http://faq.holdemmanager.com/questions/120/EV+Explained

Not sure if this has been posted already.
Watching it right now (22 minutes long).

Just watched it. Not that great. This thread is probably better.

These aren't really the case. The adjustment (not the red line but the difference between the lines) doesn't depend on getting it in good or bad, but on the actual outcome relative to expectation. As said above, the situation you describe would usually result in gaining equity without showdown, so increasing both red and green lines.

Assuming your plays were +EV, the red line (and possibly green line) may go down if you are dealt a high proportion of bad hands and/or your opponents are dealt a high proportion of good hands (for example if you keep running into aces bullying the bubble). This is an element that is not adjusted for in the red line and, as said, a reason that you still need a reasonable sample to draw meaning from even the red line.

Are you sure about that? I thought it was all in EV. HEM needs to know opponents hole cards in order to calculate the EV. Take a look at some of your hands where you pushed and scooped the blinds, even some where you gain 20% of your stack, I think you'll find that your '$EV Diff' is $0

It does so but indirectly. Think about what the red line's doing when you never get allin in a SNG. It equals the green line. So in the case where you call on the river with a losing hand early. Red line = green line = -1 buyin. Even though your $EV Diff is $0. You could conceivably do the same thing by folding/stealing into the money with never having a showdown. Get into a postflop hand and call and lose on the river, finishing 3rd. This time red line = green line = + 1 buyin.

More typically, you gain equity every time you steal. Say going from 1500 stack to a 2000 stack. Now when you get allin, you're equity is higher with the bigger stack and the red line goes higher than if you got called with your 1500 stack. Again, this is generally speaking as some ICM things could happen where your stack goes up but equity goes down along the way I suppose.

Ah ok, I see what you mean there, I obviously thought you meant different!

Interesting, the 2nd part of your reply. I never thought of it like that, but it makes sense, kind of. But couldn't it lead to potential false positives? If for example, I make a bad push but take the blinds, increasing my stack from 1500 to 2000 and then get it in with a +$EV Diff outcome (eg, I lose 77 vs Q6). With the new $EV based on 2000 instead of 1500, could that not appear to be higher because of the bad shove? Or would we not even care about that and treat each $EV Diff on a hand by hand basis?

I may be reading into this totally wrong though...

Cheers

While I haven't read every response in this thread, there does seem to still be an open ended question of "what % of the spectrum regarding overall luck spectrum does all-in luck account for?" I recently went back to an old Stoxpoker post that I made part way through last year and saw a response from phzon (one of the most amazing mathematical minds that I've come across in the poker world). I'm just going to copy and paste his response here, but I was surprised that he was able to quantify how much luck the red line accounts for. My question on the stox forums was mainly geared towards super turbos, so take that for what it's worth, but I would definitely like to hear a discussion on his line of thinking. Is this a good answer to the question about how much luck the red line accounts for? He responded after I showed a graph of 4.8k sngs played where I ran 165BI under equity (my bad run eventually topped out at -350 BI under ev). For cliff notes, basically just read the last 2 paragraphs.

Here's the original post:

4.8k tournaments is about 70^2. If your tournaments are independent, then the standard deviation for your total results after n tournament is about 160% * squareroot(n), and the standard deviation for your average result is about 160% / squareroot(n). So, after about 4.8k tournaments, the standard deviation of your total winnings is about 1.6 buy-ins * 70 ~ 112 buy-ins. The standard deviation of your observed ROI is about 160%/70 ~ 2.3%. About 2/3 of the time, you will be within 1 standard deviation of your long term average, and about 95% of the time, you will be within 2 standard deviations of your long term average.

The 160% comes from the 50-30-20 structure, which is the same for regular 9-max SNGs, turbos, and super turbos. It actually varies slightly by the rake and by your playing style, but not nearly as much as many people expect. Someone trying to double up early might have a SD of 160%, while a tight-early player may have a SD of 150%. These differences are microscopic compared with the differences in other forms of poker, and if you make plays in SNGs to reduce your $ variance, you are probably rationalizing mistakes.

Being 165 buy-ins below expectation after 70^2 tournaments is about 1.5 standard deviations below, which is bad, but not spectacular. Ah, but all-in luck is just a component of luck. If it's a small components of luck, then being 165 buy-ins below expectation on that component is really rare.

I took about 2000 9-handed regular NLHE tournaments analyzed by HEM, exported the results, and imported that into Excel. The standard deviation of that player was 1.52 buy-ins. The standard deviation of the luck adjustment was 1.06 buy-ins. The standard deviation of the luck-adjusted result was 1.01 buy-ins.

If you raise almost all-in, and put the last few chips in after the flop, that shifts the luck of the flop from the all-in category to the pre-all-in category. It doesn't reduce your variance, but it slightly reduces the effectiveness of the luck adjustment algorithms. So, style differences can affect how much luck is all-in luck.

All-in luck may be a larger component for turbos or for super turbos, but not as much larger as you might expect. Even if your only options are to push or fold, there is still a lot of luck which is not counted as all-in luck: Which hands you are dealt, which hands your opponents are dealt, and what happens when other players collide when you have folded. It's lucky for you if the big stack takes out the 2nd stack on the bubble.

165 buy-ins down after about 70^2 tournaments is probably closer to 2.0 standard deviations down rather than 1.5 standard deviations down.

If the 1.01 buy-ins from the regular tournament sample applies, then luck-adjustment factors out 50-60% of the variance, and it takes only about 40-50% as many tournaments for breaking even to be 2 standard deviations below the mean. For example, suppose your true ROI is 5%. For breaking even to be 2 standard deviations below the mean, you have to play about (2 * (160/5))^2 ~ 4000 tournaments for breaking even to be 2 standard deviations below the mean. However, it only takes about (2 * (101/5))^2 ~ 1600 tournaments before your luck-adjusted results will exceed 0 with high confidence.

----------------------------------------------------------------------

I just analyzed a set of 10k+ super turbos, and thought I'd update the calculations on this thread.

Unadjusted standard deviation: 1.49 buy-ins/tournament.
Adjusted standard deviation: 0.84 buy-ins/tournament.

HEM's luck adjustment factors out about 1- ((0.84)/(1.49))^2 ~ 73% of the luck in super turbos. The adjusted results after about 270 tournaments are about as reliable as the unadjusted results after 1000 tournaments.

----------------------------------------------------------------------

Please discuss...

It won't know the difference between any uncalled shoves. So yes 72 offsuit being pushed through 6 players for 10 BBs will look the same as AK suited when not called.

so this mean i was running bad and probably started playing bad cause of the beats and then back to almost running even ev right?

Over a sample of 150 sngs it means nothing aside from the fact that your hands didn't win as much as they should have vs. that of your opponent's holding. It's still going to take a fairly decent sample for your red line to give you any indication of your overall play.

Also, your graph can't tell us if you tilted. Maybe you played perfectly for 115 sngs and the deck was just stacked against you. Either way, I'd imagine that your red line ROI is entirely unsustainable at +20 buy-ins for every 150 games (assuming these are STTs).

I guess I run good....?

dam u. lolz

i run bad? HM is just bad imo

the EV line does not work for 18mans.

Just found this thread with google - In Holdem Manager my ev$ roi line is exactly the same as my ROI line over almost 700 45 man sng's. I'll read the full thread tonight but is this common?

edit: this is on the Trending graph so maybe this is the wrong thread

EV line doesnt work for 45 mans so will always appear to be the same. Same as how it doesnt work for 18 mans as well - any tournaments that have > 1 table - see earlier in thread for discussion.

Ok this is a pretty good post. I believe it concurs with some stuff Juk and others posted earlier in this thread wrt to standard deviations in 9 man tournaments.

However I am certainly not in the same division as Juk and some of the other posters wrt the maths side of things (plus I've had a few drinks) so I'll leave him to comment.

I will say that this is a very articulate and nicely structured post from Phzon on the subject though.

Great post! This actually falls in line almost exactly with my methodology (albeit mine is much more of a pain and takes some time to calc). Here's a link to my post and methodology:

http://forumserver.twoplustwo.com/36...=#post16527049

Since then I've played another 20,000 SNGs and ran another 160 buy-ins below EV, for a total of 100,000 buy-ins and 610 buy-ins below EV.

So using this new method, for my last 100,000 SNGs my luck adjusted results are in about the ~1.2% range. This means that I'm on a 1-100 cooler for my last 100k SNGs. Sucks, but I guess still within reason. Too bad that in real dollars it means that I'm over $250,000 below my EV lifetime.

Thanks again SayGN for the great post!

Can someone check this for me?

My green line is 960 buy-ins below my redline over my last 67,000 SNGs. What are the odds of that?

Thanks!

I'm aware the redline has limitations but does anyone have a reason for the distance between red line and profit becoming bigger the longer the sample size. it seems like the more games you play, the further apart the profit and EV line become. I thought the lines were supposed to merge. Or is it a case that green line jumps to EV red line happen much quicker than the gradual redline increase above green line?

did that make sense? idk. is that just me that noticed the pattern?

^^^

Sample size?

My graph looked like that for the first thousand. Then it reversed over the next thousand.

I suspect it's pretty unusual to get samples that don't have apparent patterns.

The difference in terms of ROI decreases with time. The difference in total $ amount does not.

so the red line will drop to the green line rather than green line up to redline?

It means that it is not expected for the total difference between adjusted and actual results to be smaller after a greater number of tournaments. In fact the opposite is true, and there is a greater chance of the lines being greater than a given distance apart after a bigger sample size.

In terms of profit per tournament or ROI, however, it is expected that the difference is smaller the bigger the sample.

See this post by Finnisher:

http://forumserver.twoplustwo.com/sh...&postcount=354

Don't thank me, thank Phzon; he did all the math, I just copy/pasted.

That being said, I actually question whether or not the red line could really equate to ~73% of the luck. While I cannot necessarily back this up with any mathematical info, I played a lot of tournaments and have graphs to compare to one another.

Bearing in mind that the graphs show below are 100% $14+1 super turbo 9man sngs, I selected two periods that occurred 7 months apart. After analyzing every blind level, I can say that my play is nearly identical with the exception of sb and button steal % (which was 15% higher at the first blind level, yet my steal success was somehow still higher than when I was stealing with a significantly tighter range). Due to that data, we can assume that I was running either very hot in unquantifiable luck in the first graph, very cold in unquantifiable luck in the second graph, or both.


Please ignore the green line. The focus of this post is to analyze the story that the red line tells.





In my first graph, my red-line ROI was 6.5% and in my second it was only 1.6%


Assuming that the level of competition remains unchanged, and the skill of our hero (me, ldo) is the same, these graphs make two contrasting arguments.

1. The red line represents roughly 73% of overall luck, which means that the sample size required for our green line to give us a good idea of what our true ROI is even for low-level super turbos is astronomically large.

or

2. The red line must represent some number significantly less than 73%

If we assume that argument #1 is correct, it would mean that for each graph shown above, the red line gives as much information as my green line would show over roughly 10,340 sngs. Even if we assume that my two graphs represent the two extremes in unquantifiable variance for someone who has whatever my true ROI is, we can see that 10,000 super turbos is not a big enough sample for our green line to give us a concrete idea of what our true ROI is.

Regardless of which argument is correct, these graphs show that a tremendous amount of tournaments are needed to accurately depict one's true win rate (at least in super turbos).

As I've played roughly 10,000 sngs at this limit since last August with an ROI of 3.5% (both green line and red line are currently about equal), I sincerely hope that argument #1 is correct because it means that there is a significant leak(s) in my game somewhere. If there is a leak, I can plug it and at least count on the red line to be of assistance for the occasional sanity check. If argument #2 is correct, it means that the red line, even over a several thousand sng sample, comes nowhere near telling the whole story and is not all that useful. It would be frustrating if that was the case.

Maybe this assumption is the problem?