Actors, Frames and Arguments: A Multi-Decade Computational Analysis of Climate Discourse in Financial News using Large Language Models

Ruiran Su, Janet B. Pierrehumbert, Markus Leippold

Abstract: Financial news media shapes trillion-dollar climate investment decisions, yet discourse in this elite domain remains underexplored. We analyze two decades of climate-related articles (2000-2023) from Dow Jones Newswire using an Actor-Frame-Argument (AFA) pipeline that extracts who speaks, how issues are framed, and which arguments are deployed. We validate extractions against 2,000 human-annotated articles using a Decompositional Verification Framework that evaluates completeness, faithfulness, coherence, and relevance. Our longitudinal analysis uncovers a structural transformation: pre-2015 coverage emphasized risk and regulatory burden; post-Paris Agreement, discourse shifted toward economic opportunity and innovation, with financial institutions becoming dominant voices. Methodologically, we provide a replicable paradigm for longitudinal media analysis with LLMs; substantively, we reveal how financial elites have internalized and reframed the climate crisis across two decades.

Introduction

Financial news is more than just reporting; it is an engine that drives global capital allocation. How the "financial elite" frame climate change—whether as a cost to be minimized or a market to be captured—directly influences trillions of dollars in investment. Despite its significance, analyzing this discourse at scale has historically been difficult. Traditional computational methods like topic modeling or dictionary-based keyword counting are often too blunt to capture the nuances of who is speaking, how they are framing the issue, or the specific arguments used to justify their positions. This research bridges that gap by leveraging the reasoning capabilities of Large Language Models (LLMs).

Methodology: The AFA Pipeline

The researchers introduced a modular Actor–Frame–Argument (AFA) pipeline designed to work in stages rather than through a single prompt. This multi-stage approach ensures higher fidelity in extracting complex information from financial texts.

Overview of the data collection, AFA pipeline, and Decompositional Verification Framework (DVF).

As illustrated in the pipeline overview, the system performs three core tasks:

  1. Actor–Stance Identification: Extracts the specific entity speaking and their stance (e.g., supportive or opposing).
  2. Frame Classification: Categorizes the thematic "lens" of the article (e.g., Economic Risk vs. Economic Opportunity) using a custom eight-frame typology.
  3. Argument Extraction: Decomposes the text into a central claim, supporting evidence, and a "warrant" (the logical link between the two).

To ensure these outputs are reliable, the team developed the Decompositional Verification Framework (DVF). This framework uses a panel of "judge" models to evaluate extractions based on completeness, faithfulness to the source text, logical coherence, and relevance. These automated scores were calibrated against a "Gold Standard" benchmark of 2,000 human-annotated articles.

Data and Sampling

The study utilized the Dow Jones Climate News Corpus, a massive collection of 980,061 climate-related articles spanning 2000–2023. From this, a strategic sample of 4,143 articles was selected through temporal stratification and thematic clustering to represent the diversity of the full corpus. The team also employed Active Learning to identify and include "hard" argumentative cases, ensuring the model was tested on complex discourse rather than simple PR snippets.

Longitudinal Insights: The Great Pivot

Using changepoint analysis, the researchers identified a statistically significant structural break in discourse between late 2015 and early 2016, aligning with the Paris Agreement. This moment, dubbed "The Great Pivot," marked the point where the narrative flipped from "compliance and cost" to "innovation and opportunity."

Longitudinal shift in the share of frames within financial news, showing the rise of 'Economic Opportunity' and the decline of 'Economic Risk'.

As shown in the frame analysis, the "Economic Opportunity" framing rose by 15.4% over the study period, while "Economic Risk" and "Regulatory Compliance" frames dropped by 13.1% and 9.9% respectively. This shift is accompanied by a change in dominant voices: while governments led the conversation in the early 2000s (~31%), financial institutions like banks and asset managers became the loudest voices by 2023 (~34%), signaling a "marketization" of climate discourse.

Key Results and Performance

The Gemini-2.5-based pipeline achieved high performance across all metrics, with an overall F1 score of 0.819 for actor identification and 0.767 for argument extraction. Notably, zero-shot LLMs significantly outperformed fine-tuned, domain-specific models like RoBERTa-Large, particularly in identifying frames (+15.2 F1 score). This highlights the value of LLM reasoning over simple pattern matching in complex financial texts.

Conclusion

This study reveals a profound shift in how the financial elite discuss climate change, moving from a focus on risk to a focus on opportunity. While this discursive shift is significant, the authors acknowledge that the study maps discourse, not action. Future work is needed to determine if this "opportunity" talk translates into actual carbon reduction or if it primarily serves as a new form of greenwashing. Linking these discursive shifts to real-world capital flows remains a critical next step for researchers.

Summary updated

Abstract

Financial news media shapes trillion-dollar climate investment decisions, yet discourse in this elite domain remains underexplored. We analyze two decades of climate-related articles (2000-2023) from Dow Jones Newswire using an Actor-Frame-Argument (AFA) pipeline that extracts who speaks, how issues are framed, and which arguments are deployed. We validate extractions against 2,000 human-annotated articles using a Decompositional Verification Framework that evaluates completeness, faithfulness, coherence, and relevance. Our longitudinal analysis uncovers a structural transformation: pre-2015 coverage emphasized risk and regulatory burden; post-Paris Agreement, discourse shifted toward economic opportunity and innovation, with financial institutions becoming dominant voices. Methodologically, we provide a replicable paradigm for longitudinal media analysis with LLMs; substantively, we reveal how financial elites have internalized and reframed the climate crisis across two decades.
BibTeX
@misc{su2026actors,
  author = {Ruiran Su and Janet B. Pierrehumbert and Markus Leippold},
  title = {{Actors, Frames and Arguments: A Multi-Decade Computational Analysis of Climate Discourse in Financial News using Large Language Models}},
  year = {2026},
  eprint = {2601.10142},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2601.10142},
}
Figure