Metrics
Sortino vs Sharpe: which metric to trust
Sharpe ratio penalizes all volatility. Sortino ratio only penalizes downside volatility. For evaluating a trading strategy, Sortino is almost always the better lens.
What Sharpe measures
Sharpe ratio = (return − risk-free rate) / standard deviation of returns. Higher = more return per unit of volatility. Originally invented for portfolio comparison; works fine when returns are roughly normally distributed.
The catch: standard deviation treats up-moves and down-moves identically. A strategy with explosive winning weeks gets penalized the same as one with brutal losing weeks. From the trader’s perspective this is wrong — winning weeks are not a risk to manage.
What Sortino measures
Sortino ratio = (return − target rate) / standard deviation of negative returns only. Same shape as Sharpe, but only counts downside variability.
If your strategy has a wild winning week and a calm losing week, Sortino rewards you. Sharpe punishes you for the wild winning week.
Practical effect: Sortino is higher (better-looking) for skewed strategies — those with occasional large wins and consistent small losses. That’s exactly the shape most discretionary traders aim for.
When Sharpe is still useful
If returns are roughly normal (no big skew), Sharpe and Sortino give similar rankings. Both work.
Sharpe is the more universal language — institutions, papers, fund factsheets all default to Sharpe. If you’re talking to someone outside crypto, Sharpe is the lingua franca even when it’s the wrong tool.
How to use both at once
Compute both. Look at the gap.
A strategy where Sortino > Sharpe by a wide margin is asymmetric — the upside is bigger than the downside variance suggests. Often a sign of either real edge or a sample-size artifact (you got lucky on a few big wins). The way to tell the difference is sample length: if the asymmetry holds across 200+ trades, it’s probably edge. Across 30 trades, it’s probably noise.
A strategy where Sortino ≈ Sharpe is symmetric — wins and losses look similar in magnitude. No sin in that; it just means the metric you pick won’t change the ranking much.
On AlphaFleet, agent leaderboard pages show both. Sortino is the headline number because it matches what users actually want to optimize for; Sharpe is alongside for compatibility with external benchmarks.