The Kelly criterion gives a mathematically optimal answer to "how much should I bet?" — assuming you know the true probability of winning. In sports betting, you never know the true probability. You estimate it. And the gap between your estimate and reality is where most bettors blow up their bankrolls, even when their general approach is right.
This article walks through the framework sharp bettors use to size bets when they don't know their true edge, why fractional Kelly is the practical standard, and the specific situations that call for sizing down further than Kelly suggests.
The setup: Kelly assumes information you don't have
A reminder from Bankroll Management: the Kelly criterion formula for sports betting is:
Stake fraction = (decimal odds × estimated win probability − probability of losing) ÷ (decimal odds − 1)
If you estimate 60% probability on a +100 bet (decimal 2.00):
(2.00 × 0.60 − 0.40) ÷ (2.00 − 1) = 0.80 ÷ 1.00 = 0.20, or 20% of bankroll.
Full Kelly says bet 20% of your bankroll. That's an enormous bet. And it assumes your 60% estimate is exactly right.
In practice, your 60% estimate is uncertain. It might be 55%, or 65%. You don't know. The true probability that the bet wins is somewhere in a range around your estimate, and the width of that range is your estimation uncertainty.
Kelly with uncertain inputs produces catastrophic results when the estimate is too high. A bettor who estimates 60% but is actually facing a 53% bet will lose money long-term, and Kelly will tell them to bet 20% of their bankroll while doing it. The math compounds losses fast.
The standard solution: fractional Kelly
Almost every serious bettor uses fractional Kelly — betting some fraction of what full Kelly suggests. The standard fractions:
Quarter Kelly (0.25x). Bet 25% of what full Kelly recommends. In the example above, instead of 20%, you'd bet 5% of bankroll. Quarter Kelly captures about 55% of full Kelly's expected growth rate while reducing variance to roughly 25% of full Kelly's variance. This is the most commonly recommended fraction for serious bettors with documented but imperfect edge estimates.
Half Kelly (0.5x). Bet 50% of what full Kelly recommends. Captures about 75% of full Kelly's growth rate with 50% of the variance. Appropriate for bettors with very well-calibrated probability estimates over a long history.
Tenth Kelly (0.1x) or smaller. Bet 10% or less of full Kelly's recommendation. Appropriate for new strategies, new sports, or models with limited validation history. Most growth is preserved in the upside but downside variance is dramatically reduced.
The honest position: most bettors should be using 0.25x Kelly or smaller for most of their bets, including bettors who feel confident they're using their estimates well.
Why fractional Kelly is mathematically right (not just safer)
The case for fractional Kelly isn't just "be conservative because uncertainty is scary." It's mathematically optimal under realistic conditions.
Two related arguments:
The variance penalty grows quadratically with bet size. Doubling your bet doesn't double your downside — it quadruples the variance. Halving your bet from full Kelly to 0.5x Kelly cuts variance by 75% while only cutting expected growth by 25%. The math favors the smaller bet for any rational risk preference.
Estimation error matters more than the central estimate. Kelly is sensitive to overestimation of edge. If your true edge is 5% but you estimate 8%, full Kelly tells you to bet roughly 60% more than the true optimum, which over time produces lower returns than betting the correct amount. Fractional Kelly automatically corrects for the typical bettor's tendency to overestimate their edge.
In aggregate, fractional Kelly is the bet sizing strategy that maximizes long-term wealth growth under realistic conditions of imperfect information. Full Kelly is theoretically optimal only under the unrealistic assumption that you know your edge exactly.
When to size down further than Kelly suggests
A few specific situations where even fractional Kelly is too aggressive:
New strategy, no track record. When you're betting a new sport, new prop type, or new model that hasn't been validated over a meaningful sample, your edge estimate has enormous uncertainty. Size as if your edge is much smaller than your point estimate. Tenth Kelly or smaller is appropriate until you have at least 100-200 graded bets in the new market.
Recent model retraining or methodology change. When you've updated the model or process producing your probability estimates, the old calibration data doesn't fully apply. Treat the post-change period as a new strategy until you've accumulated enough bets to verify the new approach is still calibrated.
Highly variable markets. Some prop types and sports are inherently noisier than others. NBA assists are noisier than NBA points; MLB win totals are noisier than NFL spread bets. Higher-variance markets warrant smaller fractional Kelly multipliers because the realized outcomes deviate more from expected.
Correlated bets in your portfolio. If you're betting multiple props in the same game, those bets are correlated — if one player has a bad game because the team got blown out, multiple of your bets are likely to suffer together. The Kelly formula assumes bet independence. When your bets aren't independent, your effective bet size on the underlying outcome (the game itself) is larger than any individual bet suggests. Size each correlated bet smaller to compensate.
Tournament or finals settings. Single-game stakes with no opportunity for "next game" recovery. NFL playoffs, NBA Finals, Super Bowl props. These deserve more conservative sizing than regular-season bets because the variance of a single game outcome can't be averaged out across the rest of a season.
When (rarely) to size up
A few situations where larger position sizes are mathematically justified:
Promotional EV. When a sportsbook is offering a clear positive-EV promotion (a boost above fair price, a bonus bet on a +EV opportunity, etc.), the EV is essentially riskless after the promo is applied. Sizing for these is constrained more by the promo's stake limits than by Kelly considerations.
Established edge with long track record. A bettor with 2,000+ graded bets showing consistent positive CLV and reasonable calibration of their probability estimates can step up from 0.25x Kelly toward 0.5x Kelly without unreasonable risk. The data has narrowed the uncertainty around their edge enough to justify the higher multiplier.
Edge cases with extreme value. Occasionally a market mispricing is large enough that even with significant estimation uncertainty, the bet is clearly +EV. A bet with implied probability 30% that you reasonably estimate at 50%+ has so much margin for error that larger sizing is justified despite the uncertainty.
These cases are exceptions. The default is fractional Kelly conservative; sizing up requires affirmative reason.
A practical decision framework
Walking through a real bet decision under uncertainty:
Step 1: Estimate your edge as a range, not a point. Don't say "I think this is a 60% bet." Say "I think this is between 55% and 65%, most likely around 60%." The range tells you about your uncertainty.
Step 2: Use the lower end of the range for sizing. If your range is 55-65%, calculate Kelly at 55%, not 60%. This builds in margin for being wrong.
Step 3: Apply fractional Kelly to the result. Use 0.25x as the default. Lower (0.1x or smaller) if the bet is in a new market or recent strategy change.
Step 4: Cap the bet at a hard percentage of bankroll. Many sharp bettors set a maximum bet size — say 5% of bankroll — regardless of what Kelly suggests. This prevents catastrophic losses from single bets even when the math suggests a large bet would be optimal.
Step 5: Adjust for correlation. If this bet is correlated with other open positions, size it smaller proportional to the correlation.
The total effect of this process: a bettor with a real edge betting carefully will rarely exceed 1-3% of bankroll on any single bet, even when full Kelly suggests double-digit percentages. That conservatism is what lets the edge actually compound over time without catastrophic drawdowns blowing up the bankroll first.
What goes wrong without this discipline
A bettor who finds a genuine 55% edge but bets full Kelly (or worse, more than full Kelly) typically experiences this trajectory:
First few months: growth. The edge is real, the bets are sized aggressively, the bankroll grows fast. The bettor feels brilliant.
First normal drawdown: catastrophic loss. A normal 10-bet losing streak — which a 55% bettor will experience every few months — at full Kelly sizes (20%+ of bankroll per bet) compounds into a 50-70% bankroll loss. The bettor has done nothing wrong; this is what full Kelly variance looks like.
Recovery is mathematically slow. A 50% drawdown requires a 100% gain to recover. Even at full Kelly with a real edge, that recovery takes many months. Many bettors quit during the drawdown, concluding their strategy was bad, when in reality their bet sizing was the problem.
The bettors who survive this trajectory typically don't survive twice. Once you've experienced a 50%+ bankroll drawdown, the appropriate response is to size down — but most bettors who haven't yet had this experience are the ones overbetting in the first place.
The summary
Position sizing in sports betting is the single biggest factor separating sustainable bettors from blow-up cases. The Kelly criterion is the theoretical optimum, but it requires information you don't have — exact knowledge of your edge.
Fractional Kelly (0.25x as a standard) captures most of Kelly's growth rate while dramatically reducing variance and protecting against the inevitable overestimation of edge that most bettors do. Smaller fractional multipliers (0.1x or less) are appropriate for new strategies, untested markets, or environments where your edge estimate has high uncertainty.
The honest principle: size your bets as if your edge is smaller than you think. The math, the variance, and the human tendency to overconfidence all push in the same direction — and almost no bettor regrets being more conservative with bet sizes in hindsight.
ParlayX provides analytics tools and educational content, not betting advice. Sports betting involves financial risk and is intended for adults only. If you or someone you know has a gambling problem, call 1-800-GAMBLER for confidential help, 24 hours a day.