Bankroll Basics: Flat Staking vs Kelly (How to Size Your Bets)
A bankroll is the money you set aside for betting. Staking is the rule you use to decide how much of that bankroll goes into a single bet. The same selection can produce very different outcomes depending on stake size.
What “flat” means in practice
Flat staking is simple: you bet the same amount each time, or the same small percentage of your bankroll (for example, 1% per bet). The goal is consistency: you keep exposure stable while short-term results fluctuate.
What “Kelly” is trying to do
Kelly staking sizes bets based on estimated edge. If your estimated win probability is meaningfully higher than the probability implied by the odds, Kelly allocates a larger fraction of bankroll. If the edge is small (or negative), Kelly reduces the stake (or says to skip).
The idea that matters most
- Odds are the market price (and the market’s implied probability).
- Your probability is an estimate and can be wrong.
- Staking risk is mostly the cost of being wrong about that estimate — not the cost of one bet.
Below, both methods are turned into a practical workflow: how to compute stakes, what can go wrong, and how to avoid the most common failure mode in staking — sizing too aggressively relative to uncertainty.
1) Flat staking: consistent sizing, controlled variance
Flat staking keeps bet sizes stable so variance does not dominate bankroll changes in the short run. It does not require precise edge estimates on every bet. The implicit assumption is simple: you want exposure to stay predictable while your selection process plays out over a larger sample.
Two common flat rules
- Fixed unit: bet the same amount each time (for example, $10), regardless of bankroll changes.
- Fixed percentage: bet a small percent of current bankroll (for example, 1% per bet), recalculated over time.
The tradeoff is that flat staking does not scale up with larger edges. It treats a marginal edge and a strong edge the same, which can be acceptable if you prefer simplicity and smoother bankroll behavior.
Main risks of flat staking
- Underbetting strong edges: if a subset of bets truly has higher edge, flat sizing does not allocate more to them.
- Unit drift: increasing the “flat” unit after a winning run is variable staking in disguise.
- Fixed units after drawdowns: if bankroll falls but unit stays fixed, risk per bet rises automatically.
2) Kelly staking: edge-based sizing (and the real failure modes)
Kelly staking sizes bets from your estimated probability. If your probability estimate is accurate and stable, Kelly aims to maximize long-run bankroll growth by betting more when the edge is larger and less when the edge is smaller.
Kelly formula (decimal odds)
Let O be decimal odds, b = O − 1, p your win probability estimate, and q = 1 − p. The full Kelly fraction is:
where b = O − 1, q = 1 − p
If f* is negative, Kelly indicates no positive edge at that price (stake = 0).
Where Kelly risk comes from
- Estimation error: if p is overstated, the Kelly fraction is overstated and the stake becomes too large.
- Variance and streaks: even with an edge, losing runs occur; large fractions translate into deeper drawdowns.
- Correlation: multiple bets can share the same drivers (model bias, league conditions, injuries). Treating them as independent inflates total exposure.
Because probabilities are estimates, many bettors use fractional Kelly (half-Kelly, quarter-Kelly) plus a maximum cap. Fractional Kelly preserves the direction of Kelly (bigger edge → bigger stake) while reducing sensitivity to estimation error.
3) Worked examples: flat vs Kelly (same bankroll, same odds)
The key step in Kelly sizing is comparing your probability estimate to the probability implied by the odds. For single-outcome markets, implied probability is often approximated as 1 / O.
Implied probability (quick approximation)
Example: odds 2.10 imply 1 / 2.10 = 0.4762 (47.62%). In real markets, prices may include margin (overround), and multi-way markets require normalization across outcomes.
Example A: small edge at 2.10
Here, b = 2.10 − 1 = 1.10, q = 1 − p = 0.50. Full Kelly:
f* = (0.55 − 0.50) / 1.10 = 0.04545… ≈ 4.55%
Flat 1% stake: $10.00 Full Kelly stake: $45.45 Half Kelly stake: $22.73
Even modest differences in p matter. If the true win probability is closer to 0.48 than 0.50, the Kelly fraction drops materially. That sensitivity is why fractional Kelly and caps are common in practice.
Example B: no edge (Kelly says “skip”)
Same odds 2.10, but your estimate is p = 0.46. The implied probability is 47.62%, so your estimate is below the price.
f* = (1.10·0.46 − 0.54) / 1.10
f* = (0.506 − 0.54) / 1.10 = −0.0309… (negative)
A negative fraction means the price does not justify a stake under your estimate. In practice: stake = $0. Flat staking can still place a bet if applied mechanically, which is why flat staking is typically paired with a “take / pass” filter.
Example C: bigger edge, bigger stake (and higher drawdown exposure)
Odds 1.80 imply 55.56%. If your estimate is p = 0.62, Kelly sizes up.
f* = (0.80·0.62 − 0.38) / 0.80
f* = (0.496 − 0.38) / 0.80 = 0.145 → 14.50%
A 14.50% fraction is large for most real-world workflows because it assumes your p is reliable. Small probability errors and correlated exposure can produce deep drawdowns. Fractional Kelly and stake caps reduce that sensitivity.
Note on push/refund markets
For lines where a push is possible (for example, some spreads/totals depending on the rules), the payoff structure changes. Kelly is still applicable, but the calculation should reflect the exact distribution of outcomes (win / push / loss) rather than a single win probability.
Flat vs Kelly: what each method assumes
| Topic | Flat staking | Kelly staking |
|---|---|---|
| Inputs | Bankroll + chosen unit or % | Bankroll + odds + probability estimate |
| Primary objective | Stable exposure | Edge-weighted exposure |
| Main failure mode | Sizing does not reflect differences in edge | Overbetting from optimistic/noisy probabilities and ignored correlation |
| Best fit | Simple rules and consistent risk per bet | Quantified edge with conservative multipliers and caps |
A practical sizing workflow (works for both)
- Decide “take / pass” first: do not size a bet until you have a reason the price is acceptable.
- Set a risk budget: many workflows treat ~0.5%–1.5% as a “standard” stake zone and require stronger evidence to go higher.
- Cap maximum stakes: a cap (for example, 2%–5% of bankroll) prevents one estimate error from dominating exposure.
- Control correlation: if several bets depend on the same assumption, reduce the combined stake rather than sizing them independently.
Kelly is not “automatically better” than flat. It is more responsive to edge estimates, and that responsiveness is only as good as the inputs. If your probability estimates are unstable, fractional Kelly or flat staking with a consistent filter usually produces more stable exposure.
Mini calculator: flat vs fractional Kelly (decimal odds)
This calculator converts inputs into a stake size. It does not evaluate whether a bet is “good” or “bad” — it only applies the sizing rules. If your probability estimate is uncertain, fractional Kelly plus a cap reduces sensitivity to estimation error.
Core formulas used
b = O − 1, q = 1 − p
full Kelly fraction f* = (b·p − q) / b
fractional Kelly stake = bankroll × max(0, f*) × kellyFraction
flat stake = bankroll × flatPercent
Interpretation: if full Kelly is negative, it means your estimate does not clear the price at those odds. Flat staking can still be used as a sizing rule, but only after you have a separate “take / pass” decision.
FAQ: bankroll and staking
1) What exactly counts as a bankroll?
Your bankroll is the money separated for betting only. Staking rules assume this amount can fluctuate with variance; it should not be mixed with essential expenses.
2) What is a reasonable flat stake size?
Many bettors implement flat staking as a small percentage per bet (often around 0.5%–2%). Higher-variance markets and higher uncertainty in selection quality typically require lower percentages.
3) What is the Kelly criterion in one sentence?
Kelly staking uses your estimated win probability and the odds to compute a bankroll fraction that allocates more stake to larger estimated edges.
4) Why do people use half-Kelly (or quarter-Kelly)?
Because probability estimates are imperfect. Fractional Kelly reduces exposure while keeping the edge-weighted logic of Kelly.
5) Can Kelly staking cause very large drawdowns?
Yes. Even with an edge, variance and correlation can produce long losing runs. If probability estimates are optimistic or correlated across bets, full Kelly can size too aggressively; fractional Kelly and caps reduce that sensitivity.
6) How should you size parlays with flat vs Kelly?
Parlays have higher variance and depend on multiple legs. Flat staking often treats them as smaller exposure. Kelly requires a probability estimate for the full parlay price; if that estimate is not reliable, sizing should be conservative.
7) If two bets both look like “value,” how do you choose stake?
Flat staking keeps sizes equal (or uses tiers). Kelly sizes by edge: larger estimated edge gets a larger fraction. If bets share drivers (same league/model assumptions), reduce combined exposure rather than treating them as independent.