Sklansky's (Poker) Tournament System
October 12, 2005 8:26 PM Subscribe
So I was reading "Tournament Poker for Advanced Players" by Sklansky, and found his system, and his modified system, pretty interesting. It's pretty easy to analyze the basic system, but the modified system is more complicated, and I have some questions/concerns about it.
First, if you're not aware of the system I'm talking about, I'll give a tiny bit of detail. His system basically involves decided whether to fold or go all in. He has a formula that is based on the number of limpers, the number of players still to act, and the ratio of the blinds to your chip stack. Basically (limpers+1) * actors * stack / blinds. If there is a raise before you, you go all in with AA KK QQ or AKs. If there is no raise, you take your output from the formula, and depending on the number, there are different sets of hands to go all in on. The lower the number is, the more relaxed the constraints. This follows from the fact that if there are fewer players to act, you are less likely to get called, and if your ratio to the blinds is low, you must play looser to catch up.
Anyway. I made some scripts that calculate this number for me in real time, and I've been experimenting with playing the system live, online. I tried in free tournaments first. It does "ok", generally coming in the top 25%. This is fairly easy to do in a freeroll though, I bet I could do as well, or better, just folding. Freerolls suck. I've also tried it in a few low dollar tourneys, $1-$5. I generally go out around the mid point. Not a great showing.
With the original system, which had a fixed set of hands to go all in wth, you would go all in about 13% of the time. So if you weren't called you'd slowly make money, picking up the blinds a little more often than you paid them. If you were called, generally you would either lose or double up. In the modified system, the call rate seems to be more like 5%. It may be that I don't have enough samples yet but that seems to be it.
I don't know how to simulate the situation. I guess I could model the players, picking values for how loose of callers they are (since the # of limpers affects the number so much) and run a bunch of trials. The other thing I don't quite know how to analyze is expected return. It seems to me that if you assume that everyone folds to you, you can model it fairly easily -- but everyone won't always fold, and in fact, in low dollar tournaments you will get called with crap hands quite a bit. So an unknown factor, to me, is how often I can expect to get called.
I am wondering also if the system needs to be modified for online play, since the levels advance every 5 minutes (turbo) or 12 (normal) compared to maybe once an hour in a large live tournament. So, online, I might expect to see 10 hands between blind increases and live I might expect to see 30-40. In the freerolls I was often knocked out (in the top 25% mind you) after only about 50 hands. In these situations it seems like you might need to loosen the restrictions, but again, not sure how to model it.
First, if you're not aware of the system I'm talking about, I'll give a tiny bit of detail. His system basically involves decided whether to fold or go all in. He has a formula that is based on the number of limpers, the number of players still to act, and the ratio of the blinds to your chip stack. Basically (limpers+1) * actors * stack / blinds. If there is a raise before you, you go all in with AA KK QQ or AKs. If there is no raise, you take your output from the formula, and depending on the number, there are different sets of hands to go all in on. The lower the number is, the more relaxed the constraints. This follows from the fact that if there are fewer players to act, you are less likely to get called, and if your ratio to the blinds is low, you must play looser to catch up.
Anyway. I made some scripts that calculate this number for me in real time, and I've been experimenting with playing the system live, online. I tried in free tournaments first. It does "ok", generally coming in the top 25%. This is fairly easy to do in a freeroll though, I bet I could do as well, or better, just folding. Freerolls suck. I've also tried it in a few low dollar tourneys, $1-$5. I generally go out around the mid point. Not a great showing.
With the original system, which had a fixed set of hands to go all in wth, you would go all in about 13% of the time. So if you weren't called you'd slowly make money, picking up the blinds a little more often than you paid them. If you were called, generally you would either lose or double up. In the modified system, the call rate seems to be more like 5%. It may be that I don't have enough samples yet but that seems to be it.
I don't know how to simulate the situation. I guess I could model the players, picking values for how loose of callers they are (since the # of limpers affects the number so much) and run a bunch of trials. The other thing I don't quite know how to analyze is expected return. It seems to me that if you assume that everyone folds to you, you can model it fairly easily -- but everyone won't always fold, and in fact, in low dollar tournaments you will get called with crap hands quite a bit. So an unknown factor, to me, is how often I can expect to get called.
I am wondering also if the system needs to be modified for online play, since the levels advance every 5 minutes (turbo) or 12 (normal) compared to maybe once an hour in a large live tournament. So, online, I might expect to see 10 hands between blind increases and live I might expect to see 30-40. In the freerolls I was often knocked out (in the top 25% mind you) after only about 50 hands. In these situations it seems like you might need to loosen the restrictions, but again, not sure how to model it.
Looking at the reviews from this book on Amazon.com, it came to the conclusion I came to:
Therefore, in the final stages of the tournament, it is often more important to facilitate eliminating other players than to maximize the number of chips that can be won during the course of a hand. Thus it is dead wrong to suggest that some Omaha High-Low hands are so strong that a player should always raise with the hand. That suggestion, which may be appropriate for a live game, could be disastrous in the last stages of a tournament.
Like in Survivor, the one who wins is the one who doesn't go after the one in the lead but the one right ahead of him. It looks (as I suspected) that this was about maximizing your chips and not advancing in tournaments.
posted by geoff. at 8:43 PM on October 12, 2005
Therefore, in the final stages of the tournament, it is often more important to facilitate eliminating other players than to maximize the number of chips that can be won during the course of a hand. Thus it is dead wrong to suggest that some Omaha High-Low hands are so strong that a player should always raise with the hand. That suggestion, which may be appropriate for a live game, could be disastrous in the last stages of a tournament.
Like in Survivor, the one who wins is the one who doesn't go after the one in the lead but the one right ahead of him. It looks (as I suspected) that this was about maximizing your chips and not advancing in tournaments.
posted by geoff. at 8:43 PM on October 12, 2005
Response by poster: Right, right, don't get me wrong. I am not planning to adopt this as a style of play. And even sklansky does not claim it's optimal, merely that it's a method for a completely brain dead player to advance to the final stages of a major tournament against expert players, without knowing how to play poker. He actually offers it as an arugment against the ascendancy of no limit tournaments, which essentially allow a novice player to avoid being outplayed, by removing all of the subtlety from the game.
Anyway, I found the system interesting and I was curious how it would perform. In real play I got some surprising results (i.e. you'd only play about 5% of the hands, which is not enough to keep from getting blinded out -- you need to play 10% or more on average).
I built a poker academy bot for this and tried that out last weekend. After doing this I realized there was *already* a PA bot that follows this system (it does both systems actually). It performs a little better than adequarely vs. the other PA players, which are fairly decent, but on average only comes in 4th. This is better than the expected result (5th) but not by much. But then again, with the version of PA I have you can only simulate 9 person sitngos.
posted by RustyBrooks at 8:57 PM on October 12, 2005
Anyway, I found the system interesting and I was curious how it would perform. In real play I got some surprising results (i.e. you'd only play about 5% of the hands, which is not enough to keep from getting blinded out -- you need to play 10% or more on average).
I built a poker academy bot for this and tried that out last weekend. After doing this I realized there was *already* a PA bot that follows this system (it does both systems actually). It performs a little better than adequarely vs. the other PA players, which are fairly decent, but on average only comes in 4th. This is better than the expected result (5th) but not by much. But then again, with the version of PA I have you can only simulate 9 person sitngos.
posted by RustyBrooks at 8:57 PM on October 12, 2005
Response by poster: I'm thinking the best way to "simulate" this is in real play. I just have to decide whether it's worth $100 for me to see how it performs in real life, just to satisfy my curiosity. If he's right, and you could often coast into the money with this sytem, then I don't have much to worry about. If he's wrong, well, $100 is not so much really to test a theory. I paid $30 for his book and it's only about the10th poker book I've bought.
posted by RustyBrooks at 8:59 PM on October 12, 2005
posted by RustyBrooks at 8:59 PM on October 12, 2005
Er, are you saying the formula tells you to go all in 5% of the time and fold 95% of the time?
If people see you playing that tight and then all of a sudden see you going all in, the only callers you'll get are those with premium hands (or the super crazies).
Out of curiosity how many play money tournaments did you play to reach your top 25% number? How many tournaments at the low $1-5 range did you play to reach your midpoint number?
posted by curbstop at 9:35 PM on October 12, 2005
If people see you playing that tight and then all of a sudden see you going all in, the only callers you'll get are those with premium hands (or the super crazies).
Out of curiosity how many play money tournaments did you play to reach your top 25% number? How many tournaments at the low $1-5 range did you play to reach your midpoint number?
posted by curbstop at 9:35 PM on October 12, 2005
I've played over 1000 one table tournaments online (most either $22 or $33 buyin) with a positive ROI. I've also been a little intrigued with Sklansky's modified system that he explains in some publication (this one contains a key number, and a number of hand ranges to push/fold based on the key number).
From putting in various situations, to plugging in the numbers as I play real time, the overall feeling I get is that the modified system simply isn't very good. It underestimates your opponents' calling ranges and overestimates your folding equity. When the stack/blinds ratio gets low it pushes far too aggressively. I would look into evaluating push/fold situations away from the table by determining your pushing EV by estimating your opponents' calling ranges and using logic/ICM calculations. Just use straight up ICM for determining your folding equity. Use the results to get to get your gut in tune with good pushing situations.
In summary: The system is too simplistic and doesn't estimate your opponents' calling ranges correctly. You're not going to make money with it. : (
posted by reishus at 9:38 PM on October 12, 2005
From putting in various situations, to plugging in the numbers as I play real time, the overall feeling I get is that the modified system simply isn't very good. It underestimates your opponents' calling ranges and overestimates your folding equity. When the stack/blinds ratio gets low it pushes far too aggressively. I would look into evaluating push/fold situations away from the table by determining your pushing EV by estimating your opponents' calling ranges and using logic/ICM calculations. Just use straight up ICM for determining your folding equity. Use the results to get to get your gut in tune with good pushing situations.
In summary: The system is too simplistic and doesn't estimate your opponents' calling ranges correctly. You're not going to make money with it. : (
posted by reishus at 9:38 PM on October 12, 2005
Response by poster: curbstop: you'd think so. But nonetheless, say I'm playing 5%, so 5 hands out of 100. I'll get called on 2 or so of them. And the POINT is to not get called. You're supposed to pick up the blinds and any limpers who have come in. That's why I think his system is a little off, because you should be going all in a bit more to pick up the blinds more often. I've played 5 freerolls with his system and 3 money tournaments. Probably only 300-400 hands is all. Oh, and I played a number of sitngos where it actually did quite well, but, I think, probably not a whole lot better than just folding. So maybe 500-600 hands. Definitely not enough to know anything for sure, but during those hands I saw some things that didn't see right to me.
reishus: that's my take too. I keep feeling like "I'm going to *fold* this here? and then alternately "I'm supposed to go all in with K6 suited here!?". But then I do go all in with K6 suited and get called with Q8o.
The funny thing is that I think the UNMODIFIED system is probably better, even though he modified it to make it better. The original system was so simple because he wanted to teach it to someone who'd never played poker and didn't want them to have to do math at the table.
I suspect I could make his system better by tweaking the ranges.
posted by RustyBrooks at 5:40 AM on October 13, 2005
reishus: that's my take too. I keep feeling like "I'm going to *fold* this here? and then alternately "I'm supposed to go all in with K6 suited here!?". But then I do go all in with K6 suited and get called with Q8o.
The funny thing is that I think the UNMODIFIED system is probably better, even though he modified it to make it better. The original system was so simple because he wanted to teach it to someone who'd never played poker and didn't want them to have to do math at the table.
I suspect I could make his system better by tweaking the ranges.
posted by RustyBrooks at 5:40 AM on October 13, 2005
Response by poster: OK, I did some admittedly quite crude simulation last night. Here's what I did (over and over):
I modelled players by their gestalt activity rather than what they'd call or raise. I think this is a fair first approximation. So, I'd set up 10 players and assign them all a calling looseness (like, player 1 calls the blind 15% of the time. Player 2 calls 35% of the time, etc). I based these off observed data that I have. I also assigned each a raising probability. Once one player raises in sklansky's system, you only call with AA KK QQ or AKs so I don't have to consider which players will call the raise or re-raise. And actually, I'm not concerned with how often I'll WIN (right now), just how often I'd be able to call with his system.
So, each iteration through I assign a dealer randomly. I give everyone a random chip stack between 5 and 75 times thebig blind, weighted a bit towards the middle. I deal myself cards (no one else's cards matter). Starting with the UTG player they randomly fold, call, or raise according to the parameters I set up before.
Obviously, this is pretty crude but I think it's close enough for what I'm after. After a buttload of simulations (100K or so) the allin percent for the system seems to be about 8%. I suspect if I fool with the looseness and raise-i-ness percents I might get higher numbers (i.e. I have the boundaries set for a table with several pretty bad players).
Sklanksy says in his book that if you have a guy on your right who raises too much you might have to change your constraints a bit. I bet with some fooling around I could come up with an alrgorithm that examines the looseness of the players around you and shifts the boundaries accordingly. 8% is a bit too low for this to work, I think.
posted by RustyBrooks at 8:18 AM on October 13, 2005
I modelled players by their gestalt activity rather than what they'd call or raise. I think this is a fair first approximation. So, I'd set up 10 players and assign them all a calling looseness (like, player 1 calls the blind 15% of the time. Player 2 calls 35% of the time, etc). I based these off observed data that I have. I also assigned each a raising probability. Once one player raises in sklansky's system, you only call with AA KK QQ or AKs so I don't have to consider which players will call the raise or re-raise. And actually, I'm not concerned with how often I'll WIN (right now), just how often I'd be able to call with his system.
So, each iteration through I assign a dealer randomly. I give everyone a random chip stack between 5 and 75 times thebig blind, weighted a bit towards the middle. I deal myself cards (no one else's cards matter). Starting with the UTG player they randomly fold, call, or raise according to the parameters I set up before.
Obviously, this is pretty crude but I think it's close enough for what I'm after. After a buttload of simulations (100K or so) the allin percent for the system seems to be about 8%. I suspect if I fool with the looseness and raise-i-ness percents I might get higher numbers (i.e. I have the boundaries set for a table with several pretty bad players).
Sklanksy says in his book that if you have a guy on your right who raises too much you might have to change your constraints a bit. I bet with some fooling around I could come up with an alrgorithm that examines the looseness of the players around you and shifts the boundaries accordingly. 8% is a bit too low for this to work, I think.
posted by RustyBrooks at 8:18 AM on October 13, 2005
Thing is, coming in the top 25% isn't relly an "ok" outcome, since you're not getting paid in the lower end of that 25%. The money is typically in the top 10%, right?
Maybe mix it up. Play Sklansky for the big hands, and Hellmuth for the rest of the top ten?
posted by solid-one-love at 8:18 AM on October 13, 2005
Maybe mix it up. Play Sklansky for the big hands, and Hellmuth for the rest of the top ten?
posted by solid-one-love at 8:18 AM on October 13, 2005
Response by poster: I am, of course, assuming that I implemented his system correctly. I could be off on something... I discovered something last night that incorrectly counted the # of limpers.
posted by RustyBrooks at 8:20 AM on October 13, 2005
posted by RustyBrooks at 8:20 AM on October 13, 2005
Response by poster: No, top 25% sucks. In freerolls sometimes the money positions are LUDICROUSLY low, like top 50 people out of 2500. My point was that I didn't think the system was right, and I'm trying to figure out why, and what could be done to make it more effective.
posted by RustyBrooks at 8:24 AM on October 13, 2005
posted by RustyBrooks at 8:24 AM on October 13, 2005
There is no "System" for poker.
But if you want to try it out, let's get together the second MetaFilter Poker Tourney.
MeFi Poker Tourney Wiki
posted by Duncan at 9:43 AM on October 13, 2005
But if you want to try it out, let's get together the second MetaFilter Poker Tourney.
MeFi Poker Tourney Wiki
posted by Duncan at 9:43 AM on October 13, 2005
Why are you focusing on the overall all in % and tweaking it? This number is not relevant at all. You need to examine each situation where you push and determine if a push is relevant.
One limper to you in the cutoff. You need to decide the EV of pushing.
You estimate the button will call a push, due to current stack sizes and his personality, with his top 15% of hands. Small blind - 10%, Big Blind - 20%, Limper - 10%. However, since the limper will call with his top 10% of hands, and he limped, he might have a hand in the 10% about 30% of the time (depends on his limping and raising ranges).
So...
push ev = .15( ev if button calls) + .85(ev if button folds)
ev if button calls = (my equity against his range)*($$ev from the resulting chip distribution if i win the all in) + (1-my equity against his range)*($$ev from the resulting chip distribution if i lose the all in)
ev if the button folds = .10(ev if small blind calls) + .90(ev if the big blind folds)...
this math problem is the only way to reasonably solve a problem like that. sklansky's just sort of groping in the dark. there are situations when skalansky's system are too aggressive, and there are situations where it's not aggressive enough. but the former are much more common.
posted by reishus at 9:46 AM on October 13, 2005
One limper to you in the cutoff. You need to decide the EV of pushing.
You estimate the button will call a push, due to current stack sizes and his personality, with his top 15% of hands. Small blind - 10%, Big Blind - 20%, Limper - 10%. However, since the limper will call with his top 10% of hands, and he limped, he might have a hand in the 10% about 30% of the time (depends on his limping and raising ranges).
So...
push ev = .15( ev if button calls) + .85(ev if button folds)
ev if button calls = (my equity against his range)*($$ev from the resulting chip distribution if i win the all in) + (1-my equity against his range)*($$ev from the resulting chip distribution if i lose the all in)
ev if the button folds = .10(ev if small blind calls) + .90(ev if the big blind folds)...
this math problem is the only way to reasonably solve a problem like that. sklansky's just sort of groping in the dark. there are situations when skalansky's system are too aggressive, and there are situations where it's not aggressive enough. but the former are much more common.
posted by reishus at 9:46 AM on October 13, 2005
Response by poster: The reason I'm concerned about the overall percentage of the time you would push all in is, if you don't do it frequently enough, you lose money consistently to the blinds. Any system that results in you going all in less than about 10% of the time will leak money through the blinds. Do this long enough and you lose the threat ability inherent in going all in. I certainly have not (generally) found sklansky to be too aggressive, rather often too passive. But this may be due to the opponents I'm playing it against (weak ones). Sklansky is approximating the kind of calculation you're doing above, dumbed down to something that a person could do in his head. It's a gross approximation, for sure. Also, to be clear, I am only interested in exploring a system where you go all in or fold, I'm trying to remove the relative skills of the players from the equation entirely.
I am not looking for The System To Beat Everyone At Poker. I saw a particular system in a book. It looked interesting. I tried it. It seems weak. I want to find out where the weakness comes from and see if it could be improved.
It is sklansky's belief, and mine too to a certain extent, that there is a "bug" in no limit tournaments, that it's possible to reduce it to a set of gambles, with little or no skill involved (well, no more skill than remembering a simple calculation and a table of hands). Now, it's still a set of gambles. No system will allow you to win all the time, BUT, I think there are systems that give you a better chance of placing in the money than randomness would dictate. And I'm interested in theoretically exploring such systems.
posted by RustyBrooks at 10:30 AM on October 13, 2005
I am not looking for The System To Beat Everyone At Poker. I saw a particular system in a book. It looked interesting. I tried it. It seems weak. I want to find out where the weakness comes from and see if it could be improved.
It is sklansky's belief, and mine too to a certain extent, that there is a "bug" in no limit tournaments, that it's possible to reduce it to a set of gambles, with little or no skill involved (well, no more skill than remembering a simple calculation and a table of hands). Now, it's still a set of gambles. No system will allow you to win all the time, BUT, I think there are systems that give you a better chance of placing in the money than randomness would dictate. And I'm interested in theoretically exploring such systems.
posted by RustyBrooks at 10:30 AM on October 13, 2005
« Older Can a collection agency re-open a nine year-old... | What do I need as proof of stock ownership in a... Newer »
This thread is closed to new comments.
The problem I have with most poker systems, such as the one you've described, abnormal behavior within a group of players (the guy that doesn't fold but should) throws off a system fairly quickly.
I've had great experience modeling blackjack systems. It's a lot easier and constrained than a poker game. I've had little success with actually using a model like you describe and sticking it to it no matter what. The differing strategies between players makes sticking to modeling impossible. This is not the advice I'm sure you're looking for and someone who doesn't do this just for the weekends might be able to help, but I use such things as rules of thumb rather than strict formulas.
posted by geoff. at 8:40 PM on October 12, 2005