階層式資產預選方法對投資組合績效之影響 -以 NASDAQ 100 成分股為例;Hierarchical Asset Preselection Approach on Portfolio Performance: Evidence from NASDAQ-100 Constituents

NCUIR > School of Management at National Central University > Graduate Institute of Information Management > Electronic Thesis & Dissertation > Item 987654321/98369

Please use this identifier to cite or link to this item: https://ir.lib.ncu.edu.tw/handle/987654321/98369

Title:	階層式資產預選方法對投資組合績效之影響 -以 NASDAQ 100 成分股為例;Hierarchical Asset Preselection Approach on Portfolio Performance: Evidence from NASDAQ-100 Constituents
Authors:	蔡容平;Tsai, Rung-Ping
Contributors:	資訊管理學系
Keywords:	資產預選;深度學習;股價預測;情感分析;投資組合最佳化;夏普率;Asset pre-selection;Deep Learning;Stock price prediction;Sentiment analysis;Portfolio optimization;Sharpe Ratio
Date:	2025-07-25
Issue Date:	2025-10-17 12:41:42 (UTC+8)
Publisher:	國立中央大學
Abstract:	過往投資組合最佳化問題中，透過有效的資金配置使投資目標最佳化，近期研究開始著重事前的資產預選加強投資效益。然而，過去研究較少整合多種類型的股票資料，且尚未有研究探討預選架構設計對投資組合績效的影響。因此，本研究將採用股票之基本面、技術面及消息面三種類型的資料，依據資料發佈的週期長短，設計出兩種標的過濾器，並疊加兩種過濾器建構出階層式資產預選方法，評估每月挑選出的投資組合績效。本研究提出新的階層式資產預選方法，結合社群媒體資料與技術指標資料預測股價並計算出報酬率作為標的篩選條件，建構出短期指標過濾器；採用公司過去基本面表現作為標的篩選條件，建構出長期指標過濾器，實驗設計上，選用Nasdaq100中的成分股，蒐集美國社群平台Reddit貼文資料以及yfinance與alpha_vantage之股價及財報資料。實驗中，首先，以RNN、LSTM及GRU預測股價以選定短期指標過濾器之最佳模型；接著，建構兩種階層式資產預選方法，分別為LS-preselector和SL-preselector，篩選具有潛力之投資組合，最後，採用GA及PSO執行資金配置，並於2023年3月至12月期間進行股市模擬回測。實驗結果顯示，LS-preselector+PSO策略有最佳的投資績效表現，與隨機選股和單層過濾器相比，本研究提出的兩階層資產預選架構有效能提升投資績效表現。 ;In traditional portfolio optimization, investment performance has typically been achieved through efficient capital allocation. Recently, however, increasing attention has been given to prior asset preselection to further improve outcomes. Despite this trend, limited research has explored the integration of multiple stock data types or examined how different preselection framework designs affect portfolio performance. To address this gap, this study proposes a hierarchical asset preselection method that integrates three types of data: fundamental, technical, and sentiment. Based on their release frequencies, two filters are designed and combined into a layered screening framework. Portfolio performance is evaluated monthly to assess the effectiveness of this approach. Specifically, sentiment data from social media and technical indicators are used to predict stock prices and compute short-term returns, forming a short-term filter. Historical fundamental metrics form the long-term filter. The experiment focuses on constituent stocks of the Nasdaq-100 index, using Reddit posts and data from Yahoo Finance and Alpha Vantage. RNN, LSTM, and GRU models are used to select the optimal model for the short-term filter. Two hierarchical filtering sequences—LS-preselector (long-term first) and SL-preselector (short-term first)—are applied, followed by capital allocation using Genetic Algorithm (GA) and Particle Swarm Optimization (PSO). Backtests from March to December 2023 show that the LS-preselector with PSO strategy achieves the best performance outperforming single-layer filters.
Appears in Collections:	[Graduate Institute of Information Management] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	30	View/Open

社群 sharing

Loading...