Robust Risk-Aware Reinforcement Learning

We present a reinforcement learning (RL) approach for robust optimization of risk-aware performance criteria. To allow agents to express wide variety risk-reward profiles, we assess the value policy using rank dependent expected utility (RDEU). RDEU allows seek gains, while simultaneously protecting themselves against downside risk. robustify optimal policies model uncertainty, not by its distribution but rather worst possible that lies within Wasserstein ball around it. Thus, our problem formulation may be viewed as an actor/agent choosing (the outer problem) and adversary then acting worsen strategy inner problem). develop explicit gradient formulae problems show their efficacy on three prototypical financial problems: portfolio allocation, benchmark optimization, statistical arbitrage.