Internal DJNA studies have reached the same conclusion. Hafez (2009), for example,
demonstrates the superiority of relevance scores of 90 or above (incidentally, if an MCQ
ranking is assigned to a company in a story, then it must have a relevance score of 90 or
above).
We are now ready to define the events that shall be our buy signals.
Definition 3.SupposeC,¼hN 1 ;:::;Nmi, andPare as above, whereN 1 ;:::;Nmall have
a relevance score of 100 relative toC. Say thatwitnesses asentiment reversal for C
relative to Pif the following conditions hold:
1.m32;
- the difference between the publication datesNm 1 andN 2 is at least 30 calendar days;
3.ðC;Ni;PÞ<0 for all 2im1;
4.ðC;N 1 ;PÞ0; and
5.ðC;Nm;PÞ0.
Remark 1.If such a sequenceexists, we will simply refer to the event itself as a
sentiment reversal, where the event date is the date the last storyNmis published.
In summary, a sentiment reversal is a news story that causes a company to emerge
from a prolonged period of negativity, as determined by at least 30 negative net
sentiment measurements over at least 30 calendar days. Monthly measurement periods
are used in order to increase the likelihood that the reversals are non-accidental.
Requiring the minimal number of measurements to be 30 likewise seemed appropriate:
if a reversal occurs within the shortest possible time of one month, it would be
reasonable to expect an average of at least one news story per day.
As far as an anomalistic hypothesis is concerned, Figure 9.1 illustrates the main idea
well enough. Namely, over an extended period of negative press about a company, its
stock price is depressed below normative levels, and, when the stock is bought at the
moment the sentiment has reversed, there is reversion to previous price levels. In this
way, the anomaly would be the result of a kind of information lag. At the event date, net
news sentiment has reached an equilibrium point; however, cautious investors are still
getting comfortable with the idea that the company has emerged from its trough of
negativity, and the stock remains cheap.
9.3 Sentiment reversal universes
The only variable left free in Definition 3 isP: the monthly timeframe over which net
sentiment is measured. It might be helpful to think of this variable as a kind of ‘‘look-
back’’. As a company marches through time, how much news history about it, looking
back in time at any given moment, should be incorporated into a sentiment measure-
ment? Is one month enough or is a broader sweep of history needed? We approach
this question by studying 24 distinct universes of sentiment reversals as determined by
1-month to 24-month trailing net news sentiment measurement periods. In particular,
forx¼ 1 ;:::;24, define the universeUxto be the sentiment reversals obtained by
sequences of the form
hðC;N 1 ;xÞ;:::;ðC;Nm;xÞi;Sentiment reversals as buy signals 235