The Economic Journal, 121 (February), F4–F32. Doi: 10.1111/j.1468-0297.2010.02413.x. 2011 The Author(s). The Economic Journal 2011 Royal
Economic Society. Published by Blackwell Publishing, 9600 Garsington Road, Oxford OX4 2DQ, UK and 350 Main Street, Malden, MA 02148, USA.
THE EFFECTS OF GOVERNMENT PURCHASES SHOCKS:
REVIEW AND ESTIMATES FOR THE EU*
Roel Beetsma and Massimo Giuliodori
In this article, we review the theoretical consequences of government purchases shocks for both
closed and open economies, followed by a discussion of the empirical literature. Next, we provide
our own estimates for the EU countries. We find that an increase in government purchases
raises output, consumption and investment and reduces the trade balance. However, the stimulating
effect is weaker and the trade balance reduction is larger for the more open EU economies,
consistent with larger leakage effects. Further, we show that government purchases shocks in
large EU economies have non-negligible consequences for economic activity in the main trading
partners.
Until the early 1980s fiscal policy was widely regarded as a useful tool for economic
stabilisation. In response to the oil shocks of the 1970s many countries relied not only
on monetary accommodation but also fiscal expansion. However, those active fiscal
policies did not prevent widespread increases in unemployment, while at the same time
they resulted in high deficits and rising public debts. This demonstrated ineffectiveness
of fiscal policies made many economists sceptical about the usefulness of fiscal policy as
a tool for macroeconomic stabilisation. Nevertheless, politicians have continued to view
an active fiscal policy as a useful instrument for solving their economic problems. Quite
recently, though without success, Japan applied large fiscal expansions in an attempt to
escape its liquidity trap. Further, after they secured membership of the Euro-area, a
number of countries relaxed fiscal policy. Since the beginning of the current crisis,
the US have engaged in unprecedented fiscal expansion, much of it associated with
the rescue of the financial sector but also a large share spent to support the ‘real
economy’. In November 2008, the EU presented its European Economic Recovery
Plan (EERP) aimed at a cumulative (over the crisis period) discretionary fiscal stimulus
of about 1.5% of the EU GDP, with 1.2% of GDP coming directly from the Member
States.
This article starts by reviewing the theoretical and empirical consequences of an
increase in government purchases in closed and open economies. We focus mostly on
the short to medium-run consequences of fiscal expansions, because the main question
we are interested in is to what extent discretionary fiscal policy is able to stabilise the
business cycle.1 We also present our own estimates for the EU economies of responses
to temporary government purchases shocks, paying specific attention to the behaviour
* Corresponding author: Roel Beetsma, University of Amsterdam, CEPR, CESifo and Tinbergen Institute,
Roetersstraat 11, 1018 WB Amsterdam, The Netherlands. Email: R.M.W.J.Beetsma@uva.nl.
We are grateful to Luca Onorante and two anonymous referees for their helpful comments on an earlier
version of this article.
1
In this respect, our article complements Gemmell et al. (2010) and Heady et al. (2010) who study the
long-run effects of fiscal policy, while, moreover, they focus on the long-run economic consequences of the
structure of the tax-spending system.
[ F4 ]
[ F E B R U A R Y 2011 ]
GOVERNMENT PURCHASES IN THE EU
F5
of the public budget balance, the trade balance, the components of GDP, the role of
trade openness and cross-border spill-overs.2 We conduct an elaborate investigation
of the potential presence of anticipation effects. However, we fail to find any evidence
of such effects.
The results of our estimation are consistent with the neo-Keynesian framework. An
increase in government spending raises output, with a multiplier exceeding unity on
average, and consumption and investment, and it reduces the public balance and the
trade balance. The real effective exchange rate appreciates with some lag and also the
long-term interest rate increases with some lag. Moreover, our findings suggest that
the stimulating effect is smaller for more open economies, because a substantial part of
it leaks away. We also find evidence of spill-overs across countries via the trade channel.
This strengthens the rationale behind the concerted fiscal expansion envisaged in the
EERP. While the benefits from unilateral expansion may be too small due to leakage to
other countries, a multilateral expansion forces countries to internalise the crossborder effects of their own efforts and may benefit all of them, in particular in times of
deep recession like we have experienced in the past two years.
The remainder of this article is structured as follows. In Section 1 we briefly discuss
the theoretical consequences of a discretionary increase in government purchases
in closed and open economies. Section 2 summarises recent empirical evidence. In
Section 3 we present our own baseline responses to government purchases shocks for
the EU, while in Section 4 we explore the robustness of our baseline results. In Section
5, we investigate the effects of government purchases shocks on the components of
GDP, while in Section 6 we estimate the strength of the cross-border spill-overs of such
shocks. Section 7 concludes the main text. An online Appendix provides details on
the handling of the data and some further results (http://www1.fee.uva.nl/toe/
beetsma.shtm).
1. Government Purchases Shocks: A Brief Theory Review
We are mostly interested in the short to medium-run effects of shocks to government
purchases on output, consumption and investment. In analysing those effects, it may be
useful to distinguish between closed and open economies.
1.1. Fiscal Expansions in Closed Economies
Baxter and King (1993) conduct a number of fiscal policy experiments in a standard
neo-classical model. Taxes are lump sum. Hence, for a given path of government
purchases the time profile of tax revenues does not matter. Consider an increase
(temporary or permanent) in government consumption. The discounted value of
future tax payments rises, which generates a negative wealth effect and induces individuals to reduce both private consumption and leisure. The labour supply increases,
hence real wages fall and total output expands. Moreover, the rise in employment raises
2
In contrast to the theoretical review Section, in which for expositional purposes, we concentrate purely
on the effect of the fiscal expansion, in our regression analysis in order to reduce the risk of omitted variable
bias, we account for the reaction of monetary policy via a change in the interest rate.
2011 The Author(s). The Economic Journal 2011 Royal Economic Society.
F6
THE ECONOMIC JOURNAL
[FEBRUARY
the (marginal) productivity of capital and thereby the real interest rate, and induces
more private investment. A boost to public investment produces additional positive
effects on the economy by raising the productivity of capital and labour. Private
consumption may then even increase, although with some lag.
Burnside et al. (2004) introduce a number of features into the neo-classical model to
improve the replication of estimated responses to a government purchases shock. In
particular, they assume that there are costs to changing the capital stock, while there is
‘habit persistence’ in consumption in the sense that individuals want to maintain a
given consumption level. This way, changes in employment and investment exhibit a
more drawn out pattern after the policy shock.
The most distinguishing feature of the neo-classical model is that it tends to predict a
fall in private consumption following an increase in government purchases, while a
large part of the empirical literature (discussed below) finds the opposite effect.3 The
main obstacle in reconciling theory with this empirical observation is the rightward
shift of the labour supply curve, which for a given labour demand curve implies a lower
wage and, hence, less consumption. Hence, to generate a positive effect on consumption it is necessary (though not always sufficient) to have a mechanism that also
shifts the labour demand curve to the right. An example is found in Devereux et al.
(1996), where higher government spending raises the equilibrium number of firms in
intermediate goods sectors characterised by increasing returns to specialisation. The
productivity of all firms in the sector rises and the resulting outward shift in the labour
demand may dominate the increase in the labour supply, hence produce a higher real
wage and an increase in consumption.
Ravn et al. (2006) introduce ‘deep habits’ into a model with monopolistic competition in goods production. ‘Deep habits’ refer to habit formation about the amount of
consumption of individual goods (rather than about aggregate consumption as is
standard under the assumption of habit persistence). The model implies that demand
for a specific good is composed of a price elastic component (as in standard monopolistic competition models) and an inelastic component that does not react to price
changes. An increase in aggregate demand raises the weight of the elastic component,
inducing producers to lower their prices. An increase in government consumption
produces the standard negative wealth effect on the one hand, resulting in a higher
labour supply. On the other hand, the higher aggregate demand induces producers to
lower goods prices, thereby also generating more demand from private agents. The
overall higher demand for goods stimulates the demand for labour, which exerts a
positive effect on the real wage that may dominate the downward wage pressure from
the higher labour supply.
An alternative route to generating a positive consumption response to a government
purchases increase is to use a New-Keynesian framework with monopolistic competition
and nominal rigidities. These could take the form of pre-set prices for products. Such a
framework has originally been used to study the effects of monetary policy shocks, but
recently it has also become a vehicle for analysing fiscal policy (Beetsma and Jensen,
3
There are ways, though, to reconcile the neo-classical model with a positive consumption response to a
government purchases impulse. Linnemann (2006) shows that when consumption and employment are
complementary in utility, such an impulse may generate a positive consumption response. The wealth effect
stimulates employment and, due to the complementarity, consumption is also stimulated.
2011 The Author(s). The Economic Journal 2011 Royal Economic Society.
2011 ]
F7
GOVERNMENT PURCHASES IN THE EU
2005). As before, an increase in government purchases generates a negative wealth
effect implying an increase in the labour supply. However, with sticky prices and because the price of output exceeds the marginal cost under monopolistic competition,
an increase in the demand for goods will be met with an increase in the supply of
output. Hence, the demand for labour increases, which has a positive effect on the real
wage rate, ceteris paribus. However, even if the overall effect on the real wage is positive,
this does not guarantee a (substantial) positive effect on consumption. The reason is
that consumers have an incentive to save at least part of the higher real wage for the
future. Owing to the rise in (current and future) taxes, consumption may still fall.
Hence, an additional imperfection is needed to generate higher consumption.
Therefore, Galı́ et al. (2007) assume that there is a group of ‘rule-of-thumb’ consumers
who consume their entire disposable income (income minus taxes) and, thus, do not
save. These consumers spend the entire increase in their real wage immediately. Provided that this group of consumers is sufficiently large, the net effect on consumption is
positive.
To summarise, Table 1 provides a systematic overview of the macro-economic effects
of an increase in government purchases in different settings.
1.2. Fiscal Expansions in Open Economies
With an open-economy framework, we can also explore the consequences of a
government purchases shock for the trade balance and the real exchange rate. The
discussed mechanisms produced by the neo-classical and New-Keynesian frameworks
remain present. However, also other factors emerge as relevant for the macro-economic
responses to the shock. In particular, we discuss below the roles of international asset
market completeness, trade openness of the economy and the substitutability of
products across borders.
Consider an extension, as in Baxter (1995), of Baxter and King’s (1993) neo-classical
model to a setting of two open economies, Home and Foreign. The countries produce
perfectly substitutable goods. With complete international asset markets, risks are
perfectly shared between the two countries. A government purchases impulse in Home
generates identical responses of output, consumption and investment in the two
countries. The Home trade balance deteriorates as a net effect of lower public saving
and higher private saving and investment. With incomplete international asset markets
in which only a risk-free bond is traded, the wealth effect is borne by Home. While the
Table 1
Predicted Effects of a Government Purchases Increase in a Closed Economy
Neo-classical
Deep habits
NK – nominal rigidities only
NK – rule of thumb
Output
Consumption
Real
wage
Labour
supply
Labour
demand
Employment
"
"
"
"
#
"
#
"
#
"
"
"
"
"
"
"
0
"
"
"
"
"
"
"
Notes. This Table is based on Pappa (2006). ‘NK’ = New-Keynesian.
2011 The Author(s). The Economic Journal 2011 Royal Economic Society.
F8
THE ECONOMIC JOURNAL
[FEBRUARY
signs of the responses are unaltered, Home labour input and production rise by more
and consumption and the real wage fall by more than before, while the opposite is the
case in Foreign.
Corsetti and Müller (2008) assume that Home and Foreign products are imperfectly
substitutable, implying that the composition of government purchases becomes
important. Government purchases fall entirely on domestic products, which is a
reasonable assumption as they are mostly made up of civil servants’ salaries, while
public procurements tend to be biased towards domestic firms. Because the spending
impulse falls on Home products, Home’s real exchange rate appreciates. Now, the
standard risk-sharing condition under complete international asset markets implies
that the fall of Home consumption exceeds that of Foreign.4
More openness stems the fall of consumption, because it dampens the increase in
the real interest rate (being the price of a unit of today’s consumption in terms of
future consumption). Intuitively, the real exchange rate appreciation lowers the price
of today’s consumption relative to what it will be in the future when the real exchange
rate has returned to its original value. This dampening effect is stronger, the larger is
the consumption share on foreign goods (i.e. the more open is the economy). The
more moderate rise in the real interest rate weakens its negative effect on the investment response. Obviously, the positive effect of openness on consumption and
investment implies a larger trade balance deterioration.
In a related framework, Müller (2008) shows that the real exchange rate appreciation
causes an ‘expenditure switching effect’ with Home and Foreign consumers switching
towards Foreign products. The switching effect is larger the higher is the elasticity of
substitution between Home and Foreign products. The increase in output will be
correspondingly smaller and the fall in the trade balance will be larger.
2. Government Purchases Shocks: A Brief Review of the Empirical Literature
The main obstacle in empirical fiscal policy analysis is to identify exogenous and
unexpected fiscal events. In this regard, the literature has followed two major
approaches. One is the so-called ‘narrative’ or ‘dummy’ methodology. Ramey and
Shapiro (1998) and Ramey (forthcoming) identify large exogenous increases in defence
spending and analyse their effects on the economy.5 However, Perotti (2007) argues
that, while this approach is appealing and war episodes are plausibly exogenous, these
are not ‘normal’ events and, therefore, they cannot be safely used to draw inference on
the effects of fiscal policy on economic activity under non-exceptional circumstances. An
alternative route might be to follow Romer and Romer (2010), who use (real-time)
information from the official US budget documents to classify exogenous tax changes.
The other major approach is to set up a structural vector autoregression (SVAR)
model. The model explains a system of macroeconomic variables by its lags and
exogenous shocks to the variables in the system. The ‘trick’ is to find an appropriate set
4
The condition is (after some normalisation), uCF =uCH ¼ RER, where the left-hand side is the ratio of the
marginal utilities of Foreign (F) and Home (H) consumption, while the right-hand side is the real exchange
rate, defined such that an increase means a depreciation of Home real exchange rate.
5
Edelberg et al. (1999) and Burnside et al. (2004) are closely-related follow ups of the work by Ramey and
Shapiro (1998).
2011 The Author(s). The Economic Journal 2011 Royal Economic Society.
2011 ]
GOVERNMENT PURCHASES IN THE EU
F9
of restrictions among the variables, for example, motivated by economic theory or
institutional features, to extract the exogenous shocks from the data and estimate the
parameters. Because the estimates that we show below build on SVAR models, we shall
limit our further discussion of the literature to this specific approach.6
2.1. Closed-Economy Models
Applying an SVAR approach to the US after World War II, Blanchard and Perotti
(2002) find that an increase in government purchases stimulates economic activity with
a peak multiplier that tends to be close to unity, that it boosts consumption, while it has
a negative effect on investment. Their main identifying assumption is that government
purchases do not contemporaneously react to output within the same quarter. The
authors also exploit detailed institutional information about the tax system, as well as
information on the elasticities of taxes and transfers to income. Fatás and Mihov (2001)
essentially confirm the results of Blanchard and Perotti (2002), except that they find a
(lagged) positive effect on investment. Auerbach and Gorodnichenko (2010) employ
the Blanchard and Perotti (2002) identification scheme to estimate regime-switching
SVARs that allow output multipliers to differ according to the business cycle situation.
Their average estimate of the peak in the multiplier is again close to unity. Mountford
and Uhlig (2009), who identify shocks by imposing theory-motivated signs on the
impulse responses, find only a weak positive effect on output and consumption, while
investment is crowded out.
Not all work focuses solely on the US. Burriel et al. (2010) follow the methodology of
Blanchard and Perotti (2002) in estimating impulse responses of government purchases for the aggregate euro area over the period 1981–2007 and comparing those
with the corresponding responses for the US. The output multipliers are similar and
typical below unity. Kirchner et al. (2010) also estimate VARs for the aggregate euro
area over roughly the same period but allow the effects of the government purchases
impulse to be time varying. They find that the short-run effectiveness of the impulse in
stimulating the economy increased until the end of the 1980s, after which it declined
again.
2.2. Open-Economy Models
Somewhat surprisingly, Kim and Roubini (2008) find that for the US an increase in
the budget deficit produces a short-run improvement of the current account,
regardless of whether the budgetary expansion is caused by an increase in public
spending or a reduction in taxes. They explain this finding by arguing that the direct
effect of the expansion on the current account is dominated by the extra private
saving (to pay for the higher future taxes) and the rise in the interest rate, which
depresses investment.
6
Favero and Giavazzi (2009) try to reconcile the narrative and SVAR methods and argue that the differences in impulse responses under these approaches are the result of the more limited information employed
under the narrative approach, in particular the neglect of the inter-temporal government budget constraint.
They show that feeding the exogenous shocks identified by Romer and Romer (2010) directly into the SVAR
yields responses comparable to those found by others under the ‘standard’ SVAR approach.
2011 The Author(s). The Economic Journal 2011 Royal Economic Society.
F10
THE ECONOMIC JOURNAL
[FEBRUARY
There is related work for other countries as well. Lane and Perotti (1998) link the
trade balance and its components to the different components of the public budget
for a large set of OECD countries over the period 1960–95. They find that the
composition of a change in fiscal policy and the exchange rate regime both matter for
the effects on the external account. In particular, higher wage government consumption produces a fall in exports and a deterioration of the trade balance, especially under flexible exchange rates.7 Imports are also negatively affected, or
unaffected, depending on the regression specification. Related analysis by Lane and
Perotti (2003), also for the OECD, essentially confirms these results. Moreover, they
find that an increase in non-wage government consumption has only a small effect on
the traded sector.
Monacelli and Perotti (2006) and Ravn et al. (2007) estimate impulse responses to an
increase in government purchases for Australia, Canada, the UK and the United States.
Estimations by the former are on a country-by-country basis, while the latter use a panel
VAR. The sample periods are roughly the same. Both studies show that an increase in
government purchases raises output and consumption but leads to a deterioration of
the trade balance.
As far as the consequences for the real exchange rate are concerned, these
articles find that it depreciates, while Bénétrix and Lane (2010) for a sample of
eleven EMU countries estimate an appreciation of the real effective exchange rate
after a positive government purchases shock.8 However, an application of their
methodology to the four-country sample of Monacelli and Perotti (2006) and Ravn
et al. (2007) does indeed confirm the appreciation and indicates that the difference
in results for the two samples may be traced to the exchange rate regime and
monetary policy.9 While the other group of countries maintains flexible exchange
rates, the EMU group has been characterised by (quasi) fixed exchange rates over
recent decades.
3. Effects of Government Purchases Shocks on GDP in EU Countries
This Section provides our estimates of the effect of government purchases shocks on
EU economies. We follow the SVAR methods, because alternative approaches are not
(practically) feasible. First, to follow the Ramey and Shapiro (1998) approach, we
would need a panel of large war-related movements in defence spending. There are
very few, if any, such events in our sample. Following an approach along the lines of
Romer and Romer (2010) is not practically possible, because a comparable dataset
based on official documents for discretionary spending in a large panel of European
countries does not exist.
7
With a fixed exchange rate regime the tendency for the real exchange rate to appreciate will be suppressed, hence the trade balance is better protected against the shock.
8
While the work of Bénétrix and Lane (2010) is related to ours, this article focuses more closely on the
consequences of government purchases shocks for output, the public budget and the (components of the)
trade balance. Moreover, in this article we explore explicitly the presence of anticipation effects, various
sample splits (in particular, between closed and open economies) and the role of direct fiscal spill-overs
among the EU countries.
9
This is also confirmed in a broader study by Corsetti et al. (2010) for a panel of OECD countries.
2011 The Author(s). The Economic Journal 2011 Royal Economic Society.
2011 ]
GOVERNMENT PURCHASES IN THE EU
F11
3.1. Methodology
Before turning to the actual estimation, some discussion about the methodology is
warranted. In the absence of any ‘active’ policy intervention, the public budget moves
automatically with the economic cycle. Government transfers, such as unemployment
benefits, increase as the economy slows down and unemployment rises, while at the
same time tax revenues on labour, capital income and consumption fall. The opposite
happens when the business cycle is improving. These automatic movements of the
government budget resulting from the business cycle are referred to as the cyclical
component of the budget. The structural component of the public budget (or the fiscal
stance) is obtained by subtracting the automatic cyclical component from the actual
budget balance. However, this structural component may also be systematically linked
to the economic cycle. For example, the government may reduce tax rates whenever
activity falls below potential. The component could also be systematically linked to
other variables such as the stock of outstanding public debt or inflation. These systematic responses will be referred to as the endogenous structural component. The other
part of the structural budget is called the exogenous component. Examples are an
increase in public purchases to finance a war or a politically-motivated increase in
transfers to the population prior to an election.
We extend the analysis in Beetsma et al. (2008) and explore the consequences of
unexpected changes in government purchases for domestic activity, its components,
the public budget and the external balance for a sample and sub-samples of 14 EU
countries over the period 1970–2004. This sample period avoids the influence on our
analysis of outliers associated with the extreme events during the recent crisis. We also
investigate the potential spill-overs on activity in the EU trading partners. The sample
countries are Austria, Belgium, Denmark, Finland, France, Ireland, Italy, Germany,
Greece, the Netherlands, Portugal, Spain, Sweden and the United Kingdom. The data
sources are the OECD Economic Outlook (EO), the Main Economic Indicators (MEI)
of the OECD Statistical Compendium and the IMF’s Direction of Trade Statistics
(DOTS). A more detailed description of the sample and the data sources is found in
the online Appendix.
Following Beetsma et al. (2008) and Bénétrix and Lane (2010), but in contrast to
some related literature, we use annual instead of quarterly data. This has a number of
advantages. First, the shocks we uncover with annual data may be closer to the actual
shocks, because new fiscal impulses do not usually take place at the quarterly frequency but more likely in the new budget and possibly in mid-year budgetary revisions. In addition, the effects of the potential anticipation of fiscal policy changes
should be less relevant, because the uncovered shocks are more likely to be truly
unanticipated. After all, a given policy shock is less likely to be anticipated one year
before it actually takes place than one quarter before it actually takes place. (The
presence of anticipation effects is formally tested in Section 4 (robustness) below.)
Further, there is less need to be concerned with the details of the institutional
setting. For example, if tax revenues are systematically larger in one quarter than in
another quarter (this is, in particular, the case for corporate tax revenues), then a
model based on quarterly data would take this feature into account. In addition, with
quarterly data, it is more likely that decisions on purchases take place in another
2011 The Author(s). The Economic Journal 2011 Royal Economic Society.
F12
THE ECONOMIC JOURNAL
[FEBRUARY
quarter than when the actual outlays are made. This raises the chance that the
identified shocks are wrongly dated. Finally, with annual data potential seasonality
effects are absent from the data.
A drawback of using annual data is that there are fewer observations available.10
Hence, to obtain more precise estimates, we estimate the VAR model in a panel format
(that is, we pool observations for a number of countries over a given sample period).
The new disadvantage then is that one needs to impose cross-country homogeneity on
the relationships among the variables. A number of econometric adjustments are made
to deal with this objection. In particular, we include country-specific constant terms and
country-specific linear time trends into the regression. Moreover, we include yearspecific effects to eliminate any cross-country contemporaneous residual correlation.
All these features reduce the amount of heterogeneity. In addition, the fact that EU-14
countries share many similarities is conducive to reducing heterogeneity. Nevertheless,
we will also investigate whether the effects of an increase in government purchases
depend on other factors, such as the degree of openness.
3.2. Baseline Impulse Responses
The set of endogenous variables in the baseline structural VAR consists of government
purchases g (the sum of government consumption and government investment),
cyclically-adjusted net taxes (with country-specific cyclical adjustment) nt, output
(GDP) y, the long-run nominal interest rate irl and the real effective exchange rate
reer.11 An increase in this latter variable amounts to a real domestic depreciation. All the
variables are real and in natural logarithms, except for the long-run interest rate, which
is in %. The variables are entered into the vector [g, nt, y, irl, reer]0 and identification is
based on a lower-triangular Cholesky decomposition according to this particular
ordering. Hence, the main identifying assumption is that government purchases are
not contemporaneously affected by the other variables, in particular not by GDP. This
assumption is not obviously satisfied. However, the cyclically-sensitive spending items
(in particular, social benefits and other transfers) are included in net taxes, which are
then cyclically adjusted. Further, changes in government purchases are usually contained in the budget law for the coming year, while adjustments during that year tend
to be of less importance. Below, we test the plausibility of this identification strategy by
imposing alternative orderings and non-zero contemporaneous restrictions. Observe
that, because we focus on impulse responses to government purchases shocks only, the
relative ordering of the other variables does not affect the impulse responses, as these
variables are all ordered after government purchases (Christiano et al., 1999).
We include cyclically-adjusted net taxes (Alesina et al., 2002) rather than unadjusted
net taxes to take account of the cross-country heterogeneity in the response of net taxes
to changes in output. If there is such heterogeneity, then not accounting for it may
10
However, even if we wanted to use quarterly data, this would not be possible for the country sample
under consideration. Truly (i.e. non-interpolated) quarterly data are only available for a subset of our sample
countries.
11
Of each variable we include two lags in the regression model, which is enough to get rid of any serial
correlation in the residuals. However, the results are robust to alternative lag lengths. We also check the
cross-country correlations in the residuals but in all instances these correlations are close to zero.
2011 The Author(s). The Economic Journal 2011 Royal Economic Society.
2011 ]
GOVERNMENT PURCHASES IN THE EU
F13
12
jeopardise the correct inferences on the dynamics of the VAR. An additional benefit
of including cyclically-adjusted net taxes is that we get a direct insight into the reaction
of the fiscal authorities to a spending shock, because it is cyclically-adjusted net taxes
that are directly under the government’s control. Obviously, since elasticities of net
taxes are not directly observable, cyclically-adjusted net taxes are imprecisely observed.
Therefore, in Subsection 4.3. we confirm our baseline findings by including unadjusted
net taxes into our model.
Figure 1 depicts the baseline impulse responses to a 1% of GDP increase in government purchases, while Table 2 provides the outcomes for specific moments after
the shock. On impact GDP rises significantly by 1.2%, while it peaks at 1.5% after one
year.13 The multiplier is quite large but not out of line with findings by others. Bénétrix
and Lane (2010) report a somewhat smaller, though not entirely comparable, response
of domestic relative to rest-of-EMU GDP. Cyclically-adjusted net taxes fall on impact,
probably because governments try to accommodate the spending impulse with a discretionary reduction in the tax burden in order to further stimulate the economy.
However, as expected, unadjusted net taxes (not reported here) increase on impact.
From the responses of its constituent variables, we can construct the response for the
primary budget. In particular, this response is calculated as
5.0
4.0
3.0
2.0
1.0
0.0
Government Spending (g)
0 1 2 3 4 5 6 7 8 9 10
Years After Shock
2.0
1.0
0.0
–1.0
–2.0
Cyclically Adjusted Net Taxes (nt)
60.0
40.0
20.0
0.0
–20.0
–40.0
0.5
0.0
–0.5
–1.0
–1.5
–2.0
0 1 2 3 4 5 6 7 8 9 10
Years After Shock
2.0
1.5
1.0
0.5
0.0
–0.5
Output (y)
0.5
Long Term Interest Rate (irl)
0 1 2 3 4 5 6 7 8 9 10
Years After Shock
Real Effective Exchange Rate (reer)
0 1 2 3 4 5 6 7 8 9 10
Years After Shock
Budget Balance (Constructed)
0.0
–0.5
–1.0
0 1 2 3 4 5 6 7 8 9 10
Years After Shock
0 1 2 3 4 5 6 7 8 9 10
Years After Shock
Fig. 1. Baseline Responses After a Government Purchases Shock of 1% of GDP
Notes. See note to Table 2. Confidence bands (the area between the dashed lines) are the
5th and the 95th percentiles from Monte Carlo simulations based on 2,000 replications.
The solid line gives the median over the replications.
12
The estimated link from output to net taxes will generally differ from the actual link in a specific
country, thereby affecting the dynamics of the VAR.
13
Statements about statistical ‘significance’ will always be based on the 10% level.
2011 The Author(s). The Economic Journal 2011 Royal Economic Society.
F14
[FEBRUARY
THE ECONOMIC JOURNAL
Table 2
Responses to a Government Purchases Increase of 1% of GDP
Impact effect
After one year
After three years
After five years
(a) Baseline
Government purchases
Net taxes
Output
Long-run interest rate
Real eff. exchange rate
Budget balance ⁄ GDP
4.15*
1.06*
1.17*
1.19
0.23
0.72*
4.42*
0.24
1.50*
2.57
0.43
0.47*
3.19*
0.21
1.19*
19.86
0.81*
0.21*
2.07*
0.13
0.66*
27.81*
0.62
0.18
(b) Baseline without time effects
Government purchases
Net taxes
Output
Long-run interest rate
Real eff. exchange rate
Budget balance ⁄ GDP
4.15*
0.98*
1.12*
6.23
0.33
0.73*
4.37*
0.16
1.48*
12.32
0.65
0.45*
3.05*
0.42
1.17*
26.33
0.91*
0.14
1.96*
0.45
0.49*
45.36*
0.63
0.16
(c) Baseline with quadratic time trends
Government purchases
4.15*
Net taxes
1.27*
Output
1.20*
Long-run interest rate
2.55
Real eff. exchange rate
0.27
0.76*
Budget balance ⁄ GDP
3.93*
0.48
1.42*
7.06
0.57
0.44*
2.23*
0.40
0.84*
5.73
0.98*
0.26*
1.07*
0.54
0.31
7.64
0.60
0.24*
(d) Baseline in first difference – no trends
Government purchases
4.15*
Net taxes
0.96*
Output
1.12*
Long-run interest rate
3.11
Real eff. exchange rate
0.14
Budget balance ⁄ GDP
0.73*
4.99*
0.39
1.54*
9.53
0.18
0.62*
5.82*
0.00
1.72*
10.78
0.79
0.66*
5.96*
0.01
1.67*
9.93
0.78
0.72*
Notes. The shock is an increase in government purchases equal to 1% of GDP. Further, * denotes statistical
significance at the 10% level. The impulse responses are expressed in percentage of the underlying variable,
except for the long-run interest rate, which is in basis points, and the budget balance ⁄ GDP ratio, which is in
percentage of GDP. The model in first differences displays the cumulative responses.
NTtNA Gt =Yt ¼ NT NA =Y N^ Tt þ nY^t Y^t ðG=Y Þ G^t Y^t ;
where NT NA, G and Y are, respectively, the anti-logs of nt NA (unadjusted or non-cyclicallyadjusted net taxes (real and in natural logarithm)), g and y, while a hat denotes the
percentage deviation from the initial value (the impulse response) and n is the average
elasticity of net taxes with respect to real output (calculated in Van den Noord, 2000; and
OECD, 2005). Hence, we construct unadjusted net taxes out of cyclically-adjusted net
taxes and output, using this elasticity. We evaluate the approximation at the overall sample
mean shares of G and NT NA over Y. Clearly, the fiscal impulse leads to a sharp
deterioration of the budget balance (0.7% on impact), which shrinks in the following
years. As far as the other variables are concerned, the long-term interest rate rises,
although with some lag after the shock, most likely as a result of the inflationary pressures
(see Subsection 4.1. below) exerted by the expansion of demand. In line with the
empirical finding of Bénétrix and Lane (2010) for the euro-area and the theoretical
prediction of Corsetti and Müller (2008) and Müller (2008), the real exchange rate
appreciates, although the appreciation becomes significant only after three years.
2011 The Author(s). The Economic Journal 2011 Royal Economic Society.
2011 ]
GOVERNMENT PURCHASES IN THE EU
F15
Panel (b) of Table 2 shows the results for the baseline model when we drop the time
effects. The impulse responses are qualitatively and quantitatively similar to those
reported in panel (a). As a further robustness check, we also estimate the baseline
model replacing the country-specific linear time trends with country-specific quadratic
time trends (see panel c, Table 2). Again, the impulse responses closely resemble those
under the baseline estimation. Only the effect on output dies out slightly faster than
under the baseline, while the long-run interest rate is no longer significant after five
years. Finally, in panel (d) we report the (cumulative) responses when we estimate
the model in first differences.14 The responses are rather similar to those under the
baseline. However, the effects are more persistent, because the temporary shock to the
change in government purchases produces a permanent shift in its level. This leads to a
larger and more persistent effect on output and the budget balance over GDP ratio in
the medium to longer run. Yet, the short-run responses of output and the budget
balance are very similar to those under the baseline.
4. Robustness Tests
4.1. Testing for Anticipation Effects
One of the problems with the use of VAR models to identify fiscal shocks is the so-called
foresight or anticipation effect. Leeper et al. (2008) explore its consequences for the
VAR analysis of fiscal shocks. In this Subsection we will show that the anticipation
problem is unlikely to be relevant for our analysis. More specifically, the recent literature has proposed a number of solutions to this problem. The first is again to follow
the narrative approach in identifying fiscal shocks (Ramey, forthcoming; Mertens and
Ravn, forthcoming). Recent literature also suggests other solutions to ameliorate
anticipation problems within the SVAR methodology. One approach, which is also
advocated by Ramey in her criticism of Blanchard and Perotti’s (2002) quarterly VAR
for the United States, is to estimate an SVAR with annual data as we do.15 If changes in
government purchases are anticipated over only a period of less than one year, then
shocks estimated at the annual level can be treated as unanticipated.
To test whether the anticipation problem is still relevant in our annual VAR, we run
Granger-causality tests as in Ramey (forthcoming). In particular, we test whether realtime fiscal forecasts have any predictive power for the government purchases shocks
estimated from our model. In her study for the US, Ramey (forthcoming) uses government spending forecasts from the Survey of Professional Forecasters. However, such
information is not available for our EU country sample. Therefore, we use an alternative
dataset consisting of real-time one-year ahead forecasts for the budget balance–GDP
14
Panel unit root tests suggest that all the variables in our baseline model are integrated of order one.
Therefore, specifying the VAR model in levels is the correct approach if there is at least one co-integration
relationship among the variables of the system. Estimating the model in first differences may be the appropriate specification in case there is no co-integration relationship among the variables. However, in a VAR
differencing leads to information being thrown away (e.g., the VAR cannot capture a co-integration relationship), whereas this produces almost no gains (Doan, 2006). Despite these drawbacks, we want to test that
the two approaches do not lead to substantially different results.
15
In a recent article, Mertens and Ravn (2010) derive a fiscal VAR estimator applicable when spending
shocks are anticipated. They apply this estimator to US quarterly data and find that anticipation effects do not
overturn existing results from the standard VAR literature.
2011 The Author(s). The Economic Journal 2011 Royal Economic Society.
F16
THE ECONOMIC JOURNAL
[FEBRUARY
ratio. These data have been produced by the European Commission twice a year since
the 1970s.16 We use the Fall forecasts, which are based on official fiscal plans by the
individual governments. In fact, using the Fall forecasts, we are stacking the odds
against ourselves in favour of the hypothesis that anticipation effects play a role, because, if present, the anticipation of policy changes in period t should be stronger in the
autumn of t 1 data than in data produced at an average arbitrary moment in period
t 1.
Before we test whether the identified government purchases shocks are anticipated,
we verify that the forecast for the budget balance–GDP ratio does indeed have predictive power for government purchases and, hence, that it is a potentially relevant
variable for capturing the anticipation of government purchases shocks. In particular,
we run a fixed-effect panel regression of government purchases g in year t on its own
two lags and the balance–ratio forecast produced in year t 1 for year t.17 The coefficient of the latter is estimated significantly with a p-value of 0.007, indicating that our
balance–ratio forecast has strong predictive power for g.
We then estimate a fixed-effect panel regression of the government purchases shocks
identified from the baseline model on two of their own lags and the balance–ratio
forecast. This latter variable does not seem to have any predictive content for the
government purchases shock (p-value of its coefficient is 0.524). These results are
consistent with the hypothesis that the annual innovations in government purchases
estimated with our VAR are not anticipated.
In order to test this preliminary conclusion further, we follow three alternative routes
suggested by the literature for dealing with anticipation effects in the context of
quarterly VAR models. Yang (2007) argues that including ‘forward-looking’ variables in
the VAR helps to assuage the foresight problem, because those variables capture
information about the future effects of fiscal shocks.18 In particular, she finds that by
augmenting a standard VAR model a la Blanchard and Perotti (2002) with short-term
interest rates and prices, the response of output to tax shocks weakens substantially.
This provides informal and indirect evidence that these additional variables capture
agents’ foreknowledge of tax shocks. Following this approach, we extend our baseline
model with the log of the GDP deflator and the short-run nominal interest rate. The
results are shown in panel (a) of Table 3. For all common variables the impulse
responses are qualitatively and quantitatively very similar to those under the baseline.
The price level is restricted to remain unaltered upon impact and then becomes
significantly positive, most likely because of the increased demand pressure resulting
from the government purchases increase.
According to Forni and Gambetti (2010) the inclusion of forward-looking variables
like stock prices into the VAR seems particularly conducive to correctly estimating the
fiscal shock and reducing fiscal-foresight problems. We also investigate this suggestion
16
We thank Melander et al. (2007) for providing this dataset to us.
The results are reported in the online Appendix.
This argument has recently also been stressed by Sims (2009), who uses simulations to show that even
when a model contains non-invertibilities (because important state variables cannot be observed by the
econometrician), VAR innovations may perform quite well in identifying shocks. To reduce any potential bias,
he recommends including as many observable states and other forward-looking variables as is feasible in the
VAR.
17
18
2011 The Author(s). The Economic Journal 2011 Royal Economic Society.
2011 ]
GOVERNMENT PURCHASES IN THE EU
F17
Table 3
Robustness: Anticipation Effects
Impact effect
After one year
After three years
After five years
(a) Baseline with short-run interest rates and prices
Price level
0.00
Government purchases
4.15*
Net taxes
0.97*
Output
1.19*
Long-run interest rate
1.53
Real eff. exchange rate
0.15
Short-run interest rate
0.04
Budget balance ⁄ GDP
0.70*
0.28*
4.34*
0.19
1.49*
2.28
0.31
7.59
0.45*
0.67*
3.10*
0.28
1.11*
28.98*
0.71*
17.43
0.21
0.94*
1.97*
0.23
0.58*
32.77*
0.46
22.16*
0.18
(b) Baseline with stock prices
Government purchases
Net taxes
Output
Long-run interest rate
Real eff. exchange rate
Stock prices
Budget balance ⁄ GDP
4.15*
0.70
1.05*
14.43*
0.37
2.14
0.70*
4.49*
0.43
1.19*
27.41*
0.48
1.11
0.66*
3.18*
0.43
0.93*
31.94*
0.76
4.50*
0.27*
2.11*
0.57
0.63*
19.51*
0.82*
3.03*
0.11
(c) Baseline on 8-country sample (B8)
Government purchases
4.06*
Net taxes
1.99*
Output
1.53*
Long-run interest rate
14.57
Real eff. exchange rate
0.26
Budget balance ⁄ GDP
0.76*
4.62*
1.90*
1.77*
13.95
0.01
0.77*
3.22*
0.56
1.36*
19.02
0.16
0.31*
1.94*
0.40
0.87*
19.68
0.44
0.00
(d) B8 with fiscal forecasts as exogenous variable
Government purchases
4.06*
Net taxes
2.09*
Output
1.54*
Long-run interest rate
7.38
Real eff. exchange rate
0.26
Budget balance ⁄ GDP
0.77*
4.54*
1.96*
1.76*
5.65
0.09
0.77*
2.82*
0.53
1.12*
2.11
0.36
0.31*
1.39*
0.26
0.51
1.24
0.49
0.05
Notes. See note to Table 2.
by augmenting the baseline model with the log of nominal stock prices. The impulse
responses, reported in panel (b) of Table 3 are not too different from those under the
baseline. Net taxes remain insignificant on impact, although quantitatively the difference from the baseline impact is limited. Output is again significant with a multiplier
larger than one and a peak reached after one year. Both are smaller though than under
the baseline. The budget balance–GDP ratio response is very similar to that under the
baseline. The main difference concerns the long-run interest rate, which now becomes
significant upon impact.
As an additional, and final, test we augment the baseline model with the balance–
ratio forecast we used earlier as an exogenous variable. This approach is in the spirit of
that followed by Auerbach and Gorodnichenko (2010) in dealing with expectations of
changes on government purchases.19 In order to have a sufficiently balanced sample
19
We include the balance–ratio forecast as an exogenous variable, because we find it conceptually difficult
to interpret the forecast as a function of lagged forecasts of the system’s variables. Nevertheless, including it as
an endogenous variable (ordered first) into the system has no effect on the results.
2011 The Author(s). The Economic Journal 2011 Royal Economic Society.
F18
THE ECONOMIC JOURNAL
[FEBRUARY
that is, moreover, in terms of sample period comparable to that for our previous
estimates, we only estimate this model for those countries for which these forecasts are
available since the 1970s. The resulting sample consists of eight countries: Belgium,
Denmark, France, Germany, Ireland, Italy, the Netherlands and the UK. To explore the
consequences of the expansion of the baseline model for this reduced set of countries,
panels (c) and (d) of Table 3 show the results for, respectively, the baseline model and
the baseline model augmented with the balance–ratio forecast. The impulse responses
are qualitatively and quantitatively very similar for the two cases. Only the reaction of
the long-run interest rate is more subdued in the augmented model, although it is
insignificant in both cases.
4.2. Testing the Identifying Restrictions
A key to identification concerns the assumption of a zero within-year response of
government purchases to output. This assumption is also made by Bénétrix and Lane
(2010) for their annual-data VAR model. While the zero response can reasonably be
expected to hold for quarterly data, its fulfilment is not a priori assured for annual data.
However, in Beetsma et al. (2006, 2009a,b) we show how to assess the plausibility of this
identifying assumption for an annual VAR, based on estimates from a quarterly model
that assumes that government purchases take at least one quarter to react to an output
shock. There we apply our method to a VAR in government purchases and output for
the largest available subset of the countries in our current sample for which we have
sufficient quarterly fiscal data. We show that it is indeed reasonable to identify the
annual fiscal SVAR by imposing a zero within-year impact of output on government
spending. These results are consistent with recent empirical evidence by Born and
Müller (2009), who perform several tests and provide evidence that annual government
spending is indeed predetermined in the US.
Following Beetsma et al. (2006, 2008), below we assess the legitimacy of our identifying restriction in two more ways. First, we show that the main conclusions of our
model are unaffected if we impose a plausible range of contemporaneous responses of
public spending to output. While our baseline assumes that the contemporaneous
reaction of government purchases to output is zero, the results reported in panel (a) of
Table 4 assume a contemporaneous response of government purchases to output
of 0.20 (a 1% increase in output lowers purchases by 0.20%), while those reported in
panel (b) impose a contemporaneous response of 0.20. Overall, this implies a rather
wide range for the purchases response, which we can reasonably expect to contain
the true within-period response, given that government purchases are largely predetermined by the official budget and contain no components that automatically
fluctuate with the business cycle. A lower contemporaneous response requires a larger
impact effect of the purchases shock on output to explain the data. However, for the
indicated range, the output multiplier is always quite substantial and above one after
one year. The response of the budget balance–GDP ratio is only mildly affected by the
imposed reaction of government purchases to output. With a positive contemporaneous reaction of purchases to output, the deterioration in the budget balance–GDP ratio
is larger than in the other cases. However, the largest difference concerns cyclicallyadjusted net taxes which respond with a smaller reduction when the imposed purchases
2011 The Author(s). The Economic Journal 2011 Royal Economic Society.
2011 ]
F19
GOVERNMENT PURCHASES IN THE EU
Table 4
Robustness: Identification Restrictions and Alternative Specifications
Impact effect
After one year
After three years
After five years
(a) Baseline with response of g to y set at 0.20
Government purchases
4.15*
Unadjusted net taxes
1.39*
Output
1.46*
Long-run interest rate
2.86*
Real eff. exchange rate
0.06
Budget balance ⁄ GDP
0.67*
4.52*
0.43
1.84*
4.77
0.42
0.39*
3.40*
0.17
1.43*
20.26
0.99*
0.16
2.26*
0.10
0.79*
29.25*
0.78*
0.18
(b) Baseline with response of g to y set at 0.20
Government purchases
4.15*
Net taxes
0.69
Output
0.86*
Long-run interest rate
1.33
Real eff. exchange rate
0.37
Budget balance ⁄ GDP
0.78*
4.31*
0.03
1.14*
2.07
0.41
0.55*
2.96*
0.24
0.93*
18.51
0.59
0.26*
1.84*
0.16
0.53*
26.01*
0.46
0.18*
(c) Baseline with private output before g
Private output
0.00
Government purchases
4.15*
Net taxes
0.84*
Long-run interest rate
0.92
Real eff. Exchange rate
0.37
Output
1.00*
0.75*
Budget balance ⁄ GDP
0.28*
4.37*
0.18
2.23
0.45
1.26*
0.55*
0.38
3.06*
0.15
17.99
0.65
1.02*
0.26*
0.17
1.94*
0.12
25.84*
0.52
0.59*
0.19*
(d) Baseline with unadjusted net taxes
Government purchases
4.15*
Unadjusted net taxes
1.16*
Output
1.18*
Long-run interest rate
2.18
Real eff. exchange rate
0.21
Budget balance ⁄ GDP
0.78*
4.45*
2.34*
1.52*
4.87
0.40
0.60*
3.26*
1.83*
1.25*
14.66
0.78*
0.42*
2.13*
0.57
0.73*
24.67*
0.65
0.41*
(e) Adding the debt ratio to the baseline
Government purchases
4.15*
Net taxes
0.75
Output
1.24*
Long-run interest rate
7.35
Real eff. exchange rate
0.14
Debt ⁄ GDP
0.28
Budget balance ⁄ GDP
0.63*
4.30*
0.46
1.61*
10.80
0.54
0.38
0.24*
3.03*
1.02*
1.21*
12.41
1.10*
0.96
0.01
1.75*
0.98*
0.60*
16.55
0.79*
1.66*
0.05
(f ) Adding the real compensation rate
Government purchases
4.15*
Net taxes
1.24*
Output
1.13*
Long-run interest rate
9.39
Real eff. exchange rate
0.13
Real compensation rate
0.56*
Budget balance ⁄ GDP
0.78*
4.30*
0.19
1.35*
15.71
0.25
0.43*
0.50*
3.01*
0.19
1.01*
30.00*
0.65
0.47*
0.25*
1.92*
0.25
0.57*
30.47*
0.44
0.56*
0.16
Notes. See note to Table 2.
response increases. Presumably, the reason is that the policymakers want to limit the
budgetary effects of the positive purchases response to output.
The second way to justify our identification assumption is to remove the government
purchases component from output and position the resulting variable, referred to as
2011 The Author(s). The Economic Journal 2011 Royal Economic Society.
F20
[FEBRUARY
THE ECONOMIC JOURNAL
20
‘private output’, before government purchases in the VAR. This way we relax the
assumption of a zero contemporaneous response of government purchases to
economic activity, while purging the automatic link between output and purchases,
because government purchases (to the extent that they are not imported) are part of
output. The (total) output response is by construction unity on impact and exceeds
unity in the next two years (see Table 4, panel c), which is qualitatively in line with the
baseline. The responses of the other variables are much in line with the baseline
responses, with net taxes being significantly negative on impact with a response that is
quantitatively similar to that under the baseline. Also the response of the budget balance is quantitatively similar to that under the baseline.
4.3. Alternative Specifications
This subsection explores the robustness of the baseline specification into a number of
dimensions. First, we re-estimate the model by including unadjusted rather than
cyclically-adjusted net taxes into the VAR. The results are found in Table 4, panel (d).
Re-assuringly, the impulse responses for the other variables are close to the corresponding responses under the baseline. Those for government purchases, output and
the real exchange rate are very close, while the budget balance–GDP ratio is slightly
more persistent in the case of unadjusted net taxes. Not surprisingly, we see that net
taxes are positively affected by the shock, because of the positive effect of the output
increase on the cyclical component of net taxes.
In panel (e) of Table 4, we report the results of adding the debt ⁄ GDP ratio as an
endogenous variable to the baseline model to control for sustainability concerns of
the fiscal authorities in setting their policies. We add the debt ratio after government
purchases into the system. However, adding it as the first variable would not have
changed the results. The impulse responses are only mildly affected by the inclusion
of debt into the model. The response of output remains quantitatively very similar,
while the real exchange rate again becomes significant after three years. The budget
balance–GDP ratio is somewhat less persistent though, while cyclically-adjusted net
taxes are no longer significant on impact and rise in the medium-run. The debt ⁄ GDP
ratio rises steadily to reach a peak of 2.2% after eight years, after which it starts to
decline slowly.
As a final alternative we extend the baseline model with the real wage rate of
the economy. As our theory section suggests, the real wage rate may be an
important transmission channel of an increase in government purchases onto the
rest of the economy. Here we proxy the theoretical concept of the real wage rate
with the real compensation rate. Panel (f ) of Table 4 reports the results. After the
shock, the real compensation rate increases on impact and remains significant for
some years. The other responses are essentially unchanged, though. Only the longterm interest rate increases faster, while the real effective exchange rate appreciates
by less and is no longer significant after three years (although it gets very close to
significance).
20
Incidentally, the results are also robust to a switch in the ordering of g and nt in the model.
2011 The Author(s). The Economic Journal 2011 Royal Economic Society.
2011 ]
GOVERNMENT PURCHASES IN THE EU
F21
4.4. Varying the Sample Period
We now explore the robustness of our findings by varying the sample period. The
first variant drops the EMU period. Hence, the sample runs over the years 1970–98.
The second variant drops the 1970s which may be considered a somewhat unusual
period because of the oil shocks. The sample period in this case is 1980–2004. This
variant may also help us in detecting potential structural changes. In particular,
ongoing financial innovation may have reduced credit restrictions in the private
sector, making fiscal expansion less effective. The impact effect on output is indeed
larger for the first sub-sample (see Table 5, panels a and b). However, after the
impact the difference in the output responses shrinks, despite the fact that the longterm interest rate rises by more and the real effective exchange rate appreciates by
more in the second sub-sample.
4.5. Varying Exchange Rate Flexibility
Recent empirical studies have explored how the effects of fiscal impulses depend on
the exchange rate regime (e.g., Lane and Perotti, 1998, 2003). However, our sample
countries have gone through different exchange regimes during our sample period.
It is not obvious in our context of a panel VAR how its parameters can be made
exchange rate regime dependent, while a further complication is that not all the
countries were in the same regime at the same moment. To obtain some indication
of the consequences of differences in exchange rate flexibility for the effectiveness
of a government purchases shock, we follow a slightly alternative route and focus on
the sub-sample of EMU countries. This is the full sample minus Denmark, Sweden
and the United Kingdom. These countries are not part of the euro-area and never
had the intention to become part of it. Hence, it is not obvious how their exchange
rate flexibility has changed. The other countries are part of a common
currency area since 1999 and in the decades before that they have followed a policy
of gradual reduction of exchange rate flexibility against the German mark. In
particular, while the early years of the European Monetary System were characterised by frequent shifts of the official fluctuation bands around the exchange rates,
after the Basle-Nyborg agreement in 1987 exchange rates became more stable.
Therefore, we split our sample period into the two sub-sample periods 1970–87 and
1988–2004, and we estimate for each sub-period a panel VAR on our EMU-country
sample. Neither of the two subsamples is characterised by perfectly flexible or
perfectly fixed exchange rate regimes. However, within our entire sample, this
provides the maximum contrast in terms of exchange rate flexibility between two
sub-periods.
Table 5, panels (c) and (d), reports the impulse responses. They are similar for
output. In the first sub-period, the long-term interest rate falls in the short run, which is
most likely the result of monetary accommodation of the shock (with the central bank’s
policy instrument moving the entire yield curve). In the second sub-period, cyclicallyadjusted net taxes accommodate the spending shock more strongly. Most interesting is
the difference in behaviour of the real effective exchange rate, which suggests that the
two sub-periods do indeed differ in terms of exchange rate flexibility. In the first
2011 The Author(s). The Economic Journal 2011 Royal Economic Society.
F22
[FEBRUARY
THE ECONOMIC JOURNAL
Table 5
Robustness: Various Sample Splits
Impact effect
After one year
After three years
After five years
(a) Sample period 1970–98
Government purchases
Net taxes
Output
Long-run interest rate
Real eff. exchange rate
Budget balance ⁄ GDP
4.10*
1.05*
1.44*
2.67
0.25
0.60*
4.65*
0.02
1.81*
4.41
0.23
0.35*
3.28*
0.20
1.36*
26.85
0.59
0.17
1.89*
0.07
0.70*
30.72*
0.57
0.17
(b) Sample period 1980–2004
Government purchases
Net taxes
Output
Long-run interest rate
Real eff. exchange rate
Budget balance ⁄ GDP
4.16*
1.19*
1.06*
13.96
0.37
0.77*
4.67*
0.08
1.56*
26.32*
0.78*
0.40*
3.75*
0.73
1.42*
41.08*
0.95*
0.07
2.62*
0.58
0.84*
35.70*
0.55
0.10
(c) EMU-11, sample period 1970–87
Government purchases
4.30*
Net taxes
0.64
Output
1.36*
Long-run interest rate
26.24
Real eff. exchange rate
0.07
Budget balance ⁄ GDP
0.63*
4.56*
0.66
1.93*
44.52*
0.77
0.23
3.10*
0.15
1.39*
6.84
0.21
0.19
1.72*
0.06
0.67
37.53
0.76
0.16
(d) EMU-11, sample period 1988–2004
Government purchases
4.44*
Net taxes
1.10*
Output
1.32*
Long-run interest rate
6.84
Real eff. exchange rate
0.52
Budget balance ⁄ GDP
0.63*
4.61*
0.25
1.61*
15.83
1.32*
0.34*
3.48*
0.81
1.18*
17.34
1.24*
0.05
2.31*
0.86
0.67
10.60
0.46
0.02
(e) ‘Closed’ economies
Government purchases
Net taxes
Output
Long-run interest rate
Real eff. exchange rate
Budget balance ⁄ GDP
4.48*
0.86
1.36*
5.74
0.05
0.62*
3.93*
1.11
1.57*
8.53
0.56
0.00
2.81*
0.74
1.15*
30.52
1.66*
0.00
1.90*
0.64
0.55
31.28
1.36*
0.07
( f ) ‘Open’ economies
Government purchases
Net taxes
Output
Long-run interest rate
Real eff. exchange rate
Budget balance ⁄ GDP
3.86*
1.06
0.79*
4.71
0.62
0.88*
4.34*
1.90*
0.88*
4.80
0.67
1.16*
2.78*
0.53
0.53
8.39
0.62
0.60*
1.27*
0.28
0.11
15.39
0.43
0.21
Notes. See note to Table 2.
sub-period the real effective exchange rate hardly moves, while in the second subperiod the real exchange rate exhibits an appreciation, because the nominal exchange
rate can no longer move to offset the price inflation.21
21
The difference in reaction of the real effective exchange rate is in line with the earlier suggestion that
the direction of movement of the real exchange rate after a government purchases shock depends on the
exchange rate regime. The average flexibility of the exchange rate regimes in our first sub-period could be
too low to produce a significant depreciation of the real effective exchange rate.
2011 The Author(s). The Economic Journal 2011 Royal Economic Society.
2011 ]
GOVERNMENT PURCHASES IN THE EU
F23
4.6. ‘Closed ’ Versus ‘Open’ Economies
Ideally, we would want to allow for country-specific heterogeneity in the VAR parameters. However, this is practically infeasible. Nevertheless, we can explore the consequences of some sample splits. One of the most significant differences among our
countries is their degree of openness. Indeed, the theoretical discussion above implies
that the effects of an increase in government purchases on the trade balance may
depend on the degree of trade openness. Therefore, in this Subsection we split our
country-sample into ‘closed’ and ‘open’ economies, where those countries for which
the ratio of exports plus imports over GDP has on average over time been in the upper
(lower) half of the sample are classified as ‘open’, while the other countries are referred to as ‘closed’. The open economies are Austria, Belgium, Denmark, Ireland, the
Netherlands, Portugal and Sweden. The average degree of openness among these
countries is 0.88, while among the other group of countries it is 0.49. Hence, on
average the two groups differ substantially in their degree of openness. Table 5, panels
(e) and (f ), reports for our baseline specification, estimated separately for each of the
two groups of countries, the impulses responses to a 1% of GDP government purchases
shock.
There are quite substantial differences between the impulse responses for the two
groups of countries. The output response is weaker on impact for the open than for the
closed economies (0.79 versus 1.36). The response for the open economies lies outside
the 90% confidence band around the response for the closed economies, and vice versa,
indicating that on impact the responses are statistically different. Also in the years after
the shock the difference in responses remains significant or close to significance. The
output response is weaker for the open economies, despite the fact that net taxes are
much more accommodative for these economies.22 They become significantly negative
after one year, while for the closed economies they become positive after one year. Also
the movement in the long-run interest rate is less accommodative for the closed economies. For this group its response is positive though insignificant, while for the open
economies its movement is a lot smaller. Finally, for the closed economies the real
effective exchange rate exhibits a substantial and long-lasting appreciation, while for the
open economies it shows relatively little movement. An explanation might be that the
open economies are more exposed to foreign competition implying that the response of
prices is more subdued (the real effective exchange rate is CPI based – see the online
Appendix). All three factors (net taxes, long-run interest rate and real effective exchange
rate) contribute to giving the closed economies a relative disadvantage in terms of output
response. The fact that the output response for this group is nevertheless quite a bit larger
suggests that the leakage effect is substantially larger for the open economies.
4.7. Country-by-Country Estimation
To investigate the potential presence of heterogeneity further, we estimate a slimmeddown version of the baseline model, in this case a VAR of the vector [g, nt, y], on a
22
We do not have an obvious explanation for the difference in the degree of net tax accommodation.
Possibly, the authorities of the open economies are aware of the limited effectiveness of an increase in
government purchases and reduce net taxes to boost the effect of the purchases increase.
2011 The Author(s). The Economic Journal 2011 Royal Economic Society.
F24
[FEBRUARY
THE ECONOMIC JOURNAL
country-by-country basis. We shrink the original model, because each country has a
maximum of 35 observations, which would be on the low side given the number of
parameters to be estimated in the original model. Again we use a Cholesky decomposition with government purchases ordered first. Figure 2 shows the mean responses
for both the full country sample and the groups of closed and open economies. All
mean responses are rather similar to the original ones based on the panel VARs,
though they are slightly muted, because the persistence in the government purchases
response is a bit lower. To give an indication of the dispersion of the responses, the
Figure also shows the next-to-lowest and the next-to-highest points in the fourteen
individual country responses. This corresponds to roughly an 85% confidence interval.
5. Effects on Government Purchases Shocks on GDP Components
Following earlier contributions in the literature, we now break up GDP into its components and we feed those separate components into the panel VAR. The vector of
endogenous variables now becomes [g, nt, x, c, i, m, reer]0 , where x is exports, c is private
consumption, i is private investment and m is imports, all variables being real and in
natural logarithms. To avoid making the VAR too large, we drop the long-run interest
rate, which in the baseline became significant only after five years. Identification is
again based on the lower-triangular Cholesky decomposition. We include the components x and m of the trade balance share of GDP (tby) as separate variables in the model, in
order to help us in tracing the sources of trade balance movements (and to shed light on
the twin-deficits hypothesis). Further, by including consumption, we can see if we can
reproduce the Keynesian effects found by Galı́ et al. (2007) among others; while by
incorporating investment, we can explore whether the positive short-run effects of fiscal
expansion on investment, as predicted by some versions of the neo-classical models
discussed above and the open-economy extension of Corsetti and Müller (2008), carry
over to the current sample.
Figure 3 and panel (a) of Table 6 show the impulse responses for our seven
endogenous variables, as well as impulse responses for output, the budget balance
and the trade balance constructed out of the responses for the endogenous variables.
For the trade balance share of GDP we construct the impulse responses as
2.5
2.0
1.5
1.0
0.5
0
–0.5
–1.0
–1.5
1
2
3
4
5
6
7
8
9
10
Years After Shock
Fig. 2. Means and Extremes of Individual Country Responses
Notes. The solid line is the mean over the full sample; the dashed line is the mean over the
closed economies; the dashed-dotted line is the mean over the open economies; and the
outer lines represent to the next-to-lowest and next-to-highest points in the 14 individual
responses.
2011 The Author(s). The Economic Journal 2011 Royal Economic Society.
2011 ]
GOVERNMENT PURCHASES IN THE EU
Government Spending (g)
4.0
2.0
0.0
2.0
0.0
–2.0
1.0
0.0
–1.0
–2.0
1.5
0 1 2 3 4 5 6 7 8 9 10
Years After Shock
Cyclically Adjusted Net Taxes (nt)
0 1 2 3 4 5 6 7 8 9 10
Years After Shock
Export (x)
0 1 2 3 4 5 6 7 8 9 10
Years After Shock
Consumption (c)
0.5
–0.5
6.0
2.0
–2.0
3.0
1.0
–1.0
0 1 2 3 4 5 6 7 8 9 10
Years After Shock
Investment (i)
F25
Import (m)
0 1 2 3 4 5 6 7 8 9 10
Years After Shock
Real Effective Exchange Rate (reer)
0.0
–1.0
–2.0
1.5
0.5
–0.5
0.5
0.0
–0.5
–1.0
0.5
0 1 2 3 4 5 6 7 8 9 10
Years After Shock
Output (Constructed)
0 1 2 3 4 5 6 7 8 9 10
Years After Shock
Budget Balance (Constructed)
0 1 2 3 4 5 6 7 8 9 10
Years After Shock
Trade Balance (Constructed )
–0.5
0 1 2 3 4 5 6 7 8 9 10
Years After Shock
–1.5
0 1 2 3 4 5 6 7 8 9 10
Years After Shock
Fig. 3. Effects on GDP Components After a Government Purchases Shock
Notes. See note to Figure 1.
^t Y^t Þ, where X and M are the anti-logs of x and m,
ðX =Y ÞðX^t Y^t Þ ðM =Y ÞðM
respectively, and evaluate this expression at the overall sample mean shares of X and M
over Y. The reaction of net taxes is very similar to that under the baseline. Consumption and private investment indeed both increase.
In particular, the reaction of investment is strong (a peak response of 3.3% after one
year). Consumption reaches a peak of 1.1% after one year. The constructed GDP
response resembles the corresponding response for the baseline model. Also in line
with our earlier finding, the real effective exchange rate appreciates, though again with
some lag. The result of this appreciation is a fall in exports that becomes significant
after one period. The fall in exports is consistent with what Lane and Perotti (1998,
2003) find if the increase in government purchases falls mainly on government wage
consumption. Imports rise in line with the higher income being partly spent on foreign
products. Moreover, they are helped by the real exchange rate appreciation. The
constructed budget balance again shows a sharp deterioration on impact, after which it
restores gradually to zero. The constructed trade balance falls on impact by more than
0.5% of GDP and reaches a minimum of 0.8% of GDP. The combined effects on the
budget balance and the trade balance provide support for the twin-deficits hypothesis
2011 The Author(s). The Economic Journal 2011 Royal Economic Society.
F26
[FEBRUARY
THE ECONOMIC JOURNAL
Table 6
Responses to a Government Purchases Increase: GDP Split into Components
Impact effect
After one year
After three years
After five years
(a) All countries
Government purchases
Net taxes
Exports
Consumption
Investment
Imports
Real eff. exchange rate
Output
Budget balance ⁄ GDP
Trade balance ⁄ GDP
4.15*
0.90*
0.46
0.91*
1.85*
1.21*
0.30
1.31*
0.63*
0.55*
4.49*
0.24
0.87*
1.13*
3.30*
1.69*
0.61
1.49*
0.49*
0.84*
3.28*
0.02
0.53
0.75*
2.59*
1.39*
0.99*
1.06*
0.34*
0.63*
2.17*
0.44
0.20
0.43
1.36
0.94*
0.76*
0.64*
0.34*
0.38*
(b) ‘Closed’ economies
Government purchases
Net taxes
Exports
Consumption
Investment
Imports
Real eff. exchange rate
Output
Budget balance ⁄ GDP
Trade balance ⁄ GDP
4.48*
0.75
0.58
1.36*
2.28*
1.08*
0.31
1.82*
0.40*
0.39*
4.04*
0.75
1.16
1.46*
4.22*
0.93
1.00
2.03*
0.09
0.49*
2.85*
0.01
0.48
0.91*
3.32*
1.26*
1.96*
1.36*
0.07
0.41*
1.96*
0.79
0.58
0.61
2.16*
0.87
1.72*
0.84*
0.25
0.34*
(c) ‘Open’ economies
Government purchases
Net taxes
Exports
Consumption
Investment
Imports
Real eff. exchange rate
Output
Budget balance ⁄ GDP
Trade balance ⁄ GDP
3.86*
0.87
0.71
0.21
0.40
0.50
0.61
0.65*
0.90*
0.54*
4.23*
1.81*
0.64
0.41
0.50
1.37*
0.64
0.53*
1.27*
0.89*
2.66*
0.36
0.63
0.15
0.01
0.55
0.43
0.25
0.66*
0.53*
1.06
0.28
0.61
0.18
0.99
0.51
0.32
0.05
0.23
0.05
(d) Only EU goods trade
Government purchases
Net taxes
Goods exports to EU
Output
Goods imports from EU
REER w.r.t. EU
Budget balance ⁄ GDP
Trade balance ⁄ GDP
4.15*
1.11*
0.22
1.18*
2.28*
0.06
0.73*
0.46*
4.36*
0.33
0.64
1.55*
2.97*
0.32
0.46*
0.67*
3.18*
0.42
0.48
1.24*
2.75*
0.47
0.14
0.60*
2.08*
0.28
0.26
0.74*
1.77*
0.15
0.12
0.37*
Notes. See note to Table 2. The trade balance is expressed in % of GDP. Further, the responses for output
(except in panel d ), the budget balance and the trade balance are constructed out of their components.
Finally, ‘REER’ = real effective exchange rate.
in the sense that there is a positive correlation between the two variables conditional on
a government purchases shock.
Again, we also split the sample into our groups of ‘closed’ and ‘open’ economies.
The impulse responses for the adjusted model confirm the differences between the two
groups found earlier for the baseline model (see panels b and c of Table 6). While for
the closed economies consumption and investment exhibit strong positive responses
2011 The Author(s). The Economic Journal 2011 Royal Economic Society.
2011 ]
GOVERNMENT PURCHASES IN THE EU
F27
(with respective peak effects of 1.5 and 4.3%), the corresponding responses are not
significant and quantitatively small (though positive) for the other group. This contrasts with the theoretical results of Corsetti and Müller (2008) we discussed earlier.
Again, the response of output is constructed out of the responses of its components.
For the closed economies it reaches a maximum of 2.0% after a year, while for the open
economies it reaches a maximum of 0.65% (achieved on impact), confirming that for
this group a large fraction of the impulse leaks away. Moreover, for this group the
stimulus to output fades away rather quickly. As far as the effects on the public budget
are concerned, for both groups it deteriorates on impact, although the effect is more
than twice as large for the open economies. Moreover, for the closed economies the
budget is back into balance after one year, while for the open economies it deteriorates
further and only returns to balance in the medium run. One reason for the larger
immediate deterioration for the open economies is the smaller increase in GDP. The
other is the substantial reduction in net taxes for these economies, which becomes
significant after one year. Finally, comparing the effects on the trade balance as a share
of GDP, we see that the reduction is larger in magnitude for the open than for closed
economies (at least over the first two years). A major source of the fall in the trade
balance is the rise in imports, which in turn is to a large extent driven by the increase in
GDP. Hence, the substantially-larger GDP response for the closed economies likely
suppresses the difference in trade balance responses between the two groups of
economies. Otherwise, the difference would have been even larger.
Because we are particularly interested in the benefits of fiscal stimulus to EU partner
countries, we limit ourselves now to exports to and imports from EU countries only.
These intra-European trade data are constructed from the bilateral trade data in the
DOTS and these only report trade in goods. However, most of the international trade is
trade in goods. In our sample imports and exports of goods are on average 28% of
GDP, while the corresponding figure for services is about 5% on average. Further, one
would generally consider data on cross-border trade in goods more reliable than data
that also cover trade in services. Hence, the non-availability of data on trade in services
should not deter us from obtaining at least reasonably reliable estimates of the consequences of a government purchases impulse on the (components of the) trade
balance with other EU countries. For the sake of parsimony we include GDP instead of
its components in the panel VAR. Hence, the vector of endogenous variables becomes
[g, nt, x EU, y, m EU, reer EU]0 , where x EU (m EU) is good exports (imports) to (from) all
other EU countries in the sample and reer EU is the real effective exchange rate with
respect to all other EU sample countries. The impulse responses are reported in panel
(d) of Table 6. Compared with the results reported in panel (a) of this Table, the
negative response of exports and the appreciation of the real effective exchange rate
are no longer significant, while imports react more strongly, suggesting a higher
elasticity (to domestic output) of imports from the EU than from other countries.
6. Direct Estimates of Fiscal Spill-Overs in the EU
According to its EERP presented at the end of November 2008 the EU intends to
respond to the current economic crisis with a discretionary fiscal expansion of 1.5%
(cumulative over the crisis period) of EU GDP in order to mitigate the fall in economic
2011 The Author(s). The Economic Journal 2011 Royal Economic Society.
F28
THE ECONOMIC JOURNAL
[FEBRUARY
activity. The EU Member States have committed 1.2% of their GDP in total (170 of the
total of 200 billion euros), while the remainder is supposed to come from EU funds.
Importantly, the initiative for the EERP was taken by the European Commission
(2008). The Commission works for the entire EU and, as such, it is supposed to
internalise all externalities. Hence, the idea was to have a concerted stimulus, so that
countries can benefit from the positive trade spill-overs vis-a-vis each other. Without the
coordinating efforts at the European level, this plan might never have taken off, because each individual country might prefer to profit from the other countries’
expansions, rather than see much of the higher demand from its own expansion leak
abroad. The latter is a legitimate concern, in particular for the more open economies,
as our estimates have shown. Some countries (in particular, Germany) were not supportive of the plan, claiming that they are already providing enough stimulus by simply
letting their automatic stabilisers operate.
To form an idea about the potential gains from the coordination of fiscal expansion,
we expand the baseline model with the variable y*, which is a measure of economic
activity in the other thirteen EU countries. Specifically, the vector of endogenous variables becomes now [y*, g, nt, y, irl, reer]0 . We allow for both a response of y* to y as well
as a response of y to y*. This introduces an extra parameter above the ‘diagonal’. With
the remainder of the identification scheme following the lower-triangular Cholesky
decomposition we need an additional restriction. This is achieved by imposing a zero
restriction on the response of g to y*. Given that we have throughout assumed that
g does not contemporaneously react to y, it seems most reasonable to also assume that
it does not contemporaneously react to y*.23
In exploring the international spill-over effects we confine ourselves to government
purchases impulses originating in the five largest EU economies (France, Germany,
Italy, Spain and the UK). Together, these economies make up a large fraction of the
entire EU economy, while, moreover, we expect the international consequences of
individual fiscal impulses in small countries to be minor. We thus estimate a panel VAR
with the five largest EU economies. Figure 4 shows the impulse responses for the two
weighting schemes used to construct the rest-of-Europe output. The first scheme is
based on GDP weights, while the second is based on the intensity of bilateral trade
relationships. In particular, we use a weighting scheme based on the amounts of imports by the impulse-giving country (see the online Appendix). The extended specification reproduces the baseline results. The effects on domestic output have become
even somewhat stronger on impact and one year after the shock, possibly because these
large economies are even more closed on average than the ‘average’ closed economy in
our sample.
When based on GDP weights, the rest-of-EU output goes up in response to the
domestic shock, although the rise is not statistically significant (see Figure 4(a)). With
foreign output based on imports weights, not surprisingly the effect on rest-of-EU
output is larger (see Figure 4(b)). Two years after the shock it reaches a significant peak
if we consider a confidence interval of minus ⁄ plus one standard error around the
mean, as has become rather standard now in this literature. The peak effect of 0.35 is
23
To avoid collinearity problems due to the high correlation between the time effects and the restof-Europe output (which is almost country-invariant), period effects are excluded from this specification.
2011 The Author(s). The Economic Journal 2011 Royal Economic Society.
2011 ]
GOVERNMENT PURCHASES IN THE EU
(a)
6.0
GDP-Based Weights of Foreign GDP
Long Term Interest Rate (irl)
Output* (y*)
100.0
50.0
0.0
–50.0
0 1 2 3 4 5 6 7 8 9 10
0 1 2 3 4 5 6 7 8 9 10
Years After Shock
Years After Shock
Government Spending (g)
Real Effective Exchange rate (reer)
2.0
2.0
–2.0
0.5
0.0
–0.5
–2.0
4.0
0 1 2 3 4 5 6 7 8 9 10
Years After Shock
Cyclically Adjusted Net Taxes (nt)
0.0
–4.0
3.0
–6.0
0 1 2 3 4 5 6 7 8 9 10
Years After Shock
F29
0 1 2 3 4 5 6 7 8 9 10
Years After Shock
Budget Balance (Constructed)
0.5
–0.5
–1.5
0 1 2 3 4 5 6 7 8 9 10
Years After Shock
Output (y)
1.0
–1.0
0 1 2 3 4 5 6 7 8 9 10
Years After Shock
(b)
1.0
0.5
0.0
–0.5
6.0
4.0
2.0
0.0
–2.0
4.0
2.0
0.0
–2.0
–4.0
3.0
2.0
1.0
0.0
–1.0
Trade-Based Weights of Foreign GDP
Output* (y*)
Long Term Interest Rate (irl)
100.0
50.0
0.0
–50.0
0 1 2 3 4 5 6 7 8 9 10
0 1 2 3 4 5 6 7 8 9 10
Years After Shock
Years After Shock
Government Spending (g)
Real Effective Exchange Rate (reer)
2.0
0.0
–2.0
–4.0
–6.0
0 1 2 3 4 5 6 7 8 9 10
0 1 2 3 4 5 6 7 8 9 10
Years After Shock
Years After Shock
Cyclically Adjusted Net Taxes (nt)
0 1 2 3 4 5 6 7 8 9 10
Years After Shock
Output (y)
Budget Balance (Constructed)
0.5
–0.5
–1.5
0 1 2 3 4 5 6 7 8 9 10
Years After Shock
0 1 2 3 4 5 6 7 8 9 10
Years After Shock
Fig. 4. System with Spill-Overs
Notes. See note to Fig. 1. The area between the dotted lines is the confidence interval based
on minus ⁄ plus one-standard error (i.e. 16–84th percentile), while the dashed lines mark the
90% confidence interval (5–95th percentile).
2011 The Author(s). The Economic Journal 2011 Royal Economic Society.
F30
THE ECONOMIC JOURNAL
[FEBRUARY
quite substantial if we realise that this number reflects the spill-over of a unilateral fiscal
expansion (though of a large economy). Hence, on the basis of these results, spill-overs
of a concerted fiscal expansion in the five largest EU countries can be expected to be
non-negligible, which might justify such an action in a common recession.
7. Concluding Remarks
This article has briefly reviewed the theoretical and empirical literature on the consequences of discretionary impulses to government purchases. Employing panel VARs
we have provided our own empirical evidence for the EU. We find a positive effect on
output with a multiplier exceeding unity. The real exchange rate appreciates and the
public budget deteriorates. A split of output into its components suggests positive
responses of private consumption and investments, while imports rise and exports fall,
implying a deterioration of the trade balance and, hence, lending support to the twindeficits hypothesis. A sample split into relatively closed and relatively open economies
suggests a lower output multiplier and a stronger and more drawn-out trade balance
deterioration for the latter group, which indicates that for this group a substantial part
of the fiscal stimulus leaks abroad.
Further investigation shows that import elasticities for within-EU trade are higher
than for trade with countries outside the EU. In our more explicit search for crossborder spill-over effects we found indications of non-negligible effects of government
purchases expansion on other EU countries, in particular the main trading partners.
These results provide some rationale for the EERP, which envisages a concerted fiscal
response to the current economic crisis. Of course, one should be careful in advising
substantial joint fiscal expansion, because we have primarily explored the short to
medium-run consequences of an increase in government purchases. Given the ensuing
rise of the public debt, the ability to boost activity significantly (in a period of economic
distress) would be a minimum requirement for enacting such an expansion.
University of Amsterdam, CEPR, CESifo and Tinbergen Institute
University of Amsterdam and Tinbergen Institute
References
Alesina, A., Ardagna, S., Perotti, R. and Schiantarelli, F. (2002). ‘Fiscal policy, profits, and investment’,
American Economic Review, vol. 92, pp. 571–89.
Auerbach, A.J. and Gorodnichenko, Y. (2010). ‘Measuring the output responses to fiscal policy’, NBER
Working Paper No. 16311.
Baxter, M. (1995). ‘International trade and business cycles’, in (G. M. Grossman and K. Rogoff, eds),
Handbook of International Economics, pp. 1801–64, Amsterdam: North-Holland.
Baxter, M. and King, R.G. (1993). ‘Fiscal policy in general equilibrium’, American Economic Review, vol. 83,
pp. 315–34.
Beetsma, R. and Jensen, H. (2005). ‘Monetary and fiscal policy interactions in a micro-founded model of a
monetary union’, Journal of International Economics, vol. 67, pp. 320–52.
Beetsma, R., Giuliodori, M. and Klaassen, F. (2006). ‘Trade spillovers of fiscal policy in the European Union: a
panel analysis’, Economic Policy, vol. 21(48), pp. 639–87.
Beetsma, R., Giuliodori, M. and Klaassen, F. (2008). ‘The effects of public spending shocks on trade balances
and budget deficits in the EU’, Journal of the European Economic Association, vol. 6(2–3), pp. 414–23.
Beetsma, R., Giuliodori, M. and Klaassen, F. (2009a). ‘Temporal aggregation and SVAR identification, with an
application to fiscal policy’, Economics Letters, vol. 105(3), pp. 253–5.
2011 The Author(s). The Economic Journal 2011 Royal Economic Society.
2011 ]
GOVERNMENT PURCHASES IN THE EU
F31
Beetsma, R., Giuliodori, M. and Wierts, P. (2009b). ‘Planning to cheat: EU fiscal policy in real time’, Economic
Policy, vol. 24(60), pp. 753–804.
Bénétrix, A.S. and Lane, P.R. (2010). ‘Fiscal shocks and the real exchange rate’, mimeo, IIIS, Trinity College
Dublin.
Blanchard, O. and Perotti, R. (2002). ‘An empirical characterisation of the dynamic effects of changes in
government spending and taxes on output’, Quarterly Journal of Economics, vol. 117, pp. 1329–68.
Born, B. and Müller, G.J. (2009). ‘Government spending shocks in quarterly and annual US time series’,
mimeo, University of Bonn.
Burnside, C., Eichenbaum, M. and Fisher, J.D.M. (2004). ‘Fiscal shocks and their consequences’, Journal of
Economic Theory, vol. 115, pp. 89–117.
Burriel, P., de Castro, F., Garrote, D., Gordo, E., Paredes, J. and Pérez, J.J. (2010). ‘Fiscal policy shocks in the
Euro Area and the United States’, Fiscal Studies, vol. 31(2), pp. 251–85.
Christiano, L., Eichenbaum, M. and Evans, C. (1999). ‘Monetary policy shocks: what have we learned and to
what end?’, in ( J.B. Taylor and M. Woodford eds.), Handbook of Macroeconomics pp. 65–148, Amsterdam:
Elsevier Science.
Corsetti, G., Meier, A. and Müller, G. (2010). ‘What determines government spending multipliers?’, mimeo,
Cambridge University.
Corsetti, G. and Müller, G. (2008). ‘Twin deficits, openness, and the business cycle’, Journal of the European
Economic Association, vol. 6(2–3), pp. 404–413.
Devereux, M.B., Head, A.C. and Lapham, B.J. (1996). ‘Monopolistic competition, increasing returns, and the
effects of government spending’, Journal of Money, Credit, and Banking, vol. 28(2), pp. 233–54.
Doan, T. (2006). User’s Guide RATS Version 6. Estima.
Edelberg, W., Eichenbaum, M. and Fisher, J.D.M. (1999). ‘Understanding the effects of a shock to government purchases’, Review of Economic Dynamics, vol. 2, pp. 166–206.
European Commission (2008). A European Economic Recovery Plan, available at http://ec.europa.eu/economy_
finance/publications/publication13504_en.pdf.
Fatás, A. and Mihov, I. (2001). ‘The effects of fiscal policy on consumption and employment: theory and
evidence’, CEPR Discussion Paper No. 2760.
Favero, C.A. and Giavazzi, F. (2009). ‘Reconciling VAR-based and narrative measures of the tax-multiplier’,
mimeo, Universita Bocconi.
Forni, M. and Gambetti, L. (2010). ‘Fiscal foresight and the effects of government spending’, mimeo,
University of Modena and University Autonoma de Barcelona.
Galı́, J., Vallés, J. and L
opez-Salido, J.D. (2007). ‘Understanding the effects of government spending on
consumption’, Journal of the European Economic Association, vol. 5(1), pp. 227–70.
Gemmell, N., Kneller, R. and Sanz, I. (2009). ‘The timing and persistence of fiscal policy impacts on growth:
evidence from OECD countries’, Economic Journal, vol. 121(550), pp. F33–58.
Heady, C., Johansson, A., Arnold, J., Brys, B. and Vartia, L. (2010). ‘Tax policy for economic recovery and
growth’, Economic Journal, vol. 121(550), pp. F59–80.
Kim, S. and Roubini, N. (2008). ‘Twin deficit or twin divergence? Fiscal policy, current account, and real
exchange rate in the United States’, Journal of International Economics, vol. 74(2), pp. 362–83.
Kirchner, M., Cimadomo, J. and Hauptmeier, S. (2010). ‘Transmission of government spending shocks in the
Euro Area: time variation and driving forces’, ECB Working Paper No. 1219.
Lane, P. and Perotti, R. (1998). ‘The trade balance and fiscal policy in the OECD’, European Economic Review,
vol. 42, pp. 887–95.
Lane, P. and Perotti, R. (2003). ‘The importance of composition of fiscal policy: evidence from different
exchange rate regimes’, Journal of Public Economics, vol. 87, pp. 2253–79.
Leeper, E.M., Walker, T.B. and Yang, S. (2008). ‘Fiscal foresight: analytics and econometrics’, CAEPR
Working Papers, Indiana University Bloomington.
Linnemann, L. (2006). ‘The effect of government spending on private consumption: a puzzle?’, Journal of
Money, Credit, and Banking, vol. 38(7), pp. 1715–35.
Melander, A., Sismanidis, G. and Grenouillea, D. (2007). ‘The track record of the Commission’s forecasts: an
update’, European Economy Economic Papers No. 291, Brussels.
Mertens, K. and Ravn, M. (2010). ‘Measuring the impact of fiscal policy in the face of anticipation: a structural
VAR approach’, Economic Journal, vol. 120(544), pp. 393–413.
Mertens, K. and Ravn, M. (forthcoming). ‘Empirical evidence on the aggregate effects of anticipated and
unanticipated US tax policy shocks’, Review of Economic Dynamics.
Monacelli, T. and Perotti, R. (2006), ‘Fiscal policy, the trade balance and the real exchange rate: implications
for international risk sharing’, mimeo, IGIER.
Mountford, A. and Uhlig, H. (2009). ‘What are the effects of fiscal policy shocks?’, Journal of Applied
Econometrics, vol. 24(6), pp. 960–92.
Müller, G. (2008). ‘Understanding the dynamic effects of government spending on foreign trade’, Journal of
International Money and Finance, vol. 27(3), pp. 345–71.
OECD (2005). Economic Outlook. Paris: OECD.
2011 The Author(s). The Economic Journal 2011 Royal Economic Society.
F32
THE ECONOMIC JOURNAL
[ F E B R U A R Y 2011]
Pappa, E. (2006). ‘Discussion of ‘‘Comparing alternative methodologies to estimate the effects of fiscal policy
by Roberto Perotti’’’, mimeo, London School of Economics.
Perotti, R. (2007). ‘In search of the transmission mechanism of fiscal policy’, NBER Macroeconomics Annual,
vol. 22, pp. 169–226.
Ramey, V. (forthcoming), ‘Identifying government spending shocks: it’s all in the timing’, Quarterly Journal of
Economics.
Ramey, V. and Shapiro, M. (1998). ‘Costly capital reallocation and the effects of government spending’,
Carnegie-Rochester Series on Public Policy, vol. 48, pp. 145–94.
Ravn, M., Schmitt-Grohé, S. and Uribe, M. (2006). ‘Deep habits’, Review of Economic Studies, vol. 73(1), pp.
195–218.
Ravn, M., Schmitt-Grohé, S. and Uribe, M. (2007). ‘Explaining the effects of government spending shocks on
consumption and the real exchange rate’, mimeo, EUI Florence and Duke University.
Romer, C.D. and Romer, D.H. (2010). ‘The macroeconomic effects of tax changes: estimates based on a new
measure of fiscal shocks’, American Economic Review, vol. 100(3), pp. 763–801.
Sims, E.R. (2009). ‘Non-invertibilities and structural VARs’, mimeo, University of Notre Dame.
Van den Noord, P. (2000). ‘The size and role of automatic fiscal stabilizers in the 1990s and beyond’, OECD
Working Paper No. 230.
Yang, S. (2007). ‘Tentative evidence of tax foresight’, Economics Letters, vol. 96, pp. 30–7.
2011 The Author(s). The Economic Journal 2011 Royal Economic Society.