New Machine Learning methods to implement

[MLQNineteen]

Hi everyone,
New to this forum. Looking forward to hearing everyone's insights
I'm looking into novel methods in portfolio management using Machine Learning for a commodities portfolio as part of a project. I have looked into some of the more classical techniques, but want to explore new methods using latest research. I found some interesting papers which employ deep reinforcement learning, but so far it seems non-applicable in practice due to transaction costs.
Does anyone have any pointers into what I could start to look at?
Thanks

 

[felixenzo]

Hi @MLQNineteen,

I'm also looking for novel methods in portfolio management. I have found and implemented a few papers regard this. For example, this one, "A Deep Reinforcement Learning Framework for the
Financial Portfolio Management Problem," where is used a convolutional neural network for portfolio management. Actualy I found it in this Coursera's course "Overview of Advanced Methods of Reinforcement Learning in Finance". This paper is focused in Cryptocurrencies. I proved it in two platforms, Catalyst and QuantConnect. But whit not good results. In QuantConnect I think the problem was insufficent training, because when I try to train the model in a big training data set, the loop is longer than 10 minutes, which trigger an error in that platform.

Another aproach for portfolio managment using Machine Learning is this: "Market Self-Learning of Signals, Impact and Optimal Trading:
Invisible Hand Inference with Free Energy", by Igor Halpering, how is the professor for this course of Coursera: "Reinforcement Learning in Finance", where he explaind the method, that in summary use Inverse Reinforcement learning for portfolio management. Maybe in next weeks I'll try to programm it.

But, I would like to get some guide and to hear from others which algoritms have proved to be useful.

Greetings.

 

[wchurch]

I have used long-short term memory (LSTM) for volatility forecasting project before, have you used LSTM before or it's something that you think you can add in your portfolio?