Deep Deterministic Policy Gradient for Portfolio Optimization

Summary

This work tests whether deep reinforcement learning can recover optimal trading strategies in environments that are simple to describe but mathematically challenging. The authors focus on deep deterministic policy gradient, a method that learns a policy for choosing actions through interaction with an environment. They select settings where the optimal strategy, or a close approximation, is already known, allowing the learned behavior and reward to be compared with a reference solution.

The reported result is that the agent recovers essential features of the known strategies and obtains rewards close to optimal. This provides evidence that the algorithm can act as a solver in the tested environments. The excerpt does not identify the specific market models, portfolio constraints, training procedures, or evaluation metrics. Its conclusion is therefore limited to the chosen controlled settings and does not establish performance in noisy live markets, with transaction costs, or where the optimal policy is unknown.

Key ideas

• The study evaluates deep reinforcement learning as a solver for trading strategies.
• It uses deep deterministic policy gradient in mathematically nontrivial but conceptually simple environments.
• The tested environments have known or nearly known optimal strategies for comparison.
• The reported agent recovers key strategy features and achieves near-optimal rewards.
• The excerpt does not establish performance under live market frictions or unknown optimal policies.

Tags

# Deep Deterministic Portfolio Optimization


# Deep Deterministic Portfolio Optimization









Can deep reinforcement learning algorithms be exploited as solvers for optimal trading strategies? The aim of this work is to test reinforcement learning algorithms on conceptually simple, but mathematically non-trivial, trading environments. The environments are chosen such that an optimal or close-to-optimal trading strategy is known. We study the deep deterministic policy gradient algorithm and show that such a reinforcement learning agent can successfully recover the essential features of the optimal trading strategies and achieve close-to-optimal rewards.

Shown in full with attribution under the source's licence. Licence: abstract CC0

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.