Drew Creal, Siem Jan Koopman, and Andre Lucas, "Generalized
Autoregressive Score Models with Applications", Journal of Applied
Econometrics, Vol. 28, No. 5, 2013, pp. 777-795.

The paper has two applications. The first application introduces new
bivariate copula models, for which we can provide the data. The second
application is to default data from Moody's, and these data are
confidential.

The bivariate Gaussian copula models introduced in this paper are
estimated using daily exchange rate data from 1986 through 2008. There
are three series with 5729 time-series observations each. The original
exchange rate data come from Datastream and are the DeuscheMark
(Euro)/Dollar, Yen/Dollar, and British Pound/Dollar.  The copula
models do not use the raw data. As explained in the paper, the raw
data are first pre-filtered using AR/GARCH models as univariate
marginal models to obtain the probability integral transforms. The
bi-variate copula models are then estimated on the probability
integral transforms. The text file contains the post-processed data
(the probability integral transforms) used to estimate the new copula 
models introduced in the paper. In the text files, there are four
columns of data. The first column contains the dates, and the last
three columns are the probability integral transforms. Each series of
probability integral transforms has 5729 observations and they are in
order across the columns: Euro/Dollar, Yen/Dollar, and British
Pound/Dollar.

The probability integral transforms are in the file prob-int-trans.cs,
which is an ASCII file in DOS format. It is zipped in the file
ckl-data.zip. Unix/Linux users should use "unzip -a".

The default and rating transition data used in the point process
application of this paper are the property of the Moody's Investor
Services, who do not permit open access, hence the data cannot be
lodged here.  The data can be bought from Moody's (www.moodys.com) as
their Default Research Database (Corporate). 

The analysis in the paper is based on the data between January 1981
and March 2010. The data can be obtained by merging the tables
MASTDFLT, MASTISSR, and SENRATG in the database.

From the original data base, all issuer ratings are obtained for all
US companies in the database. The original 18 grade ratings are mapped
into a 2-grade rating system. Upon a default day, we check whether
there are multiple defaults with the same qualitative default 
description in the database. If so, we check the bankruptcy dates,
industry classification, and ultimate parent identifier, where
available, and eliminate the younger defaults. If a company becomes
non-rated in the database and subsequently jumps to default, we skip
the transition to the non-rated class.

Drew Creal
dcreal [AT] chicagobooth.edu