Risk-Aware Multi-Armed Bandit Problem with Application to Portfolio Selection

摘要: Sequential portfolio selection has attracted increasing interests in the machine learning and quantitative finance communities recent years. As a mathematical framework for reinforcement policies, stochastic multi-armed bandit problem addresses primary difficulty sequential decision making under uncertainty, namely exploration versus exploitation dilemma, therefore provides natural connection to selection. In this paper, we incorporate risk-awareness into classic setting introduce an algorithm construct portfolio. Through filtering assets based on topological structure of financial market combining optimal policy with minimization coherent risk measure, achieve balance between return.