imperfections via ‘Rule of Thumb’ consumers or a fraction of liquidity constrained
consumers (GaH et al., 2005; Bilbiie et al., 2006). The latter models come also closer to
replicating the results of the growing empirical literature on the effects of fiscal policy.
The main result of studies that use the VAR-counterparts to DSGE-models is that they
can indeed recover significant effects of fiscal expansions on output. These are more in
line with a positive ‘Keynesian’ effect on consumption, albeit the eventual multiplier is
strongly reduced. The identification of fiscal policy is fraught with difficulties,
however.77 First, the implementation of announced changes in government policies is
subject to lengthy and visible political negotiations that are anticipated in private agents’
behaviour. As a consequence, fiscal shocks need not affect fiscal variables first. This is a
problem of the shock being non-fundamental (Lippi and Reichlin, 1994). Second,
decisions on fiscal policy affect different groups in the public via a range of different
spending and tax instruments. There exists no ‘standard’ fiscal shock: every political
discussion considers the trade-off between a range of possible taxation and spending
adjustments. The means of financing and the adjustment in expenditures and revenues
wrap empirically relevant effects of different budget components in an aggregate fiscal
shock without considering the path of public debt. Most studies focus on total spending or
revenues, and find small and positive effects of government spending on consumption,
but prolonged negative effects of higher taxation. Only a couple of studies consider the
dynamic behaviour of some particular budget components.78 Third, these identification
problems are only exacerbated by the automatic reaction of fiscal aggregates to economic
variables.
The seminal contribution of Blanchard and Perotti (2002) lies in using a semi-structural
VAR that employs external institutional information on the elasticity of fiscal variables to
output. Cleaning out the automatic cyclical reaction of the total fiscal balance leaves
shifts to the cyclically adjusted balance as discretionary fiscal shocks. Blanchard and
Perotti (2002) additionally impose some timing restrictions on the economic effects of
discretionary policy. These timing assumptions avoid to some extent anticipation effects
but would not capture these completely if implementation lags are important. Subsequent
studies have mainly attempted to verify the original approach of Blanchard and Perotti
(2002) with a variety of techniques and usually tend to confirm their findings.79
However, the empirical literature has hitherto ignored the supply and demand channels of
fiscal policy that are at front-stage of the theoretical DSGE models. Such effects are only
implicitly acknowledged in these VAR studies. Changes in tax revenues, for example, are
usually found to have lasting effects on output. There are nevertheless two other strands
of the empirical fiscal policy literature that attribute a role to supply side variables. First,
the literature on non-Keynesian effects of fiscal policy would argue that fiscal
77
78
79
A full discussion of the problems in identifying the effects of fiscal policy is provided in Perotti
(2005).
Ramey and Shapiro (1998) look into the sectoral reallocation effects following shocks. A particular
role in the transmission of fiscal policy shocks is also played by the labour market. A couple of
papers compare the effects of consumptive government purchases to increases in public employment
(Finn, 1998; Pappa, 2005; Cavallo, 2005). Perotti (2004) and Kamps (2004) examine the output and
labour market effects of government investment.
Mountford and Uhlig (2002) retrieve different types of fiscal shocks among those that conform to
some a priori sign restrictions on the entire impulse response or variance decomposition of fiscal
variables. Canova and Pappa (2002) select only those shocks that satisfy formal sign restrictions on
the conditional cross-correlation of the responses to the orthogonalised shocks of the variables in the
model.
128