Stock Market · Trading Systems

Walk-forward testing

Walk-forward testing is the gold standard for evaluating systematic strategies. Instead of optimising on the entire history and hoping it works in the future, you simulate the entire process of optimise → trade → re-optimise.

How walk-forward works

1. Use 2018-2020 data to optimise parameters

2. Trade those parameters on 2021 data (out-of-sample)

3. Now re-optimise on 2019-2021 data

4. Trade those parameters on 2022 data

5. Continue rolling forward...

Each year's out-of-sample performance is what you would have actually achieved if you re-optimised periodically and traded forward.

> A strategy that survives walk-forward testing has a much higher probability of working in live trading than one that only looks good in static backtests.

What walk-forward reveals

• Does the strategy still work as market conditions change?
• Are the optimal parameters stable, or do they jump around?
• Is the out-of-sample CAGR similar to the in-sample CAGR?

Stability of parameters

If your optimal parameter is 14-day MA in 2020, 9-day in 2021, 22-day in 2022. Your edge is unstable. The market regime is changing, and your strategy is curve-fitting to each regime.

If optimal parameters cluster (always 12-16 day range), your edge is probably real.

Tools

• Amibroker: built-in walk-forward optimisation
• Python (backtesting.py, vectorbt): full programmable walk-forward
• TradingView: limited walk-forward, mostly manual

Realistic numbers

A strategy that looks profitable in static backtests AND maintains its edge under walk-forward testing is genuinely rare. Most strategies fail walk-forward. That's a feature, not a bug. It filters out curve-fitting before you risk real money.

Takeaway. Walk-forward testing simulates the full optimise-and-trade cycle: optimise on past, trade on future, repeat rolling forward. Surviving walk-forward is the strongest evidence that a strategy has real edge.