Half-Kelly vs. Full Kelly — Why Pros Rarely Stake the Full Amount

The Kelly criterion answers one question precisely: what fraction of bankroll maximizes long-run logarithmic growth? The formula f* = (b·p − q) / b is provably optimal. Yet almost no one in sports betting practice uses the full Kelly stake. The reason: optimal does not mean comfortable.

What full Kelly actually does: It maximizes the expected value of log(bankroll). That is not the same as 'maximizes profit after 100 bets'. Full Kelly tolerates massive drawdowns on the way there. Simulation: over 1,000 bets with a true 5% edge and full Kelly, the probability of at some point losing at least 50% of the bankroll is about 50%. Read that again — one in two bettors with full Kelly and real edge experiences a 50% drawdown.

The double effect of estimation error: Full Kelly assumes your probability estimate is exactly correct. In reality even the best bettor estimates with ±2–3% error. Overestimate your edge by just 2 percentage points and logarithmic growth collapses — while drawdowns explode. Full Kelly is a perfect strategy for a world with perfect estimates. That world doesn't exist.

Half-Kelly (0.5 × Kelly): You give up roughly 25% of the theoretical growth but halve the expected drawdown depth. In the same 1,000-bet simulation, the probability of a 50% drawdown drops to about 12%. Half-Kelly is the standard in the professional betting community because the risk-reward profile is meaningfully better than full Kelly.

Quarter-Kelly (0.25 × Kelly): You keep about 60% of full-Kelly growth, and the maximum drawdown shrinks again significantly. For bettors with uncertain models (i.e. nearly all private bettors), Quarter-Kelly is the safer entry point. On a £1,000 bankroll with 5% edge on odds 2.00, full Kelly is 5% = £50, Half-Kelly £25, Quarter-Kelly £12.50.

Mathematical justification: Growth as a function of the Kelly fraction is an inverted parabola with maximum at 1 (full Kelly). But the curve is relatively flat near the peak and drops steeply past it. At 1.5× Kelly you're near zero growth despite correct edge. At 0.5× you retain most of the growth. 'Better too little than too much' is mathematically correct, not just conservative.

Practical rules from the pro betting community: Very robust model (e.g. arbitrage, validated CLV tracking over 1,000+ bets) → Half-Kelly. Solid in-house model with measured edge → Quarter- to Half-Kelly. 'Feel' estimate even if methodical → Eighth-Kelly or flat staking. Beginners with no proven edge → no Kelly at all; flat 1–2% of bankroll.

Why not just flat stake? Flat staking (e.g. always 2% of bankroll) is an approximation of Kelly at medium edge. The difference: flat doesn't react to edge size. A 1% edge bet and an 8% edge bet get the same stake. Mathematically suboptimal, but psychologically easy. Half-Kelly only beats flat staking if your edge estimate is actually calibrated.

Common mistakes: (1) Using full Kelly because 'the formula says it's optimal' — ignores estimation error. (2) Overstating edge in the Kelly formula (wishful thinking → effective 1.5× or 2× Kelly). (3) Applying Kelly to multiple parallel bets without accounting for correlation (two bets on the same match outcome are not two independent Kelly bets).

Tool practice: Our Kelly calculator displays Full, Half, and Quarter-Kelly side by side automatically. Enter your odds and probability estimate and you immediately see the three stakes. Watch the 'expected log growth' column — you'll notice Half-Kelly sits almost at full Kelly's growth, but with much lower risk.

A realistic setup: Bankroll £2,000, 5% edge on odds 1.90. Full Kelly ≈ 5.5% = £110, Half-Kelly ≈ £55, Quarter-Kelly ≈ £27.50. If this bet loses, full Kelly leaves you at £1,890, Half-Kelly at £1,945. Over 200 bets with several losing streaks the difference compounds massively — Half-Kelly survives the streaks, full Kelly often doesn't.

Related articles: 'Value betting explained', 'Calculating your edge — the formula', Kelly criterion calculator (Full/Half/Quarter Kelly side by side), Dutching calculator. External source: J. L. Kelly Jr., 'A New Interpretation of Information Rate' (1956) — the original paper defining the log-growth optimum.

Bottom line: Full Kelly is mathematically optimal only if your probability estimate is perfect. Since it never is, Half-Kelly or Quarter-Kelly is almost always the better choice — less growth on paper, more growth in reality (because you don't go broke). Experienced bettors run Half-Kelly as standard; beginners should run Quarter-Kelly or less.