應用機器學習於臺灣股市投資組合框架之研究

文化大學機構典藏 CCUR > 商學院 > 資訊管理學系暨資訊管理研究所 > 博碩士論文 > Item 987654321/48259

jsp.display-item.identifier=請使用永久網址來引用或連結此文件: https://irlib.pccu.edu.tw/handle/987654321/48259

题名:	應用機器學習於臺灣股市投資組合框架之研究 Framework for the Application of Machine Learning in Portfolio Selection of Taiwan Stock Market
作者:	林怡君
贡献者:	資訊管理學系碩士在職專班
关键词:	機器學習集成學習投資組合堆疊法 machine learning ensemble learning portfolio selection stacking
日期:	2020
上传时间:	2020-08-06 09:17:40 (UTC+8)
摘要:	本研究的研究對象為臺灣50指數成分股。研究目的是找出最佳化投資組合，並提供有效資金配置，運用機器學習中的單純貝氏、羅吉斯迴歸、支援向量機、隨機森林，以及極限梯度提升演算法分別建立模型，再使用集成學習(Ensemble Learning)的堆疊法(Stacking)獲得最終的模型，預測股價漲跌趨勢作為挑選股票的依據。最後，結合現代投資組合理論(Modern Portfolio Theory, MPT)與資本資產定價模型(Capital Asset Pricing Model, CAPM)找出最佳投資組合的資金配置，進行投資績效的分析及探討，以協助投資者降低風險、增加報酬率，並做為投資決策的參考。本研究經實證結果顯示，在五種預測模型的比較中，以單純貝氏、羅吉斯迴歸和支援向量機的預測效果較好，預測隔月股價漲跌之準確率平均達58%。比較F值的分數，三者也較其他演算法優，F值達73%。若以集成學習堆疊法建構模型，準確率達59%，及F值達73%，是預測模型中最高的，由此可知，透過集成學習方式可以獲得較優且穩定的預測效果。比較年化報酬率MPT最高達13%與CAPM最高達33%均優於臺灣50指數8%；比較累積報酬率CAPM最高達77%亦較MPT最高達27%為優。 The objective of this study is to find the optimal investment portfolio and provide effective fund allocation of Taiwan 50 index stocks. Using Naïve Bayes, Logistic Re-gression, Support Vector Machine, Random Forest, and eXtreme Gradient Boosting al-gorithms in machine learning to establish models. Then use the Stacking of Ensemble Learning to obtain the final model to predict the stock price trend as the basis for se-lecting stocks. Finally, combine Portfolio Theory and Capital Asset Pricing Model to find the optimal allocation of funds for the investment portfolio, and analyze and dis-cuss investment performance, to help investors to increase returns with low risk, and as a reference for investment decisions. The empirical results of this study show that in the comparison of the five predic-tion models, Naïve Bayes, Logistic Regression, and Support Vector Machine improve-ment are better, and the accuracy rate of predicting the stock price fluctuations every month is 58% in average. Comparing the scores of F value, the aforementioned three are also better than other algorithms, with an F value of 73%. If the model is constructed with Ensemble Learning Stacking, we observed the accuracy rate of 59%, and the F value of 73%, which is the highest among all of the prediction models. Therefore, we can conclude that a better and stable prediction result can be obtained through Ensemble Learning. The experimental result also shows that the return on investment (ROI) of MPT has 13% and CAPM has 33%, which both are better than Taiwan 50 index’s 8%. Additionally, the ROI of CAPM is also higher than that of MPT.
显示于类别:	[資訊管理學系暨資訊管理研究所 ] 博碩士論文

文件中的档案:

档案	描述	大小	格式	浏览次数
index.html		0Kb	HTML	270	检视/开启

检视Licence

在CCUR中所有的数据项都受到原著作权保护.

数据加载中.....