Why the Kelly Criterion Doesn’t Work for Pools Stakes

Somewhere along the way, every numerate pools player bumps into the Kelly Criterion. It gets passed around forums and bar-stool conversations as the “proper” mathematical way to size a bet, borrowed wholesale from poker and horse racing. It sounds authoritative, it has a clean formula, and it is almost entirely useless for a weekly pools coupon. Here is why, with the arithmetic laid bare rather than taken on faith.

What the Kelly Criterion actually assumes

The Kelly formula is written as f* = (bp − q) / b, where f* is the fraction of your bankroll to stake, b is the net odds offered on a win, p is your estimated probability of winning, and q is simply 1 − p. It was designed for situations where you know, or can reasonably estimate, both a fixed payout multiple and a stable win probability, and where you can repeat the bet many times under similar conditions. Fixed-odds betting on a single football match fits that description reasonably well. A pari-mutuel pools coupon does not fit it at all.

Running the numbers on a real coupon

Take an illustrative example. Suppose a player estimates their chance of landing a meaningful Treble Chance dividend at roughly 1 in 2,000,000 for a given coupon — a deliberately rough but not unreasonable order of magnitude for a demanding points target. Suppose further that, based on a particularly generous week, the effective payout multiple on the stake might be in the region of 500,000 to 1, again purely illustrative. Plugging those numbers into the Kelly formula:

f* = (bp − q) / b = (500,000 × 0.0000005 − 0.9999995) / 500,000 ≈ −0.0000015

That result is negative. Kelly’s own formula, applied honestly to numbers of this shape, tells you to stake nothing at all. Even if the illustrative payout multiple were pushed up to 2,000,000 to 1, the fraction it recommends is effectively zero, a rounding error rather than a usable stake size. This is not a quirk of bad luck; it is what the formula is supposed to do when the odds of winning are vanishingly small relative to the payout uncertainty. Kelly is built to protect you from exactly these bets, not to help you size them.

Why the pools breaks Kelly’s core assumptions

There is no fixed payout multiple

Pools dividends are pari-mutuel: the prize pool is built from everyone’s stakes that week, after deductions, and then split among the winners at whatever level they reached. The “b” in Kelly’s formula needs to be a known, fixed number. On a pools coupon it moves every single week depending on how many other players enter, how many matches finish as draws, and how the points fall. You cannot plug a stable number into a formula that requires one.

Your win probability is an estimate of an estimate

Kelly also assumes you have a reasonably trustworthy value for p. Form, fixture difficulty and goal trends can nudge your confidence in a line up or down, but nobody genuinely knows the true probability of a specific eight-match combination landing a dividend-paying points total to three decimal places. Feed a shaky estimate into a formula this sensitive and the output is shaky too, however precise it looks once it is through a calculator.

Trials are not independent and identical

Kelly’s long-run growth logic depends on being able to repeat a bet many times under statistically similar conditions. Each pools week is a new, unique set of fixtures, each with its own weather, injuries and competitive stakes. There is no clean “repeat this bet a thousand times” scenario to lean on.

What happens if you try to “fix” Kelly for pools anyway

Some players, on hearing that the raw formula collapses to zero, try to patch it rather than abandon it. A common workaround is to replace the true win probability with something more generous, on the theory that a skilled line looks more like a 1-in-50,000 shot than a 1-in-2,000,000 one. The arithmetic shifts, but the underlying problem does not go away.

Illustratively, raise p to 1/50,000 and keep b at 500,000: f* = (500,000 × 0.00002 − 0.99998) / 500,000 ≈ 0.0000180. That is a tiny positive fraction of a bankroll, technically “staking something,” but so small that in practice it would round to a few pence on even a generous weekly budget. The formula has not become useful; it has simply become less dramatically broken. Nudging the inputs to make Kelly produce a friendlier-looking number is really just choosing the answer you want and reverse-engineering an input to justify it, which defeats the entire point of using a formula in the first place.

A steadier way to size your coupon spend

None of this means stop thinking about budgeting. It means stop borrowing a tool built for a different game. A simpler, more honest approach for pools tends to work better in practice:

• Decide a fixed weekly amount you can genuinely afford to lose, in the same way you would budget for any regular hobby.
• Split that amount across a sensible number of lines or a perm, rather than one large single line, to spread exposure across different fixture combinations.
• Keep the amount flat week to week rather than increasing it after a loss or a near miss, which is really just Kelly-style chasing dressed up differently.
• Review your total spend monthly, not after every single result, so that one unlucky or lucky week does not distort your judgement.

Illustrative comparison of two approaches

The honest takeaway

There is a certain appeal in dressing up a hobby with serious-looking mathematics, and the Kelly Criterion is a genuinely clever piece of maths in the right context. It simply is not the right context here. Pools coupons are a low-probability, variable-payout, pari-mutuel product, and the sensible response to that is disciplined, flat, affordable staking rather than a formula imported from a different kind of bet. Treat the Kelly number as an interesting curiosity to mention down the pub, not a staking plan to follow.

A note on why this matters beyond one formula

The broader lesson here applies well past Kelly specifically. Any staking system borrowed from a different kind of gambling, whether it is a progression system from roulette, a unit-sizing method from sports spread betting, or a bankroll rule from poker tournaments, tends to carry hidden assumptions about fixed odds, known probabilities and repeatable trials. Pools rarely satisfies any of those three conditions cleanly. The healthiest habit is to ask, before adopting any staking rule, whether the game it was designed for actually resembles the one you are playing.

Whatever system you use, remember that pools should be played for enjoyment on money you can afford to lose, never treated as a source of income, and that no formula, including this one, can turn long odds into a reliable return. If you are 18 or over and choose to play, set a limit before you start and stick to it regardless of how the week unfolds.