Why Position Sizing Matters More Than Entries
Consider a system with a 60% win rate and 1:1 risk-reward. That is a genuine statistical edge. But if you risk 50% of your account on each trade, a run of three consecutive losers — which has a 6.4% probability — reduces your capital to 12.5% of its starting value. Recovery requires a 700% return. The edge was real; the sizing destroyed it.
The same system with 2% risk per trade survives that streak with a 5.9% drawdown, recoverable in normal trading within two weeks. Nothing changed except position size. Academic research consistently confirms that sizing explains more variance in long-term trading outcomes than signal accuracy.
Ralph Vince's work on optimal-f and the Kelly Criterion's adoption by Renaissance Technologies, Citadel, and other quantitative funds formalised this: the size of the bet matters more than the quality of the bet.
Over-betting relative to your true edge is far more destructive than under-betting. If your real win rate is 53% but you think it's 60%, full Kelly sizing will systematically over-bet and eventually destroy the account.
The Kelly Criterion — Full Formula
The Kelly Criterion maximises the expected logarithm of wealth, equivalent to maximising the long-term geometric growth rate of capital. For binary outcomes: f* = (p × b − q) / b, where f* is the optimal fraction, p is win probability, q = 1 − p is loss probability, and b is the reward-to-risk ratio.
For a system with 55% win rate and 1.5:1 R:R: f* = (0.55 × 1.5 − 0.45) / 1.5 = 0.25. Full Kelly says risk 25% per trade. This produces maximum long-term growth but with drawdowns that can exceed 80% before recovery.
The key insight is that Kelly optimises for terminal wealth, not for comfort. It assumes infinite time horizon, perfect knowledge of probabilities, and zero psychological constraints. None of these hold in practice, which is why professionals modify it.
Half-Kelly: The Professional Standard
Ed Thorp, who pioneered Kelly in both blackjack and equity markets through Princeton Newport Partners, advocated half-Kelly or even quarter-Kelly in practice. The mathematics support this conservatism.
Half-Kelly achieves 75% of full Kelly's geometric growth rate while cutting expected maximum drawdown roughly in half. The growth-to-drawdown ratio improves. You sacrifice 25% of the upside to avoid 50% of the downside.
The deeper reason is estimation error. Kelly assumes you know your true win rate and payoff ratio with certainty. In trading, these are estimates derived from historical data that may not represent future performance. Half-Kelly builds in a margin of safety against this uncertainty. The AI Trading Copilot's Risk Manager agent enforces half-Kelly as the default sizing methodology, using the Trader agent's probability score as the win probability estimate.
The AI Trading Copilot caps half-Kelly sizing at 3% per position, with an override requiring explicit manual confirmation for any size above 2%. This provides a hard ceiling regardless of how confident the model is.
Drawdown Mathematics and Circuit Breakers
Recovery from drawdown scales non-linearly: 10% down needs 11.1% to recover; 20% needs 25%; 50% needs 100%; 75% needs 300%. This asymmetry makes preventing large drawdowns more important than maximising returns.
The AI Trading Copilot implements a 15% drawdown circuit breaker. If the portfolio drops 15% from its high-water mark, the system halts all new signals until manual review and reset. This is enforced by the Risk Gate — the final stage of the signal pipeline.
Beyond the portfolio-level breaker, professional risk management includes daily loss limits (2–3% of equity) and weekly loss limits (5–6% of equity). These prevent tilt — the tendency to increase risk after losses in an attempt to recover quickly.
- ·A 15% drawdown circuit breaker halts all new position entries automatically
- ·Daily loss limit of 2–3% prevents compounding of losses within a single session
- ·Weekly loss limit of 5–6% catches multi-day losing streaks before they escalate
- ·Volatility scaling halves all position sizes when realised vol exceeds the 90th percentile
Correlation-Adjusted Position Sizing
Standard Kelly treats each bet as independent. In portfolio trading, positions are rarely independent. Long positions in AAPL, MSFT, and GOOGL will all decline simultaneously in a broad tech selloff. Effective risk far exceeds the sum of individual position risks.
The adjustment formula modifies the Kelly fraction: f_adjusted = f_kelly / (1 + avg_correlation × n_correlated), where n_correlated is the number of existing positions with correlation above a threshold (commonly 0.6). This ensures portfolio-level risk remains bounded.
The Correlation agent in the AI Trading Copilot computes real-time pairwise correlations across all positions and feeds this data to the Risk Manager. When a new signal triggers for an asset highly correlated with existing exposure, the position size is automatically reduced — or the signal is vetoed entirely if adding it would breach correlation limits.
Portfolio-Level Risk Controls
Individual trade sizing is necessary but insufficient. Portfolio-level controls address systemic risks that per-trade rules miss. These constraints operate as a layered defence system — no single rule is perfect, but multiple overlapping constraints create robust protection.
- ·Maximum total exposure: no more than 20–30% of capital deployed across all positions at any time
- ·Sector concentration: no more than 10% exposure to a single sector
- ·Single-name limit: no individual position exceeding 5% of equity regardless of Kelly suggestion
- ·Directional balance: net long/short exposure capped at 15% unless Regime agent confirms trending
- ·Volatility scaling: automatic position halving when realised volatility exceeds its 90th percentile
The AI Trading Copilot's Risk Gate enforces all these constraints as the final stage of the signal pipeline. A trade might pass Kelly sizing but get vetoed if it would breach any portfolio-level limit.