Position Sizing: The Smallest Candidate Wins

The most robust position sizing method for an automated trading system is to compute candidate sizes using risk-based, ATR-based, and half-Kelly formulas, then execute the smallest of the three. This approach mitigates the inherent failure modes of each individual sizing technique, providing a more resilient system.

What went wrong

We experienced a production incident where a seemingly robust strategy, designed to capture mean reversion in a volatile asset, suffered an outsized drawdown. The system was configured with a fixed risk percentage per trade, tied to a stop-loss distance. During a period of extreme, short-lived volatility, the stop-loss was triggered on a position that, due to the fixed risk percentage and a very tight stop, resulted in a significantly larger capital allocation than intended. This single, large loss eroded a substantial portion of the capital, impacting subsequent trading opportunities.

Why it happens

Each position sizing method has a characteristic failure point that can be exploited by specific market conditions:

• Risk-based sizing (fixed fraction of capital with stop distance): This method fails when the stop-loss is unusually tight. If the stop is placed very close to the entry price, a fixed percentage of capital allocated to the trade will result in a larger number of shares or contracts. In our case, a sudden spike in volatility compressed the stop-loss distance, leading to an oversized position relative to the actual risk tolerance we thought we were enforcing.
• ATR-based sizing: This method uses Average True Range (ATR) to gauge volatility and adjust position size. While it accounts for general market volatility, it can fail if the actual stop-loss placement is not aligned with the ATR. The system might size a trade based on ATR, but if the stop is placed much tighter or wider than what ATR suggests, the risk per trade can be miscalculated. It ignores the actual stop distance, which is the primary determinant of risk on a per-trade basis.
• Kelly Criterion (even half-Kelly): The Kelly criterion is theoretically optimal for maximizing long-term growth, but it is extremely sensitive to estimation errors in the strategy's win rate and payoff. Even a modest overestimate of the edge (win rate multiplied by average win size divided by average loss size) can lead to severe drawdowns. In our experience, historical distributions used to estimate these parameters are rarely perfect, and market regimes can shift, invalidating past estimates. Using half-Kelly mitigates this somewhat, but the fundamental sensitivity remains.

What we changed

To address these individual failure modes, we implemented a multi-pronged approach to position sizing within the Tradewink system. For every potential trade signal, our sizer now computes three candidate sizes:

1. Fixed-fractional risk based on stop distance: This is the traditional method where a percentage of capital is risked, and the number of units is determined by the stop-loss distance. Size = (Capital * Risk_Percentage) / Stop_Distance.
2. ATR-based size: This method uses ATR to determine a volatility-adjusted position size. A common approach is Size = (Capital * Risk_Percentage) / (ATR * Multiplier), where the multiplier accounts for how many ATRs the stop is placed away.
3. Half-Kelly: We calculate the theoretical Kelly fraction based on historical performance data (win rate and average win/loss ratios) and then take half of that fraction. Half_Kelly_Fraction = 0.5 * (Win_Rate - Loss_Rate) / Average_Payoff_Ratio.

Crucially, the system then selects the smallest of these three computed sizes. There is no weighting or averaging; we simply take the minimum. This ensures that if one method suggests an aggressive size due to a specific market condition (e.g., tight stop for risk-based, low ATR for ATR-based, or an optimistic edge estimate for Kelly), the other methods act as a brake, forcing a more conservative allocation.

Furthermore, sizes are dynamically adjusted for market regime. We employ a market proxy to detect volatile or transitioning regimes. In such periods, all computed sizes are further reduced. This adds another layer of defense against unexpected market behavior.

Finally, a sophisticated cost model is integrated. Before any size is accepted, it is adjusted downwards to account for expected slippage and commission. This means a marginal trading setup is not sized as if execution were free. The cost model subtracts expected slippage and commission before the size is accepted, ensuring that the effective risk per trade is maintained even after transaction costs.

How to check your own system

To ensure your automated trading system's position sizing is as resilient as possible, run through this checklist:

• Identify your primary sizing method: What formula do you currently use (e.g., fixed risk %, ATR, Kelly)?
• Analyze its failure modes: Under what specific market conditions has this method historically led to oversized positions or unexpected risk?
• Simulate alternative methods: For each trade signal, calculate what the position size would have been using at least two other distinct sizing methodologies (e.g., if you use risk-based, calculate ATR-based and half-Kelly sizes for the same signal).
• Implement the minimum: Configure your system to always use the smallest position size generated by your primary method and the simulated alternatives.
• Incorporate regime filtering: Does your system reduce position sizes during periods of high volatility or regime transition? If not, consider adding this logic.
• Factor in transaction costs: Does your sizing calculation account for expected slippage and commissions? If not, adjust your target risk downwards to reflect these costs before determining the final size.

By systematically evaluating and combining multiple sizing approaches, you can build a more robust system that is less susceptible to the catastrophic failures that plague single-method sizing strategies.

Disclaimer

This article describes engineering decisions in a trading system. It is not investment advice. Trading involves substantial risk of loss and is not suitable for all investors. Past performance does not guarantee future results. Always do your own research and consider your financial situation before trading.