Metrics5 min read

The Sortino Ratio

The Sortino ratio measures risk-adjusted return using only downside volatility, treating upside deviation as irrelevant to risk. It exists because the Sharpe ratio penalizes the asymmetric, fat-tailed return distributions that most strategies actually produce, conflating "volatility you want" with "volatility that hurts you."

For systematic traders, Sortino is the more honest comparison metric across strategies whose return distributions are skewed — trend-following, options selling, or anything with embedded convexity. It will not save you from path-dependent ruin, but it filters out one specific dishonesty baked into Sharpe.

The formula

Sortino = (R_p - R_f) / σ_d

Where R_p is the strategy's mean return over the measurement period, R_f is the risk-free rate (or minimum acceptable return, often called MAR), and σ_d is the downside deviation — the standard deviation computed only from returns below the threshold.

σ_d = sqrt( (1/N) * Σ min(0, R_i - MAR)^2 )

The summation runs over all N periods in the sample, but only returns below MAR contribute non-zero terms. Note the divisor: some implementations divide by the count of below-threshold observations only, others by N. The latter (full-sample N) is the Sortino convention and what produces comparable values across strategies with different downside frequencies.

Annualization follows the Sharpe convention: multiply by sqrt(252) for daily returns, sqrt(52) for weekly, sqrt(12) for monthly. The numerator must use the same period as the denominator before scaling.

Interpretation and thresholds

An annualized Sortino above 2.0 is strong for a systematic strategy on out-of-sample data. Above 3.0 is exceptional and should trigger suspicion of look-ahead bias, survivorship effects, or insufficient sample length before celebration. Below 1.0 indicates the strategy's edge does not meaningfully compensate for its drawdown profile.

Sortino values typically run 1.3x to 2x higher than Sharpe on the same return series, because eliminating upside variance shrinks the denominator. Comparing a strategy's Sortino to another strategy's Sharpe is meaningless — always compare like to like. A strategy with symmetric returns will show Sortino and Sharpe values within roughly 10 percent of each other.

Sample size matters more for Sortino than for Sharpe. If only 8 percent of your returns fall below MAR, then a five-year daily backtest gives you roughly 100 observations driving the entire denominator. That estimate is noisy. Treat Sortino computed on fewer than 200 below-threshold observations as a point estimate with a wide confidence band.

What Sortino does not capture

Sortino measures the dispersion of negative returns, not their sequencing. A strategy that loses 2 percent every Monday for a year and gains it back each Friday has identical Sortino to one that suffers a single catastrophic week — assuming matched downside magnitudes. Drawdown duration, recovery time, and consecutive-loss streaks are invisible to the ratio.

It also does not capture tail risk in the way traders sometimes assume. Downside deviation is a second moment; it weights a -5 percent return as 25 units of "badness" and a -10 percent return as 100 units. A strategy that occasionally produces -20 percent days still gets averaged into a finite, possibly attractive number. For tail behavior, look at conditional VaR, maximum drawdown, or the third and fourth moments of the loss distribution.

A high Sortino on a short-volatility strategy is a known trap. Selling tail risk produces smooth, mostly-positive returns with rare catastrophic losses — exactly the profile Sortino flatters. Always pair Sortino with maximum drawdown and tail metrics for any strategy with negative skew.

Finally, Sortino assumes the MAR is meaningful. If you set MAR to zero on a strategy benchmarked against a 5 percent risk-free rate, you are flattering the numerator while ignoring opportunity cost. Set MAR to either the actual risk-free rate or the return of the capital's next-best deployment.

In Kestrel Signal

Kestrel Signal reports Sortino alongside Sharpe, Calmar, and maximum drawdown on every backtest result, with MAR configurable per run (defaulting to the period's risk-free rate). Downside deviation uses the full-sample N convention for cross-strategy comparability. The metric is shown with its sample-size confidence interval when below-threshold observations fall under 500, so you see when the estimate is too noisy to trust.