A Reexamination of Dispersion in Analysts’ Forecasts and Future Stock Returns

This paper reexamines the negative association between dispersion in analysts’ earnings forecasts and future returns. Diether, Malloy, and Scherbina (2002) attribute this relation to the heterogeneous expectations hypothesis pro- posed by Miller (1977). In this paper, I reexamine the validity of the Miller (1977) model for the relation between forecast dispersion and future returns and provide an alternative explanation (i.e., analysts’ self-selection hypothesis) for the observationally equivalent phenomenon. Consistent with recent research, I find that high dispersion stocks are overvalued relative to low dispersion stocks over the period 1984-1999. I expect that under the heterogeneous expectations hypothesis, changes in dispersion will be negatively associated with returns. Furthermore, the negative relation is more likely to be pronounced for the group of firms with severe short-sale constraints. By contrast, under the self-censoring hypothesis, higher forecast dispersion is likely to be associated with higher stopped coverage and deteriorating future performance. Using a sample of 35,599 firm years over the period of 1984 – 1999 in the US, I investigate whether stock returns are consistent with the predictions of the heterogeneous expectations hypothesis. I find evidence inconsistent with the heterogeneous expectations hypothesis. Changes in dispersion do not differentiate future returns. The level of dispersion in analysts’ forecasts appears “sticky” from period to period so that changes in dispersion do not drive the overvaluation of firms. I also find that the negative relation between dispersion in forecasts and returns holds irrespective of short-sales constraints. Instead, my evidence is consistent with the assertion that analyst self-selection plays a role in the asso- ciation between dispersion in forecasts and returns. Consistent with the selection explanation, I find that high dispersion firms are more likely to have upward bias and right-skewed forecasts. Also those high dispersion firms experience a significantly lower analyst coverage and higher incidence of delistings than low dispersion firms. More important, I find that the relation between analyst forecast dispersion and future returns disappears once I control for proxies for self- censoring (i.e., future profitability or stopped coverage). Taken as a whole, the results of this paper are consistent with the view that the negative relation between dispersion and future returns is due to the effective overvaluation that results from the analyst self-selection phenomenon. In addition, this paper suggests that high dispersion in forecasts is informative about the degree of self-selection by analysts. Investors may use the dispersion of fore- casts to undo the bias from self-selection. Given the increased importance of security analysts, coupled with the introduction of short-sales transactions in Korea, I am hoping that this study sheds light on the effect of security analysts’ forecasts and short-sales constraints on stock valuations.