Blanchard Perotti (2002)

An Empirical Characterization of the Dynamic Effects of Changes in Government
Spending and Taxes on Output

Author(s): Olivier Blanchard and Roberto Perotti
Source: The Quarterly Journal of Economics, Vol. 117, No. 4 (Nov., 2002), pp. 1329-1368
Published by: Oxford University Press
Stable URL: https://www.jstor.org/stable/4132480
Accessed: 13-09-2018 21:50 UTC
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact [email protected].
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms
Oxford University Press is collaborating with JSTOR to digitize, preserve and extend access
to The Quarterly Journal of Economics
This content downloaded from 34.192.2.131 on Thu, 13 Sep 2018 21:50:57 UTC
All use subject to https://about.jstor.org/terms
AN EMPIRICAL CHARACTERIZATION OF THE DYNAMIC
EFFECTS OF CHANGES IN GOVERNMENT SPENDING
AND TAXES ON OUTPUT*
OLIVIER BLANCHARD AND ROBERTO PEROTTI
This paper characterizes the dynamic effects of shocks in government

ing and taxes on U. S. activity in the postwar period. It does so by usin
structural VAR/event study approach. Identification is achieved by using
tional information about the tax and transfer systems to identify the
response of taxes and spending to activity, and, by implication, to inf
shocks. The results consistently show positive government spending s
having a positive effect on output, and positive tax shocks as having a
effect. One result has a distinctly nonstandard flavor: both increases in t
increases in government spending have a strong negative effect on in
spending.
I. INTRODUCTION
The predominant, Keynesian, view of the effects of fiscal

policy that was embedded in the large-scale macroeconometric
models of the seventies and eighties has come under attack.
Theoretically, in the neoclassical approach that has developed in
the last twenty years, government spending can have drastically
different effects than in Keynesian models, particularly on pri-
vate consumption. Empirically, the response of the economy to
several episodes of fiscal retrenchment in the last fifteen years
has been at odds with conventional Keynesian wisdom: on several
occasions, private consumption and GDP increased significantly
while government spending was severely cut. Finally, the evi-
dence from large-scale econometric models has been largely dis-
missed on the grounds that, because of their Keynesian structure,
these models assume rather than document a positive effect of
fiscal expansions on output.
* We thank the editor and the two referees for useful comments. We thank
James Poterba, Robert Solow, and seminar participants at Bocconi University,
Catholic University in Milan, Columbia University, New York University, Prince-
ton University, University of Mannheim, the Bank of Italy, the ESSIM conference
in Sintra, Portugal, and the NBER Summer Institute for comments and sugges-
tions. Conversations with Christopher Sims also helped us clarify several issues.
We thank Eric Hilt and Douglas Smith for excellent research assistance, Jonas
Fisher for sharing his data, and Jonathan Gruber for help with the data. We
thank the National Science Foundation for financial support. The data, together
with the details of data construction and estimation, and the results mentioned
but not reported in the text, are available at http://www.iue.it/Personal/Perotti
and http://web.mit.edu/blanchar/www/.
o 2002 by the President and Fellows of Harvard College and the Massachusetts Institute of
Technology.
The Quarterly Journal of Economics, November 2002
1329
1330 QUARTERLY JOURNAL OF ECONOMICS
In view of these challenges, our purpose in this paper is to

characterize the dynamic effects of shocks in government spend-
ing and taxes on economic activity in the United States during
the postwar period. We do so using a structural VAR approach. To
achieve identification, we rely on institutional information about
the tax and transfer systems and the timing of tax collections to
construct the automatic response of fiscal policy to economic
activity, and, by implication, to identify the shocks to fiscal policy.
A related structural VAR approach has been used in a number of
studies to assess the effects of monetary policy (see, in particular,
Bernanke and Mihov [1998]). We believe that such an approach is
actually better suited for the study of fiscal policy, for two rea-
sons. First, in contrast to monetary policy, fiscal variables move
for many reasons, of which output stabilization is rarely predomi-
nant; in other words, there are exogenous (with respect to output)
fiscal shocks. Second, again in contrast to monetary policy, deci-
sion and implementation lags in fiscal policy imply that, at high
enough frequency-say, within a quarter-there is little or no
discretionary response of fiscal policy to unexpected contempora-
neous movements in activity. Thus, with enough institutional
information about the tax and transfer systems, one can con-
struct estimates of the automatic effects of unexpected move-
ments in activity on fiscal variables, and, by implication, obtain
estimates of fiscal policy shocks. Having identified these shocks,
one can then trace their dynamic effects on GDP and its compo-
nents. This is what we do in this paper.
In a methodological twist that was imposed on us by the data
but is likely to be useful in other contexts, we combine this
structural VAR approach with one akin to an event-study ap-
proach: a few episodes of large discretionary changes in taxation
or in expenditures are simply too large to be treated as realiza-
tions from the same underlying stochastic process and must be
treated separately. We trace the effects of these large, one-time,
changes by studying the dynamic response of output to an asso-
ciated dummy variable that we include in the VAR specification.
As we show when we look at the 1950s, not all such large fiscal
events can be used so cleanly; when they can, as in the case of the
1975 temporary tax cut, we find a high degree of similarity between
the impulse responses obtained by tracing the effects of estimated
VAR shocks, and by tracing the effects of these special events.
Our results consistently show positive government spending
shocks as having a positive effect on output, and positive tax
AN EMPIRICAL CHARACTERIZATION OF THE EFFECTS 1331
shocks as having a negative effect. The size and persistence of these

effects vary across specifications (for instance, whether we treat
time trends as deterministic or stochastic) and subperiods; yet, the
degree of variation is not such as to cloud the basic conclusion.
We also consistently find a positive effect of government
spending on private consumption, a straightforward implication
of virtually all Keynesian models but a result that is difficult to
reconcile with the neoclassical approach, except under counter-
factual assumptions about the path of taxation over time. In
contrast, we find that both increases in taxes and increases in
government spending have a strong negative effect on private
investment spending. This effect is consistent with a neoclassical
model with distortionary taxation, but more difficult to reconcile
with Keynesian theory: while agnostic about the sign, Keynesian
theory predicts opposite effects of tax and spending increases on
private investment. This does not appear to be the case.
Our paper is organized as follows. Section II presents the
main specification, and discusses identification. Section III pre-
sents the data and discusses their main properties, both at high
and low frequencies. Section IV discusses the contemporaneous
relations between shocks to government spending, net taxes, and
output. Section V presents the dynamic effects of tax shocks.
Section VI does the same for spending shocks. Section VII dis-
cusses robustness, including subsample stability, quarter depen-
dence, cointegration, and the use of alternative net tax elastici-
ties. Section VIII takes up the important issue of anticipated
fiscal shocks, i.e., the possibility that, precisely because of imple-
mentation lags in policy, what we as econometricians treat as
unexpected shocks are in fact anticipated by private agents in the
economy. Section IX extends the sample to include the 1950s, and
shows what can be learned from the Korean War buildup. Section
X relates our results to those in the recent papers on the effects of
fiscal policy, both those in the VAR and those in the event-study
tradition. Section XI takes a first pass at a more disaggregated
analysis, and presents the response of the individual output com-
ponents. Section XII concludes.
II. METHODOLOGICAL ISSUES
Both government expenditure and taxation affect GDP: since

the two are presumably not independent, to estimate the effects of
one, it is also necessary to include the other. Hence, we focus on
two-variable breakdowns of the budget, consisting of an expenditure

and a revenue variable.
We define the expenditure variable as total purchases of
goods and services, i.e., government consumption plus govern-
ment investment. We call it "government spending," or simply
"spending" for short. We define the revenue variable as total tax
revenues minus transfers (including interest payments). We call
it "net taxes," or "taxes" for short. We discuss the data in more
detail in Section III.1
A. The VAR
Our basic VAR specification is
(1) Yt = A(L,q)YtI + U,,

where Y
rithms of [Tt,Gt,Xt]' is a three-dimensional
quarterly taxes, vector
spending, and GDP, all in per
in real, thecap
log
terms.2 We use quarterly data because, as we discuss below, this
essential for identification of the fiscal shocks. We allow either for
deterministic (quadratic trends in logs), or stochastic (unit root with
slowly changing drift) trends, and for a number of dummy variables;
we defer a discussion of these issues to later.
Ut [tt, g,, xt]' is the corresponding vector of reduced-form
residuals, which in general will have nonzero cross correlations.
A(L,q) is a four-quarter distributed lag polynomial that al-
lows for the coefficients at each lag to depend on the particular
quarter q that indexes the dependent variable. The reason for
allowing for quarter-dependence of the coefficients is the presence
of seasonal patterns in the response of some of the taxes to
economic activity. Suppose, for illustrative purposes, that a tax is
paid in the last quarter of each year, for activity over the year:
then, in the last quarter, the tax revenue will depend on GDP in
the current and past three quarters; in the other three quarters,
1. Any decomposition and choice of two fiscal variables reflects one's theoretical
priors. Ours is no exception and reflects our belief that, in the short run, fiscal policy
works mainly through the effect of spending and taxes on aggregate demand and the
effect of aggregate demand on output. A researcher who viewed fluctuations instead
as real business cycles and believed in Ricardian equivalence, would likely choose a
different two-variable decomposition, such as government consumption and govern-
ment investment, or government spending and the marginal tax rate.
2. We use the GDP deflator to express the variables in real terms. This allows
us to express the impulse responses as shares of GDP. Results using the own
deflator to express spending in real terms are very similar.
it will be equal to zero and thus will not depend on GDP.3 We have
collected evidence on quarter dependence in tax collection over
the sample period from various institutional sources. The Appen-
dix lists the main relevant features of the tax code.4
B. Identification
We now discuss our identification methodology. Without loss
of generality, we can write
(2) tt = a ix + a2eg + et
(3) gt bxt + b2et + e
(4) x = Cltt + c2gt e,

where et, t, and are the mutually uncor
shocks that we want to recover.
The first equation states that unexpected movements in
taxes within a quarter, tt, can be due to one of three factors: the
response to unexpected movements in GDP, captured by a 1 xt, the
response to structural shocks to spending, captured by a2 eg, and
to structural shocks to taxes, captured by et. A similar interpre-
tation applies to unexpected movements in spending in the sec-
ond equation. The third equation states that unexpected move-
ments in output can be due to unexpected movements in taxes,
unexpected movements in spending, or to other unexpected
shocks, ex. Our methodology to identify this system can be divided
into three steps.
(1) We rely on institutional information about tax, transfer,
and
and spending programs
b1. A priori, to construct could
these coefficients the parameters al
capture two
different effects of activity on taxes and spending: the
automatic effects of economic activity on taxes and spend-
ing under existing fiscal policy rules, and any discretion-
ary adjustment made to fiscal policy in response to unex-
pected events within the quarter. The key to our
identification procedure is to recognize that the use of
quarterly data virtually eliminates the second channel.
3. Note that, because equation (1) is a reduced form, quarter dependence in

the relation of taxes to GDP can show up in all three equations; we thus have to
allow for quarter dependence in all three equations.
4. In some cases the pattern of collection lags has changed--slightly--over
time. Allowing for changes in quarter dependence in the VAR over time would
have quickly exhausted all degrees of freedom. We have not done it.
Direct evidence on the conduct of fiscal policy suggests

that it takes policymakers and legislatures more than a
quarter to learn about a GDP shock, decide what fiscal
measures, if any, to take in response, pass these mea-
sures through the legislature, and actually implement
them. The same would not be true if we used annual data:
to a degree, fiscal policy can be adjusted in response to
unexpected changes in GDP within the year.5
Thus, to construct a 1 and b1, we only need to construct
the elasticities to output of government purchases and of
taxes minus transfers. To obtain these elasticities, we use
information on the features of the spending and tax/
transfer systems:
We could not identify any automatic feedback from
economic activity to government purchases of goods and
services; hence, we take b 1 = 0.
Turning to net taxes, write the level of net taxes, P', as
P = I i, where the Pi's are positive if they correspond
to taxes, negative if they correspond to transfers." Let Bi
be the tax base which corresponds to tax Ti (or, in the
case of transfers, the relevant aggregate for the transfer
program, i.e., unemployment for unemployment bene-
fits). We can then write the within-quarter elasticity of
net taxes with respect to output, a , as
(5)~ a1 = j Th'B, i
ThY
(5) a, =I~~al-- Z q Oti ,.
where T1Ti,Bi denotes the elasticity of taxes of type i t
their tax base, and Bi,,X denotes the elasticity of the ta
base to GDP.
To construct these elasticities, we extend earlier work
by the OECD [Giorno et al. 1995], who have constructed
' T,,Bi and nSBi,x for four separate categories of taxes.

These elasticities were computed with respect to annual
5. Although the logic of our approach is similar to that of Bernanke and

Mihov [1998] and Gordon and Leeper [1994], our identification methodology is
different in one important respect. In those papers (which look at monetary, not
fiscal policy) identification is achieved by assuming that private sector variables
do not react to policy variables contemporaneously. By contrast, we assume that
economic activity does not affect policy, except for the automatic feedback built in
the tax code and the transfer system.
6. The tilde denotes the level of net taxes. We have used T earlier to denote
the logarithm of net taxes per capita.
changes; we do the same with respect to quarterly

changes. The details are presented in the Appendix. The
main difference between the OECD's and our estimated
overall elasticity a, comes from differences in the quar-
terly and annual elasticities of the tax bases with respect
to GDP. For instance, the contemporaneous elasticity of
profits to GDP, estimated from a regression of quarterly
changes in profits on quarterly changes in GDP, is sub-
stantially higher than the elasticity obtained from
a regression using annual changes-4.50 versus 2.15,
respectively.
The average value of our measure of a, over the 1947:
1-1997:4 period is 2.08; it increases steadily from 1.58 in
1947:1 to 1.63 in 1960:1 to 2.92 in 1997:4.7 Using a
different method, Cohen and Folette's [2000] work im-
plies an estimate of (annual) a, ranging from 1.7 in 1951
to 2.2 in 1998. The similarity of results, obtained with two
different methods, is reassuring.
(2) With these estimates of a1 and b1, we construct the
cyclically adjusted reduced-form tax and spending resid-
uals, t t - a1xt and gt - gt - b1x, - gt (as b1 - 0).
Obviously, t' and g' may still be correlated with each
other, but they are no longer correlated with ex. Thus, we
can use them as instruments to estimate c1 and c2 in a
regression of x, on t, and gt.
(3) This leaves two coefficients to estimate, a2 and b2. There
is no convincing way to identify these coefficients from
the correlation between t' and g': when the government
increases taxes and spending at the same time, are taxes
responding to the increase in spending (i.e., a2 # 0, b2 =
0) or the reverse? We thus present the results under two
alternative assumptions. Under the first, we assume that
7. The value ofa 1 varies over time, both because the ratios of individual taxes
and transfers to net taxes-the terms Ti/T in expression (5) above-and the tax
base elasticities of tax revenues-the terms ITq Bs-have changed over time. The
fact that a1 varies over the sample sugges& 'time variation in the dynamic
responses of spending and taxes to activity, and thus time variation of the VAR.
Except for tests of subsample stability reported later, we have not explored this
time variation further.
One implicit assumption in our construction of a, is that the relation between
the various tax bases and GDP is invariant to the type of shock affecting output.
For broad-based taxes, such as income taxes, this is probably fine. It is more
questionable, say, for corporate profit taxes: the relation of corporate profits to
GDP may well vary depending on the type of shock affecting GDP.
tax decisions come first, so that a2 = 0, and we can

estimate b2. Under the second, we assume that spending
decisions come first, so that b2 - 0, and we can estimate
a2. It turns out that, in nearly all cases, the correlation
between t' and g' is sufficiently small that the ordering
makes little difference to the impulse response of output.8
C. Impulse Responses
Having identified the tax and spending shocks, we can study
their effects on GDP. We face two issues here. The first is that the
constructed elasticity of net taxes to output, a1, varies over time.
We ignore this here and compute the impulse response using the
mean value of a1 over the sample, i.e., 2.08. Second, one of the
implications of quarter dependence is that the effects of fiscal
policy vary depending on which quarter the shock takes place. To
avoid estimating four different impulse responses, we estimate
the covariance matrix from the quarter dependent VAR, and then
use a VAR estimated without quarter dependence (except for
additive seasonality) to characterize the dynamic effects of the
shocks. Thus, our impulse responses give, admittedly only in a
loose sense, the average dynamic response to fiscal shocks.
III. THE DATA
We define net taxes as the sum of Personal Tax and Nontax

Receipts, Corporate Profits Tax Receipts, Indirect Business Tax
and Nontax Accruals, and Contributions for Social Insurance,
less Net Transfer Payments to Persons and Net Interest Paid b
Government. Government spending is defined as Purchases of
Goods and Services, both current and capital. The source is the
Quarterly National Income and Product Accounts, with the ex
ception of Corporate Profits Tax Receipts, which are obtained from
the Quarterly Treasury Bulletin. All these items cover the gen
8. As in all small-dimensional VARs, one should worry about omitted vari-

ables. Controlling for inflation, in particular, may be potentially important be
cause all our variables are expressed in real terms, but spending on goods an
services is typically budgeted in nominal terms, and personal income tax brackets
are not indexed contemporaneously. For these reasons, inflation shocks are likel
to affect both spending and taxes.
If activity shocks affect inflation within the quarter, then our constructed
values of a, and b 1 will be incorrect, as they ignore the effect of activity through
inflation on taxes and spending. However, in our data, the correlation between
quarterly inflation and real GDP shocks is small (0.01), so this does not appear t
be a major issue. We have not explored this further.
eral government, i.e., the sum of the federal, state, and local
governments, and social security funds.9 All the data are season-
ally adjusted by the original source, using some variant of the X11
method.10
A. High-Frequency Properties
Figure I displays the behavior of the ratio of government
spending (i.e., purchases of goods and services) and of net taxes
(i.e., taxes minus transfers) to GDP over the longest available
sample, 1947:1 to 1997:4. It is obvious that these series display a
few extremely large quarterly changes in taxes and spending,
well above three times their standard deviations of 4.3 percent
and 1.9 percent, respectively.
There are two particularly striking changes in net taxes.
First, the increase in 1950:2 by about 26 percent (more than six
times the standard deviation), followed by a further increase in
1950:3 by about 17 percent (or about four times the standard
deviation). These episodes represent in part the reversal of the
temporary 8 percent fall in net taxes in 1950:1, caused by a large
one-time payment of National Service Life Insurance benefits to
the war veterans; but mostly they represent a genuine increase in
tax revenues. The second episode is the large temporary tax
rebate of 1975:2, which results in a net tax drop by about 33
percent for one quarter.11
On the expenditure side, the Korean War stands out: in
1951:1, at the onset of the Korean War buildup, government
spending increases by 10 percent (more than five times the stan-
dard deviation), and continues to grow at about the same quar-
terly rate during the next two quarters, 1951:2 and 1951:3; after
this, spending continues to increase at more than twice the stan-
dard deviation in 1951:4 and again in 1952:2. It is difficult to
think of the early 1950s as being generated by the same stochas-
9. We do not have data on the corporate profit tax on a cash basis for
state and local governments. This represents about 5 percent of total cor-
porate profit tax receipts at the beginning of the sample, and about 20 percent at
the end.
10. Corporate profit tax receipts are only reported without a seasonal adjust-
ment. We use the RATS EZ-X11 routine to seasonally adjust this series.
11. See Blinder [1981] for a detailed analysis of this tax cut, and its effects on
consumption. The 1975:2 tax rebate (which was combined with a social security
bonus for retirees without taxes to rebate) corresponded to an increase in dispos-
able income of about $100 billion at 1987 prices; by comparison, the 1968 surtax
decreased disposable income by $16 billion and the 1982 tax cut increased dis-
posable income by $31.6 billion, always at 1987 prices (see Poterba [19881).
taxlgdp
0.225
0.200 -
0.175 -
0.150 -
0.125 1 47
. . . .54
. 1 I.. ..1 1 . . . . . . . . . . . . . ..1. . . . . . . . . .
61 68 75 82 89 96
gcn/gdp
0.26
0.24 -
0.22
0.20 -
0.18 -
0.16 -
47 54 61 68 75 82 89 96
FIGURE I
Net Taxes and Spending, Shares of GDP
tic process as the post-1960 period. Thus, our strategy is to

proceed in two steps. For most of the paper we run regressions
starting from 1960:1. In Section VIII we extend the sample back
to include the 1950s and look at what we can learn from this
longer sample.
Note that our benchmark sample still includes one large net
tax cut episode, the 1975:2 tax cut. This episode is a well-identi-
fied, isolated, temporary, tax cut. Hence, it can be easily and
clearly dummied. This allows us to compare two different types

impulse responses with a net tax shock: one tracing the dynamic
effects of the estimated net tax shocks (our first approach), the
other tracing the effects over time of a unit shock to the 1975:2
dummy variable (our second approach).
B. Low-Frequency Properties
The general visual impression from Figure I is one of no clear
trends, but clear low-frequency (say decade-to-decade) move-
ments in both spending and taxes. One may be surprised by th
general absence of upward trends in spending and net taxes; b
recall that we are looking at government spending not including
transfers, and that net taxes are taxes net of transfers. Thus, th
figures hide the trend increases in taxes and transfers, which
have indeed taken place during this period.
The main practical issue, for our purposes, is how to treat
these low-frequency movements in our two fiscal series in relatio
to output. We have conducted a battery of integration tests for T
G, and X. Formal tests (Augmented Dickey-Fuller and Phillips
Perron, with a deterministic time trend) do not speak strongly o
whether we should assume stochastic or deterministic trends for
each variable. We have also conducted a battery of cointegration
tests. One obvious candidate for a cointegration relation is the
difference between taxes and spending, T - G:12 in fact, the
stationarity of the deficit is the basic idea underlying the tests of
"sustainability" of fiscal policy by Hamilton and Flavin [1986] and
Bohn [1991].13 Figure II displays the logarithm of the tax/spend-
ing ratio. Again, formal test results do not speak strongly: one can
typically reject the null of a unit root at about the 5 percent level,
but no lower.14
In light of these results, we estimate our VARs under two
alternative assumptions. In the first, we formalize trends in all
three variables as deterministic, and allow for linear and qua-
dratic terms in time in each of the equations of the VAR. In the
second, we allow for stochastic trends. We take first-differences of
each variable, and, to account for changes in the underlying drift
12. Recall that T is net of interest payments.

13. Note that we test cointegration between the logarithms of taxes and
spending. This is equivalent to testing cointegration between the logarithms of the
net tax/GDP and of the spending/GDP ratios.
14. This lack of strong evidence for cointegration between T and G is consis-
tent with a number of recent empirical studies: see, e.g., Bohn [1998], who tests for
cointegration over the 1916-1995 period using annual data.
log(tax)-log(gcn)
0.2 -
0.1
-0.0
-0.1
-0.2
-0.3
-0.4
-0.5
-0 .6 1 1 1 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
47 54 61 68 75 82 89 96
FIGURE II
log (Net Taxes) - log (Spending)
terms, we subtract a changing mean, constructed as the geomet-

ric average of past first differences, with decay parameter equal
to 2.5 percent per quarter (varying this parameter between 1 and
5 percent makes little difference to the results). For brevity, in
what follows, we will refer to the two specifications as "DT" (for
"deterministic trend") and "ST" (for "stochastic trend"), respec-
tively. In both specifications, we allow for the current value and
four lags of a dummy for 1975:2.
In our benchmark specifications, we do not impose a cointe-
gration restriction between the tax and the spending variables. In
Section VII we discuss results under the restriction that T and G
are cointegrated. As it turns out, this makes little difference to
the results.
From now on, unless otherwise noted (in particular in
Section VIII), our results are based on the 1960:1-1997:4
sample.
TABLE I
LARGE CHANGES IN NET TAXES AND SPENDING
sd(A log G) = 0.019 sd(A log T) = 0.049
A log G > 3 sd A log T > 3 sd

1951:1 0.103 1950:2 0.266
1951:2 0.112 1950:3 0.171
1951:3 0.108 1975:2 -0.335
1975:3 0.240
2 sd < A log G < 3 sd 2 sd < A log T < 3 sd
1948:2 0.039 1947:3 -0.117
1948:4 0.043 1947:4 0.107
1949:1 0.049 1951:1 0.097
1949:2 0.043
1950:4 0.054
1951:4 0.051
1952:2 0.041
1967:1 0.041
IV. CONTEMPORANEOUS EFFECTS
The two panels of Table II report the estimated coefficients of

the contemporaneous relations between shocks in equation (3),
under DT and ST, and, for a2 and b2, under the two alternative
assumptions that taxes come first, or that spending comes first.15
For convenience of interpretation, while the original estimated
coefficients have the dimension of elasticities, the table reports
derivatives, evaluated at the point of means (dollar change in one
variable per dollar change in another). Table II yields two main
conclusions.
The first is that the signs of the contemporaneous effects of
taxes and of spending on GDP-c1 and c2-are those one would
expect-the former negative and the latter positive, and are
rather precisely estimated. The two coefficients have very similar
absolute values, and are also very similar across the two specifi-
cations, DT and ST. Under DT, a unit shock to spending increases
GDP within the quarter by 0.96 dollars, while a unit shock to
taxes decreases GDP by 0.87 dollars. The estimated negative
effect of taxes on output depends very much on the use of instru-
ments (as it should): the simple correlation between unexpected
15. Note that, in constructing the cyclically adjusted tax shock t', we use the
time-varying elasticity a , not its mean.
TABLE II
ESTIMATED CONTEMPORANEOUS COEFFICIENTS
cl c2 b2 a2
DT
coeff. -0.868 0.956 -0.047 -0.187

t-stat. -3.271 2.392 -1.142 -1.142
p-value 0.001 0.018 0.255 0.255

ST
coeff. -0.876 0.985 -0.057 -0.238

t-stat. -3.255 2.378 -1.410 -1.410
p-value 0.001 0.019 0.161 0.161
DT: Deterministic Trend; ST: Stochastic Trend.

Sample: 1960:1-1997:4.
c1: effect of t on x within quarter;
c2: effect of g on x within quarter;
a2: effect ofg on t within quarter (assuming b2
b2: effect of t on g within quarter (assuming a2
All effects are expressed as dollar for dollar.
movements in cyclically una

movements in output, xt, is po
the issue of the robustness of t
justed taxes to the specific va
Section VII and show that the basic conclusions hold over the
plausible range of values for a1.
The second conclusion is that the correlation between cyclically
adjusted tax and spending innovations is low (-0.09 in our sample)
yielding relatively low estimated values of a2 and b2 under either
of the two alternative identification assumptions. These small
values imply that the choice between the two orderings of taxes
and spending makes little difference to the impulse responses.
V. DYNAMIC EFFECTS OF TAXES
A. Effects of Estimated Tax Shocks

The top panel of Figure III shows the effects of a unit tax
shock assuming that taxes are ordered first (a2 = 0), under DT;
the bottom panel does the same under ST.16 For visual conve-
16. Note that the initial value of the change in taxes is not exactly 1. A unit
tax shock e' translates into a less than unit change in taxes tt, since GDP falls in
response to the shock, decreasing tax revenues.
resp. of tax, DT resp. of tpe, DT resp. of gdp
10 1010
5 0 os 05 -
5 05 - -05 -
-10o -10 - -10
-15 -15 -15
-20 -20 -20

1 4 7 10 13 16 19 1 4 7 10 13 16 19 1 4 7 1
resp. of tax, ST reap. of spe, ST reap. of gdp

10 - 10 10 -
05- 05 05 -
00 - 00 00
-05 -05 -05
-- - - - - - - - -
10 -10 0 -o
-15 1 - 15 9
-20 - I I - 2 0 -20 -. . . .
1 4 7 10 13 16 19 1 4 1 10 13 16 19 1 4 7 1
FIGURE III
Response to a Tax Shock
TABLE III
RESPONSES TO TAx SHOCKS
1 qrt 4 qrts 8 qrts 12 qrts 20 qrts peak

DT
GDP -0.69* -0.74* -0.72* -0.42* -0.22 -0.78* (5)

TAX 0.74* 0.13 -0.21* -0.20* -0.11
GCN -0.05* -0.12* -0.24* -0.26* -0.16*
ST
GDP -0.70* -1.07* -1.32* -1.30* -1.29* -1.33* (7)

TAX 0.74* 0.31* 0.17 0.16 0.16
GCN -0.06* -0.10* -0.17* -0.20* -0.20*
DT: Deterministic Trend; ST: Stochastic Trend. An asterisk indi

the two one-standard error bands. In parentheses besides the peak
All reduced-form equations include lags 0 to 4 of the 1975:2 dum
nience, the impulse responses in these an

tables are transformations of the origin
give the dollar response of each variable
the fiscal variables. The solid line give
broken lines give one-standard deviation
Carlo simulations (assuming normality),
Table III summarizes the main features of
variables. This table is useful in compa
benchmark specifications with alternativ
Under DT (top panel of Figure III an
falls on impact by about 70 cents an
quarters out, with a multiplier (defin
ratio of the trough response of GDP t
0.78. From then on, output increases s
effect of tax shocks on government spe
zons, with the largest effect being -0
but it is precisely estimated.
Under ST (bottom panel of Figure
response of output is stronger and m
have a very similar effect on output o
trough is larger (-1.33 against -0.78
after seven quarters instead of five; a
output stabilizes at around the peak resp
is slightly more persistent than in the D
spending is similar.
Thus, under both specifications, tax increases have a nega-

tive effect on output. In both cases, the effect on output takes time
to build up, with the largest response occurring after five or seven
quarters depending on the specification. The negative response of
output is more pronounced under the assumption of a stochastic
trend.
When taxes are ordered second, the results (not shown) are
virtually identical to those in Figure III. This comes from the low
correlation between cyclically adjusted tax and spending innova-
tions, which in turn leads to a small value of a2.
B. Dynamic Effects of the 1975:2 Net Tax Cut

We now study the dynamic effects of a unit shock to the
1975:2 dummy variable, under DT and ST, respectively.17 The
two panels of Figure IV display the results: as we are looking at
a tax decrease, the signs are reversed relative to Figure III. But
the effects are very similar. The impulse response captures well
the fact that this tax decrease was temporary: taxes are back to
normal after a quarter. The effects on output take some time to
build up, reaching a peak after four quarters with a multiplier of
0.75 under DT and 1.02 under ST, thus close to the multipliers in
Table III. As usual, the response of output is more persistent
under ST. The effects on spending are again small.
We find this similarity of results with the impulse responses
from identified shocks comforting. The main difference is that the
largest output response is reached sooner than in the impulse re-
sponse to the typical tax shock; this is plausibly explained by the
smaller persistence of the tax shock in the 1975:2 episode than of the
typical tax shock in Figure III. (One would also have expected the
more temporary tax cut to lead to a smaller change in the present
value of taxes, thus a smaller initial effect on consumption, and a
smaller multiplier. There is no evidence that this is the case.)
VI. DYNAMIC EFFECTS OF SPENDING SHOCKS
The two panels of Figure V show the effects of a unit spend-

ing shock on GDP when spending is ordered first (b2 0= ) under
17. Note that this exercise can be given a structural interpretation only if the
tax cut captured by the dummy variable was unanticipated and was not a
response to a contemporaneous output shock. In other words, identification in the
"event-study" approach requires the same two assumptions that allowed identi-
fication in the "structural VAR" approach.
resp. of tax, DT resp. of spe, DT resp. of gdp
10 10 1
6- 6- -
4 4- 4
2 2 -- - --- 2
-2 -2 .2
S- -6
1 3 5 7 9 11 13 15 17 19 1 3 5 7 9 11 13 15 17 19 1 3 5 7 9
1
resp. of tax, ST resp. of ape, ST resp. of gdp
1
6 a 6 /
4 4 - 4
2 -2 - 2 /2
-8 --/----
/ 2- -
1 3 5 7 9 11 13 15 17 19 1 3 5 7 9 11 13 15 17 19 1 3 5 7
FIGURE IV
Response to a Shock to the 1975:2 Dummy
TABLE IV
RESPONSES TO SPENDING SHOCKS
DT
GDP 0.84* 0.45 0.54 1.13* 0.97* 1.29* (15)

GCN 1.00* 1.14* 0.95* 0.70* 0.42*
TAX 0.13 0.14 0.17 0.43* 0.52*
ST
GDP 0.90* 0.55 0.65 0.66 0.66 0.90* (1)

GCN 1.00* 1.30* 1.56* 1.61* 1.62*
TAX 0.10 0.18 0.33 0.36 0.37
DT and, then, under ST. As

marizes the main features o
under alternative specificati
Under DT (top panel of Fi
shocks are longer lasting than
is still there after two year
dollars, then declines, and r
1.29 after almost four year
over the same horizon, prob
response of GDP (notice that
the shape of the output resp
The peak output response
of Figure V and Table IV), 0
now reached on impact rath
that the impact response is
standard error bands are also
output becomes insignifican
strong response of spending,
years.
Thus, in all specifications output responds positively to a
spending shock. Spending reacts strongly and persistently to its
own shock. Depending on the specification, the spending multi-
plier is larger or smaller than the tax multiplier. (Traditional
Keynesian theory holds that the spending multiplier should be
larger than the tax multiplier; there is no consistent evidence
that this is the case.)
As in the case of taxes, the ordering of the two fiscal variables
resp. of ape, DT resp. of tax, DT resp. of gdp
25 - 25 25
20 -20 - 20 -
15 15 15 s /
05 05 05
00 o00 . 00 o
\ /
-05 -05 - -o05

.10 -10. -10
1 4 7 10 13 16 19 1 4 7 10 13 16 19 1 4 7 1
reap. of ape, ST reap. of tax, ST reap. ofgd

25 25 215
20 -- . 20 20
Is. 15 I
I
-05 -05 -05
.10 .10 I'10 f i l l i l l I w i l l -lo

1 4 7 10 13 16 19 1 4 1 10 13 16 19 1 4 7 1
FIGURE V
Response to a Spending Shock
TABLE V
STABILITY OF RESPONSES TO TAX SHOCKS
Net taxes Spending
excl. period max. GDP response excl. period max. GDP response
DT
60-69 -1.18* (1) 60-69 1.44* (1)

70-79 -0.90* (5) 70-79 1.47* (10)
80-89 -0.49* (2) 80-89 0.96* (3)
90-97 - 1.45* (7) 90-97 1.73* (12)
ST
60-69 -1.45* (11) 60-69 1.25* (1)

70-79 -1.48* (4) 70-79 0.62* (1)
80-89 -0.83* (7) 80-89 1.80* (3)
90-97 - 1.52* (7) 90-97 0.85* (12)
does not matter for the response t

are ordered first, a shock to spe
effects on output (not shown) to t
VII. ROBUSTNESS
We take up three issues in this section, subsample stability,

the effect of imposing cointegration, and robustness to alternative
assumptions about the tax elasticity (detailed results are given on
the websites given in the acknowledgment note).
Subsample stability. Unlike monetary policy, fiscal policy
in the United States in our sample does not lend itself easily to a
periodization based on alternative policy regimes. Hence, we
check for subsample stability by dropping one decade at a time.
Table V reports the results from this exercise. The left and right
parts summarize the impulse responses of output to a net tax
shock and a spending shock, respectively, obtained by dropping
one decade at a time.
The exclusion of the eighties causes a substantial drop in the
magnitude of the tax multiplier, particularly under DT: the tax
multiplier ranges from -0.49 when the eighties are excluded to
- 1.45 when the nineties are excluded. There is also evidence of
some instability in the spending multiplier, in particular under
ST: the multiplier when the eighties are excluded is about three
times that when the seventies are excluded-1.80 against 0.62.

We do not have a convincing explanation for the pattern of change
of impulse responses over time.
Cointegration. The imposition of a cointegrating relation
between G and T yields very similar results to our benchmark
case. 18
Alternative net tax elasticities. Our identification proce-
dure depends crucially on using E' as an instrument. The con-
struction of this series in turn depends very much on the elastic-
ity of net taxes a . One may ask how sensitive the results are to
alternative values of a1. We have reestimated the impulse re-
sponse to a tax shock, under DT and ST, when taxes are ordered
first (a2 = 0), and for three different values of a,: the baseline
value minus 0.5, the baseline value, and the baseline value plus
0.5. We believe that the range implied by these three values more
than covers the relevant range for a . A simple way of describing
the results is that the general shape of the response of GDP to a
tax shock is similar across values of a,. What changes is the size
of the multiplier. Under DT, for example, the multiplier goes from
0.55 (in absolute value) for a1 equal to the benchmark value
minus 0.5, to 1.05 for a1 equal to the benchmark value plus 0.5.
We read these results as saying that the broad response of output
we characterized earlier is robust to the details of construction
of a,.
VIII. ANTICIPATED FISCAL POLICY
Fiscal policy is subject to two types of lags: decision lags,

which imply that it takes some time for policy to be changed in
response to shocks; implementation lags, which imply that it
takes some time for policy changes to be implemented. The first
source of lags is what helped us achieve identification. The second
source, however, raises a problem we have ignored until now,
namely that what we (the econometrician) identify as fiscal
shocks may be the result of earlier policy changes, and thus, in
fact, be anticipated by the private sector.
To see the problems this raises and how they may be solved,
let us go back to our original specification of the joint movement
of taxes, spending, and output in equations (1), but now allow the
18. In keeping with quarter dependence, we allow the coefficient of the

cointegrating term Gt 1 - Tt-1 to differ depending on the quarter.
private sector to know fiscal shocks one quarter in advance. For

notational simplicity, let us look at a two-variable VAR in T and
X, and ignore quarter-dependence:
(6) Tt = aXt + A11(L)T_-, + A12(L)Xt_1 + et
(7) Xt = coETTt+l + clTt + A21(L)Tt-1 + A22(L)Xt-, + et.

The first equation is as before: taxes depend on output, current
and lagged, and lagged taxes, together with fiscal shocks. The
second equation now allows output in quarter t to depend not only
on current and lagged taxes, but also on the expectation of taxes
in quarter t + 1, based on an information set which includes et +l.
In other words, this specification assumes that, because of imple-
mentation lags, the private sector knows fiscal shocks one quarter
ahead, and may therefore react to them one quarter before they
are actually observable by the econometrician.
To see the econometric problems this specification raises,
rewrite the equation for output as
(8) X = coTt+l1 + clTt + A21(L)Tt-1 + A22(L)Xt-1 + E',

where the error term is defined as E" [e-co(T
EtTt+l)].
From the assumption that et+1 is known at time t, the term
in parentheses in the composite error term is uncorrelated with
etl,+ 1but
Tt in (8)will
are typically be correlated
correlated with the compositewith
error eterm.
+1. As
Hence,
in our both Tt and
benchmark case, Tt is correlated with e', through the effect of e'
on output and in turn on tax revenues. Also, now Tt+1 is corre-
lated with both components of the error term: with e+ 1 through
the effects of the latter on taxes in quarter t + 1, and with e',
through the effect of the latter on Tt and therefore on Tt+ .
Can equation (8) be estimated by instrumental variables?
The answer is yes, if we are willing to make stronger identifica-
tion assumptions. The reasoning follows the logic of our previous
identification strategy. If we can construct a series for e, then et
and its value led once, e t1, can be used as instruments for Tt and
T,,+ 1 Both are correlated with Tt or Tt+ 1 and uncorrelated with
the two components of the composite error term. Put another
way, if we can identify et in the tax equation, we can then use it
and its value led once as instruments in the output equation.
With this in mind, let us turn to the tax equation, and for
reasons that will be clear below, rewrite it by separating X, 1

from further lagged output terms:
(9) Tt = alXt + al21Xt_- + A11(L)T,1 + A12(L)Xt_2 + e,

where A2 - L-1(A12(L) - A121). Note that both Xt and Xt_1
are likely to be correlated with the error term e': Xt because of the
effect of taxes on output, and Xt 1 because of the assumption that
fiscal shocks in quarter t are known to the private sector in
quarter t - 1.
One way then to achieve identification is to assume that
there is no discretionary response of fiscal policy to output shocks
this quarter (the assumption we made until now) nor to output
shocks last quarter (a stronger assumption than before). In other
words, decision lags prevent a discretionary response of fiscal
policy to output for two quarters.19
We proceed under this assumption, and follow the same
methodology as before: we construct a 1 and a121 from information
about the tax structure and the response of tax bases to GDP, run
the regression above imposing the values of a, and a121 so con-
structed, and obtain a time series for et as the residual.
The construction of a121 under the assumption that it only
reflects the automatic response of taxes to output is presented in
the Appendix. It suggests an average value for a121 over the
1960:1-1997:4 period of 0.16, compared with 2.17 for a1. We
estimate the system with anticipated effects under three alterna-
tive assumptions, a121 = 0.0, 0.16, and 0.5, which can roughly be
thought of as the lower bound, mean, and upper bound estimates
for al21. For simplicity of computation, we ignore quarter depen-
dence in estimation, and for simplicity of presentation, we
present the results only under stochastic detrending (ST). Also,
while we excluded spending for notational simplicity in the argu-
ment above, our VAR specification obviously includes both spend-
ing and taxes, in addition to output. We treat spending symmet-
rically with taxes, allowing for an effect of anticipated spending
next quarter on output this quarter. Extending our earlier as-
sumption about spending, we assume that there is no automatic
effect of output on spending either contemporaneously (our ear-
19. Clearly if fiscal policy were anticipated, say two quarters ahead, identi-
fication would require that policy cannot respond to output shocks this quarter
and in the last two quarters.
1.0
0.5
0.0
-0.5
-1.0
-1.5 -
[] no anticip.
a121 = .16
A a121 = 0.0
-2.0
-2.0 a121=.5 5 10 15 2
0 5 10 15 20
FIGURE VI
Response of Output to an Anticipated Tax Shock, ST
lier assumption) or at a one-quarter lag (an extension of our

earlier assumption).
For either of the three values of a121, there is not much
evidence of an effect of anticipated tax changes on output. The
estimated parameter co varies from 0.07 for a121 = 0 to -0.03 for
a121 = 0.5, and is never significant. There is a bit more evidence
of a positive effect of anticipated government spending on output.
The coefficient on anticipated spending in the output equation
varies from 0.12 when a121 = 0 to 0.19 when a1 21= 0.5, and, in
this last case, is significant at the 10 percent level.
Figure VI shows the effects, from quarter 0 on, of a tax shock
of 1 in quarter 1 under the assumption that a121 = 0.16, the
estimated sample average, or a121 - 0.0, or a121 = 0.5. It does so
by simulating the system of equations above (but extended to
allow for spending), in response to a shock ofet of 1 at t = 1. This
simulation is done under rational expectations: although the
shock occurs at t = 1, anticipations of future taxes and output
affect output and taxes at t = 0. For comparison, the figure also
2.5
2.00 00
1.5
1.0
0.5
0.0
-0.5 [] no anticip.
a121 = .16
A a121 = 0.0
a-1.0 21 = .5
0 5 10 15 20
FIGURE VII
Response of Output to an Anticipated Spending Shock, ST
reports the response in the benchmark case when we did not

allow for an effect from anticipated fiscal policy. Obviously, that
response starts at t = 1.
The response of output at t = 0 is small. After that, the
response of output is qualitatively similar to that without antici-
pated effects; the largest difference occurs when we assume that
a121 = 0, in which case the long-run effect with anticipated
effects is -0.95 against about -1.30 in the case of no effects from
anticipated fiscal policy.
The difference with the case of no anticipated fiscal policy
effects is larger in the case of government spending shocks, dis-
played in Figure VII. Because the coefficient of anticipated spend-
ing in the output equation is large, the effect on output at t = 0
is substantial, around 0.5 percentage points. And because both
the coefficients of lagged output and of contemporaneous spend-
ing in the output equation are also quite large, the response of
output rises even more at t = 1, to about 2.0, and stays there
thereafter. This compares with a response of output in the case
of no effects from anticipated fiscal policy, of about 1.0 in the

long run.
Another approach to the problem of anticipated fiscal policy
is again to make use of the tax event of 1975:2. Suppose that the
large tax cut had been anticipated by, say, one quarter; then if we
add one lead value of the 1975:2 dummy in our specification, it
will pick up the effect, if any, of the anticipated tax cut on output.
(Note that by including the first lead of the tax dummy variable,
we implicitly have to make the same assumption that we make
explicitly in our first approach to anticipated fiscal policy:
namely, that the tax cut was not implemented in response to
either output shocks in either 1975:2 or 1975:1; otherwise, our
coefficients on the dummy variable would be biased.) The coeffi-
cients on lead values of the tax dummy simply pick up the resid-
uals for GDP (in the benchmark specification) in the quarters
before the change. For both 1974:4 and 1975:1, the two residuals
are small and negative, leading to estimated negative effects, at
one and two leads, of the 1975:2 tax cut on GDP in 1974:4 and
1975:1. Thus, at least under the prior that anticipated tax cuts
should lead to either zero or positive effects on output, there is no
evidence of anticipation effects of fiscal policy in that particular
episode.
We draw two conclusions from this section. First, identifying
and tracing the effects of anticipated fiscal shocks can be done
within a VAR framework but requires stronger identification
restrictions. These restrictions may be too strong. Second, under
those identification restrictions, we find that allowing for antici-
pations of fiscal policy does not alter substantially the results we
have obtained so far, more so in the case of taxes. Although we do
not see the evidence as very strong one way or the other, our
impulse responses suggest only weak effects of anticipated tax
changes on output, a result consistent, for example, with the
findings by Poterba [1988] for the Reagan tax cuts.
IX. ADDING THE FIFTIES
So far, we have used only the post-1960 sample. The reason,

we argued, was that the 1950s, with their large spending and tax
shocks, do not appear to be generated by the same stochastic
process as the post-1960 data. Still, there are a number of ways in
which the 1950s can be used to provide information on the effects
of fiscal policy. We have explored two of them.
A. Using Dummies for the Major Tax and Spending Shocks

The first approach parallels our treatment of the 1975 tax
cut. We extend the sample to start in 1949:1, use dummies for the
major tax and spending episodes, and then trace the effects of
normal fiscal shocks in the VAR estimated using the extended
sample, as well as the effects of the dummies.
The last approach, however, runs into an identification prob-
lem which we did not face in the post-1960 sample. If we allow the
dummy associated with each major tax and spending shock to
have its own distributed lag effect, we cannot identify the effects
of each major spending or tax shock: they are too close in time to
each other. When looking at output, for example, in 1950:3, we
cannot distinguish between the effect of the 1950:1 tax cut after
two quarters, the effect of the 1950:2 tax increase after one
quarter, or the contemporaneous effect of high spending in
1950:3. Nevertheless, something-admittedly more informal-
can be learned by estimating and examining the deviations of
taxes, spending and output from normal in the early 1950s.
To do so, we estimate impulse responses using the sample
starting in 1949:1, but allowing for thirteen dummies from 1950:1
to 1953:1 (equivalently nine dummies, each with four lags; this is
another way of stating the identification problem we just dis-
cussed). Under both DT and ST, the tax and spending output
multipliers associated with e' and e' are similar to those of Tables
III and IV, respectively. The shapes of the impulse responses (not
reported) are also very similar to those from the post-1960 sam-
ple. The main difference is that the output response to a spending
shock under ST is much less persistent, stabilizing after about eight
quarters at around 0.20, instead of 0.65 in the post-1960 sample.
Turning to the effects of the large fiscal shocks of the early
1950s, Figure VIII gives the deviations of taxes and spending
from normal, starting in 1950:1, and the associated deviation of
output, as captured by the coefficients on the thirteen dummies
and their effects through the VAR dynamics.
The figure shows an initial fall in taxes, followed by a strong
increase, which after the first two quarters, largely mimics the
path of GDP. This captures the large tax cut of 1950:1, and the
larger tax increases of 1950:2 and 1950:3.
Spending initially deviates little from normal and then it
increases over time. This captures the large buildup in defense
reap. of ape, DT reap. of tax, DT reap. of gdp
20 20 20 ,
\15
5 15 15 /
10 10 10
5 s -
0
'I 0
,o
07
1
-5
3 5 7 9 11 13 15
i/ ( -
-5
17 19 1 3 5
reap. of ape, ST reap. of tax, ST reap. of gdp

7 9
-5
11 13 15 17 19 13 5
20 20 20
Is 15 - 15 /
10 10 - -- - - - 10 I
5 5 - 5--
II
- i /
0 : _-? I 0, 0
1 3 5 7 9 11 13 15 17. 19 1 3 5 7 9 11 13 15 11 19 1 3 5 7 9
FIGURE VIII
Response to a Shock to the 1950:1 Dum
spending associated with the Korean War, starting in 1951:1, and

continuing until 1952:2.
We can then think of the path of GDP as the response of GDP
to this complex path of taxes and spending. The response of GDP
shows an initial increase-presumably to the initial tax cut-
followed by further increases, presumably in response to the
Korean War buildup. The effects of spending and taxes appear
larger than in the post-1960 sample. In the DT case, the maxi-
mum increase in output is more than 2.5 times the maximum
increase in spending (and this despite the fact that the increase in
spending comes with a nearly equal increase in taxes). In the ST
case, this ratio is smaller, but still equal to 1.5.
B. Defense versus Nondefense Spending

Another way of approaching the difference between the
1950s and the rest of this sample is to note that the 1950s were
dominated by shocks to defense spending, while the rest of the
sample is dominated by shocks to nondefense spending. Hence, by
dividing government spending between its defense and nonde-
fense components, one can hope to capture explicitly the different
persistence of the spending shocks in the early 1950s and in the
rest of the sample.
We thus estimate impulse responses from a four-variable
VAR, with defense and nondefense government spending on
goods and services replacing aggregate government spending on
goods and services. The system also includes the 1975:2 dummy
variable and its first four lags.
The contemporaneous correlation between the two shocks to
defense and nondefense spending is small (0.18) so the order of
the first two variables makes virtually no difference to the re-
sults. We discuss the results when the variable whose effect is
studied is ordered first.
Under DT, we obtain large and similar multipliers for de-
fense and nondefense spending-2.50 and 2.67, respectively-
with a bell-shaped output response in both cases. This is despite
the fact that the persistence of the two spending shocks is indeed
very different: defense spending keeps increasing considerably
for one year after the initial shock, while nondefense spending
decreases after the shock.
Under ST, the responses to a defense spending shock display
a similar picture, although the output response is smaller, and
the spending response stronger and more persistent. But the
output response to a nondefense shock is now small and insignif-

icant, despite the fact that nondefense own response is stronger
and more persistent. A potential explanation is in the estimated
defense response, which becomes negative very soon and exceeds,
in absolute value, the response of nondefense spending. We do not
have an explanation for this lack of robustness of the impulse
responses to nondefense shocks across the DT and ST specifications.
X. RELATION TO OTHER STUDIES
We can now compare our methodology (the combination of a

structural VAR and an event-study approach) and results to
other recent empirical papers on fiscal policy.
The event-study approach has been used by Ramey and Sha-
piro [1997], Edelberg, Eichenbaum, and Fisher [1999], and Burn-
side, Eichenbaum, and Fisher [2000]. All three studies adopt a
"narrative approach" exploiting the exogeneity of military build-
ups. They define one dummy variable taking the value of 1 in
1950:3, 1965:1, 1980:1, and trace out its effects on a number of
macroeconomic variables, including GDP (hours in the case of
Burnside, Eichenbaum, and Fisher).
All three papers find a roughly coincidental increase of de-
fense spending and of GDP (business hours in the case of Edel-
berg, Eichenbaum, and Fisher [1999]). Based on the graphs in
Edelberg, Eichenbaum, and Fisher, defense spending peaks after
two years at about 2.5 percentage points of GDP, and slowly
declines thereafter. GDP peaks after four quarters at about 3.5
percent, and then goes back to trend within two or three years.
A nonexhaustive list of studies following a VAR or a struc-
tural VAR approach includes Barro [1981] and Ahmed and Rog-
ers [1995] (who both look at the effect of expenditures, ignoring
taxes), Blanchard and Watson [1986], and more recently Rotem-
berg and Woodford [1992] and Fatas and Mihov [1998].20
Fatas and Mihov's [1998] specification is closest to our bench-
mark specification, but includes only one fiscal variable, the ratio
of the primary deficit to GDP. Identification is achieved by as-
suming that the "sluggish" private sector variables, output and
20. Results from twelve large macroeconometric models are reported in Bry-
ant [1988]; under the assumption of an unchanged path for money, the average
effect of a permanent government spending increase of one percentage point of
GDP is an increase of 1.27 percent of GDP in the first year, declining slowly to 0.65
percent after five years.
prices, do not respond to changes in policy within a quarter; this,

in a sense, is the exact opposite of our assumption. An increase in
the primary deficit by one percentage point of GDP leads to an
increase in GDP by about one percentage point after about two
years, while the primary deficit goes back to trend very quickly.
Their implied "deficit" multiplier is similar to our spending mul-
tiplier over the same 1960-1997 period, but less persistent.
Rotemberg and Woodford's [1992] specification is closest to
our defense/nondefense decomposition. They trace the effects of
military spending and military employment shocks on output by
Choleski-decomposing a four-variable VAR in personnel military
expenditure, military purchases, output, and the real wage. They
do not control for other spending or for taxes. Their sample covers
1947 to 1989. The estimated impact elasticity of private GDP to
military purchases (when ordered first) is about 0.1, which im-
plies an impact multiplier above 1.0 (the average ratio of military
purchases to GDP has been below 10 percent after World War II),
and a bit larger if one considers total GDP, as we do. This effect
persists for about four quarters, and dies out completely after
eight quarters. Thus, they find a smaller defense multiplier than
we do.
There is, however, a dimension in which some recent papers
have gone further than we have so far, namely in looking at the
effects of fiscal policy not only on output but on a large number of
macroeconomic variables. Our methodology makes it difficult to
go very far without running out of degrees of freedom but in the
next section we take a step in that direction.
XI. EFFECTS ON OUTPUT COMPONENTS
We now go back to the post-1960 sample, and decompose the

output effects of tax and spending shocks into their effects on
each component of GDP. This exercise is interesting in itself, and
also because it is a first step in sorting out the relative merits of
alternative theories. For instance, both standard neoclassical and
Keynesian models imply a positive effect of government spending
on output. However, neoclassical models typically predict a nega-
tive effect on private consumption (see, e.g., Baxter and King
[1993]), while Keynesian models predict the opposite sign. The
implication of these models for private investment is more am-
biguous. In the neoclassical model, a shock to government spend-
ing can raise private investment if the shock is sufficiently per-
sistent and taxes are sufficiently nondistortionary; investment

can fall in the opposite case. In a Keynesian model, investment
increases if the accelerator effect prevails, falls if the effect of a
higher interest rate prevails.
We estimate a four-variable VAR, with the component of
GDP whose response we are studying ordered last. The relation
between residuals now becomes
tt = alxt + a2eg + et
(10) gt = blx, -+ b2et + eg

Xt C= ltt + c2gt + et
xi,t = dt + d2g + e',
where
generalxi,t indicates a component of GDP, so ex and et' are in
be correlated.
A. Response to a Tax Shock

Table VI displays a summary of the impulse response of
the various GDP components to a net tax shock, under DT
and ST. The impulse responses of aggregate GDP and govern-
ment spending change (slightly) with each component added
to the VAR, raising the issue of which one to report. The first line
in each case gives the response of GDP and government spending
from the three-variable model (see Table IV). The last line in each
case displays the unconstrained sum of the responses of the
individual components of GDP, which is denoted by "SUM."
Under DT, an increase in taxes reduces all the private com-
ponents of GDP. After a small decline on impact, private con-
sumption falls by a maximum of 0.35 (35 cents) after five quar-
ters; investment falls by 0.36 on impact and then increases to
become marginally positive after two years. The negative effect
on imports and exports is very small. The sum of the responses of
the components of GDP is close to the response of GDP in the
three-variable system.
A similar picture emerges under ST, with very similar peak
negative responses of private consumption and investment, and
again small responses of imports and exports. However, the dif-
ference between the GDP response in the three-variable model
and the sum of the responses of its components is more
pronounced.
TABLE VI
RESPONSES OF GDP COMPONENTS
DT, TAX
GDP -0.69* -0.74* -0.72* --0.42* -0.22 -0.78* (5)

GCN -0.05* -0.12* -0.24* -0.26* -0.16* -0.05* (1)
CON -0.18* -0.35* -0.32* -0.23* -0.20* -0.35* (5)
INV -0.36* -0.00 -0.00 0.18* 0.16* -0.36* (1)
EXP -0.02 0.01 -0.01 0.02 0.05 -0.08 (3)
IMP -0.01 0.02 -0.14* -0.06 0.04 -0.14* (7)
SUM -0.60 -0.48 -0.43 -0.23 -0.18 -0.60 (1)
ST, TAX
GDP -0.70* -1.07* -1.32* - 1.30* -1.29* -1.33*

(7)
GCN -0.06* 0.04* -0.01* -0.00* -0.00* 0.04*
(4)
CON -0.15 -0.40* -0.44* -0.43* -0.43* -0.44*
(7)
INV -0.35* -0.22 -0.30 -0.27 -0.27 -0.35*
(1)
EXP -0.00 -0.01 -0.06 -0.07 -0.07 -0.10 (3)
IMP -0.01 -0.02 -0.12 -0.12 -0.11 -0.13 (3)
SUM -0.55 -0.57 -0.68 -0.66 -0.66 -0.73 (6)
DT, SPE
GDP 0.84* 0.45 0.54 1.13* 0.97* 1.29* (15)

GCN 1.00* 1.14* 0.95* 0.70* 0.42* 1.14* (4)
CON 0.50* 0.63* 0.91* 1.21* 0.90* 1.26* (14)
INV -0.03 -0.75* -0.69* -0.41* -0.35* -1.00* (5)
EXP 0.20* -0.47* -0.76* -0.70* -0.06 -0.80* (9)
IMP 0.64* -0.19* -0.46* -0.42* -0.16* -0.49* (9)
SUM 1.03 0.74 0.86 1.22 1.07 1.39 (15)
ST, SPE
GDP 0.90* 0.55 0.65 0.66 0.66 0.90* (1)

GCN 1.00* 1.30* 1.56* 1.61* 1.61* 1.00 (1)
CON 0.33* 0.34 0.42 0.43 0.44 0.46* (2)
INV 0.02 -0.74* -0.97* -0.96* -0.95* -0.98* (9)
EXP 0.17* -0.16 -0.30 -0.37* -0.37 -0.37* (13)
IMP 0.56* 0.03 -0.06 -0.05 -0.04 -0.08 (9)
SUM 0.95 0.72 0.77 0.76 0.78 0.95 (1)
Sample: 1960:1-1997:4.
B. Response to a Spending Shock

The next two panels in Table VI
increase in government spending
component are considerably larger
Under DT, private consumption increases by 0.50 (50 cents)

on impact, while investment does not move; imports and exports
now react strongly, increasing by 0.64 and 0.20, respectively. The
positive effect on consumption builds up for fourteen quarters,
reaching a peak of 1.26; investment declines for the first five
quarters, with a peak crowding-out effect of - 1.0. After the initial
increase, exports start declining with a maximum negative effect
of 0.80 after nine quarters; similarly, after an initial increase,
imports start declining and reach a negative peak of 0.49 after
nine quarters. All these responses are precisely estimated, and
the unconstrained sum of the responses is close to the impulse
response of GDP in the three-variable system.
Under ST, the basic picture is qualitatively similar, but the
peak responses of consumption, imports, and exports are smaller;
the standard error bands for these components also become
wider. As before, the output response is very close to the sum of
the responses of the individual components.
C. What Do the Results Tell Us about Competing Macro

Theories?
We find that private consumption is consistently crowded out

by taxation, and crowded in by government spending. The latter
result is difficult to reconcile with a neoclassical model, regard-
less of the persistence of the spending shock, but it is consistent
with a Keynesian model.
We find that private investment is consistently crowded out
by both government spending and, to a lesser degree by taxation;
this implies a strong negative effect on private investment of a
balanced-budget fiscal expansion. This time, these effects are
consistent with a neoclassical model, but inconsistent with the
standard Keynesian approach. In this approach, an increase in
spending may increase or decrease investment depending on the
relative strength of the effects of the increase in output and the
increase in the interest rate; but, in either case, increases in
spending and taxes have opposite effects on investment. This is
not the case in our results. Interestingly, using a yearly panel
VAR on eighteen OECD countries over the same period, Alesina,
Perotti, and Schiantarelli [1999] reach the same conclusion on the
effects of fiscal policy on private investment.21
21. Note also the decline in imports (after a brief initial surge), which is
surprising in light of the considerable increase in GDP. This decline may consist
XII. CONCLUSIONS
Our main goal in this paper was to characterize as carefully

as possible the response of output to the tax and spending shocks
in the postwar period in the United States. From the several
specifications we have estimated and the different exercises we
have performed, we reach the following conclusions.
* The first is consistent with standard wisdom: when gov-
ernment spending increases, output increases; when taxes
increase, output falls.
* In most cases the multipliers are small, often close to one.
In the case of spending shocks, the proximate explanation
is in the opposite effects they have on the different compo-
nents of output: while private consumption increases fol-
lowing spending shocks, private investment is crowded out
to a considerable extent. Both exports and imports also fall.
* While the response of consumption is difficult to reconcile
with the neoclassical approach to fiscal policy, the response
of investment, which decreases in response to both in-
creases in taxes and increases in spending, is hard to
reconcile with the Keynesian approach. We believe that
this result deserves further investigation.
APPENDIX
A. 1. The Data
All the data, unless otherwise noted, are from the Nationa
Income and Product Accounts. The Citibase mnemonics are giv
in parentheses. FG stands for Federal Government; SLG stand
for State and Local Governments.
Government spending
Purchases of goods and services, FG (GGFE) + Purchases of
goods and services, SLG (GGSE).
Net taxes
Receipts, FG (GGFR)22 + Receipts, SLG (GGSR)23 - Federal
mostly of a decline in imports of capital goods, and therefore might be a reflection

of the decline in investment. We have not explored this explanation here, but it
clearly deserves further investigation.
22. The Quarterly National Income and Product Accounts report taxes on a
cash basis, except for the Corporate Profits Tax and the Indirect Business Tax,
which are reported on an accrual basis. We use the Quarterly Treasury Bulletin
to obtain data on federal corporate profit taxes on a cash basis.
grants-in-aid (GGAID) - Net transfer payments to persons, FG

(GGFTP) - Net interest paid, FG (GGFINT) - Transfer pay-
ments to persons, SLG (GGST) - Net interest paid, SLG
(GGSINT) + Dividend received by government, SLG (GGSDIV).
Defense spending
Purchases of goods and services, national defense, FG
(GGFEN).
A.2. Elasticities
To construct the aggregate net tax elasticity to GDP, a,, we

consider four categories of taxes: indirect taxes (IND), personal
income taxes (INC), corporate income taxes (BUS), and social secu-
rity taxes (SS). For each category, we construct the elasticity to GDP
(X) as the product of the tax elasticity to its own base, ITL,,, and
the elasticity of the tax base to GDP, IqB,x (see expression (5) in
the text). For each tax, we must take into account the possible
presence of collection lags and of quarter dependence. On the expen-
diture side, we also construct an approximate output elasticity of
total transfers. The rest of this section describes our assumptions.
Indirect taxes
We take the tax base to be GDP. This is an approximation. In
many states, food consumption is excluded. In most states, the
sale of materials to manufacturers, producers, and processors is
also excluded (see Advisory Committee of Intergovernmental Re-
lations [1995]). Hence,
BINDX = 1. O
qIND,BIND = 1.0 (from Giorno et al. [1995])
Collection lags: 0
Quarter dependence: none.
Personal income taxes
We start from the formula for the elasticity to GDP from
Giorno et al. [1995]. Let T = t(W)W(E)E(X), where T is total
revenues from the personal income tax, t is the tax rate, W is the
We do not, however, have this series for the state and local governments, so
we use accruals (the proportion of corporate profit tax receipts collected by state
and local governments is about 5 percent at the beginning of the sample, and
about 20 percent at the end of the sample). We also could not find data on indirect
taxes on a cash basis; however, the difference between receipts and accruals
appears to be very small.
23. Corporate income taxes collected by state and local governments are not
included.
wage (or earnings, in the OECD terminology), E is employment

and X is GDP. Define the tax base as BINC = W * E. Assuming
constant elasticities everywhere, define
elasticity of taxes to earnings: D = d log (tW)/d log W

elasticity of earnings to employment: F = d log W/d log E
elasticity of employment to output: H = dlog E/dlog Y.
Totally differentiating the expression for total tax revenues,
after some steps one obtains
TBINc, = HI(F + 1)
TINCBINc = (FD + 1)/(F + 1).
We obtain values of D from Giorno et al. [1995]. We estimate
F from a regression of the log change of the wage of production
workers on the first lead and lags 0 to 4 of the log change in
employment of production workers. We estimate H from a regres-
sion of the log change of employment of production workers on the
first lead and lags 0 to 4 of the log change in output. The values
of F and H are the estimated coefficients of lag zero of the
dependent variable. We find F = .62 and H = .42.
Collection lags: 0
Quarter dependence: none.
Note that for personal income taxes we assume the same
elasticity for employees and the self-employed: the former have
their taxes withheld at the source, while the latter make quar-
terly payments based on the estimated income of that quarter. So
long as there is no systematic pattern in the end-year adjust-
ments (as it should be if the tax system is well designed), our
assumption does not introduce any substantial bias in our aggre-
gate elasticity.
Social security taxes
We follow exactly the same procedure as for personal income
taxes. The only difference is in the value of the elasticity of taxes
to earnings, D, which we also obtain from Giorno et al. [1995].
Corporate income taxes
Each corporation can have its own fiscal year different from
the tax year. Large corporations are required to make quarterly
installment payments, of at least 0.8 of the final tax liability.
Until 1980, no penalty was applied if the estimated tax liability
was based on the previous year's tax liability; this exception has
been gradually phased out since 1980. Hence,
CBBUS,X: we estimate it as the lag 0 estimated coefficient from

a regression of quarterly changes of corporate profits on the
first lead and lags 0 to 4 of changes in output.
'qBUS,BBUs: Giorno et al. [1995] gives value of 1 for the annual

elasticity. This is the right value of the quarterly elasticity
for tax accruals as well. For tax receipts, however, things are
more complicated. Corporations make partial payments
throughout the year, and make up for the difference in quar-
ter 2. Hence, we estimate the elasticity of receipts to accruals
from a regression of tax receipts on tax accruals, using quar-
ters 1, 3, and 4, but not quarter 2. This gives a coefficient of
0.85 for lqBUS,BBus
Collection lags: yes. This follows from the fact that we use a
value of .85 for -lBUS,BBs,' as explained above. This value
appears stable over time.
Quarter-dependence: yes (for the same reason as above).
Transfers
Among transfers, unemployment benefits certainly have a

large within-quarter elasticity to the cycle. However, unem-
ployment benefits are a small component of transfers: on
average, unemployment expenditures (defined as the sum of
"active" and "passive" measures) represent less than 2 per-
cent of total government expenditures in the United States.
Other categories of transfers might be sensitive to the cycle.
We do not have reliable quarterly data for our sample on the
components of transfers that might be sensitive to the cycle.
Based on OECD estimates, we use a value of -0.2 for the
elasticities of total transfers to GDP.
In Section VIII our identification strategy requires us to

consider the elasticity of net taxes to lagged output, the term
a121. To do so, we assume that there are no nonquarter depen-
dent collection lags, and take into account the lags in the response
of the tax bases (the wage, employment, profit) to GDP, using the
same regressions as before.
MASSACHUSETTS INSTITUTE OF TECHNOLOGY AND NATIONAL BUREAU

OF ECONOMIC RESEARCH
EUROPEAN UNIVERSITY INSTITUTE AND CENTRE FOR ECONOMIC POLICY RESEARCH
REFERENCES
Advisory Committee of Intergovernmental Relations, "Significant Features of

Fiscal Federalism," 1, 1995.
Ahmed, S., and J. Rogers, "Government Budget Deficits and Trade Deficits: Are
Present Value Constraints Satisfied in Long-Term Data?" Journal of Mone-
tary Economics, XXXVI (1995), 351-374.
Alesina, A., S. Ardagna, R. Perotti, and F. Schiantarelli, "Fiscal Policy, Profits,
and Investment," mimeo, Boston College, 1999.
Barro, R., "Output Effects of Government Purchases," Journal of Political Econ-
omy, LXXXIX (1981), 1086-1121.
Baxter, M., and R. King, "Fiscal Policy in General Equilibrium," American Eco-
nomic Review, LXXXIII (1993), 315-339.
Bernanke, B., and I. Mihov, "Measuring Monetary Policy," Quarterly Journal of
Economics, CXIII (1998), 315-334.
Blanchard, 0., and M. Watson, "Are All Business Cycles Alike?" The American
Business Cycle: Continuity and Change, Robert Gordon, ed. (Chicago: NBER
and University of Chicago Press, 1986), pp. 123-156.
Blinder, A., "Temporary Income Taxes and Consumer Spending," Journal of
Political Economy, LXXXIX (1981), 26-53.
Bohn, H., "Budget Balance through Revenue or Spending Adjustments? Some
Historical Evidence for the United States," Journal of Monetary Economics,
XXVII (1991), 333-359.
, "The Behavior of U. S. Public Debt and Deficits," Quarterly Journal of
Economics, CXIII (1998), 949-964.
Bryant, R., Empirical Macroeconomics for Interdependent Economies (Washing-
ton, DC: Brookings Institution, 1988).
Burnside, C., M. Eichenbaum, and J. Fisher, "Assessing the Effects of Fiscal
Shocks," mimeo, Northwestern University, 2000.
Cohen, D., and G. Folette, "The Automatic Fiscal Stabilizers: Quietly Doing Their
Thing," in Fiscal policy in an era of surpluses: Economic and financial impli-
cations, Federal Reserve Bank of New York Economic Policy Review, VI
(2000), 35-68.
Edelberg, W., M. Eichenbaum, and J. Fisher, "Understanding the Effects of
Shocks to Government Purchases," Review of Economic Dynamics, II (1999),
166-206.
Fatas, A., and I. Mihov, "Measuring the Effects of Fiscal Policy," mimeo, INS
1998.
Giorno, C., P. Richardson, D. Roseveare, and P. van der Noord, "Estimating
Potential Output, Output Gaps, and Structural Budget Deficits," Economics
Department Working Paper No. 152, OECD, Paris, 1995.
Gordon, D., and E. Leeper, "The Dynamic Impact of Monetary Policy: An Exercise
in Tentative Identification," Journal of Political Economy, CII (1994),
1128-1147.
Hamilton, J., and M. Flavin, "On the Limitations of Government Borrow
Framework for Empirical Testing," American Economic Review, LX
(1986), 808-819.
Poterba, J., "Are Consumers Forward Looking? Evidence from Fiscal Experi-
ments," American Economic Review, Papers and Proceedings, LXXVIII
(1988), 413-418.
Ramey, V., and M. Shapiro, "Costly Capital Reallocation and the Effects of
Government Spending," Carnegie-Rochester Conference Series on Public Pol-
icy, 1997.
Rotemberg, J., and M. Woodford, "Oligopolistic Pricing and the Effects of Aggre-
gate Demand on Economic Activity," Journal of Political Economy, CX (1992),
1153-1207.

Blanchard Perotti (2002)

Uploaded by

Copyright:

Available Formats

Blanchard Perotti (2002)

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Blanchard Perotti (2002)

Uploaded by

Copyright:

Available Formats

An Empirical Characterization of the Dynamic Effects of Changes in Government

Spending and Taxes on Output

OLIVIER BLANCHARD AND ROBERTO PEROTTI

This paper characterizes the dynamic effects of shocks in government

The predominant, Keynesian, view of the effects of fiscal

In view of these challenges, our purpose in this paper is to

shocks as having a negative effect. The size and persistence of these

II. METHODOLOGICAL ISSUES

Both government expenditure and taxation affect GDP: since

two-variable breakdowns of the budget, consisting of an expenditure

Our basic VAR specification is

(1) Yt = A(L,q)YtI + U,,

(4) x = Cltt + c2gt e,

3. Note that, because equation (1) is a reduced form, quarter dependence in

Direct evidence on the conduct of fiscal policy suggests

' T,,Bi and nSBi,x for four separate categories of taxes.

5. Although the logic of our approach is similar to that of Bernanke and

changes; we do the same with respect to quarterly

tax decisions come first, so that a2 = 0, and we can

III. THE DATA

We define net taxes as the sum of Personal Tax and Nontax

8. As in all small-dimensional VARs, one should worry about omitted vari-

tic process as the post-1960 period. Thus, our strategy is to

clearly dummied. This allows us to compare two different types

12. Recall that T is net of interest payments.

terms, we subtract a changing mean, constructed as the geomet-

sd(A log G) = 0.019 sd(A log T) = 0.049

A log G > 3 sd A log T > 3 sd

IV. CONTEMPORANEOUS EFFECTS

The two panels of Table II report the estimated coefficients of

coeff. -0.868 0.956 -0.047 -0.187

p-value 0.001 0.018 0.255 0.255

coeff. -0.876 0.985 -0.057 -0.238

DT: Deterministic Trend; ST: Stochastic Trend.

movements in cyclically una

V. DYNAMIC EFFECTS OF TAXES

A. Effects of Estimated Tax Shocks

-10o -10 - -10

-15 -15 -15

-20 -20 -20

resp. of tax, ST reap. of spe, ST reap. of gdp

-05 -05 -05

1 qrt 4 qrts 8 qrts 12 qrts 20 qrts peak

GDP -0.69* -0.74* -0.72* -0.42* -0.22 -0.78* (5)

GDP -0.70* -1.07* -1.32* -1.30* -1.29* -1.33* (7)

DT: Deterministic Trend; ST: Stochastic Trend. An asterisk indi

nience, the impulse responses in these an

Thus, under both specifications, tax increases have a nega-

B. Dynamic Effects of the 1975:2 Net Tax Cut

VI. DYNAMIC EFFECTS OF SPENDING SHOCKS

The two panels of Figure V show the effects of a unit spend-

1 qrt 4 qrts 8 qrts 12 qrts 20 qrts peak

GDP 0.84* 0.45 0.54 1.13* 0.97* 1.29* (15)

GDP 0.90* 0.55 0.65 0.66 0.66 0.90* (1)

DT and, then, under ST. As

-05 -05 - -o05

reap. of ape, ST reap. of tax, ST reap. ofgd

.10 .10 I'10 f i l l i l l I w i l l -lo

Net taxes Spending