Conference Publications

Hou†, Xiaoxuan., Yuan†, J. Y., Leibo, J. Z. & Jaques, N. InvestESG: A Multi-Agent Reinforcement Learning Benchmark for Studying Climate Investment as a Social Dilemma Accepted at ICLR 2025

• Poster presentation at NeurIPS 2024 Workshop on Tackling Climate Change with Machine Learning

• A multi-agent reinforcement learning (RL) environment using PyTorch and JaxMARL to simulate how companies and investors (modeled as individual RL agents) respond to ESG mandates under various scenarios. The framework supports large-scale simulations with over 50 agents spanning 100 years, capturing complex climate and economic dynamics that are challenging to represent in traditional analytical models. 

Journal Papers Under Review

Hou, Xiaoxuan. & Chen, S. Collective Renewable Power Purchase Agreement: Negotiation and Risk-Sharing Mechanisms. Reject & Resubmit at Manufacturing & Service Operations Management

• Presented at INFORMS 2023, 2024

• Analytical and economic modeling to explore how corporate buyers can effectively negotiate collective long-term power purchase agreements and analyze the differences in equilibrium outcomes across various negotiation mechanisms. 

• Hierarchical optimization with game-theoretical modeling and generalized Nash bargaining of the negotiation process between multiple entities.

Hou, Xiaoxuan. & Yu, Q. Explaining Worker Schedule Quality Disparities: Evidence from an American Restaurant Chain. Under Review at Management Science

• Presented at MSOM 2024, Wharton Empirical OM Workshop 2024, POMS 2024

• Empirical research using reduced-form regression and causal inference methods to uncover gender and racial disparities in the schedule quality of restaurant workers. 

Hou, Xiaoxuan. & Wagner, M. Leveraging Embedded Time Series Forecasting in Dynamic Programming for Storage Optimization. Working paper; In Preparation for Submission to Manufacturing & Service Operations Management

• Theoretical proof for optimal storage operation policies based on time series price evolution, accompanied by the design of an efficient dynamic programming algorithm that leverages critical features of time series models for practical implementation. 

Conference Papers Under Review

Liu†, K., Hou†, Xiaoxuan, Yuan, J., Jaques, N. (2025). Evaluating LLM Agents as Human Simulators in Climate Social Dilemmas. Under Review at AAAI 2026

• First LLM-based agent framework for climate finance, extending InvestESG to compare LLMs with RL baselines, humans, and a social planner.

Work In Progress

Wang, P., Chen, Shi., Hou, Xiaoxuan, (2025). A Dynamic Multi-Agent Framework for Collective ESG Investment: Mitigation, Greenwashing, and Strategic Trade-offs.

• As a follow up to Hou, Yuan, et al. 2024 and Liu, Hou, et al. 2025, developing an analytical modeling framework for climate finance with a focus on greenwashing, complementing InvestESG studies.

• Providing benchmarks and comparisons against both RL- and LLM-based simulation approaches.

Hou, Xiaoxuan, Yu, Q. (2025). AI vs. Human Managers: Gender Disparities in Service Scheduling and Negotiation.

• As a follow up to Hou and Yu 2024, evaluating whether “AI managers” reduce or amplify gender disparities in scheduling relative to human managers using careful calibration against real scheduling data and LLM agent-based tool-calling for schedule optimization and feasibility confirmation.

• Testing negotiation dynamics, including female assertiveness and framing effects when interacting with AI vs. human managers.