Fiscal Multipliers and the State of the Economy
  • 1 https://isni.org/isni/0000000404811396, International Monetary Fund

Contributor Notes

Only a few empirical studies have analyzed the relationship between fiscal multipliers and the underlying state of the economy. This paper investigates this link on a country-by-country basis for the G7 economies (excluding Italy). Our results show that fiscal multipliers differ across countries, calling for a tailored use of fiscal policy. Moreover, the position in the business cycle affects the impact of fiscal policy on output: on average, government spending, and revenue multipliers tend to be larger in downturns than in expansions. This asymmetry has implications for the choice between an upfront fiscal adjustment versus a more gradual approach.

Abstract

Only a few empirical studies have analyzed the relationship between fiscal multipliers and the underlying state of the economy. This paper investigates this link on a country-by-country basis for the G7 economies (excluding Italy). Our results show that fiscal multipliers differ across countries, calling for a tailored use of fiscal policy. Moreover, the position in the business cycle affects the impact of fiscal policy on output: on average, government spending, and revenue multipliers tend to be larger in downturns than in expansions. This asymmetry has implications for the choice between an upfront fiscal adjustment versus a more gradual approach.

I. Introduction

Since the economic crisis a rapidly expanding empirical literature tries to estimate the effect of discretionary fiscal policy on output. However, only a few empirical studies have so far analyzed the links between fiscal multipliers and the underlying state of the economy.

This paper investigates how the effects of fiscal policy on output may vary depending on whether the economy is in an expansion or a downturn. Expansions and downturns are defined by the sign of the output gap (positive and negative, respectively). The decision to use the output gap as the threshold variable is motivated by several factors, one of them is that under a negative output gap—independently of the sign of the GDP growth rate—excess capacities are available in the economy, reducing the crowding out of private investment following a government spending shock.

The contribution of the paper is twofold. First, to the best of our knowledge, this is the first study to develop a dataset of quarterly data on government expenditure and revenue for six of the G7 economies (excluding Italy) going back to the 1970s.2 Second, country-by-country estimation allows the explanatory variables (government spending and revenue) to have differing regression slopes, depending on whether the chosen threshold variable—the output gap—is above or below a particular level, which is chosen to maximize the fit of the model.

Our analysis employs a nonlinear threshold vector autoregressive model (TVAR) which separates observations into different regimes based on a threshold variable. Within each regime, the model is assumed to be linear. However, after a fiscal shock is implemented, the regime is allowed to switch, depending on the level of the output gap. As a result, the effects of fiscal policy shocks on economic activity depend on their size, direction and timing with respect to the business cycle.

The paper shows that the position in the business cycle affects the impact of fiscal policy on output: for an average of G7 economies, government spending and revenue multipliers tend to be larger in downturns than in expansions. This asymmetry has implications for the desirability of upfront fiscal adjustment versus a more gradual approach. When the output gap is initially negative, fiscal adjustment implemented gradually has a smaller negative impact on growth (cumulative over two and one-half years) than does an up-front consolidation of the same overall size. This suggests that when feasible, a more gradual fiscal consolidation is likely to prove preferable to an approach that aims at “getting it over quickly.”

Multipliers are found to differ significantly across countries, calling for a tailored use of fiscal policies and a country-by-country assessment of their effects. In those countries where spending impact multipliers are found to be statistically significant and sizeable (Germany, Japan, and the United States), spending shocks have a significantly larger effect on output when the output gap is negative than when it is positive. In the United Kingdom, spending multipliers are small under both positive and negative output gaps. The results are generally less conclusive for revenue multipliers. The impact is more significant for Canada, France, Germany, and Japan. In Germany, revenue multipliers are slightly higher in “good times” than in “bad times”, which could suggest that individuals and firms are more willing to spend additional income when market sentiment is positive, thereby becoming less Ricardian. In Canada and Japan revenue measures work as a countercyclical tool only when the output gap is negative.

The paper is structured as follows. Section II provides background information on fiscal multipliers and summarizes the findings of other studies that have estimated regime dependent multipliers. Section III presents the data sources and outlines the methodology. The main results as well as related policy implications can be found in Section IV. Section V concludes.

II. Background and Literature Review

A. What are Fiscal Multipliers and How Large are They?

Fiscal multipliers are typically defined as the ratio of a change in output to an exogenous and temporary change in the fiscal deficit with respect to their respective baselines (Spilimbergo, Symansky, and Schindler, 2009). In spite of an extensive literature, there is still no consensus regarding the size of fiscal multipliers. They tend to be smaller in more open economies and in countries with larger automatic stabilizers (Figure 1), but as the theoretical and empirical literature suggest, they differ widely across countries.

Figure 1.
Figure 1.

Country Characteristics and Multipliers

Citation: IMF Working Papers 2012, 286; 10.5089/9781475565829.001.A001

Sources: IMF, Fiscal Affairs Department Fiscal Rules database and Fiscal Transparency database; Organization for Economic Cooperation and Development (OECD); and IMF staff estimates.Note: Multipliers are based on the OECD (2009). Openness is measured by import penetration, that is the 2008–11 average of Imports/(GDP − Exports + Imports)*100. Auto matic stabilizers are measured as the semielasticity of the budget balance and are extracted from André and Girouard (2005). The negative correlations in the panel are ro bust to outliers being removed using an automated Stata procedure based on leverage (a measure of how far an in dependent variable deviates from its mean) and residual in the equation.

A comprehensive literature review on fiscal multipliers can be found in Baunsgaard and others (2012), who extend and update Spilimbergo, Symansky, and Schindler (2009). Baunsgaard and others (2012) review a total of 37 studies including both model based (DSGE) and vector autoregressive (VAR) approaches. For those studies government spending multipliers range between 0 and 2.1, with a mean of 0.8 during the first year after fiscal measures are taken. Government revenue multipliers range from about −1.5 to 1.4, with a mean of 0.3.

B. Do Multipliers Differ in Downturns and Expansions?

Although most studies do not distinguish between multipliers in different underlying states of the economy, the effects of fiscal policy shocks on economic activity are likely nonlinear. Multipliers could be significantly larger in downturns than in expansions. In times of a negative output gap, the traditional crowding-out argument—that higher government spending displaces private spending—is generally less applicable since excess capacities are available in the economy. In addition, the proportion of credit-constrained households and firms, which adjust spending in response to a change in disposable income, is higher.

Surprisingly few studies have tried to distinguish between multipliers in downturns and expansions. These have mostly focused on a single country (Germany: Baum and Koester, 2011; and the United States: Auerbach and Gorodnichenko, 2012a) or employed a panel data approach, thereby providing average multipliers across countries, which may mask important heterogeneities in the estimation process (Auerbach and Gorodnichenko, 2012b).3

A recent study that is close to our paper and that distinguishes between multipliers on a country-by-country basis is the work by Batini, Callegari, and Melina (2012). Using regime-switching VARs with output growth as the threshold variable, the paper focuses on interactions between fiscal and monetary policies. It estimates the impact of fiscal adjustment in the United States, Europe (the Euro area as a whole, Italy and France) and Japan, allowing fiscal multipliers to vary across recessions and booms. A fiscal consolidation is found to be substantially more contractionary if made during a recession than during an expansion. First-year cumulative multipliers for consolidations that began during downturns range between 1.6 to 2.6 for expenditure shocks, and 0.2 to 0.4 for tax shocks. First-year cumulative multipliers for consolidations that began in expansions range from 0.3 to 1.6 for expenditure shocks, and -0.3 to 0.2 for tax shocks. Second-year cumulative multipliers have similar sizes to 1-year multipliers, implying that a large part of the impact of fiscal shocks on output materializes within 4 quarters. A summary of results from selected studies on fiscal multipliers that employ non-linear approaches is provided in Table 1.

Table 1.

Cumulative Fiscal Multiplier Estimates from Selected Non-Linear Approaches

article image

Multipliers reported here reflect the real GDP response (in percent) to a 1 percent spending shock.

Our paper conducts a nonlinear time-series analysis for six G7 countries, applying a threshold methodology that closely follows Baum and Koester (2011). The threshold value is determined endogenously, allowing the data to find the value of the output gap that maximizes the fit of the model in both regimes. It contrasts with Auerbach and Gorodnichenko (2012a), who use a regime-switching structural VAR (SVAR) in which the threshold value has to be determined exogenously. Furthermore, Auerbach and Gorodnichenko (2012a) use a moving average presentation of the GDP growth rate as the threshold variable. The main difference between our paper and Batini, Callegari, and Melina (2012) is the country sample as well as the threshold variable.

The reasons to employ the output gap instead of the GDP growth rate are manifold. The output gap is the measure most commonly used to identify economic cycles, as it is seen not only as reliable ex-post but also as a reliable real-time indicator for policy-makers. It is thus an appropriate choice given our focus on downturns and expansions. More importantly, one argument for fiscal policy being more effective in downturns than in expansions is that under a negative output gap, excess capacities are available in the economy, making the crowding out of private investment lower. This argument is expected to hold as long as the output gap is negative, which can hardly be captured by low or negative growth rates. The GDP growth rate has also the disadvantage that it can be positive after output has reached its trough, while a negative output gap can prevail for various further quarters (see Woo, Kinda, and Poplawski-Ribeiro, 2013). Furthermore, the usual presence of positive serial correlation in GDP growth rates plays a role in explaining business cycles length. Business cycles are often found to be shorter when one uses the GDP growth rates (Harding and Pagan, 2002).4

III. Data and Methodology

A. Data Sources and Description

The countries included in our sample are Canada, France, Germany, Japan, the United Kingdom and the United States.5 For most countries we construct quarterly datasets since at least the 1970s. Data sources include the Organization for Economic Cooperation and Development (OECD) Economic Outlook, the IMF’s International Financial Statistics and Eurostat as well as national account data. Fiscal data cover the general government. There are some caveats regarding the data sources, especially in the cases of Japan and France, for which data were interpolated for some years (see also Perotti, 2005).

The vector autoregression consists of three variables, namely real GDP, real net revenue and real net expenditure, as in the seminal paper by Blanchard and Perotti (2002). The net revenue series is equal to general government revenues minus net transfers; and government spending comprises general government investment and general government consumption (but excludes transfers and subsidies). All series are deflated with the GDP deflator. For most of the countries—except for Germany, for which the HP filter is used (see Baum and Koester, 2011)—output gap data are obtained directly from the OECD. A detailed description of the data can be found in Appendix A.

B. Threshold VAR Methodology

A threshold VAR is a simple method to model changing dynamics of a set of variables over two or more distinct regimes. The regimes are determined by a transition variable, which is either endogenous or exogenous (Hansen 1996, 1997, Tsay 1998). In general, it is possible to obtain more than one critical threshold value, but for simplicity we will focus on a model with only two regimes.

The threshold VAR can be represented as

yt=δ1Xt+δ2XtI[Zt-dZ*]+ut(1)

zt–d is the threshold variable determining the prevailing regime of the system, with a possible lag d. I[•] is an indicator function that equals 1 if the threshold variable zt–d is above the threshold value z*, and 0 otherwise. The coefficient matrices δ1 and δ2, as well as the contemporaneous error matrix ut, are allowed to vary across regimes. The delay lag d and critical threshold value z* are unknown parameters and are estimated alongside the parameters.

Whether or not system (1) offers threshold behavior is determined by means of the Tsay (1998) multi-variate threshold approach. The method applies a white noise test to predictive residuals of an arranged regression.6 A detailed description of the testing procedure can be found in Tsay (1998), as well as in Baum and Koester (2011).

Impulse response (IR) functions need to be based on well identified shocks. This study employs the Blanchard and Perotti (2002, BP) structural identification procedure, which accounts for the effect of automatic stabilization on revenues. Revenue elasticities with respect to GDP are obtained following OECD calculations (Girouard and André, 2005). Subsequently, the share of direct and indirect taxes, social security contributions, and social spending (transfers) in total net revenue are multiplied by their respective elasticities to construct quarterly weighted elasticities.

The BP approach has been subject to various criticisms (IMF, 2010). These include that it may fail to capture exogenous policy changes correctly, since changes in revenues are not only due to cyclical developments and discretionary policy, but also to asset and commodity price movements. For example, a boom in the stock market improves cyclically-adjusted tax revenues and is also likely to reflect developments that raise private consumption and investment. Such measurement error is likely to bias the analysis towards downplaying contractionary effects of deliberate fiscal consolidation. A rise (fall) in cyclically adjusted revenue (spending) may also reflect a government’s decision to raise taxes or cut spending to restrain domestic demand and reduce the risk of overheating. In this case, using the cyclically adjusted data to measure the effect of fiscal consolidation on economic activity would suffer from reverse causality and bias the analysis towards supporting the expansionary fiscal contractions hypothesis.

Alternative methods proposed include the “narrative” and “action”-based approaches by Romer and Romer (2010) and the IMF (2010), which use information from budget documents to directly identify exogenous policy changes. So far, the narrative approach has only been applied using quarterly data for the United Kingdom (Cloyne, 2011) and the United States (Romer and Romer, 2010). The IMF (2010) created a multiple country data set based on this approach (see also Devries and others, 2011), but it only includes annual data. Therefore, given the lack of quarterly data of comparable quality for the countries in our sample, the BP approach is employed in this study.7

In order to take previous criticism into account, the net revenue and expenditure series are corrected to eliminate, to the extent possible, those changes in government revenues and expenditure that are not necessarily linked to fiscal policy decisions and that cyclical adjustment methods may fail to capture (for example, large movements in asset or commodity prices).8 This removes the largest—but not all—measurement errors, as identified episodes in IMF (2010) refer to cases of fiscal consolidations and not expansions. Furthermore, the IMF (2010) only provides data on an annual basis and therefore covers only part of our dataset.9 Hence, especially the responses of output to revenue shocks have to be interpreted cautiously.

C. Impulse Response Functions

The impulse response functions (IRFs) need to reflect the nonlinearity of our model. The challenge in computing IRFs in a nonlinear model is that they should allow not only the shock impact to depend on the regime itself, but also the regime to switch after a shock has been implemented. The latter is important, as output—and the output gap—evolve over time following a fiscal policy shock. Thus, not considering regime switches in the impulse response functions could result in over- or understated fiscal multipliers.

The generalized impulse response function (GIRF), developed by Koop (1996) and Koop, Pesaran, and Potter (1996), addresses nonlinearity by being history-dependent. This implies that the IRF depends on the specific time period in which the shock occurs. Formally, we implement shocks for each period within one regime and then take regime averages to obtain the GIRFs.10 Defining εt as a shock of a specific size, m as the forecasting horizon and Ωt-1 as the history or information set at time t-1, the GIRF for each period is described as the difference between two conditional expectations:

GIRF=E[Xt+m|εt,εt+1=0,,εt+m=0,Ωt1]E[Xt+m|εt=0,εt+1=0,,εt+m=0,Ωt1](2)

Since the GIRF methodology allows the regimes to switch after a fiscal shock is implemented, the IRFs depend on the size and also the direction (sign) of the shocks. For example, a positive spending shock in a downturn could increase output for several quarters, closing the output gap and inducing a shift into the expansionary regime. A negative shock might not cause the same shift of regimes.

Nevertheless, due to various features of our GIRF generation, the differences between positive and negative shocks tend to be small:

  • First, the output gap in our sample is rather persistent. It does not close immediately after a shock of a reasonable size is implemented (2 percent shocks are used).

  • Second, the output gap has to be updated after each forecast period, which makes the forecast of the GDP trend necessary. The one-sided Hodrick-Prescott (HP)-filter could potentially be applied to update the trend, as done in Baum and Koester (2011), but this yields very little precision along the boundaries. Instead, within the forecast horizon we take the trend as given, so that it follows the evolution of the original trend series. The trend GDP is thus unaltered by fiscal shocks for several quarters, allowing for less variation between positive and negative shocks than in Baum and Koester (2011).11

  • Third, the output gap itself enters the VAR with one or more lags. For instance, in cases in which the highest threshold significance was obtained for an output gap in three lags, there will be no difference between positive and negative shocks for the first three quarters of the IRF.

Confidence bands are constructed using the standard parametrical bootstrap procedure following Luetkepohl (2000). This method randomly draws from the estimated residuals, recursively computes bootstrap time series, and re-estimates the coefficient matrices accordingly for a large number of repetitions (500). Thus, the non-linear impulse responses reported are averages of stochastic simulations, while the confidence bands are percentiles of 500 stochastic simulations of the nonlinear impulse responses. The 1 standard deviation confidence bands are taken from the distribution of the resulting IRFs.

IV. Results

A. Country-by-Country Results

Table 2 summarizes selected descriptive statistics. For all countries but Japan the majority of statistical tests suggest a specification with one lag in the VAR.

Table 2.

G7 Selected Countries: Descriptive Statistics

(Percent of GDP, unless otherwise specified)

article image
Source: Authors’ calculations.Note:

Unit value. The VAR lag length is chosen based on the majority of suggestions by the sequential modified LR test statistic (LR), the Final prediction error (FPE), the Akaike information criterion (AIC), the Schwarz information criterion (SC) and the Hannan-Quinn information criterion (HQ).

The results for the Tsay threshold approach are presented in Table 3. The estimated threshold values and the corresponding lag length are chosen based on the highest significance.12 Since the threshold output gap value for most countries is relatively small, the discussion that follows refers to the two regimes as the positive and negative output gap regimes or, simply, as expansions and downturns. Apart from the United Kingdom, the threshold value is below the average output gap (see Table 2) and negative for all countries. Consequently, for most of the countries, the majority of the observations lie in the upper output gap regime. The suggested threshold values are significant at the 10 percent level for France, at 5 percent for the United Kingdom, and at 1 percent for Canada, Germany, Japan and the United States. Therefore, we estimate a two regime threshold SVAR for all countries in our sample.

Table 3.

G7 Selected Countries: Threshold Estimation

article image
Source: Authors’ calculations.Note: *,**,*** indicate significance levels at the 10, 5, 1 percent level respectively.

Figures 2 and 3 present four and eight quarter cumulative multipliers for each country. In addition, Appendix B presents the cumulative GIRFs for each country under the two output gap regimes for a fiscal expansion (Figures B.1 and B.2) and a fiscal contraction (Figures B.3 and B.4).

Figure 2.
Figure 2.

Cumulative Fiscal Multipliers: Fiscal Expansion

Citation: IMF Working Papers 2012, 286; 10.5089/9781475565829.001.A001

Source: IMF staff estimates.Note: The striped bars correspond to those measures for which no significant multiplier is found at the time the fiscal shock is implemented.
Figure 3.
Figure 3.

Cumulative Fiscal Multipliers: Fiscal Contraction

Citation: IMF Working Papers 2012, 286; 10.5089/9781475565829.001.A001

Source: IMF staff estimates.Note: The striped bars correspond to those measures for which no significant multiplier is found at the time the fiscal shock is implemented.

We find broad supportive evidence for a nonlinear impact of fiscal policy on output. Government spending shocks have a larger effect on output when the output gap is negative (Canada being the only exception). This is particularly true for those countries where spending multipliers are statistically significant on impact (see Appendix B), and sizeable (Germany, Japan, and the United States).13

The results are generally less conclusive for revenue multipliers. However, quite a consistent result across countries is that first year revenue multipliers are small (on average well below 0.5). Their impact is statistically significant for Canada, France, Germany, and Japan. In Germany, revenue multipliers are slightly higher in “good times” than in “bad times”, which could suggest that individuals and firms are more willing to spend additional income when market sentiment is positive, thereby becoming less Ricardian. In Canada and Japan revenue measures work as a countercyclical tool only when the output gap is negative.

Using output growth as a threshold variable rather than the output gap yields results that are qualitatively similar, with the exception of France. With GDP growth as the threshold variable, fiscal expansions result in an increase in output for France and vice versa for fiscal contractions. For the other countries, the results remain comparable, although in the case of Canada, using output growth as the threshold gives much larger multipliers in the downturn regime. A detailed comparison of multipliers between models employing the different thresholds variables is shown in Appendix C.

B. Average of G7 Economies

Figure 4 shows multipliers for the average of the G7 economies (excluding Italy). It broadly supports the above findings, with both spending and revenue multipliers significantly larger in times of a negative output gap than when the output gap is positive.

Figure 4.
Figure 4.

Fiscal Multipliers in G-7 Economies

Citation: IMF Working Papers 2012, 286; 10.5089/9781475565829.001.A001

Source: IMF staff calculations.Note: Cumulative multipliers are standardized multipliers over four quarters. Only statistically significant multipliers are included in the average. Average revenue multipliers exclude France, for which the outliers are large and data limitations are particularly severe. Italy is not included in the G7 average.

Figure 4 also includes average multipliers estimated with a standard linear structural VAR for the same countries and period. We find that these multipliers lie between the positive and negative regime multipliers, and that they are very much in line with averages identified in the previous literature (Baunsgaard and others, 2012). This suggests that the linear model underestimates the effects of spending and revenue shocks during downturns and overestimates their effects in expansions.

Assuming, in line with recent fiscal adjustment packages in advanced economies, that two thirds of the adjustment comes from spending measures, a weighted average of spending and revenue multipliers in downturns yields an overall fiscal multiplier of about 1.

C. Discussion and Caveats

The results indicate that multipliers vary significantly between and within countries, which calls for a tailored use of fiscal policies and a country-by-country assessment of their effects. The empirical results are mostly in accordance with other studies on fiscal multipliers (see Favero, Giavazzi, and Perego, 2011, Perotti, 2005). We confirm the sizable spending multipliers that have been found in the previous literature for the United States. For Canada and the United Kingdom, our low expenditure multipliers are in line with Perotti (2005), who, using a structural VAR, finds that spending multipliers have decreased significantly since the 1980s.

We find that revenue multipliers in the United States and the United Kingdom are very small and not statistically significant. This could be due to a change in the impact of revenue measures on output over time, while our results reflect the historical average impact of fiscal policy. Perotti (2005) shows that prior to the 1980s, tax cuts had a significant positive impact on GDP, but in the period after 1980 this effect became negative. These results contradict the findings of Romer and Romer (2010) and Cloyne (2011), who, using a narrative approach to construct a dataset on exogenous revenue shocks, find significant and large revenue multipliers for the United States and the United Kingdom, respectively. 14 However, recent work by Favero and Giavazzi (2012), as well as Perotti (2011), demonstrate that the revenue multipliers in Romer and Romer (2010) are subject to a strong upward bias as their specification cannot be interpreted as a moving average (MA) representation of the output process. When using a “corrected” truncated MA representation, Favero and Giavazzi (2012) estimate revenue multipliers of around -0.5.

Our results are also mostly in line with the analyses that control for the state of the cycle (Auerbach and Gorodnichenko (2012a) and Batini, Callegari, and Melina (2012)). Our study confirms the state dependency of fiscal multipliers and shows that multipliers, and especially spending multipliers, are significantly larger in downturns than in expansions. Spending multipliers in the United States are found to be significantly above one during downturns.

We find that revenue multipliers are significantly smaller than spending multipliers. This can be explained with basic Keynesian theory, which argues that tax cuts are less potent than spending increases in stimulating the economy since households may save a significant portion of the additional after-tax income. However, a number of earlier studies have shown that expenditure-based fiscal consolidations have a more favorable effect on output than revenue-based consolidations (see, for example, Alesina and Ardagna, 2010). IMF (2010) reaches the same conclusion and notes that this result is partly due to the fact that, on average, central banks lower interest rates more in case of expenditure-based consolidations (perhaps because they regard them as longer-lasting).15 When interest rates are already low, the interest rate response becomes less relevant. This may imply that, in the current environment, the Keynesian positive fiscal multiplier prediction prevails.16

When thinking about the exact design of a fiscal consolidation package one needs to take into account other factors in addition to the size of multipliers. Notably, the long-term effects of specific adjustments and the efficiency of tax and expenditure changes depend on their preexisting levels and structure. For example, the current high tax pressures in some countries (particularly in Europe) suggest that the bulk of the fiscal adjustment should focus on the expenditure side (although revenue increases may be inevitable when the targeted adjustment is large).

Moreover, several important caveats apply to our analysis. First, the model only includes three variables and does not take into account possible interactions with monetary policy and public debt. For instance, Auerbach and Gorodnichenko (2012b) find that the size of government debt reduces the response of output to government expenditure shocks (see also Ilzetzki, Mendoza, and Vegh, 2010).17 Thus, the analysis may have overestimated fiscal multipliers, especially in high debt countries.18 Second, some of the country heterogeneities may be the result of different data sources. Data limitations are particularly important for France, where true quarterly data are available only since the 1990s.

D. Policy Implications: Up-front versus Gradual Implementation

An important policy implication of the found asymmetries is that if financing allows, gradual fiscal adjustment may in some cases be preferable to a more frontloaded approach. For example, when the output gap is negative, at the time the fiscal shock is implemented, a gradual spending adjustment will have a smaller negative impact on output in the short term than an up-front reduction.

Figure 5 illustrates this for an average of the G7 economies in the sample (excluding Italy). It shows the impact of a one euro (or the relevant national currency) front-loaded improvement in the fiscal deficit versus a gradual improvement that is spread evenly over two years. When the output gap is negative initially, a more gradual fiscal adjustment hurts growth less in the first two and one-half years of the simulation period. Conversely, when the output gap is initially positive, a more front-loaded shock has a smaller cumulative impact on growth. 19

Figure 5.
Figure 5.

G-7 Economies: Cumulative Impact on Output from a Negative Discretionary Fiscal Spending Shock

Citation: IMF Working Papers 2012, 286; 10.5089/9781475565829.001.A001

Sources: National sources; and IMF staff estimates.Note: The figure shows average multipliers for G7-countries with significant impact multipliers.

An explanation for this finding lies in the nonlinear nature of the impulse response functions. They allow the regime to switch after the impact of the shock. Thus, if the shock initially occurs in a negative output gap regime, over the course of the tightening there is some probability of moving into a positive output gap regime in which multipliers are lower. With a longer fiscal consolidation period, the probability of this occurring is higher. Conversely, if the impact of the shock initially occurs in a positive output gap regime, then policymakers should use the favorable conditions and tighten upfront.

The discussion of up-front versus gradual adjustment is subject to some caveats. First, our results do not include anticipation effects. Especially in case of a gradual adjustment, such effects could alter the growth impact significantly. Second, a sharp up-front fiscal adjustment might be accompanied by further negative growth effects (such as a further downward pressure on employment, human capital, and financial markets), which our model does not capture in the current specification. If such additional negative impacts were to occur, the upward sloped parts of the IRFs for the up-front fiscal adjustment might not materialize, or only much later. Third, a sharp up-front adjustment may increase market confidence. Fiscal consolidation can in general calm markets, in which case the results of the up-front adjustment might be biased downwards. However, in the current sovereign debt crisis the bond spreads seem largely driven by GDP growth prospects (Cottarelli and Jaramillo, 2012).

V. Conclusions

This paper investigates the relationship between fiscal multipliers and the underlying state of the economy on a country-by-country basis for the G7 economies (except Italy). It extends the rapidly evolving literature on fiscal multipliers using non linear estimation techniques and a new dataset for six of the G7 economies.

We find evidence that the impact of fiscal policy on economic activity varies with the business cycle and that the effect of fiscal policy on output is nonlinear. Fiscal multipliers for the six economies analyzed are on average larger in times of negative output gaps than when the output gap is positive.

However, the value of multipliers differs noticeably across countries. Spending shocks tend to have a larger effect on output when the output gap is negative, particularly in those countries where spending impact multipliers are statistically significant and sizeable (Germany, Japan, and the United States). The results are generally less conclusive for revenue multipliers. For Canada, France, Germany, and Japan the impact is statistically significant. However, in Germany revenue multipliers are slightly higher in “good times” than in “bad times”. In Canada and Japan, on the other hand, revenue measures work as a countercyclical tool only when the output gap is negative. This heterogeneity of the multipliers calls for a tailored use of fiscal policies and a country-by-country assessment of their effects.

The finding that the impact of fiscal policy on output depends on the underlying state of the economy has important implications for the choice between an upfront fiscal adjustment versus a more gradual approach. When the output gap is negative at the time the fiscal shock is initially implemented, an up-front negative fiscal spending shock will have a larger short-term impact on output than a more gradual fiscal adjustment.

Our analysis can be extended in various directions. It would be relevant to investigate the interaction between fiscal multipliers and monetary policy, particularly during periods in which interest rates are close to the zero lower bound. Moreover, the multiplier effects of different revenue and expenditure components, and how these are related to the underlying state of the economy, could be analyzed. The country sample could also be extended to other advanced and emerging economies, to investigate the state dependency of multipliers in a broader group of countries.

Fiscal Multipliers and the State of the Economy
Author: Ms. Anja Baum, Mr. Marcos Poplawski Ribeiro, and Anke Weber