Optimise Your Stakes with the Kelly Criterion

The Kelly Criterion Formula: A Step-by-Step Breakdown

The Kelly formula determines the optimal fraction of your bankroll to wager on a single bet:

Where:

• f* = the fraction of your bankroll to wager
• b = net odds received (decimal odds − 1)
• p = your estimated probability of winning
• q = probability of losing (1 − p)

The formula can also be written as f* = (p × b − q) / b or, in its more intuitive form: f* = edge / odds, where edge = (p × b) − q.

Worked Example 1: A Standard Even-Money Bet

You estimate a Premier League match result has a 55% chance of going your way. The best available odds are 2.00 (evens).

• b = 2.00 − 1 = 1.0
• p = 0.55
• q = 0.45
• f* = (1.0 × 0.55 − 0.45) / 1.0 = 0.10 or 10%

Kelly recommends betting 10% of your bankroll. On a €5,000 bankroll, that is €500 per bet.

Worked Example 2: Odds-On Selection

You estimate a 70% probability of winning at decimal odds of 1.60.

• b = 1.60 − 1 = 0.60
• p = 0.70
• q = 0.30
• f* = (0.60 × 0.70 − 0.30) / 0.60 = (0.42 − 0.30) / 0.60 = 0.20 or 20%

Worked Example 3: Slight Edge at Short Odds

You estimate a 52% probability on a bet priced at 2.10.

• b = 1.10
• p = 0.52
• q = 0.48
• f* = (1.10 × 0.52 − 0.48) / 1.10 = (0.572 − 0.48) / 1.10 = 0.084 or 8.4%

These numbers immediately reveal why most professionals use a fractional Kelly — even modest edges generate stake sizes that most bettors would find uncomfortably large.

Why Full Kelly Is Dangerous: The Case for Fractional Kelly

Kelly's formula maximises the expected logarithm of wealth, meaning it produces the highest long-term geometric growth rate. In theory, no other staking method outperforms full Kelly over an infinite number of bets with perfect probability estimates. In practice, full Kelly is used almost nowhere among serious bettors.

The core problem is that full Kelly is incredibly sensitive to overestimating your edge. If your true probability is 52% but you estimate it at 55%, you are not just slightly over-betting — you are betting roughly three times the mathematically optimal amount. Probability estimation errors of 2–5 percentage points are not exceptional; they are normal. Even professional models miss by this margin regularly.

The empirical consequences are severe. A bettor using full Kelly with a 55% win rate at evens will, on average, experience a 50%+ drawdown at some point — not as an unlikely disaster scenario, but as a mathematical expectation. The probability of halving your bankroll before doubling it using full Kelly is exactly 1 in 3. Over long betting careers, a significant majority of full Kelly bettors experience a drawdown that would have ended their betting activity had they not had very deep pockets and iron discipline.

Fractional Kelly in Numbers

The relationship between Kelly fraction, growth rate, and drawdown risk is non-linear and heavily favours fractional use:

The critical insight: moving from full Kelly to half Kelly costs you 25% of theoretical growth, but reduces your chance of a bankroll-halving event from 33% to 11%. Moving to quarter Kelly costs you 56% of theoretical growth, but that chance drops to 1.2%. For a bettor whose edge depends on staying solvent and not being emotionally derailed by severe drawdowns, this trade-off is extraordinarily favourable.

Almost every serious professional bettor uses between half Kelly and quarter Kelly. Many use a hard cap of 2–2.5% of bankroll per bet regardless of what the formula calculates, effectively implementing a fractional Kelly with a ceiling.

Estimating Your Edge: The Role of Sharp Book Odds and CLV

The Kelly formula is only as useful as the probability estimate fed into it. If your edge estimate is wrong, Kelly doesn't just give you suboptimal results — it amplifies the error directly into stake sizing. Accurate probability estimation is therefore not peripheral to the Kelly Criterion; it is the entire foundation.

The most reliable tool for edge estimation in sports betting is closing line value (CLV). Pinnacle's (PS3838's) closing line is the industry benchmark for what the true probability of an outcome was at the time of market close. Because Pinnacle accepts sharp money and adjusts its lines in response to informed betting, their closing prices are closer to true probability than any other bookmaker's odds.

The De-Vig Process

Every bookmaker's odds contain a margin that inflates the implied probability beyond 100%. To extract the true implied probability, you must de-vig the line. The Power method is the most accurate for this purpose:

1. Convert each outcome's odds to implied probability: 1 / decimal odds
2. Find the overround: sum of all implied probabilities (e.g., 1.05 for a 5% margin)
3. Apply Power normalisation to account for favourite-longshot bias
4. The resulting probabilities sum to 100% — these are the "true" market estimates

Pinnacle's 2–3% margin on major markets makes this process far more accurate than applying it to a soft book running 8–10% margins. At lower margins, the uncertainty introduced by de-vigging is smaller, meaning your edge estimates are more reliable.

Practical workflow:

1. Form your probability estimate from research, models, or a combination
2. De-vig Pinnacle's current line to get the market's best probability estimate
3. Your edge = your probability − market's true implied probability
4. Only bet when edge > 1.5–2% to account for estimation uncertainty
5. Apply fractional Kelly to size the bet
6. At game end, note the closing line and calculate your actual CLV

If you track CLV consistently and find you are averaging +2–3% CLV across hundreds of bets, your probability estimation process is working. If your CLV is negative on average, your model is systematically overestimating edge regardless of short-term results.

Why Broker Access Improves This Process

Accessing PS3838 through a betting broker gives you continuous visibility into the sharpest lines in the market — the benchmark against which all probability estimates should be validated. Pinnacle's opening lines move in response to sharp action within minutes, creating an extremely efficient market. Observing how your assessments compare to Pinnacle's movements over time is one of the best calibration tools available to a serious bettor.

Bankroll Management Strategies: Flat, Proportional, and Kelly Compared

Flat Staking

Flat staking means betting the same amount on every wager regardless of perceived edge, odds, or bankroll size. It is the simplest approach and has genuine merits: it is easy to track, emotionally predictable, and protects against the most common mistake of betting too much when you feel confident.

The limitations are significant. Flat staking does not exploit the compounding effect — as your bankroll grows, your flat stake becomes proportionally smaller, reducing your growth rate. Conversely, after losses, your fixed stake represents a larger percentage of a smaller bankroll, amplifying drawdown risk. It also allocates identical amounts to high-confidence and low-confidence selections, which is suboptimal.

Proportional Staking

Proportional staking bets a fixed percentage of the current bankroll on every bet (e.g., always 1% of whatever the bankroll is today). This automatically adjusts for bankroll growth and losses, producing more stable percentage drawdowns. It is better than flat staking but does not account for edge size — a 5% edge and a 0.5% edge receive identical stake percentages, which is mathematically suboptimal.

Kelly Staking

Kelly staking bets proportionally based on both bankroll size and the size of the perceived edge. Larger edge = larger stake percentage; smaller edge = smaller stake. This is mathematically optimal for long-term growth, subject to the edge estimation caveats discussed above.

Bankroll Simulations: What the Numbers Actually Show

Theory is useful; simulation is convincing. Consider a bettor with a genuine 55% win rate at even-money (2.00) odds — a realistic edge for a skilled professional. Kelly recommends a 10% stake per bet. The following simulations run 500 bets under each staking approach starting with a €5,000 bankroll:

Full Kelly (10% stake per bet)

Starting stake: €500. After a good run of 500 bets, the bankroll can reach €40,000–€80,000. But the journey is brutal: drawdowns of 40–60% occur multiple times per 500-bet sample. A typical 10-bet losing run at full Kelly on a €10,000 bankroll takes it to €3,487 — a 65% drawdown. Recovery takes another 15–20 bets at minimum. Most bettors would reduce stakes, panic, or stop betting entirely during such a drawdown, negating the theoretical advantage entirely.

Half Kelly (5% stake per bet)

Starting stake: €250. After 500 bets, the median outcome is €20,000–€30,000. The same 10-bet losing run on a €10,000 bankroll takes it to €5,987 — a 40% decline, still uncomfortable but not catastrophic. Recovery typically requires 10–12 bets. This is the most popular professional approach: the growth rate is substantial, and the drawdowns are survivable.

Quarter Kelly (2.5% stake per bet)

Starting stake: €125. After 500 bets, the median outcome is €12,000–€18,000. The 10-bet losing run produces a drawdown to €7,736 — a 22% decline. Recovery requires 5–7 bets. This is the preferred approach for bettors who are building a bankroll, operating under uncertainty about their true edge, or cannot tolerate the psychological pressure of larger drawdowns.

Flat €100 per bet

After 500 bets at 55% win rate, expected net profit is 500 × 0.10 × €100 = €5,000, taking the bankroll from €5,000 to €10,000. Linear, predictable, but it fails to compound — a bettor who built their bankroll to €15,000 on half Kelly stakes would be earning 3x more per bet than the flat €100 bettor, despite starting at the same place.

Handling Multiple Simultaneous Bets

Standard Kelly assumes bets are placed sequentially — one at a time. In practice, sharp bettors often have multiple open positions simultaneously across different sports, leagues, and markets. This creates a problem: if you independently apply Kelly to each simultaneous bet, the sum of all stake percentages can easily exceed 100% of your bankroll, especially on days with many qualifying opportunities.

The correct approaches depend on the correlation between bets:

• Fully independent bets: Apply fractional Kelly per bet and ensure the sum of all open positions does not exceed 100% of bankroll. Reduce each individual bet proportionally if the total would exceed this ceiling.
• Correlated bets: Bets on related outcomes (same match, same league, outcomes that move together) require lower individual stakes than Kelly suggests for each in isolation. For example, betting on a team to win and over 2.5 goals in the same match shares common outcome risk — a quarter Kelly allocation to each would effectively create a half Kelly exposure to the same match.
• Conservative universal cap: Many professionals simply set a hard rule of no more than 2% of bankroll per bet regardless of Kelly output, and accept a maximum of 20–30% total exposure across all open positions. This is simple, robust, and protects against both over-betting and model errors.

Common Kelly Criterion Mistakes and How to Avoid Them

Overestimating Edge

The most damaging error. If your true edge is 2% but you estimate it at 5%, Kelly instructs a stake 2.5× higher than optimal. Studies of professional bettor models suggest overestimation of edge by a factor of 2× is common, particularly before 500+ bets of data have been accumulated. The remedy: be conservative in edge estimation, require CLV confirmation over a large sample before trusting model output, and use fractional Kelly to buffer the error.

Using a Small Sample to Confirm Edge

50 bets with a 60% win rate feels compelling. Statistically, it tells you almost nothing — that result is within normal variance even for a bettor with zero edge. Professional validation requires 200–500 bets minimum at similar odds before updating your edge estimate upward, and 1,000+ bets for high confidence. Track CLV throughout: CLV data becomes meaningful much faster than win rate because it measures your ability to beat the market, not just to win in general.

Ignoring the Closing Line

Treating every model output as equally trustworthy is a mistake. PS3838's closing line represents the collective judgment of the sharpest money in the market. If your selections systematically receive worse odds than the closing line, your model is not generating real edge — it is most likely identifying phantom patterns in the data. If you are consistently beating the closing line, that is the strongest signal that your probability estimates are valuable. Access to PS3838 via a betting broker makes this validation process routine rather than exceptional.

Not Recalculating After Bankroll Changes

Kelly stakes are proportional to current bankroll, not starting bankroll. After a significant drawdown or a significant run of profits, recalculate your stake sizes. A bettor who started with €5,000, grew to €12,000, but continues staking at the €5,000 level is systematically under-betting. Conversely, a bettor who has dropped to €3,000 and continues staking at €5,000 levels is over-betting — this is one of the most common paths to ruin.

Betting More Than One Edge Type Simultaneously

If you run a value betting strategy and an arbitrage strategy simultaneously from the same bankroll, calculate your total exposure carefully. The combined strategy may have very different Kelly properties than either strategy in isolation. Most professionals separate their value betting and arb activities conceptually even when using a single wallet, and apply conservative total exposure limits to the combined book.

Practical Bankroll Guidelines for Irish Bettors

Translating Kelly theory into practical guidelines depends on your strategy, risk tolerance, and bankroll size:

The minimum bankroll figures assume you intend to place 5–15 bets per week. Below the stated minimums, transaction costs, minimum stake requirements, and the practical difficulty of proper stake sizing make Kelly-based management impractical. The minimum is not what you need to start — it is what you need to run the strategy correctly.

For Irish bettors, access to PS3838 via brokers like BetInAsia or AsianConnect provides both the line access needed for CLV benchmarking and the high limits required to bet the correct Kelly stake without facing rejection. Soft bookmakers will refuse or restrict large stakes from winning bettors before the Kelly amounts become significant; sharp book access through a broker eliminates this limitation. Our broker comparison guide details each option's stake limits and account policies.

How Single-Wallet Broker Access Improves Kelly Efficiency

Bankroll fragmentation is one of the least-discussed enemies of Kelly efficiency. If a bettor's capital is split across eight separate bookmaker accounts plus two exchange accounts, the practical bankroll available for any single opportunity is a fraction of their total capital. A bet that Kelly identifies as a 2% stake cannot be executed at the correct size if only a fraction of the bankroll is in the relevant account.

Brokers like BetInAsia and AsianConnect operate on a single-wallet model: one deposit provides access to PS3838, SingBet, and multiple other books and exchanges simultaneously. This means your Kelly stake calculations are based on your full available capital, not a portion of it. Capital efficiency improves dramatically, and the operational overhead of managing multiple account balances is eliminated.

Additionally, brokers' access to multiple sharp books simultaneously means you can shop for the best odds on each selection — a PS3838 line on one book, an exchange price on another, all available in the same platform. Better odds on the same probability estimate directly increases your Kelly-calculated stake (because edge is larger), compounding the benefit.