How to Size Positions Without Gambling: Kelly Fraction Math Explained
Most traders size positions with vibes. We use math. Here's the difference between fractional Kelly sizing and blowing up your account.
You know the phrase "bet too big and you go broke, bet too small and you starve"?
That's the position sizing problem. And most traders solve it by... vibes.
"Yeah, I'll put 2% of my account on this one. Actually, make it 3%, I feel good about it."
That's not a strategy. That's gambling. The math for how much to actually risk is not complicated. Most people just don't use it.
The Kelly Criterion
A guy named John Kelly figured this out in 1956 while working at Bell Labs. Here's the formula:
`
f = (p * b - q) / b`
Where:
f= fraction of bankroll to bet
p = probability of winningb = ratio of bet winnings to stake (like 2:1 odds means b=2)q = probability of losing (1 - p)Example: You have a +55% edge on a trade (55% win rate), and if you win you make 2x your bet, if you lose you lose 1x.
`
f = (0.55 * 2 - 0.45) / 2
f = (1.1 - 0.45) / 2
f = 0.65 / 2
f = 0.325`
So you should bet 32.5% of your bankroll.
That seems insane, right? 32.5%? That's a lot.
Yeah. Kelly is aggressive. It maximizes long-term growth. It also gets you crushed in the near term if you hit a bad streak.
Fractional Kelly
Nobody uses full Kelly. If you hit a few losses in a row, full Kelly will destroy you even if the edge is real.
So you use a fraction. Maybe 0.5 Kelly, maybe 0.25 Kelly.
`
ffractional = fkelly * fraction`
If Kelly says 32.5%, then 0.5 Kelly says 16.25%, and 0.25 Kelly says 8%.
In Weatherman Trading, we use a conservative fractional Kelly — well below the theoretical maximum — and scale that fraction based on forecast confidence. Here's why:
How We Actually Do It in Weatherman Trading
We're more sophisticated than just plugging numbers into Kelly, but the principle is the same.
For each contract:
1. Calculate edge (expected value per contract):
EV = (P(win) payoutifcorrect) - (P(lose) stake) - fees
If EV is negative, we don't trade. If EV is too small relative to the fee structure and execution risk, we don't trade.
2. Estimate position size from Kelly:
Using P(win) and the payout, calculate what fractional Kelly says.
3. Apply hard constraints:
4. Check liquidity:
Don't size bigger than the top of the order book. We only take what the market can absorb without significant slippage.
5. Divergence adjustment:
If GFS and ECMWF disagree significantly, we shrink position size proportionally. If they disagree sharply, we don't trade that contract at all. The exact thresholds are calibrated to the specific contract markets we trade and are part of the system.
Why This Beats Vibes
If you size by vibes, you're inconsistent. You'll bet big on high-confidence trades (which is right) but you'll also bet big on trades that feel good but have worse edges (which is wrong).
Math doesn't care how a trade feels. It just calculates what's optimal given your edge estimate.
And here's the thing: even if your edge estimate is slightly off, consistent sizing beats inconsistent sizing. A +2% edge with good sizing beats a +5% edge with garbage sizing.
The Dirty Secret
Kelly sizing assumes you know your edge.
You usually don't.
Your backtest shows 55% win rate? Reality might be 51%. Your forecast model looks perfect on historical data? It's probably overfit.
So:
The Bottom Line
Position sizing math is boring. It won't make you rich. But it will keep you from going broke.
Vibes-based trading is how accounts get liquidated. Math-based trading is how they survive long enough to actually learn something.
We built Weatherman Trading on the assumption that if you have an edge, you should size rationally. And if you don't have an edge, you shouldn't trade at all.
Everything else is just noise.