We use a panel of more than 100,000 investor accounts in US stocks over the period 1991-1995 to construct an investor-based measure of dispersion of opinion, unlike the analyst based measure used in the literature. We use this measure to test two competing hypotheses: the sidelined investors...