本文旨在透過文字探勘的方式，探討不同市場狀態下新聞報導與台灣五十指數報酬率間的關聯性。研究結果顯示，若不區分市場狀態，則新聞情緒對於指數報酬不具預測力。但區分市場狀態後，會因不同市場狀態下投資人信心程度的不同，對好、壞消息的關注度有所差異，使得新聞報導所隱含的情緒對於指數報酬具有預測力。在牛市下，投資人因投資表現較好而產生過度自信，會對新聞中隱含之正面情緒關注度較高，反應也更為迅速，而對於新聞中隱含之負面情緒的反應則較為緩慢，使得負面情緒的新聞在牛市期間對於股票報酬具有預測力；相反地，在熊市期間，投資人因信心較為低落，對於負面新聞情緒的關注度較高，反應也較為快速，而對正面情緒的消息的反應則較為緩慢，使其在熊市期間具有預測力。 我也將媒體曝光效果和新聞情緒效果分開討論。發現在熊市下考慮新聞情緒效果後，會使得媒體曝光效果變得不顯著。且在控制媒體曝光效果後，新聞情緒對於指數報酬率仍然有預測能力。整體而言，無論處於牛市或熊市下，壞消息的傳遞速度皆較好消息來得慢。;The thesis studies the relationship between stock returns and sentiment in news content through the text mining. The results show that news sentiment cannot predict index return in the full period within different horizon, but may predict weekly return in the bear and bull markets, respectively. The most intuitive interpretation is that investors have overconfidence in bull market that would make them pay more attention on good news and neglect the bad news. This would enhance the predictability of bad news under the bull market. Similarly, when it comes with bear market, investors are lack of confidence and pay more attention on bad news instead of good news, which leads to the significant predictability of good news under bear market. I also distinguish the coverage effect from sentiment effect and show that after controlling for the media coverage, negative news have much pronounced influence on index return than positive news, in both bull and bear markets.
