透過您的圖書館登入
IP:3.21.248.119
  • 學位論文

基於公司財報及產業表現基本面分析與集成模型多任務遷移學習之股價預測

Decision Support for Stock Investment with Ensemble-based Multitasking Transfer Learning centric to Fundamental Analysis on Financial Statements and Industry Status

指導教授 : 姜國輝 劉文卿
本文將於2026/07/25開放下載。若您希望在開放下載時收到通知,可將文章加入收藏

摘要


隨著資訊科技的蓬勃發展,諸多科技技術與創新應用蜂擁而出,而在應用機器學習對個別股票的短期價格進行技術分析和情緒分析的大潮中,提供股票趨勢長期預測的基本面分析仍然是機器學習尚未開發的領域。雖然基於財務報表的基本面分析能夠解決股票未來表現的複雜性,做出基於價值的股票交易策略,但財務報表中會計項目的高維度特性與不確定的勾稽關系阻礙了機器學習的應用。為了解決這個問題,本研究用不同的特徵工程技術準備了數據集,以提高預測模型的性能。 此外,對某一特定公司的盈利能力於潛力的分析通常是獨立的,不適用於其他公司。為此,我們開發了一個兩層轉移學習模型,以實現所獲知識的可轉移性並提高訓練的效率。最後,GRU被用來將獲得的特徵轉化為股票的比較內在價值,用於評估功能。利用台灣半導體行業11年來的上市公司財務報表和產業類股指數,實現了基於特徵的轉移學習和GRU的統一框架,用於基於財務報表和工業狀況的基本面分析,可以覆蓋決策任務的的三個階段,並在效率、準確性、精確性和轉移性及回報方面進行了評價。

並列摘要


In the stride of applying machine learning for short-term price prediction of individual stocks with technical analysis and sentiment analysis, fundamental analysis which provides the long-term prediction of stock trends remains unexplored territory for machine learning. Whilst fundamental analysis based on financial statements is capable of resolving the complexity of future performance of the stocks and leading to the value-based stock trading strategy, the high dimensionality and undetermined collinearity of accounting items in financial statements hinder the application of machine learning. To solve this problem, this research prepared datasets with different feature engineering techniques to improve the performance of the predictive model. Further, the machine learning analysis of profitability and potential for a specific company are usually unique and not applicable to the others. For this, a two-layer Transfer Learning model is developed for the transferability of gained knowledge and to increase the efficiency of the training. At last, GRU is used to transform the gained features into the comparative intrinsic value of the stock for the evaluation function. With the financial statements of the list companies and industrial stock indexes of the Taiwanese semiconductor industry and industrial stock index over 11 years, a unified framework of feature-based surrogate function, transfer learning, and GRU for fundamental analysis based on financial statements and industrial status which can cover the tasks of all of the three phases of decision-making was realized and evaluated with respect to efficiency, accuracy, precision, transferability and return.

參考文獻


張惠珊. (2016). 財務危機預警分析-以財務報表、公司治理、商品品質三大指標探討. (碩士). 國立中正大學, 嘉義縣. Retrieved from https://hdl.handle.net/11296/4t78jn
林佑南. (2020). 運用財務比率分析企業財務危機之研究. (碩士). 淡江大學, 新北市. Retrieved from https://hdl.handle.net/11296/vs3x65
林宜隆. (1989). 專家系統之探討. 資訊縮影管理(14&15), 83-97.
林怡君. (2020). 應用機器學習於臺灣股市投資組合框架之研究. (碩士). 中國文化大學, 台北市. Retrieved from https://hdl.handle.net/11296/h9tqa4
甄典蕙. (2015). 財務報表舞弊偵測模型之建立-以中國上市公司為例. (碩士). 國立政治大學, 台北市. Retrieved from https://hdl.handle.net/11296/f82cs6

延伸閱讀