How ICM Is Calculated: The Math Behind Chip Equity

ICM turns chip stacks into prize-pool dollars by working out how often each player finishes in each paid position, then paying them for every one of those outcomes. The model rests on a single assumption — your chance of finishing first equals your share of the chips in play — applied over and over until every finishing order is accounted for. Everything after that is bookkeeping.

Below is the full calculation run by hand on a four-handed final table, plus the reason the arithmetic keeps telling good players to fold hands they would happily stack off with in a cash game.

What ICM actually computes

Your stack is not money. It is a claim on a prize pool that gets divided by finishing order, and the only thing that converts one into the other is a probability distribution over where you finish.

ICM builds that distribution. It produces a set of numbers like "50% chance of first, 33% of second, 15% of third, 3% of fourth", multiplies each one by the payout attached to that position, and adds the results. The total is your equity, in dollars, right now.

That total is what people mean by what ICM means in poker: the value of your seat if everyone stopped playing and the tournament resolved fairly from here. Every ICM decision you will ever make is a comparison between two of these numbers.

The model is deliberately thin. It knows two things — stacks and payouts — and nothing else. For the concepts behind it rather than the arithmetic, start with the Independent Chip Model explained.

The one assumption everything rests on

Here it is, and it is the entire model:

Your probability of finishing first equals your stack divided by all chips in play.

Hold 40% of the chips, win 40% of the time. Nothing more sophisticated is going on.

Second place is that same rule applied twice. To finish second, somebody else has to win, and then you have to be the best of whatever is left. So you loop over every opponent, take their probability of winning, and multiply it by your share of the chips remaining once their stack is removed from the denominator.

Third place loops over every ordered pair of players finishing ahead of you. Fourth loops over every ordered triple. The work grows brutally fast — n players produce n factorial finishing orders — which is why ICM stops being a paper exercise past about six-handed.

A worked four-player example

Four players remain. There are 10,000 chips in play and a $1,000 prize pool paying $500 / $300 / $150 / $50. The stacks are A on 5,000, B on 3,000, C on 1,500 and D on 500.

Start with A. First place is immediate: 5,000 of 10,000 chips is 50%.

Second place needs three terms, one for each player who might win instead:

• B wins (30%), then A takes 5,000 of the 7,000 chips left → 0.30 × 71.4% = 21.4%
• C wins (15%), then A takes 5,000 of 8,500 → 0.15 × 58.8% = 8.8%
• D wins (5%), then A takes 5,000 of 9,500 → 0.05 × 52.6% = 2.6%

Add those and A finishes second 32.9% of the time. Third place repeats the exercise across all six ordered pairs of opponents finishing ahead, which comes to 14.6%. Fourth is whatever is left over: 2.6%.

Now price the distribution:

500 × 0.500 + 300 × 0.329 + 150 × 0.146 + 50 × 0.026 = $371.77

Run the identical loop for every seat and the final table prices out like this.

Player	Stack	Chip share	ICM equity	Prize-pool share
A	5,000	50.0%	$371.77	37.2%
B	3,000	30.0%	$303.32	30.3%
C	1,500	15.0%	$215.02	21.5%
D	500	5.0%	$109.89	11.0%

The four equities sum to exactly $1,000, which is the one sanity check worth running on any ICM output. Prize money cannot be created or destroyed by a chip transfer.

Why the fourth chip is worth less than the first

Compare the last two columns. A holds half the chips and 37.2% of the money. D holds a twentieth of the chips and 11.0% of the money — better than double what the chip count implies.

This is the whole reason ICM exists. Only one player collects first prize, but everyone who survives collects something, so the value of accumulating chips flattens out as your stack grows. The first chips you win buy survival, which is worth a lot. The last ones buy a bigger share of a prize you were already favourite for, which is worth much less.

The asymmetry shows up sharply when you move a fixed number of chips around. Take 1,000 chips off C and give them to A, and A gains $35.52. Move 1,000 the other way, and A loses $37.91. Same chips, different price, and the loss is bigger every time.

That divergence is the single biggest adjustment between cash games and tournaments, and it is worth being fluent in expected value in poker before you try to reason about it. In a cash game a 50/50 for stacks is exactly neutral. On a final table it is a losing play for whoever has more to lose.

What the numbers change at the table

Turn that $13 into a threshold. If A calls B's 3,000-chip shove, winning leaves A on 8,000 and busts B in fourth; losing leaves A on 2,000 and B on 6,000. Feeding both outcomes back through ICM, A needs roughly 57% equity to break even on the call rather than the 50% a chip-EV calculation would demand.

That seven-point gap is the risk premium, and it is the practical output of the whole model. Three consequences follow directly.

Big stacks tighten their calling ranges and widen their shoving ranges. They have the most dollars exposed when they lose a flip, so they avoid flips — but they can attack relentlessly, because everyone else faces an even worse price to call.

Short stacks get the best price in the room. D risks $109.89 and can only lose it once, so D's calling threshold barely moves from chip-EV. The pressure lands on the medium stacks who are trying to ladder past a short stack.

Pay jumps set the size of the premium. A flat structure barely bends the numbers; a steep one bends them enormously. This is why the money bubble and the final-table pay jumps generate almost all the ICM folds you will ever make, and why a satellite — where every paid finish wins the identical prize — produces the most extreme folds in poker.

Where the model breaks down

ICM is a useful lie, and knowing which parts are the lie keeps you from over-trusting it.

It has no concept of blinds. A 10 big blind stack about to post the big blind and a 10 big blind stack on the button price out identically, even though one of them is materially worse off. It has no concept of position, no concept of the next hand, and no concept of skill — everybody is assumed to play the same, forever.

It also ignores the future game entirely. Being short with three limpers to your left is a different tournament from being short at a table of nits, and the model cannot see either situation. Full future-game simulations correct some of this at enormous computational cost.

Treat ICM the way you treat pot odds: a price, not a decision. The price is accurate; what you do about it still depends on reads, on the structure, and on how the table is actually playing.

Running the numbers in practice

Nobody computes factorials at the table. What you build instead is calibration, and that comes from checking your instincts against the model away from the felt.

1. Price the spot before you look. Guess your risk premium, then run it.
2. Vary one thing at a time. Move a stack, or flatten the payouts, and watch which direction the threshold travels.
3. Learn the shapes, not the digits. Big stack tight, short stack loose, medium stacks squeezed. Those hold everywhere.

An ICM calculator does the whole loop in a keystroke, including the two-outcome comparison that produces your break-even threshold. Feed it the final tables you actually played and you will find the folds you missed within a session or two.

The maths is not the hard part. Believing the maths, while a stack of chips sits in front of you and first place is still available, is.