1
Training window
Use only prior observations to select the current system state.
Lesson · Backtesting and Trading Systems
Walk-Forward Testing
Repeat chronological training and test windows, choose parameters only from past data, and stitch the unseen test windows into one forward simulation.
Core concepts
What to understand
Keep each definition operational so it can be implemented and tested.
2
Forward window
Run the chosen state without modification on the next period.
3
Stitched result
Combine only forward-window outcomes for the final evaluation.
Interactive lab
Walk-Forward Analysis Lab
Change training and testing window lengths. Observe parameter stability, forward returns, and coverage across regimes.
Walk-Forward Analysis Lab
Deterministic synthetic educational data · no live market connection
BACKTEST MODEL
Process
Advance chronologically
Every fold preserves the direction of time.
Train
Compare allowed parameters using the historical window.
Freeze
Select one parameter using the documented rule.
Test and roll
Apply it to the next window, record results, then advance.
Knowledge check
Check your understanding
Select one answer for each question. Feedback appears immediately.
1. Which returns belong in the final walk-forward curve?
Training chooses parameters; forward windows estimate their next-period behavior.
2. A very short training window may produce:
Limited data can make the selected parameter highly sensitive to noise.
3. Walk-forward testing removes all overfitting risk:
It improves temporal validation but still depends on choices, data, costs, and repeated experimentation.
4. 350 bars, an 80-bar training window and a 30-bar forward window, advancing 30 bars each fold. How many forward windows are scored?
⌊(350 − 80 − 30) ÷ 30⌋ + 1 = ⌊240 ÷ 30⌋ + 1 = 9, and 9 × 30 forward bars cover every bar after the first training window exactly once.
Practical rules
- Use strictly chronological folds.
- Document parameter selection and tie-breaking.
- Evaluate stitched forward results, not training results.
Common mistakes
- Using future folds to choose earlier parameters.
- Reporting only successful windows.
- Changing the candidate set after seeing each forward result.
References
Official reading
The labs are simplified simulations. These sources cover historical performance, data, costs, or execution limitations.
FINRA · Smart Beta—What You Need to Know
Explains that historical backtesting is hypothetical and does not predict future performance.
CME Group · Historical Market Data
Describes historical market data used for market analysis and strategy validation.