本文嘗試建立多種情緒指標，基於報酬、實現波動率、交易量以及流動性等股市特徵，建構相應的情緒因子，觀察這些變數是否存在屬於各自的情緒。利用Baker & Wurgler (2006) 所提供的資料，樣本期間為1965年9月至2022年5月，通過三種不同的因子降維方法萃取各個目標變數的情緒因子。研究發現，以報酬為目標的情緒因子對該報酬具有顯著的解釋力，同樣的結果也適用於其他變數構建的情緒因子。此外，使用多個情緒因子進行預測時，進一步發現變數可能會受到不同情緒因子之間的相互影響，這些情緒因子的表現均優於通過主成分分析 (PCA) 構建的情緒指標。本文還根據各個變數在三種方法下選擇出表現最好的情緒因子，並解構了在不同景氣週期下情緒因子的表現差異。最終，本文展示了投資者在實際應用情緒因子的經濟價值，顯示以目標變數構建的情緒指標進行預測所產生的投資組合能夠獲得更高的確定等值報酬。;This paper attempts to establish various sentiment indices based on stock market characteristics such as returns, realized volatility, trading volume, and liquidity, to observe whether these variables reflect unique sentiment features. Using the data provided by Baker & Wurgler (2006) for the period from September 1965 to May 2022, we extract sentiment factors for each target variable through three different dimensionality reduction methods. The study finds that the sentiment factor targeting returns has significant explanatory power for those returns, and similar results apply to sentiment factors constructed for other variables. Additionally, when using multiple sentiment factors for prediction, it is further observed that the variables may be mutually influenced by different sentiment factors. These sentiment factors outperform those constructed using principal component analysis (PCA). The paper also identifies the best-performing sentiment factors for each variable under the three methods and deconstructs the performance differences of sentiment factors across different business cycles. Finally, this paper demonstrates the economic value of practically applying sentiment factors, showing that portfolios predicted using sentiment indices based on target variables achieve higher certainty equivalent return.
