Skip navigation

DSpace JSPUI

DSpace preserves and enables easy and open access to all types of digital content including text, images, moving images, mpegs and data sets

Learn More
DSpace logo
English
中文
  • Browse
    • Communities
      & Collections
    • Publication Year
    • Author
    • Title
    • Subject
    • Advisor
  • Search TDR
  • Rights Q&A
    • My Page
    • Receive email
      updates
    • Edit Profile
  1. NTU Theses and Dissertations Repository
  2. 管理學院
  3. 會計學系
Please use this identifier to cite or link to this item: http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/98074
Title: 透過機器學習與深度學習模型預測上市櫃公司月營收與套利-考量三因子風險溢酬
Forecasting Monthly Revenue and Arbitrage Strategies for Listed and OTC companies Using Machine Learning and Deep Learning Models - Considering the Fama-French Three-Factor Model
Authors: 陳洛霆
Lo-Ting Chen
Advisor: 蔡彥卿
Yann-Ching Tsai
Co-Advisor: 劉心才
Hsin-Tsai Liu
Keyword: 月營收預測,機器學習,深度學習,套利,投資策略,三因子模型,
Monthly Revenue Forecasting,Machine Learning,Deep Learning,Arbitrage,Investment Strategy,Fama-French Three-Factor Model,
Publication Year : 2025
Degree: 碩士
Abstract: 本研究旨在運用機器學習與深度學習模型,預測臺灣上市櫃公司之月營收,並評估建構相關套利策略之可行性與績效,同時考量資料前處理方法及市場環境變動(如 COVID-19 疫情)的影響。本研究評估隨機森林、極限梯度提升、多層感知機與 Transformer 四種模型,搭配兩種資料結構(年對年與月對月)及兩種特徵工程方法(原始月營收標準化法與去趨勢月營收標準化法),共 16 組交叉組合對預測結果的影響。基於預測結果,建構三種套利策略:未預期營收相對比例排序、未預期營收標準化、連續未預期營收,並評估其報酬表現。最後,運用 Fama-French 三因子模型檢驗策略是否具備風險調整後超額報酬(Alpha)。

實證結果顯示以下五個主要結論:第一,在不同資料結構的比較中,年對年結構普遍優於月對月結構,能提供較佳的預測穩定性與準確性。第二,去趨勢化處理在多數情況下有助於提升模型預測準確度,其中以多層感知機與 Transformer 模型受益最為顯著。第三,若去趨勢化訓練資料主要來自於受疫情擾動的時段,則樹模型(如隨機森林與極限梯度提升)之預測效果反而因基準值錯估而遭削弱,但對神經網路及深度學習模型而言,去趨勢化仍具正向效果。第四,在各種套利策略中,未預期營收相對比例排序具備最佳的穩定性,於不同模型與資料前處理組合下均能產生正向報酬;相比之下,未預期營收標準化與連續未預期營收策略則潛力較高但表現波動較大,易受到嚴格閾值設定的影響。第五,經由 Fama-French 三因子模型進行風險調整後,多數模型與策略組合均能產生統計上顯著且正向的風險調整後超額報酬(Alpha)。結果支持機器學習與深度學習模型於臺灣資本市場中,透過預測月營收並建構相應套利策略,具備擷取風險調整後超額報酬的可行性與應用潛力。
This study uses machine learning and deep learning models—Random Forest, XGBoost, MLP, and Transformer—to forecast monthly revenues of Taiwan’s listed and OTC companies, evaluating the effectiveness of three arbitrage strategies based on prediction errors. Sixteen model setups combine year-over-year and month-over-month data with both raw and detrended revenues. Performance is measured by risk-adjusted excess returns using the Fama-French Three-Factor Model.

Results show year-over-year structures generally improve prediction accuracy. Detrending further benefits MLP and Transformer, but can weaken tree-based models during COVID periods. Among strategies, Relative Unexpected Revenue is the most stable, while others provide higher but more volatile returns. Most approaches deliver statistically significant positive alphas, highlighting the potential for machine learning-driven arbitrage in Taiwan’s capital market.
URI: http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/98074
DOI: 10.6342/NTU202501462
Fulltext Rights: 未授權
metadata.dc.date.embargo-lift: N/A
Appears in Collections:會計學系

Files in This Item:
File SizeFormat 
ntu-113-2.pdf
  Restricted Access
5.06 MBAdobe PDF
Show full item record


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.

社群連結
聯絡資訊
10617臺北市大安區羅斯福路四段1號
No.1 Sec.4, Roosevelt Rd., Taipei, Taiwan, R.O.C. 106
Tel: (02)33662353
Email: ntuetds@ntu.edu.tw
意見箱
相關連結
館藏目錄
國內圖書館整合查詢 MetaCat
臺大學術典藏 NTU Scholars
臺大圖書館數位典藏館
本站聲明
© NTU Library All Rights Reserved