John Kittrell
ABSTRACT
The company-specific metric ofnet news sentimentis defined and shown to predict long-
term positive excess stock returns. A sequence of news stories about a company induces
a sequence of sentiment measurements (positive or negative relative to the company),
and when significant movements of such sequences from negative to positive are treated
as buy signals, the universe of stocks so defined outperforms the S&P 500 index over a
1-year time horizon. If the metric is further refined to incorporate extended periods
of news coverage, excess returns are improved. However, the event frequency and
subsequent outperformance of such sentiment reversals are sensitive to market
environments.
9.1 INTRODUCTION
It is clear to even the most casual investor that stock prices often move for irrational
reasons. In general, this is because the stock market is driven by human action and
naturally reflects the chronic irrationality within human nature.^1 Theefficient market
hypothesisis a pleasant academic fiction, although successful investing requires being
mindful of the inefficient (or, say, ‘‘sentimental’’) factors that lurk beneath. To under-
stand and harness the power of sentiment for profit has been a long-time dream of
money managers. The subjective nature of what constitutes sentiment is a serious
impediment to the realization of this dream; however, recent technological advances
in the field of computational linguistics provide opportunities to approach the matter in
a more rigorous, objective way. Leveraging the most up-to-date technology, we attempt
to measure the sentiment around companies as determined by their news coverage and
incorporate such measurements into a profitable investment strategy.
Notable contributions to the understanding of news sentiment are to be found
throughout the academic landscape. Themedia effect is one of the better known
empirical results connecting news and stock returns; viz., companies with no media
coverage outperform companies with high media coverage. In Fang and Peress (forth-
coming) and Peress (2008), the case is made that media coverage is directly related to the
9 Sentiment reversals as buy signals
The Handbook of News Analytics in Finance Edited by L. Mitra and G. Mitra
#2011 John Wiley & Sons
(^1) Such as a thing is, such is its act:Unumquodque enim quale est, talia operator—Aquinas (I–II, Q.55, A.2).