過去在討論股價是否存在均數回復現象,一開始是為了提出市場是無效率的證據,主要論點是依據Summers(1986) 「股價為隨機漫步及定態序列所組成的。」而均數回復現象是由於短期的股價的過度反應,透過投資者的反向投資策略促使股價回到真正的價值。但是,過去文獻在討論均數回復時,未考慮交易成本與不收斂區間。 本文使用Obstfeld and Taylor(1997)提出的Band-TAR模型為研究台灣、韓國、新加坡、香港、泰國、美國、日本的大盤指數是否存在均數回復現象,研究期間為1971年至2007年底的月資料,從模型估計結果發現台灣、韓國、新加坡、泰國的大盤指數同時存在均數回復現象與非套利區間,此現象是使用AR(1)模型偵測不到的。此外,台灣股市存在均數回復,與丁國玄(1996)和洪淑瑜(2001)結論相吻合。 當股價的行為存在均數回復現象,代表股價具有某種程度的可預測性,本文只針對存在均數回復現象的大盤指數進行樣本外預測與評估。本文選用1971年到1999年底為樣本內期間,而2000年到2007年底為樣本外預測期間,本文針對沒有截距項的隨機漫步模型、AR(1)模型,與Band-TAR模型的預測進行比較,採用Harvey,Leybourne,and Newbold(1997)提出的Modified Diebold-Mariano檢定與Clark and West(2006)提出的巢式(nested)檢定,來評估模型預測的表現。在顯著水準10%下,Band-TAR模型在預測台灣、韓國大盤指數時,有顯著優於線性AR(1)模型。預測日本1到6個月、預測台灣2到5個月,使用Band-TAR模型有顯著優於隨機漫步模型。
Detecting mean reversion is in order to evidence market inefficiency. Many studies about mean reversion are in terms of Summers’ (1986) idea-stock price is composed of random walk and stationary components. They suggest that mean reversion is due to contrarian investment strategy while stock price’s overreaction temporarily. However, they don’t import transaction cost and bands of inaction for mean reversion. This article applies the Band-TAR model by Obstfeld and Taylor(1997) to study mean reversion in stock markets over the 1971-2007 period -including Taiwan, Korea, Singapore, Hong Kong, United States, Japan, Thailand. Using Band-TAR model, we can find mean reversion and bands of inaction in Taiwan, Korea, Singapore, and Thailand. This result isn’t found in AR (1) model. Further study for predictability of stock price, we compare Band-TAR, AR (1) and Random walk without draft model in predictive accuracy. The in-sample period is 1971 to 1999 and out-of-sample period is 2000 to 2007. We utilize “Modified Diebold-Mariano test” by Harvey,Leybourne and Newbold (1997) and “nested model test” by Clark and West (2006).