Abstract
Over the years, there has developed a fairly substantial body of research on the time series of earnings. As a whole, this literature concludes that changes in (annual) accounting earnings are unpredictable, that is, earnings follow a "random walk." Based on this result, some inferences of economic substance (policy) have been claimed. In this paper we reconsider empirical issues which, at least to some extent, have been obscured by this conclusion. We argue that the above result is only true in a limited sense, since a modest enlargement of the predictive information set should allow for a rejection of the hypothesis that earnings changes are unpredictable. Specifically, we hypothesize that book rate-of-return predicts earnings changes. If this is so, past inferences based on the "random walk hypothesis" are incorrect.