本研究以2006年至2019年台灣上市上櫃之非金融業公司為研究樣本，從文獻和實務界選取影響股價之變數用以預測股票報酬率。先使用2006至2017年的季度蹤橫追蹤資料估計係數並建立預測模型，再以2018與2019年的季度資料進行樣本外預測能力之檢驗，觀察Theil不等性係數及其他預測誤差之數值來判斷模式之預測能力，以及R平方檢視模型之解釋能力。預測模型分成基本迴歸式 (包含所有自變數)、逐步迴歸分析以及動態報酬模型 (加入報酬率前一期為解釋變數) 等三種。實證結果發現：基本迴歸式與逐步迴歸分析之預測能力並不佳，因為Theil不等性係數均在0.5以上，解釋應變數之變異的能力約在20%至33%之間。動態模型方面，解釋能力比基本迴歸式和逐步迴歸式之解釋力有所提升，解釋能力約提升5%左右，故自變數中考慮前一期報酬率之模型是較佳選擇。 This study uses non-financial companies listed in Taiwan from 2006 to 2019 as the research sample, and selects variables that affect stock prices from literature and practice to predict stock returns. First use the quarterly tracking data from 2006 to 2017 to track the data to estimate the coefficients and establish a prediction model, and then use the quarterly data of 2018 and 2019 to test the out-of-sample forecasting ability, and observe the Theil inequality coefficient and other prediction error values to judge The predictive power of the model and the explanatory power of the R-squared inspection model. The prediction model is divided into three types: basic regression (including all independent variables), stepwise regression analysis, and dynamic compensation model (the explanatory variable before the return rate is added). The empirical results found that the predictive ability of basic regression and stepwise regression analysis is not good, because Theil's inequality coefficient is above 0.5, and the ability to explain the variation of strain number is about 20% to 33%. In terms of dynamic models, the explanatory power is improved compared to the basic regression and stepwise regression. The explanatory power is improved by about 5%. Therefore, the model that considers the previous return rate in the independent variable is the better choice.
