Hello, the reason I'm opening this discussion is that for a while I've been thinking about building forex/stock trading automation system(something like a bot but in a more sophisticated way). I have a decent python experience with over 1 year of hands-on deep learning and general purpose projects so it is valid to say I'm quite experienced with tensorflow 2.x, keras, numpy, pandas, matplotlib and most of standard python libraries. I'm currently learning c++ for performance needs that cannot be met with python alone. I've been intending to use my background to create a trading system I have not thought of specifics but my general overview of a problem solution might be to create a combination of deep learning models and maybe some reinforcement learning techniques as well(which something I'm currently learning). The question is do you think of a criteria to get best outcomes in terms of prediction accuracy and be able to deliver something that might turn profitable at some point? I have not started working on the project yet but when I read about the topic in kaggle and general machine learning forums, the idea of forex bots/ auto-trading systems is sometimes met with skepticism: some argue it's nearly impossible to create something useful and some others claim that they were successful to create deep learning models (my best guess is RNN/LSTM architectures) with variable accuracy(60-90%). I don't think experimentation will hurt, I mean I can start working something out and figure out the best results for myself but I thought it's a good idea to ask for guidance from those who have similar/better background/experience as/than myself as well as others who might have tried already and hear the feedback first. What do you think?
The idea sounds like a function extrapolation (if I understand it correctly).I would recommend to start with a proper definition of input parameters and output predictions format. The input parameters can be exchange rate history (for example, a 5-years window), Dow Jones index and other external factors (in fact, I think the more is the better in this case). Output can be the exchange rate, which can then be reused to calculate the next value.
Once you've defined it, you need to prepare a dataset. Your Python knowledge can be helpful to create some sort of web-crawler and download all the data you need.
I don't know anything about trading, but it seems like it is indeed impossible to predict all the external factors that affect the prices/rates (the same applies to weather forecast).
View all questions in Machine Learning forum