gxceed
← 論文一覧に戻る

Corporate Climate Narratives for Transition‐Finance Governance: A Language‐Model Index of Transition Risk and Financing Conditions in China

移行金融ガバナンスのための企業気候ナラティブ：中国における移行リスクと資金調達条件の言語モデル指標 (AI 翻訳)

Xuewen Kuang, Yinyin Liu, Jianmin Liu, Yuqi Bai

Sustainable Development📚 査読済 / ジャーナル2026-05-29#トランジション・ファイナンスOrigin: CN
DOI: 10.1002/sd.71283
原典: https://doi.org/10.1002/sd.71283

🤖 gxceed AI 要約

日本語

本研究は、中国上場企業の年次報告書の経営者による説明（MD&A）セクションを用いて、ファインチューニングされたトランスフォーマーモデルにより移行リスク指数（CTR）を構築。企業レベルの移行リスク開示が高いほど借入コストが上昇することを示し、データ制約のある環境での移行金融ガバナンスへの示唆を提供する。

English

This study constructs a narrative-based transition-risk index (CTR) by fine-tuning a transformer classifier on Chinese listed firms' MD&A sections (2010-2023). It finds that higher disclosed transition risk is associated with a 0.335-0.365 percentage-point increase in the cost of debt, providing evidence for transition-finance governance in data-constrained settings.

Unofficial AI-generated summary based on the public title and abstract. Not an official translation.

📝 gxceed 編集解説 — Why this matters

日本のGX文脈において

日本でもSSBJ対応や有報・統合報告書における気候関連開示が進む中、自然言語処理を用いた移行リスク指標の構築手法は、日本企業の開示内容の分析に応用可能。特に、定性的なナラティブから資金調達条件への影響を検証する枠組みは、投資家対応や政策立案に有用な示唆を与える。

In the global GX context

This paper demonstrates how NLP can extract transition risk from corporate narratives in emerging markets, offering a scalable approach for TCFD/ISSB adoption where emissions data is incomplete. The finding that disclosed transition risk impacts borrowing costs supports transition-finance governance and has implications for global climate disclosure frameworks.

👥 読者別の含意

🔬研究者:Methodological contribution: fine-tuned transformer for climate risk measurement outperforms dictionary methods; causal identification strategy using shift-share instrument.

🏢実務担当者:Provides a tool to monitor transition risk from textual disclosures, useful for credit risk assessment and portfolio screening in data-poor environments.

🏛政策担当者:Evidence that disclosed transition risk influences financing conditions supports using narrative-based metrics to guide transition finance governance and disclosure mandates.

📄 Abstract（原文）

ABSTRACT Achieving the Sustainable Development Goals (SDGs), especially SDG 13 (Climate Action), requires mobilizing transition finance, yet firm‐level climate transition risk is difficult to observe in emerging economies with incomplete emissions data and uneven disclosure. Using the Management Discussion and Analysis sections of Chinese listed firms' annual reports from 2010 to 2023, this study constructs a narrative‐based transition‐risk index (CTR) by fine‐tuning a sentence‐level transformer classifier and aggregating sentence‐level transition‐risk intensity to the firm‐year level. The language‐model approach outperforms dictionary methods on a held‐out hand‐labeled test set. In within‐firm fixed‐effects regressions, higher disclosed transition‐risk content, as captured by CTR, is associated with higher borrowing costs, indicating tighter debt‐market financing conditions; in the controlled baseline specifications, a one‐standard‐deviation increase in CTR is associated with a 0.335–0.365 percentage‐point increase in the cost of debt. A shift‐share instrument combining predetermined baseline carbon intensity with leave‐one‐industry‐out transition‐risk trends provides supportive, though not definitive, identification evidence. In joint disclosure‐based specifications, a narrative‐based physical‐risk measure extracted from the same source exhibits a different pricing pattern from transition‐risk narratives, consistent with the possibility that physical‐risk disclosure mixes exposure with adaptation‐ or resilience‐related content. Overall, the results indicate that disclosed transition‐risk content, rather than a cleanly observed structural risk measure, contains lender‐relevant information and can support transition‐finance governance in data‐constrained settings.

🔗 Provenance — このレコードを発見したソース

🔔 こうした論文の新着を逃したくない方は キーワードアラート に登録（無料・3キーワードまで）。

gxceed は公開メタデータに基づく研究支援データセットです。要約・翻訳・解説は AI 支援で生成されています。 最終的な解釈・検証は利用者が原典資料に基づいて行うことを前提とします。