Data are defined following ESA-95 nomenclature. Definitions for the French budget
changed in 1978. We linked the former series (going back to 1970) to the ESA-95 series
and include an impulse dummy for this data break. We treat the effects of German
Reunification in 1991 in a similar way. We further condition the models on these
deterministic terms. Before estimating the structural model, we want to check for possibly
other breaks in the VAR. We follow the method of Bai et al. (1998) and apply the
sequential sup Quandt-Andrews likelihood ratio test on the VAR model. Sample size
forces us to consider a single break date only, as the optimal search concentrates on the
central 70% of the sample and consequently leaves too few degrees of freedom for
examining multiple breaks. We correct for a possible change in volatility before and after
the break date. As in Stock and Watson (2003), we weigh each period’s residuals by their
average volatility. The lag length in the VAR is henceforth set to one year (following the
Bayesian Information Criterion).
Table 5.4 reports the results. For Germany, we could detect a further break in the data in
1976, related to the large increase in social spending under the Brandt government. For
France, Portugal and Spain in contrast, we find a significant break date that is seemingly
related to the Maastricht consolidations, albeit the confidence bounds are rather large and
span nearly the entire nineties. It is nevertheless suggestive of the change in the conduct
of fiscal policy under the effect of the Maastricht rules. Due to this imprecision, we
refrained from explicitly modelling these shifts with additional dummy variables.
Table 5.4 VAR Break Date Test (Bai et al., 1998)
|
France |
Germany |
Portugal |
Spain |
|
1992*** [1989,1996] |
1976*** [1974,1978] |
1997*** [1995,2001] |
1998*** [1996,2003] |
Notes: *** denotes significance of the break date at 1%; break date is Sup-Quandt break date, years in
brackets are the confidence interval at 33% (Bai, 1997).
The Transmission Channels of Fiscal Policy
We first discuss some general results of our small scale model, and assess the properties
of output and fiscal series, and the role of the various structural shocks. The following
paragraphs discuss the fit of the model in terms of impulse response functions and the
forecast error variance decomposition.89 We have summarised all results in Figures 5.3
and 5.4. This prepares the ground for an analysis of the fiscal indicator in section 5.4.
The effect of productivity shocks is to lift up real output permanently (Figure 5.3). The
speed of accumulation is rather fast: after five years, the major part of the shock has
worked out. In Germany, this happens even faster. The sampling uncertainty around the
effect is large, but given the large bounds we have used, the significance of most impulse
responses after some years is actually surprising. To what extent are these supply shocks
driven by fiscal developments? In France and Portugal, these shocks go hand in hand with
positive long-term effects on total expenditures and revenues as well. This effect is also
strongly significant.90 The difference between revenues and spending responses is not
89
90
Impulse responses follow a one standard shock, and are plotted over a 10 year horizon with 90%
confidence intervals, based on a bootstrap with 5000 draws.
As the long-term elasticity of both spending and revenues is larger than unity, this looks like a
‘Wagner’ style government expansion owing to economic growth.
135