their motivations such that it is difficult to differentiate between analyst herding that is
imitation- or information-driven. Since market prices are driven by expectations of
corporate fundamentals and new information leads to a revision of those expectations,
here we look at higher frequency information contained within corporate news flow as a
leading indicator of analyst revisions to understand how different types of news are
incorporated into analysts’ earnings forecasts.
Our aim is to understand what type of information causes analysts to revise their
earnings expectations, how the informational content of the signal varies according to
the news catalyst, and whether investors can use news flow signals as input into their
models.
Many of the anomalies that quantitative investors systematically exploit are based on
deep-seated behavioural biases. The academic literature points to both the market and
analyst misreaction to new information—due to delayed information diffusion, investor
inattention and investors’ limited ability to process information instantaneously, with
implications for stock-specific risk, earnings, and price-momentum-based strategies over
the short and medium term.
One focus of academic research has attempted to explain the momentum and reversal
characteristics of different investor responses to public and private signals, considering
the type of information that causes investors to change their expectations. For example,
DeBondt and Thaler (1990) argue that investor myopia results in an overemphasis on
recent earnings. Cognitive biases mean that analysts overreact and place too much
weight on new information and put less weight on long-term averages. Daniel,
Hirshleifer, and Subrahmanyam (1998) model investor behaviour by overconfidence
and biased self-attribution. In their model, investors hold too strongly their own
information and discount public signals, resulting in underreaction to public informa-
tion and overreaction to private information; while Barberis, Shleifer, and Vishny (1998)
base their model on conservatism and the representativeness heuristic, arguing that
investors change their views and overreact or underreact to company earnings based
on the past stream of realizations. Hong and Stein (1999) illustrate a model based on the
slow diffusion of information through two classes of traders with a differential in
processing news, resulting in investors who underreact to news and overreact to non-
informational price movements.
The second area of research has focused on earnings momentum, by measuring
analyst forecast errors and analyst underreaction to the continuation in returns.
Abarbanell and Bernard (1992) find that analyst forecast errors are positively correlated
with prior year changes in earnings, implying that analyst forecasts do not fully reflect
information in recent changes while Easterwood and Nutt (1999) were the first to
examine whether misreaction to new information varies depending on the nature of
the information. They show that analysts overreact to good news and underreact to bad
news.
More recent research has focused on using computational linguistic methods. For
example, Tetlock (2007) uses General Inquirer software to count the number of times
words appear within text from predetermined categories within the Harvard IV-4
Psychosocial Dictionary. Tetlock shows that firms with a high fraction of negative words
in firm-specific news stories prior to announcing earnings go on to announce low
earnings. This predictive ability is strongest when the story mentions ‘‘earnings’’ (i.e.,
news stories with a high proportion of negative returns and a reference to the word
212 News and abnormal returns