TradingCalcDeflated Sharpe Ratio Calculator
RISK/STATS · DEFLATED SHARPE RATIO · ADVANCED
Deflated Sharpe Ratio (Bailey & Lopez de Prado): the Sharpe threshold the best of N tried strategy variants would clear by luck alone, and whether your result still clears it.
Comma, space, or newline separated. No thousand separators (write 43250.50, not 43,250.50).
Comma, space, or newline separated. No thousand separators (write 43250.50, not 43,250.50).
The rigorous path: uses each trial's own Sharpe ratio, not an assumption.
The problem this solves
Try enough strategy variants and one of them will show a great Sharpe ratio by pure chance, even with zero real skill. After 1,000 independent trials, the *best* one is expected to show a Sharpe of about 3.26 by luck alone. The Deflated Sharpe Ratio (DSR) asks: does your actual result still clear that higher, trial-adjusted bar?
The formula
Two steps, Bailey & Lopez de Prado (2014):
- E[max SR] = E[SR_n] + sqrt(V[SR_n]) × [(1-γ)·Z⁻¹(1-1/N) + γ·Z⁻¹(1-1/(N·e))]
- DSR = PSR(SR*_0 = E[max SR]) - the Probabilistic Sharpe Ratio evaluated against that threshold
γ ≈ 0.5772 (Euler-Mascheroni constant). Correlated trials overstate raw N, so an average-correlation input derives an effective trial count via Kish's design effect: N_eff = N / (1 + (N-1)·ρ).
Reading the result
- DSR ≥ 95%: your Sharpe clears the luck-alone bar with high confidence
- DSR around 50%: your Sharpe is roughly at the threshold the best of your trials would reach by chance
- DSR low: your result is consistent with pure selection bias, not genuine skill
Where to go next
DSR corrects for how many variants you tried. For the plain (non-deflated) Sharpe ratio and its own confidence interval, see the Sharpe ratio calculator, or get the full risk/return picture with the portfolio tearsheet.
Use via API or MCP
This calculation is available as a deterministic API call for bots and AI agents.
What is the Deflated Sharpe Ratio?
A correction for selection bias: it compares your strategy's Sharpe ratio not against zero, but against the Sharpe ratio the best of N tried variants would show by pure luck alone, then reports the probability your true Sharpe still exceeds that higher bar.
Why does the number of trials matter?
Trying many parameter combinations or variants and reporting only the best one inflates the apparent Sharpe ratio, even with zero real skill: after 1,000 independent trials, the best is expected to show a Sharpe of about 3.26 by chance alone.
What if my trials were correlated, not independent?
Correlated trials (e.g. small parameter tweaks on the same underlying strategy) overstate the effective number of independent tests. Supplying an average pairwise correlation derives a lower, more honest effective trial count via Kish's design effect.
Which input mode should I use?
If you tracked every trial's own Sharpe ratio, use that: it's the more rigorous path, using the trials' actual empirical mean and variance. If you only know how many you tried, use the trial-count mode with an average correlation estimate.
Is a low DSR proof my strategy has no edge?
It means your result is statistically consistent with pure selection bias given how many variants you tried, not proof of zero skill. A high DSR is stronger evidence the edge is real, not a guarantee it persists out of sample.