How can you improve the prediction of asset returns with Machine Learning ?

[jdessain]

One method you can improve the prediction of asset returns with Machine Learning / Deep Learning is by using a custom loss function that echoes the objective of the algorithm !
All errors in prediction do not lead to error in investment... some lead to wrong investment decision, others have no financial consequence.
Translating this asymmetry in the algorithm improves significantly the quality of the algorithm and its efficiency once tested live.

To know more about the WHY and HOW, read the latest research paper https://ssrn.com/abstract=3973086.
Please share your comments and suggestions !

 

[PepeQuant]

you build a time machine. travel to future and get the data and fit your model with that

ps if you do build that please let me know I will invest all my life saving to your fund

 

[Daniel Duffy]

you build a time machine. travel to future and get the data and fit your model with that

ps if you do build that please let me know I will invest all my life saving to your fund

Count me in! I can give an advanced C++ course on the way there to keep people sharp. We need a good internet connection with the TAs.

 

[jdessain]

you build a time machine. travel to future and get the data and fit your model with that

ps if you do build that please let me know I will invest all my life saving to your fund

Just read the paper... you'll understand. with AI, you can avoir some downturns and marginally beat a buy & hold strategy.
Data and code for loss function on ssrn