The same return series can produce different Sharpe and Sortino values because the denominators answer different questions. Sharpe measures variability of differential returns; Sortino measures shortfalls below a target. Their numerators match only when benchmark and target match. Neither score proves that a strategy is safe.

What each formula actually measures

Sharpe divides mean differential return relative to a stated benchmark by the standard deviation of that differential return. Sortino uses return relative to a stated target and target downside deviation. Their numerators coincide only if the chosen benchmark and target coincide. State frequency, annualisation and deviation conventions before comparing values.

Standard deviation measures squared distances from the mean, so equal positive and negative returns do not necessarily contribute equally unless they are equally far from that mean. Sortino’s downside deviation measures shortfalls below its stated target. The difference can change the assessment of an asymmetric return series, but neither measure alone describes its entire risk.

When a simple denominator relationship applies

If returns are normal and the target equals their mean, downside deviation under the all-period denominator convention is standard deviation divided by √2. For targets below or above the mean, that relationship changes. Symmetry alone does not make Sortino a fixed 1.4 times Sharpe; their numerators and targets must also be specified.

Even this mathematical special case does not establish that an equity portfolio, bond fund or mean-reversion system has normal returns. At a target equal to the population mean, the Sortino numerator is zero; if that is also the Sharpe benchmark, its numerator is zero too. Do not turn the denominator relationship into a general nonzero ratio multiplier.

When the two ratios diverge dramatically

An asymmetric hypothetical series can produce different rankings because total dispersion and target shortfalls weigh its observations differently. A strategy label such as trend following does not establish the realised distribution, a frequency of winning months or the relation between these ratios.

A large positive observation affects both the arithmetic mean and the standard deviation. It contributes zero to downside squared shortfalls only if it is at or above the chosen target, while remaining in the all-period observation count. Its effect on either ratio must be calculated from the whole series; it is not a universal mark against the strategy.

A rare large loss can alter the mean and both denominators. A smooth prior sample may have omitted such outcomes, but neither statistic supplies their future frequency. Calculate the before-and-after inputs instead of assuming that Sortino always falls faster or that a specified trading style periodically loses a fixed percentage.

A worked example — same strategy, two verdicts

Retain twelve hypothetical net monthly returns: four at 2%, four at 5%, two at 15%, one at −2% and one at −4%. The sum is 52 percentage points and the arithmetic mean is 4.3333% per month; the sum is not annual compounded return. Choose a constant benchmark and target of exactly one-third of a percent in each month as exercise assumptions, not a current quoted rate. Both numerators are then exactly 4 percentage points per month. Use sample standard deviation with divisor 11 and downside deviation with divisor 12; do not annualise these results.

Sharpe and Sortino for an illustrative trend-following strategy — hypothetical figures
Sum of the monthly returnsSum 52 percentage points; arithmetic mean 4.3333% per month. The sum is not a compounded annual return.
Assumed monthly benchmark and targetExactly 1/3% in each month; a hypothetical fixed assumption for both ratios
Sample standard deviation of monthly differences√[(1,082/3)/11] = 5.726070 percentage points; no annualisation
Sharpe ratioMonthly: 4 / 5.726070 ≈ 0.698559
Downside deviation√[((7/3)² + (13/3)²)/12] ≈ 1.420746 percentage points; all 12 months count
Sortino ratioMonthly: 4 / 1.420746 ≈ 2.815423
Sortino relative to SharpeAbout 4.0303 times higher in this invented sample; no universal conversion factor

The high Sortino in this hypothetical sample reflects its small downside denominator. Twelve months with only two target shortfalls give limited evidence about future tail risk. There is no universally sufficient 36-month history: assess coverage, dependence and market conditions as well as length.

Which ratio is more useful for a retail trader

Sortino addresses shortfalls below a chosen target, while Sharpe addresses variability of differential returns. A 1 percent planned stop risk does not bound realised losses, and target reward-to-risk ratios do not establish the realised distribution. Report drawdowns and tail losses alongside either ratio.

Sharpe is widely used for performance comparisons, but implementations still differ in return frequency, benchmark, annualisation and treatment of fees. Compare like with like and report those choices. Sortino adds information about shortfalls below a chosen target; neither is the only valid cross-asset comparison measure.

State the return series, benchmark and calculation convention before using Sharpe to compare strategies.

Double scale, the same two ratios

Fictitious net series A is−1%,−1%,+2%,+4%;B doubles every return:−2%,−2%,+4%,+8%. Benchmark and target are0%. Mean and both deviations double, leaving their ratios unchanged. Worst single-period loss doubles.

Scroll sideways to see the full diagram

Two series with doubled amplitudeEach B return doubles A's absolute magnitude. With zero target and benchmark, both ratios match despite the larger period loss.Returns[%], one shared scaleA: green; B = 2 × A: coral0Period 1A−1B−2Period 2A−1B−2Period 3A+2B+4Period 4A+4B+8Mean: 1→2; each return × 2Sharpe: 0.408→0.408Sortino: 1.414→1.414Worst period: −1%→−2%
B bars are twice as long on the same axis;the ratios alone do not distinguish amplitude.

A mean1%, sample SD≈2.4495 percentage points, downside deviation≈0.7071. B has2%,4.8990 and1.4142. Thus Sharpe≈0.4082 and Sortino≈1.4142 in both. Include exposure size and losses in units relevant to your account.

Assumptions: four invented net returns;B=2×A. Benchmark and target0% in both. Sharpe uses sample SD divisor3;downside deviation divisor4. No annualisation. This scales data, not a promise of linear actual leverage, costs or execution. Formula or data references: William F.Sharpe — The Sharpe Ratio(1994); CFA Institute — The Sortino Ratio(2012).

Common mistakes when reading the two ratios

  1. Overinterpreting a short window. Ratios from few observations are sensitive to those observations. Neither twelve nor thirty-six months automatically makes them credible; state the window, frequency and limitations.
  2. Confusing variability with drawdown. Ratios use the stated return frequency, not necessarily monthly data. Maximum drawdown records the largest peak-to-trough decline in a specified account series and window. Inspect the equity path and the convention used for the Calmar ratio alongside the ratios rather than interpreting a ratio as a loss cap.
  3. Treating a high Sortino as proof of safety. It describes observed shortfalls below a target. The sample may omit a major loss or an intraperiod decline. A very small denominator can generate a large or undefined quotient without establishing low future risk.
  4. Using an unstated benchmark or target. Match units and periods to the return series. A policy rate or annual bond yield is not automatically the realised return on a suitable benchmark asset. Changing the benchmark can alter the Sharpe numerator and denominator; changing the Sortino target alters its shortfalls.

What to do tomorrow

  1. Document the calculation. Use returns adjusted correctly for deposits, withdrawals and fees. State the frequency, Sharpe benchmark and Sortino target in consistent period rates. Downside deviation is the square root of mean squared target shortfalls across all periods; a policy rate is not automatically an investable benchmark for your account.
  2. Compare matched definitions. Align windows, costs, frequency, benchmark, target and deviation convention. Do not apply fixed retail quality grades or infer asymmetry solely from a claimed typical 1.3–2 multiplier.
  3. Inspect what changed after a loss. Compare the full series, numerator and denominator before and after it. A change in a sample statistic is information about that sample; it does not identify the complete future outcome distribution.
  4. Add a path-based measure. Calmar relates a stated annualised-return measure to maximum drawdown over a specified window. Check that convention and the actual peak-to-trough history; reporting more scores does not itself validate a strategy.

Related reading: Sharpe ratio basics — formula and benchmarks; Sortino ratio basics — downside deviation and interpretation; Calmar ratio — the drawdown-based companion measure; maximum drawdown — the key path-of-risk metric.