Nikolaus Hautsch, Lada M. Kyj, and Peter Malec, "Do High-Frequency
Data Improve High-Dimensional Portfolio Allocations?", Journal of
Applied Econometrics, Vol. 30, No. 2, 2015, pp. 263-290.

The raw data used in this paper are confidential. However, the
empirical results can be replicated using the blocked correlation
matrices, variance estimates, and realized covariance matrices, as
well as the open and close prices provided.

The data used in the article are from the Trade and Quote (TAQ)
database of the New York Stock Exchange. TAQ is a collection of
intraday trades and quotes for all securities listed on the New York
Stock Exchange, American Stock Exchange, Nasdaq National Market
System, and SmallCap issues.  This study considers a subset of 400
stocks from the S&P 500 index with the longest continuous trading
history during the sampled period, which is from January, 2006 to
December, 2009. The stock symbols are listed in the file SPXTickL.txt.
A liquidity index, which sorts the stocks according to the number of
mid-quote revisions, is in the file SPXLiquIndx.txt. 

Blocked correlation matrices are estimated from raw datasets which are
mid-quote prices sampled at 1 second increments from 9:45 EST to 16:00
EST. The data filtering procedure is discussed in section 4 of the web
appendix. The estimated correlation matrices are in five files called
BRKCorrx.txt, where x corresponds to the number of liquidity groups
used (1, 2, 4, 8 and 10). These files contain 80200 by 1003 matrices,
where the rows correspond to correlation elements in vech format, and
the columns correspond to the number of trading days.  

The corresponding numbers of refresh times per block are in five files
called NRftx.txt with x defined as above. These files contain 1003 by
x(x+1)/2 matrices, where the rows correspond to the number of trading
days and the columns correspond to the number of distinct blocks.

Variance estimates based on the univariate realized kernel are in the
file RKVar.txt. This file contains a 1003 by 400 matrix, where the
rows correspond to the number of trading days, and the columns
correspond to the number of stocks.

5-minute realized covariance matrices based on previous-tick
interpolation of the mid-quote data are in the file RCov5Min.txt. This
file contains a 80200 by 1003 matrix, where the rows correspond to
covariance elements in vech format, and the columns correspond to the
number of trading days.  

Daily open and close prices are in the files OpenPrice.txt and
ClosePrice.txt, respectively. These files contain 1003 by 400
matrices, where the rows correspond to the number of trading days and
the columns correspond to the number of stocks.

Since the five BRKCorrx.txt files and the RCov5Min.txt file are
extremely large, they have been compressed using 7zip instead of zip
or pkzip. Using 7zip results in substantially smaller compressed
files. This program is freely available for many versions of Windows
at

  http://www.7-zip.org/

Command-line versions for many versions of Linux are also available.
On Debian and Ubuntu systems, they may be found in the package
p7zip-full.

Each of the files BRKCorrx.txt and the file RCov5Min.txt is 7zipped in
the file BRKCorrx.7z and RCov5Min.7z, respectively.

All other files are zipped in the file hkm-data.zip. These files are
ASCII files in DOS format. Unix/Linux users should use "unzip -a".

There is also a web appendix in the file Hautsch-Kyj-Malec-appendix.pdf.

Please address any questions to:

Peter Malec
Institute for Statistics and Econometrics
Humboldt-Universität zu Berlin

malecpet [AT] hu-berlin.de