Using machine learning to Improve systematic trading

[Rogerahan]

I have attached a document I wrote on how to improve systematic trading using machine learning. I personally found some interesting results.

Hopefully people find this useful. I am also curious as to what your thoughts are, and whether you have applied machine learning to your own trading methodology.

 

[Baron]

Thanks so much for sharing that Roger!

 

[BigTommy]

Thanks for sharing! This is interesting.

 

[fordewind]

ML doesn`t help as it curve fits stops to the past optimum, which aren`t relevant, unless you can accurately predict the upcoming volatility.But don`t know any one who`s managed to do that as yet.

 

[Rogerahan]

Thanks so much for sharing that Roger!

You're welcome Baron. Thanks for the feedback.

 

[Rogerahan]

Thanks for sharing! This is interesting.

You're welcome, and I am really glad to hear you found it interesting!

 

[Rogerahan]

ML doesn`t help as it curve fits stops to the past optimum, which aren`t relevant, unless you can accurately predict the upcoming volatility.But don`t know any one who`s managed to do that as yet.

I found in my research ML really did help, at least in terms of a smaller a drawdown, greater trade expectancy, higher win percentage etc. However, I applied ML to already defined systematic strategy in a way which doesn't include optimizing stops or anything like that.

I also made sure, as per best practice, to split my data into a training set, a cross validation set, and a test set for the final statistical analysis. Additionally, I made sure to address how to check results via a 'learning curve function' for variance (over-fitting) or bias in the model.

Also, I state in the paper that one of my justifications for using Logistic regression as my ML algorithm of choice is its in built assumptions of linearity. Which is kind off a safety net against over-fitting. I also included the code for regularised logistic regression just in case someone using the document did find inherent variance in their model.

 

[fordewind]

I found in my research ML really did help, at least in terms of a smaller a drawdown, greater trade expectancy, higher win percentage etc. However, I applied ML to already defined systematic strategy in a way which doesn't include optimizing stops or anything like that.

I also made sure, as per best practice, to split my data into a training set, a cross validation set, and a test set for the final statistical analysis. Additionally, I made sure to address how to check results via a 'learning curve function' for variance (over-fitting) or bias in the model.

Also, I state in the paper that one of my justifications for using Logistic regression as my ML algorithm of choice is its in built assumptions of linearity. Which is kind off a safety net against over-fitting. I also included the code for regularised logistic regression just in case someone using the document did find inherent variance in their model.

What is your DD, if i may?Over what period?

 

[traider]

are you currently a student?

 

[Rogerahan]

are you currently a student?

No. I know I wrote the document like a thesis, Its just that I wanted it to be be decently thorough and well referenced.