AGENT-BASED MODELING WITH REINFORCEMENT LEARNING FOR CARBON MARKET DESIGN IN UZBEKISTAN’S INDUSTRIAL SECTOR: A THEORETICAL FRAMEWORK

Abstract

Uzbekistan’s updated Nationally Determined Contribution commits to a 35% reduction in greenhouse gas intensity per unit of GDP by 2030 relative to 2010, yet the country lacks a carbon pricing mechanism to drive this transition. This paper proposes a theoretical framework integrating agent-based modeling (ABM) with reinforcement learning (RL) to simulate and optimize the design of a cap-and-trade carbon market for Uzbekistan’s emissions-intensive industrial sector. The framework addresses a critical research gap: the absence of computational models capturing strategic behavior of few, large, state-affiliated industrial emitters under carbon pricing in developing and transition economies. We define two agent classes-a Government Regulator and heterogeneous Industrial Enterprises-and formulate their decision problems as Markov Decision Processes with mixed continuous-discrete action spaces, where firms simultaneously choose green technology investment proportions (continuous) and net quota trading volumes (discrete). Using mechanism design principles, we derive a payoff matrix for regulator–industry interactions under strict versus loose cap scenarios, mathematically derive the equilibrium carbon price threshold at which firms transition from quota trading to green technology investment, and analyze two initial allocation regimes: grandfathering and auctioning. Distributional analysis highlights acute risks for emissions-intensive trade-exposed sectors and monotown labor markets in cities such as Almalyk, Navoi, and Angren. The paper concludes that Uzbekistan’s oligopolistic industrial structure necessitates hybrid allocation mechanisms with output-based benchmarking for trade-exposed sectors, and proposes computational implementation of the ABM-RL model as a critical next step for empirical parameterization.