Machine designed strategies. Do they work?

Discussion in 'Strategy Building' started by v75z52, Mar 20, 2012.

  1. I am not sure it should be anchored to absolute numbers. I would argue that Zuckerberg should have done what he has done no matter whether he tried his hands at trading and even though had he made 8 figures in trading. The goal is to optimize talent and maximize efficiencies. I highly respect those who perused their talents and created innovative fairly priced products others can use, whether I myself would purchase such products or not.

    What I DO NOT appreciate is snake oil salesmen who tout 60k products that do nothing other than overfitting to historical time series.

    Anything needed to model time series and optimize whatever you want (you can overfit the hell out of any time series if thats what you feel inclined to do) can be had here:

    http://www.r-project.org/
    http://www.python.org/
    http://www.marketcetera.com/site/
    http://code.google.com/p/tradelink/
    among many others

    The truth why such people find an audience again and again is that some of the truly lazy among us human kind believe they can shell out cash in a trade off for hard work and knowledge/intelligence. No piece of software will make up for a lacking IQ or willingness to put in hard work. Bitter truth for the amateurs in this field: This job is one of the hardest in the world to make decent money (which is why you have people like the ones here in this thread selling their crap). I would argue most at ET are better off selling themselves as male prostitutes rather than BSing everyone they are "traders", conveniently omitting the qualifiers "losing" and "impoverished".

    I am not trying to make enemies here, I don't care some disagree or have their ego hurt by what I say. The ONLY REASON I put in couple minutes to write this is to WARN NEWBIES that THERE IS NO SHORTCUT IN LIFE, ESPECIALLY NOT IN TRADING. A piece of software that is priced at such ridiculous level targeting newcomers SHOULD BE PUNISHABLE BY LAW, its along the exact same lines than having a broker push CMOs or other structured products to a grandma who is dependent on her retirement allowances.

     
    #141     Apr 17, 2012
  2. gmst

    gmst

    I read bloomberg's book and as far as I remember he was a salesman at Saloman not a trader. Not 100% sure though. Its true that a Salesman is more likely to sell a software than a trader. Anyways this was just a correction won't really matter to the discussion. Carry on guys!
     
    #142     Apr 17, 2012
  3. +1

    Target market here is the lazy & intellectually challenged. Ideal customer being a not-so-bright, but cashed up widow or orphan - windfall wealth that is unrelated to the skill of the individual.

    "I'm really not a math person, but I recon I'd be a great trader if I could just find the right indicators."

    Those guys.
     
    #143     Apr 17, 2012
  4. It makes me chuckle but a second later stories of some poor souls come to mind who drag down their wife, their families, because they can't admit they are not cut out for trading but dump hundreds and thousands on seminars, charting software, indicator manuals, 60K(!!!) curve fitting software, and the like. Those people should be ashamed of themselves to drag others into their miserable life but they rather find excuses. How many stories have we read here at ET. I am hoping to sensitize some of the newbies here that there is no way to get around the learning curve and lots of efforts that need to be put into this business. The software under discussion nor any other machine learning tool should be in the tool arsenal of a beginner, and what pisses me massively off is that those sales people are targeting beginners here. Successfully perusing machine learning algorithms requires rigorous knowledge of statistics and probabilities on top of intricate knowledge of market dynamics which most newbies sorely lack. Targeting the unassuming and wishful is what pisses me off about those sales people.

     
    #144     Apr 17, 2012
  5. luisHK

    luisHK


    It seems you're mixing up TSL and PAL here. PAL has been commented by several long term ET users in this thread and others and does offer a free demo. Cost is much lower than TSL although I'm not sure about the usefulness of the software.
     
    #145     Apr 18, 2012
  6. First of all, I would suggest that you calm down. Yes, he worked at Salomon but he worked on his software with ML, a fact that you miss:

    "In June 1983, Bloomberg and his three core partners, Duncan MacMillan, Chuck Zegar, and Tom Secunda, demonstrated their prototype machine for Merrill Lynch, which had specified that the device perform complicated calculations on government bonds as well as conventional pricing computations of Merrill Lynch's inventory. "

    http://www.fundinguniverse.com/company-histories/Bloomberg-LP-Company-History.html

    Is this why ML kept this in-house for 5 years? I do not think that you have any clue or idea of what you are talking about or that you have ever used such systems.

    No caps please. No need for shouting. Yes, instead of saying that he developed the terminal with ML I said he worked for ML. It is probably the same thing as they were the boss. You are picking up on minute details to build a huge argument.

    I agree with you on this one. You are probably referring to TSL and the fact that someone registered here and posted vague results that couldn't be reproduced by anyone except him. I more than agree with you. There are other cheaper programs that offer free demos and reproducible results. There are some other programs based on genetic programming but you cannot easily reproduce the results because of the inherent randomness of the process. They offer free demo too.

    You should also think of the possibility that something that does not work for you maybe works for someone else because he sees something or knows something that you do not. You should abandon dogma if you want to succeed in life.
     
    #146     Apr 18, 2012
  7. I did not compare different applications. My comment was geared towards sales people targeting clients for which such products are ill suited.

    Machine learning algorithms should be free of charge, as I said a little work on your end and you can pick up whatever you need through loading different freely available R packages.

     
    #147     Apr 18, 2012
  8. he never worked with ML. You really should do a bit more diligent homework. ML initially funded the venture with 30% and later on sold its stake in two tranches to BBG. Plus ML was one of the first USERS of the terminals, nothing more nothing less. Please get your facts straight.

    I worked for over a decade in the industry, among those in the past 6 years as prop trader. I am using BBG on a daily basis. I countered with facts and you make assumptions thats the difference.

    "Something that does not work for me may work for others"? This is the typical snake oil salesmen line. Give me a break, please. Machine learning algorithms were hyped in the hedge fund world for a very short while in the same way than neural networks were hyped and nobody uses them nowadays. Time to throw the garbage at unassuming retail clients, preferably newbies and hopefuls. Enough said!!!

     
    #148     Apr 18, 2012
  9. ssrrkk

    ssrrkk

    Depending on how you look at it, I think capitalizing on people's "laziness" is not immoral. If your software tool does exactly what you advertise it to do (again I am NOT talking about actual trading algorithms like PAL, just tools for enabling your own analysis and design), then I think it's a good business model. For example, I can provide a GUI-based tool that packages some of the useful analyses that are possible in R but requires no programming knowledge. This is arguably capitalizing on the user's "laziness" that prevents him or her from learning R and programming -- of course the decision to not spend time learning programming could be a strategic one, i.e., "I would rather spend time learning the markets than learning to code", etc.). Regardless, I think such software will provide a useful service as long as it does what it promises (the analysis and nothing more).

    I agree with you regarding selling a curve-fitting system -- I think it would not be moral to give the impression that it will likely find you a profitable algorithm. Perhaps it needs a black box warming like "Warning: this software is very likely to find trading algorithms that are curve-fit and statistically insignificant. This means that what appears to be a profitable system in the past will very likely be unprofitable in the future. Use at your own risk"...


     
    #149     Apr 18, 2012
  10. That right there would be my answer to you. To not re-post my earlier posts i'll say i agree that TSL is being marketed to newbies and that some poor souls end up with a product they don't know how to use - because they lack the knowledge of statistics and a lot of other skills. However, with that knowledge, the fact that 80% of systems they find truly are curvefits (and i believe that), the other 20% that work out-of-sample on a statistically significnt portion of the data may just be an underlying market mechanism that was discovered. It all comes down to statistics, since you can claim there is still 0.05% chance that the system simply got lucky after 1000 good trades OOS, but that's a much lower risk compared to many others taken by traders on a regular basis. Of course, the statistical analysis is probably a major part of the game with such software(and should be undertaken very very seriously ), but the fact that the systems are composed of price derivations (indicators) and patterns which makes it an open white-box, allows you to see if it did in fact catch onto some mean-reversion logic or trend confirmation logic. These things are obvious to see, and a curve-fit system will often use an oscillator in the inverse way or do something that's obviously stupid and you'll know it's a curve-fit even before looking at OOS data.

    I also agree that the way TSL is being sold is a minor crime, since the demos and the data presented on the site solely focus on the successul systems (the 20% or even 5% if you want to be a pessimist) that work OOS. Further, TSL has an erroneous way of counting filled limit/stop orders, no slippage, etc, further emphasising "good systems" that aren't really that good. On the newest EVORUN demos you can even see that the majority of systems in the OOS panel are squarely at around 0 net profit, going up and down around that value (if slippage and commission were used, they would mostly be negative). This is expected of curve-fitted and random systems and indicates a very low ratio of "valid" systems. When marketing TSL they don't tell you that. But what ultimately doesn't make TSL a complete fraud is the fact that with good analysis, in theory, you can find systems that have true underlying market dynamics in them. Considering it evolves 100s of systems per second, you're bound to find something after a few weeks of running the data. You can't deny this theory is valid.

    As for the poster "milewski05" i got in touch with him via e-mail and checked him out. He's a real person and a real TSL user, i know that for sure.
     
    #150     Apr 18, 2012