This is the code used in the research work I did for my master thesis with the title "Deep Learning Methods for Portfolio Optimization".
Given a dataframe of assets prices we first generate the optimal portfolio allocation thought Markowitz theory, then we train a Temporal Convolutional Network targeting this optimal allocation.
This allows the user to have flexibility in the return-risk trade off wanted. We now show some performance of the allocation generated by the network in the first semester 2020.
And the allocation generated from which the previous portfolio performance is computed: