Reinforcement Learning for Finance

Part III. Financial Applications

The third part of the book applies the algorithms and techniques introduced in the first two parts to classical financial problems:

• Chapter 6 applies deep Q-learning (DQL) to the algorithmic trading of a single financial instrument. It builds on the prediction game discussed in Chapter 3. The chapter uses Monte Carlo simulated data to train a financial Q-learning (FQL) agent called TradingAgent. The goal of the FQL agent is to maximize the profit from going long and short on a single financial instrument.

• Chapter 7 uses DQL to learn how to hedge, or rather replicate, a European call option in the seminal model by Black-Scholes-Merton (1973) for option pricing. The HedgingAgent is able to learn appropriate hedging strategies by working with market-observable data only. For example, the agent knows the current price of the underlying asset, the time to maturity, and the current option price.

• Chapter 8 applies reinforcement learning (RL) to three classical problems in investment management. The first problem is determining the optimal allocation between a risky asset and a risk-free asset, commonly referred to as two-fund separation. The second problem focuses on finding the optimal allocation between two risky, negatively correlated assets. The third problem extends this to the optimal allocation among three risky assets. The InvestingAgent developed in this chapter generates Sharpe ratios that consistently surpass those of individual risky ...