本研究利用逐筆委託單,以Patterson and Sharma(2005) 所提出的Bootstrapped runs test 來計算群聚指標,並利用揭示檔及成交單資料,以Hu(2006) 之方法估計股票報酬波動度中所含之雜訊,觀察台灣股票市場中散戶、外資、投信基金以及自營商群聚行為與股票報酬波動度之關係以及其中雜訊的變化,並推論何類投資人群聚行為屬於根據資訊的理性群聚,而何種投資人屬於盲從之非理性群聚。 實證結果發現,在整日觀察下,各類投資人群聚行為對波動度並無顯著影響,而投信則是會使雜訊降低。在日內區間,散戶之群聚行為皆會使波動度提高,開盤時投信及外資群聚行為會使雜訊降低,表示其為根據隔夜資訊所作之理性群聚;而盤中及收盤,除了投信外其餘三類投資人群聚行為均會使雜訊增加,散戶在收盤時更是明顯,此可以資訊瀑布流來解釋,此時群聚行為多屬於非理性之群聚。因此綜合來看,投信多屬於根據資訊之理性群聚者,散戶則多屬於盲從之非理性群聚者。 This study Based on intraday order book data and transaction data from Taiwan Stock Exchange(TSE), adopted the herding measurement of Patterson and Sharma(2005) and the estimation of noise by Hu(2006). Then observed the relationship between herding behavior of different type of investors and noise in Taiwan stock market, and tried to infer that investors’ herding behavior was rational or irrational. The empirical result showed that investors’ herding behavior had no significant effect on volatility, but herding behavior by Investment Trust decreased noise in daily pattern. In intraday pattern, herding by Investment Trust and Foreign Institutional Investor decreased noise but individuals increased noise at market open. At market close, herding behavior except Investment Trust increased noise especially individuals, and this can be explained by information cascade. According to the result above, herding by Investment Trust was rational, but herding by individuals was irrational.