Debt and the Eects of Fiscal Policy

Carlo Favero and Francesco Giavazzi February 14, 2008

Abstract Equilibrium structural models of scal policy are solved by imposing the government intertemporal budget constraint and are simulated under the equilibrium assumption that the real value of the debt in the hands of the public must equal the expected present value of government surpluses. Empirical models of scal policy typically do not impose this condition and usually do not even include debt. This is particularly surprising in the case of countries where the data reveal that scal variables respond to the level of the debt. In this paper we conduct VAR analysis of US scal policy by explicitly including debt and the stock-ow identity linking debt and decits. We apply our methodology to dierent identication approaches: the structural VAR approach and the narrative approach. Our main ndings are that the absence of an eect of scal shocks on long-term interest ratesa frequent and puzzling result in research based on VARs that omit a debt levelcan be explained by their mis-specication, especially over samples in which the debt to GDP ratio is very high by its historical standards. The explicit inclusion of the debt-decit dynamics does not alter sizeably the eect of scal policy on output in standard structural VARs, more sizeable dierences are obtained when the narrative approach is considered. Keywords: scal policy, public debt, government budget constraint, VAR models JEL Classication: H60, E62
We thank Olivier Blanchard, Eric Leeper and Roberto Perotti for useful comments. Francesco Giavazzi thanks the Federal Reserve Bank of Boston for its hospitality while this paper was completed. Favero: IGIER (Universita Bocconi) and CEPR. Giavazzi, IGIER (Universit Bocconi), MIT, CEPR and NBER.

Introduction

Empirical, VAR based, evidence on the eects of scal policy is designed to serve the purpose of selecting the appropriate structural model to conduct policy analysis. As recently pointed out by Chung and Leeper (2007) there is a potentially very relevant discrepancy between scal Dynamic Stochastic General Equilibrium models and empirical scal VARs. Equilibrium structural models are solved by imposing the government intertemporal budget constraint and are simulated under the equilibrium assumption that the real value of the debt in the hands of the public must equal the expected present value of government surpluses. Empirical models of scal policy typically do not impose this condition and usually do not even include debt. This is particularly surprising in the case of countries where the data reveal a positive correlation between the government surplus-to-GDP ratio and the government debt-to-GDP ratio and thus indicate that scal variables respond to the level of the debt. Bohn (1998) nds such a correlation in a century of U.S. data. In this paper we conduct VAR analysis of US scal policy by explicitly including debt and the stock-ow identity linking debt and decits. We show how impulse response analysis can be conducted in a scal VAR augmented by the stock-ow relation linking debt and decit. We treat the debt-decit relation as an identity that involves non-linear relationships between the variables normally included in a scal VAR but no additional parameters to be estimated and no additional shocks to be identied. As a consequence of these facts, the computation of impulse responses to scal shocks that takes explicitly on account the government budget constraint is accomplished with some simple modications of the standard VAR approach. As the point we make is independent of the particular assumption adopted to identify scal shocksa thorny issue in the scal VAR literature, we apply our methodology to dierent identication approaches: the structural VAR approach (such as in Blanchard and Perotti, 2002 or in Mountford and Uhlig, 2002) and the narrative approach, as in Ramey (2006) or in Romer and Romer (2007). Our main ndings are that the absence of an eect of scal shocks on long-term interest ratesa frequent and puzzling result in research based on VARs that omit a debt levelcan be explained by their mis-specication, especially over samples in which the debt to GDP ratio is very high by its historical standards. The explicit inclusion of the debt-decit dynamics does 1

not alter sizeably the eect of scal policy on output in standard structural VARs, more sizeable dierences are obtained when the narrative approach is considered. The plan of the paper is as follows. In Section 2 we explain why estimating the eects of scal policy shocks omitting the the level of the public debt is problematic and how VAR analysis with the government budget constraint can be conducted. Section 3 describes our data. In Sections 4 and 5 we evaluate the empirical relevance of our point computing impulse responses to scal shocks in models in which the variables are allowed to respond to the level of the debtwhose evolution over time is determined by the intertemporal government budget constraint. In specifying the response of scal variables to the level of the debt we follow Bohn (1998). We then compare these impulse responses with those obtained from models that omit the debt level. In Section 4 we use the identication technique proposed by Blanchard and Perotti (2002),; in Section 5 we use the tax shocks identied by Romer and Romer (2007). The methodology described in this paper to analyze the impact of scal shocks by taking into account the stock-ow relationship between debt and scal variables could be applied to other dynamic models which include similar identities. One example are the recent discussions (see e.g. Christiano et al, 2005 and Chari et al. 2005) on the importance of including capital as a slow-moving variable to capture the relation between productivity shocks and hours worked.

Fiscal policy VARs and the stock-ow constraint

The study of the dynamic response of macroeconomic variables to shifts in scal policy is typically carried out estimating a vector autoregression of the form Yt = C0 +
k X Ci Yti + ut i=1

(1)

where Y includes government spending, taxes, output and other macroeconomic variables such as interest rates, consumption and ination. 2

This specication does not consider the stock-ow relation between debt and decit and the level the level of the debt-to-GDP-ratio is usually omitted from (1). This assumption is potentially very relevant in determining the measured empirical eects of scal policy for a number reasons : a feedback from the level of debt ratio to taxes and government spending could be important in the scal reaction functions if scal authorities attach some weights to debt stabilization. Bohn(1998) shows that such a feedback is statistically signicant and economically important: in his study a century of U.S. data reveal a positive correlation between the government surplus-to-GDP ratio and the government debtto-GDP ratio; interest rates, a central variable in the transmission of scal shocks, depend on future expected monetary policy and on the risk premium: both may be aected by the debt dynamicsfor instance if a growing stock of debt raises fears of future monetization or, in an extreme case, of debt default. The impact of a given scal shock on interest rates will be very dierent depending on whether the shock produces a path of debt that is stable or tends to become explosive1 . eects of the level of debt on ination and output uctuations cannot be ruled out a priori, think of the debate on the validity of the Ricardian equivalence (Barro (1974), Kormendi (1983)) or of the literature on the inationary eects of the debt See Canzoneri et al.(2001). It seems therefore important to allow for the fact that taxes, government spending, output, ination and the rate of interestin other words the variables entering Yare linked by an identity, the equation that determines how the debt ratio evolves over time. These observations naturally lead to replacing (1) with
k X Yt = C0 + Ci Yti + i dt1 + ut i=1
1

(2)

Giavazzi, Jappelli and Pagano (2000) nd that an increase in taxes can raise private consumption if it moves the economy from an unsustainable scal path to a sustainable one. Romer and Romer (2007) also nd that the eect of a U.S. tax shock on output depends on whether the change in taxes is motivated by the governments desire to stabilize the debt, or is unrelated to the stance of scal policy

 0 where Yt = gt tt yt pt it . dt1 is the debt-to-GDP ratio at the beginning of the period, it is the nominal rate of interest (the average cost of debt nancing), yt is real GDP growth (the dierences of the log of GDP), pt is ination (the dierences of the log of the price level), tt and gt are, respectively, (the logs of) government revenues and government expenditure net of interest. Once the debt has been brought into the emprical specication, it becomes immediately clear that (2) omits an identity that links, dynamically and nonlinearly, all variables included in the VAR. This identity is the government budget constraint: dt = 1 + it exp (gt ) exp (tt ) dt1 + (1 + pt ) (1 + yt ) exp (yt ) (3)

The omission of this relation from the empirical VAR does not allow to consider the endogeneity status of dt , and leads to compute impulse response functions with respect ot the shock of interest under the untenable assumption of a given level of the debt to GDP ratio. Interestingly (3) does not contain any parameter to be identied and estimated or any structural shock. The special nature of this relation poses an interesting (and solvable) problem for the computation of impulse responses that describe the eect of scal shocks. gt tt yt pt it dt Our objective is to compute the response of to scal shocks. To this end we specify the following model: Yt
k X = C0 + Ci Yti + i dt1 + ut i=1

(4)

dt = Aut

1 + it exp (gt ) exp (tt ) dt1 + (1 + pt ) (1 + yt ) exp (yt ) = B et

where ut are the VAR innovations and et are the structural shocks of interest. (4) diers from traditional scal VARs in two aspects. First, all variables in the VAR are allowed to respond to the level of the debt. This is simply a multivariate extension of the univariate specication of the scal reaction function in Bohn (1988). Second, we endogeneize the debt, and we do so via an intertemporal relation which does not contain an error, since our 4

VAR contains all the variables that are relevant to determine the debt-decit dynamics. Note that the identication of structural shocks is unaected by the inclusion of d in the model. Since we treat the debt-decit relationship as an identity, the number of shocks remains the same, so that the identication assumptions used in traditional scal VAR are immediately applicable. Also, since there are no parameters to be estimated in the debt-dynamics equation, (4) can be estimated excluding the stock-ow identity. After all the parameters of interest have been identied and estimated, impulse responses comparable to those obtained from the traditional moving average representation of a VAR can be constructed going through the following steps: generate a baseline simulation for all variables by solving (4) dynamically forward (this requires setting to zero all shocks for a number of periods equal to the horizon up to which impulse responses are needed), generate an alternative simulation for all variables by setting to one just for the rst period of the simulationthe structural shock of interest, and then solve dynamically forward the model up to the same horizon used in the baseline simulation, compute impulse responses to the structural shocks as the dierence between the simulated values in the two steps above. (Note that these steps, if applied to a standard VAR, would produce standard impulse responses. In our case they produce impulse responses that allow for both the feedback from dti to Yt and for the endogeneity of dt modelled via (3), compute condence intervals via bootstrap methods.2 Two comments are in order before moving to the data.
Bootstrapping requires saving the residuals from the estimated VAR and then iterating the following steps: a) re-sample from the saved residuals and generate a set of observation for Yt and dt , b) estimate the VAR and identify strucutral shocks, c) compute impulse responses going thorough the steps described in the text, d) go back to step 1. By going thorugh 1,000 iterations we produce bootstrapped distributions for impulse responses and compute condence intervals.
2

1) What are the sources of potential misspecication of traditional VARs if the data are generated by (4)? In traditional scal VAR Y already contains all the variables that enter the government intertemporal budget constraint but it is unlikely that the short lags of g, t, p, y and i that enter (linearly) traditional scal VARs trace the evolution of the debt ratio accurately enough. In fact the stock-ow relation implies that dt is the result of a long and non-linear lag dynamics i K K X exp (gti ) exp (tti ) Y +
i=0 K Y i=o

dt =

1 + iti dtK 1 (1 + pti ) (1 + yti )

exp (yti )

i=o

1 + iti (1 + pti ) (1 + yti )

2) What about present value conditions? Chung and Leeper (2007) have recently shown that that the present value condition that the real value of the debt must equal the expected present-value of surpluses generates a set of cross-equation restrictions on traditional scals VARs. By augmenting traditional scals VARs with the debt decit dynamics we follow a dierent route. In fact our impulse responses satisfy by construction period-by-period the debt-decit stock-ow relationship. We can therefore evaluate the validity of the tranversality condition by considering the long-run response of dt to scal shocks that we derive explicitly and by checking if it converges to zero. In the next sections we illustrate the empirical relevance of our point on U.S. data by considering two dierent ways of identifying scal shocks, that are representatives of alternative paths researchers have followed: the technique proposed by Blanchard and Perotti (2002) and the "exogenous" tax shocks identied by Romer and Romer (2007) with a narrative approach.

The data

We use quarterly data for the U.S. economy since 1960:1, the sample analyzed in Blanchard and Perotti (2002) and extended to 2005:4 in Perotti (2007). Our approach requires that the debt-dynamics equation in (??) tracks the

path of dt accurately: we thus need to dene the variables in this equation with some care. The source for the dierent components of the budget decit and for all macroeconomic variables are the NIPA accounts (available on the Bureau of Economic Analysis website, downloaded on December 7th 2006). yt is (the log of) real GDP per capita, pt is the log dierence of the GDP deator. Data for the stock of U.S. public debt and for population are from the FRED database (available on the Federal Reserve of St.Louis website,also downloaded on December 7th 2006). Our measure for gt is (the log of) real per capita primary government expenditure: nominal expenditure is obtained subtracting from total Federal Government Current Expenditure (line 39, NIPA Table 3.2 ) net interest payments at annual rates (obtained as the dierence between line 28 and line 13 on the same table). Real per capita expenditure is then obtained by dividing the nominal variable by population times the GDP chain deator. Our measure for tt is (the log of) real per capita government receipts at annual rates (the nominal variable is reported on line 36 of the same NIPA Table). The average cost servicing the debt, it , is obtained by dividing net interest payments by the federal government debt held by the public (FYGFDPUN in the Fred database) at time t 1. The federal government debt held by the public is smaller than the gross federal debt, which is the broadest denition of the U.S. public debt. However, not all gross debt represents past borrowing in the credit markets since a portion of the gross federal debt is held by trust fundsprimarily the Social Security Trust Fund, but also other funds: the Trust Fund for Unemployment Insurance, the Highway Trust Fund, the pension fund of federal employees, etc.. The assets held by these funds consist of non-marketable debt.3 We thus exclude it from our denition of federal public debt. Figure 1 reports, starting in 1970:1 (the rst quarter for which the debt data are available in FRED), this measure of the debt held by the public as a fraction of GDP (this is the dotted line). We have checked the accuracy of the debt dynamics equation in (??) simulating it forward from 1970:1 (this is the continuous line in Figure 1). The simulated series is virtually superimposed to the actual one: the small dierences are due to the presence of siegnorage (that we ignore and it is clearly very little relevant empirically), to
3 Cashell (2006) notes that "this debt exists only as a book-keeping entry, and does not reect past borrowing in credit markets."

approximation errors in computing ination and growth rates as logarithmic dierences, and to the fact that the simulated series are obtained by using seasonally adjusted measures of expenditures and revenues. Based on this evidence we have used the debt dynamics equation to extend dt back to 1950:1. (A quarterly series for dt extending back to 1950:1 will become necessary when we compare our results with those in Romer and Romer (2007) whose sample starts just after World War II.) Figure 1 shows that this series tracks the annual debt level accurately, at least up to the early 1950s. 4

Fiscal shocks identied from SVARs

We start by estimating on our data a traditional scal Structural VAR (SVAR) as in Blanchard and Perotti (2002) and extended in Perotti (2007) (B&P in what follows). We consider the following VAR:
k X = C0 + Ci Yti + ut i=1

Yt

(5)

0 gt tt yt pt it Yt = Aut = B et Following B&P and optimal lag-length choice criteria the length of the lag polynomilal in the VAR is set to four quarters. We apply B&P methodology also to impose restrictions that allow the two structural scal shocks in (5) to be recovered from the reduced form residuals, u. The innovations t in the reduced form equations for taxes and government spending, ug t and ut , contain three terms: (i) the response of taxes and government spending to uctuations in macroeconomic variables, such as output and ination, that is implied by the presence of automatic stabilizers; (ii) the discretionary response of scal policy to news in macro variables and (iii) truly exogenous shifts in taxes and spending, the shocks we wish to identify. B&P exploit the
We are unable to build the debt series back to 1947:1, the start of the Romer and Romer sample, because data for total governemnt spending, needed to buld the debt series, are available on a consistent basis only from 1950:1.
4

fact that it typically takes longer than a quarter for discretionary scal policy to respond to news in macroeconomic variables: at quarterly frequency the contemporaneous discretionary response of scal policy to macroeconomic data can thus be assumed to be zero. To identify the component of ug t and ut which corresponds to automatic stabilizers they use institutional informat tion on the elasticities of tax revenues and government spending to macroeconomic variables. They thus identify the structural shocks to g and t by imposing on the A and B matrices in Au = Be the following structure 5 : 1 0 0 1 a31 a32 a41 a42 a51 a52 agy agp agi aty atp ati 1 0 0 a43 1 0 a53 a54 1 ug t ut t uy t pt u t ui t b11 0 0 0 0 b21 b22 0 0 0 0 0 b33 0 0 0 0 0 b44 0 0 0 0 0 b55 eg t et t e1 t e2 t e3 t

where ei t (i = 1, 2, 3) are non-scal shocks and have no structural interpretation. Since agy , agp , agi , aty , atp and ati are identied using external information 6 , there are only 15 parameters to be estimated. As there are also 15 dierent elements in the variance-covariance matrix of the 5-equation VAR innovations, the model is just identied. The ei t (i = 1, 2, 3) are derived by imposing a recursive scheme on the bottom three rows of A and B; however, the identication of the two scal shocksthe only ones that we shall use to compute impulse responsesis independent of this assumption. Finally, the identication assumption imposes b12 = 0. 7
5 Mountford and Uhlig (2002) identify government spending and revenue shocks by imposing restrictions on the sign of impulse responses. Fatas and Mihov (2001) rely on a Choleski ordering. 6 The elasticities of taxes and government spending with respect to output, ination and interest rates used in the identication have been updated in Perotti (2007) and are

Elasticities of government revenues agy agp agi Entire sample 0 -0.5 0 1960:1-1979:4 0 -0.5 0 1980:1-2006:2 0 -0.5 0
7

and expenditures aty atp ati 1.85 1.25 0 1.75 1.09 0 1.97 1.40 0

B&P provide robustness checks for this assumption by setting b21 = 0 and estimating b12 . We have also experimented with this alternative option. In practice, as the top left corner of the B matrix is not statistically dierent from a diagonal matrix, the assumption b12 = 0 is irrelevant to determine the shape of impulse response functions.

Although we use the same identifying assumptions, our choice of variables diers slightly from those used in B&P, because, as discussed above, we need to use variables that allow the debt dynamics equation to track the path of dt accurately. In particular, our measure of i is the average cost of debt nancing rather than the yield to maturity on long-term government bonds used in B&P. Our denitions of g and t are also slightly dierent: we follow the NIPA denitions by considering net transfers as part of government expenditure, rather than subtracting them from taxes. To check that our slight dierences in data denitions do not change the results we have rst estimated (1) as in B&P. Following Perotti (2007) who nds dierences in the impulse response functions before and after 1980, the sample is split in two sub-samples 1960:1-1979:4 and 1980:1-2006:2. The impulse responses are reported in Figures 2.1 and 2.2 are consistent with those reported in B&P. In particular: an (exogenous) increase in public expenditure has an expansionary effect on output, while an (exogenous) increase in revenues is contractionary. The impact of scal policy weakens in the second sub-sample, in particular the eects of tax shocks become insignicant; after 1980 scal shocks become less persistent; the eect of scal shocks on interest rates is insignicant in the rst subsample; it is small, signicant but counterintuitive in the second subsample when an increase in public spending lowers the cost of servicing the debt; scal shocks have consistently no signicant eect on ination. These results, however, should be interpreted with some caution. Figure 3 reports the responses of the debt/GDP ratio to scal shocks, obtained just adding to (1) the intertemporal budget constraint. The gure shows that temporary shocks to expenditure and revenue have a permanent eect of the debt/GDP ratio, implying, in these cases, an unstable debt dynamics. This feature is particularly prominent in the second subsample, where, as shown by Figure 4, the U.S. economy moved from a situation where GDP growth exceeded to cost of nancing the debt to a situation where the converse has been true. 10

A permanent eect of temporary scal shocks on the debt ratio is a very unwelcome feature for the empirical model. First, because impulse responses will be computed along debt paths that are very unlikely. Second, because these impulse responses cannot be used to evaluate DSGE models, since these models (as pointed out by Chung and Leeper, 2007) typically include the government budget constraint and are simulated under the equilibrium assumption that the real value of debt in the hands of the public must equal the expected present-value of surpluses. It is obviously impossible to compare the empirical evidence from a model that delivers an explosive path for the debt, with the paths of variables generated by forward looking models, since such models do not have a solution when the debt dynamics is unstable.

4.1

Estimating the eects of scal shocks in a SVAR with debt dynamics

We now consider the structural VAR that includes a debt feedback ansd the government intertemporal budget constraint Yt
k X = Ci Yti + i (dt1 d ) + A1 Bet i=1

(6)

dt =

1 + it exp (gt ) exp (tt ) dt1 + (1 + pt ) (1 + yt ) exp (yt )

The specication of (6) is suggested by the empirical ndings in Bohn (1998), who estimates a scal reaction function in which d is the the unconditional mean of the debt ratio over a U.S. sample running 1916 to 1995 (d = .35). We also identify d as the sample average of the debt ratio (which happens to be equal to .35 also in our chosen sample).We model the target level of the debt as a constant on the basis of the evidence of stationarity of d, pointed out by Bohn(1988) and conrmed in our sample. Given that all relations in the VAR contain a constant, our results are robust to alternative choices of d . The lag-length of (6) is kept to four quarters as in (1) Table 1 reports the estimated coecients on (dt1 d ) in the ve equations (taxes, spending, output, ination and the cost of debt service) in the two sub-samples. [INSERT TABLE 1 HERE] 11

The estimated coecients show a signicant stabilizing eect of the debt level on the primary surplus. In the rst sub-sample (1960:1-1979:4) this stabilizing eect works through the response of taxes to deviations of the debt from the target level; in the second sub-sample (1980:1- 2006:2) it works through the response of expenditures. The cost of nancing the debt also responds to (dt1 d ): interestingly, such response is signicant only in the rst sub-sample when the level of the debt is further away from its sample mean, our proxy for d . The nding that the coecients on (dt1 d ) are signicant suggests that the correct VAR to be used in analyses of scal policy is one in which scal variables respond to the level of the debt, or its deviation from a target level. In other words, the VARs normally used to study the eects of scal policy shocks omit a slow moving variable. We shall assess the empirical relevance of this omission by comparing the impulse responses of a traditional VAR with those generated by (6). The eects of scal shocks in a model with feedbacks Figures 5.1 and 5.2 compare the impulse responses obtained from (6) (reported with a dotted line) with those obtained in a SVAR without a debt feedback (reported with a continuous line). In both cases we use the same identifying assumptions. Figure 5.1 refers to the rst sub-sample, 1960:11979:4; Figure 5.2 to the later period. In each gure the left-hand panels refer to a one percent shock to g; the right-hand side panels refer to an equivalent shock to t. In each column the graphs show, from top to bottom, the impulse response of g , t, y , ination, and the average cost of debt service. The reported condence bounds are for the impulse responses with a debt feedback. Pre-1980: following a positive shock to g, allowing for a debt feedback results in a larger response of interest rates and ination (outside the 95% condence bounds). For interest rates the divergence widens over time, as debt accumulates, following a positive shock to t, interest rates fall more in the model with feedback and the dierence (between the two impulse responses) widens over time, 12

 the output eect of a shock to g is larger in the model with a debt feedback. Post-1980: The eect of g shocks on output is dampened in a model with feedback; t shocks have no signicant eect on output, g and t shocks are less persistent in the model with feedback as the stabilization motive acts to dampen the eect of the initial shock, the response of interest rates to a positive g shock is still negative at the beginning, but rises over time in the presence of a feedback Figure 6 compares the responses of the debt level to scal shocks in a traditional scal VAR and in a model with feedback. Including the feedback eliminates the possibility that a temporary scal shock might permanently aect d.. Table 2 complements the result in Figures 5.1 and 5.2 by computing the cumulative response of interest rates and output to a scal shock over three horizons, (4, 12 and 20 quarters) and comparing them with the responses in the absence of a debt feedback. In the rst sub-sample the eect of a 1% g shock on interest rates, cumulated over 20 quarter is 0.10 in the model with feedback, 0.04 without. This dierence however disappears after 1980. This result suggests that the response of interest rates to the debt level is stronger the further the debt is from its sample mean, our proxy for d . The expansionary eect of an increase in expenditure is less pronounced in the model with feedback in both subsamples, as the reaction of g to the level of the debt counteracts the initial shock. [INSERT TABLE 2 HERE]

13

Fiscal shocks identied from the narrative record

Romer and Romer (2007) (R&R in what follows) use the U.S. narrative recordpresidential speeches, executive-branch documents, and Congressional reportsto classify the size (dened as the estimated revenue eect of a new tax bill), timing, and principal motivation for all major postwar tax policy actions.8 They then identify, among all documented tax actions, those that could be classied as "exogenous", as opposed to those that were countercyclical, i.e. motivated by a desire to return output growth to normal. Exogenous tax changes are further divided into two groups: those that appear to be motivated by a desire to raise the potential growth rate of the economy, and those aimed at reducing a budget decit inherited from previous administrations. Since 1947 U.S. Federal laws changed taxes in 82 quarters. A number of these quarters had tax changes of multiple types. Among the 104 separate quarterly tax changes identied, 65 are classied as exogenous. In this Section we use these 65 tax changes (the R&R exogenous tax shocks) and ask what dierence it makes if the debt channel is, or is not, included in the transmission mechanism. R&R estimate the impact of tax shocks on output using a single-equation approach:
12 k X X Ttex 1 yt = 0 + i + j Ztj + et Yt1 i=1 j =1 T ex

(7)

t1 where yt is real quarterly output growth, Yt are the tax shocks, 1 measured as a percent of nominal GDP, and Ztj are controls (lags of yt , monetary policy shocks, government spending, oil prices). The Z 0 s are assumed to be exogenous, and in particular unaected by the tax shocks, not even with a lag. The R&R exercise should thus be interpreted as asking the following (hypothetical) question: Assume that the transmission mechanism

Early attempts at applying to scal policy the methodology proposed by Romer and Romer (1989) to identify monetary policy shocks were Edelberg, Eichenbaum and Fisher (1999), Burnside, Eichenbaum and Fisher (2004), Ramey (2006). These papers used a dummy variable which identies characterizes episodes of signicant and exogenous increases in government spending (typically wars).

14

of tax shocks is shut down and that such shocks only aect output directly, rather than, for instance, also via their eect on interest rates. What is their eect on output under this hypothesis? R&R nd that "exogenous" tax increases have a larger negative eect on output than countercyclical tax hikes. Among the exogenous tax increases, those motivated by the aim to rein in a budget decit are less contractionary.in fact the negative impact on output is statistically insignicant. To estimate the eects of the R&R tax shocks when scal policy is allowed to respond to the level of the debt we rst need to embed these shocks in a model that doesnt shut down the transmission mechanism. We do this using the R&R shocks in the two VARs analyzed above: (1) and (??).9 Therefore, we estimate the following two models:
k X Ttex Yt = Ci Yti + i + ut Tt i=1 k k X X Ttex = Ci Yti + i dti + i + ut Tt i=1 i=1

(8)

Yt

(9)

dt =

1 + it exp (gt ) exp (tt ) dt1 + (1 + pt ) (1 + yt ) exp (yt )

where the variables in Y are, as before, taxes, government spending, output, ination and interest rates. Including the R&R tax shocks in a VAR is a natural way of computing the dynamic response of macro variables to shocks identied outside the VAR because what matters are the impulse responses generated by the dierent shocks, not the correlation of the shocks themselves.10 The R&R shocks are valid shocks to taxes because we nd that they are uncorrelated with all lags of the variables included in the VARs and are signicant only in the equation
9 R&R scale their shocks by the level of GDP. We scale them by taxes to allow direct comparability of the eects of these shocks with those identied in a SVAR. In a SVAR tax shocks are extracted from a specication in the logarithms of the levels of real variables. Innovations thus have the dimension of a percentage change in taxes. A one per cent change in taxes is much smaller than a one per cent shock in the tax-to-GDP ratio. The re-scaling aects the size of the eects but not the shape of the impulse responses. 10 VARs have been used to compute impulse responses to shocks identied outside the VAR in the analysis of the eects of monetary shocks in Bagliano and Favero (1999).

15

for t. Thus they satisfy the properties that exogenous shocks identied in a structural VAR should fulll. Figure 6 shows the impulse response of output to an exogenous R&R tax shock equivalent to 1% of taxes. Impulse responses are computed using three dierent models: (7), the equation estimated by R&R where we have replaced
ex Tt 1 ex Tt 1 Yt1

with

Tt1

(8), a VAR that excludes a debt feedback (9), a model that allows the variables in the VAR to respond to the level of the debt. The R&R shocks start in 1947, while our data, for the reasons noted in footnote 2, only start in 1950:1: we thus miss the exogenous shocks that occurred between January 1947 and December 1949. As in the previous Section we split the sample in two parts: 1950:1-1979:4 and 1980:1-2006:2. The eects on output of the exogenous R&R tax shocks are quite dierent in the two sub-samples and depending on the model they are embedded in. In the rst sub-sample (1950:1-1979:4) the contractionary eect of a tax hike is larger when Z is endogeneized in a model that includes the level of the debt and the government intertemporal budget constraint. This probably happens becauseas documented in the previous Section (i) a single-equation partial equilibrium approach cannot take into account the persistent eect on taxation of the initial shock and (ii) the fact that scal shocks could cumulate over time amplifying the eect on output of an initial impulse. In the second sub-sample, when the burden of debt stabilization falls on expenditure -an exogenous increase in taxes is compensated by a subsequent expenditure accommodation. This explains why, analyzing the eects of shocks in a model where Z is endogenous and scal policy responds to the debt level, produces much smaller output eects compared with the R&R single equation model. Figure 7 shows that in fact, in the second sub-sample, an initial positive tax shock is accompanied by further tax changes in the opposite direction. Following the initial shock taxes fall: when this happens the eect on the budget is compensated by increases in spending. These responses are not captured in (7) because the equation sets to zero the dynamic response of all variables, with the only exception of output growth, to tax shocks. 16

Conclusions

Motivated by the empirical ndings in Bohn (1998), we have analyzed the eects of scal shocks allowing for a direct response of taxes, government spending and the cost of debt service to the level of the public debt (as a ratio to GDP). We have shown that omitting such a feedback can result in incorrect estimates of the dynamic eects of scal shocks. We suggested in particular that the absence of an eect of scal shocks on long-term interest ratesa frequent nding in research based on VARs that omit a debt feedback and do not endogeneize debt dynamicscan be explained by their mis-specication, especially over samples in which the debt to GDP ratio is very high by its historical standards. The methodology described in this paper to analyze the impact of scal shocks by taking into account the stock-ow relationship between debt and scal variables could be extended to other dynamic models which include similar identities. For instance, the recent discussions on the importance of including capital as a slow-moving variable to capture the relation between productivity shocks and hours worked (see e.g. Christiano et al, 2005 and Chari et al. 2005) could benet from an estimation technique that tracks the dynamics of the capital stock generated by the relevant shocks. The same applies to open economy models that study, for instance, the eects of a productivity shock on the current account (see e.g. Corsetti et al., 2006) and that typically omit a feedback from the stock of external debt on macroeconomic variables. This approach could also be used in the analysis of the eects of scal shocks on debt sustainability, an issue which cannot be addressed in the context of a VAR that fails to keep track of the debt dynamics. Stochastic simulations of (??) could also be used to evaluate the sustainability of current systematic scal policy and to compute the risk of an unstable debt dynamics implied by the current policy regime.

References

Bagliano, Fabio and C. Favero [1999]: "Information from nancial markets and VAR measures of monetary policy", The European Economic Review, 43, 825-837. 17

Barro R.J. [1974] "Are Government Bonds Net Wealth?" Journal of Political Economy , 82, 6, 1095-1117 Blanchard, Olivier and R. Perotti [2002]: "An Empirical Characterization of the Dynamic Eects of Changes in Government Spending and Taxes on Output", Quarterly Journal of Economics Bohn, Henning [1998]: The Behaviour of U.S. public debt and decits, Quarterly Journal of Economics, 113, 949-963. Burnside, Craig, M. Eichenbaum and J.D.M. Fisher [2003]: Fiscal Shocks and Their Consequences, NBER Working Paper No 9772 Canzoneri M.,R. Cumby and B.Diba [2001] "Is the Price Level Determined by the Needs of Fiscal Solvency, American Economic Review, Vol. 91, No. 5, 2001, pg.1221 - 1238. Cashell, Brian W. [2006]: "The Federal Government Debt: its size and economic signicance" CRS Report for Congress. Chari V.V., P.J. Kehoe and E.R. McGrattan [2005]: "A Critique of Structural VARs Using Business Cycle Theory", Federal Reserve Bank of Minneapolis Research Department Sta Report 364 Christiano, Lawrence J., M.Eichenbaum and R.Vigfusson [2005]: "Assessing Structural VARs", mimeo Chung Hess and D.Leeper [2007]: "What has Financed Government Debt?" NBER WP 13425 Corsetti, Giancarlo, L. Dedola and S. Leduc [2006]: "Productivity, External Balance and Exchange Rates: Evidence on the Transmission Mechanism Among G7 Countries, forthcoming Edelberg, Wendy, M. Eichenbaum and J. D.M. Fisher [1999]: Understanding the Eects of a Shock to Government Purchases, Review of Economics Dynamics, pp.166-206, 41 Fats, Antonio and I. Mihov [2001]: The Eects of Fiscal Policy on Consumption and Employment: Theory and Evidence, mimeo, INSEAD 18

Giavazzi, Francesco, T. Jappelli and M. Pagano [2000]: "Searching for NonLinear Eects of Fiscal Policy: Evidence from Industrial and Developing Countries", European Economic Review, 44, no. 7, June. Kormendi, Roger C. [1983] Government Debt, Government Spending, and Private Sector Behavior. American Economic Review 73 , 994-1010. Mountford, Andrew and H. Uhlig [2002]: What Are the Ecets of Fiscal Policy Shocks? CEPR Discussion Paper 3338 Perotti, Roberto [2007]: In Search of the Transmission Mechanism of Fiscal policy"; NBER Macroeconomic Annual, forthcoming Ramey, Valerie [2006]: "Identifying Government Spending Shocks: Its All in the Timing", mimeo, July Romer, Christina and David H. Romer [1989]: Does Monetary Policy Matter? A new test in the spirit of Friedman and Schwartz in Blanchard, O. and S. Fischer, (eds.), NBER Macroeconomics Annual, Cambridge, MIT Press, 4, 121-170 Romer, Christina and David H. Romer [2007]: The Macroeconomic Eects of Tax Changes: Estimates Based on a New Measure of Fiscal Shocks", mimeo, March.

19

1 .0

0 .8

0 .6

0 .4

0 .2

0 .0 55 60 65 70 75 DY 80 85 90 95 00 05

D Y_ I

Figure 1: Actual (DY) and simulated (DY_I) (dynamically backward and forward starting in 1970:1) debt-GDP ratio. Actual data are observed at quarterly frequency from 1970 onwards and at annual frequency from 1970 backward. The simulated data are constructed using the government intertemporal budget constraint (2) with observed data and initial conditions given by the debt-to-GDP ratio in 1970:1.

20

s h o c k s t o l_ g g .012 .010 .008 .000 .006 .004 .002 -. 004 .000 -. 002 2 4 6 8 10 12 14 16 18 20 -. 005 -. 006 2 4 -. 001 -. 002 -. 003 .003 .002 .001

s h o c k s t o l_ t t

10

12

14

16

1 8

20

.008

.012

.006 .004

.008

.004 .002 .000 .000 -. 004

-. 002

-. 004 2 4 6 8 10 12 14 16 18 20

-. 008 2 4 6 8 10 12 14 16 1 8 20

.004

.000

.003

-. 001

.002

-. 002

.001

-. 003

.000

-. 004

-. 001 2 4 6 8 10 12 14 16 18 20

-. 005 2 4 6 8 10 12 14 16 1 8 20

.0004

.0008

.0002

.0006

.0004 .0000 .0002 -.0002 .0000 -.0004

-. 0002

-.0006 2 4 6 8 10 12 14 16 18 20

-. 0004 2 4 6 8 10 12 14 16 1 8 20

.0005 .0004 .0003 .0002 .0001

.0003 .0002 .0001 .0000 -. 0001

.0000 -. 0002 -.0001 -.0002 -.0003 -.0004 2 4 6 8 10 12 14 16 18 20 -. 0003 -. 0004 -. 0005 2 4 6 8 10 12 14 16 1 8 20

Figure 2.1: Fiscal shocks identied from a SVAR:1960:1-1979:4. The rst column shows responses to shocks to gt ; the second column to shocks to tt .The responses reported along the rows refer, respectively, to the eects on gt , tt , yt , pt , it .

21

s hoc k s .012 .010

t o l_ g g .002 .001 .000

s hoc k s

t o l_ t t

.008

-.0 01 -.0 02

.006 -.0 03 .004 -.0 04 -.0 05 .002 -.0 06 .000 2 4 6 8 10 12 14 16 18 20 -.0 07 2 4 6 8 10 12 14 16 18 20

.005 .004

.020 .016

.003 .002 .001 .008 .000 -.0 0 1 -.0 0 2 .000 -.0 0 3 -.0 0 4 2 4 6 8 10 12 14 16 18 20 -.0 04 2 4 6 8 10 12 14 16 18 20 .004 .012

.005

.003

.004

.002

.003

.001

.002

.000

.001

-.0 01

.000 2 4 6 8 10 12 14 16 18 20

-.0 02 2 4 6 8 10 12 14 16 18 20

.0003 .0002 .0001 .0000

.0003 .0002 .0001 .0000

-.00 0 1 -.0 0 01 -.00 0 2 -.00 0 3 -.00 0 4 -.00 0 5 2 4 6 8 10 12 14 16 18 20 -.0 0 02 -.0 0 03 -.0 0 04 2 4 6 8 10 12 14 16 18 20

.0000 -.00 0 1 -.00 0 2 -.00 0 3 -.00 0 4 -.00 0 5 -.00 0 6 -.00 0 7 -.00 0 8 -.00 0 9 2 4 6 8 10 12 14 16 18 20

.0007 .0006 .0005 .0004 .0003 .0002 .0001 .0000 -.0 0 01 -.0 0 02 2 4 6 8 10 12 14 16 18 20

Figure 2.2: Fiscal shocks identied from a SVAR:1980:1-2006:3. The rst column shows responses to shocks to gt ; the second column to shocks to tt .The responses reported along the rows refer, respectively, to the eects on gt , tt , yt , pt , it .

22

. 0016

s h oc k s t o l_gg

. 0016

s hoc k s t o l_t t

. 0012

. 0012

. 0008

. 0008

1960-1979

. 0004

. 0004

. 0000

. 0000

-. 0004

-. 0004

-. 0008 2 4 6 8 10 12 14 16 18 20

-. 0008 2 4 6 8 10 12 14 16 18 20

. 003

. 000 -. 001

. 002 -. 002 . 001 -. 003 -. 004 . 000 1980-20 06 -. 001 -. 005 -. 006 -. 007 -. 002 -. 008 -. 003 2 4 6 8 10 12 14 16 18 20 -. 009 2 4 6 8 10 12 14 16 18 20

Figure 3: Response of the debt to GDP ratio to scal shocks identied from a SVAR.

.2 5 .2 0 .1 5 .1 0 .0 5 .0 0 -.0 5 60 65 70 75 80 85 90 95 00 05

a v e r a g e c o s t o f f i n a n c i n g t h e d e b t ( a n n u a li z e d ) q u a r t e r l y n o m i n a l G D P g r o w t h ( a n n u a li z e d )

Figure 4: Average cost of debt nancing and quarterly (annualized) nominal GDP growth

23

s h o c k s .0 1 2 .0 1 0 .0 0 8 .0 0 6

to

l_ g g .0 0 3 .0 0 2 .0 0 1 .0 0 0 -.0 0 1

s h o c k s

to

l_ t t

.0 0 4 -.0 0 2 .0 0 2 .0 0 0 -.0 0 2 -.0 0 4 2 4 6 8 1 0 1 2 1 4 1 6 1 8 2 0 -.0 0 3 -.0 0 4 -.0 0 5 -.0 0 6 2 4 6 8 1 0 1 2 1 4 1 6 1 8 2 0

.0 1 0 .0 0 8

.0 1 2

.0 0 8 .0 0 6 .0 0 4 .0 0 2 .0 0 0 -.0 0 2 -.0 0 4 -.0 0 4 -.0 0 6 2 4 6 8 1 0 1 2 1 4 1 6 1 8 2 0 -.0 0 8 2 4 6 8 1 0 1 2 1 4 1 6 1 8 2 0 .0 0 0 .0 0 4

.0 0 4

.0 0 0

.0 0 3

-.0 0 1

.0 0 2 -.0 0 2 .0 0 1 -.0 0 3 .0 0 0 -.0 0 4

-.0 0 1

-.0 0 2 2 4 6 8 1 0 1 2 1 4 1 6 1 8 2 0

-.0 0 5 2 4 6 8 1 0 1 2 1 4 1 6 1 8 2 0

.0 0 0 4

.0 0 0 8 .0 0 0 6

.0 0 0 2 .0 0 0 4 .0 0 0 0 .0 0 0 2 .0 0 0 0 -.0 0 0 2 -.0 0 0 4 -.0 0 0 4 -.0 0 0 6 2 4 6 8 1 0 1 2 1 4 1 6 1 8 2 0 -.0 0 0 6 2 4 6 8 1 0 1 2 1 4 1 6 1 8 2 0

-.0 0 0 2

.0 0 0 8

.0 0 0 2

.0 0 0 6

.0 0 0 0

.0 0 0 4 -.0 0 0 2 .0 0 0 2 -.0 0 0 4 .0 0 0 0 -.0 0 0 2 -.0 0 0 6

-.0 0 0 4 2 4 6 8 1 0 1 2 1 4 1 6 1 8 2 0

-.0 0 0 8 2 4 6 8 1 0 1 2 1 4 1 6 1 8 2 0

Figure 5.1: Fiscal shocks identied from a SVAR (solid line) and in model with feedbacks (dotted line). Sample 1960:1 1979:4. The rst column shows responses to shocks to gt ; the second column to shocks to tt .The responses reported along the rows refer, respectively, to the eects on gt , tt , yt , pt , it .

24

.012

s ho c k s

to

l_ g g

.0 06

s h oc k s

to

l_ t t

.010

.0 04

.008

.0 02

.006

.0 00

.004

-.0 0 2

.002

-.0 0 4

.000 2 4 6 8 10 12 14 1 6 1 8 20

-.0 0 6 2 4 6 8 10 12 1 4 16 18 20

.006

.0 20 .0 16

.004 .0 12 .002 .0 08 .0 04 .0 00 -.0 0 2 -.0 0 4 -.0 0 4 2 4 6 8 10 12 14 1 6 1 8 20 -.0 0 8 2 4 6 8 10 12 1 4 16 18 20

.000

.004

.0 03

.0 02 .003 .0 01 .002 .0 00 .001 -.0 0 1

.000 2 4 6 8 10 12 14 1 6 1 8 20

-.0 0 2 2 4 6 8 10 12 1 4 16 18 20

.0002 .0001 .0000

.000 3

.000 2

.000 1 -.0 0 0 1 .000 0 -.0 0 0 2 -.0 0 0 1 -.0 0 0 3 -.0 0 0 4 -.0 0 0 5 2 4 6 8 10 12 14 1 6 1 8 20 -.0 0 0 2

-.0 0 0 3 2 4 6 8 10 12 1 4 16 18 20

.0000

.000 8

-.0 0 0 2

.000 6

.000 4 -.0 0 0 4 .000 2 -.0 0 0 6 .000 0 -.0 0 0 8

-.0 0 0 2

-.0 0 1 0 2 4 6 8 10 12 14 1 6 1 8 20

-.0 0 0 4 2 4 6 8 10 12 1 4 16 18 20

Figure 5.2: Fiscal shocks identied from a SVAR (solid line) and in model with feedbacks (dotted line). Sample 1980:1 2006:3. The rst column shows responses to shocks to gt ; the second column to shocks to tt .The responses reported along the rows refer, respectively, to the eects on gt , tt , yt , pt , it .

25

.0015

shoc ks to l_gg

shocks to l_tt .0016

.0010

.0012

.0005 1960- 1979 .0000

.0008

.0004

-.0005

.0000

-.0010 2 4 6 8 10 12 14 16 18 20

- .0 00 4 2 4 6 8 10 12 14 16 18 20

.0010

. 00 0 -.001 -.002

.0005

.0000

-.003 -.004 -.005 -.006

-.0005 1980-2006 -.0010

-.0015

-.007 -.008 2 4 6 8 10 12 14 16 18 20 2 4 6 8 10 12 14 16 18 20

-.0020

Figure 6: Eect on the debt to GDP ratio of scal shocks identied from a SVAR (solid line) and in model with feedbacks (dotted line).

26

1 9 5 0 : 1 -1 9 8 0 : 4 .002 .000 -. 0 0 2 -. 0 0 4 -. 0 0 6 -. 0 0 8 -. 0 1 0 -. 0 1 2 -. 0 1 4 2 4 6 8 10 12 14 16 18 20 2 4 6

1 9 8 1 : 1 -2 0 0 6 : 2

10

12

14

16

18

20

R& R V A R w it h o u t IG B C V A R w it h IG B C

Figure 7: Using the Romer and Romer (2007) tax shocks. Eect on output in dierent models

.0 1 2 .0 0 8 .0 0 4 .0 0 0 -.0 0 4 -.0 0 8 -.0 1 2 -.0 1 6 2 4 6 8 g t 1 0 1 2 y D p 1 4 1 6 i d 1 8 2 0

Figure 8: Dynamic response of all variables to an R&R tax shock, in a VAR with a debt feedback estimated over the sample 1950:1-1980:1.

27

Table 1

The effect of d t1 d in a VAR (st. errors in parenthesis) gt tt yt p t 0. 006

0.027 0.0067

it 0. 029

d t1 d 1960:1-1979:4 0. 101
0.197 0.052

0. 650 0. 161
0.225 0.082 0.085 0.0196

0.0097

1980:1-2006:2 0. 129 0. 054 0. 042 0. 0089 0. 0082

0.0062

Table 2: Cumulative responses of y and i to a g and a t shock Cumulative responses to g and t shocks equal to 1 per cent (annualized). Bootstrapped confidence intervals are reported below estimates

Hor

without debt feedback

60:1-79:4 80:1-06:2 60:1-79:4 80:1-06:2 60:1-79:4

with debt feedback

80:1-06:2 60:1-79:4 80:1-06:2

g shock y t t4
t12 t20 0. 065
0.005, 0.13

t shock
0. 22
0.28, 0.15

g shock
0. 06
0.01, 0.13

t shock
0. 24
0.30, 0.18

0. 16
0.10, 0.22

0. 0008
0.06, 0.06

0. 125
0.06, 0.18

0. 007
0.06, 0.07

0. 390
0.15, 0.66

0. 70
0.48, 0.96

0. 86
1.13, 0.61

0. 160
0.12, 0.39

0. 38
0.12, 0.65

0. 617
0.42, 0.86

0. 84
1.14, 0.61 1.84,0.89

0. 122
0.15, 0.40

0. 50
0.06, 0.94

1. 24
0.82,1.72

1. 339
1.83, 0.91

0. 22
0.40, 0.74

0. 41
0.02, 0.86

1. 02
0.68, 1.46

1. 30

0. 168
0.37, 0.69

it

t4 t12 t20

0. 003
0.01, 0.01

0. 045
0.06, .0.03

0. 003
0.01, 0.013

0. 009
0.004, 0.026

0. 007
0.02, 0.001

0. 055
0.07, .0.04

0. 006
0.01, 0.001

0. 016
0.002, 0.03

0. 002
0.05, 0.05

0. 135
0.17, 0.098

0. 005
0.06, 0.05

0. 056
0.003, 0.112

0. 032
0.005, 0.06

0. 136
0.17, 0.10

0. 059
0.08, 0.03

0. 0671
0.02, 0.12

0. 04
0.02, 0.10

0. 21
0.27, 0.16

0. 04
0.12, 0.03

0. 099
0.0243, 0.177

0. 098
0.05, 0.16

0. 189
0.24, 0.14

0. 14
0.20, 0.09

0. 108
0.03, 0.19