July 13, 2011
This dissertation comprises of four chapters. The first chapter reviews the literature on the relationship between differences of opinion and stock returns, with special attention to the use dispersion of analysts’ earnings per share forecasts in asset pricing. As reviewed, the literature is split on the relationship between forecast dispersion and stock returns both in terms of the direction of the relationship and in terms of the role of forecast dispersion. Therefore, to understand the causes of the dispersion-return relationship, I conduct detail analyses of the dispersion anomaly. The second chapter sheds a new light on the empirical relationship between forecast dispersion and stock returns by examining whether the relationship is robust to different dispersion measures and whether the dispersion-return relationship is related to other well-known financial anomalies. My results strongly suggest that contemporaneous dispersion is negatively correlated with future stock returns. Moreover, the dispersion-return relationship is most pronounced in smallest market capitalization stocks and is robust across different measures of forecast dispersion. Notably, my results show that the dispersion anomaly is not explained by previously documented phenomena such as accruals quality, asset growth, capital investment underperformance, and equity issue anomalies. To understand more about the dispersion-return relationship, it is necessary to understand the causes of forecast dispersion. The third chapter fills this gap by conducting a thorough analysis on the determinants of forecast dispersion such as firm risk, information asymmetries, forecasting difficulties, analyst conflicts of interest, herding. Evidence shows that forecast dispersion has several dimensions including information asymmetries and differences of opinion. On one hand, a group of analysts can have superior information. On the other hand, even when having the same information set, for example after earnings announcements, analysts revise their forecasts not necessarily in the same direction. Regression analysis shows that forecast dispersion is a function of firm’s risk, past performance, analysts’ differing information set, forecasting difficulty, and analyst conflicts of interest and herding. In other words, forecast dispersion is a complex concept and different factors simultaneously explain why analysts disagree in their forecasts. Overall, my analysis importantly suggests that we should be cautious when using forecast dispersion as a measure of firm riskiness or firm information environment. The fourth chapter investigates whether incorporating determinants of forecast dispersion as conditioning information in asset-pricing models helps capture the impact of the dispersion effect on raw and risk-adjusted returns of individual stocks (and not portfolios). I use four different specifications of the two-pass time-series regression models with time-varying betas, where betas vary with firm’s market value of equity, book-to-market ratio, and the corporate spread. Regardless of the method used for risk-adjustment, there is a strong negative relation between average returns and forecast dispersion. Moreover, my results show that accounting for the determinants of forecast dispersion reduces but does not eliminate the predictive power of forecast dispersion on stock returns. Remarkably, the determinants of forecast dispersion account for half of the profitability of dispersion strategy, thus substantiating the importance of the determinants of forecast dispersion in understanding the dispersion anomaly.