Machine Learning and Neural Networks

Discussion in 'Technical Analysis' started by HanSolo, May 2, 2019.

  1. HanSolo

    HanSolo

    I have been reading about neural networks and machine learning lately and was wandering if anyone is using neural networks for trading? Is it a waste of time? If not what is the best way to use them. I have seen some fantastic examples of how they can be used to predict an optimal signal etc. but can't find anyone who has integrated them successfully into a profitable trading strategy.
     
    Last edited: May 2, 2019
  2. You can't train a neural network on price data and expect something to come out of it.

    The way I think about it is: if a human can spot this pattern, I can train a machine to do this, I just need to provide the right information. For example, if you're looking at a bar chart and expect to train a machine on OHLC data, you've lost already.

    Implicit when a human looks at a bar chart:

    1. Height of body
    2. Wick zones
    3. Color of candle
    4. (maybe) news events, earnings, Trump tweets

    You need to provide your algorithm these things for it to even have a hope of capturing the pattern you want it to capture.

    TL;DR It's still hard.
     
  3. I am also curious if anyone is really successful using AI to successfully trade the market? lots of quants claimed they used AI only because it made them look more sexy.
     
  4. You make a lot of uninformed comments recently. Pissijg on Dalio first and now this. I know of several hedge funds from personal interaction that they trained deep learning networks on asset pricing data and derived models they profitably trade. I myself do so as well. If hft firms for years made a killing on making markets with fully automated algorithms then you basically claim the same can't be done using dlns? Please stop as a blind to lead the blind you obviously are way out of your league in this technical area.

    For starters, all the aspects you enumerated are implicitly factors the network is fully aware of through weighting neurons. Goodness. Sometimes I get a good laugh out of this site which is primarily the reason I am here.

     
  5. rb7

    rb7

    I first started to program and use Neural Networks and Genetic Algos 25 years ago. I was never able to come up with something useful with them. That doesn't mean that they can't be useful, it only means that I wasn't able to make them work in a manner that they would be used to make money.

    There is no silver bullet. You still need to understand the markets in order to train/program a NN. It won't do it for you by itself. There are so many variables that you need to configure in a NN, and so many architecture types and so many network types.
     
  6. Did I say deep learning networks? No. I don't know much about those, so please feel free to enlighten the OP instead of following me around.

    Also, if it were easy, everyone would do it. Therefore it is still hard. Feel free to post about what you do though, I'd love to learn.
     
  7. This is what you wrote:

    "You can't train a neural network on price data and expect something to come out of it"

    Hard to trade markets with such memory deficit. Neural networks are the super group that includes deep neural networks. DNN is the sub group.

     
  8. OK. Now, please enlighten the OP and myself what you do so we can learn from you. For me, I have been in neural networks since the 90s so when I hear neural networks, I think of that generation of work.
     
  9. Well as you can see everywhere around DNNs have completely transformed how machines build causal relationships. Ever wondered why your machine language translation got so much better over the past few years? Why your Google assistant understands almost every word you speak? Why your custom tailored news freed is so much better than before? So yes, past machine learning and neural networks are pretty much worthless nowadays and are replaced by better technology just as in every aspect of life.

    Are you seriously assuming that some linguistics PhDs and philosophy PhDs at Rentech just sat around and discretionary traded markets and made a killing? Has it ever occurred to you that an incredible amount of neutral network architecture was involved as well?

     
  10. Umm no. I didn't think that. But it's also not as simple as throw price data at the problem and you're done
     
    #10     May 2, 2019