Adjustment through revenue or spending? Fiscal adjustment scenarios based on estimated multiplier effects ^♦

Ajuste via receita ou via gasto? Cenários de ajuste fiscal considerando estimativas de efeitos multiplicadores

Abstract

This paper contributes to the literature on fiscal multipliers by estimating the effects of changes in revenue and in five groups of expenditures: public investment, social benefits, subsidies, personnel expenses, and other expenditures. We find that the strongest effects arise from shocks on public investments and social benefits, with both having a positive correlation with the sign of the shock. Increases in revenues, on the other hand, have statistically significant negative effects only in the short-run. The paper estimates the effects on GDP, primary balance, and public indebtedness of different fiscal consolidation scenarios. Results indicate that it is possible to combine reductions in the public indebtedness-to-GDP ratio with increases in GDP with a rebalancing of fiscal policy towards public investment and social spending and away from subsidies.

Keywords:
Fiscal multipliers; Fiscal consolidation; SVAR; Fiscal policy; Public debt

Resumo

Este artigo contribui para a literatura sobre multiplicadores fiscais ao estimar os efeitos de mudanças na receita e em cinco grupos de despesas: investimento público, benefícios sociais, subsídios, despesas com pessoal e outras despesas. Constatamos que os efeitos mais fortes decorrem de choques nos investimentos públicos e nos benefícios sociais, ambos apresentando uma correlação positiva com o sinal do choque. Aumentos na receita, por outro lado, têm efeitos negativos estatisticamente significativos apenas no curto prazo. O artigo estima os efeitos sobre o PIB, saldo primário e endividamento público em diferentes cenários de consolidação fiscal. Os resultados indicam que é possível combinar reduções na razão dívida pública/PIB com aumentos no PIB através de um reequilíbrio da política fiscal em direção ao investimento público e aos gastos sociais, e afastando-se dos subsídios.

Palavras-chave:
Multiplicadores fiscais; Consolidação fiscal; SVAR; Política fiscal; Dívida pública

1. Introduction

In recent years, a succession of economic challenges - including the 2007 financial crisis, the Eurozone debt crisis, and the Covid-19 pandemic - alongside broader theoretical debates such as “secular stagnation” and the zero lower bound for monetary policy, has led to a renewed emphasis on the role of active fiscal policy. As highlighted by Ramey (2019), this shift has coincided with a broader resurgence in fiscal research.

A significant strand of this literature focuses on estimating the effects of fiscal consolidations on GDP. Alesina and Ardagna (2010) advanced the controversial claim that negative fiscal shocks can have positive effects on output, igniting the debate over “expansionary austerity.” However, more recent studies tend to converge on the view that fiscal consolidations, particularly in the short run, have negative effects on GDP (Ramey, 2019).

The literature has also become more nuanced, emphasizing the distinction between consolidations via tax increases and those via spending cuts. Theoretically, these measures affect output through different channels. In a standard Keynesian cross model, tax multipliers are smaller because taxes reduce disposable income, while government spending directly stimulates demand through the multiplier effect. In neoclassical models, tax changes work via supply-side incentives, while increases in public spending may raise output through a negative wealth effect that encourages higher labor supply.

Empirical estimates reflect this complexity. Government spending multipliers typically range from 0.6 to 1, though they tend to be larger when monetary and fiscal policy are aligned - for example, during periods of near-zero interest rates or wartime (Blanchard and Perotti, 2002; Auerbach and Gorodnichenko, 2012; Farhi and Werning, 2016). In contrast, estimates of tax multipliers vary widely depending on the empirical approach. Time-series estimates tend to find large negative multipliers, around -2 (e.g., Romer and Romer 2010), while calibrated New Keynesian DSGE models usually report smaller multipliers, around -1 (e.g., Zubairy 2014).

In the Brazilian context, several studies have estimated fiscal multipliers using VAR-based models. Peres (2006), Peres and Ellery (2009), Matheson and Pereira (2016), and Restrepo (2020), using SVAR methods, find relatively small spending multipliers - ranging from 0.3 to 0.81 - and negative revenue multipliers, with Matheson and Pereira (2016) estimating a cumulative effect of -2. Other studies, such as Pires (2014), Resende (2019), and Sanches and Carvalho (2022), highlight heterogeneity across spending types, finding larger multipliers for public investment (1.4 to 3.6) and social benefits (up to 4.37).

Results are also mixed regarding how multipliers behave over the business cycle. Orair et al. (2016), using a STVAR approach, and Alves (2017), using local projections, report conflicting findings. While Orair et al. (2016) identify larger multipliers during recessions - particularly for social benefits, which reach up to 8 - Alves (2017) finds no significant variation across the cycle. Similarly, Grudtner and Aragon (2017) find no clear pattern between expansion and recession periods.

This paper contributes to the literature in two main ways. First, it updates Brazil’s fiscal multiplier estimates for different types of spending and revenue using a newly constructed dataset for the federal government’s fiscal accounts from 1997 to 2023. Following the methodology proposed by Gobetti and Orair (2017), we correct the official fiscal series for accounting maneuvers and inconsistencies in revenue and expenditure classifications. This results in a more accurate and transparent representation of Brazil’s public finances. We focus on five key expenditure categories - public investment, social benefits, subsidies, personnel, and other spending - alongside aggregate revenue. These corrections allow for improved multiplier estimates and better policy evaluation.

Using this adjusted dataset, we estimate fiscal multipliers with a structural vector autoregressive (SVAR) model. We follow Lenza and Primiceri (2020) in adjusting for outliers during the Covid-19 years. Our estimates show that increases in public spending, particularly in investment and social benefits, have strong and persistent positive effects on GDP. In contrast, tax increases lead to short-term declines in output, though these effects dissipate over time. After 25 months, each real spent on public investment or social benefits increases GDP by R$ 2.6 and R$ 2.15, respectively.

The second contribution is a simulation exercise. We apply the estimated multipliers to simulate the macroeconomic effects of different fiscal adjustment strategies. These simulations evaluate the impact of changes in fiscal variables on GDP, the primary deficit-to-GDP ratio, and the public debt-to-GDP ratio. A first set of simulations examines spending-based adjustments targeting subsidies, investments, and social benefits separately. A fourth scenario analyzes a revenue-based adjustment with constant spending. Finally, three additional scenarios combine revenue increases with expanded expenditure.

Our main finding is that fiscal adjustments combining revenue increases with higher spending on investment and social benefits can lead to a net positive effect on output and improved debt sustainability. These mixed strategies yield the lowest debt-to-GDP ratio among the simulated cases. By contrast, cuts to subsidies - which have low multipliers - can help improve the primary balance with minimal cost to output. Therefore, the most effective strategy for Brazil involves reducing unproductive spending (like subsidies) and increasing both revenues and growth-enhancing expenditures (like investment and social benefits), potentially improving both macroeconomic performance and fiscal sustainability.

The paper is organized as follows. Section 2 presents the Brazilian fiscal dataset and the methodology used to correct official fiscal accounts. Section 3 details the econometric methodology for estimating fiscal multipliers. Section 4 presents the empirical results. Section 5 uses these results to simulate alternative fiscal adjustment scenarios. Section 6 concludes.

2. Brazil’s fiscal policy: broad patterns and developments

We employed the methods of Gobetti and Orair (2017) to adjust the fiscal series related to public revenue and public spending. Overall, this approach allowed for the 1) correction of the fiscal series for fiscal and accounting maneuvers (such as Petrobras’s capitalization in 2010 and inflated entries related to the FGTS supplement, sovereign fund, and PSI program), and 2) the resolution of inconsistencies regarding the composition and amount of revenues and expenditures. Several spending and revenue lines were corrected or extended using alternative data sources, especially for periods prior to 2007 or involving misrecorded items like BPC, Bolsa Família, and subsidies. These adjustments ensure consistency over time and more accurate primary balance estimates. More details on the adjustment can be found in the Appendix.¹

Figure 1
Difference between the adjusted series and the original series for the primary balance of the government (in R$ million)

In the graph above, the years with the most relevant changes in the government primary balance due to the adjustments made are 2010 (a 44.5% decrease in surplus), and 2015 (a 43.5% decrease in deficit).

For 2010, the discrepancy is mostly explained by the removal of the effects of creative accounting identified during the onerous assignment and capitalization of Petrobras in September 2010. As explained in previous sections, this accounting maneuver inflated public revenue by R$ 72.8 billion and public spending by R$ 42.9 billion, resulting in an overstatement of the government’s primary revenue by R$ 31.9 billion. The difference between the government’s primary balance using original data and the adjusted data for 2010 amounts to R$ 35.0 billion, 91% of which is attributed to the values related to the onerous assignment and capitalization of Petrobras.

For 2015, the adjusted primary deficit decreased by R$ 50.7 billion, driven largely by revisions to public spending values. As previously noted, spending under the ‘Programa de Sustentação de Investimento (PSI)’ (Investment Support Program) was reduced by R$ 14.5 billion, based on the data from the Central Bank of Brazil (BCB) compiled by Gobetti and Orair (2017). Relevant adjustments were also made in the line ‘III.3.20.1.19. Equalização de Custeio Agropecuário’ (Agricultural Financing Equalization), where spending was reduced by R$ 8.4 billion, and the FIES program, which saw a reduction of R$ 5.8 billion. The adjustments related to the payroll tax exemptions were neutral, as equal amounts were subtracted from both revenues and expenditures. The line referring to other discretionary spending was adjusted by R$ 10.1 billion.

During the period of interest, the Brazilian economy experienced relatively strong growth between 1997 and 2013, with an average annual GDP growth rate of 3.1%. This was followed by a severe economic crisis, with GDP contracting by an average of 1.1% per year between 2014 and 2020-a period that also includes the COVID-19 pandemic, which negatively impacted economies worldwide. Since then, Brazil has seen a moderate recovery, with GDP growing at an average annual rate of 3.5% between 2021 and 2023.

Regarding fiscal performance, the central government maintained primary surpluses from 1997 to 2014. However, it has recorded primary deficits since then, largely due to a continuous increase in expenditures - albeit at a slower pace since 2016 - and a marked deceleration in revenue growth since 2012. This decline in revenues reflects not only weaker GDP growth but also an increase in subsidies.

In Figure 2 below, we summarize the main trends for the key areas we analyze in this study, namely public investment, social benefits, subsidies, personnel expenses, and other expenditures, across five subperiods between 1997 and 2023 using the new dataset. We observe a vigorous expansion of public investment between 2006 and 2010. After that, for the period between 2011 and 2014, spending on subsidies grew at the highest annual rate for the entire time span. Fiscal consolidation attempts in the period between 2015 and 2016 led to decreases in both public investment and subsidy spending. With the exception of the period between 1997 and 2005, primary revenue grew at lower annual rates than primary spending over the entire time span. In the most recent period (2020-2023), spending on social benefits grew at the highest annual rate, followed by subsidy and investment spending.

Figure 2
Annual variation rates of central government spending and revenue components (% per year)

3. Methodology

We employ an SVAR approach based on Perotti (2007) and Blanchard and Perotti (2002) to model contemporary relationships. The SVAR methodology gained prominence in the literature on fiscal multipliers through Blanchard and Perotti (2002), who argue that it is suitable for fiscal policy due to the decision and implementation lags of budgetary policies. With high-frequency data (monthly or quarterly), there is minimal or no immediate fiscal policy response to unexpected output shocks, given that policymakers take more than a quarter (or a month) to perceive the output shock, decide on the next steps in fiscal policy, and present them to the legislature. In line with this literature, our identification approach aims to isolate exogenous shocks and recover the structural form of the shocks by obtaining a non-recursive orthogonalization of the error terms. In this sense, the SVAR methodology is equivalent to a VAR methodology in which the expenditures are ordered first. However, as illustrated below, the SVAR approach allows the possibility of considering possible contemporaneous responses of the revenue to GDP.

First, the VAR is estimated in reduced form. The vector of endogenous variables is three-dimensional, including time series of primary expenditures (and each component for each estimated model), primary tax revenues and output. It is a VAR model, as proposed by Sims (1980), where each variable is explained by lags of itself and the other variables of the model, capturing dynamic relationships.

According to Perotti (2007), shocks of the reduced form (or ‘surprise’ movements) can be seen as linear combinations of three components: a) the automatic response of government spending and revenue to changes in output; b) the discretionary response due to changes in endogenous variables (Perotti gives the example of tax changes in response to a recession); c) random discretionary shocks, that is, structural shocks, which are uncorrelated and unobservable - the ones that need to be recovered. Formally:

$u_t^g={\alpha }_{gy}{u}_t^y+{\beta }_{gt}{e}_t^t+e_t^g$ (1)

$u_t^t={\alpha }_{ty}{u}_t^y+{\beta }_{tg}{e}_t^g+e_t^t$ (2)

$u_t^y={\gamma }_{yt}{u}_t^t+{\gamma }_{yg}{u}_t^g+e_t^y$ (3)

The unexpected movements in the expenditure, revenue, and output variables are, respectively, denoted by $u_t^g$, $u_t^t$, and $u_t^y$. These ‘surprise’ movements are the residuals in the reduced form, as it is the part of the data that the VAR does not explain. Also, $e_t^g$, $e_t^t$, and $e_t^y$ are the structural shocks that are not correlated with each other by assumption and reflect the part of the surprise movements that is exogenous: it does not depend on policies and ‘normal’ economic evolution (Coudret, 2013). The coefficients ${\alpha }_{ij}$ reflect the response of variable $i$ to variable $j$ - the components (a) and (b) listed above are captured by the coefficients α (Jemec et al., 2013). While ${\beta }_{ij}$ measures the contemporaneous response of variable $i$ to a structural shock in variable $j$ - that is, component (c) (Perotti, 2007).

As discussed by Vdovychenko (2018), coefficients ${\alpha }_{gy}$, ${\alpha }_{ty}$, ${\gamma }_{yt}$ and ${\gamma }_{yg}$ cannot be estimated without bias due to the instantaneous mutual relationship between output, expenditures, and revenues. Two steps are necessary to solve this. First, as it is plausible to assume that discretionary fiscal responses to an output shock take longer than a quarter to be decided upon and implemented (Perotti, 2007), component (b) is removed, and coefficients α are made to reflect only the first component - the response of the automatic stabilizer. Following Perotti (2007), the second step is to use external information to the model to estimate the coefficients ${\alpha }_{gy}$ and ${\alpha }_{ty}$.

Coefficient ${\alpha }_{gy}$ reflects the contemporary elasticity of expenditure to output, and ${\alpha }_{ty}$ is the contemporary elasticity of revenues to output. The latter was estimated based on the ‘IMF method,’ as in Sanches (2020), which is a regression using dummy variables for periods and outliers. About the former, by using high-frequency data, component (b) is removed: “it typically takes longer than a quarter for discretionary fiscal policy to respond to, say, an output shock” (Perotti, 2007, p.176). In most of the literature that follows Blanchard and Perotti (2002), such an elasticity is assumed away, that is, ${\alpha }_{gy}$ is considered to be equal to zero. In other words, there is no discretionary response of fiscal variables to output.

Since $u_t^t$ and $u_t^g$ are correlated, from these separate estimations of the exogenous elasticities, the cyclically adjusted residuals, $u_t^{g,ca}$ and $u_t^{t,ca}$, are obtained - which are the shocks without the effects of the cycle:

$u_t^{g,ca}=u_t^g-{\alpha }_{gy}{u}_t^y={\beta }_{gt}{e}_t^t+e_t^g$ (4)

$u_t^{t,ca}=u_t^t-{\alpha }_{ty}{u}_t^y={\beta }_{tg}{e}_t^g+e_t^t$ (5)

The structural shocks, $e_t^g$ and $e_t^t$, can be obtained from the assumption of the ordering of the variables. Blanchard and Perotti (2002) claim that there is no reason to choose ${\beta }_{gt}=0$ or ${\beta }_{tg}=0$a priori. Regarding shocks in spending and revenue, there is no theoretical or empirical basis to decide which variable will react first. As the correlation between adjusted residuals is small, Perotti (2007) points out that the order does not change the result.${\beta }_{gt}=0$ was then assumed, and the regression of the adjusted revenue residuals on the residuals of the structural form of expenditures was estimated by ordinary least squares (OLS) to obtain ${\beta }_{tg}$ in equation (6) (Burriel et al., 2010).² We then obtain instrumental variables, the structural shocks $e_t^t$ and $e_t^g$ in equation 3, since the regressors (residuals of the reduced form) are correlated with the error term (structural shock). Those structural shocks of expenditure and revenue are used as instruments since the correlation between them and the structural shock of output, $e_t^y$, is low. The last step is estimating the impulse-response functions using the estimated coefficients.

The basic model is estimated using the vector of endogenous variables, in real terms: the logarithms of expenditures, revenue, and output. All series were seasonally adjusted, deflated by the IPCA (in December 2023 values), and log-transformed so that the econometric results can be interpreted in terms of elasticity (percentage changes), and then converted to the multiplier effect. For the estimations, the series were converted to their first differences, since they had unit roots in level according to the Augmented Dickey Fuller test. Baseline results presented in the next section do not include control variables - we include them in Appendix F. Also, residual tests indicated the models are, in general, free of problems such as heteroskedasticity and serial autocorrelation. All of them are stable. See Appendix G for more details.

As explained above, Blanchard and Perotti (2002) use the elasticity of revenues to GDP in the SVAR estimations. The methodology to estimate this elasticity closely followed the study by Sanches (2020), known as the “International Monetary Fund method” which involves estimating this elasticity through a regression using Ordinary Least Squares, incorporating controls, and assessing the model’s fit. Several equations were estimated, evaluating the degree of model fit. The model was chosen based on the absence of serial autocorrelation and heteroscedasticity, as well as having the best fit and the smallest errors. Dummy variables were included to account for breaks and outliers in the series, such as a dummy that takes the value of 1 from 2012 onwards (as suggested by Andreis, 2016), when a change in the trend of the product is observed, indicating a slowdown starting at that time. A dummy was also added for the third and fourth quarters of 2010, when a significant increase in net primary revenues occurred due to the acceleration of Brazil’s economic growth. Additionally, a dummy (dum 13) was set to 1 for the last three months of 2013, and another dummy (dum 02) was set to 1 for the first four months of 2002, to take into account outliers. A deterministic trend was also included in the model since the model is estimated with both variables, GDP and revenue, in levels (which are cointegrated). Similar methodology has been applied by Andreis (2016, 2014), Maciel (2006), Sanches and Carvalho (2022) and Sanches and Carvalho (2023). For baseline exercises, the elasticity estimated was 1.25, ³ which is in line with other studies in the literature for the Brazilian economy, such as Peres (2006) (around 2), Andreis (2014) (1.084), Gobetti et al. (2016) (1.23), Ribeiro (2016) (1.16), Casalecchi and Barros (2018) (1.13), Sanches and Carvalho (2022) (1.2).

To calculate multipliers, we need to divide the elasticity of the response by the average share of social expenditures in output (or its components). As the variables are in logarithmic form, impulse-response functions provide the elasticity of output (Y) to the fiscal variable (X):

${\xi }_{Y,X}=\frac{\frac{\Delta Y}{Y}}{\frac{\Delta X}{X}}=\frac{\Delta Y}{Y}\frac{X}{\Delta X}=\frac{\Delta Y}{\Delta X}\frac{X}{Y}$ (6)

According to Pires (2014), since $\frac{\Delta Y}{\Delta X}$ is the definition of the multiplier, which reflects a change in output given an increase of one unit in the fiscal variable, we have that:

$\frac{\Delta Y}{\Delta X}=\frac{{\xi }_{Y,X}}{\frac{X}{Y}}$ (7)

We estimate three types of multipliers (Spilimbergo et al., 2009): a) impact, which measures the effect of fiscal policy in the same period as the shock (month, in our case):$\frac{\Delta Y\left(t\right)}{\Delta X\left(t\right)}$; b) peak, which represents the highest response: max $\frac{\Delta Y\left(t+n\right)}{\Delta X\left(t\right)}$; and c) cumulative, which measures the impact over a specific number of months⁴. The latter is considered by the literature as the most appropriate measure of the fiscal multiplier, since the economy requires some time to absorb the initial shock (Ilzetzki et al., 2013; Spilimbergo et al., 2009; Restrepo, 2020). It is represented by the equation: $\frac{{{\displaystyle \sum }}_{i=1}^n\Delta Y\left(t+i\right)}{{{\displaystyle \sum }}_{i=1}^n\Delta X\left(t+i\right)}$.

4. Fiscal multipliers

We summarize our results below. In the graphs, we follow Sanches and Carvalho (2022) and present the cumulative impulse-response function of output to a shock in each expenditure and revenue variable, using confidence intervals of one and two standard deviations. This choice does not have formal justification (Ramey 2011), but much of the multiplier literature provides statistical significance based on a one standard deviation interval (including the seminal paper by Blanchard and Perotti (2002), which inspired all subsequent literature- see Ramey (2011), for more details)⁵.

For each component of public spending (social benefits, public investment, personnel expenses, subsidies, and other expenditures), as well as for total expenditures and revenues, we estimate the impulse-response functions for the 1997-2023 sample, excluding the pandemic years (2020 and 2021). This latter strategy is inspired by Lenza and Primiceri (2020), who demonstrate that the ‘ad hoc’ strategy of discarding observations during the pandemic can be acceptable for the purpose of estimating parameters of a vector autoregressive model as a way to eliminate extreme observations from the sample. In Appendix C, we present the results for the full sample, 1997-2023. In Appendix D, we present estimates considering the pre-pandemic sample (1997-2019). Also, Appendices E and F, respectively, present the results with quarterly data and control variables, as robustness checks. We included in each exercise four macroeconomic control variables, following the literature (Sanches and Carvalho, 2022; Sanches and Carvalho, 2023): inflation rate, interest rate, real exchange rate index and an index for commodities prices. Our main conclusions were robust about these changes. See more details about the data in Appendix B.

In Figure 3, we compile the cumulative impulse-response functions that illustrate the GDP response to shocks from increases in primary expenditures or revenues, as well as for each expenditure component. The dashed and dotted lines represent confidence levels of one and two standard deviations (68% and 95% confidence levels), respectively. It can be observed that the responses of GDP to social benefits and public investments are the largest and most significant: they are significant at the 5% level, whereas other expenditures do not show a significant response over time. These results are in line with Sanches and Carvalho (2022), Resende (2019) and Orair et al (2016).

It is important to highlight that only public investments and social benefits exhibit a significant multiplier effect in accumulated terms. This is not a limitation of our results, but rather a reinforcement of our central hypothesis: these two components of fiscal policy have a substantial macroeconomic impact, while other types of spending do not produce significant effects in the medium term. In fact, as we will show in the next section, a fiscal adjustment can be achieved with a positive net impact on economic activity, as long as it focuses on increasing revenues or reducing subsidies, combined with an expansion of public investments and/or social benefits, which are the spending categories with the greatest multiplier effects.

These results are, to some extent, aligned with those found in the literature referenced in the introduction. As noted by Busato and Martins (2024), since the 2000s, studies on fiscal multipliers have increasingly focused on the subcategories of public expenditure. In general, it can be stated that the multiplier effects of public investments are more significant than those associated with government consumption and the tax burden or revenue (Pires, 2014; Castelo Branco et al., 2017; Sanches and Carvalho, 2022). More recently, the literature has focused on the large effect of social spending, as indicated by Sanches and Carvalho (2023) for Brazil. Our results reinforce these findings already identified in the literature. Our contribution lies in using these findings to conduct simulations regarding the consequences of different types of fiscal adjustment.

Figure 3
Cumulative impulse-response functions of GDP to shocks in total primary expenditure (and its components) and total primary revenue

Table 2 summarizes the main results⁶, considering the impact (in the same month as the shock), peak, and cumulative multipliers. The latter takes into account the persistence of the shock⁷. Again, public investment and social benefits stand out for their positive effects on GDP, with statistical significance even over a longer term. Translating the numbers, for each real spent on investment and social benefits, GDP tends to increase by R$ 2.6 and R$ 2.15, respectively, after 25 months.

Table 2
Estimated fiscal multipliers

	Impact	Peak	Accumulated (15 or 25 months, depending on the persistence)	Accumulated 4 months	Accumulated in 12 months (for the simulations)
Total expenditure	0.11*	0.27	0.5	0.23	0.47
Total revenue	-0.43*	0.032	-0.55	-0.5	-0.53
Public investment	0.65*	1.53*	2.6*	1.4*	2.4*
Social benefits	0.35*	0.9*	2.15*	1.1*	1.81*
Personnel	-0.04	-0.07	-1.18	-0.8	-1.14
Subsidies	-0.05	0.3	0.52	0.8	0.5
Other expenditures	0.1*	0.21	0.28	0.2	0.27

• Source: authors’ elaboration. (*) denotes significance at the 5% level.

5. Simulations on fiscal consolidation: Adjustment via expenditure vs. adjustment via revenue

The estimates of multipliers help us understand the effect on GDP of different types of fiscal variables. Therefore, beyond a general understanding of these effects, we can also use them to estimate the impacts of specific policies.

Currently, there is a major debate in Brazil about the best way to implement fiscal adjustment: through increased revenues or reduced expenditures. In this section, we will examine the potential effects of each of these options based on the type of expenditure the adjustment targets. The effects estimated here are those primarily related to the dynamics of short- and medium-term aggregate demand. That is, specific expenditures and revenues may have different long-term effects, which are not estimated here. We will focus on the effects of different adjustment measures on growth and public accounts in 2023.

In this exercise, we consider a fiscal adjustment of 1% of GDP to simulate its effects on the level of GDP, the primary deficit, and public debt, both relative to GDP.

The following exercise simulates the effect of three types of fiscal adjustment. As in Marconi (2017), we use net domestic debt, as it is the one immediately affected by fiscal measures. Following Sanches (2020), we use the following equation to estimate the variation in debt in the simulated scenarios:

$\Delta D_t=\left(G_t-T_t\right)+r{D}_{t-1}$ (8)

where $\Delta D_t$ is the variation in debt from one year to the next, r is the real interest, $D_{t-1}$ is the public debt from the previous period (note that interest is applied to this debt stock), and ($\left(G_t-T_t\right)$) is the difference between the federal government’s primary expenditures and revenues (or the primary deficit). The series of the federal government’s and Central Bank’s net domestic public debt was obtained from the Central Bank, as well as its implicit interest rate (deflated by the IPCA)⁸.

The scenario estimates took into account the one-year multiplier effect, as estimated in the previous section of this paper. In 2023, the federal government’s primary deficit as a percentage of GDP was 2.11%, and the ratio of the federal government’s net domestic public debt to GDP was 59.58%. The following scenarios use 2023 data to simulate the scenarios with the respective types of fiscal adjustment.

In Table 3, presented below, we introduce three fiscal adjustment scenarios⁹:

1. Scenario 1: An adjustment of 1% of GDP on the side of public spending, that is, through spending cuts. Revenues are not affected discretionarily, varying only in relation to their connection with GDP, estimated by its elasticity of 1.25% in relation to GDP; that is, revenue will only decrease if GDP decreases, and not due to a government measure. We then simulate scenarios where there are spending cuts in public investments, social benefits, and subsidies.

2. Scenario 2: An adjustment of 1% of GDP through increased revenues, while keeping spending constant.

3. Scenario 3: An adjustment of 1% of GDP increase in both revenues and public spending on: public investments, social benefits, or subsidies. Revenue also responds to GDP according to its respective elasticity.

Table 3
Effect of a 1% of GDP fiscal adjustment in three different scenarios on: GDP, primary deficit-to-GDP ratio, and net federal government debt-to-GDP ratio (domestic)

	GDP variation	Primary deficit - GDP ratio	Debt - GDP ratio	Difference in Debt- GDP ratio (in percentage points p.p) compared to the baseline	Difference in deficit- GDP ratio (in percentage points p.p) compared to the baseline
1. Adjustment via expenditures (constant revenue)
Cut in public investment	-2.44%	1.7%	60%	0.42 p.p.	-0.41 p.p.
Cut in social benefits	-1.81%	1.54%	59.4%	-0.18 p.p.	-0.57 p.p.
Cut in subsidies	-0.5%	1.23%	58.3%	-1.28 p.p.	-0.88 p.p.
2. Adjustment via revenues (constant expenditure)
	-0.53%	1.24%	58.3%	-1.28 p.p.	-0.87 p.p.
3. Adjustment via revenues (with an increase in expenditures)
3.1. Increase in public investment	1.91%	1.63%	57.36 %	-2.22 p.p.	-0.48 p.p.
3.2. Increase in social benefits	1.28%	1.79%	57.86%	-1.72 p.p.	-0.32 p.p.
3.3. Increase in subsidies	-0.03%	2.23%	59%	-0.58 p.p.	0.12 p.p.

• Source: prepared by the authors. Baseline: primary deficit-to-GDP of 2.11% and debt-to-GDP ratio of 59.58% (2023 data).

In the first case of Scenario 1, the adjustment falls on public investments. In this context, GDP would decline by 2.4% compared to what actually occurred in 2023. Note that the public debt-to-GDP ratio would be around 60%, that is, higher than the observed debt-to-GDP ratio (59.58%). Similarly, if the adjustment prioritizes cutting social benefits, we also observe a substantial decline in GDP (1.81%), along with a debt-to-GDP ratio very close to the observed value, 59.4%. However, if the adjustment occurs through subsidy cuts, there is only a small negative effect on GDP (0.5% decrease), and this measure would result in slower growth of debt, to 58.3%, and a improvement in the primary balance.

In Scenarios 2 and 3, we examine increases in revenue. Borges (2024)¹⁰, for example, points to the recent reduction in the tax burden in Brazil¹¹ due to various factors, which creates space for the adjustment to occur, at least partially, on the revenue side. The author argues that one possibility is the use of a carbon tax, for example. A recent study from the World Bank¹² estimated that this type of tax could generate significant revenue potential for Brazil, of around 1% of GDP.¹³

In Scenario 2, we simulate an adjustment of this magnitude via revenues, without any increase in spending. The cost in terms of negative multiplier effect would be about 0.53% of GDP in one year. The debt-to-GDP ratio would be lower than in cases where there are cuts in investment and social spending. This is because an adjustment through revenue is less recessive than through spending cuts.

In Scenario 3, the initial shock is not strictly a fiscal adjustment, as there is an increase of the same proportion in both government revenues and expenditures. This setup resembles the logic of the Haavelmo multiplier, which suggests that a balanced-budget expansion - where spending and taxes increase by the same amount - can still have a positive effect on output. Interestingly, the results point to an extension of this idea: the magnitude and direction of the net effect depend on the composition of the revenue increase and the spending expansion. Significant positive effects on GDP are observed in the scenarios with increased investment (1.91%) and social benefits (1.28%). These scenarios lead to the greatest reduction in the debt-to-GDP ratio (by 2.22 and 1.72 percentage points, respectively, compared to the observed scenario). Conversely, when the adjustment compensates for the revenue increase through higher spending on subsidies, the net effect on GDP is negative. This scenario with increased subsidies shows the highest primary deficit relative to GDP.

In summary, it is interesting to note that if the adjustment occurs through increased revenues, along with increased spending on public investments and/or social benefits, it is possible to achieve a positive net effect on GDP, as well as the lowest debt-to-GDP ratios among the simulated scenarios. This is because components of spending with high multiplier effects, such as social spending and public investment, boost aggregate demand and stimulate GDP. On the other hand, the scenario with the highest debt-to-GDP ratio is the one where the entire fiscal adjustment of 1% of GDP is focused on cuts in public investment (Scenario 1.1). These conclusions are maintained in the robustness test in Appendix H (using alternative multipliers).

Therefore, fiscal adjustments that preserve high multiplier spending, such as public investments and social benefits, are better from the perspective of economic activity and also the sustainability of public debt. It is important to highlight some reasons why such spendings have a significant multiplier effect, greater than others. Social benefits, for example, are targeted at lower-income individuals who are at the base of the income distribution pyramid and contribute to stimulating the economy, as these individuals have a higher marginal propensity to consume than those with higher income levels (Kalecki, 1952; 1942). The marginal propensity to consume measures how much consumption increases with an additional unit of income. For example, Palomo, Carvalho, and Toneto (2022) estimate that, in Brazil, the marginal propensity to consume for the poorest 50% is around 0.609, while this propensity for the top 1% of the population is not statistically different from zero. Since spending on social benefits directs income to the segment of the population with a higher propensity to consume, there is a considerable stimulus to household consumption. This increased demand for goods and services, in turn, also stimulates private investment, generating a significant income multiplier effect (Sanches and Carvalho 2023).¹⁴

Public investment, in addition to its direct effect on aggregate demand, tends to encourage private sector investment and contribute to increasing economic productivity (Dutt 2013). The positive response of private investment to public investment is called the crowding-in effect. Recent literature on the Brazilian case documents this effect in various studies (Bredow et al. 2022; Iasco-Pereira and Duregger 2023; Reis et al. 2019; Sanches et al. 2025). Thus, public investment could, for example, have its multiplier effect enhanced by inducing private investment (Sanches and Carvalho 2022).

Figure 4 below shows the impact in each scenario on GDP (green bar), the primary result-to-GDP ratio (orange bar), and the impact on the debt-to-GDP ratio (blue bar). The latter is presented in terms of the difference in percentage points of each scenario compared to what actually occurred in 2023 (59.58%). We note that Scenario 3 results in substantially lower debt-to-GDP ratios than the observed ratio in 2023 in Scenarios 3.1 and 3.2, that is, in those where spending is directed towards investments and social benefits. The impact on the primary result-to-GDP ratio is shown in the graph as the difference in percentage points compared to the year 2023 (which was -2.11%).

Figure 4
Effect of a 1% of GDP fiscal adjustment in three different scenarios

6. Conclusion

The series of extreme economic shocks in the past 15 years have placed fiscal policy again as a major tool of macroeconomic stabilization and reignited research on its effects on the GDP. These two aspects are, of course, interconnected. According to Oh and Reis (2012), policymakers had little research to guide them on which policies to implement in terms of stimulus of demand in response to the Great Financial Crisis. This paper contributes to this literature by estimating fiscal multipliers of revenue increases and different types of fiscal spending using data from Brazil and employing these estimations to simulate different scenarios of fiscal adjustment measures.

More specifically, the first contribution of the paper is the construction of a new dataset on primary expenditures and revenues based on the method suggested by Gobetti and Orair (2017). This dataset allows us to correctly characterize different types of government spending and aggregate it into different groups: public investment, social benefits, subsidies, personnel expenses, and other expenditures. Using this dataset, we estimated fiscal multipliers.

We found that the components with the highest multiplier effect are public investments and social benefits. For example, R$ 1 spent on public investments generates R$ 2.6 on GDP after 25 months. For social benefits, the generated income is close: R$ 2.15. On the other hand, the multipliers associated with subsidies and revenue, for example, have much lower estimated values.

These results suggest that fiscal adjustments involving cuts in spending with higher multiplier effects, such as investments and social spending, are costly in terms of GDP and, especially for this reason, tend to lead to an increase in the debt-to-GDP ratio. Our results also suggest that adjustments through increased revenue and cuts in subsidies, for example, would be alternatives with a smaller adverse effect on economic activity.

Our second contribution is simulating different scenarios of fiscal adjustment measures. We simulated a federal government fiscal adjustment of 1% of GDP in seven scenarios: adjustment entirely based on spending cuts (in (i) subsidies, (ii) investments, or (iii) social benefits); (iv) adjustment entirely based on increased revenues; adjustment based on increased revenues, with higher spending (on (v) subsidies, (vi) investments, and (vii) social benefits. We found that it is possible to achieve fiscal adjustment with a positive net effect on economic activity, provided it is based on increased revenues or subsidies reduction along with the expansion of public investments and/or social benefits, which are the spending items with the highest multiplier effects. These scenarios also present the lowest debt-to-GDP ratios.

References

• Alesina, Alberto, and Silvia Ardagna. 2010. “Large Changes in Fiscal Policy: Taxes versus Spending”. Tax Policy and the Economy 24(1): 35-68.
• Alves, Renan Santos. 2017. “O impacto da política fiscal sobre a atividade econômica ao longo do ciclo econômico: evidências para o Brasil”. Master’s thesis. University of São Paulo.
• Andreis, Augusto. 2014. “Balanço orçamentário estrutural no Brasil: a política fiscal no Brasil é pró ou anticíclica no período pós plano Real?” XIX Prêmio Tesouro Nacional, Finanças Públicas.
• Andreis, Augusto. 2016. “O comportamento da política fiscal brasileira no período pós-plano Real e suas implicações”. Master’s thesis. Universidade do Vale do Rio dos Sinos.
• Auclert, Adrien, Matthew Rognlie and Ludwig Straub. 2024. “The intertemporal Keynesian Cross”. Journal of Political Economy 132 (12): 4068-4121.
• Auerbach, Alan J., and Yuriy Gorodnichenko. 2012. “Measuring the Output Responses to Fiscal Policy.” American Economic Journal: Economic Policy 4(2): 1-27.
• Blanchard, Olivier, and Roberto Perotti. 2002. “An empirical characterization of the dynamic effects of changes in government spending and taxes on output”. Quarterly Journal of Economics 117 (4): 1.329-1.368.
• Borges, Bráulio. 2024. “A mudança das metas e o desafio da sustentabilidade fiscal brasileira”. Observatório de política fiscal. Available at: https://observatorio-politica-fiscal.ibre.fgv.br/politica-economica/outros/mudanca-das-metas-e-o-desafio-da-sustentabilidade-fiscal-brasileira Access on 09/09/2024.
  » https://observatorio-politica-fiscal.ibre.fgv.br/politica-economica/outros/mudanca-das-metas-e-o-desafio-da-sustentabilidade-fiscal-brasileira
• Bredow, Sabrina Monique Schenato, André Moreira Cunha, and Marcos Tadeu Caputi Lélis. 2022. “Investimentos públicos e privados no Brasil entre 1996 e 2018”. Revista de Economia Política 42 (4): 957-976.
• Brenck, Clara Zanon. 2021. “Desigualdade salarial e composição do emprego no Brasil: uma estimação VEC para o período 2004-2019”. Nova Economia 31(2): 345-380.
• Burriel, Pablo, Francisco de Castro, Daniel Garrote, Esther Gordo, Joan Paredes, and Javier Pérez. 2010. “Fiscal Policy Shocks in the Euro Area and the US: An Empirical Assessment”. Fiscal Studies 31 (2): 251-285.
• Busato, Maria Isabel, and Norberto Montani Martins. 2024. “Multiplicadores fiscais no Brasil: entre consensos e dissensos”. Revista de Economia Contemporânea 28: 1-32.
• Casalecchi, Alessandro Ribeiro de Carvalho, and Gabriel Leal de Barros. 2018. “A variação da receita em resposta à atividade econômica”. Instituição Fiscal Independente, Nota Técnica n.19.
• Castelo-Branco, Marco Antonio, Elcyon C. Rocha Lima, and Luiz Fernando de Paula. 2017. “Mudanças de regime e multiplicadores fiscais no Brasil em 1999-2012: uma avaliação empírica com uso da metodologia MS-SBVAR”. Pesquisa e Planejamento Econômico 47(2): 7-62.
• Coudret, Nicolas. 2013. “Fiscal Multipliers in the Netherlands: A Structural VAR Approach”. Master’s thesis in Policy Economics, Erasmus School of Economics.
• Dutt, Amitava Krishna. 2013. “Government spending, aggregate demand, and economic growth”. Review of Keynesian Economics 1 (1): 105-119.
• Farhi, Emmanuel, and Iván Werning. 2016. “Fiscal Multipliers: Liquidity Traps and Currency Unions.” Chap. 31 in Handbook of Macroeconomics, Vol. 2, edited by John B. Taylor and Harald Uhlig, 2417-92. Amsterdam: Elsevier.
• Garcia, Jesus Gonzalez, Antonio Lemus, and Mico Mrkaic. 2013. “Fiscal multipliers in the ECCU”. IMF Working Paper 13 (117).
• Gobetti, Sergio Wulff, Frederico Dutra and Rodrigo Octávio Orair. 2016. “Resultado estrutural, ciclos econômicos e os desafios para avaliar o desempenho fiscal”. XXII Prêmio Tesouro Nacional.
• Gobetti, Sergio Wulff, and Rodrigo Octávio Orair. 2017. “Resultado primário e contabilidade criativa: reconstruindo as estatísticas fiscais acima da linha do governo geral”. Texto para discussão 2.288. Brasília: - IPEA.
• Grudtner, Vanessa, and Edilean Kleber da Silva Bejarano Aragon. 2017. “Multiplicador dos gastos do governo em períodos de expansão e recessão: evidências empíricas para o Brasil”. Revista Brasileira de Economia 71 (3): 321-345.
• Iasco-Pereira, Hugo, and Rafael Duregger. 2023. “Public investment, infrastructure, and private investment in Brazil: is there a crowding-in effect?”. Anais do 51º Encontro Nacional de Economia. Anpec.
• Ilzetzki, Ethan, Enrique Mendoza, and Carlos Végh. 2013. “How Big (Small?) Are Fiscal Multipliers?”. Journal of Monetary Economics 60: 239-254.
• Jemec, Natasa, Andreja Strojan Kastelec, and Ales Delakorda. 2013. “How Do Fiscal Shocks Affect the Macroeconomic Dynamics in the Slovenian Economy?”. Bank of Slovenia. Working Paper. Ljubljana: Bank of Slovenia.
• Kalecki, Michal. 1942. “A Theory of profits”. The Economic Journal 52 (206/207): 258.
• Kalecki, Michal. 1952 [1978] “Teoria da dinâmica econômica”. In Coleção Os Economistas. São Paulo: Abril Cultural.
• Lenza, Michele, and Giorgio Primiceri. 2020. “How to estimate a VAR after March 2020”. European Central Bank. Working Paper n. 2461.
• Maciel, Pedro. 2006. “Proposta à regra fiscal brasileira: orientação pelos ciclos econômicos”. XI Prêmio do Tesouro Nacional.
• Marconi, Nelson. 2017. “Uma meta para o investimento público com equilíbrio fiscal”. Conjuntura Econômica.
• Marques, Pedro Romero, Theo Ribas Palomo, and Laura Barbosa Carvalho. “Como neutralizar os efeitos regressivos de uma tributação sobre carbono no Brasil?”. 2020. Nota de Política Econômica n. 005. Centro de Pesquisa em Macroeconomia das Desigualdades, Made /USP.
• Matheson, Troy, and Joana Pereira. 2016. “Fiscal multipliers for Brazil. International Monetary Fund”. IMF Working Papers 79.
• Oh, Hyunseung, and Ricardo Reis. 2012. “Targeted transfers and the fiscal response to the great recession”. Journal of Monetary Economics 59: S50-S64.
• Orair, Rodrigo Octávio, Fernando de Faria Siqueira, and Sergio Wulff Gobetti. 2016. “Política fiscal e ciclo econômico: uma análise baseada em multiplicadores do gasto público”. XXI Prêmio do Tesouro Nacional.
• Palomo, Theo Ribas, Laura Carvalho, and Rodrigo Toneto. 2022. “Marginal propensity to consume heterogeneity and redistributive policies: The Brazilian case”. Working Paper 10. Centro de Pesquisa em Macroeconomia das Desigualdades, Made /USP.
• Peres, Marco Aurélio Ferreira. 2006. “Os efeitos dinâmicos da política fiscal sobre a atividade econômica: um estudo para o caso brasileiro”. Master’s thesis. University of Brasília.
• Peres, Marco Aurélio Ferreira, and Roberto de Goes Ellery Junior. 2009. “Efeitos dinâmicos dos choques fiscais do governo central no PIB”. Pesquisa e Planejamento Econômico 39 (2): 159-206.
• Perotti, Roberto. 2007. “In Search of the Transmission Mechanism of Fiscal Policy.” In NBER Macroeconomics Annual 2007, Volume 22, edited by Daron Acemoglu, Kenneth Rogoff, and Michael Woodford: 169-226. Chicago: University of Chicago Press.
• Pires, Manoel Carlos de Castro. 2014. “Política fiscal e ciclos econômicos no Brasil”. Economia Aplicada 18 (1): 69-90.
• Ramey, Valerie Ann. 2011. “Identifying government spending shocks: It’s all in the timing”. Quarterly Journal of Economics 126 (1): 1-50.
• Ramey, Valerie Ann. 2019. “Ten years after the financial crisis: what have we learned from the Renaissance in fiscal research?” Journal of Economic Perspectives 33(2): 89-114.
• Reis, Cristina, Eliane Araújo, and Erica Oliveira Gonzales. 2019. “Public investment boosted private investment in Brazil between 1982 and 2013”. Journal of Economic Issues 53 (3): 813-840.
• Resende, Carolina Tavares. 2019. “Impulso fiscal: Uma abordagem de multiplicadores fiscais com aplicação para a economia brasileira”. Master’s thesis. Escola de Economia de São Paulo. Fundação Getulio Vargas.
• Restrepo, Jorge. 2020. “How big are fiscal multipliers in Latin America?”. IMF Working Paper 17.
• Ribeiro, Lívio. 2016. “Sobre arrecadação e atividade econômica”. Nota Técnica FGV IBRE.
• Romer, Christina D., and David H. Romer. 2010. “The Macroeconomic Effects of Tax Changes: Estimates Based on a New Measure of Fiscal Shocks.” American Economic Review 100(3): 763-801.
• Sanches, Marina da Silva. 2020. “Política fiscal e dinâmica do produto: uma análise baseada em multiplicadores fiscais no Brasil”. Master’s thesis. University of São Paulo.
• Sanches, Marina da Silva, and Laura Barbosa Carvalho. 2022. “A contribuição da política fiscal para a crise brasileira de 2015-2016: uma análise baseada em multiplicadores de despesas e receitas primárias do governo central no período 1997-2018”. Nova Economia 32 (1).
• Sanches, Marina da Silva, and Laura Barbosa Carvalho. 2023. “Multiplier effects of social protection: A SVAR approach for Brazil”. International Review of Applied Economics 23 (1): 93-112.
• Sanches, Marina da Silva, Tainari Taioka, Clara Zanon Brenck, and Gustavo Pereira Serra. 2025. “How does income distribution affect the impact of public investment on private investment? Empirical evidence from Brazil”. EconomiA, Selecta 2024, 26 (2): 193-207.
• Sims, Christopher. 1980. “Macroeconomics and Reality”. Econometrica 48: 1-48.
• Spilimbergo, Antonio, Steve Symanski, and Martin Schindler. 2009. “M. Fiscal multipliers. Technical report”. IMF Staff Position Note.
• Vdovychenko, Artem. 2018. “How Does Fiscal Policy Affect GDP and Inflation in Ukraine?”. Visnyk of the National Bank of Ukraine 244. National Bank of Ukraine.
• Zubairy, Sarah. 2014. “On Fiscal Multipliers: Estimates from a Medium Scale DSGE Model.” International Economic Review 55(1): 169-195.

• ♦
  Os autores agradecem a Gilberto Tadeu Lima, Lucca Rodrigues, Clara Brenck, Amanda Resende e aos revisores anônimos pelos comentários e sugestões, que melhoraram o artigo substancialmente. Agradecem também o apoio do Centro de Pesquisa em Macroeconomia das Desigualdades (Made-FEA/USP), da Universidade de São Paulo.
• DECLARAÇÃO DE DISPONIBILIDADE DE DADOS
  Os dados utilizados neste estudo estão disponíveis mediante solicitação ao autor. Dados adicionais e informações complementares também poderão ser fornecidos para fins de verificação ou replicação. A disponibilização está condicionada à inexistência de restrições de acesso público.
• Classificação JEL
  E62, H5, H2.
• 1
  Regarding revenue-side adjustments, such as those aimed at addressing non-recurring revenues, the focus of this work is to reduce issues in the assessment of public account components rather than to obtain a structural result. In other words, the goal is to correctly determine the primary balance and its components, rather than to produce a cycle-adjusted estimate or one net of non-recurring operations. In this sense, the most relevant revenue-side adjustments pertain specifically to the transfer of exploration rights (cessão onerosa), payroll tax exemptions, and the sovereign fund (fundo soberano).
• 2
  Models were also estimated assuming tg=0, that is, that decisions relating to revenue occur before those relating to expenditure. This procedure indicated the robustness of the results to different specifications, with minor variation in impulse response functions, as is usual in the literature.
• 3
  The exercise with the extended sample (up to 2023) resulted in an elasticity of 1.36, compared to 1.20 in the exercise up to 2019. In the sample up to 2019, the estimated elasticity was similar (1.22). In the 1997-2023 sample, excluding 2020 and 2021 (baseline), the elasticity was around 1.25. For the quarterly exercise (Appendix E), the estimated elasticity was 1.7.
• 4
  The definition of the number of months is based on Garcia et al. (2013, p.11): “The long-run multiplier is defined as the cumulative multiplier when J->∞, but in practice is used the number of periods needed for the multiplier to stabilize as its long-run value ”.
• 5
  In the full sample, for the social benefits exercise, 7 lags were included-6 lags for the others. In the sample up to 2019 (and for the sample excluding only 2020 and 2021), 6 lags were included for all exercises.
• 6
  Statistical significance is assessed based on the confidence intervals of the cumulative impulse response functions, following the standard approach in the SVAR literature (Blanchard and Perotti, 2002). In the graphs, we highlight periods where the 68% and 95% confidence bands do not include zero, indicating statistically significant responses. While much of the literature (see Ramey (2011)) relies on 68% intervals (equivalent to one standard deviation), we adopt a more rigorous approach by reporting both 68% and 95% bands. Unlike the fiscal multiplier estimates, the impulse responses themselves are not subject to additional transformations.
• 7
  Following Sanches and Carvalho (2022), for social benefits and public expenditure, we consider 25 months to calculate the cumulative multiplier, given that these types of expenditure have higher persistence; and 15 months for the other components and for the revenue.
• 8
  The implicit interest rate for this concept of public debt, in cumulative terms for the year 2023 and already deflated by the IPCA, was 5,429% (Central Bank of Brazil). This rate was held constant throughout the simulations. While this is a simplifying assumption, we consider it reasonable given the results presented in Appendix F. When the interest rate is included as a control variable in the VAR, the estimated multipliers change very little, and the interest rate is not statistically significant in any specification. Even in more complete models with four control variables (see Appendix F), interest rate impacts are generally insignificant, with the sole exception of a negative effect at the 10% level in the public investment equation - an effect that does not meaningfully alter the fiscal multipliers.
• 9
  We acknowledge that total spending on subsidies amounts to less than 1% of GDP. The simulation presented in the paper is purely illustrative and aims to provide a standardized basis for comparing the macroeconomic effects of different types of public spending. The 1% of GDP shock should be interpreted as a hypothetical, uniform scenario designed for analytical purposes only.
• 10
  https://observatorio-politica-fiscal.ibre.fgv.br/politica-economica/outros/mudanca-das-metas-e-o-desafio-da-sustentabilidade-fiscal-brasileira.
• 11
  Borges (2024) points out that there is a decline in revenues on average (between 2005-2008 and 2021-2023) of around 1.6% of GDP.
• 12
  https://documents1.worldbank.org/curated/en/099050123155511882/pdf/P1761580a79b5b0c80b34c01afa40534151.pdf.
• 13
  Despite its regressive effect, a carbon tax can finance social policies that reduce inequality (Marques et al., 2020).
• 14
  The importance of heterogeneity in marginal propensities to consume is also emphasized by Auclert et al. (2024), who argue that, when this heterogeneity is taken into account, a deficit-financed fiscal policy results in cumulative spending multipliers greater than one.
• 15
  This subsidy is added to the other subsidies that are explicitly indicated in the government accounts.
• 16
  For more details on the methodological adjustment decisions, see Gobetti and Orair (2017).
• 17
  Even though the accumulated multiplier for public investment lacks significance at 5%, it is significant at 10% level.
• 18
  The normality test Jarque Bera rejects the null hypothesis that residuals are normally distributed. Normality is hard to achieve in the short run time series. As pointed out by Brenck (2021, p.378), “The Law of Large Number states that when sample size tends to infinity, the sample mean converges to the population mean, and the error term becomes normally distributed so that the lack of normality can be due to shorter term data”.

Appendix A - Adjustments to the database

A.1) Database

A.1.1. Public revenue

A.1.1.1. Primary revenue

For the line ‘I.4.8. Operações com ativos’ (Operations with assets), we removed the effects of creative accounting identified during the onerous assignment and capitalization of Petrobras in September 2010. This accounting maneuver inflated public revenue by R$ 72.8 billion and public spending by R$ 42.9 billion, resulting in an overstatement of the government’s primary revenue by R$ 31.9 billion. Consequently, our adjustments involved subtracting R$ 72.8 billion from the lines ‘‘I.4.8. Operações com ativos’ and R$ 42.9 billion from ‘III.4. Despesas discricionárias’ (Discretionary spending) with reference to September 2010.

Additionally, the impact of the sovereign fund on the primary balance was removed by deducting the values provided by the National Treasury in the relevant line (‘V. Fundo Soberano do Brasil’) for the years with available data.

We subtracted the values recorded in the line ‘I.4.7. Complemento para o FGTS (LC nº 110/01)’ (‘Supplement to the FGTS by Complementary Law No. 110/2001) under primary revenue due to fiscal maneuvers adopted by the government. A 2012 decree allowed the Brazilian Treasury to temporarily retain the transfers of an additional severance fine intended for the federal fund aimed at protecting workers (FGTS), artificially inflating the primary balance. For the line ‘I.4.4. Cota-parte de Compensações financeiras’ (Share of Financial Compensation), we based our adjustments on the data provided by Gobetti and Orair (2017) for the period from January 1997 to December 2000.

A.1.2. Public spending

A.1.2.1. Primary Spending

The ‘Programa de Sustentação de Investimento (PSI)’ (Investment Support Program) was created in 2009 in the aftermath of the international financial crisis. We adjusted the public spending line ‘III.3.20.1.16 Programa de Sustentação de Investimento (PSI)’ for the years 2009 to 2015, in order to ensure that expenses were accounted for in the year they were incurred. In the official accounts, the recorded values related to the Investment Support Program for 2015 are inflated because the National Treasury settled liabilities with BNDES related to loans from previous years, particularly between 2010 and 2014. Specifically, we adjusted the values by adding R$ 1.3 billion in 2010, R$ 2.2 billion in 2011, R$ 2 billion in 2012, R$ 4.1 billion in 2013, and R$ 5.9 billion in 2014. Correspondingly, we reduced the values for 2015 by R$ 14.5 billion, based on data from the Central Bank of Brazil (BCB) collected by Gobetti and Orair (2017). No similar discrepancies were found for years other than 2015 that merited adjustments.

In recent years, the Brazilian government has implemented several policies of payroll tax relief. The compensation amounts paid by the National Treasury to the National Institute of Social Security are recorded under a specific line in the public accounts labeled ‘III.3.9 Compensação ao RGPS pelas desonerações da folha’ (Compensation to the General Social Security System for payroll tax exemption). Given that the tax exemptions were recorded as an increase in public spending rather than a reduction in revenue, public accounts are inflated for the period between 2012 (the year the payroll tax exemption policy was first recorded) and 2021 (the most recent data on this policy from the National Treasury). To address this issue, we subtracted the recorded values related to tax exemptions from the relevant line under primary spending, as well as from the line ‘I.3 Arrecadação Líquida para o RGPS - Urbana’ (Net revenue to the General Social Security System - Urban), under primary revenue.

Since the line ‘III.1.1 Benefícios Previdenciários - Urbano’ (Social Security Benefits - Urban) is inflated for the period from 1997 to 2003 due to the inclusion of values related to the ‘Renda Mensal Vitalícia (RMV)’ (Lifetime Monthly Pension) program, which has been phased out by the Brazilian government, we adjusted the relevant line by subtracting the RMV program values obtained by Gobetti and Orair (2017).

We subtracted the values recorded in the line ‘III.3.7. Complemento para o FGTS (LC nº 110/01)’ (‘Supplement to the FGTS by Complementary Law No. 110/2001) under primary spending, as the series has been inflated since 2012 because of accounting maneuvers as explained above.

Relevant data for the line “III.3.20.1.19. Fundo Nacional de Desenvolvimento (FND)” (National Development Fund) are available for the period from 2004 to 2010, with some residual values in 2012. The FND was a development fund created in 1986 and terminated in 2010, aimed at providing resources for the Union to make capital investments necessary to stimulate national development. We excluded these values from the original database to correct for the accounting maneuvers employed by the government between 2008 and 2009. By channeling resources from the FND to the private sector through BNDES loans, the government aimed to keep the primary balance indicator intact.

We use the data collected by Gobetti and Orair (2017) from SIAFI - Sistema Integrado da Administração Financeira (Integrated Financial Administration System) to adjust the line ‘III.3.14. Fundo Constitucional do DF’ (Constitutional Fund of the Federal District), for the year 2015, as these values were recorded under an unusual entry. This adjustment preserves the total expenditure amount and ensures that the composition of the expenditure is consistent. We also refer to the data provided by Gobetti and Orair (2017) to adjust the line ‘III.3.20.1.19. Equalização de Custeio Agropecuário’ (Agricultural Financing Equalization), from January 2002 to December 2015.

A.1.2.2. Personnel expenses

The values for the line ‘III.2. Pessoal e Encargos Sociais’ (Personnel and Social Charges) were updated based on the figures provided by Gobetti and Orair (2017) from Brazil’s Central Bank reports. These adjustments were necessary to address methodological changes in the structure of the fiscal accounts.

A.1.2.3. Social benefits

The ‘Benefícios de Prestação Continuada (BPC)’ is a social benefit aimed at elderly and disabled low-income citizens, supporting more than 4.7 million people. For the line in the public accounts related to BPC spending (‘III.3.6 Benefícios de Prestação Continuada da LOAS/RMV’), we extended the series provided by the National Treasury to also cover the period from January 1997 to February 1999, using data collected by Gobetti and Orair (2017) from the ‘Anuário de Estatísticas da Previdência Social (AEPS InfoLogo)’. This extension of the series provides additional observations for estimating fiscal multipliers and simulating the effects of different fiscal adjustment policies that follow.

The Bolsa Família program (PBF) is Brazil’s main income transfer policy, supporting approximately 21 million families in 2023. We obtained data on social spending for the PBF from the National Treasury, covering the period from January 2008 to December 2023. To extend this series back to June 2001, we incorporate data from Gobetti and Orair (2017). For the years 2008 to 2017, we harmonize both data sources by adjusting values using a factor derived from the average expenditures with the PBF from both sources.

The lines ‘Abono Salarial’ (Wage Bonus), ‘Seguro Desemprego’ (Unemployment Benefit), as well as the line which records the total of both assistance benefits ‘III.3.1. Abono e Seguro Desemprego’ were adjusted using Orair and Gobetti (2017) data. We also used the authors’ data to adjust the lines ‘IV.4.1.1. PAC’, ‘d/q MCMV’ and ‘Min. do Des. Social’.

A.1.2.4. Subsidies

The implicit cost of BNDES is defined as the difference between the interest rate paid by BNDES to the National Treasury for borrowed funds and the Treasury’s borrowing cost. We used data from the Secretaria do Tesouro Nacional to estimate this ‘implicit subsidy’¹⁵. For the period from January 2001 to December 2017, we used the values calculated by Gobetti and Orair (2017). From 2018 onwards, we extended the series by incorporating spending data from the National Treasury reports. As the values are annual, we distribute them monthly with reference to the pace of BNDES resource disbursements, using sectoral disbursement data (BNDES, 2025) to reflect the speed of fund distribution over time.

A.1.2.5. Investments

We obtained data on investment spending from the National Treasury, covering the period from January 2007 to December 2023. To extend this series back to January 1997, we incorporated data from Gobetti and Orair (2017). For the years 2007 to 2017, we harmonized both data sources by adjusting values using a factor derived from the average expenditures with the PBF from both sources¹⁶.

Appendix B - GDP and control variables data

For the GDP series, we follow Sanches and Carvalho (2022) and use series 4380 from the Central Bank (BCB-Depec) (monthly GDP - current values) in monthly frequency as a measure of GDP. By using a monthly index for GDP, we conduct the analysis with monthly frequency series, all deflated by the IPCA. A similar approach was adopted by Sanches and Carvalho (2022) and Orair et al. (2016). This choice offers two advantages, as pointed out by Sanches and Carvalho (2022): i) it increases the number of observations; ii) the Brazilian institutional framework is regulated by the Fiscal Responsibility Law, whose targets and revisions are bimonthly. However, there are two disadvantages: i) the use of the IPCA to deflate the output data; ii) the use of an interpolated series estimated by the Central Bank.

As a robustness check, we present in Appendix E the analysis using GDP quarterly data from the National Accounts System from IBGE, deflated by the GDP deflator. As shown by Sanches and Carvalho (2022) and Sanches and Carvalho (2023), the estimated value of the multiplier is sensitive to variations in the empirical specification of the exercise, such as changes in frequency. Although in some cases there are larger variations in the point estimate, spending with high multipliers -public investment and social benefits- remains high and statistically significant at the 5% level. In contrast, other types of spending continue to show no relevant multiplier effect. In sum, our main conclusions are robust to changing the frequency of data.

In Appendix F we retrieve data from the Central Bank to include four variables as control: i) monthly variation of IPCA index (inflation rate) ; ii) interest rate (Selic) accumulated in a month, annualized; iii) index for real effective exchange rate; iv) commodities index. Following the Augmented Dickey Fuller for unit root test, the commodity price index and the exchange rate index are included as first-difference. Our main conclusions are also robust to including control variables¹⁷.

Appendix C - Results for the sample 1997-2023

Table C1
Multipliers

	Impact	Peak	Accumulated (15 or 25 months, depending on the persistence)
Total expenditure	-0.04	0.14	0.16
Total revenue	-0.45*	-0.06	-0.7
Public investment	0.9*	1.4*	2.43*
Social benefits	0.29*	0.91*	2.1*
Personnel	0.11	0.29	-0.51
Subsidies	-0.64	0.27	-0.06
Other expenditures	-0.1	0.31	0.12

• Source: prepared by the authors. (*) indicates significance at 5%.

Figure C1
Accumulated impulse-response functions of output to shocks in primary expenditures and revenues.

Appendix D - Results for the pre-pandemic sample (1997-2019)

Table D1
Multipliers

	Impact	Peak	Accumulated (15 or 25 months, depending on the persistence)
Total expenditure	0.14*	0.14	0.22
Total revenue	-0.45*	0.03	-0.6
Public investment	1.1*	1.84*	2.7*
Social benefits	0.47*	1*	2.25*
Personnel	-0.08	0.03	-1.23
Subsidies	-0.29	0.4	0.02
Other expenditures	0.11	0.11	-0.4

• Source: prepared by the authors. (*) indicates significance at 5%.

Figure D1
Accumulated impulse-response functions of output to shocks in primary expenditures and revenues

Appendix E - Exercise using quarterly data

Table E1
Multipliers

	Impact	Peak	Accumulated (5 or 8 quarters, depending on the persistence)
Total expenditure	0.02	0.85*	0.76
Total revenue	-0.37*	-0.09	-0.38
Public investment	2.2*	3.7*	3.5*
Social benefits	0.67*	2.5*	3.1*
Personnel	-0.32	0.94	0.38
Subsidies	-1.1	-1.1	-1.87
Other expenditures	-0.51	0.15	-0.18

• Source: authors’ elaboration. (*) denotes significance at the 5% level.

Figure E1
Accumulated impulse-response functions of output to shocks in primary expenditures and revenues

Appendix F - Exercise using control variables

Table F1
Multipliers

	Impact	Peak	Accumulated (15 or 25 months, depending on the persistence)
Total expenditure	0.1*	0.17	0.3
Total revenue	-0.42*	0.09	-0.33
Public investment	0.67*	1.1*	1.9
Social benefits	0.38*	0.78*	1.75*
Personnel	-0.16	-0.16	-2
Subsidies	0.04	1	1.5
Other expenditures	0.02	0.02	-0.3

• Source: authors’ elaboration. (*) denotes significance at the 5% level.

Figure F1
Accumulated impulse-response functions of output to shocks in primary expenditures and revenues

Appendix G - Residual analysis for baseline exercises¹⁸

Social benefits exercise:

6 lags included (AIC, HQ, FPE, LR criteria)

White heteroskedasticity test p-valor: 0.949

LM Autocorrelation test p-valor: 0.7717 (lag 1), 0.1247 (lag 2), 0.2027 (lag 3), 0.0516 (lag 4), 0.1311 (lag 5), 0.0462 (lag 6)

Stability: all roots of the characteristic polynomial are in the unit circle.

Public investment exercise:

6 lags included (AIC, FPE, LR criteria)

White heteroskedasticity test p-valor: 0.3329

LM Autocorrelation test p-valor: 0.8542 (lag 1), 0.7066 (lag 2), 0.6989 (lag 3), 0.7054 (lag 4), 0.4664 (lag 5), 0.8246 (lag 6)

Stability: all roots of the characteristic polynomial are in the unit circle.

Total expenditure/ total revenue exercise:

6 lags included (AIC, FPE, LR criteria)

White heteroskedasticity test p-valor: 0.5278

LM Autocorrelation test p-valor: 0.1751 (lag 1), 0.0790 (lag 2), 0.0576 (lag 3), 0.2964

(lag 4), 0.2277 (lag 5), 0.0244 (lag 6)

Stability: all roots of the characteristic polynomial are in the unit circle.

Personnel expenditure exercise:

6 lags included (AIC, FPE, LR criteria)

White heteroskedasticity test p-valor: 0.5191

LM Autocorrelation test p-valor: 0.3158 (lag 1), 0.6180 (lag 2), 0.2415 (lag 3), 0.4907 (lag 4), 0.2274 (lag 5), 0.2573 (lag 6)

Stability: all roots of the characteristic polynomial are in the unit circle.

Subsidies expenditure exercise:

6 lags included (AIC, FPE criteria)

White heteroskedasticity test p-valor: 0.2573

LM Autocorrelation test p-valor: 0.8496 (lag 1), 0.5087 (lag 2), 0.2867 (lag 3), 0.4238 (lag 4), 0.0822 (lag 5), 0.103 (lag 6)

Stability: all roots of the characteristic polynomial are in the unit circle.

Other expenditures exercise:

6 lags included (AIC, FPE, LR criteria)

White heteroskedasticity test p-valor: 0.7212

LM Autocorrelation test p-valor: 0.1802 (lag 1), 0.1463 (lag 2), 0.0944 (lag 3), 0.0965(lag 4), 0.0122 (lag 5), 0.0193 (lag 6)

Stability: all roots of the characteristic polynomial are in the unit circle.

Appendix H

Table H1
Effect of a 1% of GDP fiscal adjustment in three different scenarios on: GDP, primary deficit-to-GDP ratio, and net federal government debt-to-GDP ratio (domestic) - using alternative multipliers (from Appendix F)

	GDP variation	Primary deficit - GDP ratio	Debt - GDP ratio	Debt - GDP ratio (difference)	Deficit - GDP ratio (difference)
1. Adjustment via expenditures (constant revenue)
1.1. Cut in public investment	-1.7%	1.52%	59.3%	-0.28 p.p.	-0.59 p.p.
1.2. Cut in social benefits	-1.6%	1.49%	59.21%	-0.37 p.p.	-0.62 p.p.
1.3. Cut in subsidies	-1.3%	1.42%	58.9%	-0.68 p.p.	-0.69 p.p.
2. Adjustment via revenues (constant expenditure)
	-0.32%	1.19%	58.16 %	-1.42 p.p.	-0.92 p.p.
3. Adjustment via revenues (with an increase in expenditures)
3.1. Increase in public investment	1.38 %	1.76 %	57.78 %	-1.8 p.p.	-0.35 p.p.
3.2. Increase in social benefits	1.28%	1.84 %	57.9 %	-1.68 p.p.	-0.27 p.p.
3.3. Increase in subsidies	0.095%	1.95%	58.2%	-1.38 p.p.	-0.16 p.p.

• Source: prepared by the authors. Baseline: primary deficit-to-GDP of 2.11% and debt-to-GDP ratio of 59.58% (2023 data).

As we can see, the main conclusions remain unchanged despite variations in the estimated multipliers. In Scenarios 1.1 and 1.2, the debt-to-GDP ratios are closer to the observed value (59.58%), while Scenario 1.3 continues to yield the lowest debt-to-GDP ratio. Scenario 2 still results in a lower debt-to-GDP ratio than Scenarios 1.1 and 1.2, for example.

Finally, Scenario 3 continues to show GDP gains in cases 3.1 and 3.2, along with reductions of 1.8 and 1.68 percentage points in the debt-to-GDP ratio compared to the baseline. Scenario 3.3, although now displaying a positive effect on GDP, still shows a modest impact, and remains the scenario with the highest deficit-to-GDP ratio.

Edited by

• EDITOR-CHEFE
  Dante Mendes Aldrighi https://orcid.org/0000-0003-2285-5694
  Professor - Department of Economics University of São Paulo (USP)

• EDITOR ASSOCIADO
  Marcelo Milan https://orcid.org/0000-0001-7586-6528

Data availability

Os dados utilizados neste estudo estão disponíveis mediante solicitação ao autor. Dados adicionais e informações complementares também poderão ser fornecidos para fins de verificação ou replicação. A disponibilização está condicionada à inexistência de restrições de acesso público.

Publication Dates

• Publication in this collection
  03 Apr 2026
• Date of issue
  Jan-Mar 2026

History

• Received
  11 Sept 2024
• Reviewed
  26 June 2025
• Accepted
  01 Sept 2025