Cournot–Bertrand Hybrid Competition Model with Machine Learning Market Prediction

Source abstract

This study proposes a simulation-based framework for analyzing hybrid Cournot–Bertrand competition under dynamic market conditions with market-clearing sales and inventory accumulation. Unlike traditional formulations, the demand side is price-driven, with realized sales determined endogenously as the minimum of produced quantity and market demand. This accounts for unsold stock, inventory dynamics, and bankruptcy-like regimes resulting from loss-dominated trajectories. Firms operate under heterogeneous strategic modes, where a Cournot-type firm adapts production while prices adjust via market feedback, and a Bertrand-type firm adapts price while production adjusts operationally. To accelerate large-scale exploration across parameter spaces, a machine learning classifier is trained on trajectory-based features (profit statistics, inventory indicators, spectral descriptors). The resulting neural surrogate enables rapid regime identification (equilibrium, cyclic, chaotic, and bankruptcy) and risk screening without repeated long-run integration, supporting interpretable analysis of competitive behavior in energy and financial markets.