You have a setup that wins 58% of the time with a 1.6:1 average reward-to-risk ratio. How much of your account should you put on it? Risk 1% and you leave money on the table. Risk 10% and three consecutive losers wipe out a month of gains. The Kelly criterion gives you an exact answer to this question, and the answer is almost always larger than you expect.
Kelly criterion position sizing connects two things most traders treat separately: your edge (how often you win and by how much) and your bet size (how much capital you commit per trade). The formula was originally developed by John Kelly at Bell Labs in 1956 for signal noise problems in telecommunications. Ed Thorp brought it into gambling and finance, and Ralph Vince later extended the concept into optimal-f for trading portfolios. The math is simple. The practical application for swing traders requires several modifications that the textbook version does not mention.
I started using a fractional Kelly framework about two years ago after blowing past my quarterly drawdown limit twice. The formula itself was not the breakthrough. The discipline of measuring my actual win rate and payoff ratio before sizing was.
The Kelly Formula and What Each Input Means
The standard Kelly fraction for a simple win/loss bet is:
f^* = \frac{p \cdot b - q}{b}
Where f^* is the fraction of capital to risk, p is your win probability, q = 1 - p is your loss probability, and b is the ratio of average win to average loss (your payoff ratio).
Each variable deserves scrutiny. Win probability is not your backtest win rate across all conditions. It is your win rate for the specific setup type you are about to trade, measured over a meaningful sample. If you trade pullback-to-moving-average entries separately from breakout entries, each setup has its own p. Combining them produces a blended number that accurately represents neither.
The payoff ratio b is average winner divided by average loser, measured in R-multiples or dollar amounts. A payoff ratio of 1.5 means your average win is 1.5 times your average loss. This number is sensitive to outliers. One outsized winner can inflate your average and push Kelly’s recommendation higher than your typical trade justifies.
Worked Example: Estimating Your Kelly Fraction
Suppose you review your last 80 swing trades on breakout setups. You find a 55% win rate and a 1.5:1 average payoff ratio. Plugging in:
f^* = \frac{0.55 \times 1.5 - 0.45}{1.5} = \frac{0.825 - 0.45}{1.5} = \frac{0.375}{1.5} = 0.25
Kelly says to risk 25% of your capital on this trade. Let that sink in. A quarter of your account on a single swing position. On a $100,000 account that is $25,000 at risk. If your stop is 5% below entry, that means a $500,000 notional position. Five times your account on one trade.
This is mathematically optimal for long-run geometric growth if your estimates are perfectly accurate, your edge is perfectly stable, and you can tolerate 50-80% drawdowns along the way. None of these conditions hold for a swing trader.
Why Full Kelly Is Too Aggressive for Swing Trading
Full Kelly maximizes the long-run growth rate of capital. It does not minimize drawdowns. It does not account for the psychological reality of watching your account drop 60% before recovering. And it assumes you know your true edge with perfect precision.
The practical problems stack up fast. First, drawdowns under full Kelly are severe. Simulation studies consistently show that full Kelly produces peak-to-trough drawdowns of 50% or more, even with a genuine edge. Ed Thorp himself noted that full Kelly is “too aggressive for most investors” and recommended half-Kelly as a practical maximum.
Second, your edge estimate contains error. You are estimating p and b from a finite sample of past trades. With 80 trades, your win rate has a standard error of roughly \sqrt{p(1-p)/n}, which for p = 0.55 and n = 80 gives about 5.6 percentage points. Your true win rate could plausibly be anywhere from 49% to 61%. At p = 0.49 with a 1.5 payoff ratio, Kelly recommends risking only 0.7% of capital. At p = 0.55, it recommends 25%. That is a 35x difference from a 6-point uncertainty in your win rate estimate.
Third, Kelly assumes each trade is independent. Swing trades are not. If the market gaps down hard, your long breakout in AAPL and your long breakout in MSFT both lose simultaneously. Correlated losses magnify drawdowns beyond what the Kelly formula predicts for any single position.
Fractional Kelly: The Practical Fix
The standard remedy is to use a fraction of the Kelly bet. Half-Kelly (risking 50% of what full Kelly recommends) reduces the growth rate by only 25% while cutting drawdowns roughly in half. Quarter-Kelly cuts growth by about 44% but produces drawdowns that most traders can actually survive.
Using the same example with f* = 0.25:
Half-Kelly: 0.50 \times 0.25 = 0.125 or 12.5% of capital at risk.
Quarter-Kelly: 0.25 \times 0.25 = 0.0625 or 6.25% of capital at risk.
On a $100,000 account with a 5% stop distance, quarter-Kelly puts you in a $125,000 notional position. That is still a concentrated bet by most standards, but it is survivable. Three consecutive losses at quarter-Kelly cost you about 17% of your account. Painful, but recoverable. Three consecutive losses at full Kelly cost you about 58%. Most traders quit before recovering from that.
I run quarter-Kelly as my baseline. In practice, even that gets further reduced by the drawdown cap and correlation adjustment I describe below. My actual position sizes typically land between 2% and 5% of equity at risk, which is where most professional swing traders operate regardless of what formula they use.
Adding a Maximum Account-Risk Cap
Kelly tells you the optimal size for a single independent bet. Swing traders carry multiple positions simultaneously. If you have four positions each sized at quarter-Kelly (6.25%), your total portfolio risk is 25%. A correlated selloff that stops out all four positions takes a quarter of your account in one day.
The fix is a hard cap on total account risk. I use 10% maximum total risk across all open positions. If I have two positions each consuming 4% risk, the third position can use at most 2%, regardless of what Kelly recommends for that setup.
The math for per-position risk with a total cap:
f_{\text{position}} = \min\left(\text{Kelly fraction}, \frac{\text{Risk budget remaining}}{\text{Number of remaining slots}}\right)
Suppose your Kelly fraction says 6% per trade, your total cap is 10%, and you already have one position consuming 4% risk. Your remaining budget is 6%. If you want room for two more trades, each gets at most 3%, not the 6% Kelly recommends.
This creates a tension: Kelly says bet big, but portfolio risk management says bet small. The right answer depends on how correlated your positions are. If you trade across uncorrelated sectors and setups, you can push closer to the Kelly number. If your positions move together in a drawdown, the cap saves you.
Drawdown Budgets and When to Reduce Size
A drawdown budget sets a maximum acceptable drawdown for a defined period, typically a month or a quarter. Mine is 8% per month and 15% per quarter. When I hit 5% monthly drawdown, I cut all position sizes in half. When I hit 8%, I stop trading for the rest of the month.
This is not a Kelly modification. It is a circuit breaker that overrides Kelly when results suggest your edge has degraded or market conditions have shifted. The connection between drawdown budgets and Kelly is this: if your actual results diverge sharply from what your Kelly inputs predict, your inputs are probably wrong. Cutting size while you reassess is the rational response.
A simple drawdown scalar:
S_{\text{dd}} = \max\left(0, 1 - \frac{\text{Current drawdown}}{\text{Drawdown limit}}\right)
At zero drawdown, S_{\text{dd}} = 1 and you size normally. At 50% of your drawdown limit, S_{\text{dd}} = 0.5 and your sizes halve. At 100% of your drawdown limit, S_{\text{dd}} = 0 and you stop. This is a smoother version of the binary cut I described above.
If you want to layer volatility regime awareness on top of Kelly and drawdown budgets, the VIX-regime position sizing framework applies a regime scalar that further adjusts your fractional Kelly output based on whether the VIX sits in a calm, normal, or stressed state.
Handling Uncertain Edge: What to Do When You Do Not Trust Your Numbers
Every Kelly calculation assumes you know p and b. You never do. The question is how wrong you might be and what happens to your sizing when you are.
Start by computing Kelly at the lower bound of your confidence interval. If your 80-trade sample gives a 55% win rate, compute Kelly at 50% (the lower end of a rough confidence band). With p = 0.50 and b = 1.5:
f^* = \frac{0.50 \times 1.5 - 0.50}{1.5} = \frac{0.75 - 0.50}{1.5} = \frac{0.25}{1.5} = 0.167
Compare this to the 0.25 you get at p = 0.55. Using the conservative estimate and applying quarter-Kelly gives 0.25 \times 0.167 = 0.042, or about 4.2% of capital at risk. That is a reasonable swing-trade size.
Another approach: compute Kelly with your best estimates but then apply a “confidence discount” based on sample size. With fewer than 30 trades of a given setup, use one-fifth Kelly at most. Between 30 and 100 trades, use up to quarter-Kelly. Above 100 trades with consistent results, consider up to half-Kelly. These are not rigorous statistical thresholds. They are practical guardrails I have found useful.
The worst outcome is not using too little size. It is using too much size on an edge that does not exist. When Kelly returns a negative number, your expected value is negative. Do not take the trade. When Kelly returns a very small positive number (below 2%), the edge is probably noise. Trade it small or skip it.
A Complete Sizing Workflow for a Single Trade
Here is how I size a swing trade from setup identification to order entry:
Step 1: Identify the setup type (breakout, pullback, reversal). Look up my trailing 60-trade stats for that setup type. Current breakout numbers: 57% win rate, 1.4:1 payoff.
Step 2: Compute Kelly.
f^* = \frac{0.57 \times 1.4 - 0.43}{1.4} = \frac{0.798 - 0.43}{1.4} = \frac{0.368}{1.4} = 0.263
Step 3: Apply quarter-Kelly. 0.25 \times 0.263 = 0.066, or 6.6% of capital at risk.
Step 4: Check the drawdown budget. I am currently down 3% for the month against an 8% limit. My drawdown scalar: S_{\text{dd}} = 1 - (3/8) = 0.625. Adjusted risk: 0.066 \times 0.625 = 0.041, or 4.1%.
Step 5: Check total portfolio risk. I have two open positions consuming 5% combined risk. My 10% cap leaves 5% available. My 4.1% Kelly-adjusted risk fits inside the remaining budget. Proceed.
Step 6: Convert risk percentage to position size. On a $100,000 account, 4.1% risk = $4,100. If my stop distance is 4% below entry, position size = $4,100 / 0.04 = $102,500 notional. That is roughly one full-sized position.
Step 7: Sanity check. Is this position more than 20% of my account value? If yes, cap it there regardless. Is the stock liquid enough to exit at my stop price without major slippage? If not, reduce further.
The entire process takes about two minutes once you have current stats. I keep a spreadsheet with trailing setup-type performance that updates automatically from my trade log.
Common Mistakes with Kelly Sizing
Using backtest stats instead of live trade stats. Backtests overestimate edge because they include hindsight in pattern selection. Your live win rate is probably 5-10 percentage points lower than your backtest win rate. Size from live results.
Ignoring the payoff ratio and optimizing only for win rate. A 70% win rate with a 0.5:1 payoff ratio gives a Kelly fraction of zero. You have no edge despite winning most trades. The payoff ratio matters as much as the win rate.
Updating stats too frequently. Recalculating Kelly after every trade creates noisy oscillations in position size. Use a rolling window of at least 40-60 trades per setup type. Update monthly, not daily.
Applying one Kelly fraction to all setup types. Your breakout setups and your pullback setups almost certainly have different win rates and payoff ratios. A single blended Kelly fraction oversizes the weaker setup and undersizes the stronger one.
Forgetting that Kelly assumes reinvestment. The formula assumes you immediately reinvest all gains and losses compound continuously. In practice, most swing traders withdraw profits periodically and do not let the account compound indefinitely. This makes full Kelly even more aggressive than the theory suggests.
Kelly Sizing When the Edge Comes and Goes
Swing trading edges are not stationary. A breakout strategy that works well in trending, moderate-volatility markets might show a negative Kelly fraction during choppy, range-bound conditions. The drawdown-quality momentum filter is one way to measure whether conditions favor your setup before you size the trade.
If you track your setup-type performance by market regime, you can compute conditional Kelly fractions. My breakout setups show a 62% win rate with a 1.6 payoff when the Ulcer Index on SPY is below 5 (calm, trending market). When the Ulcer Index exceeds 10, the win rate drops to 48% with a 1.2 payoff. Kelly at the favorable conditions:
f^* = \frac{0.62 \times 1.6 - 0.38}{1.6} = \frac{0.992 - 0.38}{1.6} = \frac{0.612}{1.6} = 0.383
Kelly at the unfavorable conditions:
f^* = \frac{0.48 \times 1.2 - 0.52}{1.2} = \frac{0.576 - 0.52}{1.2} = \frac{0.056}{1.2} = 0.047
The difference is massive. In favorable conditions, even quarter-Kelly gives 9.6% risk. In unfavorable conditions, quarter-Kelly gives 1.2% risk, barely worth trading. This is the Kelly framework telling you what your instinct already suspects: trade big when conditions favor you, trade tiny or sit out when they do not.
Combining conditional Kelly fractions with VIX regime scalars gives you a two-layer filter: the Kelly fraction adjusts for setup-specific edge, and the regime scalar adjusts for market-wide volatility conditions. The two are not redundant. You can have a high-edge setup in a high-vol market (size moderately) or a low-edge setup in a calm market (size small or skip).
Position Sizing Is the Only Edge You Fully Control
You cannot control whether your next trade wins. You cannot control whether the market gaps overnight. You cannot control whether your 57% win rate holds next quarter. The one thing you control completely is how much you risk on each trade.
Kelly criterion position sizing turns that control into a framework. Not a rigid formula you follow blindly, but a starting point that forces you to quantify your edge before committing capital. The fractional Kelly with drawdown caps described here will not make you rich quickly. It will keep you in the game long enough for your edge to compound.
The traders who blow up are rarely the ones with bad setups. They are the ones who sized their good setups as if their edge estimates were certain.
Educational content only. Not investment advice. Trading involves risk. You are responsible for your decisions.