這篇研究在探討將藤耦合(vine copula)與網絡方法結合應用於投資組合優化。我們首先使用de-GARCH技術消除每個財務時間序列中的自相關、條件異方差性和波動聚集等內在特徵。接著，我們計算多變量de-GARCH序列的相似矩陣，以構建整體的最小生成樹（MST），這有助於識別適合投資組合的股票。隨後，我們為選定的股票構建局部最小生成樹（LMST），並基於 LMST 使用各種藤耦合模型來描述選定股票的聯合分佈。然後，使用這種基於聯結網路的分佈來設定投資組合中所選股票的權重。我們採用 2019 年至 2023 年 S&P100 指數的成分股，透過移動視窗的架構進行實證研究。數值結果表明，與競爭對手相比，所提出的方法獲得令人滿意的累積報酬。;This study explores the application of vine copulas combined with network methods for portfolio optimization. We begin by eliminating inherent features such as autocorrelation, conditional heteroscedasticity, and volatility clustering in each financial time series using the de-GARCH technique. We then calculate the similarity matrix of the multivariate de-GARCH series to construct the global Minimum Spanning Tree (MST), which helps identify suitable stocks for the portfolio. Subsequently, we build the local MST (LMST) for the selected stocks and employ various vine copulas based on the LMST to model the joint distribution of the selected stocks. This copula network-based distribution is then used for setting the weights of the selected stocks in the portfolio. Our empirical investigation involves the component stocks of the S&P100 index from 2019 to 2023, using a rolling-window framework. The numerical results demonstrate that the proposed method yields satisfactory cumulative returns compared to competitors.
