Strategy Expander: 380 Configurations from 63 Templates

Abstract

Complementing our Model Factory, the Strategy Expander automatically generates 380 trading strategy configurations from 63 core templates. This paper details the template system, expansion logic, and performance-based selection mechanism.

The Template System

Our Strategy Expander starts with 63 carefully designed core strategy templates across multiple categories:

Technical Analysis Strategies (15 templates)

• Moving Average Crossover variants
• RSI Reversal strategies
• MACD Signal strategies
• Bollinger Band breakout/mean-reversion
• Ichimoku Cloud strategies
• Volume-price divergence
• Multiple timeframe analysis

ML-Driven Strategies (2 templates)

• Ensemble prediction integration
• Feature importance trading

Arbitrage Strategies (3 templates)

• Cross-exchange statistical arbitrage
• Triangular arbitrage
• Funding rate arbitrage

Event-Driven Strategies (1 template)

• News/announcement reaction trading

Grid Bot Strategies (5 templates)

• Fixed grid trading
• Dynamic grid trading
• Volatility-adaptive grids
• Trend-following grids
• Mean-reversion grids

Hedge Fund Strategies (22 templates)

• Statistical pair trading
• Mean reversion (various lookbacks)
• Momentum (various timeframes)
• Trend following
• Counter-trend
• Volatility targeting
• Risk parity
• Factor investing

Microstructure Strategies (6 templates)

• Order flow imbalance
• VPIN-based trading
• Microprice deviation
• Market making
• Queue position optimization
• Iceberg detection

Advanced Strategies (7 templates)

• Regime detection and switching
• Correlation breakdown trading
• Options gamma scalping
• Volatility surface trading
• Delta hedging
• Carry strategies
• Basis trading

Expansion Logic

Each template is expanded across multiple parameter sets:

Parameter Categories

1. Timeframes: 1m, 5m, 15m, 1h, 4h, 1d
2. Lookback Windows: 10, 20, 50, 100, 200 periods
3. Risk Parameters: Conservative, moderate, aggressive
4. Entry Thresholds: Tight, standard, loose
5. Exit Rules: Time-based, signal-based, trailing

Expansion Example: MA Crossover

Base template: Moving Average Crossover

Expansions:

• Fast MA: 10, 20, 50 periods
• Slow MA: 50, 100, 200 periods
• Timeframes: 1h, 4h, 1d
• Entry filter: None, RSI, Volume

Total variants: 3 × 3 × 3 × 4 = 108 configurations

(After pruning invalid combinations: ~30 viable configurations per template)

The 380 Configuration Count

63 templates × ~6 average expansions = 378 ≈ 380 configurations

Some templates expand more (MA crossover: 30+), others less (specific microstructure: 2-3).

Configuration Structure

Each configuration includes:

python
@dataclass
class StrategyConfiguration:
    # Identity
    template_id: str
    config_id: str
    name: str
    
    # Parameters
    timeframe: str
    lookback: int
    risk_profile: str
    
    # Entry conditions
    entry_rules: List[EntryRule]
    entry_threshold: float
    
    # Exit conditions
    exit_rules: List[ExitRule]
    stop_loss_pct: float
    take_profit_pct: float
    
    # Position sizing
    position_size_method: str
    max_position_pct: float
    
    # Execution
    execution_algo: str
    slippage_model: str
    
    # Performance tracking
    sharpe_ratio_30d: float
    win_rate_30d: float
    max_drawdown_30d: float
    is_active: bool

Performance-Based Selection

All 380 configurations run continuously in evaluation mode.

Rating System

Each configuration receives a composite score:

python
def calculate_composite_score(config: StrategyConfiguration) -> float:
    """
    Calculate composite score for configuration ranking.
    Higher is better.
    """
    # Sharpe ratio (weighted 40%)
    sharpe_score = config.sharpe_ratio_30d * 0.4
    
    # Win rate (weighted 20%)
    win_rate_score = config.win_rate_30d * 0.2
    
    # Drawdown penalty (weighted 20%)
    drawdown_penalty = (1 - config.max_drawdown_30d) * 0.2
    
    # Consistency bonus (weighted 10%)
    consistency_score = config.profit_factor_30d / 3 * 0.1
    
    # Diversification bonus (weighted 10%)
    correlation_penalty = config.avg_correlation_with_active * 0.1
    
    return (
        sharpe_score +
        win_rate_score +
        drawdown_penalty +
        consistency_score -
        correlation_penalty
    )

Selection Algorithm

Every 24 hours:

1. Calculate composite scores for all 380 configurations
2. Rank by composite score
3. Select top 15 that meet minimum thresholds:
   • Sharpe > 1.0
   • Win rate > 50%
   • Max drawdown < 25%
4. Ensure diversification:
   • Max 3 from same template family
   • Max correlation 0.6 between any two active strategies
5. Activate selected configurations
6. Demote others to evaluation-only mode

Continuous Adaptation

The Strategy Expander adapts continuously:

5-Minute Updates

• Performance metrics updated
• Risk metrics recalculated
• Signal generation continues

24-Hour Rebalancing

• Full ranking recalculation
• Active configuration review
• Promotions and demotions

Weekly Deep Analysis

• Template performance review
• Expansion rule updates
• Dead configuration pruning

Results

Strategy Expander performance across market regimes:

The system automatically adjusts strategy mix based on market conditions.

Integration with Model Factory

Strategy Expander works alongside Model Factory:

• Model Factory: Generates price predictions and signals
• Strategy Expander: Determines how to act on signals
• Combined: 486 ML configurations × 380 strategy configurations

Not all combinations are valid—we pair compatible models with appropriate strategies based on signal type and frequency.

Conclusion

The Strategy Expander provides:

1. Breadth: 380 configurations covering diverse market approaches
2. Depth: Careful parameter expansion from proven templates
3. Adaptability: Automatic selection based on market conditions
4. Efficiency: Resources focused on top performers

Together with the Model Factory, this creates a comprehensive, adaptive trading infrastructure.

---

For more on our infrastructure approach, see our blog post on building transparent trading systems.