This paper examines whether investor sentiment expressed in online user generated content ( UGC) in the stock message boards of Easymoney． com has a predictive power for Chinese stock market． Naive Bayes classifier is used to obtain the expectation of every message，and then a new indicator of investor sentiment is put for ward that combines the idea of bullishness and investor attention． Granger causality test，instantaneous Granger causality test and intertemporal regression analysis are applied to study the predictability of investor sentiment for stock returns，volatility，and trading volume． It is found that investor sentiment has no predictable power for returns，volatility，or trading volume; while investor sentiment does have a significant positive effect on the contemporaneous stock price，and disagreement of investor sentiment can affect contemporaneous trading volume． Inventor sentiment using the messages before market opening has the predictive power of open price，and inventor sentiment from the trading hours is more valuable for closing price and trading volume． Moreover，stock returns are the Granger cause of investor sentiment． Our results can help to understand the behavior of investors and their effects on stock market．