Adaptive Portfolio Optimization with Multi-Agent Deep Reinforcement Learning

This repository contains the implementation of Adaptive Portfolio Optimization using Multi-Agent Deep Reinforcement Learning (DRL) with short-term performance analysis.

Overview

Traditional portfolio optimization often relies on static models or a single DRL agent, which fail to adapt to dynamic market conditions. Our approach introduces an adaptive multi-agent framework that:

• Trains multiple DRL agents (A2C, SAC, TD3, DDPG, PPO).
• Selects the best-performing agent based on short-term returns (10-day window).
• Dynamically allocates portfolio weights to improve robustness and adaptability.

Key Results

• Backtested on Dow Jones data.
• 11.43% average annual return.
• 38.29% cumulative return.
• 0.832 Sharpe ratio, outperforming individual DRL agents.

Publication

This work is published in the International Journal of Information and Communication Technology Research (IJICTR): http://ijict.itrc.ac.ir/article-1-738-en.html

Sharbaf Movassaghpour S, Kargar M, Bayani A, Assadzadeh A, Khakzadi A.
Adaptive Portfolio Optimization with Multi-Agent Deep Reinforcement Learning and Short-Term Performance Analysis.
IJICTR 2025; 17(3):58-69.

© 2025 Masoud Kargar