trading-platform/docs/02-definicion-modulos/OQI-006-ml-signals/especificaciones/ET-ML-015-backtesting-framework.md

---
id: "ET-ML-015"
title: "Backtesting Framework"
type: "Technical Specification"
status: "Approved"
priority: "Alta"
epic: "OQI-006"
project: "trading-platform"
version: "1.0.0"
created_date: "2026-01-25"
updated_date: "2026-01-25"
task_reference: "TASK-2026-01-25-ML-TRAINING-ENHANCEMENT"
---

# ET-ML-015: Backtesting Framework

## Metadata

| Campo | Valor |
|-------|-------|
| **ID** | ET-ML-015 |
| **Epica** | OQI-006 - Senales ML |
| **Tipo** | Especificacion Tecnica |
| **Version** | 1.0.0 |
| **Estado** | Aprobado |
| **Ultima actualizacion** | 2026-01-25 |
| **Tarea Referencia** | TASK-2026-01-25-ML-TRAINING-ENHANCEMENT |

---

## Resumen

El Backtesting Framework proporciona una infraestructura completa para evaluar estrategias ML en datos historicos. Incluye **BacktestEngine**, **PositionManager**, **MetricsCalculator**, validacion walk-forward, y position sizing basado en **Kelly Criterion**.

### Componentes Principales

- **BacktestEngine**: Motor principal de backtesting
- **PositionManager**: Gestion de posiciones y ordenes
- **MetricsCalculator**: Calculo de metricas de performance
- **EffectivenessValidator**: Validacion de efectividad (target 80%)
- **Walk-Forward Validator**: Validacion temporal robusta

---

## Arquitectura

### Diagrama General

```
┌─────────────────────────────────────────────────────────────────────────┐
│                      BACKTESTING FRAMEWORK                               │
├─────────────────────────────────────────────────────────────────────────┤
│                                                                          │
│  ┌───────────────────────────────────────────────────────────────────┐  │
│  │                      BACKTEST ENGINE                               │  │
│  │                                                                     │  │
│  │   ┌─────────────────────────────────────────────────────────────┐ │  │
│  │   │                     DATA HANDLER                             │ │  │
│  │   │  - Load historical data                                      │ │  │
│  │   │  - Manage data streams                                       │ │  │
│  │   │  - Handle multiple timeframes                                │ │  │
│  │   └─────────────────────────────────────────────────────────────┘ │  │
│  │                              │                                      │  │
│  │                              ▼                                      │  │
│  │   ┌─────────────────────────────────────────────────────────────┐ │  │
│  │   │                  STRATEGY EXECUTOR                           │ │  │
│  │   │  - Execute ML predictions                                    │ │  │
│  │   │  - Generate signals                                          │ │  │
│  │   │  - Manage signal queue                                       │ │  │
│  │   └─────────────────────────────────────────────────────────────┘ │  │
│  │                              │                                      │  │
│  │                              ▼                                      │  │
│  │   ┌─────────────────────────────────────────────────────────────┐ │  │
│  │   │                  POSITION MANAGER                            │ │  │
│  │   │  - Open/close positions                                      │ │  │
│  │   │  - Kelly position sizing                                     │ │  │
│  │   │  - TP/SL management                                          │ │  │
│  │   │  - Risk controls                                             │ │  │
│  │   └─────────────────────────────────────────────────────────────┘ │  │
│  │                              │                                      │  │
│  │                              ▼                                      │  │
│  │   ┌─────────────────────────────────────────────────────────────┐ │  │
│  │   │                  METRICS CALCULATOR                          │ │  │
│  │   │  - Calculate returns                                         │ │  │
│  │   │  - Risk metrics (Sharpe, Sortino, Max DD)                   │ │  │
│  │   │  - Trade statistics                                          │ │  │
│  │   │  - ML-specific metrics                                       │ │  │
│  │   └─────────────────────────────────────────────────────────────┘ │  │
│  └───────────────────────────────────────────────────────────────────┘  │
│                              │                                           │
│                              ▼                                           │
│  ┌───────────────────────────────────────────────────────────────────┐  │
│  │                  WALK-FORWARD VALIDATOR                            │  │
│  │                                                                     │  │
│  │  ┌───────────┐  ┌───────────┐  ┌───────────┐  ┌───────────┐      │  │
│  │  │  Fold 1   │  │  Fold 2   │  │  Fold 3   │  │  Fold N   │      │  │
│  │  │ Train/Test│  │ Train/Test│  │ Train/Test│  │ Train/Test│      │  │
│  │  └───────────┘  └───────────┘  └───────────┘  └───────────┘      │  │
│  │                              │                                      │  │
│  │                              ▼                                      │  │
│  │        ┌─────────────────────────────────────────────────┐        │  │
│  │        │         EFFECTIVENESS VALIDATOR                  │        │  │
│  │        │         Target: 80% Effectiveness                │        │  │
│  │        └─────────────────────────────────────────────────┘        │  │
│  └───────────────────────────────────────────────────────────────────┘  │
│                              │                                           │
│                              ▼                                           │
│  Output: BacktestResults                                                 │
│          - Performance metrics                                           │
│          - Trade history                                                 │
│          - Equity curve                                                  │
│          - Walk-forward analysis                                         │
│          - Effectiveness score                                           │
└─────────────────────────────────────────────────────────────────────────┘
```

---

## BacktestEngine

### Configuracion

```python
@dataclass
class BacktestConfig:
    # Data
    symbol: str
    start_date: str
    end_date: str
    timeframe: str = '5m'

    # Capital
    initial_capital: float = 10000.0
    currency: str = 'USD'

    # Position sizing
    position_sizing: str = 'kelly'  # 'fixed', 'kelly', 'volatility'
    max_position_pct: float = 0.1   # Max 10% per position
    kelly_fraction: float = 0.5     # Half-Kelly

    # Risk management
    max_drawdown_pct: float = 0.2   # Stop trading at 20% DD
    max_daily_loss_pct: float = 0.05
    max_open_positions: int = 3

    # Execution
    slippage_pct: float = 0.001     # 0.1% slippage
    commission_pct: float = 0.001   # 0.1% commission

    # Walk-forward
    walk_forward_folds: int = 5
    train_ratio: float = 0.8
```

### Clase Principal

```python
class BacktestEngine:
    """Motor principal de backtesting"""

    def __init__(self, config: BacktestConfig):
        self.config = config
        self.data_handler = DataHandler(config)
        self.position_manager = PositionManager(config)
        self.metrics_calculator = MetricsCalculator()
        self.trade_log: List[Trade] = []
        self.equity_curve: List[float] = []

    def run(
        self,
        strategy: MLStrategy,
        data: pd.DataFrame
    ) -> BacktestResults:
        """Ejecutar backtest completo"""

        self.equity_curve = [self.config.initial_capital]

        for i in range(len(data)):
            bar = data.iloc[i]

            # 1. Update positions with current prices
            self._update_positions(bar)

            # 2. Check exit conditions (TP/SL)
            self._check_exits(bar)

            # 3. Generate signal from strategy
            signal = strategy.generate_signal(data.iloc[:i+1])

            # 4. Execute signal if valid
            if signal and self._can_trade():
                self._execute_signal(signal, bar)

            # 5. Record equity
            self.equity_curve.append(self._calculate_equity(bar))

            # 6. Check risk limits
            if self._check_risk_limits():
                break

        return self._compile_results()

    def _execute_signal(self, signal: Signal, bar: pd.Series):
        """Ejecutar senal de trading"""
        # Calculate position size
        size = self.position_manager.calculate_size(
            signal=signal,
            current_price=bar['close'],
            account_value=self.equity_curve[-1]
        )

        if size > 0:
            # Apply slippage
            entry_price = self._apply_slippage(
                bar['close'],
                signal.direction
            )

            # Open position
            position = self.position_manager.open_position(
                direction=signal.direction,
                size=size,
                entry_price=entry_price,
                stop_loss=signal.stop_loss,
                take_profit=signal.take_profit,
                timestamp=bar.name
            )

            self.trade_log.append(position)
```

---

## PositionManager

### Kelly Criterion Position Sizing

```python
class PositionManager:
    """Gestion de posiciones con Kelly sizing"""

    def __init__(self, config: BacktestConfig):
        self.config = config
        self.positions: List[Position] = []
        self.closed_positions: List[Position] = []
        self.win_rate: float = 0.5  # Updated dynamically
        self.avg_win: float = 0.01
        self.avg_loss: float = 0.01

    def calculate_kelly_fraction(self) -> float:
        """Calcular fraccion de Kelly optima"""
        # Kelly formula: f* = (p * b - q) / b
        # p = win probability
        # q = loss probability (1 - p)
        # b = win/loss ratio

        p = self.win_rate
        q = 1 - p
        b = self.avg_win / self.avg_loss if self.avg_loss > 0 else 1

        kelly = (p * b - q) / b

        # Apply fraction (half-Kelly recommended)
        kelly *= self.config.kelly_fraction

        # Cap at max position
        kelly = min(kelly, self.config.max_position_pct)
        kelly = max(kelly, 0)  # No negative

        return kelly

    def calculate_size(
        self,
        signal: Signal,
        current_price: float,
        account_value: float
    ) -> float:
        """Calcular tamano de posicion"""

        if self.config.position_sizing == 'fixed':
            position_value = account_value * self.config.max_position_pct

        elif self.config.position_sizing == 'kelly':
            kelly_fraction = self.calculate_kelly_fraction()
            position_value = account_value * kelly_fraction

            # Adjust for signal confidence
            position_value *= signal.confidence

        elif self.config.position_sizing == 'volatility':
            # Risk parity based on volatility
            volatility = signal.expected_volatility or 0.02
            target_risk = 0.02  # 2% target volatility contribution
            position_value = (target_risk / volatility) * account_value
            position_value = min(position_value, account_value * self.config.max_position_pct)

        # Convert to units
        units = position_value / current_price

        return units

    def update_statistics(self):
        """Actualizar estadisticas de trading"""
        if len(self.closed_positions) < 10:
            return

        wins = [p for p in self.closed_positions if p.pnl > 0]
        losses = [p for p in self.closed_positions if p.pnl <= 0]

        self.win_rate = len(wins) / len(self.closed_positions)

        if wins:
            self.avg_win = np.mean([p.pnl_pct for p in wins])
        if losses:
            self.avg_loss = abs(np.mean([p.pnl_pct for p in losses]))
```

### Position Class

```python
@dataclass
class Position:
    id: str
    symbol: str
    direction: str           # 'LONG' or 'SHORT'
    size: float              # Units
    entry_price: float
    entry_time: datetime
    stop_loss: Optional[float] = None
    take_profit: Optional[float] = None
    exit_price: Optional[float] = None
    exit_time: Optional[datetime] = None
    exit_reason: Optional[str] = None  # 'TP', 'SL', 'SIGNAL', 'MANUAL'

    @property
    def is_open(self) -> bool:
        return self.exit_price is None

    @property
    def pnl(self) -> float:
        if self.exit_price is None:
            return 0
        if self.direction == 'LONG':
            return (self.exit_price - self.entry_price) * self.size
        else:
            return (self.entry_price - self.exit_price) * self.size

    @property
    def pnl_pct(self) -> float:
        if self.exit_price is None:
            return 0
        if self.direction == 'LONG':
            return (self.exit_price - self.entry_price) / self.entry_price
        else:
            return (self.entry_price - self.exit_price) / self.entry_price

    @property
    def duration(self) -> Optional[timedelta]:
        if self.exit_time is None:
            return None
        return self.exit_time - self.entry_time
```

---

## MetricsCalculator

### Metricas Implementadas

```python
class MetricsCalculator:
    """Calculo de metricas de performance"""

    def calculate_all_metrics(
        self,
        equity_curve: List[float],
        trades: List[Position],
        benchmark: Optional[pd.Series] = None
    ) -> BacktestMetrics:
        """Calcular todas las metricas"""

        returns = self._calculate_returns(equity_curve)

        return BacktestMetrics(
            # Return metrics
            total_return=self.total_return(equity_curve),
            cagr=self.cagr(equity_curve),
            annualized_return=self.annualized_return(returns),

            # Risk metrics
            volatility=self.volatility(returns),
            max_drawdown=self.max_drawdown(equity_curve),
            avg_drawdown=self.avg_drawdown(equity_curve),
            calmar_ratio=self.calmar_ratio(equity_curve),

            # Risk-adjusted returns
            sharpe_ratio=self.sharpe_ratio(returns),
            sortino_ratio=self.sortino_ratio(returns),

            # Trade metrics
            n_trades=len(trades),
            win_rate=self.win_rate(trades),
            profit_factor=self.profit_factor(trades),
            avg_trade_return=self.avg_trade_return(trades),
            avg_win=self.avg_win(trades),
            avg_loss=self.avg_loss(trades),
            largest_win=self.largest_win(trades),
            largest_loss=self.largest_loss(trades),
            avg_trade_duration=self.avg_trade_duration(trades),

            # ML-specific
            prediction_accuracy=self.prediction_accuracy(trades),
            signal_quality=self.signal_quality(trades)
        )
```

### Formulas de Metricas

#### Sharpe Ratio

```python
def sharpe_ratio(
    self,
    returns: np.ndarray,
    risk_free_rate: float = 0.02,
    periods_per_year: int = 252 * 24 * 12  # 5-min candles
) -> float:
    """Calcular Sharpe Ratio anualizado"""
    if len(returns) < 2:
        return 0

    excess_returns = returns - (risk_free_rate / periods_per_year)
    if np.std(excess_returns) == 0:
        return 0

    sharpe = np.mean(excess_returns) / np.std(excess_returns)
    sharpe_annualized = sharpe * np.sqrt(periods_per_year)

    return sharpe_annualized
```

#### Sortino Ratio

```python
def sortino_ratio(
    self,
    returns: np.ndarray,
    risk_free_rate: float = 0.02,
    periods_per_year: int = 252 * 24 * 12
) -> float:
    """Calcular Sortino Ratio (solo downside risk)"""
    if len(returns) < 2:
        return 0

    excess_returns = returns - (risk_free_rate / periods_per_year)
    downside_returns = excess_returns[excess_returns < 0]

    if len(downside_returns) == 0 or np.std(downside_returns) == 0:
        return 0

    sortino = np.mean(excess_returns) / np.std(downside_returns)
    sortino_annualized = sortino * np.sqrt(periods_per_year)

    return sortino_annualized
```

#### Maximum Drawdown

```python
def max_drawdown(self, equity_curve: List[float]) -> float:
    """Calcular maximo drawdown"""
    equity = np.array(equity_curve)
    peak = np.maximum.accumulate(equity)
    drawdown = (peak - equity) / peak

    return np.max(drawdown)
```

#### Profit Factor

```python
def profit_factor(self, trades: List[Position]) -> float:
    """Calcular profit factor (gross profit / gross loss)"""
    gross_profit = sum(t.pnl for t in trades if t.pnl > 0)
    gross_loss = abs(sum(t.pnl for t in trades if t.pnl < 0))

    if gross_loss == 0:
        return float('inf') if gross_profit > 0 else 0

    return gross_profit / gross_loss
```

---

## Walk-Forward Validation

### Implementacion

```python
class WalkForwardValidator:
    """Validacion walk-forward para backtesting"""

    def __init__(
        self,
        n_folds: int = 5,
        train_ratio: float = 0.8,
        gap_periods: int = 0
    ):
        self.n_folds = n_folds
        self.train_ratio = train_ratio
        self.gap_periods = gap_periods

    def validate(
        self,
        strategy_class: Type[MLStrategy],
        data: pd.DataFrame,
        backtest_config: BacktestConfig
    ) -> WalkForwardResults:
        """Ejecutar validacion walk-forward"""

        n_samples = len(data)
        fold_size = n_samples // (self.n_folds + 1)

        fold_results = []

        for fold in range(self.n_folds):
            # Define train/test split
            train_end = (fold + 1) * fold_size
            test_start = train_end + self.gap_periods
            test_end = min(test_start + int(fold_size * (1 - self.train_ratio)), n_samples)

            train_data = data.iloc[:train_end]
            test_data = data.iloc[test_start:test_end]

            # Train strategy
            strategy = strategy_class()
            strategy.fit(train_data)

            # Backtest on test period
            engine = BacktestEngine(backtest_config)
            results = engine.run(strategy, test_data)

            fold_results.append({
                'fold': fold + 1,
                'train_size': len(train_data),
                'test_size': len(test_data),
                'sharpe': results.metrics.sharpe_ratio,
                'return': results.metrics.total_return,
                'max_dd': results.metrics.max_drawdown,
                'win_rate': results.metrics.win_rate,
                'n_trades': results.metrics.n_trades
            })

        return self._aggregate_results(fold_results)

    def _aggregate_results(
        self,
        fold_results: List[Dict]
    ) -> WalkForwardResults:
        """Agregar resultados de todos los folds"""

        sharpes = [f['sharpe'] for f in fold_results]
        returns = [f['return'] for f in fold_results]
        max_dds = [f['max_dd'] for f in fold_results]
        win_rates = [f['win_rate'] for f in fold_results]

        return WalkForwardResults(
            n_folds=len(fold_results),
            fold_results=fold_results,
            avg_sharpe=np.mean(sharpes),
            std_sharpe=np.std(sharpes),
            avg_return=np.mean(returns),
            std_return=np.std(returns),
            avg_max_dd=np.mean(max_dds),
            avg_win_rate=np.mean(win_rates),
            stability_score=1 - np.std(sharpes) / (np.mean(sharpes) + 1e-6)
        )
```

### Visualizacion Walk-Forward

```
Time ──────────────────────────────────────────────────────────────▶

Fold 1: [========== TRAIN ==========][gap][===TEST===]
                                                       ▼
Fold 2:      [============== TRAIN ==============][gap][===TEST===]
                                                                    ▼
Fold 3:           [================== TRAIN ==================][gap][===TEST===]
                                                                                 ▼
Fold 4:                [====================== TRAIN ======================][gap][===TEST===]
                                                                                              ▼
Fold 5:                     [========================== TRAIN ==========================][gap][===TEST===]
```

---

## Effectiveness Validator

### Target: 80% Effectiveness

```python
class EffectivenessValidator:
    """Validar efectividad de estrategia ML"""

    EFFECTIVENESS_TARGET = 0.80  # 80% target

    def __init__(self, min_trades: int = 30):
        self.min_trades = min_trades

    def validate(
        self,
        results: BacktestResults,
        predictions: List[Prediction]
    ) -> EffectivenessScore:
        """Calcular score de efectividad"""

        if len(results.trades) < self.min_trades:
            return EffectivenessScore(
                score=0,
                is_valid=False,
                message=f"Insufficient trades: {len(results.trades)} < {self.min_trades}"
            )

        # Component scores
        prediction_accuracy = self._prediction_accuracy(predictions)
        directional_accuracy = self._directional_accuracy(results.trades)
        risk_adjusted = self._risk_adjusted_score(results.metrics)
        consistency = self._consistency_score(results)

        # Weighted final score
        weights = {
            'prediction': 0.25,
            'direction': 0.25,
            'risk_adjusted': 0.30,
            'consistency': 0.20
        }

        final_score = (
            prediction_accuracy * weights['prediction'] +
            directional_accuracy * weights['direction'] +
            risk_adjusted * weights['risk_adjusted'] +
            consistency * weights['consistency']
        )

        return EffectivenessScore(
            score=final_score,
            is_valid=final_score >= self.EFFECTIVENESS_TARGET,
            components={
                'prediction_accuracy': prediction_accuracy,
                'directional_accuracy': directional_accuracy,
                'risk_adjusted': risk_adjusted,
                'consistency': consistency
            },
            message=self._generate_message(final_score)
        )

    def _prediction_accuracy(self, predictions: List[Prediction]) -> float:
        """Accuracy de predicciones ML"""
        correct = sum(1 for p in predictions if p.was_correct)
        return correct / len(predictions) if predictions else 0

    def _directional_accuracy(self, trades: List[Position]) -> float:
        """Accuracy direccional de trades"""
        correct = sum(1 for t in trades if t.pnl > 0)
        return correct / len(trades) if trades else 0

    def _risk_adjusted_score(self, metrics: BacktestMetrics) -> float:
        """Score basado en metricas de riesgo"""
        sharpe_score = min(1.0, max(0, metrics.sharpe_ratio) / 2.0)
        sortino_score = min(1.0, max(0, metrics.sortino_ratio) / 2.5)
        dd_score = 1.0 - min(1.0, metrics.max_drawdown * 2)

        return (sharpe_score + sortino_score + dd_score) / 3

    def _consistency_score(self, results: BacktestResults) -> float:
        """Score de consistencia"""
        # Variabilidad de returns mensuales
        monthly_returns = self._calculate_monthly_returns(results.equity_curve)
        if len(monthly_returns) < 3:
            return 0.5

        positive_months = sum(1 for r in monthly_returns if r > 0)
        consistency = positive_months / len(monthly_returns)

        return consistency
```

### Criterios de Efectividad

| Componente | Peso | Target | Descripcion |
|------------|------|--------|-------------|
| **Prediction Accuracy** | 25% | >= 55% | ML predictions correctas |
| **Directional Accuracy** | 25% | >= 55% | Trades en direccion correcta |
| **Risk-Adjusted** | 30% | Sharpe >= 1.0 | Retorno ajustado por riesgo |
| **Consistency** | 20% | >= 60% meses positivos | Estabilidad temporal |

---

## API y Uso

### Uso Basico

```python
from backtesting import BacktestEngine, BacktestConfig, WalkForwardValidator

# Configuracion
config = BacktestConfig(
    symbol='XAUUSD',
    start_date='2024-01-01',
    end_date='2024-12-31',
    initial_capital=10000,
    position_sizing='kelly',
    kelly_fraction=0.5
)

# Cargar datos
data = load_historical_data(config.symbol, config.start_date, config.end_date)

# Crear estrategia
strategy = MyMLStrategy()
strategy.load_model('models/pva/XAUUSD/v1.0.0')

# Ejecutar backtest
engine = BacktestEngine(config)
results = engine.run(strategy, data)

# Mostrar resultados
print(f"Total Return: {results.metrics.total_return:.2%}")
print(f"Sharpe Ratio: {results.metrics.sharpe_ratio:.2f}")
print(f"Max Drawdown: {results.metrics.max_drawdown:.2%}")
print(f"Win Rate: {results.metrics.win_rate:.2%}")
print(f"Profit Factor: {results.metrics.profit_factor:.2f}")
```

### Walk-Forward Validation

```python
# Configurar walk-forward
wf_validator = WalkForwardValidator(
    n_folds=5,
    train_ratio=0.8,
    gap_periods=12  # 1 hour gap
)

# Ejecutar validacion
wf_results = wf_validator.validate(
    strategy_class=PVAStrategy,
    data=data,
    backtest_config=config
)

print(f"Average Sharpe: {wf_results.avg_sharpe:.2f} +/- {wf_results.std_sharpe:.2f}")
print(f"Stability Score: {wf_results.stability_score:.2f}")

# Fold breakdown
for fold in wf_results.fold_results:
    print(f"Fold {fold['fold']}: Sharpe={fold['sharpe']:.2f}, Return={fold['return']:.2%}")
```

### Effectiveness Validation

```python
# Validar efectividad
eff_validator = EffectivenessValidator(min_trades=30)
eff_score = eff_validator.validate(results, predictions)

print(f"Effectiveness Score: {eff_score.score:.2%}")
print(f"Target Achieved: {eff_score.is_valid}")

for component, value in eff_score.components.items():
    print(f"  {component}: {value:.2%}")
```

---

## Estructura de Archivos

```
apps/ml-engine/src/backtesting/
├── __init__.py
├── engine.py               # BacktestEngine
├── position_manager.py     # PositionManager, Position
├── metrics.py              # MetricsCalculator, BacktestMetrics
├── walk_forward.py         # WalkForwardValidator
├── effectiveness.py        # EffectivenessValidator
├── kelly.py                # Kelly Criterion implementation
├── data_handler.py         # DataHandler for historical data
└── visualization.py        # Plotting utilities
```

---

## Visualizaciones

### Equity Curve

```python
def plot_equity_curve(results: BacktestResults):
    """Graficar curva de equity"""
    fig, axes = plt.subplots(3, 1, figsize=(12, 10))

    # Equity
    axes[0].plot(results.equity_curve)
    axes[0].set_title('Equity Curve')
    axes[0].set_ylabel('Account Value ($)')

    # Drawdown
    drawdown = calculate_drawdown_series(results.equity_curve)
    axes[1].fill_between(range(len(drawdown)), drawdown, alpha=0.3, color='red')
    axes[1].set_title('Drawdown')
    axes[1].set_ylabel('Drawdown %')

    # Monthly returns
    monthly_returns = calculate_monthly_returns(results.equity_curve)
    colors = ['green' if r > 0 else 'red' for r in monthly_returns]
    axes[2].bar(range(len(monthly_returns)), monthly_returns, color=colors)
    axes[2].set_title('Monthly Returns')
    axes[2].set_ylabel('Return %')

    plt.tight_layout()
    return fig
```

### Trade Distribution

```python
def plot_trade_distribution(results: BacktestResults):
    """Distribucion de trades"""
    fig, axes = plt.subplots(2, 2, figsize=(12, 10))

    trades = results.trades
    returns = [t.pnl_pct for t in trades]

    # Return distribution
    axes[0, 0].hist(returns, bins=50, edgecolor='black')
    axes[0, 0].axvline(0, color='red', linestyle='--')
    axes[0, 0].set_title('Trade Return Distribution')

    # Win/Loss ratio
    wins = sum(1 for r in returns if r > 0)
    losses = len(returns) - wins
    axes[0, 1].pie([wins, losses], labels=['Wins', 'Losses'], autopct='%1.1f%%')
    axes[0, 1].set_title('Win Rate')

    # Cumulative returns
    cumulative = np.cumsum(returns)
    axes[1, 0].plot(cumulative)
    axes[1, 0].set_title('Cumulative Trade Returns')

    # Trade duration
    durations = [t.duration.total_seconds() / 3600 for t in trades if t.duration]
    axes[1, 1].hist(durations, bins=30, edgecolor='black')
    axes[1, 1].set_title('Trade Duration (hours)')

    plt.tight_layout()
    return fig
```

---

## Consideraciones de Produccion

### Performance Optimization

```python
# Usar numba para calculos intensivos
from numba import jit

@jit(nopython=True)
def fast_max_drawdown(equity: np.ndarray) -> float:
    """Calculo optimizado de max drawdown"""
    peak = equity[0]
    max_dd = 0.0

    for i in range(len(equity)):
        if equity[i] > peak:
            peak = equity[i]
        dd = (peak - equity[i]) / peak
        if dd > max_dd:
            max_dd = dd

    return max_dd
```

### Parallel Backtesting

```python
from concurrent.futures import ProcessPoolExecutor

def parallel_backtest(
    strategies: List[MLStrategy],
    data: pd.DataFrame,
    config: BacktestConfig,
    n_workers: int = 4
) -> List[BacktestResults]:
    """Ejecutar backtests en paralelo"""

    def run_single(strategy):
        engine = BacktestEngine(config)
        return engine.run(strategy, data)

    with ProcessPoolExecutor(max_workers=n_workers) as executor:
        results = list(executor.map(run_single, strategies))

    return results
```

### Result Persistence

```python
def save_results(results: BacktestResults, path: str):
    """Guardar resultados de backtest"""
    output = {
        'metrics': results.metrics.to_dict(),
        'trades': [t.to_dict() for t in results.trades],
        'equity_curve': results.equity_curve,
        'config': results.config.to_dict(),
        'timestamp': datetime.now().isoformat()
    }

    with open(path, 'w') as f:
        json.dump(output, f, indent=2)
```

---

## Referencias

- [ET-ML-001: Arquitectura ML Engine](./ET-ML-001-arquitectura.md)
- [ET-ML-010 to ET-ML-014: Strategy Specifications](./ET-ML-010-pva-strategy.md)
- [Kelly Criterion (Wikipedia)](https://en.wikipedia.org/wiki/Kelly_criterion)
- [Walk-Forward Analysis (Pardo)](https://www.amazon.com/Evaluation-Optimization-Trading-Strategies-Pardo/dp/0470128011)

---

**Autor:** ML-Specialist (NEXUS v4.0)
**Fecha:** 2026-01-25