We begin by examining whether analysts do, in fact, condition their use of stock returns and other analysts’ forecast
revisions on how informative these signals are likely to be of future earnings changes. In contrast to much of the prior research
that examines analysts’ responses to these two public signals separately, we study the effect of both signals simultaneously
around earnings announcements because it allows us to reduce the likelihood of correlated omitted variables. As Welch
(2000) explains, ‘‘[l]acking access to the underlying information flow, [one] cannot discern if the influence of recent revisions
is either a similar response by multiple analysts to the same underlying information or is caused by direct mutual imitation’’
(p. 393). Consequently, in addition to studying these signals together, we also control for a number of other factors, such as
each analyst’s individual surprise in forecasting current period earnings, to further ensure that the analysts are in fact looking
to the reactions of investors and other analysts and not simply responding similarly to the event that caused their reactions
(i.e., the earnings announcement).
A key innovation in our study is that we develop proxies for the informativeness of each signal. For a measure of the
informativeness of returns, we use Baker and Wurgler’s (2006, 2007) monthly index of investor sentiment. The sentiment
measure is constructed to capture the speculative trading demand for stocks and investors’ optimism or pessimism about the
market. In other words, investor sentiment can be thought of as investor optimism or pessimism that is not justified by a
stock’s underlying fundamentals (e.g., earnings, cash flows, or growth rates). When sentiment is extremely high or low,
changes in stock prices are not as representative of the underlying fundamentals and are, therefore, likely to be less
informative about future earnings. For a measure of the informativeness of the average analyst revision, we use the number of
analysts who have contributed to the average revision. Intuitively, as more analyst opinions are added, idiosyncratic noise
gets averaged out. We use these two measures of informativeness because they are ex-ante knowable to analysts. However,
they are only proxies for informativeness, so we also use ex-post measures of informativeness as a robustness check. As we
explain in more detail below, these proxies play a central role in our study because they enable us to make much stronger
inferences about the underlying drivers of analyst behavior.