Aller au contenu

Sharpe, Sortino, drawdown

Judging performance against the risk taken.

10 min read · Advanced

“I made +40% this year.” Well done — but at what risk? A raw return says nothing without its cost in risk. Three measures let you judge a performance honestly.

The Sharpe ratio: return per unit of risk

Sharpe = (Portfolio return − Risk-free rate) / Volatility

Above 1, the return/risk pair is considered good; above 2, excellent.

Two portfolios both up 12% are not worth the same: the one that achieved it with 10% volatility is far superior to the one that endured 30% of turbulence. The Sharpe ratio puts everyone on an equal footing.

Sortino: punishing only bad volatility

The Sharpe ratio's flaw: it penalises volatility on the upside as well as the downside. Yet nobody complains about a surge upwards. Sortino keeps only downside volatility, which is closer to what an investor actually feels.

Max drawdown: the most honest measure

Maximum drawdown measures the worst fall endured between a peak and the following trough. It averages nothing and smooths nothing: it shows the worst moment lived through. It is often the most useful figure, because it is the one that makes people panic-sell.

sommetcreuxmax drawdownLa pire perte subie entre un sommet et le creux suivant.
Max drawdown: the maximum pain between a peak and the trough that follows.

The cruel mathematics of losses

A 50% loss requires +100% to get back to breakeven. An 80% loss requires +400%. That is why limiting drawdown often matters more than maximising return.

In Earnnest

The portfolio's Risk tab computes Sharpe, annualised volatility and max drawdown over the period of your choice.

À retenir

  • ✓A return without a measure of risk is incomplete information.
  • ✓Sharpe = return/risk; Sortino = penalises only the downside; drawdown = the worst real pain.
  • ✓Deep losses are asymmetric: −50% requires +100% to recover.

Educational content for information only: neither investment advice nor a personal recommendation. Past performance does not predict future performance.