are different animals, and comparing them is meaningless.
- Always look at max drawdown next to it. Sortino compresses the pain into one number, and a single number can hide a 60 percent hole you would never have sat through.
When we surface risk-adjusted metrics inside Blockcircle's scorecards, we show Sharpe and Sortino side by side for exactly this reason. One number is a summary, and every summary throws away detail. Two numbers that disagree tell you where to go look, which is usually the more valuable thing. Run both on your own series, watch where they split, and you will start to trust the split more than either figure on its own.