such as the EGARCH model of Nelson (1991) and the Diagonal VECH and Diagonal
BEKK models.^5 Finally, given that our sample period (October 2003 to September
2009) covers the Global Financial Crisis, we also analyze the two subsample periods,
before and after the crisis. Since it is difficult to point out the starting date of the crisis,
we apply the Quandt–Andrews Unknown Breakpoint Test (Andrews, 1993; Andrews
and Ploberger, 1994), which tests for one or more unknown structural breakpoints in the
equation’s sample.^6 The breakpoint is identified as the last interval of the date Novem-
ber 1, 2007. Therefore, we perform our analysis for the two subsample periods: October
1, 2003 to November 1, 2007 and November 2, 2007 to September 30, 2009. A detailed
discussion of our methodology is given in the technical appendix in Section 12.A (p. 283).
12.4 Results
Table 12.1 presents descriptive statistics for the 30-minute return of the S&P/ASX 200
Index and the SPI 200 Futures, absolute returns, and the number of news items. The
results show that all the return series and absolute return series are not normally
distributed. This is evident in the whole sample period, as well as in each year of the
sample. From Panels C and D, volatility in the equity and futures market, as proxied by
the absolute value of return, experiences a sharp increase from 2007. This finding reflects
the turbulent time during the Global Financial Crisis and provides the motivation for
investigating the news arrival–volatility relation in two separate subperiods. Finally, the
descriptive statistics in Panel E indicate that, on average, there are about nine company
announcements in each 30-minute interval. The number of news items variable also
exhibits a very high level of skewness. This finding implies that there are times when the
number of news events is much higher than its median level (i.e., news clustering).
The results of autocorrelations of the absolute 30-minute index and futures return as
well as news variables are presented in Table 12.2. From Table 12.2, we observe a
statistically significant serial correlation for the number of news items and the absolute
30-minute index and futures return. Similar to Kalev et al. (2004), the highest levels of
autocorrelation are evident in lags 12 and 24. This observation reflects accumulated
news arriving overnight. Overall, the results in Tables 12.1 and 12.2 suggest that the
number of news variables possesses a certain level of similarity with the volatility
clustering of return series. This provides some support for using the number of public
news arrivals to explain the persistence of volatility.
We initially start with an examination of the impact of news arrival on volatility based
on a censored regression of the absolute value of seasonally adjusted returns on news
arrivals. The results of this investigation are given in Table 12.3. The findings in Table
12.3 indicate that news arrivals have a positive effect on volatility. The coefficient
estimate for the news arrivals variable is positive and significant at the 1% level for
both the S&P/ASX200 Index and the SPI 200 Futures for the whole sample period as
well as in both subsample periods. We also document a positive relation between lagged
276 News and risk
(^5) These two models are the simpler form of the VECH model (Bollerslev, Engle, and Wooldridge, 1988) and the BEKK model
(Engle and Kroner, 1995) in which conditional variance and covariance depend only on their own lags and cross-products of
the error term.
(^6) Quandt (1960) developed the OLS-based test in case of unknown break location. Andrews (1993) and Andrews and
Ploberger (1994) provided the limiting distribution of the test statistic and critical values, and Hansen (1997) developed a
method to calculatep-values.