State–Reward Alignment and Rich State Representations in Financial Reinforcement Learning

Independent research investigating the impact of state representation richness and reward-function design on reinforcement-learning-based trading agents. The study compares traditional Raw OHLC inputs with a custom 89-dimensional engineered feature representation across multiple reward formulations, random seeds, and realistic trading environments, introducing the concept of State–Reward Alignment and demonstrating the importance of jointly designing state representations and reward functions for financial reinforcement learning.