Rearview Download

Is my backtest luck?

Try enough variations of a strategy on the same history and one will look good by chance. Enter your backtest’s figures below to compute the Deflated Sharpe Ratio (Bailey & López de Prado, 2014) right in your browser; nothing leaves your device.

Deflated Sharpe Ratio calculator

Your backtest’s figures

Adjust the numbers below; the verdict updates as you type.

Annualised Sharpe of the best run you picked.
Every backtest, grid trial and tweak you tried.
Length of the backtest in years.
Converts between annual and daily Sharpe.
Distribution & trial assumptions
Std. dev. of annual Sharpe across your trials.
0 = unrelated ideas; 0.5–0.8 = nearby params.
0 for normal returns; negative for crash-prone tails.
3 for normal returns; higher for fat tails.

Deflated Sharpe verdict

—

Deflated Sharpe Ratio (DSR)
—
Luck’s best (expected max Sharpe)
—
Effective independent trials (Neff)
—
Return observations (T)
—
Normal score (z)
—

Uses Rearview’s exact thresholds: Probably not luck at DSR ≥ 95%, Probably luck under 50%, and Can’t tell in between. How Rearview computes it ›

Counting every try is the hard part.

Most traders underestimate N by an order of magnitude, forgetting last week’s parameter tweaks, deleted runs, or the 150 variations of a script an AI coding agent tried before handing over a winner. And Deflated Sharpe alone can’t tell whether the winning parameter holds up out of sample.

Rearview logs every run, edit preview, search trial and rvw run --project run into a chained tries log automatically, pairs Deflated Sharpe with the Probability of Backtest Overfitting (PBO), and lets you seal the latest years in a lockbox until your research is done.

What the inputs mean

  • Variations tried (N): every setting or rule variation you looked at on the same history. Even deleted runs count, because seeing them shaped what you kept.
  • Average correlation (ρ): nearby parameter values (say RSI 8, 10, 12) move together, so Rearview shrinks N to Neff = ρ + (1 − ρ)N. Unrelated ideas have ρ near 0.
  • Spread of trial Sharpes (σ): how widely the trials’ Sharpes scattered. Wider scatter raises luck’s bar (SR0).
  • Skew and kurtosis: negative skew and fat tails (kurtosis > 3) make a high Sharpe ratio less reliable over a finite sample.