Sortino Ratio

The formula and what each piece means

The Sortino ratio is structurally identical to Sharpe, except the denominator changes:

Sortino  =  (R_p − MAR)  /  σ_downside

R_p        = portfolio return over the measurement period
MAR        = minimum acceptable return (usually r_f, sometimes 0%, sometimes a target)
σ_downside = sqrt( (1/N) · Σ max(MAR − R_i, 0)² )

Three design choices distinguish it from Sharpe:

• The denominator only counts deviations below the MAR. Positive surprises contribute zero, not negative or positive, to the shortfall sum. Upside volatility is invisible.
• The shortfalls are squared. Big shortfalls dominate small ones quadratically, preserving the second-moment structure that makes Sharpe and Sortino comparable on symmetric distributions.
• The denominator divides by N, not by the count of below-MAR observations. This matters: dividing by the count alone inflates the metric and breaks comparability across portfolios with different "loss frequencies." Sortino and Price were explicit about this in 1991.

The single benchmark MAR is investor-specific. A pension liability-driven portfolio targets the discount rate. A long-only equity manager often uses the risk-free rate. A hedge fund targeting absolute return uses 0%. The choice changes the number; it does not change the structural insight.

Worked example: where Sharpe and Sortino disagree

Consider two strategies measured over 24 monthly periods, both with mean return 1.0% and standard deviation 3.0%. Strategy A is symmetric — returns roughly normal. Strategy B is positively skewed — most months gain modestly, two months gain a lot, no big losses. Both Sharpe ratios are identical: (1.0% − 0.4%) / 3.0% = 0.20 per month.

Now compute Sortino with MAR = 0.4% (the monthly risk-free rate). For Strategy A, shortfalls accumulate roughly symmetrically with upside; downside deviation is around 2.0%, and Sortino = 0.6% / 2.0% = 0.30. For Strategy B, with its handful of large positive outliers and no large losses, downside deviation drops to about 1.1%; Sortino = 0.6% / 1.1% = 0.55 — nearly double. Both strategies look identical on Sharpe; Sortino correctly favours the one that gets its volatility from the right side of the distribution.

                    Strategy A         Strategy B
                    (symmetric)        (positive skew)
Mean monthly:        1.00%              1.00%
Stdev monthly:       3.00%              3.00%
Skewness:            0.0                +1.4
Downside dev:        2.00%              1.10%
Sharpe (monthly):    0.20               0.20
Sortino (monthly):   0.30               0.55

The Sharpe ratio gives a tied verdict on two strategies that are not interchangeable. Sortino captures the distinction.

Sortino versus other risk-adjusted-return metrics

Metric	Numerator	Denominator (risk)	Best for	Weakness
Sharpe (1966)	R_p − r_f	σ (total stdev)	Approximately symmetric returns	Penalises upside volatility identically with downside
Sortino (1991)	R_p − MAR	σ_downside (semi-stdev)	Skewed, asymmetric, option-rich	MAR is investor-specific; can be gamed
Treynor (1965)	R_p − r_f	β to market	Diversified equity portfolios	Ignores non-market risk; useless for hedge funds
Information ratio	R_p − R_b	tracking error (σ of R_p − R_b)	Active managers vs benchmark	Bound to the chosen benchmark
Calmar (Young 1991)	Annualised R_p	Maximum drawdown	Trend-followers, CTAs	One-point dependence; very noisy
Omega (Keating-Shadwick 2002)	E[max(R − τ, 0)]	E[max(τ − R, 0)]	Full-distribution sensitivity	Less familiar; sensitive to fat tails
Burke ratio (Burke 1994)	R_p − r_f	sqrt(Σ drawdown²)	Drawdown-sensitive strategies	Same one-point criticism as Calmar

In practice, professional performance reports show several of these side by side. The Sortino ratio's contribution is the cleanest second-moment alternative to Sharpe — same family, fixing the one well-defined defect.

History: Roy, Markowitz, Sortino, Price

The intellectual case for downside-only risk measurement is older than the Sortino ratio. A. D. Roy's 1952 Econometrica paper "Safety-First and the Holding of Assets" — published the same year as Markowitz's mean-variance paper, and reportedly without either author knowing about the other's work — proposed that investors minimise the probability of disastrous loss rather than minimise variance. Roy's "safety-first" criterion is the conceptual ancestor of every downside-risk metric. Markowitz himself wrote in Portfolio Selection (1959) that semi-variance — the average of squared shortfalls — was "a more plausible measure of risk than variance," but used variance because computers of the era could not invert covariance matrices fast enough to handle semi-covariance.

Frank Sortino, then at the Pension Research Institute, took up that thread when computers caught up. His 1991 paper with Robert van der Meer in the Journal of Portfolio Management formalised "downside risk" with the target semi-deviation. The current canonical form — Sortino and Lee Price 1994 — fixes details about how the MAR enters and how the denominator normalises. By the late 1990s the ratio had become a standard companion to Sharpe in hedge-fund and pension-fund performance reports.

Implementation pitfalls

• Dividing by the count of below-MAR observations. A natural-looking mistake. Standard deviation divides by N; semi-deviation does too. Some software libraries divide by the number of negative deviations, inflating Sortino by a constant factor that depends on how often the portfolio underperforms. CFA Institute Performance Standards (GIPS) discourage this variant.
• Confusing MAR with the risk-free rate. MAR is whatever target the investor cares about — could be 0%, could be inflation, could be a discount-rate hurdle. Reporting a Sortino without stating the MAR is uninterpretable.
• Using insufficient observations. Like all second-moment estimators, σ_downside has high sample-size requirements. Three years of monthly returns (36 observations) is the rough minimum for stability. Annual returns over 5 years (5 observations) produce statistically meaningless Sortinos and yet are widely published.
• Treating skewness as constant. Many strategies (selling out-of-the-money puts, merger arbitrage, carry trades) appear positively-Sortino during quiet markets and reveal severe negative skew only in stress. Reporting Sortino on a 5-year quiet sample is the most common way to make a tail-risk strategy look like a free-lunch generator.
• Smoothing artifacts. Illiquid assets (private equity, real estate, infrequently-priced hedge fund holdings) understate volatility through stale pricing. The resulting σ_downside is artificially low, inflating Sortino. Adjustments (Geltner unsmoothing, Getmansky-Lo-Makarov 2004 autocorrelation correction) exist but are rarely applied.
• Ignoring asymmetric scaling. Annualising Sortino is non-trivial. Multiplying by sqrt(12) assumes iid returns and a stable MAR — both are routinely violated.

When Sortino changes the answer — and when it doesn't

For approximately symmetric, lightly-tailed return distributions — broad equity indices, balanced 60/40 portfolios, large-cap mutual funds — Sortino and Sharpe rank portfolios essentially identically. The two metrics differ by a constant scaling factor and the choice rarely matters.

The gap opens up for skewed return distributions. Strategies that systematically take asymmetric risk — selling premium (volatility short), carry trades that pay off until they don't (FX carry, EM debt), tail-insurance writing, merger arbitrage — show artificially attractive Sharpe ratios because their few large losses are diluted by symmetric-volatility scaling. Sortino corrects this. Strategies that systematically generate asymmetric gain — long volatility, long out-of-the-money calls, trend-following CTAs — look unattractive on Sharpe and reasonable on Sortino. Hedge fund performance attribution increasingly relies on Sortino plus a fat-tail diagnostic (Cornish-Fisher VaR, modified Sharpe) precisely because of this asymmetry.

When Sortino isn't the right tool

• Drawdown-driven mandates. If the investor's binding constraint is maximum drawdown (e.g. fund of funds with a high-water mark or a fund with a 10% drawdown trigger), Calmar or Burke are more directly aligned with the constraint.
• Benchmark-relative mandates. Active managers paid to beat the S&P 500 are evaluated against the Information ratio, not Sortino — the benchmark, not absolute return, is the right point of comparison.
• Distribution-aware decision making. When the entire return distribution matters — e.g. tail-risk budgeting, scenario analysis — Omega ratio or full-distribution diagnostics are more informative than any single second-moment ratio.
• Insufficient sample. No risk-adjusted metric is reliable on small samples. With less than ~36 monthly observations, narrative due diligence outperforms ratio-comparison.