significantly higher trading volumes for earnings news, guidance, and restructuring
stories.
Lastly, the business cycle also affects the dominance of news items. Academic research
by Antweiler and Frank (2005) finds that on average news during an expansion is good
news, whereas news during a recession is bad news. In most cases the market response is
stronger during a recession that it is during an expansion. When we run regressions of
short-term returns post the news citation conditional upon whether it was good or bad
news, we observe that in periods of economic slowdown (2001–2002 and 2008–2009) bad
news relating to credit downgrades or earnings is deemed more important.
8.5 News flow and analyst revisions
Next we consider the impact of such misreaction to information by focusing on the
implication for earnings momentum strategies. From a quantitative perspective,
investors have traditionally relied upon earnings momentum factors to incorporate
corporate news flow. A consistent strategy based on buying companies each month
that have seen the most broker upgrades, and selling those with the most downgrades
would have generated an annualized return of 9.1% pa since 1990. Earnings revisions
strategies, however, typically do not identify the piece of information that has triggered
the change in forecasts. We only observe the actions of analysts, rather than the motive
behind analysts’ actions.
Here we consider whether earnings expectations change following certain news flows
and, if news does lead revisions, how can investors exploit this effect?
To proxy for changes in the market’s expectation of earnings around news, we focus
on revisions ‘‘clusters’’ (detailed below) using detailed analyst EPS revisions. Using
individual analyst EPS forecasts we calculate the dates when revision clusters are
formed. Our aim is to match revisions clusters to news items to understand what triggers
changes in analyst forecasts.
Following Bagnoli, Levine, and Watts (2005a, b), we define a revisions cluster as
occurring when at least three different analysts have revised their EPS forecasts within
three trading days for a given company. Once a cluster begins, the end date is marked
when more than three trading days have passed between sequential revisions. The end
date is then the date of the last revision within the cluster. This approach means that we
restrict our analysis to instances of significant revision activity and implicitly focus on
larger cap companies, thereby avoiding the issue that less liquid companies have less
news flow.
Having identified the revisions clusters, we then search for corporate news flow within
a 5-day window around the start date which may have triggered the first analyst
revision. Figure 8.7 shows the number of matched revisions clusters over time, while
Table 8.1 shows for each news type the percentage that result in a revision cluster and the
importance of that news type as a percentage of total revisions.
Our results show that significantly more clusters occur after earnings announcements,
implying that analysts are reluctant to change forecasts without first seeing hard
evidence. Twice as many revisions clusters are associated with accounting-related news
than strategic news. Table 8.1 shows that we have seen revisions clusters form 43% of
the time following earnings-related news. These revisions account for 31% of all analyst
The impact of news flow on asset returns: An empirical study 221