Backtesting Pattern Strategies

Backtesting Pattern Strategies

Back-testing applies your trading rules to historical data to measure how the system would have performed. Done rigorously, it validates your edge. Done poorly, it creates false confidence.

Why Back-Test?

Without it, your system is a hypothesis. Back-testing answers: Does it have positive expectancy? What is the win rate? Maximum drawdown? How many trades? Which conditions favor or damage it?

Manual vs. Automated

Manual: Scroll through charts, identify patterns, record results. Slow but extremely educational.

Automated: Code rules into TradingView, Python/Backtrader, or QuantConnect. Fast, covers more data, requires programming.

Start manual (100+ trades), then automate for larger validation.

The Process

Define rules completely -- every rule unambiguous and mechanical.

Select data -- 3-5 years minimum including different market conditions.

Split data -- 70% in-sample (development), 30% out-of-sample (validation). Never peek.

Apply rules to in-sample data chronologically.

Calculate metrics.

Validate on out-of-sample. Similar results = robust. Much worse = likely overfit.

Cardinal Sins

Overfitting: Optimizing rules to perfectly fit historical data. Signs: too many filters (>5-7), specific numbers without justification, in-sample dramatically better than out-of-sample. Prevention: keep rules simple, use round numbers, validate out-of-sample.

Survivorship Bias: Testing only currently-listed stocks. Prevention: use survivorship-bias-free databases.

Look-Ahead Bias: Using information unavailable at trade time. Prevention: only use data available at the decision point.

Ignoring Costs: Spreads, commissions, slippage consume 0.1-0.5% per trade, adding up over hundreds of trades.

Walk-Forward Analysis

The gold standard: multiple rolling in-sample/out-of-sample tests. Optimizes on 12 months, tests on next 3, rolls forward, repeats. Simulates periodic re-optimization.

Monte Carlo Simulation

Randomize trade order thousands of times to see the range of possible drawdowns, probability of profit targets, and worst-case scenarios.

Sample Size

30 trades = absolute minimum. 100 = reasonable. 200+ = good statistical significance. 500+ = excellent, can analyze sub-categories.

Interpreting Results

Expectancy must be positive after costs (0.3-0.5R is good). Profit factor above 1.3. Maximum drawdown must be psychologically tolerable. Expect live performance to be 20-30% worse than back-test.

> Key Takeaway: Back-testing bridges theory and practice. Done rigorously with out-of-sample validation and realistic costs, it gives evidence-based confidence. Methodology matters more than results.