Greener on the other side?
Mapping China’s overseas co-financing and financial innovation
Download the report here.
About The Report
I’m pleased to announce the publication of a new paper with Yunnan Chen and ODI Global titled “Greener on the Other Side? Mapping China’s Overseas Co-financing and Financial Innovation.”
This paper examines whether China’s overseas lending has truly “gone green” and maps the networks of institutions financing this transition. We used a novel LLM-powered approch to do a first-of-its-kind comprehensive analysis of China’s green overseas lending portfolio, uncovering important patterns in sectoral distribution and financing modalities.
Key Policy Findings
Our analysis revealed several important insights about China’s overseas financing patterns:
- Of China’s $1.5 trillion in overseas development finance since 2000, only $86.5 billion (5.8%) went to “green” investments, primarily in hydropower (71.6%) and nuclear projects (12.0%)
- Despite rhetoric about a “Green Belt and Road” and declining coal investments, we haven’t observed a significant pivot toward renewable energy financing
- As policy bank infrastructure lending declines, co-financing through syndicated loans remains resilient ($180+ billion between 2013-2021)
- Our social network analysis revealed two distinct financing ecosystems with limited overlap: policy-driven lenders dominate green project co-financing, while commercial banks operate in separate networks
- Chinese commercial banks like ICBC occupy strategic “bridge” positions between these ecosystems but haven’t fully leveraged this potential to channel commercial capital toward green investments
Expanding Policy Research with AI - A Beginning
What makes this work exciting is how it demonstrates new possibilities for policy research through large language models:
LLMs allowed us to analyze nearly 18,000 project descriptions in just 15 hours at a cost of $1.58—work that would have required approximately 1,500 hours of human labor and $22,500 using traditional approaches (17,957 x 5 minutes per observation x $15 per hour).
This isn’t just about efficiency; it’s about making previously infeasible analyses possible for policy researchers.
However, we recognize that powerful tools require responsible use. Can we trust these results? To address this question, we developed a practical validation workflow comparing multiple models against human evaluations on a 300-project sample. Our chosen model achieved 91.8% agreement with human raters—a promising result.
We see this as just the beginning of a longer journey in developing best practices for AI-powered policy research. While we’re proud of what we’ve accomplished, we know there’s much more work to be done. Our hope is that by sharing our approach openly, others can build upon this foundation and advance the field further.
Raising the Bar on Transparency in Chinese Lending Research
Research on Chinese overseas lending is often politically charged, making transparency and reproducibility particularly valuable. We’ve taken meaningful steps to raise the bar in this field:
- Open Methods and Assumptions: We’ve documented our classification criteria, analytical processes, and limitations in a detailed methodological annex, exposing our work to scrutiny
- GitHub Repository: Our public repository includes the code used for analysis, enabling others to examine, test, or build upon our work
- Version Control and Package Management: Using GitHub for version control and renv in R for package management ensures others can see exactly which versions of each package we used
We recognize that reproducibility exists on a spectrum. With limited time and budget, we’ve taken several important steps—version control, documented code, pinned packages—but acknowledge there’s room for improvement, such as containerization or fully scripted data extractions. By making these aspects of our work transparent, we hope not only to add credibility to our specific findings but also to empower others to challenge, verify, or extend our research with data-driven approaches.
Resources
This research reflects my commitment to combining rigorous data science with domain expertise to create evidence-based insights on complex policy questions. I’m excited about the possibilities these approaches open up and look forward to continuing this work.
Citation Information
Citation
@report{chen2025,
author = {Chen, Yunnan and Emery, Teal},
publisher = {ODI Global},
title = {Greener on the Other Side?},
date = {2025-04-09},
url = {https://odi.org/en/publications/greener-on-the-other-side-mapping-chinas-overseas-co-financing-and-financial-innovation/},
langid = {en}
}