Sharpe vs. Sortino vs. Calmar: Which Risk Metric Actually Fits Your Strategy?

Most investors reach for Sharpe first because it is the cleanest number on the page. That is also why it misleads so often. Sharpe treats a 10% upside swing and a 10% downside swing as equally bad, even though only one of them should keep you awake at night ^[1].

Sortino and Calmar fix different parts of that problem. Sortino ignores upside volatility and focuses on downside deviation ^[2]. Calmar ignores day-to-day noise and goes straight for the throat: the worst peak-to-trough loss ^[3]. If you compare a stock index, a 60/40 portfolio, and a trend-following CTA with the same metric, you are often ranking the wrong thing. For a broader framework on how to judge a strategy, see the benchmarking problem and why drawdowns matter more than returns.

Sharpe, Sortino, and Calmar measure three different kinds of pain

The formulas are simple, but the interpretation is not.

Sharpe was introduced by William Sharpe in 1966 as a reward-to-variability measure ^[1]. Sortino came later because investors kept objecting to the obvious flaw: not all volatility is bad ^[2]. Calmar, named after the California Managed Accounts Report, became popular because managers and allocators cared less about variance than about the size of the hole they might have to climb out of ^[3].

That distinction matters. A strategy with frequent small gains and rare large losses can look fine on Sharpe until the tail event arrives. A strategy with lumpy upside, like trend following, can look mediocre on Sharpe because it spends long stretches doing nothing and then jumps when markets break ^[4]. If you are studying systematic strategies, pair this with the backtest checklist and point-in-time backtesting; a clean metric cannot rescue dirty data.

The exact formulas, with the assumptions people skip

Here are the standard definitions used in practice.

Sharpe’s denominator is standard deviation. That is elegant, and also blunt. It treats a strategy with frequent small gains and occasional crashes the same way it treats a strategy with noisy but harmless upside. Sortino narrows the lens by using downside deviation around a target return, usually zero or the risk-free rate ^[2]. Calmar goes one step further and asks a more human question: how bad was the worst stretch? ^[3]

The catch is that each metric smuggles in a different philosophy. Sharpe says volatility is the enemy. Sortino says only bad volatility is the enemy. Calmar says the path matters more than the distribution. None is universally right. Most investors overread whichever one flatters their favorite strategy.

For a broader discussion of volatility itself, see volatility measurement and volatility targeting in stock research.

Worked example: SPY, 60/40, and a trend-following CTA do not rank the same way

To make the comparison concrete, use three real return streams from widely studied proxies: SPY for U.S. equities, a 60/40 stock-bond mix, and a managed-futures/trend-following CTA proxy. The exact numbers below are AIBROKER analysis using monthly total returns from publicly available ETF proxies and index series over 2010-2024, with a 3-month T-bill proxy as the risk-free rate. This is illustrative, not audited performance. Methodology details matter; see backtesting pitfalls and survivorship bias.

The ranking is not stable. SPY often wins on raw return. The 60/40 mix usually looks smoother. The CTA proxy often looks best on Calmar because it tends to cut losses faster in crisis periods, even if its monthly path is choppier and its upside arrives in bursts ^[4]. That is exactly why a single metric can be a trap.

Here is the worked logic. Suppose a strategy earns 8.5% annualized, has 9.0% annualized volatility, 6.0% downside deviation, and a 7.7% maximum drawdown. Its Sharpe is roughly 0.94, its Sortino is about 1.42, and its Calmar is about 1.10. Change only the drawdown to 20%, and Calmar collapses to 0.43 while Sharpe barely moves. That is not a rounding error. It is a different story.

Trend-following often looks better on Calmar than on Sharpe because its edge shows up in crisis convexity, not in smooth month-to-month gains. If you judge it like a bond fund, you will miss the point.

Why Sharpe breaks on lottery-like and negatively skewed strategies

Sharpe is most fragile when returns are not close to normal. That is not a niche problem. Many option-selling, carry, and short-volatility strategies produce a stream of small gains punctuated by rare, ugly losses. They can post attractive Sharpe ratios for years and then blow up in one month ^[5][9].

Lopez de Prado’s deflated Sharpe ratio was designed for exactly this environment: many trials, non-normal returns, and selection bias from testing lots of variants until one looks good ^[5]. The point is not that Sharpe is useless. The point is that a high Sharpe from a backtest can be mostly a mirage if you searched enough parameter combinations.

The uncomfortable implication is simple: a strategy can be “good” on Sharpe and still be a terrible thing to own. If the left tail is fat enough, the average return per unit of volatility is a poor summary of the actual risk you bear. That is why a serious backtest review should also check skew, kurtosis, drawdown depth, and the number of independent trials. A metric is not a due-diligence process.

If you want the broader context for this problem, read overfitting and walk-forward analysis.

Sortino helps, but the target return can move the goalposts

Sortino is often the better metric when upside volatility should not count as risk. That is true for many long-only investors. It is also why Sortino tends to favor strategies with asymmetric upside, including trend following and some quality or momentum approaches ^[2][4].

But Sortino has a built-in judgment call: the target return T. Set T to zero, and you get one answer. Set it to the risk-free rate, and you get another. Set it to a required hurdle, and the ratio changes again. That is not a bug. It is the metric admitting that “bad” depends on the investor. It also means two analysts can quote different Sortinos for the same strategy without either one lying.

That flexibility is useful, but it is also the trap. Sortino is not objective in the way Sharpe pretends to be. It is conditional on the target you choose. In practice, that makes it more honest than Sharpe for some strategies and less comparable across strategies. If you are building a systematic process, document the target in your research notes. Otherwise the ratio is just a number with a costume on.

For investors who care about process as much as outcome, systematic vs. discretionary and three numbers that matter are useful companions.

Calmar is the metric allocators use when they care about the path, not the spreadsheet

Calmar is brutally intuitive. Divide annualized return by maximum drawdown. If a strategy makes 12% a year but once lost 24% from peak to trough, its Calmar is 0.50. If another strategy makes 8% a year and its worst drawdown is 8%, its Calmar is 1.00. The second strategy may be less exciting. It is often easier to own.

That is why Calmar shows up so often in managed-futures and hedge-fund conversations. Investors do not just buy return. They buy a path they can survive. Young’s original work on the Calmar ratio was aimed at evaluating managed accounts, where drawdown tolerance is a real business constraint ^[3].

Still, Calmar has a blind spot. It only cares about the single worst drawdown. A strategy with many medium-sized losses can look better than one with one large loss, even if the first is more exhausting to hold. It also ignores the duration of the drawdown. A 15% drawdown that lasts three months is not the same as a 15% drawdown that lasts three years. The ratio does not know the difference.

Most investors underweight this point. They obsess over variance because it is easy to compute. They should care more about whether they can actually hold the strategy through the worst stretch. That is not a philosophical preference. It is a survival test.

For a deeper look at the mechanics of drawdowns, see drawdowns: why they matter more than returns and Sharpe vs. Calmar.

A decision tree for choosing the metric that matches your strategy

Use the metric that matches the shape of the return stream, not the one that flatters the backtest.

1. If returns are roughly symmetric and you mainly want a quick comparison: start with Sharpe.
2. If upside volatility should not count as risk: use Sortino, but write down the target return.
3. If the worst drawdown is the real constraint: use Calmar, then inspect drawdown duration separately.
4. If the strategy is heavily optimized or tested many times: adjust Sharpe with deflation or at least discount it aggressively ^[5].
5. If the strategy has fat tails, short-vol exposure, or crash risk: do not rely on Sharpe alone.

That decision tree is not academic. It changes what you buy, what you size, and what you can tolerate. A trend-following CTA may deserve a lower Sharpe than a bond proxy if it delivers crisis protection and a better Calmar. A short-vol strategy may deserve a high Sharpe discount if the left tail is doing all the work. A 60/40 portfolio may look boring on every metric and still be the right answer for a liability-matching investor.

Worked checklist for a backtest review:

• Compute all three ratios on the same return series.
• Check skew and maximum drawdown alongside Sharpe.
• State the risk-free rate and Sortino target explicitly.
• Look for parameter mining and multiple testing.
• Compare the strategy to a real benchmark, not a fantasy one.

If you need a broader framework for comparing strategies, the benchmarking problem and the momentum premium are good next stops.

So which metric should you trust first?

Trust none of them first. Then trust the one that matches the failure mode you care about most. That is the honest answer.

Sharpe is the fastest screen, but it is the easiest to game with a smooth-looking backtest. Sortino is better when upside noise should be ignored, but it depends on a target that can be moved around. Calmar is the most emotionally realistic for many allocators because it speaks the language of drawdowns, but it can miss the grind of long, shallow losses.

For a strategy with near-normal returns, Sharpe is usually fine as a first pass. For a strategy with asymmetric upside, Sortino is often more informative. For a strategy where capital preservation and investor behavior matter most, Calmar is the number that tells you whether the ride is survivable. The best research process uses all three, then asks which one is being flattered by the data.

That is the uncomfortable implication. A single “best” risk-adjusted metric does not exist. There is only a best metric for a specific return stream, investor, and failure mode.