ESG特徴量を統合した財務危機早期警告のための解釈可能な機械学習モデルに関する研究
Research on Interpretable Machine Learning Model for Financial Distress Early Warning Integrating ESG Features (原題)
(著者不明)
🤖 gxceed AI 要約
日本語
会計データに偏りがちな従来の財務危機予測に、ESGスコア・炭素強度・企業規模・監査意見といった非財務指標を融合したハイブリッド予測枠組みを中国上場企業向けに提案。XGBoostとRandom Forestを予測エンジンとし、不均衡データをオーバーサンプリングで補正、SHAPで各特徴量の寄与と方向性を解釈。ESG≥6.25かつCI<1.24で危機確率が有意に低下することを示した。
English
This paper proposes a hybrid ML framework fusing financial indicators with ESG scores, carbon intensity, firm size and audit opinions to predict financial distress among Chinese listed firms. Using XGBoost and Random Forest with oversampling and SHAP explainability, it finds distress probability drops significantly when ESG ≥ 6.25 and CI < 1.24. Results show ESG-augmented models improve predictive accuracy, informing risk management and policy.
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
SSBJ基準や有報でのサステナビリティ開示が進む日本では、ESG指標を財務リスク評価に組み込む実務的意義が高まっている。SHAPによる解釈可能性は、投資家対応や統合報告書でのリスク説明にも応用可能な示唆を提供する。
In the global GX context
Amid ISSB/CSRD-driven disclosure, this work demonstrates that ESG and carbon-intensity features carry genuine predictive signal for financial distress, supporting the case that sustainability data is financially material. The SHAP-based interpretability aligns with growing regulatory demand for explainable climate-risk models.
👥 読者別の含意
🔬研究者:ESG・炭素強度を特徴量に組み込んだ解釈可能MLによる財務危機予測の実証例として、AI×ESG研究の方法論的参照点となる。
🏢実務担当者:ESGスコアや炭素強度を自社の信用・リスク管理モデルに取り込む際の閾値や特徴量設計の参考になる。
🏛政策担当者:ESG・炭素情報の開示が金融安定に資することを示す証拠として、開示制度設計の根拠づけに活用できる。
📄 Abstract（原文）
Precise early-warning of corporate financial distress is fundamental to maintaining capital-market stability and ensuring sustainable firm-level development. Addressing the limitation that traditional financial-distress studies rely almost exclusively on accounting data while overlooking non-financial determinants, this paper proposes a hybrid predictive framework that fuses conventional financial indicators with sustainability-related non-financial metrics for Chinese listed companies. Financial variables are selected across five dimensions—solvency, profitability, operating efficiency, growth potential and cash-flow generation—to capture core financial characteristics, whereas non-financial variables incorporate environmental, social and governance (ESG) scores, carbon intensity (CI), firm size(Size) and audit opinions(AO), thereby enriching the information set. Machine-learning algorithms—namely eXtreme Gradient Boosting(XGBoost)and Random Forest(RF)—serve as the primary predictive engines, and the severe class imbalance induced by the low prevalence of distress events is alleviated through oversampling techniques. Model performance is evaluated comprehensively via accuracy, precision and the area under the ROC curve (AUC), while SHAP explainability is exploited to quantify both the marginal impact and the directional effect of each feature on distress probability. Empirically, we document that the likelihood of financial distress decreases significantly when ESG ≥ 6.25 and CI < 1.24, whereas financial-risk exposure is materially attenuated for firms whose leverage is below 0.68 and total-asset turnover exceeds 1.7. The findings corroborate that integrating multi-dimensional indicators with advanced machine-learning techniques substantially enhances predictive accuracy, offering robust scientific guidance for corporate risk management, investor decision-making and regulatory policy design.
🔗 Provenance — このレコードを発見したソース
- semanticscholar https://doi.org/10.1145/3801438.3801472first seen 2026-09-11 05:50:54 · last seen 2026-09-21 05:21:08
🔔 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。