Bridging the gap between Markowitz planning and deep reinforcement\n learning

Abstract

While researchers in the asset management industry have mostly focused on\ntechniques based on financial and risk planning techniques like Markowitz\nefficient frontier, minimum variance, maximum diversification or equal risk\nparity, in parallel, another community in machine learning has started working\non reinforcement learning and more particularly deep reinforcement learning to\nsolve other decision making problems for challenging task like autonomous\ndriving, robot learning, and on a more conceptual side games solving like Go.\nThis paper aims to bridge the gap between these two approaches by showing Deep\nReinforcement Learning (DRL) techniques can shed new lights on portfolio\nallocation thanks to a more general optimization setting that casts portfolio\nallocation as an optimal control problem that is not just a one-step\noptimization, but rather a continuous control optimization with a delayed\nreward. The advantages are numerous: (i) DRL maps directly market conditions to\nactions by design and hence should adapt to changing environment, (ii) DRL does\nnot rely on any traditional financial risk assumptions like that risk is\nrepresented by variance, (iii) DRL can incorporate additional data and be a\nmulti inputs method as opposed to more traditional optimization methods. We\npresent on an experiment some encouraging results using convolution networks.\n