where
G
t is government expenditure,
T
t is the tax variable which
is being forecasted (income and social security taxes paid by
employees, or the same plus indirect taxes), TT
t is total tax
revenues, and
Q
t is GDP. All variables are expressed in real per
capita terms, and all first differences are normalized by the lagged
value of real, per capita disposable income. The second specification of the forecasting systems adds DGt00 to the list of regressors
of the government consumption equation, and
D
T
t
0
0 to the list of
regressors of the tax equation. The third specification of the
system adds
D
Q
t
0
0 to both regressions. In the benchmark regressions that I present below, for each country and for each variable I
choose the specification with the highest
R
0. In each regression
the constant is allowed to change in 0000