Walk-Forward Analysis (WFA)

Walk-Forward Analysis is the gold standard for testing whether optimized parameters generalize to unseen data. It splits the date range into rolling windows and alternates between optimization and validation.

Key Concepts

Term	Meaning
In-Sample (IS)	The training window — parameters are optimized here
Out-of-Sample (OOS)	The testing window — parameters are validated on unseen data
Fold	One IS + OOS pair
Step Size	How many candles the window advances between folds
WFE	Walk-Forward Efficiency — ratio of OOS to IS performance

How It Works

              Fold 1                    Fold 2                    Fold 3
  ├────── IS ──────┤── OOS ──┤
              ├────── IS ──────┤── OOS ──┤
                          ├────── IS ──────┤── OOS ──┤

The date range is divided into overlapping folds
For each fold: a. Optimize on the IS window (full grid search) b. Backtest the IS window with the best parameters → IS metrics c. Backtest the OOS window with the same parameters → OOS metrics
Compare IS vs OOS performance across all folds to measure robustness

Configuration

Setting	Description
In-Sample Period	Length of the training window (days or candles)
Out-of-Sample Period	Length of the test window
Step Size	How far to advance between folds (defaults to OOS size)
Parameter Ranges	Same min/max/step as Optimization
Optimization Target	Metric used to select best parameters per fold

Parameter Policy

Mode	Behaviour
Optimize (default)	Re-optimize parameters for each fold's IS window
Fixed	Use one set of parameters for all folds (tests robustness of a known configuration)

Results

Each fold reports IS and OOS metrics independently. The summary includes:

Average OOS return and Sharpe ratio across all folds
Win rate — fraction of folds with positive OOS return
Consistency — how stable OOS returns are across folds (higher is better)
Robustness rating — Excellent, Good, Fair, or Poor

Robustness Rating

Rating	Criteria
Excellent	OOS win rate >= 70% AND consistency >= 0.6
Good	OOS win rate >= 60% AND consistency >= 0.5
Fair	OOS win rate >= 50% AND consistency >= 0.4
Poor	Below Fair thresholds

Overfitting Metrics

Metric	What It Measures
Walk-Forward Efficiency (WFE)	`OOS annualised return / IS annualised return`. Values near 1.0 mean the strategy performs similarly on unseen data. Values far below 1.0 suggest overfitting.
Performance Drop	Percentage of IS performance lost in OOS
Sharpe Ratio Drop	Difference between IS and OOS Sharpe ratios
Win Rate Drop	Difference between IS and OOS win rates

Tips

A strategy rated Good or Excellent is far more trustworthy than one that only looks great in a single backtest
If WFE is below 0.5, the optimized parameters are likely overfitting to historical noise
Use IS percentages between 60–80% of the total window — too small won't optimize well, too large won't leave enough OOS data
Run WFA on the same date range you optimized on to verify the optimization was not a fluke
If results are Poor, simplify the strategy (fewer parameters, longer indicator periods) before tuning further