Staking: flat, percentage or Kelly — which suits you
You can have the best model in the world detecting value and still lose money if you size your bets badly. Staking —how much you put on each play— is the forgotten half of the game. Most people obsess over what to bet and neglect how much, which is exactly what decides whether you survive a bad streak or not.
There are three staking methods that dominate the landscape: flat stake, percentage stake and Kelly. None is "the best" in the abstract; each fits a different level of experience, risk tolerance and model quality. Let's compare them with numbers and give you a clear recommendation.
The three methods in a table
| Method | How it works | Pros | Cons | For whom |
|---|---|---|---|---|
| Flat | Same fixed amount on each bet (e.g. $100 always) | Simple, disciplined, easy to audit | Ignores your confidence level and the real value | Beginners, validating a new model |
| Percentage | A fixed % of the current bankroll (e.g. 2%) | Adapts to the bankroll, protects in bad streaks | Doesn't distinguish a bet with little or much edge | Intermediate level, solid management |
| Kelly | Variable % based on your edge and the odds | Mathematically optimal for growth | Requires estimating the probability well; high variance | Advanced with a validated model |
Flat stake: the disciplined base
You bet the same amount on each play, regardless of the odds or your confidence. If your unit is $100, you bet $100 at odds of 1.80 and $100 at odds of 4.00.
Example: bankroll of $5,000, unit of $100 (2%). After 100 bets with a yield of 5%, your profit is 100 × $100 × 0.05 = $500, regardless of the order of the wins.
The great virtue of the flat stake is that it isolates the quality of your selections. Since the size never changes, your result purely reflects whether you pick well. That's why it's the ideal method to validate a new model before putting more capital into it: if you're not profitable at flat stake, you won't be with anything more sophisticated.
If you're starting out, the flat stake isn't a "rookie" method you quickly graduate from. It's the most honest way to know if you have a real edge.
Percentage stake: the bankroll breathes
You bet a fixed percentage of the current bankroll, which you recalculate as it goes up or down. If your stake is 2% and you have $5,000, you bet $100. If the bankroll drops to $4,000, your next bet is $80.
This creates a natural protective effect: you bet less when you're losing and more when you're winning. It's almost impossible to go completely bust, because the amounts shrink as the bankroll drops.
Comparative example of a drawdown. Imagine 5 losses in a row:
- Flat (unit $100): you lose $500 fixed. From $5,000 you drop to $4,500.
- Percentage (2%): you lose $100, then $98, $96... the damage is cushioned and you drop to about $4,520.
The difference looks minor over 5 bets, but over long streaks the percentage method preserves much more capital and speeds up recovery when the bankroll rises again.
Kelly: the mathematical optimum (with an asterisk)
Kelly goes a step further: the percentage varies according to the value of each specific bet. More edge and better odds, you bet more; less edge, you bet less. It's the only one of the three that incorporates how much edge you have on each play.
The problem is that Kelly amplifies your estimation errors. If you think you have more edge than the real one, you overbet. That's why practically everyone uses fractional Kelly (half or quarter Kelly) to tame the variance. We develop it in depth in the Kelly criterion explained simply.
Example: with your model giving 55% probability and odds of 2.10, full Kelly calls for ~14% of the bankroll. Half Kelly: ~7%. Compare it with the fixed 2% of the percentage method: Kelly makes you bet much more when the value is big, which accelerates growth... and the scares.
Recommendation based on your level
Straight to the point:
- You're starting out or your model isn't proven → flat stake at 1–2% of your bankroll. Period. You need clean data to know if you have an edge before complicating things.
- You've had months of records and you're profitable at flat stake → move to percentage (2–3%). You gain protection in drawdowns without taking on Kelly's variance.
- You have a validated model, you measure your CLV and you tolerate variance → half or quarter Kelly. Here you really squeeze your edge.
The most common trap is jumping straight to Kelly without having validated anything: you end up betting hard on an imaginary edge.
At EDGE we accompany each value bet with a calculator that supports all three methods —flat, percentage and fractional Kelly— so you choose based on your moment, not the fashion. It's part of the "measured without hype" DNA: correct staking is invisible when you win and saves your life when you lose.
Choose the method that matches your real level, not the one you'd like to have. And record everything: without a bet tracking you won't know whether your staking is helping or getting in the way.
EDGE is an analysis tool, not a bookmaker. Betting carries risk. 18+. Play responsibly.
Put this into practice with EDGE
EDGE's AI finds the value for you and measures your edge with no smoke. Start free, in paper mode.