Experimental

RL-Driven Portfolio Rebalancer

Reinforcement-learning agent that learns to rebalance a multi-asset portfolio under transaction-cost and risk constraints.

About this project

Existing rebalancing rules were static and ignored regime changes; the team wanted to explore whether RL could adapt continuously.

Solution

PPO agent trained on simulated and historical market data with a transaction-cost-aware reward; backtested against rule-based baselines.

Technology

  • Python
  • PyTorch
  • Stable-Baselines3
  • Pandas
  • Optuna

Impact

Out-of-sample Sharpe +0.18 versus rule-based benchmark; experimental — not deployed to client capital.