Backtest & Optimize

How partiqon tests strategies on historical data, finds optimal parameters, and validates robustness. Includes backtesting, optimization, walk-forward validation, and WFO.

Backtesting

A backtest simulates a strategy on historical data. For each bar, we check if entry or exit signals fire, and record all trades with their P&L. No real money is used — it's pure simulation to measure past performance.

How It Works

1. Load data: Historical bars (OHLCV) from BAR6 files
2. Reset state: Strategy parameters, indicators, and positions reset to start
3. Iterate bars: For each bar in the period, feed it to the strategy
4. Check signals: Does the strategy generate entry or exit signals?
5. Execute trades: Entry opens a position; exit closes it. Record entry/exit price, P&L
6. Calculate metrics: Win rate, drawdown, profit factor, etc.
7. Output results: Trade list, equity curve, performance statistics

Key Metrics

• Total Return: Ending equity − starting capital (in %)
• Win Rate: Winning trades / total trades (higher is better, but not everything)
• Profit Factor: Gross profit / gross loss. >1.5 is respectable; >2.0 is strong
• Max Drawdown: Largest drop from peak equity to trough. Risk metric; lower is better
• Sharpe Ratio: Return per unit of risk. Adjusts for volatility; higher is better
• Trades per year: Frequency. More trades = more data to validate; fewer trades = less slippage/fees

PSO Optimization

PSO (Particle Swarm Optimization) is a search algorithm that finds good parameter values automatically. Instead of trying random combinations, particles "swarm" toward the best regions of the parameter space.

How PSO Works

1. Define searchable parameters: Which block parameters can be tuned? (e.g., EMA period 10–100, RSI level 20–50)
2. Initialize swarm: Create N particles, each with random parameter values (e.g., 30 particles)
3. Evaluate each particle: Run a backtest with those parameters. Record profit factor (or your chosen metric)
4. Track best: Each particle remembers its best result ever. The swarm remembers the global best.
5. Update velocities: Each particle adjusts its parameters toward its personal best and the global best (with randomness)
6. Repeat: Run multiple generations (e.g., 60 iterations). The swarm converges toward good parameters.
7. Return champion: The best parameters found across all generations

Why PSO?

• Efficient: Fewer backtests than grid search (which tries every combination)
• Flexible: Handles any combination of parameters, discrete or continuous
• Robust: Avoids getting stuck in local optima better than pure gradient descent
• Transparent: You see which parameters the algorithm favored and why

Walk-Forward Validation

Walk-forward validation prevents overfitting by splitting data into blocks: optimize on the first block (in-sample), then test on the second block (out-of-sample), but never mix them.

The Problem: Overfitting

If you optimize on all your data, you can fit the noise in that specific history. The parameters look great on the past but fail on the future. Walk-forward breaks the data to catch this.

Walk-Forward Process

1. Define blocks: Split your 5-year history into 3 blocks (e.g., each ~1.5 years)
2. Block 1 (in-sample): Run PSO optimization on this period. Find best parameters.
3. Block 2 (out-of-sample): Test those best parameters on the next block. How do they perform on unseen data?
4. Block 3 (out-of-sample): Slide the window. Use Block 2 to optimize; test on Block 3.

Interpreting Results

• Good strategy: In-sample profit factor 2.0+, out-of-sample profit factor >1.0. Out-of-sample is lower, but still profitable and consistent.
• Overfitted: In-sample factor 5.0+, out-of-sample <1.0. The parameters don't generalize.
• Underfitted: Both in-sample and out-of-sample are poor. The strategy idea isn't strong enough.

Walk-Forward Optimization (WFO)

WFO combines walk-forward validation with optimization: optimize on each in-sample block, then test the result on the next out-of-sample block. This simulates real trading where you re-optimize periodically.

WFO Process

1. Split data: 3 blocks (or more). Blocks 1, 2, 3 for example.
2. Cycle 1: Optimize on Block 1 (in-sample) → test optimized params on Block 2 (out-of-sample)
3. Cycle 2: Optimize on Block 2 (in-sample) → test optimized params on Block 3 (out-of-sample)
4. Aggregate results: Combine all out-of-sample trades from cycles 1, 2, etc.
5. Calculate metrics: Profit factor, drawdown, etc. on the full aggregated out-of-sample result

WFO Interpretation

• Strong signal: WFO profit factor >1.5, consistent across all cycles. Parameters generalize.
• Weak signal: WFO profit factor <1.0 or highly variable. Parameters are fragile.
• Quarterly breakdown: If 3 out of 4 out-of-sample periods are profitable, the strategy is robust. If only 1 out of 4, it's brittle.

Strategy Development Workflow

The recommended approach:

1. Build: Design your strategy using blocks in the graph editor
2. Test: Run a simple backtest with default parameters (2020–2026)
3. Optimize: Run PSO on a clean in-sample block (e.g., 2020–2023)
4. Validate: Backtest the optimized params on out-of-sample (2023–2026). Profit factor >1.0?
5. Walk-Forward: Run WFO with 3 blocks. Profit factor >1.5 and consistent? Ready to trade.
6. Paper trade: Demo the strategy on live data for 1–2 weeks before real money
7. Redeploy: Every quarter, run WFO again on the latest data to re-optimize parameters

Common Pitfalls

• Using future data: Indicators or signals that "know" tomorrow's close. Always check for lookahead bias.
• Ignoring slippage: Backtests assume instant fills at OHLC prices. Real trading has slippage. Add 0.5–1 ATR as a buffer.
• Too many parameters: Optimizing 20 parameters on 3 years of data = overfitting. Keep parameters minimal.
• Wrong metric: Chasing max profit factor can hide negative skew. Use Sharpe ratio or profit factor with max drawdown together.
• In-sample only: Never optimize and test on the same period. Always validate out-of-sample.

Next Steps

Ready to build a strategy? Start in the graph editor on the /graph page. Or read the block reference to understand what's available.