Episode 108: David Einhorn

26 thoughts on “Episode 108: David Einhorn”

1. I liked this very much. Good job guys.

   For the 100th episode you asked for “dream guests”. David was my choice, and here he is 8 episodes later.

   I’ll have to think of who to list next, but one of the people David mentioned, Andy Frankenberger, sounds interesting.

   • Is he like your Wall Street man crush? I’d love to see you get giddy over him like Andrew was with Doyle.

     I love studying billionaires and otherwise uber successful people. I found it odd that he didn’t feel the urge to give courtesy laughs. Sometimes I’m like that too, but only in person.

     I love his take on figuring out the table as if it was a puzzle. This is something I learned from the TPP Premium episodes and have started to think this way more often recently.

     • I follow him and I respect most of what he’s done as an investor, but he is a crush in the sense that he is a great combination of two worlds I care about, poker and investing.

       • Aaron Brown would be a great guest for that intersection.

         • the Aaron Brown suggestion is worth considering. guy is totally fascinating.

2. I was surprised you guys didn’t bring up Andy Beale when David was discussing how he gains an edge by not caring about the stakes. I wonder if he has read Craig’s book.

   I thought David’s outlook on poker was pretty fascinating, but also somewhat peculiar. It was an odd combination of viewing the game seriously enough to think about the development of his 3 and 4-betting ranges and wanting to test himself against the best in the world, but flippantly enough to only play a handful of times a year and see it as nothing more than a fun game to be played purely for recreation. It has to be hard to take any game that seriously in such limited exposures.

   Maybe his personal wealth dictates a lot of that, but even controlling for that, it’s odd. Maybe Richard Brodie would have some insight?

   • I’m surprised we didn’t ask about that, too. I meant to. Mea culpa.

3. I think Nate’s vocabulary setting was turned up a notch or two for this one.

4. Congrats! Your podcast just got linked on money.cnn.com!

   http://money.cnn.com/2014/12/30/investing/david-einhorn-poker-versus-investing/index.html?section=money_topstories

   • I think this type of publicity is also a factor in his limited responses. Also it appears to me that he might be a little shy, which is fine but helps explain the contradiction Matt noted.

     • I could sense either that or some slight social awkwardness. Same thing with Kevin. I kinda like that in guests the though. Easy to be yourself here.

5. Really great strategy piece this week, tremendous clarity of thought

   • Thanks! That’s great to hear.

6. I’ve spent my spare time over the last six months writing a counterfactual regret minimization code to find the GTO solution for a given pair of ranges, flop and game tree for NLHE. It’s (mainly) working now and I’m looking for examples to try it out on, so I ran it for the example 3bet pot hand in this podcast.

   It’s quite tricky to work out what preflop ranges Nate and Andrew are talking about, so I’ve gone for a mainly wide value range for BTN, but with polarized suited Aces, and a narrow midstrength range for CO.

   CO: 22-TT, AJo, ATo, AJs-A9s,KQs, KJs, KQo, QJs, JTs, T9s, 98s,87s,76s
   BTN: TT+, AQo+, AJs+, A5s-A2s, K9s+, K8o+, Q9o+, Q8s+, J9o+, JTs, T9s, 98s, 87s,76s,65s

   I’ve allowed a 2/3 pot bet or raise as well as jamming at every decision node.

   After a few hours of computing, I get an approximation to the GTO solution. It’s exploitable by less than 2% of the pot, which is 60c in this case. It includes the effect of rake on NL200 at Pokerstars. I came up with my own algorithm for bucketting hands which is not completely lossless, as it misses some subtle blocker effects, but which seems to work pretty well. My CFRM algorithm is mainly standard, but with a few small tweaks I came up with. The whole thing is written in MATLAB, my computing environment of choice, with some C++ to do the computational heavy lifting. More details at http://www.pokerstrategy.com/forum/thread.php?threadid=324733 .

   First of all, although CO has about 40% equity with his range on the AhQhTd flop, the EV is split about 4bb to CO, 9.5bb to BTN and 1bb to rake at equilibrium, so only about 28% of the pot to CO. I also tried a Jh7c2d flop, and the EV was split about equally, with 1bb to rake. Pokerstars always seems to make at least 1bb per hand at equilibrium. Nice work if you can get it.

   On the flop, CO should check his entire range, after which BTN should bet about 50% of his range. A lot of hands have mixed strategies, but hands that should usually bet include 9d8d, 9h8h, K8, K9, KQ, but not KdQd, AK, but not AdKd, QT, AQ, including AdQd, AA, KJ, but not KdJd, J9, JT.

   The exact solution for KhTs is check 71%, bet 29%, so I’ll assume we check.

   On the Js turn, CO should check his entire range. If he does bet, the solution has BTN overall folding 22%, calling 61%, raising 13% and jamming 5%. As you can, see, it starts to get complicated to write about. The majority of jamming is with Kx, including KhTs, which should jam 13% of the time, raise 19%, and call 69%. I’ll assume we call.

   On a Qs river I have no strategy for CO since he should have checked the turn. If he bets, the solution has KhTs folding 40%, calling 57% and jamming 3%.

   If we go back to the turn and make CO check, Kx is a BTN check 14% and a bet 86%. After a bet, CO should usually jam KhJh, call with JhTh-7h6h, Kx, 98s, TT, AJ and most of his AT combos. If he calls and the river is Qs, he checks 95% of his range, mainly betting with some TT. After a check, BTN should check back 77% of his Kx, bet 23% and call a jam with 44% of these. Overall BTN checks back 55%, bets 23% and jams 22% of his range. He jams air plus some full houses. He bets with air, some straights and some full houses.

   I liked the thought process that made Andrew jam the river, but it appears to be a purely exploitative strategy, since a balanced CO will still have enough full houses in his range on the river.

   As you can see, computing the solution is one thing, interpreting it another, and as for learning from it, it needs a lot of work……and this is all for one particular turn and river. The algorithm gives the solution for any turn and river.

   I’ll probably try to do this again for any HU pots that come up on the podcast, and would be interested to see whether anybody finds this helpful.

7. Yes this is very interesting and well explained with the right amount of details. I have also written a solver based on Will Tipton video pack but it is too slow to get complete trees like this…
   I would be interested on the sensibility of your solution to slight changes in ranges (mainly for CO) do you get same cutoff hands? And mixed strategy/ pure ones on the same spots?
   More is welcome.

   • If you suggest some ranges, I’ll run it again and see what happens.

     • Maybe just considering CO calling the 3bet with both or a combination of AQo and AQs would be interesting. PokerSnowie also think it is sensible, but does not like 3betting with KT on the button as it must be an exploitative play against a too wide CO.

       • Given the board and the runout, my prediction is that this will ship a load of EV CO’s way! :^)

         I’ll try to set the calculation going tonight. My kids monopolise my laptop at the weekend, and I’ve also been very distracted by those clever people in Alberta and their solution of HU FLHE. They used something called CFR+, which is just a small change in how they do the regret matching, and I’ve been trying it in my code. I haven’t noticed much performance improvement, but I don’t know whether they’re using some sort of sampling too, so I’m looking forward to seeing the details when they publish them.

       • Ran it. CO’s EV went down. Checked it. I’d forgotten to given him KJ. (facepalm)

         I’ll get back to you later!

8. OK. Done it. CO’s EV goes up by about 0.5bb, but he should still check his whole range. However, BTN checks behind rather more, 68%, now that CO’s range is stronger. I couldn’t see much difference in BTN’s betting range, just his frequencies. My graphical user interface is rubbish (I’ve never written one before because I’m over 40 and hate computers), so it’s not easy to compare. The solution for KhTs is now check 95%, bet 5%, consistent with this observation (about BTN’s frequencies, not my age).

   On the Js turn, CO should still check, but if he bets instead, BTN’s frequencies don’t change much. If CO check/calls and the river is Qs, CO’s frequencies change a lot, because of the AQ combos now in his range. He checks 73%, raises 14% and jams 13%. He jams with some fraction of his AQ and some JT as a balancing bluff. His raising range is mainly TT, some AQ and some bluffs. Note that he checks 41% of his AQ combos to check/raise. If CO bets, BTN calls with 63% of his Kx and folds the rest. My bucketting algorithm puts all of BTN’s Kx into one bucket, which neglects any blocker effects (e.g. personally, I’d rather call with KQ then K9 on that river).

9. Thx a lot! This is great and confirms your analysis.
   I tried to install your software, but even the small display one is too huge for my laptop… any way you could reduce it even more?

   I don’t play LH so I don’t know how the results can be interpreted (limping AA 0.5% for example) and how it translates into NLH. Anyway you can find more information on http://webdocs.cs.ualberta.ca/~johanson/publications/poker/2015-science-hulhe/2015-science-hulhe-supplement on how CFR+ is different from CFR.

   • I’m going to release a new version of my solution viewer soon, which I’ll make a bit smaller…or you could get a bigger monitor. :^)

     I don’t think you can interpret the LH results in terms of NLH. Even with only one bet size allowed, the decision tree is a couple of orders of magnitude bigger in single raised pots, and so different that the ranges will change completely.

     I know how CFR+ works. I’ve implemented it in the latest version of my code. I’m currently testing whether it speeds it up appreciably.

   • I’ve updated the software. Have another try. The interface is smaller now.

     I feel like we’re using these comments for a private conversation, so maybe go to the thread on Pokerstrategy if you have any more questions, or my Skype is jbpatzer.

     • FWIW I’m pretty interested in this and you’re welcome to continue the discussion here. I’m sorry I haven’t chimed in, but I’ve been kind of overwhelmed lately and also don’t have a lot to add, but I definitely enjoy being a fly on the wall.

       • OK. I’ll try to post analysis of any other suitable hands that you discuss in upcoming podcasts.

         • It interests me and I think the community would benefit if you do continue this.

Comments are closed.