Fama (1970) 一文提出效率市場假說，認為有效率的資本市場沒有資產錯誤定價獲利的機會，然而這與現實情況有所不同，過去諸多文獻探討股票市場屬於何種效率市場，但其文獻多使用單一變數，本文同時使用總體經濟面、市場資金面與股價技術面變數為預測因子預測臺灣股票市場超額報酬率。 本文實證結果發現，被篩選出最能有效預測臺灣股市超額報酬之變數以臺灣股票市場資金面與股價技術面預測因子為主。整體而言，預測模型中簡單組合法預測表現最為傑出，接著為組合彈性網路法與平均預測法，其中組合彈性網路法與簡單組合法之變異度表現最佳、預測累積年化報酬率以簡單組合法與主成分分析法最高。 本文預測結果，機器學習預測模型可有效捕捉股票市場超額報酬，然而當前所擁有資訊隨著時間經過不斷進行校正，使得向前預測二個月與三個月之預測報酬率最高，然而隨著預測期數越長，資訊會逐漸反應於股價上，累積超額報酬率越低，當前所擁有資訊僅能於短期內獲取額外的超額報酬。;Fama (1970) proposed the efficient market hypothesis, arguing that an efficient capital market has no opportunity to gain any profit from mispricing. However, in the past, many literatures have discussed what kind of efficient market the stock market belongs to, but many literatures used the single part of variables, we are different from the past, simultaneously uses the macroeconomic, market capital, and technical variables as predictors to predict the excess return rate of the Taiwan stock market.
The empirical results found that the selected variables that are most effective in predicting excess returns in the Taiwan stock market are mainly cash flow and technical variables. Overall, the simple combination method has the best forecasting performance among forecasting models, then the combination elastic net method and the average forecasting method, combined elastic network method especially had the best forecasting performance in recession.
The most important results this article observes is that the information is continuously corrected over time, making the two-month and three-month forecast returns the highest. However, the longer ahead forecast period we predict, the more information will gradually be reflected in the stock price. In other words, the information just can only obtain additional excess returns in the short term.
