Building a Trading System: Rules, Backtesting, Refinement
Building a Trading System: Rules, Backtesting, Refinement
Most traders have a collection of concepts they like. Fewer have a system. A system is a complete framework that could be handed to a stranger who has never traded before, and they would know exactly what to do in every scenario. If your system requires your judgment at any step beyond following the rules, it's not a system — it's a preference.
The Components of a Complete System
A complete trading system defines all of the following without ambiguity:
1. Setup Criteria: What exactly constitutes a valid setup? This should be a checklist, not a description. For an SMC system:
- HTF bias defined (daily bullish/bearish — exact structural criteria)
- 4H zone identified (OB or FVG within the discount/premium zone)
- Session timing (within killzone — specific hours)
- Liquidity sweep confirmed (Asian range sweep or equal high/low sweep)
- ChoCH confirmed on 15M
- Entry OB or FVG identified on 5M
- All items checked: valid. Any item missing: pass.
2. Entry Rules: Where exactly do you enter? Limit order at the OB body? Market order at ChoCH candle close? The answer must be specific enough to program.
3. Stop-Loss Rules: Where does the trade fail? Beyond the OB wick? Below the swing low of the sweep? One rule, consistently applied.
4. Target Rules: Where do you exit? Nearest HTF liquidity pool? Fixed R target (2R, 3R)? First FVG above entry? Again — one rule.
5. Risk Rules: Maximum % per trade, maximum trades per day, daily loss limit. Non-negotiable.
Backtesting Protocol
Backtesting requires rigor to be meaningful. The common mistake is cherry-picking — going through history and "finding" setups that would have worked. The test must be applied systematically:
Step 1: Define all setup criteria in writing before opening the chart.
Step 2: Scroll through the past 3–6 months on your chosen instrument.
Step 3: At each point where ALL setup criteria are met, record the trade.
Step 4: Apply entry, stop, and target rules exactly as written — no adjustments based on outcome.
Step 5: Record: Win/Loss, R gained/lost, entry quality score (1–5).
Step 6: After 50+ trades, calculate: Win Rate, Average R, Max Drawdown, Expectancy.
Target metrics for a viable system:
- Win rate: 45–65% (lower acceptable with high RRR)
- Average RRR: 1:2 minimum
- Expectancy: Positive
- Max drawdown in test period: <20%
If the system doesn't meet these on backtesting, refine the criteria — not the outcomes.
Avoiding Curve Fitting
Curve fitting is the #1 backtesting error. It happens when you add rules specifically because they would have filtered out the losses you just observed.
Example of curve fitting: Your backtest shows losses when certain news events are nearby. You add a rule "don't trade within 2 hours of news." This might improve backtest results but adds a rule that is specific to those historical losses — it won't generalize.
How to avoid it: Keep the system simple. SMC gives you enough filtering through structure, zone, and timing alignment. If you need more than 5–7 criteria to filter setups, the system is likely being over-optimized.
Robust systems have fewer rules, not more. Complexity is often disguised curve-fitting.
Forward Testing Before Going Live
Never jump from backtest to live trading with significant capital. The path is:
- Backtest (3–6 months of historical data, 50+ trades minimum)
- Forward test on demo (at least 30 trades in real-time, following all rules)
- Live micro-lot testing (real money, minimum size, 30+ trades)
- Scale up only after micro-lot results match backtest expectations within acceptable variance
The forward test often reveals execution issues that backtesting doesn't: spread impact on tight stops, slippage during news events, psychological difficulty following rules in real-time.
The Refinement Loop
A system is not built once and used forever. After every 50–100 live trades, run a performance review:
- Has win rate changed significantly from backtest?
- Are any setup types consistently underperforming?
- Are any setup types consistently outperforming?
If underperforming types exist, either eliminate them from the system or add one specific filter that addresses the observed weakness without curve-fitting the history.
Refinement is small, deliberate, and evidence-based. It is not the frantic rule-adding that happens after a losing week. The losing week is data; wait until the data is statistically meaningful (50+ trades) before drawing conclusions.