Position Sizing and Drawdown Management

Why Position Sizing Matters More Than Entry Timing

You can have the best trading signals in the world, but without proper position sizing:

        - Single bad streak can wipe out months of gains

• Over - leveraging leads to forced liquidations

  • Under - sizing means opportunity cost
  • Inconsistent sizing creates unpredictable volatility

  ** The truth:** Position sizing determines 70 - 80 % of your long - term returns.Entry timing is only 20 - 30 %.

Kelly Criterion: Mathematical Position Sizing

At the heart of our risk management is the ** Kelly Criterion **, a mathematical formula for optimal position sizing:

``` f* = (p × b - q) / b ```

Where:

• p = probability of winning (win rate)
• q = probability of losing (1 - p)
• b = ratio of win to loss (usually 1 for symmetric payoffs)

Practical Example

Let's say we have:

• $10,000 account
• 55% win rate (p = 0.55)
• 70% signal confidence
• 2% current market volatility

Full Kelly: 10% of capital = $1,000 position

Problem: Full Kelly is too aggressive. A losing streak will hurt badly.

Our Approach: Fractional Kelly (0.25-0.5)

Using 0.25 fractional Kelly:

• Base size: 2.5% of capital = $250
• Volatility adjustment: ×1.5 (moderate volatility)
• Confidence scaling: ×0.7 (70% confidence)
• Final position: $262.50 (2.625% of account)

Why Fractional Kelly?

Fractional Kelly:

• Reduces returns by 10-15%
• But significantly improves risk-adjusted metrics
• Maintains max drawdown under 15% in stress tests
• Better sleep at night

Multi-Layer Position Sizing

We don't just use Kelly Criterion in isolation. Position size is determined by multiple factors:

1. Kelly Criterion (Base)

Starting point based on win probability and confidence.

2. Volatility Scaling

```python volatility_multiplier = baseline_vol / current_vol

if current_vol > 2 * baseline_vol: position *= 0.5 # Halve position in high volatility elif current_vol < 0.5 * baseline_vol: position *= 1.5 # Increase in low volatility (max 1.5x) ```

3. Correlation Adjustment

If we already have highly correlated positions: ```python if avg_correlation > 0.7: position *= 0.7 # Reduce position 30% elif avg_correlation > 0.5: position *= 0.85 # Reduce position 15% ```

4. Drawdown State

If currently underwater: ```python if unrealized_drawdown > 10%: position *= 0.5 # Halve position elif unrealized_drawdown > 5%: position *= 0.75 # Reduce 25% ```

5. Confidence Weighting

Signal confidence directly scales position:

• 90%+ confidence: 1.0x multiplier
• 70-90% confidence: 0.7-1.0x
• 50-70% confidence: 0.5-0.7x
• <50% confidence: No trade

Drawdown Throttling Mechanisms

Drawdowns are inevitable. How we respond determines survival.

Tiered Drawdown Response

Implementation

```python def check_drawdown_throttle(current_equity, peak_equity): drawdown_pct = (peak_equity - current_equity) / peak_equity

if drawdown_pct >= 0.25:
    # Kill switch
    flatten_all_positions()
    pause_trading(duration_hours=24)
    alert_admin("CRITICAL: 25% drawdown")
    
elif drawdown_pct >= 0.20:
    # Flatten everything
    flatten_all_positions()
    reduce_position_sizing(multiplier=0.0)
    
elif drawdown_pct >= 0.15:
    # Halve position sizes
    reduce_position_sizing(multiplier=0.5)
    
elif drawdown_pct >= 0.10:
    # Reduce 25%
    reduce_position_sizing(multiplier=0.75)
    
return drawdown_pct

```

Portfolio-Level Risk Management

Beyond individual trades, we manage portfolio-level risks:

Concentration Limits

• Per asset: Max 10-20% of portfolio
• Per sector: Max 30% of portfolio
• Geographic: Diversification across regions

Leverage Limits

• Crypto: Max 3x leverage
• Stocks: Max 2x leverage
• Options: Based on total Greeks exposure

Dynamic adjustment based on VaR: ```python def calculate_allowed_leverage(portfolio_var_95): if portfolio_var_95 > 0.05: # >5% VaR return 1.0 # No leverage elif portfolio_var_95 > 0.03: # 3-5% VaR return 2.0 else: return 3.0 # Low VaR, allow max leverage ```

Real-World Examples

Example 1: Moderate Win Rate Strategy

Strategy specs:

• Win rate: 58%
• Average win: $100
• Average loss: $100
• Signal confidence: 75%

Position sizing:

• Kelly suggests: 16% of capital
• Our fractional Kelly (0.25): 4% of capital
• With volatility adjustment: 3.5% final position

Result over 100 trades:

• Full Kelly: +45% (with 22% drawdown)
• Our approach: +38% (with 12% drawdown)
• Better Sharpe: 2.4 vs 1.9

Example 2: High Volatility Period

Bitcoin volatility spikes from 40% to 80%:

Our response:

1. Volatility multiplier kicks in: ×0.5
2. Positions automatically halved
3. New signals require 80%+ confidence
4. Max portfolio exposure reduced to 50%

Impact:

• Drawdown limited to 8% (vs 18% without adjustment)
• Missed some upside, but preserved capital
• Positioned to buy the dip when volatility normalizes

The Cost of Safety

Our conservative approach has trade-offs:

Advantages:

• ✅ Consistent returns with lower volatility
• ✅ Survivability through market crashes
• ✅ Regulatory compliance-ready
• ✅ Sleep-at-night peace of mind

Disadvantages:

• ❌ Lower maximum returns vs aggressive strategies
• ❌ Miss some explosive opportunities
• ❌ More complex system architecture
• ❌ Higher computational overhead

Performance Comparison

Aggressive vs Conservative position sizing over 3 years:

The conservative approach wins over multi-year periods because staying in the game matters more than maximizing short-term gains.

Lessons Learned

After extensive testing and live trading:

1. Position sizing matters more than entry timing
2. Fractional Kelly (0.25-0.5) is optimal for most strategies
3. Drawdown throttling is mandatory, not optional
4. Volatility scaling prevents blow-ups
5. Correlation monitoring protects against diversification illusion

Conclusion

In trading, you don't get rewarded for taking maximum risk. You get rewarded for taking optimal risk—the sweet spot where you maximize risk-adjusted returns while ensuring long-term survival.

Our risk-first approach may not be flashy, but it's built to last.

---

Want to see our risk management in action? Try our live demo or read about our system architecture.