Disentangling Greenwashing from Authentic Environmental Performance: A Double Machine Learning Approach to Corporate Sustainability Disclosure
グリーンウォッシュと本物の環境パフォーマンスの分離：企業のサステナビリティ開示へのダブルマシンラーニングアプローチ (AI 翻訳)
M. Pinarci
🤖 gxceed AI 要約
日本語
本論文は、企業のサステナビリティ報告におけるグリーンウォッシュを検出するための、BERTとダブルマシンラーニングを組み合わせた因果推論フレームワークを提案する。合成データを用いた実証で、高開示企業の約6.5%が潜在的なグリーンウォッシュであることを特定した。手法は実データへの応用が可能で、再現可能なコードを提供している。
English
This paper proposes a causal inference framework combining BERT and Double Machine Learning to detect greenwashing in corporate sustainability reports. Using synthetic panel data, it identifies approximately 6.5% of high-disclosing firms as potential greenwashers based on disclosure-outcome mismatch. The method is generalizable and includes reproducible code.
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が開示基準を策定中であり、グリーンウォッシュ検出は重要。本手法は開示と実績の不一致を因果推論で捉える点で、日本の投資家や規制当局にとって有用。
In the global GX context
Globally, regulators (SEC, ESMA, FSA) are cracking down on greenwashing. This paper provides a causal method to statistically identify potential greenwashing, which could be adopted by auditors and regulators to enhance disclosure credibility.
👥 読者別の含意
🔬研究者:Researchers can apply this framework to real datasets to empirically estimate greenwashing prevalence and improve detection methods.
🏢実務担当者:Corporate sustainability teams can use this approach to internally audit their own disclosure-performance alignment and reduce greenwashing risk.
🏛政策担当者:Policymakers can leverage this methodology to statistically screen filings for potential greenwashing and inform enforcement actions.
📄 Abstract（原文）
Sustainability reports have become standard in corporate disclosure. Yet the relationship between disclosed environmental commitment and actual performance remains unclear. This paper develops and demonstrates a novel methodological framework combining Double Machine Learning with Natural Language Processing (BERT) to detect greenwashing through causal inference. We introduce an integrated approach: (1) BERT-based quantification of disclosure intensity from sustainability reports, (2) Double Machine Learning with 5-fold cross-fitting to handle 250+ high-dimensional controls, and (3) Causal forest estimation of heterogeneous treatment effects. Using synthetic panel data structured to match real-world characteristics (850 firms × 8 years = 6,800 observations), we demonstrate the complete methodology pipeline. Results show that our DML framework successfully recovers causal treatment effects with valid confidence intervals, and we identify approximately 6.5% of high-disclosing firms as potential greenwashers based on disclosure-outcome mismatch patterns. The framework is generalizable to real corporate data and other disclosure-outcome applications. We provide reproducible code and full transparency on data construction to enable future empirical applications with access to actual disclosure and environmental outcome data.
🔗 Provenance — このレコードを発見したソース
- semanticscholar https://doi.org/10.2139/ssrn.6710601first seen 2026-05-15 18:04:47
🔔 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。