Kelly Criterion

The fundamental equation is f = p - (q / b)* , where f* is the optimal fraction of the bankroll, p is the probability of winning, q is the probability of losing (1-p), and b is the net odds received on the wager (i.e., you win $b for every $1 bet).

• Edge Calculation: The numerator (bp - q) represents the mathematical edge.
• Growth Maximization: The criterion maximizes the expected value of the logarithm of wealth, not the simple arithmetic return.
• Example: For a bet with a 60% chance of doubling your money (b=1), f* = 0.6 - (0.4/1) = 0.20, suggesting a 20% allocation.

Unlike strategies maximizing arithmetic mean return, the Kelly Criterion optimizes the expected logarithm of wealth, which corresponds to the compound growth rate.

• Long-Run Dominance: A Kelly bettor's bankroll will asymptotically exceed that of any essentially different strategy over a sufficiently long sequence of bets.
• Volatility Drag: The logarithmic utility function inherently penalizes large drawdowns, recognizing that a 50% loss requires a 100% gain to recover.
• Time Horizon: The strategy assumes an infinite or very long sequence of independent, identical opportunities.

In practice, traders rarely use the full Kelly fraction due to estimation errors in inputs. Fractional Kelly involves betting a fixed fraction of the full Kelly amount (e.g., Half-Kelly).

• Volatility Reduction: Betting half the Kelly fraction reduces portfolio volatility by 50% while only sacrificing 25% of the expected growth rate.
• Error Buffer: This provides a margin of safety against overestimating one's edge or win probability.
• Drawdown Control: Fractional strategies significantly reduce the probability and magnitude of severe peak-to-trough drawdowns.

The criterion extends beyond discrete binary bets to continuous financial markets via the Multivariate Kelly Criterion.

• Continuous Form: In markets, the optimal allocation vector f = (Σ⁻¹)μ, where Σ⁻¹ is the inverse covariance matrix and μ is the vector of excess returns.
• Leverage Constraint: The sum of the weights often exceeds 1.0, requiring explicit leverage constraints in portfolio optimization.
• Signal Integration: It naturally integrates with alpha models, converting a vector of expected returns and a covariance matrix directly into position sizes.

Despite its mathematical elegance, strict application faces practical hurdles in quantitative finance.

• Parameter Uncertainty: The output is highly sensitive to errors in estimating expected returns (μ), often leading to unstable and concentrated portfolios.
• IID Assumption: The standard model assumes independent and identically distributed returns, ignoring serial correlation and volatility clustering.
• Non-Stationarity: Market edges are transient; the assumption of a repeating, identical opportunity set is violated in non-stationary financial environments.

PMPT and the Kelly Criterion share a philosophical alignment with the Sortino Ratio, which replaces standard deviation with downside deviation.

• Downside Focus: Both frameworks recognize that only returns falling below a specified threshold (MAR or 0%) constitute risk.
• Portfolio Selection: A strategy maximizing the Sortino ratio is often mathematically linked to a Kelly-optimal solution under specific return distributions.
• Asymmetric Utility: They formalize the behavioral reality that investors possess asymmetric risk preferences, fearing losses more than they value equivalent gains.

A foundational framework for constructing portfolios by mathematically balancing expected returns against the variance of returns as a measure of risk. Unlike the Kelly Criterion, which focuses on maximizing geometric growth rate, MVO operates on a single-period horizon and assumes normally distributed returns. The framework generates an efficient frontier of optimal portfolios, allowing investors to select the mix that matches their risk tolerance. Key inputs include expected return estimates, variance calculations, and covariance matrices between all asset pairs.

An allocation strategy that weights assets so that each component contributes an equal amount of risk to total portfolio volatility. While the Kelly Criterion sizes bets based on edge and odds, Risk Parity focuses on balancing risk contributions across uncorrelated sources. This approach prevents any single asset class from dominating the portfolio's risk profile, often resulting in higher Sharpe ratios than traditional capital-weighted allocations. Implementation requires decomposing total portfolio risk into marginal risk contributions from each holding.

A coherent risk measure that quantifies the expected loss in the worst-case scenarios beyond the Value-at-Risk threshold. When applying the Kelly Criterion, practitioners often constrain bet sizes using CVaR to avoid ruin from tail events not captured by the formula's assumptions. CVaR satisfies all four properties of a coherent risk measure: monotonicity, sub-additivity, positive homogeneity, and translational invariance. This makes it mathematically superior to VaR for portfolio optimization under extreme market conditions.

A machine learning-based portfolio optimization method that uses hierarchical clustering to allocate capital without requiring inversion of the covariance matrix. HRP addresses a key limitation of the Kelly Criterion in multi-asset contexts: the instability of correlation estimates. By organizing assets into a tree structure based on distance metrics, HRP produces more robust allocations that are less sensitive to small changes in input parameters. The method proceeds through three stages: tree clustering, quasi-diagonalization, and recursive bisection.

The maximum observed loss from a peak to a trough before a new peak is attained, measuring the largest historical capital impairment. Kelly bettors often use fractional Kelly strategies to reduce MDD, accepting lower geometric growth in exchange for smoother equity curves. A full Kelly strategy can produce drawdowns exceeding 50% during adverse sequences, while half-Kelly typically reduces MDD by more than half with only a 25% reduction in expected growth rate. This trade-off between growth and psychological tolerance is critical in practical implementation.

A method for solving complex multi-period optimization problems by breaking them down into recursive sub-problems governed by the Bellman Equation. The Kelly Criterion naturally extends to sequential decision-making through dynamic programming, where the value function represents the logarithm of wealth. This framework handles path-dependent constraints like margin requirements, position limits, and transaction costs that the static Kelly formula cannot address. Applications include optimal execution and dynamic asset allocation across changing market regimes.