When you’re building algorithmic trading strategies—especially for passing prop firm challenges—how you handle your backtesting data is crucial.
Using StrategyQuant, one of the most powerful tools for strategy development, you’ll need to split your data into In-Sample and Out-of-Sample (OOS) periods to make sure your system isn’t overfitted and stands a real chance of performing well in live markets.
Why Split Your Data?
If you only train your strategy on a single block of historical data (In-Sample), it may “memorize” patterns from that period and look great on paper—but fall apart in real-world trading.
To avoid this, you must include Out-of-Sample data: a separate chunk of unseen data to test how your strategy behaves in new, unseen market conditions.
Recommended Data Setup for Forex in StrategyQuant
Most forex pairs in StrategyQuant have data going back to around 2003. Here’s a solid structure for splitting your data:
🔧 Building & Training Phase
- Training Period (In-Sample):
2009/January – 2022/January
This is where your strategy is actually being built. You want a solid number of years here to catch different market cycles and avoid curve-fitting. - Validation / Out-of-Sample Period:
2018 – 2022
StrategyQuant will reserve this part to test how well your strategy generalizes to new data it hasn’t seen during the build process.
📊 Strategy Filtering & Ranking Criteria
To make sure only high-quality strategies make it through the build and test phases, use the following ranking and filtering rules during your workflow:
🔨 During the Building Stage (StrategyQuant Builder Filters)
Forex Build Filtering Criteria:
- Number of Trades (In-Sample): > 200
- Average Trades Per Month (In-Sample): > 2
- Profit Factor (In-Sample): > 1.3
- Return/Drawdown Ratio: > 5
These filters ensure that your strategy has enough trade frequency, strong profitability, and solid risk-adjusted returns during the build process.
🔍 Post-Build Testing
Once your strategy is built and passes the OOS validation inside StrategyQuant, you can test it even further using different periods:
- Out-of-Sample Retest #1:
2022/January – 2024/January
This simulates how your strategy might have performed in very recent conditions, like the post-COVID recovery, inflation volatility, etc. - Out-of-Sample Retest #2:
2003/January – 2008/December
This older data shows how your strategy performs in a completely different market environment—back when spreads were wider and conditions were less algorithmic. - Final Walk-Forward or Real-Time Test:
Use any remaining data (or re-test on live/demo forward testing) to see if the system holds up outside of historical testing.
✅ Out-of-Sample Validation (Inside StrategyQuant or External Retests)
OOS Filtering Criteria:
- Profit Factor (Out-of-Sample): > 1.0
This ensures the strategy performs decently on unseen data during validation, indicating it’s not just curve-fitted to the build sample.
🌐 Market Robustness Testing (Across Other Forex Pairs)
Cross-Market Testing Criteria:
- Profit Factor: > 1.1
This is a bonus test to check how the strategy performs on similar Forex pairs. While not required, it gives you insight into how adaptable your system is in various market environments.
Final Tips
✅ The most recent data (e.g., 2022–2024) is the most relevant for prop firms, especially those like FTMO or 5%ers, which are judging you on current market behavior.
✅ The oldest data is still useful—it helps test long-term robustness, but don’t expect it to perfectly match today’s markets.
✅ Keep your testing honest. Don’t tweak the strategy after seeing OOS results unless you completely reset your test structure.
Summary
Proper data splitting isn’t just a nerdy detail—it’s the foundation of whether your strategy will thrive or die in live trading. If you’re building a system to pass a prop firm challenge, this level of discipline could be what sets you apart from everyone else.
Want more help building your own quant strategies for Forex?
👉 Check out more blogs in the future for more guides and walkthroughs.