> ## Content Index > Fetch the complete content index at: https://quantdojo.ai/llms.txt > Use this file to discover other available public pages before exploring further. # Sharpe Ratio vs Sortino Ratio vs Calmar Ratio: Which Risk-Adjusted Metric Should You Actually Optimize For? - URL: https://quantdojo.ai/sharpe-sortino-calmar-ratio-which-to-optimize/ - Published: 2026-09-04T09:52:38.000Z - Updated: 2026-09-04T09:52:38.000Z - Description: Three metrics, three different definitions of risk — and optimizing for the wrong one can make a losing strategy look brilliant. Here's which one actually matters for your strategy. - Author: Wolfgang Lämmle - Tags: metrics, backtesting, risk management, strategy validation, Sharpe Ratio, Sortino ratio, Calmar ratio Most traders optimize for the Sharpe ratio because it's the default output in every backtest engine. That's a mistake — and not a subtle one. Sharpe, Sortino, and Calmar each embed a different assumption about what "risk" means. Optimize for the wrong one and you're not finding the best strategy — you're finding the strategy that scores best on an irrelevant test. Worse, each metric has a specific failure mode that backtesters accidentally — or deliberately — exploit. Here's what the math actually says, where each metric breaks, and how to pick the right one for your strategy type. ## What Each Metric Is Actually Measuring Before comparing them, nail down what each ratio computes. The differences are not cosmetic. ### Sharpe Ratio The Sharpe ratio is excess return divided by total return volatility — where volatility is the standard deviation of all periodic returns (daily, weekly, whatever your bar is). **What it assumes:** returns are normally distributed, and all volatility is bad. A strategy that posts consistent +3% months interrupted by a +15% outlier month is *penalized* for that outlier the same way it would be penalized for a -15% month. That assumption is fine for assets that roughly follow a normal distribution. For trading strategies — especially ones with skewed payoff profiles like options or breakout systems — it's actively misleading. **The gaming vector:** inflate Sharpe by smoothing daily P&L. Strategies that mark illiquid positions to a stale mid-price show artificially low volatility. So do monthly-reporting hedge funds with side-pocketed assets. In a backtest, overly optimistic fill assumptions can do the same thing — a point worth cross-referencing with [how slippage quietly inflates backtest results](https://quantdojo.ai/slippage-backtest-model-execution-costs/). ### Sortino Ratio The Sortino ratio replaces the denominator with *downside deviation* — the standard deviation computed only on returns that fall below a target (usually zero or the risk-free rate). **What it assumes:** upside volatility is not risk. A blowout winning month should not count against you the same way a blowout losing month does. This is intuitively correct for most traders. The Sortino ratio will always be >= the Sharpe ratio for a strategy with any positive skew. Which means it will almost always look better — and that's exactly the problem when vendors cherry-pick which metric to report. **The gaming vector:** the fewer large down-returns you have, the lower the downside deviation — and the higher your Sortino. Strategies that cut losers very aggressively (tight stops, small loss caps) can produce a high Sortino while concealing that they have a terrible win/loss payout ratio. That's the same trap as [a 90% win rate that still loses money](https://quantdojo.ai/a-90-win-rate-that-loses-money-the-win-rate-trap/) — just expressed in ratio form. ### Calmar Ratio The Calmar ratio is annualized return divided by maximum drawdown (stated as a positive number). It doesn't care about volatility at all. It only asks: for every dollar of peak-to-trough loss the strategy has ever inflicted, how much annual return did it generate? **What it assumes:** the number that matters most is the worst loss sequence a strategy has ever produced — because that's what kills accounts, breaches prop-firm limits, and causes traders to abandon systems at exactly the wrong moment. **The gaming vector:** max drawdown is a function of the sample period. A strategy backtested over a short, benign window will have a small max drawdown and therefore a sky-high Calmar — even if it would have been wiped out in a different regime. Extend the lookback or stress-test via [Monte Carlo simulation](https://quantdojo.ai/monte-carlo-simulator/) to see whether that Calmar holds up. ## How the Math Breaks Under Real Return Distributions Trading strategy returns are almost never normally distributed. They tend to have: - **Fat tails** — extreme losses (and gains) occur far more often than a normal distribution predicts. - **Negative skew** in many systematic strategies — lots of small wins, rare catastrophic losses. - **Positive skew** in breakout and trend-following strategies — many small losses, rare large wins. Here's how each metric handles those realities: #### Negative skew — short vol, carry, mean reversion Sharpe Overstates edge penalizes the skew but never sees how deep the tail goes Sortino Overstates edge downside deviation stays low right up until the tail hits Calmar Most honest the drawdown eventually contains the tail event #### Positive skew — trend following, breakout Sharpe Understates edge punishes the large winning outliers that are the whole point Sortino Most accurate rewards upside without charging for it Calmar Fair but harsh captures the drawdown, ignores the size of the winners #### Fat tails, no clear skew Sharpe Misleads models no tail probability at all Sortino Misleads same blind spot, one side only Calmar Least bad a realised drawdown is at least observable The practical conclusion: no single metric is universally correct. But the *choice* of metric should be driven by your strategy's return distribution, not by which number looks biggest. ## Which Metric Fits Which Strategy Type ### Mean-reversion and market-making strategies These strategies post frequent small wins and occasional sharp losses — negative skew by design. Sortino looks great on them right up until the inevitable gap or squeeze that the downside deviation never anticipated. **Use Calmar as the sanity check** and watch it degrade as the sample period includes a volatile regime. ### Trend-following and breakout strategies Positive skew, low win rate, long streaks of small losses interrupted by large wins. The Sharpe ratio punishes them for volatility they shouldn't be penalized for. **Sortino is the most appropriate primary metric** here, but back it up with Calmar to confirm the drawdowns are survivable. ### Options-selling and carry strategies The most dangerous case. These strategies can produce a Sharpe of 2+ and a Sortino that looks even better — until they don't. The tail event that brings them down is structurally underrepresented in any single backtest window. **Calmar is the most honest metric**, but even it will look fine until the blowup sample is included. Pair it with a [Monte Carlo analysis](https://quantdojo.ai/monte-carlo-simulation-trading-strategies-explained/) that stress-tests the tail. ### Prop-firm challenge strategies Prop firms enforce hard drawdown limits, not Sharpe-ratio targets. Optimizing for Sharpe when your actual constraint is a 10% max drawdown rule is a category error. **Calmar is the correct primary metric** — it directly mirrors the prop-firm loss constraint. ## The Deeper Problem: Optimizing for a Metric Is Not the Same as Finding Edge This is the part most explainers skip. When you run a parameter sweep and rank configurations by Sharpe — or Sortino, or Calmar — you are fitting that metric to in-sample data. The metric with the highest in-sample score will almost always look worse out-of-sample. That's not a property of any specific ratio. It's a property of optimization itself. The [probability of backtest overfitting](https://quantdojo.ai/probability-backtest-overfitting-pbo-cscv/) rises sharply as the number of parameter combinations tested increases — regardless of which objective function you used. A high Sortino achieved by testing 500 parameter sets is almost certainly noise. The [Deflated Sharpe Ratio](https://quantdojo.ai/deflated-sharpe-ratio-multiple-testing-noise/) formalizes exactly this problem: it adjusts a strategy's Sharpe for the number of trials it took to find it. The right workflow: 1. **Pick the metric that matches your strategy's return distribution** — before you run a single optimization. 2. **Optimize in-sample on that metric**, but hold out a genuine out-of-sample window. 3. **Verify the ratio degrades gracefully** out-of-sample — not catastrophically. 4. **Ask what the metric is hiding.** A high Sortino might be hiding a bad Calmar. A high Calmar might be hiding a short sample period. For an independent read on whether a strategy's backtest metrics are genuine, [run it through QuantCheck](https://quantdojo.ai/quantcheck/) — it stress-tests backtest results for overfitting, multiple testing bias, and regime sensitivity without requiring you to write a single line of code. And if you're evaluating someone else's strategy rather than your own, the [red flags in a vendor's backtest](https://quantdojo.ai/strategy-vendor-backtest-red-flags-checklist/) are almost always visible in the metric selection: vendors who report only Sortino, only on a cherry-picked date range, with no Calmar in sight, are telling you something. ## A Quick Note on Computing These Ratios Definitions vary by implementation and bar size. Annualization factors differ between daily, weekly, and monthly returns. Some platforms compute Sortino using a fixed minimum acceptable return; others use zero; others use the risk-free rate. Before comparing ratios across platforms or vendors, confirm the definition. For a consistent calculation you control, use the [Sharpe ratio calculator](https://quantdojo.ai/sharpe-ratio-calculator/) at QuantDojo — and cross-reference against your own implementation to catch platform discrepancies. ## Frequently Asked Questions ### Is a higher Sortino ratio always better than a higher Sharpe ratio? Not always — it depends on what you're comparing. For strategies with positive skew (breakout, trend-following), a higher Sortino relative to Sharpe is expected and meaningful. But a Sortino that is *much* higher than Sharpe in a mean-reversion strategy is often a warning sign: it may mean the downside deviation is artificially low because the sample period hasn't included the strategy's worst-case tail event yet. ### What is a good Calmar ratio for a trading strategy? Contextually, a Calmar above 1.0 means the strategy earns more annually than its worst historical drawdown — a reasonable minimum bar. Calmar ratios above 3.0 are exceptional and should be scrutinized carefully: they usually reflect a short or benign sample period, not genuine risk control. Always extend the backtest window and stress-test with Monte Carlo before trusting a high Calmar. ### Can you optimize for multiple ratios at the same time? Yes, through multi-objective optimization — Pareto frontiers that trade off Sortino against Calmar, for example. In practice, the simplest approach is to optimize on one primary metric (chosen to match your return distribution), then filter results by a minimum threshold on a second metric as a constraint. Optimizing on too many objectives simultaneously increases the risk of overfitting — you're adding more degrees of freedom to the search. ## The Honest Bottom Line The metric you optimize for defines what "good" means to your backtest engine. Choose wrong and you find strategies that score well on an irrelevant test — not strategies with real edge. Sharpe is the default. Sortino is better for skewed payoff profiles. Calmar is the most honest for drawdown-constrained accounts. None of them is a substitute for genuine out-of-sample validation. Pick the metric that matches your strategy's return distribution. Verify it degrades gracefully out-of-sample. And treat any ratio that looks too good as a hypothesis — not a result. *Nothing in this article is financial advice — it's a framework for asking better questions of your own backtest.*