What Is the Kelly Criterion?
The Kelly Criterion solves one of the most important problems in betting: not whether to bet, but how much. Even with a genuine edge, overbetting a bankroll leads to ruin. Underbetting leaves profit on the table. Kelly provides the mathematically optimal solution — the stake size that maximizes the long-term geometric growth of a bankroll given a known edge and odds.
The formula is:
Kelly % = (bp – q) / b
Where:
- b = net odds received (decimal odds minus 1)
- p = estimated probability of winning
- q = estimated probability of losing (1 – p)
Example: A horse you estimate at 30% win probability is offered at 4/1 (5.0 decimal, so b = 4.0).
Kelly % = (4 × 0.30 – 0.70) / 4 = (1.20 – 0.70) / 4 = 0.50 / 4 = 12.5%
At full Kelly, you would wager 12.5% of your bankroll on this horse. At $1,000 bankroll, that is a $125 bet.
Why Most Bettors Should Use Fractional Kelly
Full Kelly is mathematically optimal but carries significant volatility. A losing streak — even with a genuine edge — can draw a bankroll down by 30-40% before recovering. Most professional bettors use Half Kelly (50% of the calculated stake) or Quarter Kelly (25%), which reduces variance dramatically while sacrificing only a small portion of expected long-term growth.
At Half Kelly, the example above becomes a $62.50 bet — a meaningful reduction in variance with minimal long-term cost.
The Edge Requirement
Kelly only works when you have an accurate estimate of your true edge. If you overestimate your probability of winning, Kelly will recommend a stake that is larger than optimal — this is called overbetting, and it leads to ruin faster than flat-betting with no system. The Kelly formula is only as good as the probability estimate that feeds it.
This is why accurate fair-odds calculation — not just picking winners — is the foundational skill in professional horse racing betting.
Dynamic Kelly in StrideOdds
StrideOdds uses a Dynamic Kelly model that adjusts stake recommendations based on two factors: the calculated edge (the gap between fair-odds probability and tote odds) and the Confidence Score (a model-level quality signal for that race's data). High-confidence overlays receive a higher Kelly band recommendation; low-confidence signals receive a fractional Kelly recommendation regardless of edge size.
This prevents the system from recommending large stakes in races where data quality is poor, even if the calculated edge appears significant.
The 2% Rule vs Kelly
For bettors who prefer simplicity, the 2% Rule — never wager more than 2% of total bankroll on a single bet — is a conservative flat-staking approach. It avoids overbetting but does not scale stake size to edge. Kelly is strictly superior in terms of long-term growth when probability estimates are accurate. For bettors still developing their handicapping accuracy, the 2% Rule provides a useful safety floor.