ICM is often flawed

[curtains]

The ICM has problems with short stacks, especially very short stacks. A lot of people who ask about ICM calculations, give situations where very short stacks are involved, and the ICM results in these cases are often a bit far from reality.

[rachelwxm]

[ QUOTE ]

Yes the problem is that a lot of the time the numbers are not reasonable. If the blinds are huge and you have 200 chips and your opponent has 10.
You dont have a 20x greater share of the prize pool than your opponent, however this is what the ICM tells you.

[/ QUOTE ]

unfortunately in this case ICM is correct, sometimes intuition could be totally wrong.

Here is the math if you care (assuming hugh blinds):
1. If your chips ratio is 1:1 you have 50% prize pool, plain and simple.
2. If you have 3:1 chip lead, and your opponent take 50% chance to double up to situation 1 and 50% chance OTM. Therefore, he has 25% equity and you have 75%.
Etc...

Some people like play poker using intuitions, some people like to use math. Nothing wrong with either approach imo.

Any other examples? [img]/images/graemlins/smile.gif[/img]

[curtains]

Rachel the ICM isnt correct in this case. The situation Im referring to is when there are 4 players remaining and two players are so short they cannot afford the blinds.
Just because one is 20x shorter doesnt mean they should have 20x less equity. Sometimes the player with ten chips will actually be the favorite to finish in the money, depending on who plays the blind first. Either way, the 10 chipper is never a 20-1 underdog to make 3rd place, with another short stack around.

[gumpzilla]

How could you have 75% equity in the prize pool? Assuming you're describing heads up on the end with a standard 5-3-2 payout structure, the winner only gets 5/8 = 62.5% of the prize pool. Winning every time can't do better than that.

[rachelwxm]

[ QUOTE ]

Rachel the ICM isnt correct in this case. The situation Im referring to is when there are 4 players remaining and two players are so short they cannot afford the blinds.
Just because one is 20x shorter doesnt mean they should have 20x less equity. Sometimes the player with ten chips will actually be the favorite to finish in the money, depending on who plays the blind first. Either way, the 10 chipper is never a 20-1 underdog to make 3rd place, with another short stack around.

[/ QUOTE ]

If you mean bubble with stack 4000 4000 50 20 with blinds 200/400, I agree I would be cautious with ICM. But Heads Up ICM and Chip EV converge and lead to same conclusion.

[rachelwxm]

[ QUOTE ]
How could you have 75% equity in the prize pool? Assuming you're describing heads up on the end with a standard 5-3-2 payout structure, the winner only gets 5/8 = 62.5% of the prize pool. Winning every time can't do better than that.

[/ QUOTE ]

For all simplification, I assume winner takes all. As it really does not matter since you can subtract second prize in Heads Up situation.

[curtains]

Just take a look at the thread "Interesting bubble hand" that was posted earlier in this thread. That's a great example of a situation where the numbers are just bogus.

[se2schul]

ICM will often be wrong just as the weather forecast is wrong. ICM is just a model which is a simplistic representation of a complex problem. If poker situations were simple, we wouldn't need to use a simplifying model. If you were to make ICM perfectly accurate, taking into account position, tendencies of opponents and such, you'd be on the verge of writing a bot.

ICM seems like a good tool for its purpose. In many cases, it will tell you what the correct move is. Somtimes it won't work, and that's just the nature of a model. It's important to recognize when a model is applicable to real life and where it deviates, and make the appropriate play.

I'd be interested in seeing examples of ICM not working. Do you have several specific examples that we could look at and maybe find patterns?

Steve

[FishBurger]

It seems like the main problem is that ICM is a linear function based on the number of chips you and your opponents have while the blinds are not linear. The blinds are some type of step function that can take a big chunk out of your stack with each orbit.

ICM breaks down near the "edges." In my limited experience with it, it seems like it works best in the fat part of the bubble where everyone has greater than 4x BB. Once you ener a situation where two players have less than 2xBB, position and who is in the best shape to survive their blinds becomes more important.

P.S. I'm talking out of my ass.

[eastbay]

[ QUOTE ]
He means the ICM is the ICM and gives correct ICM results. Just like calculating chip EV gives correct chip EV results.

Determining the actual value of a play in an SNG is not possible. There are too many variables. It matters how you and the other players are going to play in the following hands, which is not the case in a cash game.

All we can do is refer to models. Chip EV is one such model, but it is pretty poor. The ICM is a great improvement. If you have reasons, in a specific hand, why the ICM results do not apply then all you have to do is state them. But I think it's a stretch to say the ICM "often" gives wrong results. Also I think it's a waste of time trying to develop a model that takes into account position, etc. Those variables are too complex to be modelled.

[/ QUOTE ]

Finally some sanity in this thread.

eastbay