Return Predictability Conditional on the Characteristics of Information Signals

This dissertation examines whether simultaneously conditioning on themultidimensional characteristics of information signals can help predict returnsthat are of economic significance. We use large price changes, publicannouncements, and large volume increases to proxy for the magnitude,dissemination, and precision of information signals. Abnormal returns followinglarge price change events are found to be unimportant. As we condition on othercharacteristics of information signals, the abnormal returns become large.Large price change events accompanied by both a public announcement and anincrease in volume have a 20-day abnormal return of almost 2% for positiveevents and -1.68% for negative events. The type of news provides furtherrefinement. If the news relates to earnings announcements, management earningsforecasts, or analyst recommendations then the 20-day abnormal returns becomesmuch larger: ranging from 3% to 4% for positive events and about -2.25% fornegative events. For these news events, we also find that the underreaction isgreater for positive (negative) event firms that underperformed (overperformed)the market in the prior period, earning 20-day post-event abnormal returns of4.85% (-3.50%). This evidence is consistent with the Barberis, Shleifer, andVishny (1998) model of investor sentiment that suggests that investors are slowto change their beliefs. The evidence from our sample does not provide muchsupport for strategic trading models under information asymmetry. Finally, anout-of-sample trading strategy generates 20-day post-event statisticallysignificant abnormal return of 2.18% for positive events and -2.40% for negativeevents. Net of transaction costs, the abnormal returns are a statisticallysignificant 1.04% for positive events and a statistically significant -1.51% fornegative events.