The Sortino Denominator Nobody Agrees On

Sortino changes two things about Sharpe, not one

The familiar summary is that Sortino replaces total volatility with downside volatility. True, and incomplete: it also replaces the risk-free rate in the numerator with a minimum acceptable return, which is a target you choose rather than a rate the market quotes. Those two changes are independent, and conflating them is how people end up comparing a Sortino computed against a zero target with one computed against a benchmark and treating the difference as information about the funds. If you want the comparison to mean anything, fix the target first and state it, then let the denominator do its work.

When the denominator does its work, it can reverse a ranking, and that reversal is the reason the ratio exists. Set our steady Fund A against a Fund B whose twelve months are 0.5, 8.0, 0.4, 0.6, 7.5, 0.5, −0.4, 9.0, 0.5, 0.4, −0.2 and 0.6 percent — mostly flat, with three explosive months. Fund B has an annualised standard deviation of 12.3881 percent against Fund A's 4.3338, so Sharpe ranks A first, 2.5843 to 1.8728. But almost all of B's variability is upward: only two months fall below the target and the sum of their squared shortfalls is 0.8650 against A's 2.0875, so B's annualised downside deviation is 0.9301 percent against A's 1.4448. Sortino ranks B first, 24.9448 to 7.7518. Neither measure is malfunctioning. Sharpe is answering a question about how much the return wobbles; Sortino is answering a question about how often and how badly it disappoints. Those are different questions, and a manager who never loses money but frequently makes a great deal of it will be punished by the first and rewarded by the second.

Two denominators, one factor, and a ranking that flips because of it

The relation between the two conventions is mechanical. Both start from the same sum of squared shortfalls; one divides by the count of all periods, the other by the count of below-target periods; so the second denominator is the first multiplied by the square root of the total count over the below-target count, and the second ratio is the first divided by that same factor. On Fund A with three of twelve months below target the factor is exactly two. On Fund B with two of twelve it is 2.4495. Notice what that means: the second convention penalises a fund more heavily the fewer bad periods it had, because a smaller denominator count inflates the measured downside risk. A fund with a single bad month in twelve is penalised by the square root of twelve, or 3.4641, purely for the rarity of its bad month.

That property is not a curiosity; it can reverse a ranking on its own. Compare a Fund C with returns of 2.0, −0.2, 2.1, −0.3, 2.2, −0.1, 2.0, −0.2, 2.3, −0.3, 2.1 and −0.1 percent — six of twelve months just below the target, none of them serious — against a Fund D of 1.6, 1.7, 1.5, 1.8, −0.9, 1.7, 1.6, 1.8, 1.5, 1.7, −0.8 and 1.7 percent, which misses only twice but misses by more. Dividing by all twelve months, Fund D wins, 6.2996 to 5.3598. Dividing only by the below-target months, Fund C wins, 3.7900 to 2.5718. Same data, same target, opposite conclusions, and the whole reversal comes from C being penalised by the square root of two while D is penalised by the square root of six. If you are comparing two Sortino ratios, they must have been computed the same way, and if the source does not say which way, the number is not comparable to anything.

Move the target and the same fund scores 17.90 or −1.83

The minimum acceptable return is doing at least as much work as the denominator convention, and unlike the convention it is meant to be a decision. Hold Fund A's twelve returns fixed and walk the target upward. Against a target of zero the Sortino ratio is 17.9021. Against 3 percent a year it is 9.7199. Against 4.20 percent, our headline case, it is 7.7518. Against 6 percent a year it is 5.5199, against 12 percent it is 1.2944, and against 18 percent it turns negative at −0.7195. One series, one denominator convention, and a number that spans from eighteen to below zero. Two things move as the target rises: the numerator shrinks directly, and the denominator grows because more periods fall below the line and the existing shortfalls get deeper.

This is an argument for choosing the target on purpose rather than accepting a default. A target of zero asks whether the fund ever lost money and is the right choice for capital preservation. A target equal to the risk-free rate reproduces the question Sharpe asks and makes the two ratios comparable in intent. A target equal to inflation asks whether purchasing power was preserved. A target equal to what your liabilities require — a pension payout, a drawdown rate, a hurdle in a fee agreement — is the version of the question with actual consequences, and it is the version Sortino's own framing was built for. State the target every time you quote the number, in the same breath, because a Sortino ratio without its target is not a weakly specified figure, it is an unspecified one.

Where the ratio degenerates, and what to do about it

The failure modes are all failures of the denominator, and they are worth recognising because they look like exceptional performance. If no period fell below the target, the sum of squared shortfalls is zero, the downside deviation is zero, and the ratio is infinite. That is not a fund with unlimited risk-adjusted return; it is a sample that contains no information about downside risk, and the honest report is the count of periods and the fact that none of them missed, not a number. If exactly one period fell below, the ratio is finite but rests entirely on that one observation: our example of eleven months around 1.2 percent and one month at −9.0 percent gives an annualised downside deviation of 9.3500 percent under the all-periods convention and 32.3894 percent under the other, a factor of 3.4641 on a single data point. Small samples do this to every risk measure, but downside measures suffer worse because they discard most of the sample by construction.

Two habits fix most of it. First, report the count of below-target periods next to the ratio; it is one extra number and it tells a reader immediately whether the denominator rests on three observations or thirty, and it also lets them convert between the two conventions themselves by multiplying or dividing by the square root of the total over that count. Second, treat the annualisation as the assumption it is. Multiplying a monthly deviation by the square root of twelve assumes the periods are independent and identically distributed, which is exactly the assumption that fails for the strategies people most often measure with Sortino — anything with autocorrelated returns, illiquid marks or a short volatility profile will have its risk understated by that scaling. The ratio is a comparison tool between series measured the same way over the same window, not a physical constant, and it should be quoted with its target, its convention, its frequency and its sample length attached.

Frequently asked questions

Which convention does this calculator use, and how do I convert to the other one?

It divides the sum of squared shortfalls by the number of periods in the series, not by the number of below-target periods, so the downside deviation it reports is the square root of a second-order lower partial moment about your target. Converting is one multiplication. To get the below-target-only figure, multiply the reported downside deviation by the square root of the total number of periods divided by the number that fell below the target — equivalently, divide the reported Sortino ratio by that same square root. With three shortfalls in twelve periods the factor is two; with two in twelve it is 2.4495; with one in twelve it is 3.4641. The calculator does not report the count of below-target periods directly, but it is easy to obtain from your own series, and it is the number you should record alongside the ratio for exactly this reason.

Is a high Sortino ratio better evidence than a high Sharpe ratio?

It is better evidence about a narrower question, and worse evidence about robustness. Sortino is the right instrument when the return distribution is genuinely asymmetric and you do not want to be told that a large gain is a risk — the case for it is real and our Fund B demonstrates it. But it is estimated from fewer observations than the Sharpe ratio, because it throws away every period above the target, so it has a wider sampling error on the same data and is easier to flatter with a short window. It is also easier to game: any strategy that sells insurance — writing options, carry trades, credit risk, illiquid marks — produces a long run of small gains and rare large losses, which is exactly the shape that scores well on Sortino right up until the rare loss arrives. The honest use is to read them together with the count of below-target periods and the worst single period, and to be suspicious rather than impressed when they disagree in the direction that favours the manager.

How does the Treynor ratio fit alongside these two?

It answers the same shape of question with a third denominator, and the choice between the three is really a choice about what you consider risk to be. Sharpe divides the excess return by total volatility, so it treats every wobble as risk. Sortino divides it by downside deviation about a target, so it treats only disappointment as risk. Treynor divides it by beta, so it treats only exposure to the market as risk and implicitly assumes everything else has been diversified away. That assumption is the tell: Treynor is the right measure for a component inside an already-diversified portfolio, where the specific risk genuinely does cancel against the other holdings, and the wrong measure for a standalone holding or a concentrated fund, where the risk that Treynor discards is the risk you are actually carrying. One practical consequence follows for anyone ranking managers: a fund that scores well on Treynor and badly on Sortino is telling you its problems are not market-driven, and that is precisely the case where the Treynor number is least informative about what you will experience.

How many periods do I need before the number means anything?

Ask the question about the denominator rather than about the sample, because the denominator is where the estimate is thin. What matters is not how many periods you have but how many of them fell below the target, since those are the only ones carrying information about downside risk. Twelve monthly observations with three shortfalls give you a denominator estimated from three numbers, and one different month would move it substantially. Thirty-six monthly observations with ten shortfalls are a meaningfully different proposition. There is no threshold that turns a bad estimate into a good one, but there is a discipline that helps: compute the ratio over several overlapping windows and look at the spread rather than at the point estimate, and drop the target down to zero as a cross-check, because a fund whose ratio collapses when the target moves a little was never being measured, it was being flattered by the position of the line.