How to Size Positions with Kelly Under Uncertainty
The Kelly criterion gives the bet size that maximizes long-run growth when you know your edge exactly. In trading you never know it exactly, and that gap is dangerous: full Kelly on an overestimated edge overbets badly and can destroy capital. The practical question is not what full Kelly says but how much to discount it for uncertainty. The edge estimation, the fractional discount, and the drawdown check that keep sizing survivable are covered below.
On This Page
Before You Start
Set up the inputs that make the next steps easier
Guide Steps
Move through it in order
Each step focuses on one decision so you can keep momentum without losing the thread.
- 1
Estimate the edge and its uncertainty
Kelly needs an expected edge, but the honest version also needs how confident you are in that edge. Estimate both from out-of-sample data, not the in-sample fit that produced the strategy. A small sample or a wide spread of results means high uncertainty, which should pull your bet size down. Treating the edge as a point estimate when it is really a distribution is the root cause of Kelly overbetting.
Use the lower end of a confidence interval for the edge, not the central estimate. Sizing on the optimistic case is how good strategies get overbet into ruin.
Use The ToolCalculatorsPosition Sizing under Edge Variance
Bayesian-Kelly bet sizing when your edge is itself uncertain. Compare deterministic Kelly, Bayesian-adjusted, and conservative lower-bound versions.
ToolOpen -> - 2
Compute the full Kelly fraction
Full Kelly is the bet size that maximizes the expected logarithm of wealth. For a simple bet it is the edge divided by the odds; for a continuous strategy it is roughly the expected excess return divided by the variance of returns. This number is the theoretical ceiling. It is correct only if your edge estimate is exact, which is why it is a starting point for discounting rather than a sizing decision in itself.
Full Kelly is famously aggressive: it routinely implies drawdowns of fifty percent or more even when the edge is real. Treat it as a maximum you will scale down from.
Use The ToolCalculatorsFractional Kelly Sizer
Map conviction tiers to fractional Kelly bet sizes with a drawdown Monte Carlo simulator. Client-side. Private by default.
ToolOpen -> - 3
Apply a fractional Kelly discount
Bet a fraction of full Kelly, commonly one half or one quarter. The reason is mathematical, not timid: because growth is concave near the Kelly optimum, halving the bet sacrifices only a small slice of expected growth while roughly halving the volatility of wealth. Under estimation uncertainty, fractional Kelly also corrects for the systematic overbetting that comes from sizing on an edge you measured with error.
The more uncertain your edge, the smaller the fraction. A well-measured edge from thousands of trades justifies a larger fraction than a fresh strategy with fifty trades.
- 4
Use a Bayesian-adjusted size when the edge is shaky
When the edge estimate is genuinely uncertain, a Bayesian approach formalizes the discount: it places a prior on the edge, updates it with your data, and sizes on the resulting posterior rather than a single point. With little data the posterior pulls heavily toward the prior, automatically shrinking the bet. As evidence accumulates, the size grows toward the data-driven Kelly. This is the principled version of fractional Kelly when you can quantify the uncertainty.
Compare the deterministic Kelly, the Bayesian-adjusted Kelly, and a conservative lower-bound version. The gap between them is a direct readout of how much your uncertainty is costing in size.
- 5
Stress the drawdown before committing the fraction
Even a correct Kelly fraction produces deep drawdowns, and the relevant question is whether you can survive them psychologically and financially. Simulate the path of wealth under your chosen fraction, given the strategy's return distribution including skew and kurtosis, and read the time and depth of the typical and tail drawdowns. If the recovery time or the depth would force you to abandon the strategy, cut the fraction until it would not.
Size to the drawdown you can sit through, not the one you can describe in a spreadsheet. The constraint that matters is the one that makes you capitulate at the bottom.
Use The ToolCalculatorsDrawdown-Recovery Markov Simulator
Time to recover from an N% drawdown given monthly Sharpe + skew + kurtosis. Cornish-Fisher Monte Carlo, percentile distribution of recovery months.
ToolOpen ->
Common Mistakes
The misses that undo good inputs
Sizing on the in-sample edge
The in-sample edge is inflated by the same selection that produced the strategy. Feeding it into Kelly overbets from the start, before any live uncertainty is even considered.
Treating full Kelly as the target rather than the ceiling
Full Kelly assumes a perfectly known edge. With real estimation error it overbets, and the resulting drawdowns are far deeper than the strategy's edge can justify, often leading to capitulation at the worst time.
Ignoring correlation across simultaneous positions
Kelly sizing each position independently overbets the portfolio when positions are correlated. Several correlated bets at individual-Kelly size behave like one oversized bet during a stress event.
Try These Tools
Run the numbers next
FAQ
Questions people ask next
The short answers readers usually want after the first pass.
Sources & References
- A New Interpretation of Information Rate — J. L. Kelly Jr., Bell System Technical Journal (1956)
- The Kelly Criterion in Blackjack, Sports Betting, and the Stock Market — Edward O. Thorp (2007)
Related Content
Keep the topic connected
Kelly Criterion
What the Kelly criterion is, when full Kelly blows up, and why most working quants size at half- or quarter-Kelly.
Kelly Sizer Worked Examples
Worked Kelly criterion examples: how win probability and payoff ratio set the optimal bet fraction, and why practitioners scale down to half or quarter Kelly.
Drawdown
Drawdown explained: peak-to-trough decline, why max drawdown alone is misleading, and the recovery math that actually matters.
Expected Shortfall (CVaR)
Expected shortfall: the average loss given a VaR breach. Why regulators are migrating from VaR and what ES catches that VaR misses.