A Deep Learning Approach to Goal-Based Portfolio Optimization in Non-Stationary Environments

Goal-based portfolio optimization is a portfolio design technique that tailors investment strategies to an investor’s specific financial objective. Traditional approaches to this paradigm often assume that market dynamics are stationary, meaning that factors such as mean returns, volatility, and asset correlations remain constant over time. In reality, market dynamics are often non-stationary, posing challenges to traditional methods. In this paper, we present a deep reinforcement learning framework that adapts to evolving market conditions, enabling more robust investment strategies. Additionally, we leverage deep probabilistic regression techniques for data generation and market state estimation, ensuring the model accurately reflects changing financial environments. The presented approach offers a flexible solution for goal-based investing in dynamic markets and outperforms benchmarks on historical multi-asset data.