← 論文一覧に戻る

To what extent can spreadsheets shape sustainability? A machine learning approach to ESG score prediction

スプレッドシートはどこまで持続可能性を形成できるか？機械学習によるESGスコア予測アプローチ (AI 翻訳)

Hussam Musa, Zdenka Musová, Frederik Rech, Janka Grofčíková

Transformations In Business & Economics📚 査読済 / ジャーナル2026-07-01#AI×ESGOrigin: EU対象セクター: manufacturing
DOI: 10.15388/tibe.2026.25.2.21
原典: https://doi.org/10.15388/tibe.2026.25.2.21

🤖 gxceed AI 要約

日本語

スロバキア製造業974社のESGスコアを財務指標から予測するため、XGBoostとSHAP分析を適用。レバレッジ、流動性、債務返済能力、企業年数、税関連指標が重要な予測因子であることを発見。中程度のESGカテゴリは高い精度で予測できるが、極端なカテゴリはクラス不均衡により困難。財務データはESG評価に対して補完的な情報を提供するが、ESG固有の定性的情報を代替できない。

English

This study applies XGBoost and SHAP analysis to predict ESG ratings of 974 Slovak manufacturing firms using multi-year financial data. Leverage, liquidity, debt-servicing capacity, firm age, and tax-related indicators emerged as important predictors. The model predicted the middle ESG rating category reliably but struggled with extreme categories due to class imbalance. The findings suggest that financial data can support but not replace qualitative ESG-specific information for ESG score prediction.

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

スロバキア製造業の事例だが、日本企業でも有報データを用いたESGスコア予測に応用可能。ただし、日本独自のESG評価枠組み（例：SSBJ基準）への適合性は別途検討が必要。

In the global GX context

This paper demonstrates the potential of machine learning for ESG prediction using financial data, which is relevant for investors and rating agencies globally. It highlights the complementarity and limitations of purely financial models in ESG assessment, contributing to the broader discussion on integrating quantitative and qualitative factors in ESG evaluation.

👥 読者別の含意

🔬研究者:Shows applicability of XGBoost and SHAP for ESG prediction with financial features; useful for developing more comprehensive ESG models that combine financial and non-financial data.

🏢実務担当者:Indicates that financial data alone can provide rough ESG score estimates, but qualitative ESG data remains necessary for accuracy; useful for initial screening or identifying key financial drivers of ESG performance.

📄 Abstract（原文）

This study examines whether corporate ESG ratings can be predicted using multi-year lagged financial indicators. The objective is to evaluate the extent to which historical financial information explains ESG performance among Slovak manufacturing firms. The analysis uses a sample of 974 Slovak manufacturing firms with ESG ratings for 2023 and financial data from 2018–2022. The primary model applies XGBoost with recursive feature elimination and SHAP analysis, while a PCA-based one-vs-rest XGBoost model is used as a robustness check. The findings show that historical financial indicators provide meaningful but incomplete information about ESG performance. Leverage, liquidity, debt-servicing capacity, firm age, and tax-related indicators emerge as important predictors. Older lagged variables also remain significant, suggesting that ESG outcomes reflect longer-term financial patterns. The alternative model improves prediction, particularly for the Environmental and Governance pillars, whereas the aggregate ESG score remains more difficult to classify. The models predict the middle ESG rating category most reliably, while performance for extreme categories is limited by class imbalance and the ordinal structure of ESG scores. The study concludes that financial data can support ESG prediction, especially at the pillar level, but cannot replace qualitative and ESG-specific information.

🔗 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。