MODS: I wanted to get the STT input specifically, but if this is in the wrong forum, just delete...

I know that there are always posts about variance and how bad people have run. Hopefully this post will be different since I've got the volume and the math behind me to make it a little more interesting. I'd like some input as I'm totally lost on what to do.

I wrote a program that goes through every hand history and grabs all the preflop all-ins that I'm involved in. It doesn't factor in 3-way all-ins since that's too complicated and obscure for me. So if I have KK and villain has AKs, I'm a 87.86% favorite. If I win that hand, then my program says I was 12.14% lucky. Basically, I should only win 87.86% of the time, but I actually won 100% of the time, so therefore 12.14% lucky. If I had lost, I would've been 87.86% unlucky.

So anyways, after 81,000 all-in hands, I'm about -21,000% unlucky. Just to put that into layman's terms... If I win my 72o vs. villain's AKo 312 times in a row, then my luck would be back to even.

Now if you add in ICM and weight each hand to see the actual dollar amount of luck vs. unluck, it gets much worse: for my $55 tourneys, I'm +$813 lucky. $105s = -$19,148 unlucky. $210s = -$5,400 unlucky. $315s = -$6,123 unlucky. $525s = -$28,792 unlucky. $1ks = -$13,550 unlucky. $2ks = -$8,749 unlucky.

Juk wrote a program exactly like mine to measure these stats (there's a great post about it in the software forum) and I downloaded it and ran it against my program, and it produced identical results. That, plus I've tested my program for many months, makes me believe that there is not a flaw in the programs.

My problem is that I play online poker for a living. I know that variance is part of poker and I've had my upswings along with downswings, just like everyone else. My play on PokerStars is abnormally unlucky though. It is measurably unlucky. Should I put on the blinders and just keep playing and relegate myself to losing $10,000/month due to luck? This data set is about 75,000 SNG's (yes, that's not a typo). At what point do you have to admit that something is not right and you should move on? I'm a SNEx2, so I have alot invested with Stars. Should I throw that all away and start playing somewhere else?

I think there are 4 possible reasons why my numbers are so poor:
1) Mine and Juk's programs are wrong (unlikely)
2) Someone has hacked Stars and is fixing the cards/games (unlikely)
3) Stars is deliberately screwing the sharks and redistributing to the fish (unlikely)
4) I'm on the worst 75,000 game stretch of all time (unlikely)

So what would you do if you were in my shoes? Why do you think my numbers are so bad? Note that this has NOTHING to do with my playing style nor skill level (or lack thereof :-). This data is strictly preflop all-ins and simply a mathematical computation.

Any thoughts would be greatly appreciated!!!
-BK

i recall a bunch of supernova cash game players (p0kerb0y1 comes to mind) running their massive db's through a similar software and getting similarly disturbing results. it seemed to some that the regulars in the game were getting the business end of variance for huge sample sizes and the lucky recipients were the casual/recreational players. there was quite a debate about the idea of "redistributing wealth" to the fishies.

http://forumserver.twoplustwo.com/28...thread-264915/

A thread titled: Official Poker Site Data Analysis and Discussion Thread

in the internet gambling forum

MicroBob has a theory why your program is expected to give negative luck results for certain players.

To have played that many STT's and at those stakes I'm guessing you must still be a winning player?

I could be wrong, but if not, just look at your luck results as a compliment to your game in that despite the fact that you've had so much bad luck you're game is still good enough to grind out a profit?

Also I may be wrong here, because I don't really get Maths, but if you say you've -21000% in luck over 81000 hands doesn't that mean that you're losing 0.25% every time you go all in. So basically in 50/50's you're running at 49.875/50.125? Which isn't too bad?

If this is part of your program, please revise.

Well actually your model is wrong.

You are simplifying this little too much.

Your main profit comes form folding, not from pushing.

Here is a hand that should explain it:

You have 100t and are dealt 72o and your HU opponent has 100t and AA.
He raises(push, doesn't matter) you fold.
He should be winning 90% of your chips on average but is not.
So in your model you are gaining lot of "luck", but you are not counting this, you only count when you call.

So I believe that you folded 72o to AKs more than 312 times, so your "luck" should be ok.

GL and HF

nice first post! missing the point much? getting ****ed for 75k sngs is not something to be excited about.

it is completely valid to discuss the results vs. expectation of ONLY all-in hands and to anticipate that results will approach expectation in them as n approaches infinity. in other words, an analysis of only all-in scenarios is valid.

Bumpking,

Sorry, I forget your stars name. Are you dna?

His calculator only evaluates preflop all-in EV. In other words, although his calculator doesn't measure total luck, it does measure all-in luck, of which he seems to be running very much below expectation.

His calculator wouldn't count your example, then, of an opponent pushing AA and him "gaining a lot of luck" by folding.

If both his and Juk's program are wrong, then it is either proof that Stars is rigged or that the programs aren't correct in some way or possibly some other explanation that I haven't thought of.

Perhaps though he isn't running as bad as he thinks: if he is only running below expectation of ~4 buy-ins at the 2ks, and he somehow uses this in an overall evaluation which includes, say, only $50s, then the results will look very skewed.

75000 games total is a lot. But if you run really bad at the nosebleed buy-ins (where even the best players have minimal edges) then it can definitely skew your results in a very "Stars is rigged" way.

Ok, done a few more calculations to add to my previous post:

From your figures you say you are -$80949 in luck? Yet you've played 75,000 STT's... so you lose... just over $1 per game.

Whilst that would be pretty horrific if you were playing $11 games it's not so bad when you consider the stakes you're playing. This would be better with a more accurate estimation of the number of STT's you've played at each stake but if you take the average of the stakes you've played then it comes out at about $601 per buyin.

75,000 games at $601 a piece is around $45,000,000 in buyins that you have paid, of which luck has cost you $80,949 or around 0.18%...

Am I missing the point here or does the size of the sample in fact prove that this is not so bad?

I think your program has at least one error; it doesn't count the hands that were folded pre flop. It's just a tiny detail and I'm not saying that it totally explains your results, but when dealing with this large samples it might start to matter.

I think Juk has made a mod to his program that filters out only the HU hands. HU you have always all the cards in play so you could perhaps try to run that.

I hope I made sense, I'm in a bit hurry now.

He won't tell you.

I think the problem is that you take 20 seconds to make every decision and therefore starts purposely doomswitches you. So obviously the solution is to not timebank every time you are folding a random hand utg at 10/20.

This is wrong. You're more like 67 % IIRC, anyway much less than 87%.

Please explain exactly your method of calculation to allow math guys here to check (I guess I'm one).

Please also provide your sample sizes for each buyin. What does matter mathematically is not the absolute value of what you lost in bad luck, but the probability of that kind of losses, and we need your exact sample size for this. I mean we need to find out what is the probability of running that bad (or worse) on such a sample size.

81,000 hands is fairly significant.... but cash game players go on bad stretches that last easily 100K hands.

that being said, 81K is still pretty relevant and is the only number here that's even worth discussing because it's the only number that we're measuring on. 75K games means nothing here, those hands have been isolated.

also, buyin stakes have nothing to do with this being that it's all measured in tournament dollars. and considering the amount of T$ that have been gambled over 81K hands with starting stacks of T1500, being -T100K (or whatever it was) i don't think is very significant or unlikely after 81K hands.

that's just my interpretation on my assumptions

1 - I think this can be a good reason why you got these results:


2 - Sample sizes matter. 75K games is a good sample, but you can't look into absolute numbers forgetting that sample. You probably profited at most 5% less than you were expected to profit. Maybe 2-3% considering rakeback.


3 - Half of your unlucky amount comes from $500-$1000s. What's your sample in them? Losing $13k more than expected in $1000s seems very reasonable for a big sample and even more reasonable for a small sample, in this case due to variance.

Since you have already run juk's program on your sample, post that result and the graph. Remember that you have to run each buy-in with a different % of rake separately. If you need help compiling the results I can help you if you send me the raw output of each buy-in. I remember having something like a $15k luck downswing over about 500 SNGs and a $10k uptick over 1000, playing 55s and 109s.

Right off the bat, it doesn't look like your results are all that interesting, but I really need to see it as a % of the prize pool to tell.

Bumpking is not DNA as DNA posts under another 2p2 handle, and would never complain this much

i don't consider this a whine post. he has a huge sample and wants to know if it is normal variance. i can see the reason for the post

bk - i am not really much of a math expert so hard for me to say. just looking at it, 81K hands, -21K luck, it is just about 0.25% precent bad luck per allin. does not seem that crazy to me but i bet as this thread goes along someone can help with the math. of course it is a lot $$$ and that really sucks.

not sure how long the downswing has been going on but you should consider playing 2 or 3 less tables and maybe avoiding the real high stakes games until you get things turned around. i think the games are gonna be really tough these last couple of months as all the mid/high stakes players scramble for SNE.

Originally Posted by bumpking
So if I have KK and villain has AKs, I'm a 87.86% favorite.

KK is only a 66.3% favorite..

Thanks for the great responses and ideas guys. Here's some more info:

Sorry, I meant AA vs AKs. The fact that my program is corroborated by Juk's means i'm 99.9% sure it is accurate.

Yes, that seems to be a fair statement. But the problem lies in that the 50.125's are happening on all-ins with few chips and little $EV and the 49.875's are on the large $EV hands.

I can only measure known variables. An argument to your example would be all the times that I had AKs and villain folded 72o should be the same.

LOL. That is a good 5th possible reason!

Cash is different than SNG's. If a cash player goes on a 100k hand downswing, they weren't all-in every single hand, right?

55's: 14920 SNGs with +$813 luck = 15 BI's lucky
105's: 17388 SNGs with -$19,148 luck = 182 BI's UNlucky
210's: 13894 SNGs with -$5,400 luck = 26 BI's UNlucky
315's: 4787 SNGs with -$6,123 luck = 20 BI's UNlucky
525's: 2786 SNGs with -$28,792 luck = 55 BI's UNlucky
$1k's: 209 SNGs with -$13,550 luck = 13 BI's UNlucky
$2k's: 64 SNGs with -$8,749 luck = 4 BI's UNlucky

I'll start digging through previous year's data as this is about the last year's worth. The results from the previous years were very similar.

Can someone classify the odds of these results? What are the odds of getting 285 buyins unlucky over 54,000 SNG's? If I were playing a HU SNG and we went AI preflop and I had 72o vs. their AA (88%), then I would have to win (285/.88) = 325 SNG's to get back to even. My results have to be way outside of a standard deviation, right?

I looked a little at p0kerb0y1 and MicroBob's post that was mentioned and found some good ideas. There's a theory that there's a rig against regulars to redistribute wealth to fishes to keep the pool. I doubt this is the case, but I'm gonna run my program through and see how the top 20 regulars run to disprove the theory.

I am a winning player in overall dollar terms. I like Star's software, VIP program, etc. Can I continue to win and make a living with this sort of "unluckiness"? Yes. I just wanted to see what people thought the reasons could be for this ugly data.

I really appreciate everyone chipping in with their ideas!
Thanks,
-BK

Make a graph of $EV (actual,expected,luck) vs. SNG. A picture is worth a thousand words.

75k SNGs is a very large sample. I think I was averaging about 6 all-in hands per SNG using juk's, which accounts for all all-in hands, and maybe about 1.8 per that actually involved me. 75k SNGs is plenty. I want you to post some graphs so we can all see the "noise."

OK, so -285 buy-ins, playing 9-max is -31.7 prize pools, which is -3167%, well within what we've seen as "normal." Juk posted a lot of numbers in the big software thread and IIRC he saw a number of +-3000% swings.

The "odds" don't really matter. Just proving that a 1/100 or whatever event occurred in a sample of who knows how many winning regulars doesn't prove anything. Not to sound like a dick, but someone has to be the unluckiest bastard out there, right? We've all just assumed it was suzzer99, but maybe not.

Just from a model validity standpoint, use juk's program, not yours, unless you want to spend a lot of time benchmarking your program against his.

Thanks a lot slim. I just sprayed coffee all over my keyboard you bassturd. Besides, we all know the real answer is Pinapple.

When I pointed to the "theory" in the pokerboy/microbob thread I was not referring to the redistribution of wealth but the fact that your program is slightly flawed in that it doesn't put hand ranges on the folded hands.

For example Ax is better against a pair than your program calculates from the simple fact that no one mucked AA (and other A hands are not in the range of folders).

I believe for most winning players they are on the better side of Ax vs Pair. If that is the case for you then your program would think you had more ev than you actually had.

I would be interested in your "luck" results strictly when the tournament is heads up.