Changes: - Add FAL_KEY and GEMINI_API_KEY to .env.example - Update picture-it to use ~/.config/opencode/.env (unified creds) - Remove shodh-memory skill (no longer used) - Remove alphaear-* skills (deprecated) - Remove thai-frontend-dev skill (replaced by website-creator) - Remove theme-factory skill - Add mql-developer skill (MQL5 trading) - Add ecommerce-astro skill (Astro e-commerce) - Add website-creator skill (Next.js + Payload CMS) - Update install script for new skills
7.8 KiB
7.8 KiB
Backtesting & Optimization
Table of Contents
- Strategy Tester Modes
- Minimum Requirements for Valid Backtest
- OnTester() - Custom Optimization Criterion
- TesterStatistics() Constants
- Key Performance Metrics
- Walk-Forward Analysis
- Avoiding Overfitting
- Monte Carlo Simulation
- Multi-Currency Testing (MQL5)
- Optimization Modes (MT5)
- Frame Functions (Inter-Pass Communication)
- Practical Backtesting Workflow
Strategy Tester Modes
| Mode | Speed | Accuracy | When to Use |
|---|---|---|---|
| Every tick based on real ticks | Slowest | Most realistic | Final validation with broker-specific ticks |
| Every tick | Slow | High | Final validation, scalpers, intra-bar logic |
| 1-minute OHLC | Medium | Good | Most strategies (4 ticks per M1 bar) |
| Open prices only | Fast | Low | Bar-open strategies, rapid optimization |
| Math calculations | Instant | N/A | Pure computation without ticks |
Best practice: Optimize with Open Prices first, validate winners with Every Tick.
Minimum Requirements for Valid Backtest
- 1,000+ trades minimum (ideally 2,000+)
- 2+ full market cycles (bull + bear + range)
- Multiple years of data
- Consistent spread settings (or variable spread)
OnTester() - Custom Optimization Criterion (MQL5)
double OnTester() {
double profit = TesterStatistics(STAT_PROFIT);
double profitFactor = TesterStatistics(STAT_PROFIT_FACTOR);
double sharpeRatio = TesterStatistics(STAT_SHARPE_RATIO);
double recoveryFactor = TesterStatistics(STAT_RECOVERY_FACTOR);
double maxDrawdown = TesterStatistics(STAT_EQUITY_DD_RELATIVE);
double totalTrades = TesterStatistics(STAT_TRADES);
if(totalTrades < 100) return 0; // Reject small samples
if(maxDrawdown > 25) return 0; // Reject excessive DD
if(profitFactor < 1.3) return 0; // Reject low PF
return sharpeRatio * recoveryFactor * MathSqrt(totalTrades);
}
TesterStatistics() Constants
Complete table:
- STAT_PROFIT - Net profit
- STAT_GROSS_PROFIT - Gross profit
- STAT_GROSS_LOSS - Gross loss
- STAT_TRADES - Total trades
- STAT_PROFIT_TRADES - Winning trades
- STAT_LOSS_TRADES - Losing trades
- STAT_PROFIT_FACTOR - Gross profit / gross loss
- STAT_EXPECTED_PAYOFF - Expected payoff per trade
- STAT_SHARPE_RATIO - Sharpe ratio
- STAT_RECOVERY_FACTOR - Net profit / max drawdown
- STAT_EQUITY_DD - Max equity drawdown in money
- STAT_EQUITY_DD_PERCENT - Max equity drawdown %
- STAT_EQUITY_DD_RELATIVE - Relative equity drawdown %
- STAT_BALANCE_DD - Max balance drawdown
- STAT_BALANCE_DD_PERCENT - Max balance drawdown %
- STAT_MAX_PROFITTRADE - Largest profitable trade
- STAT_MAX_LOSSTRADE - Largest losing trade
- STAT_CONPROFITMAX - Max consecutive profit
- STAT_CONLOSSMAX - Max consecutive loss
- STAT_SHORT_TRADES - Short trades
- STAT_LONG_TRADES - Long trades
- STAT_WIN_SHORT_TRADES - Winning short trades
- STAT_WIN_LONG_TRADES - Winning long trades
Key Performance Metrics
| Metric | Formula | Good Value | Excellent |
|---|---|---|---|
| Sharpe Ratio | (Mean Return - Rf) / StdDev | > 1.0 | > 2.0 |
| Profit Factor | Gross Profit / Gross Loss | > 1.5 | > 2.0 |
| Recovery Factor | Net Profit / Max DD | > 3.0 | > 5.0 |
| Max Drawdown | Peak-to-trough % | < 20% | < 10% |
| Expected Payoff | Net Profit / Total Trades | > 0 | Context |
| Win Rate | Winning / Total | Context | Context |
Walk-Forward Analysis
Methodology
Window 1: [===== IN-SAMPLE =====][= OOS =]
Window 2: [===== IN-SAMPLE =====][= OOS =]
Window 3: [===== IN-SAMPLE =====][= OOS =]
Window 4: [===== IN-SAMPLE =====][= OOS =]
Steps
- Split data into overlapping windows (e.g., 12-month IS, 3-month OOS)
- Optimize on in-sample period
- Test optimized params on out-of-sample period
- Slide window forward, repeat
- Accept if OOS performance >= 50-80% of IS
- MT5 supports natively: Strategy Tester > Settings > Forward period
Walk-Forward Efficiency
WFE = (OOS annualized return) / (IS annualized return)
- WFE > 0.5 = acceptable
- WFE > 0.7 = good
- WFE < 0.3 = likely overfitted
Avoiding Overfitting
Principles
- Parameter stability: Good params have good neighbors. If X=14 works but X=13 and X=15 don't, it's curve-fitted
- Fewer parameters: Each extra parameter increases overfitting risk exponentially
- Out-of-sample testing: Reserve 25-30% of data untouched
- Cross-market validation: Test EURUSD strategy on GBPUSD
- Regime awareness: Test across trending AND ranging markets
Signs of Overfitting
- Sharp performance drop in forward test
- Many parameters (>5 optimized)
- Strategy only works on specific date range
- Unrealistically high backtest metrics
- Parameter sensitivity (small changes cause large performance swings)
Monte Carlo Simulation
Purpose
Test if results are due to skill or luck by randomizing trade sequence.
Implementation
double OnTester() {
// Collect all trade P&Ls
double trades[];
// ... populate from history ...
int simulations = 1000;
double worstDD95;
for(int sim = 0; sim < simulations; sim++) {
// Fisher-Yates shuffle
// Calculate max drawdown for shuffled sequence
}
// Sort DDs, find 95th percentile
// Reject if 95th percentile DD > 30%
return profit / (1 + worstDD95);
}
What Monte Carlo Tells You
- Expected range of drawdowns (not just the one historical path)
- Probability of ruin at different risk levels
- Confidence interval for returns
- If strategy is fragile (high variance across simulations)
Multi-Currency Testing (MQL5)
MT5 Strategy Tester supports multi-symbol natively:
int OnInit() {
// Reference other symbols to include them in test
int handle_eur = iMA("EURUSD", PERIOD_H1, 14, 0, MODE_SMA, PRICE_CLOSE);
int handle_gbp = iMA("GBPUSD", PERIOD_H1, 14, 0, MODE_SMA, PRICE_CLOSE);
// Tester auto-synchronizes all referenced symbols
return INIT_SUCCEEDED;
}
Optimization Modes (MT5)
| Mode | Description |
|---|---|
| Slow (Complete) | Tests every combination (exhaustive) |
| Fast (Genetic) | Genetic algorithm, finds near-optimal efficiently |
| Custom max/min | Optimizes by OnTester() return value |
Cloud Computing
MQL5 Cloud Network distributes optimization across thousands of agents worldwide.
Frame Functions (Inter-Pass Communication)
// In EA (agent): send data at end of each pass
double OnTester() {
uchar data[];
// serialize results into data
FrameAdd("Results", 1, profit, data);
return profit;
}
// In terminal: receive during optimization
void OnTesterPass() {
ulong pass; string name; long id; double value; uchar data[];
while(FrameNext(pass, name, id, value, data)) {
PrintFormat("Pass #%d: profit=%.2f", pass, value);
}
}
// Control functions
void OnTesterInit() { /* before optimization starts */ }
void OnTesterDeinit() { /* after optimization finishes, aggregate results */ }
Practical Backtesting Workflow
- Quick scan: Open Prices Only, wide parameter ranges, genetic optimization
- Narrow down: Reduce ranges around promising areas, complete optimization
- Validate: Every Tick mode with best parameters
- Walk-forward: Set forward period, verify OOS performance
- Monte Carlo: Randomize trade sequence, check robustness
- Multi-market: Test on correlated instruments
- Demo forward test: Run on demo account for 1-3 months minimum