The parameters of equation (5) and (6), notably β, δ, λ and the cut-off points c1 to c16 are
estimated using maximum likelihood. Since we are working in a panel data setting, the
generalization of ordered probit is not straightforward, because instead of having one
error term, we now have two. Wooldridge (2002) describes two approaches to estimate
this model. One “quick and dirty” possibility is to assume we only have one error term
that is serially correlated within countries. Under that assumption one can do the normal
ordered probit estimation but a robust variance-covariance matrix estimator is needed to
account for the serial correlation. The second possibility is the random effects ordered
probit model, which considers both errors εi and μit to be normally distributed, and the
maximization of the log-likelihood is done accordingly. This second approach should be
considered the best one, but it has as a drawback the quite cumbersome calculations
involved. In STATA this procedure was created by Rabe-Hesketh et al. (2000) and
substantially improved by Frechette (2001a, 2001b), and we will use such procedures in
our calculations.
4. Empirical analysis
4.1. Data
We build a ratings database with sovereign foreign currency rating attributed by the
three above-mentioned main rating agencies. For the rating notations we covered a
period from 1970 to 2005. The rating of a particular year is the rating that was attributed
at 31st of December of that year. In 2005 there are 130 countries with a rating, though
only 78 have a rating attributed by all three agencies (see Appendix 2 for rating
coverage description).2
In Figure 1 we can see the evolution of the number of countries rated by each agency
and it is possible to notice a significant increase in mid 1990’s of the number of
countries with rating, especially from S&P and Moody’s.
[Insert Figure 1 here]
2 The full historical rating dataset that we compiled, including foreign and local currency ratings as well
as credit rating outlooks, is available from the authors on request.
15