Academy/Advanced Quant/Portfolio Construction & Kelly Criterion
Advanced QuantLesson 4

Portfolio Construction & Kelly Criterion

Mathematically optimal position sizing and portfolio allocation.

13 minute read
4 key takeaways

The Kelly Criterion: Optimal Sizing

How much of your bankroll should you bet per trade? The Kelly Criterion answers this mathematically.

The Formula

f* = (bp - q) / b

Optimal fraction of bankroll to bet

  • b = win/loss ratio (average win / average loss)
  • p = win probability
  • q = 1 - p = loss probability
python
# Example strategy
win_probability = 0.55
avg_win = 2.0
avg_loss = 1.0
win_loss_ratio = avg_win / avg_loss

kelly_fraction = (win_loss_ratio * win_probability - (1 - win_probability)) / win_loss_ratio
# = (2.0 * 0.55 - 0.45) / 2.0
# = (1.10 - 0.45) / 2.0
# = 0.325 = 32.5%

half_kelly = kelly_fraction / 2  # 16.25%
# Use Half-Kelly in practice

Full Kelly vs. Half-Kelly

ApproachGrowthMax DrawdownWhen to Use
Full KellyHighest-60%+Only if you can handle it
Half-KellyHigh-20-30%This one! (recommended)
Quarter-KellyModerate-10%If very risk-averse

Full Kelly has incredible growth but devastating drawdowns. You'll go broke psychologically. Use Half-Kelly.

Portfolio-Level Diversification

Instead of betting everything on one strategy, diversify across multiple uncorrelated strategies.

python
# Portfolio of strategies
strategies = {
    'trend_following': {'allocation': 0.30, 'expected_return': 0.12},
    'mean_reversion': {'allocation': 0.30, 'expected_return': 0.10},
    'pairs_trading': {'allocation': 0.25, 'expected_return': 0.14},
    'cash': {'allocation': 0.15, 'expected_return': 0.05}
}

# Portfolio return = sum(weights * returns)
portfolio_return = sum(
    s['allocation'] * s['expected_return']
    for s in strategies.values()
)
# = 0.30*0.12 + 0.30*0.10 + 0.25*0.14 + 0.15*0.05 = 0.108 = 10.8%

# Correlation matters more than individual strategy returns
# Low correlation = lower portfolio volatility

The Correlation Problem

In normal times, stocks are 50-70% correlated. In crashes, correlation goes to 1.0 (everyone sells everything).

Correlation illusion in Bull Markets

During 2010-2021 bull market, diversification seemed pointless (everything went up 20%/year). In 2022 crash, suddenly diversification mattered. Always prepare for regime change.

Key Takeaways
  • Kelly Criterion: f* = (bp - q) / b
  • Full Kelly has wild drawdowns; use Half-Kelly
  • Correlation matrix changes in crashes (watch out)
  • Diversify across uncorrelated strategies, not just stocks