Cross-Validation Strategy Analysis
Evaluate a trading strategy's temporal consistency by splitting historical trades into multiple folds and measuring performance in each held-out segment. Unlike a simple backtest, cross-validation reveals whether results are driven by a few lucky periods or represent stable, repeatable edge across the full trade history.
K-Fold Time-Series
Purged K-Fold
Blocked K-Fold
Leakage Detection
Strategy to validate. Data uses only closed simulator trades linked to the selected strategy.
Cross-Validation Method
Time-Series Split: each fold tests on the next chronological block. Purged: adds an embargo gap around fold edges to prevent leakage. Blocked: each period is tested independently.
How many data splits to create. K=5 is standard. Higher K = smaller test folds per split. Must be < number of trades.
Evaluate: tests each fold at fixed scale 1.0 to measure strategy consistency. Optimise: grid-searches the best position scale on training folds.
Only used with Purged K-Fold. Trades removed from fold boundaries to prevent information leakage. Try 2–5 for most strategies.
Scale Grid (only used when Optimise mode is selected)
Smallest position scale tested on training folds. 0.5 = half size.
Largest position scale tested on training folds. Keep conservative to avoid overfitting.
Increment between tested scales. Smaller = finer grid, slower. 0.25 is a good balance.
Cost Model
Starting balance for each fold's equity simulation. Use your actual account size.
Flat commission deducted from each trade's P&L. Enter zero if no flat fee applies.
Execution slippage per trade. 1 basis point = 0.01% of trade value. 2–5 bps is typical.
Waiting for simulation input.
Per-Fold Test Return
Return achieved in each held-out test fold. Consistent positive bars indicate genuine edge; a large positive/negative mix signals high variance or data-snooping bias.
Positive Return
Negative Return
Mean Return
Per-Fold Sharpe Ratio
Risk-adjusted performance per fold. High variability suggests regime-dependent returns. Mean Sharpe >0.5 with low dispersion is a strong stability signal.
Fold Sharpe
Mean Sharpe
Fold-by-Fold Detail
Full per-fold breakdown of training and test metrics. Compare scale, returns, and risk metrics across folds to identify unstable periods or regime-specific performance.
| Fold # | Method | Train Trades | Test Trades | Scale Used | Test Return | Test Sharpe | Test Max DD | Test Win % | Test Profit Factor | Consistency Flag |
|---|---|---|---|---|---|---|---|---|---|---|
| Run the cross-validation engine to see fold results. | ||||||||||
K-Fold Time-Series Split
Trades are sorted chronologically and split into K blocks. Each fold uses all prior blocks as training data and the current block as the test set. Preserves temporal ordering and simulates progressively more data becoming available. Produces K−1 test evaluations.
Purged K-Fold
K equal blocks are created. For each test fold, the training set consists of all other folds with an embargo gap (N trades) removed from both sides of the test boundary. Prevents information leakage caused by overlapping holding periods or correlated trades near fold edges.
Blocked K-Fold
The full trade history is split into K contiguous blocks. Each block is held out once as the test set while the remaining K−1 blocks form the training set. Useful when regime changes are suspected, as it tests each time period independently without strict temporal-ordering constraints.