Download This PDF is a selection from a published volume from... Economic Research Volume Title: NBER Macroeconomics Annual 2007, Volume 22

This PDF is a selection from a published volume from the National Bureau of
Economic Research
Volume Title: NBER Macroeconomics Annual 2007, Volume 22
Volume Author/Editor: Daron Acemoglu, Kenneth Rogoff and Michael
Woodford, editors
Volume Publisher: University of Chicago Press
Volume ISBN: 978-0-226-00202-6
Volume URL: http://www.nber.org/books/acem07-1
Conference Date: March 30-31, 2007
Publication Date: June 2008
Chapter Title: Cyclical Budgetary Policy and Economic Growth: What Do We
Learn from OECD Panel Data?
Chapter Author: Philippe Aghion, Ioana Marinescu
Chapter URL: http://www.nber.org/chapters/c4081
Chapter pages in book: (p. 251 - 278)
4_
Cyclical Budgetary Policy and Economic
Growth: What Do We Learn from OECD
Panel Data?
Philippe Aghion,
Ioana Marinescu,
1
University
of Chicago, Harvard
University
of Chicago
University
and NBER
Introduction
is that there is a decoupling
among macroeconomists
macroeconomic
taxation, money
(e.g., budget deficit,
policy
which
should primarily
affect price and income stability1 and
A common
between
supply),
view
if anything,
should depend
growth, which,
only
long-run economic
structural
characteristics
the
of
upon
economy
(property right enforce
and so forth). That macroeco
ment, market
structure, market mobility,
should not be a key determinant
is further
of growth
policy
at by recent contributions
et al. (2003) and
such as Acemoglu
Easterly (2005), who argue that the correlation between macroeconomic
et al.) or those between
and growth
volatility
(Acemoglu
growth and
nomic
hinted
macroeconomic
variables
(Easterly),
become
insignificant
once
one
con
trols for institutions.
The question
of whether macroeconomic
policy does or does not af
fect (productivity)
In particular,
not
is
it un
growth
purely academic.
derlies the recent debate on the European
Stability and Growth Pact as
well as criticisms against the European Central Bank for allegedly pur
suing price stability at the expense of employment
In this paper we question
that view by arguing
the budget deficit is significant
in explaining GDP
and growth.
that the cyclicality
with
of
amore
growth,
the
countercyclical
budgetary
policy being more
growth
enhancing
lower the country's level of financial development.
We also identify eco
nomic factors that tend to be associated with more
countercyclical
poli
cies. These results hold in a sample of OECD countries with comparable
institutional
The
environments.
idea that cyclical macroeconomic
affect productiv
policy might
is suggested
ity growth
by Aghion,
Angeletos,
Banerjee, and Manova
252 Aghion
and Marinescu
in AABM
is that credit
The argument
AABM).
(2006; henceforth
is
constrained
firms have a borrowing
that
condi
capacity
typically
earn
tioned by current earnings
between
(the factor of proportionality
with a higher multiplier
ing and debt capacity is called credit multiplier,
a
in the economy).
In
of
financial
reflecting
development
higher degree
a recession, current earnings are reduced, and so are firms'
ability to bor
row in order to maintain
investments
growth-enhancing
(e.g., in skills,
or
extent
structural
To
the
that
macroeconomic
R&D).
capital,
higher
translates into deeper recessions,
it should affect firms' incen
volatility
tives to engage in such investments.
This prediction
finds empirical sup
in
first
who show, on
AABM,
cross-country
port,
panel regressions
by
that structural investments
of cross-country
panel regressions,
the
lower
the
level of financial develop
country's
procyclical
and second, in firm-level evidence by Berman et al. (2007). Using
the basis
are more
ment;
French
firm-level
Berman
straints,
total investment
on R&D investments
and on credit con
panel data
over
et al. show that: (a) the share of R&D investment
credit constraints;
(b) the
firms are credit constrained;
(c) this
in
the
effect is only observed during down-cycle
is,
pres
phases?that
ence of credit constraints, R&D investment
share plummets
during re
share
turns more
cessions
is countercyclical
procyclical when
but doesn't
These
without
increase proportionally
during up-cycle periods.2
in turn, suggest
macroeconomic
that countercyclical
findings,
or lower nominal
interest
investment
policies, with higher government
rates during recessions, may foster productivity
growth by reducing the
failures (in particular,
market
of
loss
induced
the
output
by
magnitude
in turn should al
in a recession, which
imperfections)
by credit market
invest
their growth-enhancing
firms to preserve
low credit-constrained
decide
the
ments over the business
For
may
government
example,
cycle.
for private firms' products by increasing spend
to stimulate the demand
and thus make
ing. This could further increase firms' liquidity holdings
without
shocks
it easier for them to face idiosyncratic
having to
liquidity
invest
sacrifice R&D or other types of longer-term
growth-enhancing
face unemploy
the other hand, in a recession, more workers
could help
so their earnings are reduced. Government
ment,
spending
credit constraints either directly (social programs,
them overcome
etc.) or
this re
and
therefore
labor
demand
employment;
indirectly, by fostering
to
make
workers
would
allow
in
laxation of credit constraints,
turn,
investments
in human capital, relocation, and so on.
growth-enhancing
the more
faced by firms and workers,
The tighter the credit constraints
be.3
should
such
policies
countercyclical
growth enhancing
ments.
On
Cyclical Budgetary Policy and Economic Growth
253
in this paper is three fold. First, we compute and an
contribution
coun
of the budget deficit on a panel of OECD
the
cyclicality
alyze
in
out
to
the
fluctuations
deficit
how
the
tries?that
is,
responds
budget
some
we
over
time.
determinants
Second,
potential
put gap
investigate
of the budget deficit. Third, we use these yearly
of the countercyclicality
Our
to assess the relationship
between
growth and the counter
at
of financial develop
various
levels
of
policies
budgetary
cyclicality
as follows: (a) the
ment. Our main findings can be summarized
budget
in
most
countries
OECD
deficit has become
increasingly
countercyclical
panel
data
over
the past
less
twenty years, but this trend has been significantly
a
more
in
within
the EMU; (b)
countries,
countercyclical
pronounced
a
with
is
associated
level of finan
higher
policy
positively
budgetary
a lower level of openness,
of an
and the adoption
cial development,
a
more
countercyclical
policy
regime; (c)
budgetary
inflation-targeting
is
financial development
has a greater positive
impact on growth when
from
argue that our results likely reflect the causality
at
to
the
least
document
statistical
very
they
budgetary
policy
growth,
variables
that are consistent with
between macroeconomic
relationships
on
credit constraints,
and
the theory and microevidence
volatility,
lower. While
we
investments.
growth-enhancing
While we do not know of any previous
the
attempt at analyzing
de
effects
of
of
the
budgetary
policies,
analyses
growth
countercyclical
terminants of the cyclicality of budgetary
policies already exist in the lit
and Tabellini
For example, Alesina
(2005) argue that more cor
run
a
more
to
will
fiscal policy. The
tend
democracies
rupt
procyclical
in
voters
is
demand
that
the government
that,
hypothesis
good times,
cut taxes or provide more public services instead of reducing debt, be
erature.
cause they cannot observe
the debt reduction
ernment of appropriating
the rents associated
In equilibrium,
this leads to a more
ditions.
moral
in the sense
hazard
and can suspect the gov
con
good economic
with
procyclical
policy as the
are more
that governments
They also show that this
problem worsens,
to
divert
in booms.
likely
public resources
in explaining
tends to be more powerful
mechanism
ob
than in borrowing
constraints
alone. While Alesina
are
a
and explore
(2005)
using
large sample of countries
in this study we use panel analysis on OECD
variations,
and Tabellini
cross-sectional
countries.
This makes
reasons.
First,
indices within
dices
the variation
in the data
served
across
the use of corruption
indices impractical
for two
no cross-sectional
variation
in corruption
there is almost
Second, there is even
time for individual
countries.
the OECD.
less variation
of these in
254 Aghion
and Marinescu
In a similar vein, Calderon, Duncan,
and Schmidt-Hebbel
(2004) show
more
are more
that emerging-market
economies with
stable institutions
a
fiscal policy.4 Their empirical analysis
countercyclical
on the International
the varia
Country Risk Guide. Although
tion in this indicator is limited across OECD countries and time, it pres
able to conduct
is based
ents somewhat
more
than corruption
variation
indices.5
as
such
Perotti
Gali
Lane (2003) focus, as
and
and
(2003)
papers,
we do, on OECD countries. Gali and Perotti investigate whether
fiscal
more
in the EMU has become
after the Maastricht
policy
procyclical
Other
for such a development.
treaty. They find no evidence
They do find,
a more coun
a
in
that
while
there
trend
OECD
is
the
toward
however,
over
the
is
fiscal
EMU
that trend.
behind
time,
tercyclical
policy
lagging
in the third
(2003) paper comes closer to the analysis developed
of our paper. Lane examines
the cyclical behavior of fiscal policy
the OECD. He then uses trade openness,
within
output volatility, output
per capita, the size of the public sector, and an index for political power
Lane's
section
to examine cross-country
in cyclicality. The rea
differences
dispersion
a
son
is
taken
from Lane and Tor
role
why power dispersion may play
nell (1998): when multiple
for
political groups compete
public spending,
more
wants
to
let any substan
become
No
group
may
they
procyclical.
subsist because
they are afraid that this will not lead to
to
rather
that surplus.
but
other
groups appropriating
repayment,
trade
that GDP growth volatility,
evidence
Lane finds, in particular,
a
more
to
lead
divisions
and
openness,
procyclical
spending
political
tial fiscal surplus
debt
even
is not present for all
though the effect of political divisions
to
this
literature
We
contribute
of
spending.
by using yearly
categories
of budgetary
the cyclicality
policy and its deter
panel data to analyze
pattern,
countries, and we show that the degree of finan
to explain within-country
element
is an important
cial development
in such policies, while
future or present EMU membership
variations
minants
within
OECD
we show that inflation tar
variations. Moreover,
cross-country
a
more
deficit.
budgetary
countercyclical
geting is associated with
coun
Most closely related to our second-stage
analysis of the effect of
are Aghion, Angeletos,
on
Banerjee,
policy
tercyclical budgetary
growth
Manova
and Aghion,
Bacchetta, Ranciere, Rogoff
(2006;
(2005; AABM),
amodel
volatil
to explain why macroeconomic
ABRR). AABM develop
explains
correlated
ity is more negatively
and
the financial development,
the lower
growth,
productivity
test
this
using cross
prediction
they
from a closed real to an open monetary
with
panel data. ABRR move
and show that a fixed nominal
economy
country
exchange
rate regime
or lower
Cyclical Budgetary Policy and Economic Growth 255
are more
rate volatility
associated with pro
positively
and the lower the
ductivity growth, the lower the financial development
ratio of real shocks to financial shocks.
real exchange
as follows. Section 2 dis
The remaining part of the paper is organized
the estimation
of the budget deficit for
of the countercyclicality
each OECD country and each year covered by our panel data set. Section
some main determinants
3 uncovers
of the countercyclicality
of the bud
on
4
deficit.
Section
GDP
de
financial
per capita growth
regresses
get
in
the
section
coefficients
2, the
velopment,
countercyclicality
computed
cusses
interaction
the two, and a set of controls.
between
Finally,
section
5 con
cludes.
2
The Countercyclicality
of the Budget
in the Cross-Country
Panel
Deficit
In this section we
measures
compute
time-varying
in our cross-country
of
of the cyclicality
the extent to
panel and compare
over
countercyclical
budgetary
policy
which budgetary
time in some
policy became more
countries
than in others. A main
is
that
finding
policy in the
budgetary
United States and the United Kingdom
have become
significantly more
countercyclical
EMU area.
2.1
over
the past
twenty
years, whereas
it has not
in the
Data
ment
data on GDP, the GDP gap (ygap), the GDP deflator,
gross debt (ggfl) are taken from the OECD Economic
nual
series.6 Our measure
Panel
and govern
an
Outlook
of budgetary
is the first difference
of
policy
debt divided by the GDP, which
is the same as the budget deficit over
GDP. Note
that debt and other government
data refer to general gov
ernment. Financial
is
measured
development
by the ratio of private
credit to GDP, and annual cross-country
data for this measure
of finan
can be drawn from the Levine database.7
cial development
In this latter
private credit is all credit to private agents, and therefore in
credit to households.
The "average years of education
in the pop
over
25
ulation
from the Barro-Lee
years old" series is directly borrowed
is only available every five years and has been lin
dataset; this measure
to obtain a yearly series. The openness variable
is de
early interpolated
fined as exports and imports over GDP, and data on it come from the
Penn World Tables 6.1. The population
share of
government
growth,
measure,
cludes
256 Aghion
Table
4.1
Statistics
Summary
Obs.
Mean
GDP756
gap
Gross
Budget
and Marinescu
debt/GDP
government
deficit/GDP
Countercyclicality
of budget
deficit
Countercyclicality
of budget
deficit
rolling window)
weighted
of budget
Countercyclicality
of GDP
Private
credit/GDP
population
Population
capita
of schooling
over 25 years
years
Average
per
0.000
0.019
0.548
0.295
756
0.048
0.046
0.511
0.563
2.686
-0.342
756
0.578
0.752
3.337
-1.112
532
0.608
1.065
8.972
-3.011
689
0.025
0.026
585
0.801
0.392
0.128
1.989
2.510
share of GDP
Investment/GDP
Inflation-targeting
-0.070
0.046
-0.065
-0.092
old
1.686
0.321
0.116
2.240
585
8.236
605
Openness
53.633
35.641
756
Inflation
0.061
0.066
-0.025
689
0.006
0.005
-0.018
605
12.440
5.709
3.008
27.848
605
24.106
4.983
12.867
41.635
756
0.112
0.316
01
(in %)
%)(in
dummy
8.705
to observations
the Countercyclicality
where
of budget
Sample
is not missing.
computed
using Gaussian
weighted
rolling windows
and Perm World
Sources: OECD Economic Outlook, Levine dataset, Barro Lee dataset,
6.1.
Note:
0.071
for the
growth
Government
Max.
deficit
rolling window)
(10-years
Growth
Min.
756
641
(AR[1])
(Gaussian
Std. Dev.
restricted
12.250
266.883
0.762
0.047
deficit
Tables
share of GDP also come from the Perm World
GDP, and investment
is defined using
the dates
Tables 6.1. The inflation targeting dummy
in Vega and
inflation targeting, as summarized
when countries adopted
Winkelried
(2005). All nominal variables are deflated using the GDP de
statistics can be found in table 4.1. The sample is an un
flator. Summary
countries: Australia,
balanced
the following
Austria,
panel including
Spain, Finland, France, United Kingdom,
Belgium, Canada, Denmark,
New Zealand,
Iceland, Italy, Japan, Netherlands,
Norway,
Germany,8
Portugal,
2.2
Sweden,
Public
Deficit
and the United
and Growth:
States.
The Empirical
Challenge
are interested
in evaluating
the impact of the cyclicality
of the bud
on
and
of
GDP
how
this
effect
deficit
the
per capita,
may de
get
growth
on
is that a
Our expectation
the degree of financial development.
pend
We
Cyclical Budgetary Policy and Economic Growth 257
Low
financial
t -1
in period
development
L
Increased|
/f
\ countercyclicality |
of budgetdeficit |
V
I
in 'Dase Period t"1
szz*~.
Lr
financial
High
| Increased j ^
] countercyclicality
| of budgetdeficitJ^*^/^1^*.
period t
j s?^
|
1
k
Time
v-v-*-v-'
t-l
t -1
in period
development
in base
t
t-l
t
t-l
^f
y^'^^J^^^
v->-*-v-'
t-l
t
Time
t
Legend
Trend GDP growth
Realized GDP growth
.
Figure 4.1
The Impact
more
of an Increase
Budget
in the Countercyclicality
countercyclical
budget
financial development
when
this effect from time variation
ure
4.1
illustrates
the situation
between
ment
is low
ment
is high
deficit
this
idea
where,
of the Budget
Deficit
on Growth
is more
likely to enhance growth
to identify
is lower. Empirically, we wish
of budgetary
within
countries.
policy
Fig
case: we
for a hypothetical
distinguish
in the base period t-l,
financial develop
deficit
financial develop
(upper panels), and the situation where
a
We
start
in the left
with
baseline
(lower panels).
depicted
hand side panels of figure 4.1: the budget deficit is thus initially assumed
to be procyclical.
The right-hand
side panels of figure 4.1 illustrate the
in period 2 after an increase in the countercyclicality
of
growth response
in period
the budget
such
that
deficit
the
deficit
becomes
1,
budget
If financial development
is low, then trend
strongly
countercyclical.
in period 2 increases substantially
in figure
growth
(upper left panel
is high, then trend
If, on the other hand, financial development
a
increases
amount
smaller
left
(lower
growth
by
panel of figure 4.1).9
at
one can think of to
most
the
obvious method
Looking
figure 4.1,
4.1).
compute
cyclicality
is to regress
the public
deficit
on the GDP growth
us
258 Aghion
and Marinescu
in period t. In prac
least squares (OLS) on the observations
ing ordinary
seems
on the GDP
more
to
it
reasonable
the
deficit
tice,
regress
public
as
where GDP* is the trend GDP)
[GDP
GDP*]/GDP*,
gap (defined
like a
rather than the GDP growth.
Indeed, the GDP gap is very much
of the GDP growth, and a forward-looking
detrended measure
govern
budgetary
policy should respond to shortfalls from trend rather
than to GDP growth per se (for a theory of why fiscal policy should de
on the GDP gap, see Barro [1979]).
pend
to measure
the cyclicality
of
This type of regression-based
approach
in the literature and can be found, for ex
is now common
fiscal policies
the
and Tabellini
(2005). However,
ample, in Lane (2003) and Alesina
ment's
used in these papers give rise to only one (or a few) observa
tion of cyclicality per country. Since we want to investigate
the impact of
in cyclicality, we need to compute,
for each country, time
time variation
measures
the
of
for
countercyclicality
budget deficit. Specifi
varying
methods
to use a yearly panel of countries, we need ameasure
cally, as we wish
t -1 and
that varies yearly. This means
that period
of countercyclicality
are
one
is
reduced to
single year each! A regression
period t in figure 4.1
so we must use observations
from
not defined
for a single observation,
The next
a few years in order to compute countercyclicality.
can be used to compute
discusses what methods
2.3
Methods
Econometric
one would
Generally,
to Compute
like estimate
subsection
countercyclicality.
Countercyclicality
the following
equation
for each
i:
country
~
*
!'"1 =
ifit
?i?W
+ <hu+ e,v where
~
ei( N(0,
<r*), (1)
of budgetary
the countercyclicality
alitmeasures
policy.
is in
in
when
the
front
of
there is aminus
economy
sign
ygapit:
the opposite
of the GDP gap
and the GDP gap is negative,
that the budget deficit increases
and so a positive
alitmeans
where
is in a
economy
Both alit and the
to denote the
ajit
The variables
bit:Gross
that
is positive,
the
when
that is, the budget deficit is countercyclical.
recession;
iswhy we write
which
constant a2it10are time-varying,
coefficient on the variable ; in country i at year t.
1 are defined as follows:
in equation
government
yit: The GDP
Note
a recession
in country
debt
in country
i and year
i at year
t, in value
t
Cyclical Budgetary Policy and Economic Growth
259
as (yit- y%) /
it:The GDP gap in country i and year t. It is computed
y
filter. A
of yit using the Hodrick-Prescott
y%,where y% is the prediction
that
of 25 was chosen, following OECD
lambda parameter
(1995). Note
the GDP
proach
a
function ap
gap computed
production
by the OECD using
a Hodrick-Prescott
so that in
is also smoothed
by
technique,
measure
the difference between
of the GDP gap and
the OECD
practice
the measure
used here is very limited: the correlation between
the two
of the GDP gap is, as expected, positively
is 77%. Our measure
correlated with the GDP per capita growth: the correlation
is, however,
variables
not so strong,
at 36%.
that bit - bit_x is exactly equal to the opposite
of the budget bal
so
our
to
the
that
is
left-hand
side
variable
ance,
equal
budget deficit as
a share of GDP, which we will
as
budget deficit. We now
simply refer to
Note
examine
how
One way
the coefficients
to implement
can be estimated
econometrically.
ajit
this is to compute
finite (for example,
ten
least squares (OLS) estimates. The ten
ordinary
to estimating
OLS method
the
simply amounts
t
at
deficit
of
the
in
year
(bit bit_Jy{^
budget
country
countercyclicality
iby running the following
for each country i, and all possible
regression
years)
year
rolling-window
rolling window
t:
years
~
"
''"' =
if it
-?,?W
+ ?2?+ ?Wfor t
(t
-
5, t+ 4). (2)
That is, one uses a ten-year centered rolling window
to estimate the coun
t.
at
of
deficit
date
This
method
suffers, how
any
tercyclicality
budget
ever, from serious shortcomings.
First, by definition, we lose the first five
years and the last four years of data for each country. Second, because the
a coefficient
method
involves estimating
at each time pe
by discarding
riod one old observation
and taking into account a new one, the coeffi
cient can vary substantially when
the new observation
is very different
from the one it replaces. This implies that the series may be jagged and
a sudden
by noise and transitory changes; moreover,
jump in the
not be coming from changes
series would
in the immediate neighbor
hood of date t, but from changes five years before and four years after.
To deal with
the shortcomings
of the ten-years
rolling window
one
can
use
are used for
such that all observations
method,
smoothing
each year, but those observations
closest to the reference year are given
affected
greater
method
The local Gaussian-weighted
weight.
is one way of achieving
this. It consists
least squares
ordinary
co
in computing
the
ajit
260 Aghion
and Marinescu
available
for each country i and
by using all the observations
one regression
then performing
for each date t,where
the observations
are
a Gaussian
at
t.u
centered
date
weighted
by
efficients
~
it
/o\
, ,
+
i,t~l -=
-?i?yOT,-T
if it
where
eh
~
N[0,
v2/wt(i)\
a2u
+
(3)
e,-T,
and wt(j)
=
i
r (T-o2_i
^
exp-^~r~
is less noisy than the
OLS method
the local Gaussian-weighted
a
window
it
similar shortcoming
suffers
from
method,
ten-years rolling
it comes to testing the idea illustrated
in figure 4.1. Indeed, these
when
While
use observations
two methods
from both the past and the future (previ
ous years and future years) to calculate yearly countercyclicality.
Ulti
we
in counter
want
to look at the impact of year t 1 changes
mately,
on year t growth, but if countercyclicality
is
using
computed
cyclicality
some future observations,
then in practice we are examining
the impact
on
of both past and (some) future countercyclicality
growth. Thus, it is
causes
hard to be certain that year t -1 countercyclicality
year t growth,
a
and reverse causality becomes
problem. One way to address this issue
is to use longer lags of countercyclicality
(t 2 or t 3 or t 4, etc.), but
on
that the effects of countercyclicality
this requires us to assume
a
are
for
delayed
specific number of years.
growth at year t
this problem by making
is
upon past observations,
depend essentially
countercyclicality
the
follow an AR(1) process. Namely,
to assume
that coefficients
using
notation
from equation
1, for each country i and for each coefficient;:
An
alternative
method,
which
gets
around
current
+
^^^-i
e^eg^N^aJ.).
The main
is to estimate a2
in implementing
this method
challenge
as
same
at
the
variance of the
the
time
the
of
coefficients)
1.
of equation
that is, the variance a2 in the formulation
(the variance
observation,
Once these variances
best
estimates
(4)
for
are estimated,
applying
the Kalman
filter gives
the
a-it.
are extremely hard to com
for these variances
The optimal estimates
turns out to be vir
solutions
pute. While
finding analytical closed-form
Carlo
Markov
Chain
Monte
(MCMC) methods
pro
tually impossible,
vide
a feasible
the method
We implement
approximation.
variances
and
the
that the
of
coefficients
equation
numerical
Matlab,
assuming
the same for all countries.12 We
are thus left with
three variances
in
are
to esti
Cyclical Budgetary Policy and Economic Growth
261
=
1,2) and one for the vari
in the equation
the MCMC method
(a;:). Intuitively,
aMarkov
hence
the
chain,
name) awide spec
(using
randomly explores
trum of possible values
and one then retains a set of
for the variances,
mate:
two for the coefficient
processes
(v2a.,j
ance of the error
that is representative
of probable values, given the data.13 An ad
over maximum
likelihood
vantage of the MCMC method
type methods
is that it does not get stuck in local solutions
and thus properly
repre
sents uncertainty
about the variances.14 Once we obtain the estimates of
values
the
three variances,
Kalman
filter.
these
ajit
coefficients
can be calculated
using
the
over the previous methods
is to be preferred
for two
AR(1) MCMC
reasons. First, it reflects a reasonable
about
is,
assumption
policy?that
on the immediate past. Second,
that policy changes slowly and depends
and most
in that itmakes
it is econometrically
appealing
coefficients mainly depend on the past (because
ajit
are used as
the
coefficients
thus, when
specification);15
ajit
variables
in panel regressions,
it is less likely that there
importantly,
reflected in the
policy
of the AR(1)
explanatory
should be a reverse
2.4
causation
problem.
Results
now use the AR(1) method
as previously
to characterize
described
the level and time path of the countercyclicality
of budget deficits in the
in our sample. We also report some basic results with
OECD countries
We
the ten-year rolling window
and Gaussian-weighted
OLS methods.
Table 4.1 summarizes
the descriptive
statistics of our main variables
of interest. For all three measures,
the budget deficit is countercyclical
is consistent with Lane's (2003) finding that
(positive coefficient), which
the primary surplus is procyclical.
It isworth noting that the three dif
ferent methods
used in the first stage to estimate countercyclicality
give
in terms of the mean: amean of about .5means
similar
results
that,
very
on average,
in our sample a 1 percentage
in
increase
the
point
opposite
of the GDP gap (i.e., a worse
leads to about
.5 percentage
recession)
points increase in the budget deficit as a share of the GDP. In terms of the
we can see that the standard error is
of these measures,
largest
as
for the ten-years rolling window
it is smaller for
method,
expected;
the Gaussian method,
and even smaller for the AR(1) MCMC method.
We now look at the evolution
of the countercyclicality
of budget
the
estimated
1.
coefficients
from
deficit, as measured
alit
by
equation
4.2
shows
the
evolution
of
the
of
the
Figure
countercyclicality
budget
variance
262 Aghion
cm-
and Marinescu
K
LO_
\^
J
V^
O]
1I
' ' I I I I I I
' I
1960
1970
I I .j
I
1990
1980
2000
2010
Year
-
10 years rolling window-Gaussian-weighted
rollingwindow
-AR(1)MCMC
Figure 4.2
The Countercyclicality
Note:
gap
Source: OECD
of the Budget
Deficit
in the United
i.e., the coefficients
plots the alit coefficients,
estimation
1, using various
techniques.
The graph
in equation
Economic
States
on the opposite
of the output
Outlook.
described
States, estimated
by the three methods
can readily see that, as expected,
the
construction
given
previously.
standard
and their empirical
of these measures
errors, the ten-years
results, and the AR(1) method
rolling window
yields the most volatile
deficit
for the United
We
in be
OLS method
with
the Gaussian-weighted
is the smoothest,
show an increase in countercyclical
tween. Overall,
all three methods
a recent trend toward decreasing
countercyclicality,
ity over time, with
OLS
and Gaussian-weighted
shown by the ten-years
rolling window
methods.
of the budget deficit esti
In figure 4.3, we show the countercyclicality
a
in our sample.
countries
method
for
few
the
AR(1)
through
mated
tend to increase
countercyclicality
Kingdom
over time, especially
since the 1980s. On the other hand, the average
in EMU countries
of
slightly de
policy
budgetary
countercyclicality
between EMU
creases over time. Also, one can observe some divergence
the counter
at the beginning
of the period,
countries:
and non-EMU
United
States
and United
of the budget deficit in EMU countries was very similar to
cyclicality
as of the 1990s, the United
States
States: however,
that in the United
Cyclical Budgetary Policy and Economic Growth 263
/"
/
/
^
"*^-,_^
?j?i?i?i?i?i?i?i?i?i?|?i?i?i?i?i?i?i?i?i?|?i?i?i?i?i?i?i?i?i?|?i?i?i?i?i?i?i?i?i?|?i?i?i?i?i?i?i?i?i?|?
1970
1960
1980
1990
2000
2010
Year
-
United States-United
countries
-EMU
Kingdom
4.3
Figure
The Countercyclicality of Budget Deficits Using theAR(1) MCMC Method
on the
The graph plots the alit coefficients,
i.e. the coefficients
of the output
opposite
the AR(1) MCMC method.
For EMU countries
1, using
gap in equation
(i.e., countries who
are or will be part of the EMU), the line represents
the average of the estimated
coefficients
in the sample;
for the EMU countries
is only computed
the average
for those years
present
where
all EMU countries
have nonmissing
observations.
Note:
Source: OECD
and
Economic
the United
Outlook.
Kingdom
more
significantly
Union did not.
became
whereas
the European Monetary
In figure 4.4, we plot the same
countercyclical,
this time based on coeffi
evolution,
the
OLS. Trends in
using
Gaussian-weighted
similar to those obtained using the AR(1) method.
that are estimated
cients
estimates
These
are very
results are consistent
with Gali and Perotti (2003), who
show,
their
in
that
in the
fiscal
deficits
decades,
splitting
sample by
general,
OECD have become more
so
in
less
EMU
countercyclical,
although
countries. Here we confirm these results, using a
time-series
full-fledged
measure
of countercyclicality.
To
summarize
deficit
has become
United
Kingdom
tion we
investigate
our
found that the budget
results, we
descriptive
more
in
the
United
States and the
countercyclical
than in EMU countries since the 1990s. In the next sec
possible
explanations
for these observed
differences
264 Aghion
CM
and Marinescu
/
/
/
/
/
^4
-
o-_
:
/
|?i?i?i?i?i?i?i?i?i?|?i?i?i?i?i?i?i?i?i?|?i?i?i?i?i?i?i?r-i?|?i?i?i?i?i?i?i?i?i?|?i?i?i?i?i?i?i?i?i?|?
1960
1970
1990
1980
2000
2010
Year
-
United States-United
countries
-EMU
Kingdom
Figure 4.4
The Countercyclicality
of Budget Deficits
the Gaussian-Weighted
OLS Method
using
on the opposite
of the output
i.e., the coefficients
plots the alit coefficients,
For EMU
the Gaussian-weighted
OLS method.
1, using
gap
rolling window
are or will be part of the EMU),
the aver
the line represents
countries
(i.e., countries who
in the sample;
the average
for the EMU countries
coefficients
age of the estimated
present
all EMU countries
have nonmissing
observations.
is only computed
for those years where
Note:
The graph
in equation
Source: OECD
Economic
Outlook.
in the countercyclicality
of budgetary
deficit
across
countries
and over
time.
3 First Stage: Determinants
of Budgetary
Policy
of the Countercyclicality
derived using
coefficients
In this section, we use the series of cyclicality
of budget
and regress the countercyclicality
the AR(1) MCMC method
our sample is re
a
over
Since
set
of
macroeconomic
variables.
ary policy
stricted
to OECD
little variation
countries,
or other institutional
variables
corruption
so far.16 Instead, we
development,
has adopted
be expected
from the
considered
by the literature
should
candidate variables:
financial
focus on the following
and
whether
the
EMU
country
openness,
membership,17
inflation
targeting. We
also
include GDP
growth
volatility
Cyclical Budgetary Policy and Economic Growth
265
as measured
error of GDP growth,
by the standard
lag of log-real GDP
as
share of GDP
control variables.
per capita, and the government
as
Financial development
it influences both the
is a plausible
suspect,
to
and
of
the
borrow
governments
ability
willingness
during recessions
in order to finance a budget deficit. Lower financial development
should
into lower countercyclicality
thus translate
of budget
deficit. While
are arguably
countries
less subject to borrowing
constraints,
there is still a fair amount of cross-country
variation
in financial devel
OECD
is also a plausible
candi
opment between OECD countries. Openness
as
one
can
to
in
flow
date,
expect foreign capital
during booms and flow
out during recessions,
the
cost
that
of
capital is higher during
implying
recessions
rather than booms. This, in turn, tends to increase the long
run cost of
financing countercyclical
taining the overall debt constant,
budget deficit policies while main
on average, over the
long run. The
in
candidate, given: (a) our observation
EMU dummy
is also a plausible
4.2
and
4.3, that the budget
figures
is less countercyclical
in the
(b) the deficit
Kingdom;
and debt restrictions
imposed by the Stability and Growth Pact and also
the restrictions
that individual
countries
in or
imposed on themselves
eurozone
than in the United
deficit
States or the United
der to qualify for EMU membership.
Inflation
should also
targeting
improve
a
country's
or
willingness
to conduct
one
countercyclical
budgetary
policy. In particular,
that
reces
factor
to
borrow
in
governments
potential
might discourage
sions is people's expectation
that such borrowing might
result in higher
ability
in the future?for
example, as away for the government
reduce
This, in turn, would
tially default on its debt obligations.
on
pact of current government
borrowing
private
(long-term)
inflation
ment.
Inflation
increases
the effectiveness
to par
the im
invest
of government
less reasonable.
targeting
in recession by making
such expectations
borrowing
Table 4.2, where
measures
are derived
the countercyclicality
shows results that are consistent with
AR(1) MCMC method,
using the
these con
(a)while countries that are more financially developed
jectures, namely:
tend to have a less countercyclical
(column 1), as a
budgetary
policy
more
more
it exhibits a
country gets
financially developed,
significantly
deficit
the
results
from
column
(column 2); using
countercyclical
budget
increase in private
2, our estimates
imply that a 10 percentage
points
credit over GDP is associated with an increase of about 0.0196 in the
of the budget deficit; in other words,
it is precisely
countercyclicality
when
the countercyclicality
of the budget deficit ismore
cor
positively
related with growth, namely when
financial development
is low, that
266 Aghion
Table
and Marinescu
4.2
The Determinants
of the Countercyclicality
of Budget
Deficits
_(1)_(2)
Year
Private credit/GDP
f .e.
Country
-0.453
f .e.
0.196
(0.115)***
EMU
year
(0.018)***
-0.127
country
(0.038)***
error of GDP
Standard
-3.364
growth
(0.818)***
Lag(log [realGDP per capita])
0.011
0.072
Government share of GDP (in%)
(0.017)
0.000
(0.071)
0.004
(0.005)
(0.001)***
Inflation targeting
0.292
0.112
(0.081)***
(0.015)***
-0.007
Openness
-0.002
(0.001)***
Observations
515
(0.001)***
515
R-squared
0.21
0.99
on the opposite
is the coefficient
The explained
variable
the AR(1) MCMC method.
EMU country
1, estimated
using
1 for all countries
that are part of the EMU as of 2006.
errors in parentheses.
Robust
standard
Note:
*Significant
of the GDP
is a dummy
gap in equation
variable
equal to
at 10%
**Significant
***Significant
Sources: OECD
at 5%
at 1%
Economic
Outlook,
Levine
dataset,
Barro Lee dataset,
and Perm World
Tables
6.1.
seems hardest
to achieve; (b) when
countercyclicality
more
trade openness
is
effects (column 2)
using country- and year-fixed
counter
and
with
associated
deficit
positively
significantly
budgetary
budgetary
deficit
coefficient on openness);
(c) EMU
(the table shows a positive
and countries with a larger standard error of GDP growth ap
deficit countercyclical
pear to have a harder time achieving budgetary
on
the
EMU
that,
average, EMU coun
ity (column 1);
dummy
implies
is
is
lower
tries' budgetary
countercyclicality
by 0.127, which
policy
cyclicality
countries
about
more
a fourth of a standard
deviation;
is
the effect of the EMU dummy
already imposed by the precur
likely to be explained by rigidities
sor EMS regime and then reinforced
Treaty, rather
by the Maastricht
than the 1999 implementation
of the EMU itself;18 further investigation
Cyclical Budgetary Policy and Economic Growth
267
of this question
the scope of this paper; (d) a higher
is, however, beyond
in the GDP is associated with amore countercycli
share of government
cal budgetary
inflation targeting is associated with
policy; (e) pursuing
amore
budget deficit. Note that the coefficient
in column 2 is of the same magnitude
dummy
countercyclical
on the in
as the co
targeting
in column 1, but of opposite
efficient on the EMU dummy
sign.
a lower level of financial development,
a higher
Hence,
degree of
to
tar
the
and
the
EMU
absence
of
inflation
openness,
group,
belonging
are
a
in the
all associated with
lower degree of countercyclicality
geting
we
move
a
In
next
to
deficit.
the
section
budget
second-stage
analysis of
flation
the effect of budget
4
Second
deficit
on growth.
cyclicality
Stage: Cyclical
Budget
Deficit
and Growth
In this section we
for bud
coefficients
regress growth on the cyclicality
in
derived
section 2, financial development,
the interaction
getary policy
between
the two variables,
and
microeconomic
theory
and a set of controls.
evidence
in the introduction
R&D/growth
the greater
Our
discussion
of the
on volatility, credit constraints,
and
that
the
lower
financial
de
suggests
the correlation
should
be between
growth and
a
more
coun
of budgetary
the
is
idea
that
policy:
can
the
reduce
that
effect
neg
tercyclical budgetary
help
policy
negative
ative liquidity shocks impose on credit-constrained
firms that invest in
velopment,
the countercyclicality
R&D and innovation.
4.1
Empirical
Our
empirical
ty*
=
PA?i
Specification
specification
+
P2rVi
+
and Results
is:
Ps^t-iPu-i
+ XA
+ 7, + 8, + eit,
(5)
of the log of real GDP per capita in coun
Ayit is the first difference
i
and
the
is
of the budget deficit as es
year t; alit_x
try
countercyclicality
timated by the AR(1) MCMC method.
Since ait_x is an estimated
coeffi
where
each observation
cient, we weigh
by the inverse of the variance of this
coefficient
in
thus
to coefficients
Stata),
(aweights
giving higher weight
that are more precisely
in the first stage. The ratio of
estimated
private
from Levine and Demirguc-Kunt
credit, p/>1Lto GDP is borrowed
(2001);
con
that vary with
the specification
Xit is a vector of control variables
a
is
and
is
the
sidered, 7. is a country-fixed
effect, St
effect,
eit
year-fixed
error
term.
268 Aghion
and Marinescu
In table 4.3, we first report results with a limited set of controls, repre
of growth: lag of log real
senting the most widely
accepted determinants
over GDP (column
GDP per capita, population
growth, and investment
trade openness,
infla
1).We then add more controls, namely schooling,
share
and
inflation
Note
of
tion, government
GDP,
targeting (column 2).
that since we control for inflation, we indirectly control for the impact of
monetary
policy on growth.
is that of a positive
The prediction
for the effect on
(^ coefficient
of
the
of
the
deficit
when
growth
budget
private
countercyclicality
on
credit over GDP is 0, and of a negative
(33coefficient
countercyclical
In the first column of table
financial development.
ity interacted with
coeffi
4.3, using a limited set of controls, we see that the corresponding
a
are
more
and
cients have the anticipated
signs
statistically
significant:
correlated with growth, but
countercyclical
budget deficit is positively
term between
and financial develop
countercyclicality
a
in column 2 does not
set
of
controls
richer
Including
are
a coeffi
If
estimates
the
results.
the
point
anything,
larger:
change
means
that
cient of 0.11 of the lagged countercyclicality
of budget deficit
the interaction
ment
is negative.
if private credit over GDP is 0, then increasing
the countercyclicality
of
0.11
increases
the budget deficit by one percentage
per
point
growth by
over
centage points. For each percentage
point increase in private credit
diminishes
effect of countercyclicality
GDP, this positive
by 0.0004. The
is thus small: private credit over GDP would
of the interaction
to be larger than 2.75 for a countercyclical
budgetary
policy to be
come growth reducing. Such a high value of private credit over GDP is
effect
need
not observed
in our sample:
of this variable
States
the United
our
in
sample.
in 2000, at 2.24, has
highest value
Then, in columns 3 and 4, we repeat the same specifications
umns 1 and 2, but allow the impact of the interaction between
the
as in col
the coun
of the budget deficit and private credit over GDP to differ
over GDP (the first quartile is then the
by quartiles of the private credit
the dummy
For example,
excluded
"2ndq (Private Credit/
category).
to
1
if
ratio lies in the second
the
Private
is
Credit/GDP
GDP)"
equal
tercyclicality
zero otherwise. As the results in these columns
quartile, and is equal to
is
and financial development
show, the interaction between
cyclicality
ra
credit
the
when
nonlinear, with a significant
private
jump occurring
tio moves
from the second to the third quartile. In other terms, it is only
bud
that countercyclical
development
less growth enhancing.
getary policy becomes noticeably
of a positive
effect of a
Table 4.3 is thus consistent with the prediction
at fairly high
levels of financial
Table
4.3
The Effect
of the Countercyclicality
of Budget
Deficits
on Growth,
(2)
_(1)
lag(Countercyclicality of budget deficit)
0.058
0.081
(0.016)***
(0.018)***
-0.005
lag(Countercyclicality
deficit*3rdq[Private
lag(Countercyclicality
deficit*4thq[Private
of budget
-0.009
(0.003)***
-0.024
(0.008)***
(0.007)***
-0.030
(0.008)***
-0.140
(0.010)***
-0.132
-0.142
(0.022)***
(0.022)***
(0.022)***
0.002
0.002
0.002
0.002
(0.000)***
(0.000)***
-1.702
(0.268)***
for the
(0.284)***
(0.000)***
-1.484
(0.272)***
0.002
population over 25 years old
Government share of GDP (in%)
(0.000)***
(0.000)**
-0.053
(0.022)**
(0.021)**
-0.004
Openness
Observations
477
0.60
R-squared
(0.290)***
(0.003)
-0.001
-0.049
targeting
(0.000)***
-1.635
0.003
(0.003)
-0.001
Inflation
Inflation
-0.131
(0.022)***
-1.490
Population growth
(0.007)
(0.014)***
-0.023
credit/GDP])
of schooling
(0.007)**
-0.008
-0.040
(0.003)**
credit/GDP])
of budget
Investment/GDP (in%)
years
-0.014
-0.022
lag(log [realGDP per capita])
Average
(0.008)
-0.006
deficit*2ndq[Private credit/GDP])
(4)
0.110
(0.012)***
of budget
(3)
(0.024)***
(0.008)
lag(Countercyclicality
Method
0.075
-0.030
of budget
lag(Countercyclicality
deficit*Private
credit/GDP)
MCMC
(0.021)***
-0.010
lag(Private credit/GDP)
AR(1)
-0.001
(0.005)
0.001
(0.005)
0.001
(0.000)**
(0.000)***
467
0.64
477
0.62
467
0.65
Note:
The explained
variable
is the first difference
of the log of real GDP per capita. All
include country
and year fixed effects. Columns
3 and 4 allow for the effects
specifications
of countercyclicality
of the budget
to differ with
deficit
of private
credit/GDP.
quartiles
errors in
Robust
standard
parentheses.
*Significant
at 10%
**Significantat5%
***Significant
Sources: OECD
6.1.
at 1%
Economic
Outlook,
Levine
dataset,
Barro Lee dataset,
and Perm World
Tables
270 Aghion
countercyclical
budget
interaction
significant
and Marinescu
on
and
growth, whereas we see a negative
credit
and
counter
between
the
private
deficit
effect
a country is, the
cyclicality variable. Thus, the less financially developed
more
it is for the government
to be countercyclical
in
growth-enhancing
we
its budgetary
In
that
EMU
observe
countries
have
policy.
particular,
budgetary
the United
that are, on average,
far less countercyclical
than in
policies
States (0.37 vs 0.61), even though the United States ismore fi
than the EMU: thus, the ratio of private credit to
developed
nancially
GDP in 2000 in the EMU
Then,
to the extent
policy
to growth,
is equal to 1.02, against 2.24 in the United States.
that it reflects the causality
from cyclical budgetary
in table 4.3 suggests
the regression
that increasing
the
of the budgetary
countercyclicality
policy would
the
EMU
the
States.
for
than
for
United
hancing
4.2
Robustness
This
section
mates. We
ification
ternative
be more
growth
en
Tests
discusses
various
potential
issues with
our
table 4.3 esti
take as the reference
for this discussion
the spec
specification
in table 4.3, column 2. Therefore, when we report on al
on this reference specification.
specifications,
they are all based
shown
A potential
first source of concern for our estimation
strategy is auto
is typical in panel growth
correlation
of residuals, which
regressions.
This implies that the standard errors may be biased. To correct for this
errors to adjust the standard
we used Newey
plea of the potential bias,
errors in the reference specification.
of er
for autocorrelation
Allowing
rors up to lag 1 increased the standard errors very slightly and left the
at the 1% level. Allowing
for autocorrelation
up
significant
of the budget deficit at
to 5 lags leaves the effect of the countercyclicality
but makes
the interaction be
the same level of statistical
significance,
coefficients
of the budget deficit and private credit be
the countercyclicality
it does not
at the 2% instead of the 1% level. Globally,
only significant
seem that autocorrelation
affects the standard
of residuals substantially
tween
errors
of
Second,
get deficit
our
estimates.
the reader may wonder
increase growth when
about what
components
of the bud
For ex
they are more countercyclical.
that ul
of total government
spending
ample, is it the countercyclicality
and
social
matters
about
transfers
What
for
security
growth?
timately
as for the
run the same analysis
for these variables
spending? We have
budget deficit,19 and found
cant in explaining
economic
was not signifi
that their countercyclicality
growth. This indicates that the cyclical be
Cyclical Budgetary Policy and Economic Growth 271
to fully account for our results:
is unlikely
stabilizers
case
it
not
is
the
that
just increasing transfers and social security
namely,
in
increases
economic
recessions
for
spending
growth. What matters
se
is not the countercyclicality
of spending
(be it discre
per
growth
havior
of automatic
is financed
this spending
tionary or not) but rather the degree towhich
to
which
the
debt?that
the
is,
government
injects extra liq
by
degree
into
the
economy.
uidity
in knowing
what happens
if
of countercyclicality
the
by
OLS. In the case of
rolling window
the reader may be interested
the AR(1) MCMC
estimate
replace
Third,
we
or the ten-years
Gaussian-weighted
on the
the Gaussian,
the coefficients
of the budget
countercyclicality
deficit and on its interaction with private credit have the same sign as in
at the 1% level. The only
the reference specification
and are significant
on the countercyclicality
difference
is that the value of the coefficient
of the budget
dow method,
statistically
are much
deficit
is lower.
the coefficients
significant,
In the case of the ten-years
rolling win
of interest are of the same sign, but are not
which
is not
surprising,
since
these
estimates
noisier.
of
Fourth, one may still be skeptical about the causal interpretation
our estimates. As mentioned
in section 2, our AR(1) MCMC
estimate of
with the
uncorrelated
should, in principle, be mostly
countercyclicality
in
First, to check whether,
is independent
of growth, we include
deed,
measure
both the lag and the lead of the countercyclicality
in the refer
ence specification.
so
not
does
the
coefficient
Doing
significantly
change
on the
an
but
and
lagged countercyclicality
yields
insignificant
positive
coefficient on the lead of procyclicality.
These results are consistent with
causing growth, not the reverse. Second, we noticed
countercyclicality
that inflation targeting
is associated with a less countercyclical
budget
deficit
but
is
in explaining
(table 4.2)
(table 4.3).
insignificant
growth
This raises the possibility
of using inflation targeting as an instrument
a
in GMM framework.
for countercyclicality
In the GMM estimation, we
future,
the endogeneity
reducing
future countercyclicality
problem.
instrument
both the countercyclicality
variable and the lagged GDP per
we
use
For
the
the
classic
instruments'
second and third lag
latter,
capita.
of GDP per capita. Excluded
instruments
in our GMM regression
are
thus second
and third lag of GDP per capita and the inflation
targeting
our GMM estimates allow for
errors of lag 1.
Moreover,
dummy.
Newey
We have thus re-estimated
the reference specification
using GMM. First
are
estimates
but
the explanatory
stage
power of inflation
significant,
targeting
for countercyclicality
is limited. Overidentifying
restrictions
272 Aghion
and Marinescu
are not
we do not reject the counter
rejected by the J test. However,
with
and
its
interaction
which
private credit are exogenous,
cyclicality
on
means
that GMM is not more appropriate
than OLS. The coefficients
and its interaction with private credit are of similar
countercyclicality
as in the reference specification,
but they are not significant
magnitudes
that our counter
exercise
thus
around
This
confirms
30%).
(P-values
measure
to
is
be
unlikely
endogenous.
cyclicality
of our effects: when
in a given year, how
the effect on growth persist? One way to answer this
the reference specification by replacing the lag of the
Finally, one may be interested in the time horizon
of the budget deficit changes
the countercyclicality
far in the future does
question
is tomodify
of the budget deficit, private credit over GDP, and the
countercyclicality
interaction of the two by further lags. When using the second lag of these
the coefficients of interest (P: and 03) are still significant and of
variables,
same
the
slightly. When using the third lag of
sign, but the R2 diminishes
of the deficit is still
the coefficient on the countercyclicality
these variables,
no
with
credit
is
but
interaction
the
private
longer significant.
significant,
in
even further lags makes
of interest become
the coefficients
Using
an
of bud
increase in the countercyclicality
significant. Thus, it seems that
getary policy
5
affects GDP growth
up to two or three years
later.
Conclusion
and determinants
the dynamics
of the
In this paper we have analyzed
on a yearly panel of OECD countries, and
policy
cyclicality of budgetary
and
financial development,
this cyclicality,
between
the relationship
as follows:
can
be
summarized
Our
economic
first,
findings
growth.
increased over time.
of budget deficits has generally
countercyclicality
less
the budget
deficit became
in EMU countries,
However,
slightly
of budgetary
Second, countercyclicality
policy appears
a lower de
a
financial
of
level
development,
higher
by
a
to inflation
to trade, and monetary
policy committed
gree of openness
are more
we found that countercyclical
budget deficits
targeting. Third,
of finan
level
the
lower
the
with
associated
country's
growth
positively
cial development.
interest
in this paper bears potentially
The line of research pursued
our second-stage
In
regres
particular,
implications.
ing growth policy
sions suggest that growth in EMU countries could be fostered if the bud
more countercyclical.
Our first-stage
get deficit in the eurozone became
that this, in turn, could be partly achieved by hav
suggests
regression
countercyclical.
to be facilitated
Cyclical Budgetary Policy and Economic Growth 273
area move
toward inflation targeting; for example,
follow
ing the EMU
in
UK
the
lead
this
and
also
the
coordination
respect,
ing
by improving
in the eurozone on fiscal policy over the cycle
between
finance ministers
so as tomake
it become more countercyclical.20
in this paper should be seen as one step in a broader re
The analysis
search program.
the same kind of anal
First, one could try to perform
countries
ysis for other groups of countries; for example, middle-income
or in Central and Eastern Europe. Second, one could
in Latin America
to volatility,
take a similar AABM-type
of approach
financial develop
to further explore the relationship
and growth
between
ment,
growth
to
and the conduct of monetary
For
what
allow
extent,
policy.
example,
can
for
short-term
of
nominal
interest
rates,
ing
higher procyclicality
in recessions and/or
R&D investments
improve govern
to
ability
implement
growth-enhancing
countercyclical
budget
one
could
the
interactions
ary policies?
Finally,
investigate
possible
in growth
between
and
regressions
countercyclical
budgetary
policy
structural reforms in the product and labor markets.
firms maintain
ments'
Appendix
1 The AR(1) MCMC Method for Calculating
Cyclicality
in the First Stage
is to give a brief description
of how we used the
Kalman
filter together with Markov
Chain Monte
Carlo methods
the coefficients
from equation
(MCMC) in order to estimate
1, under
ajit
the assumption
that they follow an AR(1) process as described
by equa
was carried out in Matlab.
tion 4. The implementation
the
Estimating
means
and variances
of the coefficients
of interest?that
is
in equa
ajit
tion 4?involves
two procedures:
com
Kalman
and
MCMC.
To
filtering
the
with
coefficients
the
Kalman
we
filter
each
for
to
need
pute
country,
The aim of this section
know
the values
of three variances:
in equation 4, for j = 1,2; that is, the process
of the Kalman
filter
minology
a2
variances
in the ter
a2 of the error term e, in equation
1; that is, the measure
error variance
in the
of
the
Kalman
filter.
terminology
the variance
ment
Moreover,
observation
to use the Kalman
filter, we need a prior for the first period of
for each country?that
of our expectation
is, a specification
274 Aghion
over
and Marinescu
at the first time step. As we do not have any mean
ajit
at the first observed
about cyclicality
information
period,
the values
ingful prior
we use a very high variance around the prior mean, so that this prior has
a
the set of initial values
effect on the estimates.
Specifically,
negligible
of the coeffi
for the coefficients were chosen to be the OLS estimates
cients using the first ten years of data for each country, and the value
is set to be 100,000 times the estimated
variance
the initial variance
these coefficients.
error
ment
are
<j2z
unknown
ful prior over them. We therefore need
ues for these three unknown
variances.
can think of MCMC
are useful. One
case
of
we
simulation,
know
the
we
and
do
not
amethod
This
have
any
meaning
to find reasonable
iswhere
as the opposite
parameters
of
a2 and the measure
variances
the process
However,
variance
of
of
MCMC
of simulating:
our
val
methods
in the
process?for
ex
it
and every time we run a simulation
program,
ample, the variances,
a set of
us
data.
observed
More
the
specifically,
probabil
gives
possible
set
data
is
the
defined
of
of
observed
any
by the
ity
getting
probability
model
sume
is the opposite: we as
and the parameters. MCMC
a set of pos
a given dataset, and we are producing
a
in
that
of ac
This is done
fashion
the probability
such
that we have
that we
have
sible parameters.
a parameter
to the probability
that this
is identical
value
cepting
our im
in
the
data.
value
has
actually produced
Specifically,
parameter
we
use
the
classic
(MH) sampler to
Metropolis-Hastings
plementation,
see
an
to
MCMC
and
introduction
do MCMC
(for
Metropolis-Hastings,
and Greenberg
[1995]). InMH, one starts with arbi
a random change
values. Every iteration one proposes
trary parameters
a
our
case
is
This iswhat
the
of
small
Gaussian
(in
parameters.
change)
for example
called
cepted
*
Accept
Chib
this change
the proposal distribution. Subsequently,
or rejected. The probability
is:
of acceptance
= min
T
is either
ac
I
p(data new_parameters)
1/ -T7?1-:-:?7
(*)
I
L p(data previous_parameters)
|
from the cor
is actually sampling
It is easy to prove that this procedure
over the parameter
values. MCMC
rect posterior
distribution
algo
In
the
first
rithms go through two different
stage, the sampler
stages.
of the data in terms of the pa
to a probable
converges
interpretation
in
stage is called burn-in, and took about 500 iterations
our case. Within
increased dramati
these 500 iterations, probabilities
a stable high level. Afterward,
the MCMC
to
and
then
converged
cally
three runs,
Over
of
relevant
the
space
parameters.
explores
algorithm
we took 10,000 samples per run after the end of burn-in. To avoid the au
rameters.
This
Cyclical Budgetary Policy and Economic Growth 275
aMarkov
chain, we only retain
to
the final estimates.
compute
samples
From these three runs, we thus get a total of 300 essentially
uncorrected
we wish
to estimate. Conver
samples for each of the three parameters
that typically characterizes
in order
every 100 iterations
tocorrelation
chain was
gence of the Markov
the across-chain
correlation with
the within-chain
comparing
From these 300 samples,
correlation.
and variances
estimate means
of the three parame
we
can then directly
ters of interest. In order
growth in our second-stage
the value of the cyclicality
assessed
on
of cyclicality
to
not
determine
regressions,
only
(alit), but also the uncertainty we have about
to correctly
infer
the effect
we need
in other words,
the standard devia
to consider the relevant
is necessary
it. To estimate
this uncertainty?or
tion of the cyclicality estimates?it
sources
Two sources are relevant in our case. One is the
of uncertainty.
that
is
filter that stems from the
uncertainty
represented
by the Kalman
finite number
uncertainty
of noisy observations.
The other source of uncertainty
is
about the three parameters
that are modeled
the
MCMC
by
To
process.
combine
4-
varianceMCMC
them,
varianceKalman,
we
where
use
the
=
approximation
varianceKalman
denotes
variancetotal
the
average
over
the 300 Kalman
filter runs using the 300 samples that we
retained from the MCMC
estimates of the three variances. This approx
correct if the variance, as estimated by the Kalman
imation becomes
fil
variance
ter, is similar over different runs of the Markov
chain, which
our
a
data. Finally,
for
full general statistical
approximation
used here can be found in Kording-Marinescu
of the methods
was
a
good
description
(2006).
Acknowledgments
is indebted
to Robert Barro, who contributed
abundant
ad
comments
and editorial
of
vice, and to the very helpful
suggestions
Daron Acemoglu,
Olivier Blanchard, Ken Rogoff,
and Michael Wood
ford. We also thank Ricardo Caballero
and Anil Kashyap
for their useful
This work
discussions.
At earlier
from fruitful conver
stages this project benefited
Tim
Laurence
Bacchetta,
Bloch, Elie Cohen,
Philippe
Besley,
Romain
and our col
Moutot,
Ranciere,
Jean Pisani-Ferry,
sations with
Philippe
in the Institutions,
leagues
Organizations
Canadian
Institute for Advanced
Research.
group at the
are very
grateful to Ann
and
re
Anne-Laure
for
Helfman,
Kolev,
Julian
Piganeau
outstanding
we
search assistance.
thank
Konrad
for
his
collabora
Finally,
Kording
tion on the first-stage analysis section, and more
for helping
specifically,
us implement
the MCMC methodology.
and Growth
We
276 Aghion
and Marinescu
Endnotes
in an
Lucas (1987) analyzes
costs of income volatility
the welfare
rate as given.
for individual
insurance,
complete markets
taking the growth
in an
and Phelan
the welfare
from countercyclical
(1994) analyze
policy
gains
with
insurance markets
but no growth.
Both find very small effects
incomplete
1. For example,
with
ity (or of countercyclical
2. As
at
aimed
policies
out by several authors,
come
that may
pointed
problem
endogeneity
sales and investment.
economy
Atkeson
economy
of volatil
it) on welfare.
reducing
results may be biased because
of an
simultaneous
determination
of
some
of these
from
the potential
and Cette
Berman,
(BEAAC) check the
Eymard, Aghion,
Askenazy,
rate ex
in sales by an exchange
of their results by instrumenting
the variation
on
rate variations
and firms' export status. This
variable, which
posure
depends
exchange
sales variation
without
is strongly
correlated with
variable
by investment
being affected
decisions.
Their results are robust to this instrumentation.
robustness
in an economy
increase aggregate
sub
intervention
government
might
efficiency
to
been
out
Holmstrom
and
shocks
has
credit
constraints
already
by
ject
aggregate
pointed
to explore po
in this section may be seen as a first attempt
and Tirole
(1998). Our analysis
and public
idea
for
between
tential empirical
of
this
the
relationship
implications
growth
over the cycle.
spending
3. That
4. There
is most
is also
likely
ning a budget
fore minimizes
by
by Talvi
the paper
to generate
and Vegh
procyclical
pressures
surplus generates
political
that surplus and becomes
a volatile
and
output,
(2000), who
government
output volatility
is that run
hypothesis
the government
there
The
spending.
to spend more:
procyclical.
a volatile
tax base.
therefore
that high
argue
This movement
is then accentuated
no
in our analysis. However,
sig
they typically have
are less widely
as
our
in
Moreover,
they
sample.
growth
the available
restricts
of interest,
their use considerably
than our main variables
available
not to use these indi
We
to
have
therefore
decided
estimates.
less
leading
precise
sample,
cators in the results reported here.
5. We
have
nificant
also used
effect
these
in parentheses
6. Codes
indicators
on GDP
over
Data
is available
scribers
7. Data
the names
indicate
at www.oecd.org.
to that service.
from Ross
downloadable
time
of variables
in the dataset.
can be downloaded
Levine's
home
page,
from
http/
Full documentation
for sub
sourceoecd.org
/: www.econ.brow.edu/fac/
Ross_Levine/Publications.htm.
involves
are
The adjustment
for the German
reunification.
level variables
adjusted
1991
in the ten years before
of interest on time and a constant
each variable
regressing
use the estimated
to predict
the
coefficients
(data based on West Germany
only). We then
in
values
ratio between
actual and predicted
values
for 1991 to 2000. We take the average
1991.
the years 1991 to 2000. We use this ratio to proportionally
adjust values before
8. All
in the countercyclicality
effect of a decrease
at high enough
levels of financial
development,
more
and spending.
efficient private borrowing
9. The
of public deficit could become
negative
out
crowds
if the government's
deficit
as a measure
of structural
constant
a2it can be interpreted
to the part of budget
deficit
it corresponds
by construction
the business
cycle.
10. The
deed,
upon
In practice, we
ing this smoothing
11.
chose
value
a value
slightly
of 5 for a. While
does
this choice
not qualitatively
affect
budgetary
that does
is somewhat
the results.
deficit:
in
not depend
arbitrary,
chang
Cyclical Budgetary Policy and Economic Growth 277
12. This
is reasonable,
since the OECD
in our sample
countries
share similar
assumption
and degrees
of economic
this assumption
is similar
Moreover,
development.
no
across
a
to assuming
heteroskedasticity
panels when
estimating
panel
regression,
is the standard
which
variances
would
assuming
assumption.
Finally,
country-specific
institutions
make
estimates
observations
more
much
would
have
due to the fact that our relatively
imprecise,
to be used to identify many more parameters.
1 for more
13. See appendix
on the
implementation
details
small
number
of
of this method.
to use maximum
14. It is indeed also possible
to estimate
methods
the
likelihood-type
but these are precisely
liable to get stuck in local solutions.
In a previous
version
variances,
so that it does not
of this paper, we used such amethod,
amended
get stuck
systematically
in a local solution.
are
that method
In practice,
the estimates
of the coefficients
correlated with
the ones obtained
here
highly
we had obtained
ajit
using MCMC.
using
on the future, inasmuch
as their variance
us
coefficients
also depend
is calculated
of available
observations.
the GDP
because
Moreover,
gap is con
ing the full sample
an HP filter, future GDP
structed
trend GDP as computed
re
is also partially
using
by
on the GDP gap.
in the coefficients
flected in the GDP gap and hence
15. The
16. As mentioned
earlier, using
ICRG indicators
variable
takes a value
dummy
and 0 for all the other countries.
turns out not to be of interest
17. This
of 1 for all countries
EMU,
This
years, so that the countries
many
fore the EMU is fully effective.
that would
but
20. The
funds,
Sapir
supervised
1,we
replaced
the EMU
eventually
have
in equation
19. Specifically,
of each of these variables.
to the
that currently
belong
has been prepared
for
even be
be different
join might
is because
with an interaction
between
experimented
was
this interaction
typically
insignificant,
the full implementation
change occurring with
18. We
dummy,
stantial
for our analysis.
the EMU
dummy
that
indicating
of the EMU
the first difference
(Sapir et al. [2003]) recommended
Commission.
by the European
report
and a post-1999
is no sub
there
in 1999.
of debt by the first difference
the setting-up
of
"rainy
day"
References
D.,
Acemoglu,
macroeconomic
S. Johnson,
symptoms:
J. Robinson,
Volatility,
and Y. Thaicharoen.
crises,
and growth.
2003.
Journal
Institutional
ofMonetary
causes,
Economics
50:49-123.
A. Banerjee,
and K. Manova.
2005. Volatility
P., M. Angeletos,
Aghion,
constraints
and productivity-enhancing
investment.
NBER Working
MA: National
Bureau of Economic
Research.
Cambridge,
Aghion,
ductivity
P., P. Bacchetta,
The
growth:
Cambridge,
MA:
R. Ranciere,
and K. Rogoff.
2006. Exchange
rate volatility
and pro
role of financial
NBER Working
no. 12117.
development.
Paper
National
Bureau of Economic
Research.
1996. Fiscal
Alesina,
A., and R. Perotti.
macroeconomic
effects. NBER Working
of Economic
Alesina,
Paper
and growth: Credit
Paper no. 11349.
in OECD
countries:
and
adjustments
Composition
MA: National
Bureau
Paper no. 5730. Cambridge,
Research.
A., and G. Tabellini.
no. 11600.
Cambridge,
2005. Why
is fiscal policy often
procyclical?
National
Bureau of Economic
Research.
MA:
NBER Working
278 Aghion
Atkeson,
A.,
plete markets.
and C. Phelan.
1994. Reconsidering
the costs of business
cycles with
Annual
MIT
Press.
9,187-207.
Cambridge:
the determination
B. 2001. Econometric
Batalgi,
Berman, N., L. Eymard,
the cyclicality
of R&D
Working
Paper
of public
no. 2007-4.
Journal of Political
of panel data, 2d ed. New
analysis
P. Aghion,
investment:
debt.
P. Askenazy,
and G. Cette.
Evidence
Cambridge,
from French
MA:
Harvard
1995. Understanding
Chib, S., and E. Greenberg.
American
Statistician
49:327-35.
Economy
2007. Credit
firm-level
constraints
J., and R. Perotti.
18:533-72.
2003.
Fiscal
University.
and cyclical properties
no. 285,
Santiago.
the Metropolis-Hastings
governance
Germany:
Holmstrom,
cal Economy
K.,
Kording,
lished paper.
B., and J. Tirole.
106 (1): 1-40.
and
I.Marinescu.
2006.
Sampling
R. 1987. Models
of business
Evidence
Latin American
cycles. New
structure
Economic
York: Basil
Vega, M., and D. Winkelried.
Journal
story? International
2005.
of Central
Inflation
Banking
Unpub
learning.
the OECD.
Journal
rates procyclical?
and economic
Cambridge,
MA: MIT
growth:
Press.
A cross
(OECD). 1995. Estimating
po
OECD Economics
Department
fiscal policy. NBER Work
and procyclical
Research.
of Economic
Bureau
and
targeting
1 (3): 153-75.
inflation
behavior:
of
Jour
Blackwell.
and Development
Cooperation
and
structural
gaps
output
budget balances.
no. 152. New York: OECD.
Paper
E., and C. Vegh. 2000. Tax base variability
no. 7499.
MA: National
Cambridge,
ing Paper
from
savings
for Economic
Talvi,
to Kalman
approaches
A. 2001. Financial
R. and Demirguc-Kunt,
comparison of banks, markets, and development.
Organization
tential output,
Working
of
Journal of Politi
of liquidity.
supply
of fiscal policy:
1998. Why
aren't
Lane, P., and A. Tornell.
nal of Development
Economics 57:185-200.
Lucas,
and public
1998. Private
Lane, P. 2003. The cyclical behavior
Public Economics
87:2661-75.
country
In Handbook
Elsevier.
in Europe.
integration
The
Algorithm.
and J. von Hagen.
2004. The design
of fiscal rules and forms of
M., R. Strauch,
am Main,
in European
Union
countries. Working
Paper Series 419. Frankfurt
Bank.
Central
European
Hallerberg,
Levine,
and monetary
policy
and
data. WCFIA
panel
and economic
Easterly, W. 2005. National
policies
growth: A reappraisal.
Economic Growth, ed. P. Aghion
and S. Durlauf,
1015-56. The Netherlands:
Policy
87:940-71.
York: Wiley.
and K. Schmidt-Hebbel.
2004. Institutions
Calderon,
C, R. Duncan,
of macroeconomic
Bank of Chile Working
Central
policies.
Papers
Gali,
incom
NBER Macroeconomics
R. 1979. On
Barro,
and Marinescu
A successful