A Generative Probabilistic Approach for Goal-Based Portfolio Optimization

Goal-based portfolio optimization seeks to design investment strategies that maximize the likelihood of achieving specific financial objectives. A major challenge in this domain is data scarcity and non-stationary market dynamics, which undermine the effectiveness of traditional approaches. To address this, we propose a generative modeling framework that integrates probabilistic regression with deep reinforcement learning. The probabilistic model estimates evolving market return distributions for state representation and generates realistic synthetic market trajectories, enabling the agent to train efficiently on diverse market scenarios and adapt to dynamic environments. Experiments on multi-asset historical data demonstrate that our approach achieves superior goal-attainment probabilities compared to established benchmarks, highlighting the value of synthetic market generation for robust goal-based portfolio optimization.