The Wiley Finance Series : Handbook of News Analytics in Finance

3.5.5 The^2 test for goodness of fit

Another test for changes between pre-event returns/volatility and post-event returns/
volatility is the^2 goodness-of-fit test. Below we describe the test for averaged event
window log returns, but the same test could be applied to volatilities as well.
Recall that the^2 test consists of the following steps:

.Create histograms off^rrwþ 1 ;:::;^rr 0 gandf^rr 1 ;:::;^rrwgsuch that they have the same bins
and every bin in the first histogram has at leastncounts, wheren>0.
.Denote the bin frequencies of the pre- and post-event histograms byff 1 ;:::;fkgand
ff 1 þ;:::;fkþg, respectively, wherekw.
.Finally, define thestatistic by

  i

ðfifiþÞ^2

fi

ð 3 : 16 Þ

The shape of the distribution changes with ð 1  Þ100% confidence if
> 21 ð 1 ;k 1 Þ.

3.6 News indices and FX implied volatility

In Section 3.5, we showed that event indices, on average, have an impact on realized FX
volatility. Since FX implied volatility indices also forecast realized volatility (see Pong et
al., 2004; Taylor, 2005), this suggests that implied volatility and news indices might be
related. On the other hand, there is an important difference between the two: while event
indices are calibrated to predict volatility over 30-minute periods, implied volatility
indices forecast volatility over much longer periods, typically about 30 days. The event
study methodology was employed to determine whether a relationship between the two
does, in fact, exist. No evidence to that effect was found; this suggests that implied
volatility and event indices may function as complementary sources of information for
risk management, each focused on a different time horizon.

3.6.1 Data pre-processing

Bank quotes for implied euro volatility were obtained from Thomson Reuters for 2005
to mid-2007. Preliminary exploration revealed that the major banks quote persistent, yet
statistically different implied volatilities (it is not uncommon for different banks to quote
implied volatilities that differ by 3 standard deviations or more). This means that one
could easily mistake changes in quote provider for genuine changes in implied volatility.
To preempt such errors, and to focus on the relationship between implied volatility and
news, we select one source of quotes for our analysis, and choose the most frequent
provider, Socie ́te ́Ge ́ne ́rale, which was responsible for 20,691 of the total 53,959 quotes
in our sample. Quotes from other banks were ignored.^5 Each tick in the time-series
contained both a bid volatility and an ask volatility. We used the arithmetic mean of
these two values.

Managing real-time risks and returns: The Thomson Reuters NewsScope Event Indices 89

(^5) The next-most-frequent providers were BNP Paribas at 12,475, Broker at 4,980, and RBS at 4,341 quotes.