利用變異數比率檢定台灣股票市場效率性之研究

Abstract

本研究採用Lo & MacKinlay(1988)與Chow & Denning(1993)的多重變異數 比率檢定對台灣股票市場是否符合隨機漫步進行研究，檢定台灣股票市場 的股價指數變動率前後期之間是否具有相依性或關連性。因此可視為弱式 效率的檢定。在研究方法上，Lo & MacKinlay 的變異數比率檢定改進了 傳統方法在時間序列呈現變異數之異質狀況下產生較大預估誤差的缺點， 而Chow & Denning 的多重檢定法則有效界定了多重檢定的聯合信賴區間 。研究的資料乃取自教育部AREMOS資料庫台灣證券交易所各類股價指數作 為分析之標的，期間為1985年至1995年。資料取樣區間為週，並以1990年 股市的崩盤做一分界線，探討市場之股價產生行為有無顯著的改變。本研 究的實證結果有幾點值得注意：1.就全期間及前半段期間(1985-1989)而 言，變異數檢定結果否定了台灣股票市場加權股價指數報酬率為隨機漫步 的假定；但對後半段期間(1991-1995)，則無法拒絕報酬率為隨機漫步的 假定。2.就全期間及前半段期間而言，股價報酬率的一階自身相關係數為 正相關；但對後半段期間，則呈現負相關。此似乎透露著早期較小的漲跌 幅度限制可能造成的正相關連性，致後期時已減輕影響，取而代之的可能 是投資者的過度反應造成之均數回復效果。 This study utilizes the multiple variance ratio test as developed in Lo & MacKinlay(1988) and Chow & Denning(1993) to examine the Taiwan stock market.Weekly index returns are derived from the composite value weighted priceindex of Taiwan Stock Exchange recorded in the AREMOS tape. The time periodcovers eleven years from 1985 till 1995. This study also devides the period into two subperiod of 1985-1989 and 1991-1995 respectively to investigate theimpact of the market crash in 1990 on the stock return generating process.Several results are worthy of noting. First, for the overall period and the first half period, the random walk model is rejected for most time lags. How-ever, for the second half period, the test fails to reject the random walkmodel for all lever of time lags. Second, the 1-st order autocorrelations as implied by the variance ratio in the case of two-week-lag suggests the thereis positive autocorrelation for weekly holding period returns for the overallperiod and the first half period. However, the autocorrelation become negativewhen the data of the second half period are used. This seemingly puzzling results maybe interpreted that the positive correlation caused by the problemof smaller price change limits might have declined in the second half periodwhile the mean reverting process caused by the overreaction of the investorsmight have set in.