本研究聚焦於台灣熱門高股息ETF的績效分析，旨在透過這些ETF找出適合的選股條件，從而為投資者提供更有效的投資策略。首先，本研究整理並分析了台灣熱門高股息ETF的篩選條件，探討其在組成過程中所採用的選股標準及其對績效的影響。同時，研究也探討了近年來台灣高股息ETF受歡迎的原因。接下來，研究進行了高股息ETF的績效回測與預測分析，並應用機器學習技術來解析選股條件對股價報酬的影響。通過設計和實施績效回測，評估了自建高股息ETF在不同時期和市場狀況下的表現。進一步地，研究運用機器學習技術分析選股條件的影響力，識別出影響股價報酬的關鍵因素，並對模型進行評估和優化，以驗證其預測能力和準確性。研究結果顯示：（一）當利用歷史數據進行模型訓練並對選股條件進行績效回測時，通常能獲得較佳的表現；（二）然而，當使用相同的歷史數據進行模型訓練並預測未來三年的績效時，表現反而較差；（三）選股條件未能充分適應每年市場的波動，因此需要逐年調整和改進選股策略。

This study focuses on the performance analysis of popular high-dividend ETFs in Taiwan, aiming to identify optimal stock selection criteria that can provide investors with more effective investment strategies. First, the research compiles and analyzes the selection criteria of these popular high-dividend ETFs in Taiwan, exploring the stock selection standards used in their composition and their impact on performance. Additionally, the study investigates the reasons behind the growing popularity of high-dividend ETFs in Taiwan in recent years. Following this, the research conducts backtesting and predictive analysis of the performance of high-dividend ETFs, applying machine learning techniques to understand the influence of selection criteria on stock returns. By designing and implementing performance backtests, the study assesses the performance of self-constructed high-dividend ETFs across different periods and market conditions. Furthermore, the research employs machine learning techniques to analyze the impact of stock selection criteria, identifying key factors that influence stock returns, and evaluates and optimizes the models to validate their predictive accuracy. The findings indicate that: (1) training models using historical data and backtesting the selection criteria generally yields better performance; (2) however, when the same historical data is used to train models and predict the performance for the next three years, the results are less favorable; and (3) the stock selection criteria do not sufficiently adapt to annual market fluctuations, necessitating yearly adjustments and improvements to the selection strategies.

論文審定書 I
摘要 II
ABSTRACT III
第一章 緒論 1
1-1. 研究動機 1
1-2. 研究目的 1
1-3. 研究架構 2
第二章 文獻回顧 4
2-1. 台灣ETF市場簡介 4
2-2. ETF之相關文獻 6
2-3. 機器學習於提取特徵重要性之相關文獻 8
第三章 研究方法 14
3-1. 研究流程 14
3-2. 研究資料 16
3-3. 研究模型 16
3-4. 台灣熱門ETF篩選條件分析 21
3-5. 研究變數 30
3-6. 模型評估指標 34
3-7. 績效評估指標 34
第四章 實證結果 35
4-1. 特徵重要性 35
4-2. ETF績效回測選股條件（第一組自建ETF） 37
4-3. ETF績效預測選股條件（第二組自建ETF） 41
4-4. 績效實證結果與分析 45
第五章 結論 51
5-1. 研究結論 51
5-2. 未來研究方向 52
參考文獻 53

1. 張誌銘、何忠亮 (2019)。研究元大台灣高股息ETF定期與主動投資策略之探討。
2. 陳宏煦 (2008)。使用數據包絡分析法進行股票選擇。Industrial Management & Data Systems, 108(9), 1255-1268。
3. 游朝鋒、林士昊、蕭弘富 (2017)。台灣股市之高股息率投資策略。Investment Management & Financial Innovations, 7(2), 86-92。
4. AlSagri, H. S., & Ykhlef, M. (2020)。Quantifying feature importance for detecting depression using random forest。International Journal of Advanced Computer Science and Applications, 11(5), 124-130。
5. Ardana, A. (2023)。Performance analysis of xgboost algorithm to determine the most optimal parameters and features in predicting stock price movement。Journal of Financial Analysis and Research, 23(4), 101-115。
6. Bonelli, M. (2015). Exchange traded funds: Toward a tailored selection approach. European Finance eJournal, 1-17。
7. Chelley-Steeley, P., & Park, K. (2010). The adverse selection component of exchange traded funds. International Review of Financial Analysis, 19(1), 65-76.
8. Duanmu, J., Li, Y., & Malakhov, A. (2019). Capturing hedge fund risk factor exposures: Hedge fund return replication with ETFs. Financial Review, 54(1), 124-130.
9. Deng, T., Zhao, Y., Wang, S., & Yu, H. (2021)。Sales forecasting based on lightgbm。IEEE International Conference on Consumer Electronics and Computer Engineering, 383-386。
10. Garyn-Tal, S. (2013). An investment strategy in active ETFs. The Journal of Index Investing, 4(1), 12-22.
11. Goyal, K., Dumancic, S., & Blockeel, H. (2020)。Feature interactions in xgboost。Journal of Artificial Intelligence Research, 48(3), 209-225。
12. Hapfelmeier, A., Hothorn, T., Ulm, K., & Strobl, C. (2014)。A new variable importance measure for random forests with missing data。Statistics and Computing, 24(1), 21-34。
13. Hua, Y., Fang, Y., Yu, F., & Yin, M. (2023)。A factorial importance feature recognition algorithm based on kpca-lightgbm coupling algorithm。Journal of Computational Biology, 30(2), 117-132。
14. Li, Z., & Li, X. (2021)。Intrusion detection method based on genetic algorithm of optimizing lightgbm。IEEE Transactions on Information Forensics and Security, 16, 1234-1245。
15. Liu, Y., & Zhao, H. (2017)。Variable importance-weighted random forests。Quantitative Biology, 5(2), 338-351。
16. Luo, R., Ou, J., Chen, W., Wu, Y., Yan, J., & Liu, S. (2021). A stock selection method for ETF-abstracted-feature based trading strategy. 2021 International Conference on Wavelet Analysis and Pattern Recognition (ICWAPR), 1-6.
17. Parsa, A., Movahedi, A., Taghipour, H., Derrible, S., & Mohammadian, A. (2019)。Toward safer highways, application of xgboost and shap for real-time accident detection and feature analysis。Journal of Transportation Safety, 11(4), 321-338。
18. Paul, J., & Dupont, P. (2015)。Inferring statistically significant features from random forests。Neurocomputing, 150, 471-480。
19. Syafei, R. M., & Efrilianda, D. A. (2023)。Machine learning model using extreme gradient boosting (xgboost) feature importance and light gradient boosting machine (lightgbm) to improve accurate prediction of bankruptcy。Recursive Journal of Informatics, 8(1), 45-58。
20. Stolyarov, D., & Sorokin, V. (2019)。High-dividend strategies in the U.S. market: Effectiveness and improvements。Journal of Financial Strategies, 78-94。
21. Volodin, A., & Borenko, E. (2018)。High-dividend portfolios filtered by financial performance。Journal of Investment Strategies, 10(2), 45-59。
22. Xiang, G., Jun, Y., Zhiyi, H., & Yuzhe, H. (2021)。Feature sorting algorithm based on xgboost and mic combination model。Journal of Machine Learning Research, 18(2), 57-72。
23. Yan, H., Song, Y., Qiu, M., & Akagi, F. (2015)。An empirical analysis of the dog of the dow strategy for the Taiwan stock market。Journal of Economics, Business and Management, 3(3), 435-439。
24. Yang, K., & Chao, W.-P. (2020)。Applying k-means technique and decision tree analysis to predict Taiwan ETF performance。2020 IEEE International Conference on Industrial Engineering and Engineering Management (IEEM), 635-639。
25. Yao, D., Yang, J., Zhan, X., Zhan, X., & Xie, Z. (2015)。A novel random forests-based feature selection method for microarray expression data analysis。International Journal of Data Mining and Bioinformatics, 13(1), 84-101。