What Transfers Under Source Shift? Definitions, Examples, and Fine-Tuning for Climate Disclosure Classification

2026-07-20

AI Summary

This study addresses the distributional discrepancies in climate-related disclosure classification across diverse textual sources—such as annual reports, press releases, and earnings call transcripts—and investigates the transferability of large language models (LLMs) in cross-source settings. While existing approaches perform well within a single source, their effectiveness when transferred across sources remains unclear. The work presents the first systematic evaluation of LLM adaptation strategies—including structured prompt definitions, few-shot examples, and fine-tuning—across eleven models and two heterogeneous corpora sharing common label spaces. Results demonstrate that strategies aligning task granularity with structured definitions and employing random few-shot examples exhibit greater transfer robustness than more complex alternatives. Notably, under source shifts, simpler adaptation methods prove safer and more effective, offering practical guidance for cross-domain analysis of climate disclosures.

Abstract

Climate disclosure classification is a fundamental task for analysing corporate climate disclosures, yet such disclosures appear in many different sources -- annual reports, press releases, and earnings calls -- that differ in length, purpose, and writing style. Existing evaluations are mostly conducted within a single source, leaving open whether common LLM adaptation strategies remain effective under source shift. We reframe climate disclosure classification as a cross-source adaptation problem and study three widely used adaptation strategies -- definitions, examples, and fine-tuning -- across eleven open- and closed-source LLMs, using two corpora that share the same label space but come from different sources. We find that all strategies bring positive cross-source gains on average, but the strongest in-source strategies are not the strongest cross-source ones: similarity-based retrieval and LoRA fine-tuning gain most in-source but lose most of that advantage under source shift; randomly selected few-shot examples, a weaker in-source baseline, retain their advantage more reliably; definitions transfer most consistently, though only when their granularity matches the target text. Across these strategies, when the source changes, simpler is often safer.

Problem

• source shift
• climate disclosure classification
• cross-source adaptation
• LLM adaptation
• domain transfer

Innovation

• cross-source adaptation
• source shift
• climate disclosure classification
• LLM adaptation strategies
• definition prompting