No, I'm a bit nitty too. Probably because I'm just a casual player and not looking at my hourly all the time.

I don't know where this ladder originates from but it could be a handy basic guideline for a beginner.

cool pic/saved.
does it use kelly?

divide by 2 obv (if it says $40k for quarter Kelly, then it's $20k for half Kelly and $10k for full Kelly)

well playing w a 10% ROI at $20s is far better than playing w a 15% ROI at $10s iyam. tho playing w scared money can lead to problems such as quitting when up, quitting when down etc. which ultimately limit the play time, but that's a mental crutch you can let go of (hopefully), while the actual act of winning money is so to say an exact science

this is imo a good idea for a beginner who doesn't have a sample to estimate his winrate, but later on obv ppl need to re-evaluate their winrate and find out which stakes they can play according to Kelly

I never miss an occasion to make a fool of myself. That was a dumb question

Hi JSpazz,

I'm trying to do this for cash also. But I'm a little mixed up on the ptBB vs. bb distinction.

So for a particular level of NLHE 6max, my WR is 2.32pt, and my stn dev is 1672.81pt.

So my R is:
R = (1672.81 / 2.32) * 2 = 1442.08 ~ 14.4 bi-ins? or is 1442.08 in ptbb in other words is it 2884.16 ~ 28.8 bi-ins? 14.4 bi-ins seems like an absurdly low number.

Thanks so much for doing this, I think it's really cool and could reduce massive amounts of stress in my life.

Quick calculation shows I'm a huge nit. I should be playing 84's while I'm playing 5's with a 4 digits roll. Aggressive BRM fail.


It can't, it would need your ROI in input.

your roi at the 80s would be alot lower than the 5s so you probably shouldn't be playing the 80s just yet :P

but i definitely shouldn't be playing the 20s with 200buyins :X

no problem dude I like sharing my thoughts w other ppl

1672.81ptbb/100 would be a really huge standard deviation, so I assume you're talking about variance (SD^2), in which case yes you calculated the bankroll correctly. If you divide it in half (14.4 buyins), you'll get the correct Kelly strategy but playing by half-Kelly is more often and much less stressful

What is the ROR for the Kelly strategy? 10 buy ins seems ridiculously dangerous. Am I missing something? Are we expected to drop down if we lose a certain number of buy ins.

I have a 5% ROI at the $100s. My worst downswing was around 20 buy ins. I've been playing the $50s then $30s then $20s (turbos) and I've had a downswing of 45 buy ins. I assume my ROI in these is higher than that at the $100s.

every time your bankroll changes size, you have to adjust the stakes you're playing. you move up and down aggressively.
there is 0 RoR because the stakes you play are always a function of your bankroll size.

So at what point do I move down when my bankroll shortens? Say my B is 10, at what point do I move down?

should i factor in rakeback in my ROI % when doing these calculations?

your B is the percentage of your bankroll you should bet on your next husng. with a B of 10, you move down once the buyin for the sng level you're playing is greater than 10% of your bankroll.
after every husng you finish, your bankroll is a new number, you simply figure out what stakes your B suggests that you play as a function of your bankroll. (this assumes that your roi is constant across stakes which is, of course, untrue, but you can still vaguely predict your winrate at adjacent stakes)

I believe so. Rakeback is part of your ROI.

I'm a bit confused on this. Taking 5% rake into account, shouldn't this be up 0.95 buyins or down 1.05 buyins? Unless I'm missing something it seems to me like the rake would make a pretty big difference in the formula. But then again I'm not great with maths.

I'm afraid I'm gonna look like a total idiot here.

An easier way to use the Kelly criterion for husngs is choosing the stake that maximizes the expected log wealth. This was shown to me by Aaron Brown.

Say you win a match with probability w (you can use the ITM % stat from HEM).
The expected log wealth is U = w*log(bankroll + stake - rake) + (1-w)*log(bankroll - stake)

You then choose the stake that gives a higher value for the expected log wealth.

Ex: Let's say you want to decide between $60 and $100 dollar stakes for Hyper-Turbos. $60s rake is $1.26 and $100 rake is $1.88. HEM says you win 52% of the games. Your current bankroll is $6000.

U(60s) = 0.52*log(6000 + 60 - 1.26) + (1-0.52)*log(6000 - 60) ~ 8.69976

U(100s) = 0.52*log(6000 + 100 - 1.88) + (1-0.52)*log(6000 - 100) ~ 8.69988

You have better expectation playing the 100s since U(100s) > U(60s).

With this formula you can easily choose the optimal stake to play given a bankroll and a winning probability.
Note that this is the Kelly criterion which is pretty aggressive and you might want to play Half Kelly instead so divide the stake by 2 to reduce the risk of ruin.

If you divide the stakes by 2 (1/2 kelly), doens't it affect the calculation as the rake on that stakes is different (in %) ?

Sorry if I didn't explain it properly. You do not divide the stakes by 2. You determine the appropriate stake for your bankroll using the formulas but you actually play at half that stake. For example if you use the formula for 30s, 60s and 100s and the stake that maximizes your growth is 100 you actually play the 60s (if you want to play approximately half kelly). If you want to play exactly half kelly which would be a stake of $50 in the example, then you would have to play mixed stakes since Pokerstars does not offer $50 stakes but that wouldn't bring you any advantage to playing $60 and moving up/down according to Kelly.

I like the explained kelly and 1/2 kelly strategies, but it also makes me nervous.

I used to play with a border-line ******ed BRM strategy(around 20bi's with random shot taking when I felt like it), then kinda smartened up and did the complete opposite with strict 100bi management.

Now I play pokerstars HU hyper-turbos. I'd love to move up to the 60s, but figured I should wait to get to 6k to not regret anything later.

But kelly tells me that with my current r.o.i of 4% after 2k games, (it's actually higher but I assume it's more along 4% long-term.) I should be playing 100s, or if assuming 1/2 kelly, i should be playing 60s with a BR of a bit over 3k.

Does this seem fine for hyper-turbos? I might try the 1/2 kelly strat, but i want a relatively small/comfortable risk of ruin. But at the same time, if I'm wasting my time at the 30s, maybe its time.

Cheers,
CK

I don't think you can assume a long termb 4% ROI unless your name is Serkules ;-)
Anyway I wouldn't play Hypers with 50 buy ins because it is not uncommon to lose that amount due to variance unless you are disciplined enough to move down according to Kelly.

I guess 5.5% is out of the question? I'll update you after 2k more games


But ty for the advise, I knew kelly was too good to be true with hypers, so I'll just stick to my 100bi as long as I can. Degen at though


Cheers,
CK

Ok, I think I get it.

And if my calculations are correct, the expected log wealth method is even more aggressive dan regular kelly.

If you have a $3000 roll, you should be playing the $100 in this example (52% ITM) as U(100) > U(60). That is 30 buy-ins.

If you use regular kelly, you should have 50 buy-ins (ROI 2%).

And the 30 buy-ins isn't prob even the lowest as I just took $3000 as an example.


Also, a question about the variance in the kelly method in general, variance for 1 table husng's is always 1?
As the variance in regular speed differs from turbo which again differs from hypers (?)

52% win rate =! 2% ROI. To calculate ROI you have to use rake and stuff.

Also to the discussion of hypers, if you can move down then you drastically lower your risk of ruin, so if you have discipline you can do something like move up with 50BI move down when you fall below 35BI.

actually you're right, but the difference is really small and assuming it's +1 or -1 is actually a bit on the safe side anyway, so you most probably won't be missing anything by going w +1/-1 aside from not taking a shot w 25 buyins when +1/-1 says you should take it w 26 buyins or smth like that

Kelly has no risk of ruin since you can always drop down levels, and you def have a nice cushion even at 100s, but if you're looking for a sane grinding experience, it's probably not gonna cut it at hypers. the problem w kelly in supers/hypers is that in order to use it efficiently you need to always be aware of your bankroll and have the discipline to stop playing and move up or down when necessary. in regspeeds/turbos your regular session is like 50 games at most and you can usually keep some sort of a score on how much you won/lost. in supers/hypers you play much more games in a much shorter time frame so actually keeping track is hard and it's also a big mental strain as you will see swings of 10-20% of your roll in half an hour sometimes, and the downswings that take days in regspeeds give you time to process the new events and calm down, while in hypers playing full kelly you might have to move down two levels in one days etc

if you have discipline and confidence in your play, you might play full kelly, but I think 1/2 or 1/3 kelly is better for sanity

yeah the logaritmic utility is the backbone of kelly, kelly is a simplification that errs on the safe side, that is missing out on opportunities to wager slightly more than what Kelly says for slightly higher "variance"

fwiw the example is wrong since the rake figure is wrong (you don't win 100-1.88 but 100-3.76, opponent has to pay rake too ). the real "tipping point" is at $3233 (http://www.wolframalpha.com/input/?i...%29%5E0.52%3D0)

the variance which I'm talking about is the mathematical variance, which is a number that depends on the distribution of outcomes. Since the outcome of a single HU match is +0.95-0.99 BI or -1.01-1.05 BI and the probabilities of either event are pretty close together, the mathematical variance doesn't vary much between games and is ~1 for all formats (it doesn't get significantly lower until you get to ~65% winrate)

the problem is that most ppl see "variance" and think of what we call variance and that's a term for "zomg my graph is a rollercoaster" and not the same term as the number that's called variance in statistics

Yeah I know, it is ~2% ROI ( 0.52*(200-3.75)-100 = 2.0448).
But for the example took just 2.

Ok, so that is were the variance of regular vs hypers is shown. You have to move up and down a lot more in hypers.