Deep Reinforcement Learning (DRL) for Portfolio Allocation

Deep reinforcement learning (DRL) has reached an unprecedent level on complex tasks like game solving (Go [], StarCraft II []), and autonomous driving. However, applications to real financial assets are still largely unexplored it remains open question whether DRL can reach super human level. In this demo, we showcase state-of-the-art methods for selecting portfolios according environment, with a final network concatenating three individual networks using layers of convolutions reduce network’s complexity. The multi entries our enables capturing dependencies from common indicators features risk aversion, citigroup index surprise, portfolio specific previous allocations. Results test set show approach overperform traditional optimization results available at demo website.