Jurgis

Putin is basically trying to go after those who put sanctions on his cronies, and he knows that even if he doesn't get them, he's creating a huge chilling effect by trying to scare other officials who might think about going after the Russian Mafia by making it known that he's going to try to reach them even outside of Russia (as he has done repeatedly to Browder). Translation is rather crappy - I guess Google Translate? - but hopefully the meaning is somewhat understandable. Like Liberty says, Russians want to interrogate a number of Americans who they claim are connected to "Bill Browder case".

I'm just gonna grab popcorn and watch how any company will try to compete with FB data portability or not. What may happen is data slurping by big cos like LinkedIn/FB?/Google?/etc. already do with your email contacts. "Give us your FB social graph, we gonna spam people you may know to get connected to you on NotReallyNewSocialNetwork". Oh wait, doesn't the privacy laws preclude that? Doesn't person Y have anything to say when you "port" the fact that you are connected to them? And BTW, this has been attempted multiple times via email harvesting. I've been invited to a bunch of NewSocialNetworks by relative/friend spam when they "port" their email data to some crappy/spam social net wannabe. Good luck. ::) Not that FB should rest on laurels. Like the paper said rightly, social network dies because it falls behind, is not used, UI sucks, etc. It's definitely not a "do nothing and collect checks" business. But then almost nothing is. Edit: what Liberty and merkhet said too.

Good paper. Hits quite a few important points that some people don't consider when they talk about data portability and/or FB/GOOGL regulation.

Jedi will crush the Galactic Empire and expand beyond! 8) Sorry, not picking on you.... well maybe a bit. 8)

https://en.wikipedia.org/wiki/Sergio_Marchionne

I've heard this argument before and personally I don't buy it. Yes, if I go to superrandom site that I use only once and I have a choice of PP or CC, I use PP. But if it's a site where I buy more than once, I choose CC. PP for me is a hassle because I have to do extra login, something may go wrong, etc. rather than CC being a hassle. But then maybe I'm not the common case... ::) Anyway, I believe glorysk87 argues that the PP competitive edge is with merchants. At least that's how I understood. I think interesting questions are - and I'm sorry, but I probably won't have time/interest/etc. to dig out these: - What are PP terms with eBay? Why is eBay leaving? Do these reasons apply to other merchants? If big merchants like eBay get low fees from PP, what these fees are? I doubt they pay 2.7% like random PP one-person-merchant account. OTOH, I did not find offhand what the big merchant fees are. Maybe they don't disclose. I think it also would be interesting to learn more about CC processing competition. Not just nonames, but companies like BAC, etc. Anyway, I think overall I'll admit that I don't know enough to have strong conviction where PP goes from here. They definitely have some brand name, some moat, some good areas. But I'm not confident enough to value these or calc growth/profitability for long term. Good luck

http://www.leadersmerchantservices.com/sq/index.php?st=top10bestccp&vendorSubId=AFF_pM2Aeokxgj I don't know who these guys are and if they are legit and whether their ads are bait and switch. So caveat emptor. Just showing I did not invent the pricing. 8)

Pretty profound misunderstanding of what PayPal does. To your point about accepting CC's - yes, they would switch to accepting CC's. That's a large part of the value proposition of PYPL. If you have an online store, you can use Braintree to set up a merchant account and payment gateway, allowing you to accept CC transactions. If you have a bricks & mortar store, you will be able to use iZettle to accept CC's and other forms of payment. If you have an app-based store, you can use Paydiant to enable those capabilities. Same for Venmo and P2P. The list goes on. To boil it down - PYPL is positioning as the service provider that will allow any company to accept any payment in any form at any time. And they're positioning themselves to capture value every time a transaction occurs anywhere along that chain. So you are saying that merchants have no choice but go through PayPal and pay them extra if they want to accept CCs? At which size can a merchant say "screw you" to PayPal and install infra that does not cost them PayPal fees while accepting CCs? What? No I didn't say they have no choice. I was responding to your original post in which you were asserting that PayPal and CC's were mutually exclusive and the fact that you seemed to not know that PayPal offers a service and makes money by enabling businesses to accept CC's. As for your most recent response - there are a few alternatives that a business can use to allow them to accept CC payments. I'm not sure what you mean by saying "install infra that does not cost them PayPal fees while accepting CCs" - any service that a business uses to accept CC payments will cost them money in the form of transaction fees. Maybe you could clarify what you mean by this? I think you misunderstood my original message. Yeah, I know that merchants can accept CCs through PayPal, i.e. customers pay with CC that goes through PayPal. I used shorthand "accept CCs" and meant "accept CCs not using PayPal". I see how that was not very readable. Regarding second message, maybe that was not clear either. From what I've seen PayPal charges pretty high fees for accepting CCs through them. I was asking at which point in time it's worthwhile to go for other CC processing solutions. With a quick search, I see tons of solutions being advertised. Some provide free hardware + software (infra). Some of them at the level of 0.15% fees. (I don't want to post promotional links here, since I have no clue how good and reliable these are). Some don't disclose fees, so I can't say offhand what the fees are. I'd guess if merchant is big enough, they can setup their own infra too (some websites claim as much). Though likely even big cos mostly outsource if they can get free HW/SW and 0.15% or lower fees. Although perhaps PayPal also offers way lower fees for bigger merchants. And clearly there are other questions of the support where PayPal can come out on top (or not): - Fraud prevention - Cash deposit time - Cash holdout for fraud/returns/etc. - Analytics/etc. Anyway, if I understand correctly, you say that PayPal provides good service for merchants, but that the PayPal-pay (not CC through PayPal, but the PayPal-account) ability is not that important for customers. And Venmo is separate topic. Yes? Edit: this is presumably the latest PayPal merchant fees: https://www.paypal.com/us/webapps/mpp/merchant-fees?&51376578128&PID1=sHCsdwMbB%7Cdc&PID2=HCsdwMbB&act=%5BAccount%5D&adgroup=TM-SMB+-+PayPal+-+Fee+(e)&adposition=1t1&adtype=%7Badtype%7D&campaign=TM-SMB+-+PayPal+Fee&ch=SEM&creative=51376578128&ct=SMB&devicemodel=&e&eu84e2fpi0&gclid=CJTOr9WAqMICFZKBaQod50MAbw&geo_country=US&geo_region=NA&kw=paypal+fees&matchtype=e&mpch=ads&mplx=27722-205546-2056-1&mv=google&network=g&obj=Acquisition&paypal+fees&spid=870902502189460310&target=&tt=TM 2.7% seems quite high and only worthwhile if there are no other choices. That's why I assumed any merchant would migrate away from PayPal asap. Am I wrong? Edit2: It's possible that PayPal 2.7% is a complete fee (i.e. merchant gets price minus 2.7% and that's that), while other services quote 0.15% fee that they charge but there's other fees, so merchant gets price minus 0.15% minus ??%. This is something that I don't know. :)

Well said. 8)

Pretty profound misunderstanding of what PayPal does. To your point about accepting CC's - yes, they would switch to accepting CC's. That's a large part of the value proposition of PYPL. If you have an online store, you can use Braintree to set up a merchant account and payment gateway, allowing you to accept CC transactions. If you have a bricks & mortar store, you will be able to use iZettle to accept CC's and other forms of payment. If you have an app-based store, you can use Paydiant to enable those capabilities. Same for Venmo and P2P. The list goes on. To boil it down - PYPL is positioning as the service provider that will allow any company to accept any payment in any form at any time. And they're positioning themselves to capture value every time a transaction occurs anywhere along that chain. So you are saying that merchants have no choice but go through PayPal and pay them extra if they want to accept CCs? At which size can a merchant say "screw you" to PayPal and install infra that does not cost them PayPal fees while accepting CCs?

I think I bought a bit when PYPL was spun off and I still have that position. That said, I don't really know why this is attractive. OK, I am not Venmo user, so perhaps I don't get that. Regarding PayPal, I have used it about once per year in the last 2-3 years. What is the use case for this? Who is using it? Is it all consumers purchasing from small online businesses and international? Isn't that a bad space to be, since any businesses that grow start taking credit cards? Are there any users who prefer to use PayPal vs using a credit card directly? (Maybe international, IDK). I see some posters above think that PayPal bring a lot to the table from business owner perspective I guess. But really wouldn't you switch to accepting CCs if you grew to any reasonable size? Would your customers keep using PayPal if they had a choice of PayPal vs CCs?

CANSLIM ad really... ::)

I have to compliment you: you do hit the right spots. 8) I also thought that their data cleaning/preparation was suspect. So I ran on data without their "missing dates" transformation. I thought I'll get bad results with that and then the results will jump when I do their transformation, which would prove that their results are caused by bad data preparation. No cigar. I get the same bad results with original data and with data prepared based on their description. Which does not prove that their preparation wasn't broken... it might be broken in a way that's not described in the paper. Or maybe the description does not match what's in the code. Anyway, maybe this is enough time spent on this paper. Maybe the right thing to do is to wait if they gonna publish the code (or ask for the code). Or just conclude that their results are broken and we just don't know why. 8) I'd still be interested to discuss with someone my implementation and where it might be different from theirs. Just to see if I missed something obvious or did something wrong. But my code is hacked up mess, so it's probably not a high ROI for anyone to look at it. 8) I could put my code on github... oh noes, I'm too ashamed of the quality to do it... ::) Anyway, thanks for discussion so far. 8) Ah, regarding I write this off as academia. People may be more interested in results/papers/thesis (I think this was master's thesis for one author) than in applying it in real life. Almost nobody from the people I know transferred their thesis/papers into actual startup work. Maybe it's more common nowadays and in certain areas, but it's likely not very prevalent. I guess this area/paper would be easier to transfer into money making than other theses, but they still might not be interested. A valid question though. I'm not cynical enough to suggest that they know their results are broken and that's why they published instead of using them themselves. I somewhat believe people don't consciously publish incorrect results. But who knows. ::)

I dont know what the authors did but ill reiterate from before vanilla LSTMs do little better than guess on the stock market. They probably had like 1000 GPU and tested thousands of hyperparameter configurations and "overfit" the test set. This is why typically papers like this are not believed anymore in the ML literature. Try adding some stuff like attention or skip connections and whatever else is hot now (I'm not sure) and didnt someone recommend GRUs instead. I have some other ideas you can use like Gaussian Processes to estimate realtime covariance matrices, but your better off looking at the literature first than trying out hairbrained ideas that might not work. It's really not a trivial excerse to outperform the market with ML. Ah, I think I see where there is a miscommunication between us. :) My goal is not to outperform market with ML. My goal is to understand whether what is proposed in this paper works and if it does not then why. 8) You are possibly completely right that what authors propose does not work. I just want to understand how they got the results they got. You've said "probably had like 1000 GPU and tested thousands of hyperparameter configurations and "overfit" the test set." before. I don't think that's the case at all. If you read the paper - which you haven't so far - you can see that their training is really simple and there's no "thousands of hyperparameter configurations". Which is baffling in itself. I have some suspicions of what could be wrong, but it's not productive to discuss it if you just dismiss the paper offhand. Which is BTW your prerogative - if that's where you stand, that's fine and I won't bother you with this further. 8) You are entirely correct that I haven't read the paper and maybe I was too hasty in dismissing the paper. I wouldn't mind a copy of the paper if you don't mind sending me one. That being said here is my reasoning in more depth. The authors seem like they are in ML acadamia, so I made a couple assumptions. 1.) It didnt look like their paper made it to one of the premier conferences and maybe its because they aren't big names but likely its because people have been training LSTMs on stocks for a long time and vanilla LSTMs dont work well and I think everyone in the ML community is suspicious of 80% hit rates using a vanilla LSTM on indices for good reason and they likely didn't do anything special to assume that they didn't just get "lucky" with their model. the reason they got "lucky" is number 2) typically papers dont discuss the hyperparameter search they go through to find the exact correct configuration, so even if they didn't say they tested 100s/1000s of hyperparameters they might have and likely did (although yes i didnt read the paper). Unless they specifically say there were few or no hyperparameters to test or they tested only a few of them, you should assume they did test many. This is a dirty secret in ML, you come up with a new technique and you dont stop testing hyperparameter choices the model until you get good results on both the test set and validation set. Then you submit to to a journal saying this method did really well because it outperformed on both the validation set and test set. But you stopped right after you get a hyperparameter choice that met those criteria which strongly bias your results upward. This is related to p-hacking. This is a perfectly natural, but bad thing people do and usually means most papers have performance that can't be matched when trying to reproduce them. You can pick basically any method of the thousands that have been proposed and if it doesn't have over 1000 citations (and the method actually seems useful) this is probably one reason why. Now you maybe you are right and something else may be missing, but if I had to guess I think its a good chance the authors just got "lucky". BTW why dont ask the authors for their code. Its customary to either give this stuff out or post it on github. As a side note: Even a vanilla LSTM has many hyperparameters: number of states, activation type, number of variables to predict, test/train/validation breakdown, number of epochs, choice of k in k fold validation, size of batches, random seed, how they intialized weights (glorot, random nomal, variance scaling..) for each weight in the ANN, the use of pca or other methods to whiten data, momentum hyperparameter for hillclimbing, learning rate initialization, choice of optimizer... My point is that even with a vanilla LSTM the author can pull more levers than can be hope to be reproduced if you don't know absolutely everything maybe even down to the version of python installed to reproduce the pseudorandom number generator. No doubt some of these choices will be mentioned in the paper, but many of these choices won't be typically, which makes any reproduction difficult. And typically the authors are the only ones who are incetivized to keep trying hyperparam configurations until one works. The real papers that are sucessful are typically methods where either its not impossible to get a reproducible and externally valid hyperparamter configuration, or something that is relatively robust to hyperaprameter choices. I sent you the link to the paper. If you look at table 1, there's couple things to notice: Yeah, for Adagrad, the accuracies are all over the place. But for Momentum and Rmsprop they are all quite similar and way higher than 50% (which would be random guess). So I think this somewhat shows that they did not just pick a single lucky combination of what you call hyperparams. You can still argue that perhaps there's a lucky hyperparam that is not shown in Table 1. That's possible, but I guess it's becoming less convincing. ;) OTOH, I did not run all the combinations they presented in Table 1, but from what I ran, the results were way more stable and clustered at 48-52% range. So I wonder why they are getting much wider dispersion than I do and why their results have so much better accuracy. So I wonder if their results are correct. In other words, you question their results because you think they hyperparam hacked. I question their results because I think there's another issue somewhere. But I don't know what it is. I think you're a bit overstating the instability of the runs. Yeah, there's definitely hyperparam hacking, but IMO - and I'm not a huge expert - the big difference comes from network architecture hacking rather than version of python, random seed, etc. Also I think you're mostly talking about papers/work where someone tries to squeeze out couple % gain on a widely studied problem where tons of methods have been applied in the past. I'd be more inclined to agree with you if these guys were at 53% accuracy in single or couple tests. But with the number of results in 70% range, I think there's something else going on. But since I don't know what it is, your argument might be still weightier than mine. ;)

I dont know what the authors did but ill reiterate from before vanilla LSTMs do little better than guess on the stock market. They probably had like 1000 GPU and tested thousands of hyperparameter configurations and "overfit" the test set. This is why typically papers like this are not believed anymore in the ML literature. Try adding some stuff like attention or skip connections and whatever else is hot now (I'm not sure) and didnt someone recommend GRUs instead. I have some other ideas you can use like Gaussian Processes to estimate realtime covariance matrices, but your better off looking at the literature first than trying out hairbrained ideas that might not work. It's really not a trivial excerse to outperform the market with ML. Ah, I think I see where there is a miscommunication between us. :) My goal is not to outperform market with ML. My goal is to understand whether what is proposed in this paper works and if it does not then why. 8) You are possibly completely right that what authors propose does not work. I just want to understand how they got the results they got. You've said "probably had like 1000 GPU and tested thousands of hyperparameter configurations and "overfit" the test set." before. I don't think that's the case at all. If you read the paper - which you haven't so far - you can see that their training is really simple and there's no "thousands of hyperparameter configurations". Which is baffling in itself. I have some suspicions of what could be wrong, but it's not productive to discuss it if you just dismiss the paper offhand. Which is BTW your prerogative - if that's where you stand, that's fine and I won't bother you with this further. 8)

The paper I was talking about is "Dow Jones Trading with Deep Learning: The Unreasonable Effectiveness of Recurrent Neural Networks" to be presented at http://insticc.org/node/TechnicalProgram/data/presentationDetails/69221 The paper is not publicly available, but you can ask the authors for copy. I have a copy and can send it to people interested, but I won't post it here publicly. PM me if you want a copy. Couple comments on various things previously mentioned now that the paper is semi-public: - The paper predicts daily close of DJIA from daily open value + opens of previous n days (2-10). - The trading algorithm is simply buy if predicted close > open and sell otherwise. If you cannot buy (already have position), then hold. If you cannot sell (already hold cash), then hold cash. - Authors use training data from 01/01/2000-06/30/2009 and test data from 07/01/2009 and 12/31/2017. This somewhat answers the critique that training is from bull market: it's not. Testing is not completely from bull market either. - Authors use pretty much vanilla LSTM, so IMO the critique that "1000s of academics looking for signals and the winners publish a paper" or that they have tweaked/over-fitted the model until it worked does not seem to apply. (It's possible that they messed up somehow and used testing data in training, but they seem to be careful, so it doesn't seem very likely). This is really vanilla IMO without much tweaking at all. Which makes the results surprising BTW. - I have some other comments, but I'd rather discuss this further with people who have read the paper, so I won't post them now. 8) As I mentioned, I spent a bunch of time reimplementing what these guys presumably implemented. I do not get their results. My results are pretty much at level of random guessing, i.e. the accuracy is around 48-52% while they get up to 80% accuracy. It's quite possible I am not doing something the same way they did. It's possible that their implementation or testing is messed up somehow too. But it's hard to prove that. Maybe they'll opensource their implementation sometime in the future. 8) If anyone is interested to get together (online through some tools?) and go through the paper and/or my implementation, we can do it. PM me and we'll try to figure out what would work best. 8)

IMO it really depends on the person and the circumstances. There are people who do just fine by starting their own shop without working elsewhere. There are people who do fine starting their shop after working in corporate not-legendary offices. There are people who do great after working for legends. There are also people who don't do well in exact the same scenarios. I don't want to psychoanalize - so please forgive me - but it looks like you're looking for others to say "yeah, you have to work for legend first". Yeah, sure working for legend first would open a lot of doors. But you have to realize that percentage-wise the number of people who get to work for legends is very low. So it's up to you to try to get a position with legendary investor, but likely you won't. There's only one Tracy Britt Cool in the last 30 years+ (Todd and Ted don't count - they went to work for Buffett after being successful investors not before). So IMO you're severely handicapping yourself if you think that working for a legend is the only way to get a good career/results/etc. Edit: regarding spotting-future-Klarman - this is as hard as spotting-future-Klarman to invest with them. I think people already suggested to you to go to BRK/Fairfax/DJCO/BOMN/SYTE/whatever annual investor meetings and network with CoBF/etc/investor/hedge crowd there. This might give you opportunities. You might even corner Whitney Tilson or Mohnish Pabrai in one of these. Though I still think that discovering future Klarman ... and clicking with them enough so that they'll hire you ... is not going to be easy. Good luck.

@racemize: Do you know/have you met Brian? Any stories/anecdotes?

??? ::) ... Actually, don't.

Yeah, good article.

In Soviet Russia investing legends notice you.

I wonder if the buyback announcement is related to Gates foundation. Perhaps BRK wants to buyback blocks that Gates foundation plans to sell.

Good talk.

https://www.wired.com/story/alphabet-google-x-innovation-loon-wing-graduation/

As expected perhaps: https://www.ccn.com/gamers-relief-bitcoin-bear-period-is-bringing-down-high-end-gpu-prices/ I did not doublecheck the prices claimed in the article.