學術產出-Theses
Article View/Open
Publication Export
-
題名 使用半監督學習預測範疇三碳排建構投資組合之分析
Predicted Scope 3 Carbon Emissions Portfolios Analysis: Semi-supervised Learning Approach作者 閻家緯
Yen, Chia Wei貢獻者 楊曉文
Yang, Sharon S.
閻家緯
Yen, Chia Wei關鍵詞 半監督學習
範疇三碳排
碳密度策略
績效回測
Semi-supervised Learning
Scope 3 Carbon Emissions
Portfolio Analysis日期 2023 上傳時間 2-Aug-2023 14:11:49 (UTC+8) 摘要 本研究旨在使用半監督式學習 COREG 模型，以公司的財報資料當作解釋 變數預測範疇三碳排，並使用碳密度當作因子，建構出具有高碳和低碳性質的 投資組合進行績效回溯。本研究的第一部份比較使用隨機森林和 COREG 模型 在預測範疇三碳排上的解釋力，資料母體為台灣 2010 年至 2021 年中有揭露範 疇一和範疇二碳排的上市櫃公司，解釋變數使用公司每一年公開揭露的財報資 料，在預測期間的 2016 年至 2021 年間發現，在每年平均絕對誤差(MAE)相 同的情況下，COREG 模型的均方根對數誤差(RMSLE)平均每年相對於隨機 森林模型減少約40%，顯示隨機森林模型傾向低估碳排放，且 COREG 模型 在捕捉尾端碳排時能有比較優異的表現。在第二部分裡，本研究進一步利用碳 密度當作篩選因子，建立低碳和高碳密度的投資組合進行回溯測試，在回測期 間的 2018 年至 2022 年之間，主要有以下兩點發現，第一，使用直接碳排建構 的碳密度投資組合有低碳高報酬現象;而使用間接碳排建構的碳密度投資組合 則有高碳高報酬現象。第二，使用已揭露碳排建構的碳密度投資組合之夏普比 率皆無法打敗大盤;而使用 COREG 模型預測的範疇三碳排所建立的投資組 合，除了呈現高碳至低碳的排序性之外，這之中的前 10% 碳密度投資組合不 僅在報酬上勝過台灣加權報酬指數，其夏普比率也是唯一一組在回測期間勝過 報酬指數的碳密度投資組合。
The study proposes the use of semi-supervised learning COREG model to predict scope 3 carbon emissions using company financial data as explanatory variables. Also, we conduct backtesting to evaluate the performance of high and low carbon density portfolios. In the first part, we compare the performance of random forest and COREG models in predicting scope 3 carbon emissions. The dataset comprises listed and over-the-counter (OTC) companies in Taiwan from 2010 to 2021 that have disclosed both scope 1 and scope 2 carbon emissions. The explanatory variables consist of financial data disclosed by these companies. Results show that the COREG model achieves an average reduction of approximately 40% in RMSLE compared to the random forest model, with same MAE each year. This indicates that the random forest model tends to underestimate carbon emissions, while the COREG model performs better in capturing extreme carbon emissions. In the second part, we construct backtesing for low and high carbon density portfolio. From the period 2018 to 2022, firstly, we found portfolios constructed based on direct carbon emissions show negative relationship between carbon density and returns, while portfolios constructed based on indirect carbon emissions show a positive relationship between them. Second, portfolios constructed using disclosed carbon emissions failed to outperform the TAIEX Total Return Index in terms of Sharpe ratio. However, we found the predicted scope 3 carbon density portfolios exhibited a high-carbon-to-low- carbon returns pattern. Furthermore, the top 10% carbon density portfolio had the highest returns and Sharpe ratio among all the portfolios including market benchmark.參考文獻 1. Zhou, P., Ang, B. W., & Han, J. Y. (2010). Total factor carbon emission performance: a Malmquist index analysis. Energy Economics, 32(1), 194-201.2. Choi, B. B., Lee, D., & Psaros, J. (2013). An analysis of Australian company carbon emission disclosures. Pacific Accounting Review.3. Kılıç, M., & Kuzey, C. (2018). The effect of corporate governance on carbon emission disclosures: Evidence from Turkey. International Journal of Climate Change Strategies and Management.4. Zhang, Y. J. (2011). The impact of financial development on carbon emissions: An empirical analysis in China. Energy policy, 39(4), 2197-2203.5. Acheampong, A. O., Amponsah, M., & Boateng, E. (2020). Does financial development mitigate carbon emissions? Evidence from heterogeneous financial economies. Energy Economics, 88, 104768.6. Lee, O., Joo, H., Choi, H., & Cheon, M. (2022). Proposing an integrated approach to analyzing ESG data via machine learning and deep learning algorithms. Sustainability, 14(14), 8745.7. Zhao, Y., Liu, R., Liu, Z., Liu, L., Wang, J., & Liu, W. (2023). A Review of Macroscopic Carbon Emission Prediction Model Based on Machine Learning. Sustainability, 15(8), 6876.8. Assael, J., Heurtebize, T., Carlier, L., & Soupé, F. (2023). Greenhouse gases emissions: estimating corporate non-reported emissions using interpretable machine learning. Sustainability, 15(4), 3391.9. Nguyen, Q., Diaz-Rainey, I., & Kuruppuarachchi, D. (2021). Predicting corporate carbon footprints for climate finance risk analyses: a machine learning approach. Energy Economics, 95, 105129.10. Nguyen, Q., Diaz-Rainey, I., Kitto, A., McNeil, B., Pittman, N., & Zhang, R. (2022). Scope 3 Emissions: Data Quality and Machine Learning Prediction Accuracy11. Serafeim, G., & Velez Caicedo, G. (2022). Machine Learning Models for Prediction of Scope 3 Carbon Emissions. Available at SSRN.12. Lacoste, A., Luccioni, A., Schmidt, V., & Dandres, T. (2019). Quantifying the carbon emissions of machine learning. arXiv preprint arXiv:1910.09700.13. De Lucia, C., Pazienza, P., & Bartlett, M. (2020). Does good ESG lead to better financial performances by firms? Machine learning and logistic regression models of public enterprises in Europe. Sustainability, 12(13), 5317.14. Chen, Q., & Liu, X. Y. (2020, October). Quantifying ESG alpha using scholar big data: an automated machine learning approach. In Proceedings of the First ACM International Conference on AI in Finance (pp. 1-8).15. De Franco, C., Geissler, C., Margot, V., & Monnier, B. (2020). Esg investments: Filtering versus machine learning approaches. arXiv preprint arXiv:2002.07477.16. Bolton, P., & Kacperczyk, M. (2021). Do investors care about carbon risk?. Journal of financial economics, 142(2), 517-549.17. Jacobsen, B., Lee, W., & Ma, C. (2019). The alpha, beta, and sigma of ESG: Better beta, additional alpha?. Journal of Portfolio Management, 45(6), 6-15.18. Misani, N., & Pogutz, S. (2015). Unraveling the effects of environmental outcomes and processes on financial performance: A non-linear approach. Ecological economics, 109, 150-160.19. Guo, T., Jamet, N., Betrix, V., Piquet, L. A., & Hauptmann, E. (2020). Esg2risk: A deep learning framework from esg news to stock volatility prediction. arXiv preprint arXiv:2005.02527.20. Sokolov, A., Caverly, K., Mostovoy, J., Fahoum, T., & Seco, L. (2021). Weak supervision and black-litterman for automated esg portfolio construction. The Journal of Financial Data Science, 3(3), 129-138.21. Pedregosa, F., Varoquaux, G., Gramfort, A., Michel, V., Thirion, B., Grisel, O., ... & Duchesnay, E. (2011). Scikit-learn: Machine learning in Python. the Journal of machine Learning research, 12, 2825-2830.22. Zhou, Z. H., & Li, M. (July). Semi-supervised regression with co-training. In IJCAI (Vol. 5, pp. 908-913). 描述 碩士
國立政治大學
金融學系
110352033資料來源 http://thesis.lib.nccu.edu.tw/record/#G0110352033 資料類型 thesis dc.contributor.advisor 楊曉文 zh_TW dc.contributor.advisor Yang, Sharon S. en_US dc.contributor.author (Authors) 閻家緯 zh_TW dc.contributor.author (Authors) Yen, Chia Wei en_US dc.creator (作者) 閻家緯 zh_TW dc.creator (作者) Yen, Chia Wei en_US dc.date (日期) 2023 en_US dc.date.accessioned 2-Aug-2023 14:11:49 (UTC+8) - dc.date.available 2-Aug-2023 14:11:49 (UTC+8) - dc.date.issued (上傳時間) 2-Aug-2023 14:11:49 (UTC+8) - dc.identifier (Other Identifiers) G0110352033 en_US dc.identifier.uri (URI) http://nccur.lib.nccu.edu.tw/handle/140.119/146602 - dc.description (描述) 碩士 zh_TW dc.description (描述) 國立政治大學 zh_TW dc.description (描述) 金融學系 zh_TW dc.description (描述) 110352033 zh_TW dc.description.abstract (摘要) 本研究旨在使用半監督式學習 COREG 模型，以公司的財報資料當作解釋 變數預測範疇三碳排，並使用碳密度當作因子，建構出具有高碳和低碳性質的 投資組合進行績效回溯。本研究的第一部份比較使用隨機森林和 COREG 模型 在預測範疇三碳排上的解釋力，資料母體為台灣 2010 年至 2021 年中有揭露範 疇一和範疇二碳排的上市櫃公司，解釋變數使用公司每一年公開揭露的財報資 料，在預測期間的 2016 年至 2021 年間發現，在每年平均絕對誤差(MAE)相 同的情況下，COREG 模型的均方根對數誤差(RMSLE)平均每年相對於隨機 森林模型減少約40%，顯示隨機森林模型傾向低估碳排放，且 COREG 模型 在捕捉尾端碳排時能有比較優異的表現。在第二部分裡，本研究進一步利用碳 密度當作篩選因子，建立低碳和高碳密度的投資組合進行回溯測試，在回測期 間的 2018 年至 2022 年之間，主要有以下兩點發現，第一，使用直接碳排建構 的碳密度投資組合有低碳高報酬現象;而使用間接碳排建構的碳密度投資組合 則有高碳高報酬現象。第二，使用已揭露碳排建構的碳密度投資組合之夏普比 率皆無法打敗大盤;而使用 COREG 模型預測的範疇三碳排所建立的投資組 合，除了呈現高碳至低碳的排序性之外，這之中的前 10% 碳密度投資組合不 僅在報酬上勝過台灣加權報酬指數，其夏普比率也是唯一一組在回測期間勝過 報酬指數的碳密度投資組合。 zh_TW dc.description.abstract (摘要) The study proposes the use of semi-supervised learning COREG model to predict scope 3 carbon emissions using company financial data as explanatory variables. Also, we conduct backtesting to evaluate the performance of high and low carbon density portfolios. In the first part, we compare the performance of random forest and COREG models in predicting scope 3 carbon emissions. The dataset comprises listed and over-the-counter (OTC) companies in Taiwan from 2010 to 2021 that have disclosed both scope 1 and scope 2 carbon emissions. The explanatory variables consist of financial data disclosed by these companies. Results show that the COREG model achieves an average reduction of approximately 40% in RMSLE compared to the random forest model, with same MAE each year. This indicates that the random forest model tends to underestimate carbon emissions, while the COREG model performs better in capturing extreme carbon emissions. In the second part, we construct backtesing for low and high carbon density portfolio. From the period 2018 to 2022, firstly, we found portfolios constructed based on direct carbon emissions show negative relationship between carbon density and returns, while portfolios constructed based on indirect carbon emissions show a positive relationship between them. Second, portfolios constructed using disclosed carbon emissions failed to outperform the TAIEX Total Return Index in terms of Sharpe ratio. However, we found the predicted scope 3 carbon density portfolios exhibited a high-carbon-to-low- carbon returns pattern. Furthermore, the top 10% carbon density portfolio had the highest returns and Sharpe ratio among all the portfolios including market benchmark. en_US dc.description.tableofcontents 第一章 緒論 8第一節 研究背景 8第二節 研究目的 9第三節 研究架構 10第二章 文獻回顧 12第一節 碳排相關的研究 12第二節 機器學習與碳排 13第三節 機器學習與 ESG投資組合 14第三章 碳排預測方法之介紹：機器學習模型 17第一節 隨機森林模型 17第二節 COREG 模型 17第三節 模型評估指標 21第四章 範疇三碳排預測結果 23第一節 研究資料 23一、資料來源 23二、變數介紹 23三、敘述性統計 24第二節 範疇三碳排預測結果 25一、碳排預測架構 25二、隨機森林預測結果 26三、COREG預測結果 27四、碳排預測結果比較 28第五章 考量碳排放密度之投資組合績效評估 30第一節、碳密度投資組合之建構 30第二節 碳密度投資組合之績效評估 30一、總碳密度投資組合之分析 30二、範疇一碳密度投資組合之分析 35三、範疇二碳密度投資組合之分析 39四、範疇三碳密度投資組合之分析 42第六章 結論與建議 47第七章 參考文獻 50附錄一 特徵篩選 53附錄二 資料處理 62附錄三 前 10% 碳密度投資組合比較 66 zh_TW dc.format.extent 5183940 bytes - dc.format.mimetype application/pdf - dc.source.uri (資料來源) http://thesis.lib.nccu.edu.tw/record/#G0110352033 en_US dc.subject (關鍵詞) 半監督學習 zh_TW dc.subject (關鍵詞) 範疇三碳排 zh_TW dc.subject (關鍵詞) 碳密度策略 zh_TW dc.subject (關鍵詞) 績效回測 zh_TW dc.subject (關鍵詞) Semi-supervised Learning en_US dc.subject (關鍵詞) Scope 3 Carbon Emissions en_US dc.subject (關鍵詞) Portfolio Analysis en_US dc.title (題名) 使用半監督學習預測範疇三碳排建構投資組合之分析 zh_TW dc.title (題名) Predicted Scope 3 Carbon Emissions Portfolios Analysis: Semi-supervised Learning Approach en_US dc.type (資料類型) thesis en_US dc.relation.reference (參考文獻) 1. Zhou, P., Ang, B. W., & Han, J. Y. (2010). Total factor carbon emission performance: a Malmquist index analysis. Energy Economics, 32(1), 194-201.2. Choi, B. B., Lee, D., & Psaros, J. (2013). An analysis of Australian company carbon emission disclosures. Pacific Accounting Review.3. Kılıç, M., & Kuzey, C. (2018). The effect of corporate governance on carbon emission disclosures: Evidence from Turkey. International Journal of Climate Change Strategies and Management.4. Zhang, Y. J. (2011). The impact of financial development on carbon emissions: An empirical analysis in China. Energy policy, 39(4), 2197-2203.5. Acheampong, A. O., Amponsah, M., & Boateng, E. (2020). Does financial development mitigate carbon emissions? Evidence from heterogeneous financial economies. Energy Economics, 88, 104768.6. Lee, O., Joo, H., Choi, H., & Cheon, M. (2022). Proposing an integrated approach to analyzing ESG data via machine learning and deep learning algorithms. Sustainability, 14(14), 8745.7. Zhao, Y., Liu, R., Liu, Z., Liu, L., Wang, J., & Liu, W. (2023). A Review of Macroscopic Carbon Emission Prediction Model Based on Machine Learning. Sustainability, 15(8), 6876.8. Assael, J., Heurtebize, T., Carlier, L., & Soupé, F. (2023). Greenhouse gases emissions: estimating corporate non-reported emissions using interpretable machine learning. Sustainability, 15(4), 3391.9. Nguyen, Q., Diaz-Rainey, I., & Kuruppuarachchi, D. (2021). Predicting corporate carbon footprints for climate finance risk analyses: a machine learning approach. Energy Economics, 95, 105129.10. Nguyen, Q., Diaz-Rainey, I., Kitto, A., McNeil, B., Pittman, N., & Zhang, R. (2022). Scope 3 Emissions: Data Quality and Machine Learning Prediction Accuracy11. Serafeim, G., & Velez Caicedo, G. (2022). Machine Learning Models for Prediction of Scope 3 Carbon Emissions. Available at SSRN.12. Lacoste, A., Luccioni, A., Schmidt, V., & Dandres, T. (2019). Quantifying the carbon emissions of machine learning. arXiv preprint arXiv:1910.09700.13. De Lucia, C., Pazienza, P., & Bartlett, M. (2020). Does good ESG lead to better financial performances by firms? Machine learning and logistic regression models of public enterprises in Europe. Sustainability, 12(13), 5317.14. Chen, Q., & Liu, X. Y. (2020, October). Quantifying ESG alpha using scholar big data: an automated machine learning approach. In Proceedings of the First ACM International Conference on AI in Finance (pp. 1-8).15. De Franco, C., Geissler, C., Margot, V., & Monnier, B. (2020). Esg investments: Filtering versus machine learning approaches. arXiv preprint arXiv:2002.07477.16. Bolton, P., & Kacperczyk, M. (2021). Do investors care about carbon risk?. Journal of financial economics, 142(2), 517-549.17. Jacobsen, B., Lee, W., & Ma, C. (2019). The alpha, beta, and sigma of ESG: Better beta, additional alpha?. Journal of Portfolio Management, 45(6), 6-15.18. Misani, N., & Pogutz, S. (2015). Unraveling the effects of environmental outcomes and processes on financial performance: A non-linear approach. Ecological economics, 109, 150-160.19. Guo, T., Jamet, N., Betrix, V., Piquet, L. A., & Hauptmann, E. (2020). Esg2risk: A deep learning framework from esg news to stock volatility prediction. arXiv preprint arXiv:2005.02527.20. Sokolov, A., Caverly, K., Mostovoy, J., Fahoum, T., & Seco, L. (2021). Weak supervision and black-litterman for automated esg portfolio construction. The Journal of Financial Data Science, 3(3), 129-138.21. Pedregosa, F., Varoquaux, G., Gramfort, A., Michel, V., Thirion, B., Grisel, O., ... & Duchesnay, E. (2011). Scikit-learn: Machine learning in Python. the Journal of machine Learning research, 12, 2825-2830.22. Zhou, Z. H., & Li, M. (July). Semi-supervised regression with co-training. In IJCAI (Vol. 5, pp. 908-913). zh_TW
