From the results presented above one may see that there is some evidence in favour of a
Ricardian regime, of monetary dominance, and that the EU-15 governments have a
tendency to use primary budget surplus to reduce the debt-to-GDP ratio, since we get a
negative sign for the estimated γ coefficient (-0,766, in the fixed effects model) for
equation (28). There is therefore no evidence that can be regarded as supporting the
FTPL for this set of European countries.
The fixed effects model is a typical choice for macroeconomists, and may eventually be
more adequate than the random effects model. For instance, if the individual effects are
somehow a substitute for non-specified variables, it is probable that each country
specific effects are correlated with the other independent variables. In addition, and
since the country sample includes all the relevant countries, the EU-15 countries, it is
less obvious that one might want to consider this set of countries as a random sample of
a larger universe of countries.
In other words, and as reminded by Greene (1997) and Judson and Owen (1997), when
the individual observations sample (countries in our case) comes from a larger
population (which could be instance all the developed countries), it would be suitable to
consider the specific constant terms as randomly distributed through the cross section
units. However, and even if the present country sample includes a small number of
countries, it is sensible to admit that the EU-15 countries have similar specific
characteristics, not shared by the other countries in the world. In this case, it would
seem adequate to choose the fixed effects formalization, even if it is not correct to
generalize the results afterwards, to the entire population, which is not the purpose of
the paper.
In the previous specification there is nevertheless an implicit assumption that the
underlying model is homogeneous that is, the coefficients are the same for all countries.
As a matter of fact, one of the problems with panel data estimations, as mentioned
namely by Haque, Pesaran and Shrama (2000), is the possibility of the real model might
be heterogeneous, with different coefficients for the explanatory variables in the cross-
section dimension. Assuming the same coefficients for all the countries, with the
exception of the intercept, may give rise to non-linearity in the estimations, even if the
relation between the variables is linear. An alternative estimator, proposed by Pesaran
26