本研究旨在探討銀行 ESG（環境、社會與公司治理）表現對資產品質風險分類的靜態預測效能，並運用機器學習方法驗證其可行性與應用潛力。以台灣 34 家本國上市銀行作為研究樣本，蒐集 106 年至 113 年之 ESG 評分與財務資料，並將不良貸款比率（NPL）依中位數進行二元分類，作為資產品質風險指標。模型部分採用五種分類技術，包括羅吉斯迴歸、決策樹、隨機森林、極限梯度提升與支持向量機，進行訓練與測試資料之靜態預測比較，並分析各變數於模型中的重要性排序。
實證結果顯示，相較於 ESG 分數，歷史 NPL、貸款比率、備抵呆帳覆蓋率與資產報酬率等財務變數在各模型中具有較高的重要性，對風險分類預測貢獻較大。ESG 總分及其三構面表現則未呈現顯著預測力，顯示 ESG 評等在台灣銀行情境下之應用尚未成熟。儘管如此，機器學習方法仍展現出良好且穩定的分類效能，為未來風險監理與永續金融整合提供可行途徑。最後，亦針對研究限制與未來研究方向提出建議。

This study investigates the static predictive effectiveness of banks’ ESG (Environmental, Social, and Governance) performance on asset quality risk classification using machine learning methods. The research focuses on 34 publicly listed domestic banks in Taiwan, utilizing ESG ratings and financial data from 2017 to 2024. The non-performing loan (NPL) ratio is used as a proxy for asset quality and is converted into a binary classification based on the median. Five classification models are employed—Logistic Regression, Decision Tree, Random Forest, XGBoost, and Support Vector Machine—to evaluate prediction performance and variable importance.
Empirical results reveal that traditional financial indicators, such as historical NPL levels, loan ratio, allowance coverage ratio, and return on assets, play a more significant role in prediction than ESG scores. ESG indicators exhibit limited predictive power under current disclosure practices in Taiwan’s banking sector. Nevertheless, the machine learning models demonstrate robust and stable classification capabilities, offering practical insights for integrating risk assessment and sustainability practices in the future. The study concludes with recommendations for future research and acknowledges certain methodological limitations.

目錄
第一章 緒論	1
第一節 研究背景與動機	1
第二節 研究目的	3
第三節 研究架構	5
第二章 文獻探討	7
第一節 ESG 與銀行資產品質之關聯研究	7
第二節 影響銀行資產品質之相關因素	9
第三節 機器學習方法於銀行資產品質預測之應用	11
第三章 研究方法	13
第一節 研究對象與資料來源	13
第二節 研究變數介紹	15
第三節 研究模型與分析方法	18
第四節 模型分類效能評估指標	24
第四章 實證結果與分析	26
第一節 敘述性統計分析	26
第二節 變數組間差異與相關性分析	30
第三節 機器學習模型預測結果與重要變數呈現	33
第四節 模型預測表現綜合比較與分析	49
第五章 研究結論與後續建議	50
第一節 研究結論	50
第二節 研究限制	52
第三節 實務建議	54
第四節 未來研究方向	55
參考文獻	56
附錄	58



	
圖目錄
圖 1-1　研究架構圖	6
圖 4-1　研究變數間皮爾森相關係數熱力圖	32
圖 4-2　決策樹模型變數重要性排序圖	38
圖 4-3　隨機森林模型變數重要性排序圖	42
圖 4-4　XGBoost 模型變數重要性排序圖	45

 
表目錄
表 3-1　各變數名稱、定義與計算方式	17
表 4-1　變數敘述統計表	28
表 4-2　NPL 類別樣本分布表	29
表 4-3　Mann–Whitney U 檢定結果	31
表 4-4　羅吉斯迴歸模型混淆矩陣	33
表 4-5　羅吉斯迴歸預測效能指標	34
表 4-6　羅吉斯迴歸模型參數估計結果	36
表 4-7　決策樹模型混淆矩陣	37
表 4-8　決策樹預測效能指標	38
表 4-9　隨機森林模型混淆矩陣	40
表 4-10　隨機森林預測效能指標	41
表 4-11　XGBoost 模型混淆矩陣	44
表 4-12　XGBoost 預測效能指標	45
表 4-13　SVM 模型混淆矩陣	47
表 4-14　SVM 預測效能指標	48

Abdullah, M., Chowdhury, M. A. F., Uddin, A., & Moudud-Ul-Huq, S. (2023). Forecasting nonperforming loans using machine learning. Journal of Forecasting, 42(7), 1664–1689.
Athanasoglou, P. P., Brissimis, S. N., & Delis, M. D. (2008). Bank‑specific, industry‑specific and macroeconomic determinants of bank profitability. Journal of International Financial Markets, Institutions and Money, 18(2), 121–136.
Bătae, O. M., Dragomir, V. D., & Feleagă, L. (2020). Environmental, social, governance (ESG), and financial performance of European banks. Accounting and Management Information Systems, 19(3), 480–501.
Beck, R., Jakubik, P., & Piloiu, A. (2013). Non‑performing loans: What matters in addition to the economic cycle? ECB Working Paper Series, 1515.
Bofondi, M., & Ropele, T. (2011). Macroeconomic determinants of bad loans: Evidence from Italian banks. Bank of Italy Occasional Paper, 89.
Bouattour, A., Kalai, M., & Helali, K. (2024). The non-linear relationship between ESG performance and bank stability in the digital era: New evidence from a regime-switching approach. Humanities and Social Sciences Communications, 11, Article 15.
Boudriga, A., Taktak, N. B., & Jellouli, S. (2009). Banking supervision and nonperforming loans: A cross‑country analysis. Journal of Financial Economic Policy, 1(4), 286–318.
Bouvatier, V., & Lepetit, L. (2008). Banks’ procyclical behavior: Does provisioning matter? Journal of International Financial Markets, Institutions and Money, 18(5), 513–526.
Fekadu, R., Getachew, A., Tadele, Y., Ali, N., & Goytom, I. (2022). Machine learning models evaluation and feature importance analysis on NPL dataset. arXiv preprint.
Foos, D., Norden, L., & Weber, M. (2010). Loan growth and riskiness of banks. Journal of Banking & Finance, 34(12), 2929–2940.
Krisciukaityte, K., Balezentis, T., & Streimikiene, D. (2023). Linking financial performance and efficiency to sustainability in banking sector: A literature synthesis. Journal of Business Economics and Management, 24(3), 506–526.
Laeven, L., & Majnoni, G. (2003). Loan loss provisioning and economic slowdowns: Too much, too late? Journal of Financial Intermediation, 12(2), 178–197.
Liu, J., & Xie, J. (2024). The effect of ESG performance on bank liquidity risk. Sustainability, 16(12), 4927.
Louzis, D. P., Vouldis, A. T., & Metaxas, V. L. (2012). Macroeconomic and bank‑specific determinants of non‑performing loans in Greece: A comparative study of mortgage, business and consumer loan portfolios. Journal of Banking & Finance, 36(4), 1012–1027.
Messai, A. S., & Jouini, F. (2013). Micro and macro determinants of non‑performing loans. International Journal of Economics and Financial Issues, 3(4), 852–860.
Nkusu, M. (2011). Non‑performing loans and macrofinancial vulnerabilities in advanced economies. IMF Working Paper, 11/161.
Serengil, S. I., Imece, S., Tosun, U. G., Buyukbas, E. B., & Koroglu, B. (2022). A comparative study of machine learning approaches for non-performing loan prediction with explainability. International Journal of Machine Learning and Computing, 12(5), 208–217.