Walk-Forward Analysis

Protect against overfitting by testing strategy on out-of-sample data. Each fold optimizes on train period and tests on unseen test period.

How Walk-Forward Works

1. Data is split into multiple folds (e.g., 4 folds)

2. Each fold has a train period (e.g., 75%) and test period (e.g., 25%)

3. Strategy parameters are optimized on train data

4. Best parameters are tested on out-of-sample test data

5. Results are aggregated across all test periods

|------ Train 1 ------|-- Test 1 --|
      |------ Train 2 ------|-- Test 2 --|
            |------ Train 3 ------|-- Test 3 --|
                  |------ Train 4 ------|-- Test 4 --|

Run Walk-Forward Analysis

Strategy

Symbols (comma-separated)

Start Date

End Date

Number of Folds

Train Percentage (%)

Mode

Rolling Anchored

Rolling: train window slides. Anchored: train always from start.

Optimization Method

Optimize For

Workers

Max Iterations (Random Search)

Initial Capital ($)

Understanding Results

Efficiency Ratio: Test Sharpe / Train Sharpe. Values > 0.7 indicate good generalization.
Parameter Stability: Low CV (< 0.3) means parameters are stable across folds.
WF vs Naive: Walk-forward typically shows lower but more realistic performance.