Analyze a backtest

Drop a CSV with a 'return' column (QuantConnect, MT4/5, pandas export). Full verdict in seconds.

Drop your returns CSV here, or click to choose one

Statistical Methodology: Why the Classical Sharpe Ratio Is Not Enough

A backtest displaying a high annualized Sharpe Ratio (> 2.0) is frequently the artifact of selection bias (multiple testing / data-snooping) or parameter overfitting on a finite historical sample. The TrueVerdikt auditing engine subjects your return series to three institutional econometric filters:

1. Deflated Sharpe Ratio (DSR — Bailey & López de Prado, 2014)

DSR = Φ( ((SR̂ − SR₀) · √(T − 1)) / √(1 − γ₃·SR̂ + ((γ₄ − 1)/4)·SR̂²) )

Adjusts the observed Sharpe Ratio (SR̂) for the number of independent trials tested (N), return skewness (γ₃), excess kurtosis / fat tails (γ₄), and sample length T.

2. Combinatorial Purged Cross-Validation (CPCV)

PBO = Prob(SR_OOS < Median | Optimal In-Sample)

Partitions the return series into training and Out-of-Sample (OOS) folds while purging overlapping serial observations to prevent look-ahead and serial correlation leakage.

3. Fractional Kelly Sizing & 95% Conditional Value-at-Risk (CVaR)

f* = ½ · (μ / σ²) | CVaR₉₅ = E[R | R ≤ VaR₉₅]

Computes a conservative half-Kelly theoretical capital allocation ceiling alongside Expected Shortfall (CVaR 95%) across the worst 5% of historical return observations.

Methodological Disclaimer (YMYL): This tool performs a purely statistical audit on the historical return series you upload. A 'GO' verdict indicates historical sample robustness against overfitting but never guarantees future live trading performance.

Frequently Asked Questions About Statistical Backtest Auditing

What is the Deflated Sharpe Ratio (DSR)?

Introduced by Bailey and López de Prado (2014), it adjusts the observed Sharpe ratio for the number of variants tested, skewness, fat tails and sample length, in order to limit selection bias.

What is combinatorial purged cross-validation (CPCV) used for?

It splits the series into training and out-of-sample test blocks while purging overlapping observations, which removes information leakage and allows estimating the probability of backtest overfitting (PBO).

Does a “GO” verdict guarantee future performance?

No. It indicates that the historical sample provided is statistically robust against overfitting. It guarantees no future performance and is not financial advice.

Must my returns follow a specific format?

Provide a numeric series of periodic returns. A short or irregular series reduces the reliability of the measures; the longer the sample, the more informative the audit.

Quant Lab Hub

Deepen your statistical strategy audit with our other Quant Lab engines and quantitative validation research guides.