請用此 Handle URI 來引用此文件:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/83367
標題: | 財務報告分析及盈餘預測 – 機器學習模型之應用 Financial Report Analysis and Earnings Forecast – Application of Machine Learning Models |
作者: | Han Lee 李涵 |
指導教授: | 蔡彥卿(Yann-Ching Tsai) 蔡彥卿(Yann-Ching Tsai | yanntsai@ntu.edu.tw | ), |
關鍵字: | 預測每股盈餘,機器學習,深度學習,財務報表分析,投資建議, Forecasting Earnings Per Share,Machine Learning,Deep Learning,Financial Statement Analysis,Investment Suggestion, |
出版年 : | 2022 |
學位: | 碩士 |
摘要: | 本論文以公司過去資產負債表、綜合損益表和現金流量表中的會計科目資訊以及過往的每股盈餘和稀釋每股盈餘資料預測公司下一年年末的每股盈餘或稀釋每股盈餘,並建立了144個機器學習模型以及12個深度學習模型。 分析各模型的預測結果後,我獲得了以下四點結論。第一點為122個機器學習模型和全部的深度學習模型皆優於基準預測值,顯見本論文建立的模型具有一定的預測能力。第二點為模型認為預測稀釋每股盈餘是個較簡單的任務。第三點為不同種類的機器學習模型對於自變數欄位和資料特徵處理方法的偏好不同。最後一點則是訓練資料不充足以及市場和產業變化會導致模型預測較不準確,但透過增添更多歷史資料和與公司前景預測相關之自變數欄位至資料集能提升模型的預測表現。 This essay used accounting information from companies' past balance sheet, comprehensive income statement and cash flow statement, as well as historical earnings per share (EPS) and diluted EPS data to forecast companies' EPS or diluted EPS for the next year, and established 144 machine learning models and 12 deep learning models. After analyzing the results of each model, I came to the following four conclusions. First, 122 machine learning models and all deep learning models are better than benchmark prediction values, which shows that models built in this essay have predictive ability. Secondly, models consider forecasting diluted EPS to be a simpler task. Thirdly, different types of machine learning models have different preferences for independent variable fields and data feature processing methods. Lastly, insufficient training data and changes of markets and industries can lead to less accurate model predictions, but performance of models can be improved by adding more diverse data and future-related independent variable fields to dataset. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/83367 |
DOI: | 10.6342/NTU202203124 |
全文授權: | 未授權 |
顯示於系所單位: | 會計學系 |
文件中的檔案:
檔案 | 大小 | 格式 | |
---|---|---|---|
U0001-0409202201280500.pdf 目前未授權公開取用 | 5.91 MB | Adobe PDF |
系統中的文件,除了特別指名其著作權條款之外,均受到著作權保護,並且保留所有的權利。