Fiscal Policy Effectiveness in a Small Open Economy
Estimates of Tax and Spending Multipliers in Paraguay1
Author: Mr. Antonio David

Contributor Notes

Author’s E-Mail Address: adavid@imf.org
Publication Date:
22 Mar 2017
eISBN:
9781475588262
Language:
English
Keywords:
WP; multiplier; government expenditure; tax shock; B. tax multiplier; expenditures in Paraguay; government expenditures in Paraguay; government spending shock; expenditure shock
Download PDF (1.6 MB)

This paper presents estimates of fiscal multipliers in Paraguay following different econometric techniques and identification approaches. The results point to multipliers for capital expenditure that are substantially higher than multipliers for current expenditure. In addition, the evidence suggests that tax multipliers are close to zero when using conventional identification approaches, but estimates can be much larger when considering the “narrative” approach. One implication of the results is that the balanced budget multiplier for Paraguay i.e. the effect of on output of an increase in expenditures (in particular capital expenditure) financed by taxes is likely to be positive.

Abstract

This paper presents estimates of fiscal multipliers in Paraguay following different econometric techniques and identification approaches. The results point to multipliers for capital expenditure that are substantially higher than multipliers for current expenditure. In addition, the evidence suggests that tax multipliers are close to zero when using conventional identification approaches, but estimates can be much larger when considering the “narrative” approach. One implication of the results is that the balanced budget multiplier for Paraguay i.e. the effect of on output of an increase in expenditures (in particular capital expenditure) financed by taxes is likely to be positive.

I. Introduction

This paper presents estimates of fiscal multipliers in Paraguay following different estimation techniques and identification approaches. It also discusses selected recent studies with a special focus on empirical results for developing countries (see Annex A). An examination of multipliers in Paraguay is timely for a number of reasons, including the government’s desire to boost investment spending in light of the country’s sizeable infrastructure needs as well as several recent revenue mobilization efforts with technical assistance support from the IMF and other partners.

Estimating multipliers for expenditures and taxes empirically is warranted because different theoretical frameworks yield opposing predictions regarding the impact of discretionary fiscal policy on output. For example, in neoclassical models, an unexpected increase in government expenditures represents a negative wealth shock (given higher expected taxes in the future to satisfy the government’s intertemporal budget constraint) leading to a fall in consumption, increase in labor supply, and a decrease in real wages for a given level of labor demand (Engemann, Owyang and Zubairy, 2008). Nevertheless, in models with nominal rigidities and credit constraints, consumption and real wages are expected to rise in response to a positive government spending shock.

Furthermore, it is not certain how specific characteristics of emerging economies like Paraguay affect the size of fiscal multipliers compared to estimates obtained for advanced economies (Batini et al, 2014). For example, credit constraints are likely to be more prominent in developing countries, which would a priori suggest higher multipliers, but these economies also tend to be smaller and more open, which could imply lower multipliers because of “leakages” in demand through higher imports. In addition, standard models also suggest that countries with flexible exchange rate regimes would tend to present smaller multipliers as real exchange rate movements may offset some of the impact of fiscal policy on output.

We follow Ramey and Zubairy (2016) and define multipliers as the cumulative response in GDP relative to the cumulative government spending/taxes during a given period (“cumulative” multipliers). Alternative definitions of fiscal multipliers have been frequently considered in the literature, such as the ratio of the peak of the output response to an initial government spending/tax shock (Blanchard-Perotti, 2002) or the ratio of the average output response to an initial fiscal policy shock (Auerbach & Gorodnichenko, 2012). Nevertheless, we believe that cumulative multipliers considered here are more directly linked to the question of relevance to policy makers (i.e. calculating by how many Guaranies GDP changes if government expenditures or taxes change by X Guaranies). Moreover, it is important to calculate multipliers by comparing the integral of output responses to the integral of expenditures/revenues (rather than impact effects) because frequently the effects of fiscal policy build over time.

The paper is organized as follows. In section II, we present estimates of government spending and tax multipliers using VAR models and conventional identification strategies. Subsequently, in section III we discuss estimates of cumulative multipliers obtained using the local projection method. In Section IV, we explore two attempts to address the problem of identification. We replicate the strategy of Kraay (2014) and use predicted disbursements by official creditors to identify exogenous government spending. On the tax side, we use a narrative approach as in Romer and Romer (2010). Finally, section V presents conclusions and policy implications.

II. Estimating Multipliers with VARs

A. Government Spending Multipliers

We begin by estimating multipliers using VAR models at a quarterly frequency, which has been the econometric approach most commonly used in the literature. We follow the Blanchard & Perotti (2002) identification of expenditure shocks, which assumes that government expenditures are not affected by shocks to GDP within the same quarter (i.e. government expenditures are weakly exogenous). We try to consider separately multipliers for current expenditure and multipliers for capital expenditure.

We estimate a number of three-dimensional VARs with seasonally adjusted data in log levels2 over the period 1998:Q1-2015:Q2 that include real current government expenditure (deflated by the GDP deflator); or real capital expenditure; real GDP; or real non-agro non-energy GDP; the change in the real effective exchange rate; and a deterministic linear trend (included as an exogenous variable). Annex B provides a brief description of the variables used and their respective sources.

Figure 1 presents cumulative orthogonalized impulse responses over 20 quarters for the different VARs estimated. As expected, output responds positively to both types of government expenditure shocks, but confidence bands are wide and impulse responses are only significant at longer horizons. Stability checks presented in Annex C confirm that all the eigenvalues of the companion matrix lie inside the unit circle and therefore the estimated models satisfy the stability condition.

Figure 1.
Figure 1.

Cumulative Orthogonalized Impulse Responses to Government Spending Shocks

Citation: IMF Working Papers 2017, 063; 10.5089/9781475588262.001.A001

Bootstrapped standard errors (500 repetitions). Shaded area represents 95 percent confidence interval. Sample period 1998:Q1-2015:Q2. Lag length selection based on AIC, SBC, and LR test. Higher order VAR used when conflicting results were obtained.

Table 1 presents estimates of the multiplier built from the impulse responses of the different models. More specifically, the elasticities obtained from the VARs in log levels were converted to changes in Guaranies by dividing by the average ratio of current or capital expenditures to GDP over the sample, such that ΔYΔG=Δ⁢ln⁡(Y)Δ⁢ln⁡(G)*YG.

Table 1.

Estimates of Government Spending Multipliers in Paraguay (Cumulative)

article image
Note: ** indicates statistical significance at the 5 percent level. * indicates significance at the 10 percent level.

Nevertheless, as discussed by Ramey & Zubairy (2015) multipliers calculated through this transformation are subject to mismeasurement and biases (most notably related to the fact that the ratios of expenditure to GDP tend to change over the sample). The estimated impact multipliers are close to zero in magnitude and are not statistically significant in all cases i.e. when considering total GDP, non-agro non-energy GDP, current expenditure and capital expenditure.

These results are in line with the ones obtained by Ilzetzki, Mendoza, & Vegh (2013) for government consumption in developing countries, but not for capital expenditure, as these authors find a multiplier of around 0.6 on impact using a panel VAR approach (see Annex A). Estevao & Samake (2013) also present estimates of multipliers for government expenditures of around 0 on impact for a number of Central American countries, whereas Matheson & Pereira (2016) using a similar approach to the one described in this paper obtain an expenditure multiplier of 0.5 on impact for Brazil.

Over longer horizons, estimates of spending multipliers in Paraguay increase to between 0.2 to 0.5 for current expenditure and between 0.5 to 2.1 for capital expenditure, although statistical significance varies depending on whether we consider total or non-agriculture non-energy GDP. The cumulative multipliers obtained for capital expenditure are in the same order of magnitude of the estimates presented by Ilzetzki, Mendoza & Vegh (2013) for developing countries, but are somewhat higher than the ones obtained by Estevao & Samake (2013) for Central America and Vtyurina & Leal (2016) for Peru. The results obtained for Paraguay contrast with the cumulative multiplier of around 0 for capital expenditure estimated by Matheson & Pereira (2016) for Brazil.

To check the robustness of the results obtained, we also estimate VAR models using more granular quarterly fiscal data from the Ministry of Finance following the GFS 2001 presentation that allows us analyze current primary expenditure (excluding interest payments) as well as capital expenditure. This data is available for the period 2003:Q1-2016:Q1. Re-estimating the models over this period is also useful because of the possibility of a structural break in GDP in 2003 (see Colman, Franco, and Tello, 2016 for a discussion and Annex B for a depiction of the GDP series). The estimated impulse responses and multipliers are presented in Annex D. While impact multipliers continue to be estimated at around zero, the point estimates over longer horizons are generally lower than the ones previously reported. Multipliers are estimated to be around 0.1 for real current primary expenditure and between 0.4 to 0.9. for capital expenditure.

B. Tax Multipliers

We now turn our attention to the estimation of tax multipliers by including different measures of tax revenue shocks in the VARs previously described. Riera-Crichton, Vegh, & Vuletin, (2016) discuss the pitfalls in identification and measurement of tax shocks and advocate the use of “narrative” approaches (for further discussion of the narrative approach see section IV of this paper and Romer and Romer, 2010 for an application to the United States) rather than imposing restrictions in VAR models. As far as the measurement of tax policy is concerned, these authors favor the use of statutory tax rates over revenue-based measures.

We follow two different strategies within the VAR framework in terms of identification assumptions in this section. For models that include cyclically-adjusted revenues3, we assume that tax policy does not respond contemporaneously (within the same quarter) to output shocks. Nevertheless, in models where unadjusted revenues are included, we allow for a contemporaneous response to GDP shocks in order to capture automatic stabilizers.

Figure 2 presents the impulse responses obtained. The models include four endogenous variables: real total government expenditures; real GDP (or real non-agriculture non-energy GDP); real tax revenues (or the change in cyclically adjusted tax revenues); and changes in the real effective exchange rate. A deterministic linear time trend is included as an exogenous variable. As in the previous section, stability checks show that the eigenvalues of the companion matrix for the models estimated lie inside the unit circle, satisfying the stability condition (Annex C).

Figure 2.
Figure 2.

Paraguay: Cumulative Orthogonalized Impulse Responses to Tax Shocks

Citation: IMF Working Papers 2017, 063; 10.5089/9781475588262.001.A001

Bootstrapped standard errors (500 repetitions). Shaded area represents 95 percent confidence interval. Sample period 1998:Q1-2015:Q2. Lag length selection based on AIC, SBC, and LR test. Higher order VAR used when conflicting results were obtained.

The results point to a negative response of output to tax shocks (as expected) in most cases with the exception of the bottom left panel, although confidence bands are wide. The multipliers implied by these impulse response functions are close to zero in magnitude and are not statistically significant in most cases, as shown in Table 2. The notable exception is the cumulative multiplier of cyclically adjusted tax revenues on non-agro non-energy GDP, which is estimated at -0.4 after 5 years (-0.2 after two years) and is significant at conventional levels. The finding that tax multipliers are larger (in magnitude and statistical significance) for non-agro non-energy GDP is not surprising, given the virtual absence of taxation in the agricultural sector up to recent periods and the relatively large share of this sector in total GDP.

Table 2.

Estimates of Tax Multipliers in Paraguay (Cumulative)

article image
Note: ** indicates statistical significance at the 5 percent level. * indicates significance at the 10 percent level.

In addition, the results obtained differ from Gunter, Riera-Crichton, Vegh & Vuletin (2016) and Matheson & Pereira (2016), who find cumulative multipliers of around -2 for developing countries and Brazil, respectively. While Matheson & Pereira (2016) use a similar approach to the one presented here, Gunter, Riera-Crichton, Vegh & Vuletin (2016) identify policy shocks through the narrative approach focusing on changes in VAT rates. Colman, Franco, & Tello (2016) using a structural VAR approach also find tax multipliers in Paraguay that are close to zero.

As in the previous section, we also estimate the models using quarterly fiscal data from the Ministry of Finance over the period the period 2003:Q1-2016:Q1 for robustness. The results are presented in Annex D. Estimates of the tax multiplier over this period are also close to zero.

III. Estimating Multipliers with the Local Projection Method

In this section, we use Jorda’s (2005) local projection (LP) method as an alternative econometric approach to obtain estimates of multipliers for Paraguay. The method entails the estimation of a series of regressions for each horizon h, such that:

Zt+h=αh+Σi=1nφihXt−i+βh⁢ shockt+ɛt+h, for  h=0,1,2,…

Where z is the variable of interest, x is a vector of control variables (which includes in our case lags of GDP, lags of government spending, and changes in the real effective exchange rate) and “shock” is the identified shock to government spending. The coefficient βh is the response of z at horizon t+h to the shock at time t. This method can be adapted to estimate non-linear state-dependent models (Ramey & Zubairy, 2015 and 2016; Dell’Erba, Koloskova, & Poplawski-Ribeiro, 2014).

One of the advantages of this approach is that it does not constrain the shape of the impulse response functions and is therefore less sensitive to misspecification than standard VAR models. Nevertheless, estimates also tend to be more erratic than the ones obtained from VARs. In addition, for state-dependent models one does not need to make assumptions about transitions from different states and feedback of shocks to states, as the estimates already incorporate transitions that occur on average in the data (Ramey & Zubairy, 2015).

Furthermore, in the LP method, multipliers can be obtained directly, rather than calculated from elasticities. Ramey & Zubairy (2016) argue that in a dynamic setting, the relevant multipliers that are of direct interest to policy makers should be calculated as the cumulative response of GDP divided by the cumulative government spending over a given horizon, rather than considering peak responses or average responses relative to an initial policy shock, because often the effects of government spending build-up over time. This “integral” multiplier can be estimated in a simple way under the LP method by focusing on the following instrumental variable regression (with the fiscal shock serving as an instrument for Σj=0hgt+j):

Σj=0hYt+j=αh+φh(L)Xt−1+mhΣj=0hgt+j+ωt+h,for h=0,1,2,…

In Figure 3 we present estimates of cumulative spending multipliers using the LP method and the Blanchard-Perotti identification. In this case, the shock is given by the residuals of a regression of real government expenditures on lagged expenditures; lagged real GDP and lagged changes in the real effective exchange rate (akin to innovations in a VAR model). Ninety-five percent heteroscedasticity and autocorrelation consistent (HAC) standard errors bands are reported because of the serial correlation of error terms introduced by the use of successive leads of the dependent variable. GDP and government expenditures were normalized by trend GDP (as in Ramey & Zubairy, 2016), where the trend was obtained using the HP filter. The top two panels present estimates for multipliers on total GDP and the bottom panels estimates for multipliers on non-agro non-energy GDP.

Figure 3.
Figure 3.

Cumulative Spending Multipliers using the LP Method

Citation: IMF Working Papers 2017, 063; 10.5089/9781475588262.001.A001

Shaded area represents 95 percent confidence interval (HAC standard error bands). Sample period 1998:Q1-2015:Q2. Multipliers estimated correspond to the cumulative response of GDP divided by the cumulative government spending over a given horizon.

We can confirm some of the previous evidence from VAR models that multipliers for capital expenditure appear to be higher than multipliers for current expenditure. In general, multipliers seem to be more precisely estimated (smaller standard errors) under this approach. While the estimated multiplier for capital expenditure on total GDP is zero on impact, it reaches 2 after four quarters and is statistically significant at the 5 percent level, a result that is broadly in line with VAR models.

In the case of current expenditure, the cumulative multiplier for total GDP only becomes statistically significant after seven quarters and reaches a value of 0.9, which is substantially larger than what was obtained using VAR models. Multipliers for current and capital expenditures on non-agro non-energy GDP (bottom panels) are of similar size to the ones obtained for total GDP under this approach and contrary to VAR models, multipliers for capital expenditure on non-agro non-energy GDP are statistically significant.

In Annex E, we undertake some robustness checks around the baseline specification. More specifically, we consider two alternative model specifications: one in which the real effective exchange rate is removed from the regressions and another where the real effective exchange rate is included in levels rather than first differences. The main results still hold, with medium-term multipliers being substantially higher for capital expenditure relative to current expenditure. Moreover, impact multipliers continue to be statistically insignificant at the five percent level.

Overall, the results from both LP and VAR models point to higher multipliers for capital expenditure relative to current expenditure. One characteristic that might help explain this differential is that capital expenditure is to a large extent financed by external borrowing in Paraguay, therefore mitigating crowding-out effects. Interestingly, we also find that, from a cyclical perspective, the correlation between the cyclical component of capital expenditure and the cyclical component of output (obtained through the HP filter) is negative and statistically significant at around -0.3 for the period 2003-2015 (Figure 4). This suggests that capital expenditure has behaved in a countercyclical manner. Nevertheless, the negative correlation discussed previously is not statistically significant for the entire sample period (1998-2015). These results are in line with the conclusions of Correa, Colman, and Tello (2016), who find that fiscal policy has been countercyclical in the more recent period.

Figure 4.
Figure 4.

Correlation between cyclical component of output and capital expenditure

Citation: IMF Working Papers 2017, 063; 10.5089/9781475588262.001.A001

Furthermore, we also use the LP method to obtain estimates of tax multipliers (Figure 5). As in the previous section, we consider two types of tax shocks. The first shock is built from the residuals from a regression of real tax revenue on lagged tax revenue; contemporaneous and lagged government expenditures; contemporaneous and lagged GDP; lagged changes in the real effective exchange rate; and a deterministic trend. The second shock follows a similar approach but uses cyclically adjusted revenues rather than real tax revenues. Similarly, to the case of government expenditures, multipliers are calculated as the cumulative response of GDP divided by the cumulative tax revenue over a given horizon.

Figure 5.
Figure 5.

Cumulative Tax Multipliers using the LP Method

Citation: IMF Working Papers 2017, 063; 10.5089/9781475588262.001.A001

Shaded area represents 95 percent confidence interval (HAC standard error bands). Sample period 1998:Q1-2015:Q2. Multipliers estimated correspond to the cumulative response of GDP divided by the cumulative tax revenue over a given horizon.

In the case of Paraguay, our estimated multipliers are not statistically significant at conventional levels in all cases. This contrasts with the results from VAR models for non-agriculture and non-energy GDP, where we found negative and significant multipliers at longer horizons, but is in line with VAR results for total GDP. Ramey (2016) presents estimates using the LP method for the US that point to tax multipliers between -2 and -3, following a variety of identification approaches, including narrative methods. Moreover, the robustness checks presented in Annex E confirm the finding that tax multipliers in Paraguay are not significant both from an economic and statistical perspective.

IV. Further Attempts to Address the Identification Problem

In order to obtain reliable estimates of fiscal multipliers it is crucial that the measures of tax and spending shocks used truly reflect discretionary policy changes that are unrelated to output fluctuations. So far in the paper, spending and tax shocks have essentially been identified through timing restrictions and exogenous assumptions. We try to further address the identification issue firstly by replicating the strategy of Kraay (2014) and using predicted disbursements by official creditors to identify exogenous government spending shocks, obtaining estimates of multipliers using an instrumental variable approach. Subsequently, we focus on the tax side and use a narrative approach as in Romer and Romer (2010) to identify exogenous tax shocks.

A. Identifying Spending Shocks through Disbursements by Official Creditors

In this sub-section, we replicate the approach proposed by Kraay (2014) to empirically estimate government spending multipliers. Kraay uses predicted disbursements by official creditors (multilateral banks and bilateral development agencies) to isolate the source of variation in government spending that is uncorrelated with contemporaneous macroeconomic shocks. More specifically, this author uses the timing between project approval and eventual disbursements to obtain the predetermined component of public spending in developing countries (the sample comprises 102 countries in the period 1970-2010 in the original paper). This identification strategy would work well if official creditors are a significant source of financing for government expenditures and if loan disbursements follow a schedule specified at the time of loan approval that is uncorrelated with subsequent macroeconomic shocks.

Kraay’s dataset contains annual data for Paraguay covering the period 1972-2010. Disbursements by official creditors represent 13.7 percent of total government spending in Paraguay over the period compared to an average of 13.4 percent for the entire sample of countries. In Table 3, we replicate the baseline regressions presented in Kraay (2014). OLS and Two-Stage Least Squares point estimates of the multiplier range from 0.2 to 0.6, but none of the coefficients are statistically significant at conventional levels, even after weak-instrument-consistent confidence intervals are constructed. In fact, the first stage regressions (specifications 2 and 3) point to a weak link between predicted disbursements and government expenditures in Paraguay.

Table 3.

Spending Multipliers using Kraay’s (2014) Identification Strategy

article image
Heteroskedasticity-consistent standard errors in parentheses. * p< 0.10, ** p<0.05, *** p<0.01

B. Identifying Tax Shocks in Paraguay through the Narrative Approach

As discussed in previous sections, Riera-Crichton, Vegh, & Vuletin, (2016) favor the use of statutory tax rates as a measure of tax policy. Nevertheless, the estimation of multipliers identified through changes in statutory rates is likely to be difficult in the case of Paraguay because of the small number of instances of changes in tax rates over the sample period. In fact, Vegh and Vuletin (2015) only include four changes in tax rates in Paraguay in their dataset covering the period 1979-2013: the introduction of the VAT in 1991, a reduction in corporate taxes from 30 to 20 percent in 2005 and from 20 to 10 percent in 2006, and the introduction of the personal income tax of 10 percent in 2007.4 In addition, statutory tax rates might not be a good proxy for tax policy for a number of reasons, such as changes in coverage over time (for example, the elimination of VAT exemptions or reduced rates for certain products), tax administrations issues, among other concerns.

Therefore, we opted to follow an alternative strategy in this section and adopt the methodology proposed by Romer and Romer (2010) to create a “narrative” measure of exogenous fiscal shocks based on the estimated impact of policy changes. Annex F presents a detailed description of the shocks identified and the relevant sources (mainly IMF staff reports on Paraguay). Following Romer and Romer (ibid.) we consider four main types of motivation for fiscal policy changes. “Spending Driven” tax changes are motivated by a change in government spending (within the same year), a typical example would be a tax increase because the country is fighting a war. These are typically endogenous to output and therefore are not included in the regressions. “Countercyclical Action” are tax changes designed to return output growth to normal (close the output gap) and are by definition endogenous to the cycle. “Deficit Driven” tax changes refer essentially to tax increases designed to reduce an inherited budget deficit. These are discretionary fiscal policy changes not linked to the cycle and are included as exogenous shocks. Finally, there can also be exogenous tax changes motivated by a desire to raise long-run growth (potential output).

As far as the timing of the shocks is concerned, we consider the effects of tax changes at the time of implementation as in Romer and Romer (2010). We follow the convention that if a tax change occurs before the mid-point of the quarter, it is assigned to that quarter, otherwise it is assigned to next quarter. Therefore, one important aspect to consider is that we are measuring the effect of “anticipated” tax changes, which can differ significantly from the effect of unanticipated changes both empirically and in theory, as discussed in Ramey (2016). We are able to identify 11 instances of “tax shocks” in this way over the period 2001-2015 with most shocks being positive ones (aiming at increasing revenues). The size of the tax shocks is based on estimates of the revenue impact of the given tax policy measure at the time of implementation (expressed in annual terms) and at the prevailing level of GDP. This measure is then converted to local currency units in real terms (deflated by the GDP deflator) as appropriate for the regressions.

Figure 6 depicts estimates of tax multipliers obtained under the narrative identification strategy at different horizons using the LP method for total GDP and for non-agro non-energy GDP. In both cases, the point estimates obtained are much higher than the ones found under the alternative identification approaches. Impact multipliers are -0.1 and -0.2 for total and non-agro non-energy GDP, respectively, although only the latter is (marginally) significantly different from zero. Over the medium-term cumulative multipliers are in the order of -2, which is in line with the estimates of Gunter, Riera-Crichton, Vegh & Vuletin (2016) using the narrative approach for a larger set of countries and also similar to the results discussed in Ramey (2016) for advanced economies.

Figure 6.
Figure 6.

Multipliers using Narrative Tax Shocks and the LP Method

Citation: IMF Working Papers 2017, 063; 10.5089/9781475588262.001.A001

Shaded area represents 95 percent confidence interval (HAC standard error bands). Sample period 1998:Q1-2015:Q2.

Nevertheless, these results on tax multipliers should be taken with caution. In fact, when including the narrative tax shock in conventional VAR models, the impulse responses point to multipliers that are not statistically significant, undermining the robustness of the estimates obtained under the LP method (Figure 7).

Figure 7.
Figure 7.

Cumulative Orthogonalized Impulse Responses to Narrative Tax Shocks

Citation: IMF Working Papers 2017, 063; 10.5089/9781475588262.001.A001

Bootstrapped standard errors (500 repetitions). Shaded area represents 95 percent confidence interval. Sample period 1998:Q1-2015:Q2. Lag length selection based on AIC, SBC, and LR test. Higher order VAR used when conflicting results were obtained.

V. Conclusions and Policy Implications

Our results point to multipliers for current expenditure that are substantially lower than multipliers for capital expenditure in Paraguay. Medium-term multipliers for capital expenditure range from 0.4 to 2, whilst multipliers for current expenditure range from 0.1 to 0.9. One characteristic that might help explain this differences is that capital expenditure is to a large extent financed by external borrowing in Paraguay, therefore mitigating crowding-out effects. These differences and the fact that estimated multipliers for current expenditure are frequently close to zero in magnitude support recent IMF policy advice to contain growth in current primary expenditure while preserving capital expenditure.

In addition, when conventional identification approaches are used, the results suggest that tax multipliers are low relative to other countries and seem to affect non-agro non-energy GDP more prominently. Estimates of tax multipliers range from 0 to -0.4 over the medium-term. Nevertheless, when we consider exogenous tax shocks under the narrative approach, estimates of multipliers can reach -2, which is more in line with the international evidence (Gunter, Riera-Crichton, Vegh & Vuletin, 2016). Nevertheless, this larger estimate does not seem to be robust to different econometric estimation techniques.

Overall, the challenge to identify fiscal shocks in a satisfactory way remains open and therefore the estimates of expenditure and tax multipliers presented in this paper should be taken with caution. Bearing these caveats in mind, one implication of the results seems to be that the “balanced budget” multiplier for Paraguay i.e. the effect of on output of an increase in expenditures (in particular capital expenditure) financed by taxes is likely to be positive. We also discussed evidence indicating that capital expenditure behaved in a countercyclical manner in Paraguay, particularly over the 2003-2015 period and might have been used effectively as a stabilization tool with the execution of expenditures being accelerated during downturns and restrained during the expansionary phases of the cycle.

Annex A. Selected Recent Studies on Fiscal Multipliers

article image
article image

Annex B. Variable Definitions and Sources

article image
A01ufig1

Fiscal Variables Normalized by Trend GDP

Citation: IMF Working Papers 2017, 063; 10.5089/9781475588262.001.A001

A01ufig2

Real GDP, Non-Agro Non-Energy GDP and Real Effective Exchange Rate

Citation: IMF Working Papers 2017, 063; 10.5089/9781475588262.001.A001

Annex C. VAR Stability Checks

A01ufig3

Stability Checks for VARs used for Government Spending Multipliers

Citation: IMF Working Papers 2017, 063; 10.5089/9781475588262.001.A001

A01ufig4

Stability Checks for VARs used for Tax Multipliers

Citation: IMF Working Papers 2017, 063; 10.5089/9781475588262.001.A001

Annex D. Additional Results: Alternative Definition of Fiscal Variables

A01ufig5

Annex Figure. Cumulative Orthogonalized Impulse Responses to Government Spending Shocks

Citation: IMF Working Papers 2017, 063; 10.5089/9781475588262.001.A001

Bootstrapped standard errors (500 repetitions). Shaded area represents 95 percent confidence interval. Sample period 2003:Q1-2015:Q4. Lag length selection based on AIC, SBC, and LR test. Higher order VAR used when conflicting results were obtained.
Annex Table.

Estimates of Government Spending Multipliers in Paraguay (Cumulative)

article image
Note: ** indicates statistical significance at the 5 percent level. * indicates significance at the 10 percent level.
A01ufig6

Annex Figure. Paraguay: Cumulative Orthogonalized Impulse Responses to Tax Shocks

Citation: IMF Working Papers 2017, 063; 10.5089/9781475588262.001.A001

Bootstrapped standard errors (500 repetitions). Shaded area represents 95 percent confidence interval. Sample period 2003:Q1-2015:Q4. Lag length selection based on AIC, SBC, and LR test. Higher order VAR used when conflicting results were obtained.

Annex Table. Estimates of Tax Multipliers in Paraguay (Cumulative)

article image
Note: ** indicates statistical significance at the 5 percent level. * indicates significance at the 10 percent level.

Annex E. Additional Results: Variations in Model Specification

A01ufig7

Annex Figure. Cumulative Multipliers using the LP Method (Real Exchange Rate Excluded)

Citation: IMF Working Papers 2017, 063; 10.5089/9781475588262.001.A001

Shaded area represents 95 percent confidence interval (HAC standard error bands). Sample period 1998:Q1-2015:Q2. Multipliers estimated correspond to the cumulative response of GDP divided by the cumulative government expenditure or tax revenue over a given horizon.
A01ufig8

Annex Figure. Cumulative Multipliers using the LP Method (Real Exchange Rate in Levels)

Citation: IMF Working Papers 2017, 063; 10.5089/9781475588262.001.A001

Shaded area represents 95 percent confidence interval (HAC standard error bands). Sample period 1998:Q1-2015:Q2. Multipliers estimated correspond to the cumulative response of GDP divided by the cumulative government expenditure or tax revenue over a given horizon.

Annex F. Narrative Exogenous Tax Shocks

article image
DD= A deficit-driven tax change is a tax increase designed to reduce an inherited budget deficit. Exogenous to output. LR =A long-run tax change is one aimed at raising long-run growth. Exogenous to cyclical considerations. Methodology to identify tax shocks follows Romer and Romer (2010).

VI. References

  • Auerbach, A. and Gorodnichenko, Y. (2012) “Measuring the Output Responses to Fiscal Policy” American Economic Journal: Economic Policy 4:1–27.

    • Search Google Scholar
    • Export Citation

  • Batini, N., Eyraud, L., Forni, L., and Weber, A. (2014) “Fiscal Multipliers: Size, Determinants, and Use in Macroeconomic Projections” IMF Fiscal Affairs Department Technical Notes and Manuals 14/04 (Washington: International Monetary Fund).

    • Search Google Scholar
    • Export Citation

  • Blanchard, O. and Perotti, R. (2002) “An Empirical Characterization of the Dynamic Effects of Changes in Government Spending and Taxes” The Quarterly Journal of Economics, 117 (4): 1329–1368.

    • Search Google Scholar
    • Export Citation

  • Correa, D., Colman, H., and Tello, J. (2016) “Coordinación de las Políticas Monetaria y Fiscal” in Paraguay: Más Allá de la Estabilidad Macroeconómica. Logros y desafíos. Banco Central de Paraguay, Asunción.

    • Search Google Scholar
    • Export Citation

  • Colman, H., Franco, R. and Tello, J. (2016) “Fiscal Multipliers in Paraguay”, mimeo, Ministerio de Hacienda, Asuncion.

  • David, A. C. & Novta, N. (2016) “A Balancing Act: Reform Options for Paraguay’s Fiscal Responsibility Law” IMF Working Paper WP/16/226 (Washington: International Monetary Fund).

    • Search Google Scholar
    • Export Citation

  • Dell’Erba, S. Koloskova, K. and Poplawski-Ribeiro, M. (2014) “Medium-Term Fiscal Multipliers during Protracted Recessions” IMF Working Paper WP/14/213 (Washington: International Monetary Fund).

    • Search Google Scholar
    • Export Citation

  • Engemann, K. M., Owyang, M. T. and Zubairy, S. (2008) “A Primer on the Empirical Identification of Government Spending Shocks” Federal Reserve Bank of St. Louis Review, 90: 117–132.

    • Search Google Scholar
    • Export Citation

  • Estevao, M. and Samake, I. (2013) “The Economic Effects of Fiscal Consolidation with Debt Feedback” IMF Working Paper WP/13/136 (Washington: International Monetary Fund).

    • Search Google Scholar
    • Export Citation

  • Gunter, S., Riera-Crichton, D., Vegh, C. A., and Vuletin, G. (2016) “Non-linear effects of tax changes on output: A worldwide narrative approach”, mimeo, Johns Hopkins University.

    • Search Google Scholar
    • Export Citation

  • Ilzetzki, E. Mendoza, E. and Vegh, C. A. (2013) “How big(small?) are fiscal multipliers?” Journal of Monetary Economics, 60: 239–254.

    • Search Google Scholar
    • Export Citation

  • Jorda, O. (2005) “Estimation and Inference of Impulse Responses by Local Projections.” American Economic Review, 95: 161–182.

  • Kraay, A. (2014) “Government Spending Multipliers in Developing Countries: Evidence from Lending by Official Creditors” American Economic Journal: Macroeconomics, 6: 170–208.

    • Search Google Scholar
    • Export Citation

  • Matheson, T. and Pereira, J. (2016) “Fiscal Multipliers for Brazil” IMF Working Paper WP/16/79 (Washington: International Monetary Fund).

    • Search Google Scholar
    • Export Citation

  • Ramey, V. (2011) “Identifying Government Spending Shocks: It’s all about the Timing” The Quarterly Journal of Economics, 126 (1): 1–50.

    • Search Google Scholar
    • Export Citation

  • Ramey, V. (2016) “Macroeconomic Shocks and Their Propagation” NBER Working Paper 21978 (Cambridge: National Bureau of Economic Research).

    • Search Google Scholar
    • Export Citation

  • Ramey, V. and Zubairy, S. (2015) “Are Government Spending Multipliers State Dependent? Evidence from Canadian Historical Data”, mimeo, University of California, San Diego.

    • Search Google Scholar
    • Export Citation

  • Ramey, V. and Zubairy, S. (2016) “Government Spending Multipliers in Good Times and in Bad: Evidence from U.S. Historical Data”, mimeo, University of California, San Diego.

    • Search Google Scholar
    • Export Citation

  • Riera-Crichton, D., Vegh, C. A. and Vuletin, G. (2016) “Tax Multipliers: Pitfalls in Measurement and Identification”, Journal of Monetary Economics, 79: 30–48.

    • Search Google Scholar
    • Export Citation

  • Romer, C. and Romer, D. (2010) “The Macroeconomic Effects of Tax Changes: Estimates Based on a New Measure of Fiscal Shocks” American Economic Review, 100: 763–801

    • Search Google Scholar
    • Export Citation

  • Valencia, F. (2016) “Fiscal Multipliers in Mexico” Mexico Selected Issues Paper, IMF Country Report No. 15/314 (Washington: International Monetary Fund).

    • Search Google Scholar
    • Export Citation

  • Vegh, C. A. and Vuletin, G. (2015) “How is Tax Policy Conducted over the Business Cycle?” American Economic Journal: Economic Policy, 7: 327–370.

    • Search Google Scholar
    • Export Citation

  • Vtyurina, S. and Leal, Z. (2016) “Fiscal Multipliers and Institutions in Peru: Getting the Largest Bang for the Sol” IMF Working Paper WP/16/144 (Washington: International Monetary Fund).

    • Search Google Scholar
    • Export Citation
1

I would like to thank Humberto Colman, Hamid Faruqee, Geoffrey Keim, Marika Santoro, and Jose Tello for excellent comments provided to previous drafts of the paper. I benefited from very useful discussions with Valerie Ramey on the calculation of integral multipliers. I am also grateful to Ehab Tawfik for outstanding research assistance. The usual caveats apply.

2

Ramey (2011) discusses the advantages of estimating the models in log levels to calculate multipliers.

3

Defined using the equation below, with an elasticity (η) of 1 for Paraguay following David and Novta (2016). Δadjusted revenue= ln(tax revenuet)-ln(tax revenuet−1)-η*[ln(Yt)-ln(Yt−1)]

4

As shown in Annex F, the personal income tax was only effectively implemented in 2012.

Same Series

Other IMF Content

Other Publishers

Cover IMF Working Papers
Fiscal Policy Effectiveness in a Small Open Economy: Estimates of Tax and Spending Multipliers in Paraguay
Author: Mr. Antonio David
Volume/Issue:
Volume 2017: Issue 063
Publisher:
International Monetary Fund
ISBN:
9781475588262
ISSN:
1018-5941
Pages:
29
DOI:
https://doi.org/10.5089/9781475588262.001
  • View in gallery

    Cumulative Orthogonalized Impulse Responses to Government Spending Shocks

  • View in gallery

    Paraguay: Cumulative Orthogonalized Impulse Responses to Tax Shocks

  • View in gallery

    Cumulative Spending Multipliers using the LP Method

  • View in gallery

    Correlation between cyclical component of output and capital expenditure

  • View in gallery

    Cumulative Tax Multipliers using the LP Method

  • View in gallery

    Multipliers using Narrative Tax Shocks and the LP Method

  • View in gallery

    Cumulative Orthogonalized Impulse Responses to Narrative Tax Shocks

  • View in gallery

    Fiscal Variables Normalized by Trend GDP

  • View in gallery

    Real GDP, Non-Agro Non-Energy GDP and Real Effective Exchange Rate

  • View in gallery

    Stability Checks for VARs used for Government Spending Multipliers

  • View in gallery

    Stability Checks for VARs used for Tax Multipliers

  • View in gallery

    Annex Figure. Cumulative Orthogonalized Impulse Responses to Government Spending Shocks

  • View in gallery

    Annex Figure. Paraguay: Cumulative Orthogonalized Impulse Responses to Tax Shocks

  • View in gallery

    Annex Figure. Cumulative Multipliers using the LP Method (Real Exchange Rate Excluded)

  • View in gallery

    Annex Figure. Cumulative Multipliers using the LP Method (Real Exchange Rate in Levels)

Cumulative Orthogonalized Impulse Responses to Government Spending Shocks

Paraguay: Cumulative Orthogonalized Impulse Responses to Tax Shocks

Cumulative Spending Multipliers using the LP Method

Correlation between cyclical component of output and capital expenditure

Cumulative Tax Multipliers using the LP Method

Multipliers using Narrative Tax Shocks and the LP Method

Cumulative Orthogonalized Impulse Responses to Narrative Tax Shocks

Fiscal Variables Normalized by Trend GDP

Real GDP, Non-Agro Non-Energy GDP and Real Effective Exchange Rate

Stability Checks for VARs used for Government Spending Multipliers

Stability Checks for VARs used for Tax Multipliers

Annex Figure. Cumulative Orthogonalized Impulse Responses to Government Spending Shocks

Annex Figure. Paraguay: Cumulative Orthogonalized Impulse Responses to Tax Shocks

Annex Figure. Cumulative Multipliers using the LP Method (Real Exchange Rate Excluded)

Annex Figure. Cumulative Multipliers using the LP Method (Real Exchange Rate in Levels)