本文主要研究移動平均在台股市場預測能力之實證，探討市場不同期間框架的股價移動平均（Moving Averages, MA）之預測能力，利用陳柔君等人（2022）的研究方法，使用滾動窗口之Fama-MacBeth迴歸估計MA與股票報酬的關係，結合主成份分析法，預測樣本外股票報酬（the out-of-sample forecasted return strategies, OSFRS），並建構投資策略，藉此獲得優異績效。研究結果：第一、MA的聯合模型對於股票報酬率的解釋力較佳，代表Fama-MacBeth迴歸的結果容易受到多種現象（如共線性）干擾。第二、模型中若納入預測模型加入過多無解釋力的自變數，可能會增加雜訊和預測誤差，故非越複雜越好。第三、比較其他模型的投資方法，以包含1至24個月MA及三種股票特徵建構之樣本外預測投資策略有較佳的績效。 This paper mainly studies the empirical evidence of the forecasting ability of moving averages in the Taiwan stock market, and discusses the forecasting ability of moving averages (MA) of stock prices in different periods of the market. Using the research method of Chen Roujun et al. (2022), using the rolling window of Fama- MacBeth regression estimates the relationship between MA and stock returns, combined with principal component analysis, predicts the out-of-sample forecasted return strategies (OSFRS), and constructs investment strategies to achieve excellent performance. Research results: First, the MA joint model has a better explanatory power for the stock return rate, which means that the results of the Fama-MacBeth regression are susceptible to interference from various phenomena (such as collinearity). Second, if the prediction model is included in the model and too many independent variables without explanatory power are added, noise and prediction errors may increase, so the more complex the better. Third, comparing the investment methods of other models, the out-of-sample predictive investment strategy that includes 1-24 month MA and three stock feature constructions has better performance.
