Goyal and Welch (2008) 提及文獻上常見的財務與經濟變數在運用單變量迴歸模型下，對於股價溢酬之向外預測表現不盡理想。本文根據文獻上常見的11個解釋變數，採用Lasso模型並運用遞迴方式 (Recursive)進行股價溢酬的向外預測，其預測期間包含1、12、36、60與120個月。本文實證結果發現Lasso模型針對每一次的樣本內估計所挑選出的解釋變數不盡相同，同時，隨著預測期間越長，Lasso模型所挑選出的解釋變數越多，代表財務與經濟變數可能更適合解釋長期股價溢酬的變動。其此，在最近的估計期間，Lasso模型認為財務與經濟解釋變數對於股價溢酬皆不存在顯著的影響。最後，不論短期或長期的預測期間，Lasso模型對於股價溢酬的向外預測表現皆優於傳統的OLS模型。 Goyal and Welch (2008) argued that the financial variables with the univariate regression could not beat the random walk with drift in out-of-sample forecasting of stock premiums. Based on the 11 explanatory variables commonly found in the existed literature, this paper adopts the Lasso approach with the recursive scheme to predict the stock premium about 1, 12, 36, 60 and 120 months ahead. Firstly, the empirical results show that the Lasso model has different significantly independent variables for each in-sample estimation. Secondly, the longer the prediction period, the more explanatory variables selected by the Lasso model which may represent financial variables are suitable for explaining the changes of long-term stock premiums. Thirdly, during the most recent sample period, we argued that financial variables did not have a significant impact on stock premiums. Finally, regardless of the short- or long-term forecast period, the Lasso model outperforms the traditional OLS model for the out-of-sample performance of stock premiums.
