Estimating Carbon Intensity Using Text-Based Indicators: A Machine Learning Approach for Endowment Managers and Private Investment Portfolios

Abstract

Climate change poses a critical threat to global political and economic stability. Addressing this challenge requires deeper engagement from institutional investors, whose capital allocation decisions can significantly influence the emissions trajectories of industries. This project investigates innovative approaches to improving greenhouse gas (GHG) emissions transparency in institutional portfolios, with a particular emphasis on private assets. Using Duke Management Company (DUMAC) as a case study, it examines how machine learning tools can support more climate-conscious endowment management. Specifically, the research applies term frequency (TF) and sentiment analysis to assess how variables derived from these techniques correlate with emissions intensity across selected sectors, with the aim of developing a methodology for estimating the GHG profiles of DUMAC’s less transparent private holdings. Ultimately, this project seeks to equip institutional investors with stronger tools for integrating climate risk into financial decision-making.