The Position Sizing Formula That Survives Black Swan Events

The Kelly Criterion calculates the optimal fraction of capital to risk on each trade to maximize long-term growth by balancing win probability and payout ratios into a single number. However, it assumes that future market behavior will mirror the past. Unpredictable events, often referred to as Black Swan events, invalidate this assumption. Such events can lead to catastrophic losses for traders relying solely on the Kelly Criterion, as the concept does not account for fat tails that characterize extreme market fluctuations.

This article examines the limitations of the Kelly Criterion during significant market events, methods for stress-testing position sizing, and alternative frameworks that can withstand market stress.

Quick answer: Why Kelly Criterion assumes stability that Black Swan events disrupt

The Kelly Criterion formula, f = (bp - q) / b, takes into account two key inputs: the probability of winning (p) and the win/loss ratio (b). These inputs are typically based on historical data, leading to the calculation of the optimal fraction of capital (f) to risk per trade.

The formula's critical flaw lies in its assumptions. It presumes that both p and b are constant, and it expects extreme outcomes to fall within predictable boundaries. While this approximation performs sufficiently in stable market conditions, Black Swan events produce outcomes that greatly exceed historical averages. For example, a trader may estimate a 60% win rate and a 2:1 reward-to-risk ratio based on 200 past trades, only to encounter a single event that erases months of gains in a few hours. Historical data fails to accommodate such extremes, making traders vulnerable when relying solely on the Kelly Criterion.

The concept: How Kelly Criterion works and where fat tails undermine the math

How Kelly Criterion is calculated

The Kelly Criterion formula is:

f = (bp - q) / b

Where:

• f = fraction of capital to risk per trade
• b = net odds (win/loss ratio). For instance, if a trade risks $100 to potentially earn $200, then b = 2
• p = probability of winning (expressed as a decimal). For example, 0.60 represents a 60% chance of winning.
• q = probability of losing (1 - p). For example, 0.40 for a 60% win rate.

Worked calculation:

Assume a trader has a system with a 55% win rate (p = 0.55) and an average win of $200 against an average loss of $100 (resulting in a reward-to-risk ratio of 2:1, so b = 2).

• q = 1 - 0.55 = 0.45
• f = (2 × 0.55 - 0.45) / 2 = (1.10 - 0.45) / 2 = 0.65 / 2 = 0.325

Kelly's criterion suggests risking 32.5% of the capital per trade.

For a $100,000 account, this means risking $32,500 for each trade. However, if the trader experiences 10 consecutive losses (a 1-in-1,000 occurrence when assuming a 55% win rate, although more common than standard distributions might suggest), the account could plummet by $325,000, effectively wiping it out.

Where the math breaks down

Kelly's assumptions include:

1. Known, stable edge. Traders must have accurate estimates for p and b. Changing market conditions can diminish trading edges and alter volatility. A 55% win rate during one market phase may decline to 45% in another.

2. Continuous distribution. Kelly assumes outcomes are predictable and smooth. It fails to consider sudden market movements, such as limit-up/limit-down situations or flash crashes, which are typical features of Black Swan events, as discussed in common features.

3. Non-existent fat tails. Normal distributions suggest that extremely rare events (like a 10-standard-deviation occurrence) happen once every 34 million years. However, Black Swans occur more frequently due to fat tails, where extreme outcomes cluster more than standard statistics would imply. Kelly does not account for this risk.

4. Costless and instant leverage. Kelly assumes that traders can adjust their position sizes instantaneously. In reality, margin calls and liquidity issues during crises can lead to liquidations at unfavorable prices.

When fat tail events occur, following Kelly's position sizes can convert a trading edge into a complete loss.

The process, step by step: A framework for stress-testing your position sizing

Traders should assess any position sizing method against potential stress scenarios before use. Here is a repeatable framework:

Step 1: Calculate your Kelly recommendation

Employ the formula with your system's historical values for p and b. This step establishes a baseline.

Step 2: Identify your three worst historical drawdown scenarios

Analyze your trading history or backtest results to pinpoint the three steepest drawdown periods. Document for each:

• The maximum consecutive losing trades
• The largest single loss
• The total loss as a percentage of your capital

Step 3: Model worst-case outcomes

Design a scenario that is 50% worse than your historical worst case for stress testing. For example:

• If your worst historical drawdown was -15%, test it for -22.5%
• If your longest losing streak was 7 trades, test for 10-12 trades

Step 4: Run your position sizing rule against the stress case

Apply your Kelly recommendation (or an alternate rule) to a simulated $100,000 account through the stress scenario. Calculate the balance after each loss.

Example: A trader using full Kelly (32.5% risk per trade) facing 10 consecutive losses of $32,500 would see their account decrease to -$225,000, resulting in a total loss.

Step 5: Adjust position sizing downward until account survival is assured

Lower the risk fraction until even the worst-case scenario leaves the account above a survivable threshold (typically 10-20% of initial capital remaining). This adjusted fraction becomes the deployed rule.

Step 6: Document and monitor

Record your Kelly input assumptions, stress case parameters, and final deployed position size. Recalculate p and b based on recent data quarterly. If significant changes arise in either, re-run the stress test and adjust position size accordingly.

Worked example: Kelly vs. fixed-fractional sizing in a market crash

Consider traders Alice and Bob, each with $100,000 accounts and identical edges: a 55% win rate and a 2:1 reward-to-risk ratio.

Alice uses full Kelly: 32.5% risk per trade ($32,500).

Bob employs fixed-fractional sizing: 5% risk per trade ($5,000).

Both face a market shock triggered by a central bank decision, resulting in a liquidation cascade. They endure 12 consecutive losing trades, surpassing any historical period indicated in their backtesting.

Trade #	Alice (32.5% Kelly)	Bob (5% Fixed)	Alice Balance	Bob Balance
Start	,	,	$100,000	$100,000
1	−$32,500	−$5,000	$67,500	$95,000
2	−$21,938	−$4,750	$45,562	$90,250
3	−$14,807	−$4,513	$30,755	$85,738
4	−$9,996	−$4,287	$20,759	$81,450
5	−$6,747	−$4,073	$14,012	$77,377
6	−$4,554	−$3,869	$9,458	$73,508
7	−$3,074	−$3,676	$6,384	$69,833
8	−$2,075	−$3,492	$4,309	$66,341
9	−$1,400	−$3,317	$2,909	$63,024
10	−$945	−$3,151	$1,964	$59,873
11	−$638	−$2,993	$1,326	$56,880
12	−$431	−$2,844	$895	$54,036

Outcome after the shock:

Alice's account remains but is reduced to 0.9% of her starting capital, leading to a state of effective bankruptcy. Any further losing streak could trigger margin calls.

By contrast, Bob’s account retains 54% of its initial capital, allowing him to continue trading and recover. His conservative sizing strategy has safeguarded his ability to survive the crisis.

This stress scenario illustrates that Kelly and its variants often prioritize growth during stable market conditions, potentially sacrificing safety during extreme events. Fixed-fractional sizing sacrifices short-term gains in favor of long-term survival.

Common mistakes: Misestimating edge, ignoring transaction costs, and over-leveraging

Mistake 1: Misestimating edge

Traders frequently overstate p and b due to:

• Data snooping. Rigorous optimizations based solely on historical data might yield inflated performance metrics not replicable in real conditions, with a 60% win rate potentially dropping to 52%.
• Survivorship bias. Backtesting often neglects data on halted stocks, bankrupt companies, or delisted instruments, leading to an inflated view of historical win rates.
• Ignoring shifting market regimes. A win rate of 55% in one year could fall to 45% in a different market environment due to changes in volatility, correlations, or policies.

Solution: Use out-of-sample testing. Reserve 20-30% of your historical data for validation and avoid its use during parameter optimization. Validate your p and b measures through real trading or genuinely unseen backtest data.

Mistake 2: Ignoring transaction costs and slippage

The Kelly Criterion does not explicitly account for:

• Commissions and fees, essential for frequent traders.
• Slippage, which refers to the differences between intended execution and actual prices.
• Bid-ask spreads that may arise during larger trades.
• Market impact, where sizeable orders can modify prices unfavorably.

These costs can diminish both b (reducing potential gains) and p (converting potential wins into losses due to costs).

Solution: Adjust Kelly calculations using net b after accounting for transaction costs. For instance, if the gross reward-to-risk ratio is 2:1 and the average transaction cost is $0.10 per trade, the net b might need modification to 1.85:1.

Mistake 3: Over-leveraging

One common mistake involves adopting full Kelly in live circumstances. Although theoretical Kelly maximizes growth, it lacks a safety net against estimation errors or extreme market disruptions.

Professional risk managers often recommend utilizing half-Kelly (half of the computed Kelly percentage) or quarter-Kelly (one-quarter) in live trading. This tactic trades off some growth for a substantially reduced risk of total loss.

Solution: Cap the maximum employed Kelly fraction at no more than 50% of the calculated Kelly percentage. For a Kelly recommendation of 32.5%, limit risk to 16.25%. Likewise, for a recommendation of 10%, the maximum should be 5%.

FAQs

Q: Should I ever use full Kelly?

A: Full Kelly is seldom advantageous in live trading. While it optimizes theoretical growth, it relies on precise inputs and neglects the risks associated with Black Swan events. It is primarily valuable for academic inquiries or backtesting to comprehend the upper limits of growth. In live accounts, half-Kelly or quarter-Kelly* balances growth potential with survival risk. Learn more about rule-based frameworks that reduce risk in funded environments.

Q: How often should I recalculate Kelly inputs?

A: Recalculate the win probability (p) and win/loss ratio (b) at least quarterly, utilizing data from your most recent 100-200 trades. If there’s a noticeable shift in market conditions (like heightened volatility or major policy changes), consider updating monthly. If p or b deviates by over 5%, rerun your stress test and adjust your position size accordingly.

Q: What is a safe fixed-fractional position sizing rule?

A: A conservative fixed-fractional sizing rule suggests risking 2-5% per trade. For a $100,000 account, this translates to risking $2,000 or $5,000 per trade under a 2:1 reward-to-risk ratio. This approach allows for automatic adjustments to risk after losses, facilitating straightforward monitoring and minimizing the challenges in edge estimation.

Q: Can I combine Kelly with volatility-adjusted sizing?

A: Yes. First, calculate Kelly based on a standard volatility environment, then proportionally reduce the position size as realized volatility rises. For example, if volatility doubles, a reduction of position size by half may be prudent.