Advanced Maths and Simulation

Kelly Criterion and Why It Fails Without an Edge

What the Kelly criterion formula says, a worked example with a positive edge, and why it recommends betting nothing when the expectation is negative.

On this page
  1. What the formula is for
  2. A worked example with an edge
  3. What happens when the edge is negative
  4. Why a clever stake size cannot fix it
  5. Where Kelly might apply and its limits
  6. A sensible takeaway

What the formula is for

The Kelly criterion is a rule from probability theory for choosing how much of a bankroll to stake on a bet that has a positive expected return. Its goal is to maximise the long-run growth rate of the bankroll. The key word is positive: the formula presupposes that the bet is in your favour.

For a bet that pays b units for each 1 unit staked, and that wins with probability p and loses with probability q = 1 - p, the Kelly fraction of the bankroll to stake is:

f = (b x p - q) / b

The numerator, b x p - q, is the expected profit per unit staked. So the formula is simply the edge divided by the odds.

A worked example with an edge

Imagine an even-money bet (b = 1) that wins 55 percent of the time. This is a made-up illustration, not a blackjack situation.

  • p = 0.55, q = 0.45
  • Expected profit per unit: 1 x 0.55 - 0.45 = 0.10
  • Kelly fraction: 0.10 / 1 = 0.10, or 10 percent of the bankroll

If the bankroll were 1,000 units, the formula would suggest staking 100 units. After a win the bankroll rises and so does the next stake; after a loss both fall. Staking a proportion rather than a fixed amount means the bankroll can shrink but never reaches exactly zero, at least in theory.

Betting more than the Kelly fraction does not raise the growth rate. Past a point it lowers it, and heavy over-betting leads to a shrinking bankroll even though each bet has a positive edge. Many people who use the idea bet only a fraction of the figure, because the formula is very sensitive to the edge being estimated correctly.

What happens when the edge is negative

Now change the bet so that it is unfavourable: an even-money bet that wins 49 percent of the time.

  • p = 0.49, q = 0.51
  • Expected profit per unit: 0.49 - 0.51 = -0.02
  • Kelly fraction: -0.02 / 1 = -0.02

A negative fraction means that the best stake is zero. In effect, the criterion is saying not to make the bet. Nothing else can be read from it. It does not offer a smaller stake, a safer stake, or a way to turn the game around. The same applies to blackjack played with a typical rule set: with a house edge against the player, the formula returns zero or less.

Why a clever stake size cannot fix it

It is tempting to think that the right betting plan can overcome a negative expectation. The arithmetic says otherwise. The expected value of a sequence of bets is the sum of the expected values of each bet. If every bet has a negative expected value, any rule for choosing the stake, whether proportional, fixed, increasing or decreasing, still produces a negative total expectation. Changing the stake changes how large the swings are, and how fast the expected loss accumulates, but never the sign.

This is also why betting systems that rely on raising stakes after losses cannot work in the long run. They may produce many small wins, but they trade them for the rare large loss, and the average stays negative.

Where Kelly might apply and its limits

In blackjack, the only setting where the criterion becomes relevant is for a player who has a genuine positive expectation on some hands, as the topic of card counting describes in theory. Even there, caution is warranted:

  • The edge has to be estimated, and estimation errors push the formula toward over-betting.
  • Table limits, a bet spread that attracts attention, and rule differences restrict what stakes are possible.
  • Blackjack outcomes are not simple win or lose bets, because doubles, splits and naturals have different payouts, so the plain formula is only an approximation.
  • Casinos may refuse service, and rules differ by place.

For ordinary play at a table with a house edge, Kelly gives no help. It sits alongside bankroll management as a concept in finance and betting theory, but it does not apply to a negative game.

A sensible takeaway

The formula is a good example of a tool that is only valid under its assumptions. When someone presents a staking method as a path to profit from blackjack, ask the first question the formula asks: where is the positive edge? If the answer is that there is none, no staking plan can create one.

Blackjack has a house edge, outcomes are uncertain, and gambling can be harmful. Only wager money you can afford to lose, and contact a local support service if gambling is becoming a problem.

The complete guide to advanced maths and simulation

Simulation Basics: How Odds Tables Are Built
Read the guide

Keep exploring

Guides from other topics

All topics →