We examine the long-term stock performance following dividend initiations and resumptions from 1927 to 1998. We show that postannouncement abnormal returns are significantly positive for equally weighted calendar time portfolios, but become insignificant when the portfolios are value weighted. Moreover, the equally weighted results are not robust across subsamples. We also document postannouncement reductions in the risk factor loadings of underlying stocks. Cross-sectionally, these reductions are negatively related to the contemporaneous price drifts, suggesting the price drifts may be a sample-specific result of chance. Our results underscore the importance of testing for changes in risk loadings in future long-term event studies.