Quantitative trading through automated systems has been vastly growing in recent years. The advancement in machine learning algorithms has pushed that growth even further, where their capability in extracting high-level patterns within financial markets data is evident. Nonetheless, trading with supervised machine learning can be challenging since the system learns to predict the price to minimize the error rather than optimize a financial performance measure. Reinforcement Learning (RL), a machine learning paradigm that intersects with optimal control theory, could bridge that divide since it is a goal-oriented learning system that could perform the two main trading steps, market analysis and making decisions to optimize a financial measure, without explicitly predicting the future price movement. This survey reviews quantitative trading under the different main RL methods. We first begin by describing the trading process and how it suits the RL framework, and we briefly discuss the historical aspect of RL inception. We then abundantly discuss RL preliminaries, including the Markov Decision Process elements and the main approaches of extracting optimal policies under the RL framework. After that, we review the literature of QT under both tabular and function approximation RL. Finally, we propose directions for future research predominantly driven by the still open challenges in implementing RL on QT applications.

We propose Conditional Value-at-Risk (CVaR) investment agents to solve the problems of single asset trading and assets allocation under the Direct Reinforcement Learning framework. We propose two convex CVaR-based agents, the CVaR-constrained and the unconstrained CVaR-sensitive. Convexity allows conveniently implementing incremental learning, leading to an adaptive investing agent at an efficient computational cost with the merit of guaranteed policy convergence. Our experiments with frictional investment under various markets reveal the CVaR-constrained potency in improving investment return per unit of risk. The unconstrained CVaR-sensitive agent, on the other hand, exhibits robustness in handling intense market pullbacks, with both CVaR-based agents showing superior risk management to a risk-insensitive one. Our approach also showed superiority over state-of-the-art methods, demonstrating the potential of CVaR-based RL investment models. We finally show how our agents are extendable to learn investing under the most general investment problem of optimizing a multi-asset portfolio.