12
3.2.3 The Two Approaches Compared
We first applied the ICSS algorithm to the four stock returns series over 1997-2002,
and to the interest rates series over 2000-April 2002. Figures 2 and 3 show the evolution of
each return series (first difference, in the case of interest rates) along with ±3 standard
deviations. A discontinuity in the dotted lines indicates the presence of a volatility
breakpoint.
[Figures 2 and 3]
For Emerging Asia, the ICSS algorithm detects breakpoints in August 1997,
October 1997, February 1998, and October 1998. For Latin America, breakpoints are found
for October 1997, October 1998, February 1999, and June 2000. For North America,
breakpoints took place in October 1997, October 1998, May 2002, and October 2002.
Finally, for Europe, breakpoints are found in October 1997, April 1998, October 1998,
January 2000, September 2001, and October 2002. Only in this latter case, does the ICSS
algorithm detect a permanent shift in volatility around September 11, 2001.
For interest rates, the ICSS algorithm detects breakpoints in variance in the 30-day
series in February 2001, June 2001, and August 2001; in May 2001 and August 2001 for
the 60-day series; in March 2000, May 2000, August 2000, March 2001, April 2001, July
2001, September 2001, October 2001, February 2002, and March 2002 for the 90-day rate;
and, in May 2000, August 2000, March 2001, and September 2001 for the 180-day rate.
In order to compare the ICSS algorithm with wavelet analysis, we took as a
benchmark the dates of breakpoints detected by the ICSS algorithm, and tested whether
wavelet analysis also found breakpoints around the same time periods. In particular, for the
stock indices, we focused on 1997-1998, 1999-2000, and 2001-2002. For interest rates, we
concentrated on 2000-1/2001 and 2/2001-4/2002. Tables 2 and 3 present our results.