Unmasking climate risk in earnings calls: traversing storms and fire with a Taxonomy-GPT prompting approach

Rethinking physical risk exposure: from geographic mapping to firm disclosures

As climate change intensifies, businesses face increasing financial and operational risks from extreme weather events. While traditional risk assessments often focus on geographic mapping of assets, this approach overlooks broader supply chain disruptions, demand fluctuations, and indirect financial impacts. This research provides a novel framework for understanding corporate exposure to climate hazards by analyzing earnings call transcripts using advanced Natural Language Processing (NLP) techniques and Generative AI.

Extracting firm-specific exposure to hazards

The study investigates how global firms have discussed climate risks and opportunities in earnings calls over the past two decades (2001–2021). By combining machine learning with linguistic analysis, the researchers identify corporate references to four major climate hazards: storms, heatwaves, wildfires, and cold waves. Their methodology consists of two key steps.

First, they extract climate-related text from earnings call transcripts using a hazard-specific taxonomy informed by scientific sources (e.g., IPCC, NOAA), which is further refined with Word2Vec word embeddings.

Second, they apply Generative AI (GPT-3.5) to analyze these excerpts, extracting firm-level climate risk indicators and classifying each mention as either a risk (negative impact) or an opportunity (positive impact).

Additionally, the study distinguishes between direct impacts—such as infrastructure damage—and indirect impacts, which arise from supply chain disruptions or market fluctuations. This is the first research to differentiate corporate exposure to distinct physical climate risks while also identifying whether the impact occurs through direct or indirect channels.

Heterogeneous exposure across sectors and market reactions: implications for firms and investors

Their findings reveal that physical climate risks are highly sector-specific, with industries such as utilities, agriculture, and construction often experiencing both risks and opportunities. For example, firms in the energy sector may see increased demand during extreme cold events, while construction firms might benefit from rebuilding efforts after climate disasters. Crucially, the study finds a strong negative relationship between stock returns and climate risk disclosures in earnings calls, suggesting that financial markets incorporate climate-related risks into asset pricing. This effect persists across multiple climate hazards, reinforcing the importance of integrating climate considerations into corporate strategy and investment decisions.

This research contributes to the growing field of climate finance by providing a systematic and scalable approach to measuring climate risk through corporate disclosures. By leveraging AI-driven text analysis, the paper offers investors, policymakers, and analysts a new tool to assess how businesses navigate climate-related financial risks.

Full paper

Read the full paper: Unmasking Climate Risk in Earnings calls: Traversing Storms and Fire with a Taxonomy-GPT prompting approach

Authors

• Lorenzo Prosperi, Prometeia SpA and Visiting Fellow at the University of Edinburgh Business School
• Lea Zicchino, Prometeia and SAIS Europe
• Maria Paola Priola, Universita di Cagliari
• Annalisa Molino, Prometeia SpA
• Michele Cimino, Prometeia SpA