Financial Bandits - Development of Thompson Sampling for Financial Data

Reinforcement learning is an emerging branch of machine learning that has typically been used for applications of autonomous vehicles. However, the framework provides a natural fit for financial applications due to the agent-environment interaction that is an investor and the market. In this paper, we develop a new class of bandit algorithms dubbed financial bandits which expand on standard bandit algorithms that are the foundation of reinforcement learning. Of particular focus are Bayesian bandits and Thompson sampling. We loosen assumptions about resources and adopting non-parametric models, we create a more appropriate class of bandit algorithms for applications to financial time series.