StatsAI.bet

Strategy 07 of 11 · free e-book

Kelly criterion for betting

Kelly Criterion: stake by your edge

The one staking formula mathematicians respect. It works only with a real edge and an honest estimate of the chance. Here is what happens with and without one.

4.5%Kelly stake at odds 2.10 and a 50% chance
82.5%full Kelly: fell 50% from a peak
716median after a 3-point overestimate

8 pages · 2.5 MB · no sign-up

Kelly Criterion, page 1 of the PDF
Type
proportional staking
Built for
bets with a measured edge
Goal
fastest long-run growth
Main danger
overestimating your edge

How the Kelly criterion works

Kelly answers one question: if a bet has a positive expected value, what share of the bankroll should you stake to grow it as fast as possible over the long run?

  1. bodds − 1 (net payout)
  2. pyour chance of winning
  3. q1 − p, chance of losing
  4. f(b·p − q) ÷ b = share to stake

Example: odds 2.10 and a real chance of 50%. Then b = 1.10 and f = (1.10 × 0.50 − 0.50) ÷ 1.10 = 4.5% of the current bankroll. If the result of the formula is zero or negative, the bet has no edge and Kelly says: do not bet.

Unlike the progressions in this series, Kelly does not react to wins and losses. It reacts to the bankroll: the stake is always a share of what you have now, so it shrinks when you lose and grows when you win.

Where it comes from

John L. Kelly Jr., a scientist at Bell Labs, published the formula in 1956 in a paper on information theory. Mathematician Edward Thorp later used it in blackjack and in investing, which made it famous.

10,000 bettors with a real edge

Now a case with a genuine edge: odds 2.10, a real chance of 50% (expected value +5%), 500 bets, starting bankroll 1,000. Five plans, the same results for everyone.

PlanMedianWorst 10%Best 10%In profitFell 50%+ from a peak
Full Kelly (4.5%)1,7644646,69868.9%82.5%
Half Kelly (2.3%)1,5317852,98477.9%15.0%
Flat stake 101,2509561,54485.3%0.3%
Overestimate: think 53% (10.3%)7163514,61845.0%100.0%
Double Kelly (9.1%)1,0006914,42151.7%100.0%

Full Kelly grows the bankroll fastest in the middle, but 82.5% of bettors saw it fall by half from a peak at some point, even with a real edge. Half Kelly gave up a little growth for far calmer results.

The last two rows are the warning. Overestimating the chance by just 3 points drops the median to 716. At double Kelly one win and one loss cancel out exactly, so the median ends where it started: 1,000.

Verdict

Kelly is the only plan in this series that is built on an edge rather than on hope. That is its strength and its limit.

Riskhighwith full Kelly; half Kelly lowers the swings a lot

What it does well

  • Mathematically the fastest long-run growth with a real edge
  • The stake shrinks automatically after losses
  • Says "do not bet" when there is no edge

Where it breaks

  • Needs an honest, calibrated chance for every bet
  • Full Kelly swings hard: 82.5% fell 50% from a peak
  • Overestimating the edge can turn growth into decline
  • A real edge over the margin-free market is rare

Three steps

1 · Chancestart from the fair odds and adjust only with a measured reason
2 · Formulaf = (b·p − q) ÷ b; zero or less means no bet
3 · Fractionstake half or a quarter of f, never more than f

If you want to use it

  1. Use half or a quarter of the Kelly stake.
  2. Start from the fair odds and move away only with a measured reason.
  3. Record every estimate and check later whether your 60% really happened about 60% of the time.
  4. If the formula says zero, do not bet.

Look inside

Eight pages: how it works, an example, the maths, a simulation of 10,000 bettors, the verdict and rules for anyone who still wants to try it.

Kelly Criterion, page 1 of the PDFKelly Criterion, page 3 of the PDFKelly Criterion, page 6 of the PDF

Kelly Criterion · Kelly Criterion: stake by your edge

Frequently asked questions

What is the Kelly criterion?

Kelly answers one question: if a bet has a positive expected value, what share of the bankroll should you stake to grow it as fast as possible over the long run?

Does the Kelly criterion work?

Only with a real edge. At odds 2.10 and a real 50% chance, full Kelly stakes 4.5% of the bankroll; in our simulation 82.5% of bettors still fell 50% from a peak, and overestimating the chance by 3 points dropped the median to 716.

What is the biggest risk of the Kelly criterion?

With full Kelly; half Kelly lowers the swings a lot. Needs an honest, calibrated chance for every bet.

How should I use the Kelly criterion?

Use half or a quarter of the Kelly stake. Start from the fair odds and move away only with a measured reason.

Is the PDF free?

Yes. The Kelly Criterion e-book has 8 pages and can be downloaded from this page without registration, in English or Serbian.

Before you try it

Every strategy stands or falls on two numbers: the real chance and the odds. StatsAI shows both for today's matches.

Learn more

More strategies from the series