The Wiley Finance Series : Handbook of News Analytics in Finance

The first is the time-series oflog returns, denotedfrigi. Since we only have banks’ quote
data, this series is derived by taking the logarithm of the geometric mean of bid and ask
quotes (as described in Section 3.3). The second time-series we consider is that ofde-
seasonalized squared log returns, denotedfsigi, which is a measure of volatility in
exchange rates. Since exchange rate volatilities exhibit strong weekly seasonalities
(see Section 3.A.1 on p. 100), this volatility measure considers only excess volatility
over typical seasonal volatility. In particular, this series is constructed by first consider-
ing the squared log returnsfr^2 igi;from which we compute theweekly seasonality:

^rr^2 i ¼

1

n

Xn

j¼ 0

r^2 ðimodWÞþj W ð 3 : 7 Þ

wherenis the number of weeks in the data (220 in this case), andWis the number of
samples in a week (12 6024 7 for 5-second returns). Finally, we definede-seasonalized
volatilityto be:
si fr^2 i^rr^2 igi: ð 3 : 8 Þ

Using the events defined above, we test the null hypothesis that the distributions of
returns and de-seasonalized squared log returns before events are the same as after the
events.
For example, if we begin with the series of volatilitiesfsigi, then we denote bysðijÞthe
sample from timeiþtj, wheretjis the time of eventj, and we consider the time-series~ss
during a 1-hour window centered at each event; that is,

s

ð 1 Þ

 30 ;...;s

ð 1 Þ

0 ;...;s

ð 1 Þ

30

s

ðkÞ

 30 ;...;s

ðkÞ

0 ;...;s

ðkÞ

30 :

From these samples we can create an averaged event window:

^ss 30 ;...;^ss 30 ; ^ssi 1 k

Xk

j¼ 1

sðijÞ: ð 3 : 9 Þ

Then by studying the averaged event window we can assess the impact of the events
comprising the event study. Naturally, this analysis can be applied to analyze log returns
frigias well as volatilitiesfsigi(as exemplified above), and we consider both.

3.5.2 Examples of event studies

For concreteness, we present some illustrative examples of event studies that motivate
the tests described later in this section. Figure 3.3 shows the graphical interface to our
event study engine. The events being studied are surges in our macroeconomic keyword
index. The currency pair being considered is EUR/USD and, in particular, we are
studying the impact of events on exchange rate volatility (de-seasonalized squared
log returns). The large plot at the top shows the averaged event window (i.e.,^ssiin
the above notation) with the pre-event samples displayed to the left of the 0-minute
mark and the post-event samples displayed to the right of the 0-minute mark. Immedi-
ately, we see a peak in the center of the plot, representing a significant increase in

Managing real-time risks and returns: The Thomson Reuters NewsScope Event Indices 83