Every comparison page on this topic lands on the same sentence: the Sharpe ratio penalizes all volatility, the Sortino ratio penalizes only downside volatility. Then it stops. That sentence is correct but incomplete, because it hides two decisions that are left to the person computing the number — the target rate you measure shortfall against, and the denominator you divide the downside sum by. Change either one and the ranking between two strategies can flip, using the exact same return series. That flip, not the formula, is the part worth forty minutes of a trader's time.
The One Real Difference: Total Deviation vs Downside Deviation
Both ratios share the same numerator: average return minus a target rate. Sharpe's denominator is the standard deviation of every return in the sample — gains and losses both count against the strategy. Sortino's denominator is the downside deviation: only the returns that fell below the target rate contribute, and each one is squared before it's averaged. A strategy with a string of large wins and zero losses scores identically on both ratios in theory, since there is nothing below target to penalize — but that case almost never shows up in a live record, which is exactly why the two numbers normally diverge.
Quantzee's own backtesting documentation treats this divergence as a diagnostic, not a decoration. If Sharpe and Sortino move together across a family of related configurations, the strategy's losses are shaped like its gains. If they diverge sharply, the loss tail is doing something the mean and total variance don't capture — and that's worth a second look before sizing a position off either ratio alone.
Both ratios are also silent about periodicity unless someone states it. A Sharpe of 0.3 computed on daily data and a Sharpe of 0.3 computed on monthly data are not the same strength of result once annualised, because the sampling frequency changes how much estimation noise sits inside the standard deviation. The same is true of Sortino, with an extra wrinkle: the below-target count that drives its denominator also shrinks or grows with the sampling frequency, so a daily record and a monthly record of the identical underlying strategy can produce Sortino ratios that disagree even when their Sharpe ratios roughly agree.
The Target Rate Is a Choice, Not a Constant
Neither ratio has one fixed target built in. Zero is the simplest choice — anything negative counts as downside. A short-dated government bill rate is a second common choice, treating any return below that bill rate as the real failure line. A required return — the minimum return a strategy has to clear to be worth running — is a third. Each produces a different number from the identical series, and page one rarely says which one it used.
We measured this on one illustrative monthly return series, twelve periods, expressed in percent: 2.1, -1.0, 1.8, 0.9, -3.2, 2.5, 1.1, -0.5, 1.9, -2.8, 2.2, 1.4. The mean is 0.533% and the sample standard deviation is 1.96%.
- Target = 0%: Sharpe = 0.272. Four of the twelve months fall below target, giving a downside deviation of 1.269% (squared shortfalls divided by all 12 periods) and a Sortino ratio of 0.420.
- Target = 0.3% (a stand-in for a short-dated bill rate): Sharpe drops to 0.119. Sortino drops to 0.164.
Both ratios fell by more than half from a 0.3-point change in one assumption, on the same twelve months of returns. A ratio quoted without its target rate stated is not comparable to any other ratio, including one computed from the same data a week later under a different convention.
The Denominator Convention Nobody Names
This is the gap that costs the most and gets mentioned the least. Once you've picked a target and identified which periods fell below it, there are two accepted ways to build the downside deviation:
- Divide by every period in the sample (N). A strategy with few losing months gets a smaller downside deviation partly because most of the denominator is zero-contribution gain months, not because the losses themselves were mild.
- Divide by the count of only the below-target periods (n). This measures the typical size of a loss, given that a loss happened, and ignores how often losses happened at all.
Using the same twelve-month series at target = 0%, with four below-target months: dividing by all 12 periods gives a downside deviation of 1.269% and a Sortino of 0.420. Dividing by the 4 below-target periods only gives a downside deviation of 2.198% and a Sortino of 0.243. That's a 42% difference in the published ratio from the identical series, caused entirely by which denominator convention was used — and most public write-ups never say which one they picked. We disclose ours below in the reporting checklist; readers checking a vendor's number should ask for the same disclosure before trusting a comparison.
One Return Series, Two Rankings
The convention argument is abstract until two strategies actually swap places. Take two ten-period monthly series, Strategy C and Strategy D, both profitable, both scored at target = 0% with the all-periods convention:
Strategy C: 2.2, 1.5, 1.6, -1.9, 0.8, 1.4, 2.7, 2.8, 1.6, 0.9 — mean 1.360%, standard deviation 1.328%, one losing month.
Strategy D: 2.5, 2.0, -0.4, -0.8, 0.7, 1.0, 0.3, 0.8, 1.9, 0.4 — mean 0.840%, standard deviation 1.053%, two losing months.
Sharpe ranks C ahead of D: 1.024 against 0.798 — C has the higher mean and a tolerable total variance, so the Sharpe ratio likes it more. Sortino reverses the call. C's single loss is a sharp -1.9%, which drives its downside deviation up to 0.601% despite happening only once; D's two losses are mild, -0.4% and -0.8%, so its downside deviation is a much smaller 0.283%. Sortino scores D at 2.970 against C's 2.264 — the opposite order from Sharpe, on the same two months of returns, with nothing changed except which deviation you divide by. A desk that screens on Sharpe and a desk that screens on Sortino would staff two different strategies from this exact pair.
Annualisation Is a Convention Too, and It Behaves Worse Here
Scaling a monthly Sharpe to an annual figure by multiplying by the square root of 12 assumes returns are independent and identically distributed period to period. That assumption is already a simplification for total-volatility measures; applied to a downside measure it gets worse, because downside deviation is built from a skewed, truncated subset of the distribution — by construction it isn't symmetric, so the square-root scaling that works reasonably well for a roughly normal return stream distorts further as the loss tail gets fatter. We still annualise Sortino the same way Sharpe is annualised, for comparability across strategies reported at the same periodicity, but we flag it in the same breath: a downside measure that's been square-root-scaled should be read as directionally useful, not as a precise annual figure, especially for any strategy with a visibly asymmetric return distribution. The CFA Institute's performance-measurement standards make the same point about scaling risk-adjusted ratios across periodicities — the convention is a modelling choice, not a law of statistics, and it should be disclosed alongside the number. Investor-education material from Investor.gov makes a related, simpler point worth repeating here: any single risk-adjusted ratio, however it's built, is a summary of past behavior, not a forecast of what a strategy will do in 2026 or any later year.
Estimation Noise: Why a Thin Loss Sample Lies
A Sortino ratio is computed from a subset of the sample — only the below-target periods feed the denominator. That subset is almost always smaller than the full sample Sharpe uses, so Sortino carries more estimation noise at the identical record length. The fewer the losing periods, the less stable the number, and that's exactly the situation that makes a short, lucky-looking record flatter itself the most.
Take a six-period monthly series with a single large loss: 3.0, 2.5, -5.0, 2.8, 3.2, 2.1. Mean is 1.433%, standard deviation 3.175%, Sharpe 0.451. The single loss drives a downside deviation of 2.041% (dividing by all 6 periods) and a Sortino of 0.702. Now change that one month's loss from -5.0% to -6.5% — a plausible revision if the record gets one more bad week folded in, or if slippage on that single trade was underestimated the first time around — and the mean drops to 1.183%, Sharpe falls to 0.313 (a 30.6% move), while Sortino falls to 0.446 (a 36.5% move). One data point, roughly a sixth of the sample, moved both ratios by close to a third just by being restated. On a six-month record that's not an edge case; it's the normal behavior of a ratio built on four or five below-target observations. We would not size a live position off a Sortino computed on fewer than roughly 20 below-target periods without flagging that instability explicitly in the report.
Where the "Above 1.00 Is Good" Benchmarks Actually Come From
One widely cited competitor review of this exact ratio reports that the commonly repeated quality bands — above 1.00 is good, above 2.00 is very good, above 3.00 is excellent — trace back to fund-analysis work on monthly mutual fund data. Those thresholds were built for monthly-rebalanced, diversified fund portfolios, not daily intraday trading records with dozens or hundreds of entries a month. Per the data Quantzee has reviewed internally, an intraday options-selling book sampled daily will often show a materially different Sharpe-to-Sortino spread than a monthly fund return series purely because of the sampling frequency, independent of whether the underlying strategy is actually better or worse. Carrying a monthly-fund benchmark table over to a daily trading log without adjustment is a comparison between two different measurement regimes dressed up as one number.
Selection Bias: The Ratio Nobody on This Topic Mentions
Every comparison above assumes the ratio you're looking at is the only one that was computed. In practice, a reported Sortino or Sharpe is usually the strongest result out of many parameter variants that were tested and discarded along the way — a stop-loss distance, an entry filter, a target rate, a lookback window, each tried several ways before one combination got published. A ratio selected as the strongest of many tested variants is inflated relative to its true, out-of-sample value, and the correction is a haircut sized to how many effective, independent tests produced it.
Our own internal testing documentation gives a concrete illustration. Our methodology was to run the full parameter grid for one strategy family and count every independent configuration that was actually evaluated, not just the one that got published. Across 396 effective tests, collected over 24 months of ongoing research in 2026, the appropriate haircut factor worked out to 3.4587 — meaning the headline ratio from the single strongest-performing configuration needs to be divided down by roughly that factor before it reflects what an investor should actually expect going forward, rather than what one fortunate configuration happened to show. We found that this correction mattered more for Sortino than for Sharpe in our own data set, because Sortino's smaller effective sample (the below-target periods only) makes it easier for one parameter sweep to land on a configuration that looks good mostly by chance. A vendor or a strategy page quoting a single standout ratio without disclosing how many variants were tried to find it is presenting the strongest outcome of an unreported experiment as if it were the only outcome.
What to Report Instead of a Bare Number
A ratio by itself, Sharpe or Sortino, answers nothing a reader can check. Four labels turn it into something verifiable:
- The target rate used in the numerator — zero, a bill rate, or a required return, stated as a number.
- The denominator convention for Sortino — divided by all periods or by below-target periods only.
- The periodicity the raw returns were measured at — daily, weekly, or monthly — since annualisation assumptions ride on this.
- The sample length, in periods, and ideally the count of below-target periods specifically, since that count is what actually drives Sortino's stability.
Quantzee reports both ratios side by side wherever we publish them, each one carrying these four labels, rather than a single unlabeled figure next to a verdict like "good" or "excellent." If a quoted ratio arrives without this context, the honest response is to ask for it before using the number to compare two strategies, and to check whether it survived a realistic account of how many configurations were tried to produce it. For a deeper look at how much of a strategy's reported edge can disappear once you correct for drawdown shape and testing overhang, see our max drawdown glossary entry and our overfitting in trading breakdown, and for the step-by-step process we use to stress a configuration before trusting any single-number summary, see our stress-testing checklist.
The Sharpe ratio itself — its formula, its standard annualisation, and a short summary of this Sortino comparison — is covered in our Sharpe ratio glossary entry; this page focuses on the consequence of the comparison rather than repeating the definitions.
