Flat, Percentage or Kelly: Which One You Should Actually Use

Bet sizing does not change whether you are right. It changes whether you are still solvent by the time being right pays.

Three plans cover almost everything anyone actually uses. Here they are, with an opinion at the end that is not neutral.

Flat

The same stake every time, usually one unit set at one percent of your starting bankroll. Short price or long, confident or hesitant, one unit.

What it buys you is legibility. Your profit and loss in units becomes a clean statement about your selections, uncontaminated by sizing decisions, which means that after four months you can actually tell whether you are any good. Nothing else on this list gives you that.

The cost is real and worth naming. Flat staking will feel too slow, permanently. When you are right about something obvious you will wish you had bet more, and that feeling never goes away. It is the price of a bankroll that survives an ordinary seven-loss run without drama.

Percentage

The same idea, except the percentage recalculates off your current bankroll rather than the starting figure. Two percent of 1,000 is 20. After the roll grows to 1,200 it is 24. After it drops to 800 it is 16.

The mechanism is self-correcting. Winning compounds, losing automatically shrinks your exposure, and in the strict mathematical sense you cannot go broke, because you are always betting a fraction of what remains.

The trade-off arrives on the way back up. After a drawdown your stakes are smallest exactly when you would like them largest, so recovery is slower than the descent was. That asymmetry is a feature. It is what stops a bad month becoming a terminal one. People find it demoralising anyway and start overriding it, which defeats the plan entirely.

Kelly

Kelly sizes the bet to the edge: f = (bp − q) / b, where b is decimal odds minus one, p is your estimated win probability and q is 1 − p. Bigger edge, bigger stake.

It maximises the long-run growth rate of a bankroll. That is a theorem rather than a claim, and it holds under one condition: p has to be right.

That condition is where esports kills it. Your probability estimate is built on small samples, rosters that changed last month, and a patch that may have altered the game since the data was generated. Overestimate the edge slightly and full Kelly stakes are not slightly too large, they are dramatically too large, because the formula scales hardest on precisely the bets you are most likely to be overconfident about.

Which is why the people who genuinely use Kelly mostly use a half or a quarter of it. Fractional Kelly keeps most of the growth while cutting the volatility and the estimation damage sharply. If the formula returns zero or a negative number it is telling you there is no bet here. Believe it.

The opinion

Use flat until you have a few hundred logged bets saying your selections beat the closing line. Almost nobody reading this has that yet, and flat is the only plan that will tell you honestly whether you do.

Percentage after that, if you want the compounding and the automatic brake.

Kelly only with a tested model producing calibrated probabilities, and then at a fraction. Kelly applied to gut-feel percentages is not a staking plan, it is confidence with a formula attached to it.

And the thing all three share: none of them converts a losing method into a winning one. Better sizing on negative-edge bets makes you lose more slowly and more smoothly. That is a real benefit and it is not the benefit people are hoping for.