How does Bookie Bullies use the Kelly Criterion in Sports Betting in its model?

Our sabermetric model factors in the Kelly Criterion in Sports Betting alongside 34 other signals (FIP, BABIP regression, Pythagenpat priors, park factors, weather, umpire boosts, bullpen quality) to produce every daily pick. See /methodology.html for the full factor list.

Are picks using the Kelly Criterion in Sports Betting free on Bookie Bullies?

Yes. Every pick is free, no signup, no paywall, no email required. Full model output (pick, confidence, expected value, edge vs market, Kelly stake) is visible to anyone.

● EXPLAINER · BY MARCDUCK

What is the Kelly Criterion in Sports Betting?

Kelly criterion is the mathematically optimal stake size given your win probability and odds. Full Kelly maximizes long-run growth but amplifies variance. Here's the formula, when to use fractional Kelly, and why we use 1/8 Kelly.

The Formula

f* = (p × b - q) / b

Where:

f* = fraction of bankroll to stake
p = your probability of winning
q = 1 - p = your probability of losing
b = decimal odds minus 1 (the profit ratio)

If you have 60% prob at -110 (decimal 1.909, b = 0.909):

f* = (0.60 × 0.909 - 0.40) / 0.909 = 0.16 — bet 16% of bankroll.

Why Full Kelly is Dangerous

Kelly assumes your probability is exact. It rarely is. Real probabilities have noise. Full Kelly amplifies estimation error catastrophically — if your "60%" is actually 55%, you've wildly overstaked and your bankroll cratters fast.

Real sharps use fractional Kelly: a fraction (1/2, 1/4, 1/8) of the optimal stake. Lower fractions trade some growth rate for much lower variance.

Why We Use 1/8 Kelly

Bookie Bullies uses 1/8 Kelly with variance adjustment. The reasoning:

Our model's probability has real noise (we display the uncertainty band).
Calibration is unproven on a per-bucket basis (we publish Brier score; it's around 0.231 — better than a 0.25 coin-flip baseline but not perfectly calibrated).
1/8 Kelly limits drawdown risk while still sizing meaningfully on high-conviction picks.
If our true edge is +5% but we mis-estimate as +3%, full Kelly loses money; 1/8 Kelly stays profitable.

Variance-Adjusted Kelly

We use the Bayesian uncertainty band (mean ± sigma from the four-component blend) to compute Kelly with the LOWER BOUND probability. Wider uncertainty → smaller stake. Tight consensus → bigger stake. This is "uncertainty-aware Kelly" — standard practice in any sharp shop.

Practical Stakes

At 1/8 Kelly, most picks land at 0-3 units (where 1u = 1% of bankroll). The bankroll calculator on every picks page converts units to dollar stakes for your input bankroll. A 2.5u Lock at $1000 bankroll = $25 — meaningful but not bankroll-risking.

When to Increase Kelly Fraction

If you've verified your edge over hundreds of picks (positive ROI, positive CLV), 1/4 Kelly becomes reasonable. 1/2 Kelly only after thousands of bets. Full Kelly is for closed-form games (poker hand math, blackjack card counting) where the probability is mathematically exact — never for sports betting where probabilities are estimated.