Elena Andreou and Eric Ghysels, "Detecting Multiple Breaks in Financial
Market Volatility Dynamics", Journal of Applied Econometrics, Vol. 17,
No. 5, 2002, pp. 579-600.

The sources of data used in this paper are Datastream and Olsen and
Associates. Both databases are proprietary and therefore redistribution of
these data is not permitted. 

Datastream access can be obtained via subscription to the Datastream
International Limited. Four international stock market price (Pt) indices
were downloaded from this database, namely, the Financial Times Stock
Exchange 100 index (FTSE), the Hang-Seng Index (HSI), the Nikkei 500 index
(NIKKEI), and the Standard and Poors 500 index (S&P500). The names in the
parentheses refer to the variable names in the paper and tables. The sample
period of 4/1/1989 - 19/10/2001 at daily frequency yielded a sample size of
T=3338 and the returns series for each index is defined as

       rt = log P_t - log P_t-1.
       
The choice of the sample is based on the recent experience of the Asian and
Russian financial crises. 

We also study the high-frequency Yen vis-a-vis the US dollar returns time
series. The data were purchased from Olsen and Associates

	http://www.olsen.ch/
	
and enquiries about data can be directed to

    data-info@olsen.ch
    http://www.olsendata.com/
    
and
  
    Olsen Data AG, 
    Seefeldstrasse 233, 
    CH-8008 Zurich, 
    Switzerland.

The sample period of 1/12/1986-30/11/1996 at 5-minute sampling frequency is
examined. The original sample provided was 1,052,064 five-minute return
observations, equivalent to 2653 days times 288 five-minute intervals per
day. Olsen and Associates construct these data from interbank FX quotes that
appear on Reuter's FXFX page. Each quote consists of a bid and an ask price
together with a "time stamp" to the nearest even second. The price at each
5-minute mark is obtained by linearly interpolating from the average of the
log bid and log ask for the two closest ticks. The continuously-compounded
returns are then simply the change in these 5-minute average log bid and ask
prices.

The returns for some days were removed from the sample to avoid having
regular and predictable market closures and calendar events which affect the
characterization of the volatility dynamics. These refer to the thin trading
of the weekends for which we exclude the observations from Friday 21:05 GMT
until Sunday 21:00 GMT, as well as the fixed holidays (Christmas, 24--26
Dec., New Year, 31 Dec -- 2 Jan and 4th of July) and finally moving holidays
(Good Friday, Easter Monday, Memorial Day, Labor and Thanksgiving days).

Moreover, we also exclude the daily sample when FX activity slows which
covers 21:05GMT the night before to 21:00 GMT that evening. Finally, we
deleted some returns contaminated from brief lapses in the Reuters data feed
which are shown as at least a long sequence of 15 zeros or constant 5-minute
returns (in place of the missing quotes) in the series. The final sample
includes 705,024 five-minute returns reflecting T=2448 trading days. A very
similar method and further explanations for the construction of this
high-frequency data series can be found in Andersen, Bollerslev, Diebold and
Labys (2001) and Andreou and Ghysels (2002), as referenced in the paper.