estimate precisely the long run effects of shocks using a short data sample. Faust and
Leeper (1997) show SVARs that achieve identification through long run restrictions may
perform poorly when estimated over sample periods typically utilised and various other
papers have illustrated this. The authors do not describe the estimation of the SVAR
model in great detail, and many of the problems surrounding SVAR identification may
not apply to their model, but judging on some of their results the proper identification of
cyclical and supply shocks seems problematic.
The authors use a mixture of short and long term restrictions to identify economic and
fiscal shocks. They identify three structural shocks: (i) "supply" shocks that drive long
term trend rise in output (also including fiscal policy, i.e. a combination of shocks);
(ii) "cyclical" shocks that capture short term fluctuations around steady state equilibrium
(transitory); (iii) "fiscal" demand shocks distinguished by the elasticity approach (the
discretionary policy stance, but excluding those policies that have long term growth
effect, like productive spending and tax policies). When this identification scheme is
applied to the data the resulting IRFs and forecast error variance decomposition seem at
first sight somewhat puzzling. The chapter would benefit from further explanation of
some of the results: e.g. the negligible contribution of "cyclical" shocks to the variation in
output, the persistent contribution of discretionary "fiscal" shocks (although these are
restricted to have no long run output effect) to the total variance in output even at longer
horizons, and various other features in country specific IRFs like the non-Keynesian
effects of fiscal shocks in France and the relation between positive supply shocks and a
decline in expenditure in Germany.
I suspect part of the problem of not correctly identifying cyclical, discretionary fiscal and
supply shocks is due to the use of annual data by the authors and the estimation of the
model with a one year lag structure. The question is whether this allows for the proper
identification of cyclical shocks and its distinction from supply shocks and fiscal shocks?
It would be interesting to see the results when quarterly data are used instead (like
Blanchard and Perotti, 1999 and Perotti, 2005).
Another problem is that the authors do not distinguish compositional detail. Different
revenue measures have different long term supply effects, and the same applies for
different expenditure measures. Looking at total revenues and total expenditure is not
very revealing in this context. This is regrettable, as the authors rightly argue that it is
highly desirable that an economic indicator of fiscal policy takes into account supply side
effects. In its current form though, without disaggregating expenditure and revenues, it
remains something of a black box and one cannot distinguish long term fiscal policy
effects from other long term shocks that are included in the "supply" shock.
The resulting structural deficit is defined as structural expenditure minus structural
revenues (both driven by discretionary fiscal shocks and supply shocks). The fiscal
indicator is smaller than standard CABs and much more volatile. The volatility makes it
difficult to interpret the authors' results in any depth and hampers a comparison with the
cyclically adjusted balances calculated by the Commission and the OECD. As it appears
that the indicator is surprisingly insensitive to assumptions on newly estimated budget
elasticities, it seems likely that this volatility is related to the problems in correctly
identifying cyclical shocks, as so much is attributed to discretionary fiscal policy shocks.
I suspect that if the shock identification problems can be overcome, this method could
deliver some very useful insights into the underlying past trends behind developments in
public finances. I look forward to seeing future extensions of the authors' approach.
160