Built a discrete-time market environment in Python where buyer and seller agents independently learn pricing strategies via Q-learning (tabular RL), modelling reward as profit after transaction costs.
Studied emergent macro-level behaviour — bid-ask spread convergence, impact of transaction cost magnitude on equilibrium price, and policy divergence under asymmetric information — across 10,000+ episode runs.
Structured the environment to mirror OpenAI Gym conventions (step / reset / render) for extensibility; visualized agent reward curves, Q-value heatmaps, and price trajectories with Matplotlib.
Tech: Python, NumPy, Matplotlib