ABSTRACT In this paper, we provide new, updated estimates for Brazilian public sector’s structural primary fiscal balance. Our structural primary fiscal balance series differ markedly from unadjusted budget results. The numbers point to a tightening fiscal stance in the first part of the 2000s and an easing fiscal stance in the latter part of the decade. Our calculations also reveal a considerable fiscal effort in 2011 (up to Q3). According to our estimates, judging the fiscal stance only from the standpoint of unadjusted primary fiscal balance can lead to misleading conclusions about policymaking. Our results confirm Brazil’s pro-cyclical fiscal drive evidenced in the literature, with a negative correlation between the policy stance (measured by the structural balance) and the action of automatic stabilizers such as tax collection (measured by the cyclical balance). We also note a recurrent use of budget-enhancing one-off operations in times of fiscal consolidation or cyclical downturns. We conclude that the current policy setting – based on non-structural primary balance targets – produces a pro-cyclical fiscal policy bias: in booming years, it leads to overspending; in recession years, it leads to tightening and a search for extraordinary revenues. The current fiscal framework needs to include incentives to raise public savings in expansion years. The use of structural primary balance targets could make fiscal policy “lean against the wind”. In our view, this new policy setting would help boost public savings and investment, increasing potential GDP growth. The opinions expressed in this Working Paper are those of the author(s), not necessarily those of Itaú Unibanco. The author thanks Ilan Goldfajn and Samuel Pessoa for their comments, Aurelio Bicalho and Giovanna Rocca for the potential GDP and long-term commodity price series, as well as Ítalo Franca and Kim Vanderbilt for the editing. BRAZIL’S STRUCTURAL FISCAL BALANCE Mauricio Oreng* [email protected]* Economist - Itaú Macro Research Team
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
ABSTRACT
In this paper, we provide new, updated estimates for Brazilian public sector’s structural primary fiscal balance. Our structural primary fiscal balance series differ markedly from unadjusted budget results. The numbers point to a tightening fiscal stance in the first part of the 2000s and an easing fiscal stance in the latter part of the decade. Our calculations also reveal a considerable fiscal effort in 2011 (up to Q3). According to our estimates, judging the fiscal stance only from the standpoint of unadjusted primary fiscal balance can lead to misleading conclusions about policymaking.
Our results confirm Brazil’s pro-cyclical fiscal drive evidenced in the literature, with a negative correlation between the policy stance (measured by the structural balance) and the action of automatic stabilizers such as tax collection (measured by the cyclical balance). We also note a recurrent use of budget-enhancing one-off operations in times of fiscal consolidation or cyclical downturns.
We conclude that the current policy setting – based on non-structural primary balance targets – produces a pro-cyclical fiscal policy bias: in booming years, it leads to overspending; in recession years, it leads to tightening and a search for extraordinary revenues.
The current fiscal framework needs to include incentives to raise public savings in expansion years. The use of structural primary balance targets could make fiscal policy “lean against the wind”. In our view, this new policy setting would help boost public savings and investment, increasing potential GDP growth.
The opinions expressed in this Working Paper are those of the author(s), not necessarily those of Itaú Unibanco.
The author thanks Ilan Goldfajn and Samuel Pessoa for their comments, Aurelio Bicalho and Giovanna Rocca for the potential GDP and long-term commodity price series, as well as Ítalo Franca and Kim Vanderbilt for the editing.
2. Definitions and Literature Review.…………..…………......6 2-A. Structural Fiscal Balance: Definition, Applications, Limitations………..….6 2-B. The Literature on Budget Adjustment Applied to Brazilian Data………...…8
3. Methodology …………….……………………….……….….10 3-A. Estimating the Structural Fiscal Balance…………………..…………..……...10 3-A.1. The Aggregated Approach (IMF)………………………………………….......……………..….…..10 3-A.2. The Indirect, Disaggregated Approach (OECD)…..………………….…….………..…………….11 3-A.3. The Direct, Disaggregated Approach (ECB)………………………………………………............13
3-B. Pre-Modeling Choices: Budget Concept and Scaling…...……....…...……..14 3-C. The Dataset ……….....…………………………………………………….….…....15 3-C.1. Setting Up the Public Sector Budget Database …………......….….….….….….….….….….… 15 3-C.2. Adjusting the Budget Data……..…………..…..………………………………………………........16 3-C.3. Sampling: Revenue Classes, Tax Bases, Time Window ……………………………………..….20
3-D. Estimation Procedures……………………………...………….………..….…… 21 3-D.1. Overview…………………………………………..…………......…..….….….….….….….…..…… 21 3-D.2. Estimating Revenue Elasticity Parameters…..………………….………………………….….......23 3-D.3. Trend Filtering: Finding the Equilibrium Path for the Tax Bases .…….......……………….…….27
4-B. Structural Primary Fiscal Balance Results….…………………………….…..33 4-C. Dissecting Budget Results…………………………………………….………....35 4-C.1.The Structural Balance for Central Government and Regional Governments…......………...…35 4-C.2.Structural Fiscal Balance: Cyclical Balance vs. Structural Balance………………..………........36 4-D. The Fiscal Impulse……………………………………………….……….………..40 4-E. Simulating Alternative GDP and Commodity Trends.…………….………....43 4-F. Comparing Our Estimates with the Literature ………………..………..……..45
3-C.1. Setting Up the Public Sector Budget Database
We calculate the structural balance for the entire Brazilian public sector. Our sample uses inputs
from different data sources, including the Finance Ministry’s (―above the line‖6) and the Central
Bank’s official (―below the line‖6) statistics. This is necessary to overcome the scarcity of
information on revenues and expenditures at a sub-national level. To present our results, we use
the BCB’s fiscal accounts structure and fill the information gaps with other data, in order to make
our structural balance estimates fully comparable with the BCB’s official primary fiscal balance
statistics7.
As for the central government, the Treasury data on revenues and spending suffices for this
work.
For sub-national entities, however, the lack of data makes sampling quite cumbersome. There is
incomplete information on revenues, and no information on spending. We took a similar
approach to Gobetti et al (2010): we structurally adjust a subset of regional revenues, available
mostly at the states level. In order to create a relevant dataset of regional government revenues,
we use the National Treasury’s data on federal transfers and the figures on states’ tax receipts
published by the Finance Ministry’s National Council of Economic Policy (Confaz). Confaz
reports the collection of ICMS (value-added) tax, IPVA (car ownership) tax and other regional
government levies.
There are shortcomings in this strategy. First, our dataset only covers a part of total regional
revenues. We are leaving behind, for instance, some municipal taxes and regional government
firms’ dividends. Moreover, the data on states’ tax receipts is subject to frequent revisions,
making our structural regional revenues estimates relatively fluid in the very short term. Despite
these problems, this presents the best available data on a significant share of public sector
revenues outside the central government.
We calculate regional government expenditures as residual from the gap between our revenue
subset and the ―below the line‖ regional primary budget surplus calculated by the Central Bank.
Data restriction is even more acute for government-owned companies, where no data on
revenues or spending is available. This is a manageable problem, though, considering the low
share of this government sphere in the public-sector fiscal results – especially after the removal
of Petrobras and Eletrobras8 from the statistics. In fact, we decided not to structurally adjust the
primary balance from this entity, considering the lack of cyclicality in its headline budget results.
Importantly, our structural primary fiscal balance results are reported including the (unadjusted)
values for the statistical discrepancy of central government results and government-owned firms’
balance, which are clearly non-cyclical variables. The idea, as mentioned before, is to guarantee
comparability of our estimates with the official data.
6 In the ―above the line‖ criterion, the fiscal balance is calculated through the gap between revenues and spending. In the
―below the line‖ methodology, the fiscal balance is calculated through changes in net government indebtedness,
excluding the impact of the FX rate. For more details on Brazil’s fiscal statistics, refer to Gerin (2010). 7Our budget database is available on the following link:
http://www.itaubba-economia.com.br/content/interfaces/cms/anexos/ITABBA_WP_6_Annex.pdf 8 The Central Bank’s revised fiscal statistics stripped of Petrobras and Eletrobras start in December 2002. Here, we
appended to this series the one excluding only Petrobras, to enlarge our series history. From 2002 to 2009, the average
absolute gap between the ex-Petrobras and ex-Petrobras/Eletrobras series is small (0.06% of GDP).
As an example of problems in the RPP data, social security payments as of December 2008 were mistakenly added to the stock of RPPs (―restos a pagar processados‖). This event boosted the statistics by about BRL 21 billion in early 2009. Our database is already adjusted for this event. 16
More details on the database adjustment can be found in Appendix 2.
variables used a revenue base, showing our treatment of these series so as to prepare them for
the subsequent steps in the structural balance estimation.
As for the sample size, we estimate tax elasticity parameters for the period spanning 2002 to
201019
. Given the fluid lags involved in the transmission of macroeconomic conditions to tax
collection, we opted to use quarterly data in our analysis.
3-D. Estimation Procedures
3-D.1. Overview
We estimate Brazil’s structural fiscal balance with two alternative methodologies. Following the
IMF-styled aggregated approach, we structurally adjust total (cyclical) government receipts for
real GDP, retail sales and oil-price cycles. In the ECB-inspired disaggregated approach, we
separately adjust segmented revenue classes according to the cycles in numerous tax bases
working as proxy for business and financial conditions, such as industrial production and
manufacturing orders, wage bill, retail sales, imports, new loans and the CRB index.
There are two reasons for using two different methodologies to estimate the structural balance:
firstly, to guarantee robustness of the results; secondly, to test (and account for) the relevance of
output-composition effects in Brazilian budget results.
As shown in Appendix 4, the cycles of GDP and other activity subcomponents in Brazil are
moderately well correlated20
, suggesting some relevance of output-composition effects, yet to a
limited extent. In any case, the change in the composition of Brazilian economic growth over the
years – with domestic spending becoming significant as a driver – was a good reason to apply
the ECB methodology on Brazilian data for a first time.
Let i index public sector entities, with ―Fed‖ standing for the federal government21
, ―Reg‖ for
regional governments, ―Fir‖ for state-owned firms. We can re-write expression (1) as follows:
(8) f – fc = fs =ifsi = fs
Fed + fsReg + f Fir
Expression (8) shows the aggregation of the public sector’s structural primary fiscal balance
estimates. It results from the addition of structural fiscal results recorded by the central
government, regional governments (i.e., states and municipalities) and state-owned firms.
The absence of subscript in the term denoting government firms’ primary surplus means that we
are taking those as fully structural (or non-cyclical). We found no evidence of swings in state-
owned firms’ fiscal balance related to economic cycles (Appendix 5). This assumption has
negligible effect on our estimates, since annual government firms’ balance has averaged out at
0.13% of GDP since 2002 according to the new series stripped of Petrobras and Eletrobras. In
19
The choice of the time window took two criteria into consideration: (1) data scarcity: retail sales and bank lending series begin in 2000. Government-owned firms’ primary balance data, free of Petrobras and Eletrobras, begin in 2002; (2) series breaks: Chow breakpoint tests pointed to structural changes occurring early in the 2000sfor various revenue categories. The decision to pick 2002 as a cutoff date for all models meant to guarantee comparability of results across methodologies. 20
This exercise uses the average of our estimates for trend variables in the two filtering approaches that we tested: a
statistical one (HP Filter) and a ―theory-based‖ one, which we detail ahead. 21
Our structural balance calculations for the federal government include the statistical discrepancy between the primary
fiscal results calculated by the Treasury and the Central Bank (BCB). As discussed earlier, the objective is to maintain
the comparability between our structural balance estimate and the official BCB data. The statistical error has averaged
practice, our structural primary fiscal balance estimates for the public sector will reflect the fiscal
stance of the general government (i.e., central and regional governments).
Suppose a general government entity i has J types of cyclical revenues (i.e., subject to structural
adjustment) and K categories of non-cyclical revenues (i.e., not subject to structural adjustment).
Then, the structural primary fiscal balance for entity i will be calculated as:
J K
(9) fis = ri
s,j +ris,k - e
* j=1 k=1
Note that expression (9) changes slightly the notation of revenues and expenditures presented
in Section 3.A, where ―t‖ denoted revenues and ―g‖ expenditures. Here ‖r― means receipts after
the stripping out of one-off revenues (denoted by ot), and ‖e― means expenditures after the
database treatment (spending one-offs, denoted by og). Formally (for all revenues):
(10.A) r = t – ot | e = g – og (10.B) R = T – Ot | E = G – Og Just as in Gobetti et al (2010), we do not make structural adjustment to government spending.
Brazilian data points to a pro-cyclical pattern in unemployment benefit expenses – the only
expenditure class usually subject to cyclical adjustment (Appendix 5). This result, which
contradicts economic intuition, probably follows the increased job-market formalization and the
Brazilian government policies to boost the minimum wage.
We calculate the structural level of revenue category j using the following general formula:
W
(11) RAs,j = Rj
A .(Bjw* / Bj
w)
Rj,Bw w=1
where Rj
A is the seasonally-adjusted real level
22 of Rj, defined in expression (10.B). B
jw is the
level of the base for revenue j, already adjusted for seasonality and inflation, with Bjw* being its
long-term consistent values. Our notation for he index w is general enough to comprise the
representation of an activity variable or commodity price, in different lags. Importantly,
expression (11) considers a same variable in different lags as a different tax base, with structural
balance calculations reflecting exactly the same lag structure as the one used in elasticity
estimation models23
.
In formula (11), we assume that the structural amount of revenue j is a function of the ratio
between the revenue base’s equilibrium level to its observed value (accounting for the due lags).
This ratio is adjusted by the elasticity of revenue j with respect to the base. In the particular case
of non-cyclical receipts, the elasticity equals zero, implying that RA
s,j = RjA (i.e., structural
revenues equal actual revenues).
22
To present our estimates, we convert structural revenues back to nominal values and then calculate those as
percentage of nominal potential GDP. 23
In some empirical works, researchers structurally adjust revenues using the steady-state elasticity (i.e., adding lagged
elasticity parameter estimates). In our view, this simplification of the lag structure distorts structural balance estimates.
(e.g., cycles) harms the consistency of elasticity estimates. Since both use of levels or cycles as
regressands are consistent with the structural balance assumption represented by expression
(11) – if we assume that cyclical elasticities are unchanged in equilibrium – we opted for a
method that makes more sense econometrically.
The strengths of our procedures:(1) the simplicity of econometric procedures, improving
transparency of results, and (2) an artisan modeling approach focused on good large-sample
properties and economic validity of estimates. The main shortcomings of our approach: (1) the
failure to allow for endogenous regime30
changes and (2) the absence of steady-state
convergence rules31
.
Below, we present noteworthy specifics about the econometrics for each structural fiscal balance
methodology. Appendix 7 and Appendix 8 bring further details on all tax-revenue models (i.e.,
estimates, residuals and tests).
Aggregated Approach
The models for central government revenues (Model I.1) and total federal transfers (Model I.2)
boasted satisfactory results, with intuitive elasticity estimates. The results confirm the statistically
positive relationship between federal revenues and GDP or oil prices. The latter was the asset
price exhibiting the best explanation power for broad federal receipts (among the CRB index,
iron ore and soybean costs, as well as the terms of trade). In our view, the significant revenue
impact of oil costs reveals the influence of financial and commodity cycles on the budget
performance.
The models for transfers to states and municipalities also show a statistically positive
relationship between transfers and collection of income taxes. The proportionality reflected in a
unit-elasticity estimate suggests an accurate replication of Brazil’s revenue-sharing framework.
For these models, there was no need to use dummy variables to control for regime changes.
Results were not so good for the states’ tax models that used GDP as base (Model I.3).
Although the initially chosen model looks well-specified, with residuals neither serially correlated
nor heteroskedastic (signaling consistent estimates), robustness tests pointed to certain
30
Another criticism applying to every econometric approach estimating budget elasticity is the possible endogeneity of explanatory variables, as Bouthevillain et al (2001) point out. 31
We do not include error-correction terms in our regressions for two reasons: first, to avoid an excessive number of
parameters, which could damage to the stability of slope coefficients and the easy interpretation of results. Second,
considering recent changes in economic formalization and, possibly, in tax-collection technology, we believe Brazilian
tax collection must be far from a steady-state (as we know it).
The ―economic‖ de-trending approach entails heterogeneous methods that vary according to the
variable used as a base. As a general rule, these procedures are oriented by economic theory or
intuition, instead of statistical techniques. An illustrative example of what we call economic
filtering is the calculation of potential (or trend) GDP using a production function framework (in
this paper, we use the Itaú Macro Research Team’s series of potential GDP estimated through
the production function method).
For the other variables characterizing business conditions in the disaggregated methodology35
,
our strategy was more intuitive. We were inspired by numerous works in the literature, as
surveyed by Bornhorst et al (2011), using benchmarks set upon historical ratios to determine
equilibrium values for some bases. We also assume a cointegration relationship between GDP
and activity subcomponents, meaning that these variables should not drift away from each other
for too long.
The long-term consistent levels of these business-cycle variables will be set upon their average
ratios to real GDP for the time spanning 1Q01 to 3Q11. In formal terms, the equilibrium level for
tax base w (for any class of revenue, anytime) is calculated via the following expression:
(13) Bw* =wY* Where w is the average ratio of tax base w to real GDP in the sampling period, and Y* is the potential output estimated under the production function framework (to keep the coherence of this type of economic filtering). Expression (13) shows that the equilibrium level for any base (i.e., activity subcomponent) in any period will be determined by the product between potential GDP in the same period and the observed average ratio. As an example, the trend value estimated for retail sales in 1Q11 will be determined by the average historical ratio of retail sales to real GDP (for the time spanning 1Q01 to 3Q11) multiplied by the estimated level for real potential GDP in the same quarter.
Since our procedure defines a constant, linear relationship between the business cycle variables
used as tax base and trend output (at a rate determined by the average historical ratio), the
resulting equilibrium levels for activity subcomponents will accompany the moves in potential
output, which makes perfect sense economically. Moreover, despite possible fluctuations in the
share of consumption in total GDP, or in the participation of labor income in total income, the use
of average ratios doesn’t seem to be a strong assumption, given the relatively low quarterly
volatility of these ratios in our sample (Exhibit-5). The only exception is the imports’ GDP-ratio
series, the increased volatility of which reflects the currency movements (remember, we use the
inflation-adjusted BRL-value of imports)
Despite the statistical sufficiency of our method – as we update average ratios (and equilibrium
levels) as new data comes in –one shortcoming of this procedure is that it may take a while to
reflect sudden structural changes in the relationship between activity subcomponents and GDP
in the short run. However, this problem also applies for other de-trending procedures.
35
The set of variables include: wage bill, retail sales, imports (in BRL), new loans, industrial orders and output.
Owing to the (yet) greater uncertainty regarding the estimation of equilibrium levels for asset
prices, and also following some non-intuitive results obtained in our tries, we decided not to use
econometric filters to de-trend oil and commodity prices, in both methodologies.
To eliminate idiosyncratic distortions and at the same time reflect different movements captured
by each de-trending methodology, our structural fiscal balance estimates use a combination of
trend values obtained from both economic and statistical de-trending methods. That applies for
both aggregated and disaggregated methodologies. Most trend values will be determined by the
average between statistical and economic filters (with the exception of prices of commodities –
Brent oil and CRB indexes – which are only de-trended via statistical methods, the HP filter).
The final series for all tax bases used in our models are displayed in Appendix 9.
4. RESULTS
4-A. Elasticity Estimates
Our estimates from the aggregated methodology point to a steady-state36
GDP elasticity of
revenues of 1.48 for the public sector37
, resulting from an elasticity of 1.45 for the central
government and 1.55 for regional governments. Our estimate falls a tad below the values found
in Gobetti et al (2010): between 1.6 and 1.8 using the OLS approach and around 2.0 using
Markov-Switching models. The evidence of elastic revenues for the period of 2002 to 2011
reflects, among other factors, the rising formalization of the economy and the increased
contribution of domestic spending to total GDP growth.
36
Steady-state elasticity means the sum of elasticity coefficients for different lags of a same regressor, accounting for
possible lags of the dependent variable. It denotes the all-time impact of the explanatory variable on the regressand. 37
Since government firms’ primary balance is assumed to be non-cyclical (i.e., zero elasticity), the revenue elasticity for the public sector is the same as the one for the general government, which includes federal and regional governments.
Exhibit-7: General Government Budget Sensitivity to Output
Source: Itau, Mello and Moccero (2006), Girouard and Andre (2005)
These numbers suggest that Brazilian fiscal results are considerably sensitive to business
cycles. This probably reflects the country’s relatively high level of tax collection and public
expenses as share of the economy, compared with its emerging-market peers. Policy-wise, this
large budget cyclicality signals the need to create fiscal buffers to protect the budget from abrupt
reversals in economic conditions.
Our aggregated structural revenue models also show that tax receipts are significantly affected
by commodity prices, with the cost of oil capturing quite well the fiscal impact of fluctuations on
commodity prices and financial conditions38
. Our estimates point to a statistically significant
elasticity of broad federal revenues with respect to Brent oil price, equal to 0.08(Exhibit-8).
In our view, the lower revenue elasticity with respect to oil, compared with GDP, is due to two
facts: first, the higher volatility of raw material costs; second, Brazil’s relatively diversified
economy, with natural resources being only part of the business picture.
The significance of oil prices in public sector revenues contrasts with the fact that oil-related
receipts have accounted for no more than 5% of total federal takings in recent years. In our
interpretation, the significant impact of oil cycles on government revenues reflects more than just
the direct effects on dividends, royalties and corporate taxes. The high correlation of oil price
with other financial variables (such as equities and other commodity prices), suggests that oil
costs also work as a proxy for the broad indirect fiscal impact from wealth effects related to
commodity and financial cycles. The latter influences government revenues via consumer
spending, capital gains and corporate investment.
38
As pointed out in previous sections, we alternatively tested the CRB commodity index, prices of other commodities (iron ore, soybean) and the terms of trade as proxies for the impact of financial cycles on government budget results. We obtained the best results in terms of explanation power and residual properties using oil costs.
Exhibit-8: Activity Variables: Fluctuations Relative to Real GDP
Source: Itaú
Elasticity estimates from disaggregated models also bring interesting insights, basically
confirming the elevated cyclicality of key revenue categories in Brazil. Given the lack of previous
work applying the ECB’s approach to Brazilian data, we use as benchmark the estimates from
Bouthevillain et al (2001) for European Union economies39
. We compared estimates for social
security contributions, corporate income taxes, personal income taxes and indirect taxes.
As Exhibit-9 shows, our revenue elasticity estimates with respect to the most commonly used
bases stood just a tad below the average of European economies (see column ―One Base‖).
But our disaggregated revenue models incorporate more explanatory variables than our
European counterpart. Since we worked with structural adjustment instead of cyclical
adjustment, as in Bouthevillain et al (2001), we used commodity-price proxies as additional tax
bases for each tax category. When we add all tax elasticity estimates with respect to all
variables, just as an approximation (see column ―More Bases‖), it appears that the cyclical
sensitivity of various revenue classes in Brazil is larger than in Europe.
Exhibit-9: Disaggregated Elasticity Estimates – An International Comparison
Source: Itaú E.U. estimates taken from Bouthevillain et al (2001) * Refers to our long-term (or steady-state) revenue elasticity estimates. ** This column’s results mean a simple addition of cyclical elasticity coefficients with no discrimination for different explanatory variables. While these estimates are not addable, rigorously speaking, these numbers provide an indicative value for the total cyclical sensitivity of tax revenue
classes when we incorporate more tax bases into the models. *** Weighted average between elasticity estimates for federal sales taxes and states’ VAT.
39
While Bouthevillain et al (2001) estimate the cyclically adjusted fiscal balance using the same ECB methodology, there are important differences in the estimation of budget elasticity parameters compared to our work. The European application obtains elasticity parameters both via econometric analysis and derivation from the tax code. Additionally, Bouthevillain et al (2001) use a framework of cyclically adjusted balance, with no treatment for financial cycles’ impact on fiscal results.
1.48
0.08
0.00
0.50
1.00
1.50
2.00
GDP Oil Price
Elasticity of General Government Revenues With Respect to....
This subsection presents and discusses our Brazilian public sector’s structural primary fiscal
balance estimates for both aggregated and disaggregated approaches. A methodological
analysis of our results can be found in Appendix 10, as we explain the reasons for the final
configuration of models, which yielded the numbers shown below.
Exhibit-10 displays our estimates, comparing those with the official (unadjusted) data. In order to
eliminate idiosyncrasies from each methodology, we display our baseline estimate, which takes
the average between estimates under the aggregated and disaggregated approaches.
Exhibit-10: Structural Primary Fiscal Balance Estimates (vs. the Official or Unadjusted Data)
Source: Itaú (A) Twelve months to September 1/ A few disaggregated models structurally adjust revenues based on cycles of activity subcomponents with a short data series (e.g., retail sales, bank lending). In these cases, to estimate results for the year of 2000, we filled the information gaps with structural balance estimates using GDP as the revenue base. The GDP-elasticity results follow the re-estimation of the model with GDP replacing the respective tax base.
A first noteworthy point about the numbers is the relative robustness of our estimates across the
two different structural balance approaches. In the quarterly series, the maximum absolute gap
between our aggregated and disaggregated estimates is 1.2% of GDP, which occurs in late
2009 and early 2010. These data points should be seen as outliers – reflecting the impact of a
sudden increase in BNDES dividends after large loans from the National Treasury.
For the time spanning 1Q00 to 3Q11 (on a quarterly frequency), the average absolute gap
between aggregated and disaggregated estimates is 0.36% of GDP (or 0.30%, disregarding the
―outliers‖). For annual data (using a same time window), the difference stands below 0.2% of
GDP for about half of the years in the sample; for a third of those years, the gap is 0.3%–0.5% of
GDP; for only two calendar-years did the difference lie within 0.6%–0.8% of GDP. In all, these
numbers suggest that estimates from our aggregated and disaggregated methods are quite
The similarity of results under both approaches is favored by adaptations we made to apply the
aggregated and disaggregated methodologies to Brazilian data40
. These adaptations aimed at
improving the meaningfulness and robustness of the results. Additionally, the small estimate gap
also reveals a relatively low influence from output composition effects in Brazil’s structural
balance, confirming ex-ante expectations following pre-tests (refer to Appendix 4).
Our final results show that, for the period spanning 1Q00 to 3Q11, the average structural primary
balance recorded by Brazil’s public sector was 2.9%, a bit short of the unadjusted primary
balance average, 3.2%. This means that the average annual impact of cyclical fiscal revenues
(including one-off transactions) on budget results has been 0.3% of GDP since 2000.
The relatively small difference in the average of structural and official budget balance hides
periods of considerable decoupling between these series across the last decade (Exhibit-11).
Exhibit-11: Structural Fiscal Balance Estimates for Brazil’s Public Sector (Quarterly Data)
Source: Itaú
From 2000 to 2008, the unadjusted primary surplus stood around 3.5% of GDP, with a low
variance in this period. In the post-Lehman period, the observed primary fiscal balance suddenly
tumbled to 1.1% of GDP in 3Q09, and then rapidly initiated a staggered recovery: to 2.0% of
GDP in 4Q09, 2.8% in 4Q10 and 3.5% in 3Q11.
In fact, as opposed to a relatively stable path of the observed primary surplus up to 2008, our
baseline structural primary fiscal balance estimates point to significant changes in the fiscal
policy stance across the decade.
Our baseline numbers signal a significant contraction in the early days of Brazil’s fiscal
adjustment. As the official data show, the policy stance turned tighter in 1999 with the reversal of
primary fiscal deficits. According to our structural balance estimates, the fiscal contraction
continued into the first half of the 2000s and, after a temporary drop in late 2000 and early 2001,
40
Key examples are: the use of retail sales as a tax base in the aggregated approach and the inclusion of GDP in the set of revenue bases in the disaggregated methodology. Both turn our procedures a bit hybrid, as compared with the pure theoretical or empirical propositions by the IMF and ECB.
Exhibit-12: Structural Primary Fiscal Balance (Baseline) Estimates By Entities
Source: Itaú
It is possible that the incomplete data on regional governments’ budgets has some influence on
this high correlation of structural balance across general government entities. Still, in our view,
the similarity between the fiscal policy stance of central and regional governments reflects some
aspects of the fiscal framework currently in place. One example is the exposure to similar
budgetary shocks – as federal and regional government share an important amount of tax
collections. Another factor is the limited room for fiscal policy maneuvering by states and
municipalities in the wake of the Fiscal Responsibility Law, enacted in 2000. This new legislation
left regional governments more dependent on decisions taken at the federal level, which seems
to have successfully improved intra-government fiscal policy coordination. Refer to Appendix 12
for more details on the structural fiscal balance registered at the regional level.
4-C.2. Structural Fiscal Balance: Cyclical Balance vs. Structural Balance
This subsection breaks down the primary fiscal balance, revealing the impact of cyclical drivers
on Brazil’s budget performance recently. For the sake of simplicity and intuition, we work only
with aggregated methodology estimates. Appendix 13 shows results using baseline estimates.
The structural fiscal balance (SFB) – which is the share of budget results consistent with output
and asset prices in equilibrium – measures the fiscal policy stance. The cyclical fiscal balance
(CFB) – which reflects the impact of business and financial cycles on budget results –reveals the
impact of automatic stabilizers (e.g., tax collection), also interpreted as a passive fiscal policy
response41
.
41
Since we estimate the CFB as residual (i.e., observed balance minus structural balance), our calculations for the CFB will be influenced by variables not necessarily linked to the cycles of activity and commodity prices. The key examples: (1) non-recurring fiscal transactions; (2) small distortions generated by the use of potential GDP (instead of actual GDP) in the scaling of structural budget result.
0.0%
0.5%
1.0%
1.5%
2.0%
2.5%
3.0%
3.5%
20
00
-III
20
01
-III
20
02
-III
20
03
-III
20
04
-III
20
05
-III
20
06
-III
20
07
-III
20
08
-III
20
09
-III
20
10
-III
20
11
-III
1.0%
1.5%
2.0%
2.5%
3.0%
3.5%
4.0%
4.5%
5.0%
Federal Government (left) Regional Governments (left) Public Sector (right)
Exhibit-14.A: Structural vs. Cyclical Primary Fiscal Balance Estimates (Aggregated Estimate)
Exhibit-14.B: Contributions to the Cyclical Primary Fiscal Balance (Aggregated Estimate)
Source: Itaú * For some years, contributions shown in Exhibit-14.B may not add, because we use nominal potential GDP as deflator (instead of actual nominal GDP, as in the official series). The annual impact stands within 0.1% of GDP for the entire time window.
The results show that, for the years of 2000–2001, the budgetary impact of economic cycles
added to nearly1% of GDP, accounting for a third of the official (unadjusted) average primary
budget results. About 60% of the cyclical budget balance in that period stems from concession
takings, which were used to improve fiscal results at the time. As explained earlier, these
revenues were stripped out of our structural budget database. The remaining 40% of the cyclical
balance reflects favorable activity and oil price cycles estimated for the year 2000. Our estimates
show a reasonable structural balance in the early 2000s – averaging at 2.3% of GDP –indicating
a tightening fiscal policy stance (see Exhibit-14.A and Exhibit-14.B).
budget results (as targeted) became a less-difficult task, paving the way for a less-tight fiscal
stance. As an upshot, the structural balance dropped to 3.4% by end-2006.
Economic conditions for Brazil kept improving in 2007–2008. According to our trend estimates,
GDP and oil price gaps were largely positive in this period, signaling that the overheating
created significant tailwinds for budgetary results. There was a positive cyclically led fiscal
balance of 0.8% per year, out of which 0.6% stemmed from an overheated economy and 0.2%
from a commodity price spike. In our calculations, 2008 was the year when the commodity cycle
helped Brazilian budget results the most, adding around 0.3% of GDP.
As the cyclical push from automatic stabilizers made the achievement of a constant fiscal target
a good deal easier, especially compared with the uphill situation faced in the beginning of the
decade, the government continued to ease the fiscal stance: the structural balance fell to 2.5%
of GDP in end-2008, the lowest level since 2001.
In the post-Lehman period, policymakers further eased the fiscal stance to avert local
contamination from the global recession. In 2009, the structural balance declined by 0.5% of
GDP42
(to 2.0%), and a local slowdown brought about a lower cyclical balance. In fact, the action
of automatic stabilizers was the main reason for the decline in the observed surplus to 2.0% of
GDP in 2009, from 3.4% in 2008.
An important feature of Brazil’s fiscal policy in the after-crisis is the use of non-recurring budget
operations to improve fiscal results, especially in 2009 and 2010. Based on these temporary
budget-enhancing transactions, the government maintained a relatively fast pace of expenses.
In our calculations, the impact of one-off fiscal operations rose to 0.6% of GDP in 2009 and
peaked at 1.4% of GDP in 2010. One-off revenues and accounting events provided the greatest
contribution for the rise in the official primary fiscal balance. The latter touched 2.8% of GDP in
2010 (from 2.0% in 2009), reflecting the impact of the activity cycle on tax collection – +0.8% of
GDP in 2010 (from -0.5% in 2009) – and despite the 1.2% decline in the structural primary
surplus (reaching 0.8%).
In 2011, fiscal policy changed gears. Most of the increase in the observed primary budget
balance – to 3.2% of GDP in 3Q11 from 2.8% in 4Q10 – was due to a tighter fiscal stance by the
federal government. As an upshot, the public sector’s structural balance rose by 1.3 percentage
points, to 2.1% of GDP, according to our aggregated methodology estimates. While one-off
revenues remained at a relatively high historical level (0.5% of GDP) and although the impact of
activity drivers on automatic stabilizers remained considerable (0.6% of GDP), the total cyclical
contribution to the primary budget result in 2011 was lower than in 2010.
In sum, the breakdown of the official primary budget balance (into structural and cyclical
components) confirms the negative relationship between the fiscal policy stance and the action
42
This relatively timid budgetary action, reflected in a smooth decline in the structural fiscal balance, contrasts with bolder off-budget fiscal interventions. The main example is the significant Treasury loans to BNDES, as the government decided to use a mix between fiscal and quasi-fiscal instruments to stimulate the economy in 2009.
Exhibit-15.B shows that a lax stance towards expenditures was the key element behind the
decline in the structural balance, as estimated for 2004 to 2010.Our calculations show that the
average fiscal impulse from public-sector spending was 0.6% of GDP per year in this period.
That contrasts with an average drag of 0.1% per year on the revenue side resulting, for instance,
from a tightening in Cofins rules in 2005 and an ―easing‖ due to the expiration of the CPMF tax in
2008.Our calculations show that the spending impulse was especially strong in 2005–2006 and
2009–2010: it was worth about 1.0% of GDP annually for these years.
We also broke down the fiscal impulse by government entities, in order to observe the fiscal
stance of different public-sector spheres and how these added to the overall structural balance.
Exhibit-16 shows that the fiscal contraction as of 2001–2003 was evenly shared by the federal
and regional governments: both contributed to the fiscal consolidation by about 0.5% of GDP per
year. These numbers contrast with a fiscal expansion of 0.3% of GDP registered by government-
owned firms43
.
Federal and regional governments were also equally accountable for the reduction of the
primary structural fiscal balance in subsequent years. The fiscal impulse averaged 0.5% per
year for the period of 2004–2010, with each government level recording an average annual
fiscal impulse of about 0.25% of GDP in that period. However, in 2011 (up to Q3), there was a
decoupling in the fiscal impulse from federal and regional governments, as the central
government tightened the fiscal stance and prompted a fiscal drag of 1.4% of GDP. Meanwhile,
regional governments eased the policy stance by about 0.1% of GDP.
The numbers on the fiscal impulse by government entities highlight the phenomenon described
in section 4-C.1– the high correlation of structural results by federal government, and states and
municipalities. In our view, this reflects a greater policy coordination following the Fiscal
Responsibility Law (2000), which limited the room for fiscal maneuvering at the regional level.
Exhibit-16: The Fiscal Impulse by Government Entities (Aggregated Estimate)
Source: Itaú
43
The fiscal expansion of state-owned companies in 2001–2002 might bea spurious one, reflecting the removal of Petrobras and Eletrobras from the fiscal statistics in 2002. These companies used to add around 0.3%–0.5% of GDP per year to the consolidated primary fiscal surplus.
We also tested scenarios combining these alternative paths for equilibrium GDP and oil costs
altogether. The outcome is a 3Q11 structural balance estimate ranging from 0.5% to 3.5% of
GDP, with a similar range (i.e., around 3pps) valid for almost the entire sampling window.
Overall, the simulations show that alternative paths of potential GDP growth and oil price may
have a significant impact on structural balance estimates. In this exercise, we conclude that a
2.5% increase (decrease) in potential GDP levels may produce a change of one percent rise
(fall) in primary structural fiscal balance estimates. For oil price, a 35% rise (drop) in equilibrium
Brent costs may upwardly (downwardly) impact structural balance estimates by 0.6% of GDP.
The simulations reveal that structural balance estimates are more dependent on equilibrium
GDP levels than on oil price trends. The greater GDP sensitivity is quite intuitive, as Brazil has a
relatively well-diversified economy, where raw materials play a significant role but are far from
accounting for the bulk of the story.
4-F. Comparing Our Estimates with the Literature
Considering the limited amount of published work on Brazil’s structural fiscal balance in recent
years, we take Gobetti et al (2010) as a benchmark to our results.
Our structural primary fiscal balance estimates are qualitatively similar to the ones obtained by
those researchers, despite some important differences in key procedures followed in this paper.
For the period of 1Q00–2Q10, the shape of the structural balance curve estimated in Gobetti et
al (2010) resembles our baseline results (Exhibit-19). The absolute gap of mean estimates for
the general government is, on average, 0.4% of GDP.
Our methodology points to more pronounced fiscal policy cycles and impulses. There are
significant differences in specific years. For 2000–2001, for instance, we calculate that the
structural primary balance for the general government (in our case, using baseline estimates44
)
was 1.4% of GDP, while estimates in the benchmark paper stand around 2.2%. This gap likely
reflects our decision to remove concession revenues from the structural balance database.
44
As explained earlier, our baseline estimates refer to the average between the structural primary fiscal results obtained from aggregated and disaggregated methodologies.
The Net Budgetary Impact of Database Adjustments: Annual Data
Sources: Press sources, Brazilian Sovereign Wealth Fund, Brazil Revenue Service, National Treasury, Itaú
(A) Past twelve months; (B) Year-to-date (up to September) * Positive values mean upward budget impact after the adjustment; negative values mean lower results after the adjustment. ** Data on spending deferrals (“restos a pagar processados”) start in 2000, so that we assume zero deferrals for 1997 to 1999. *** We correct the spending deferrals series for a distortion affecting the data from 2009 onwards, which relates to a mistaken computation of INSS pension outlays in the last month of 2008 as a delayed payment. We subtract BRL 21.2 billion from the stock of budget deferrals registered in the official data for each year since 2009. The value of this adjustment is equivalent to the total pension outlays in December 2008.
For more on our budget database, refer to the following link: http://www.itaubba-economia.com.br/content/interfaces/cms/anexos/ITABBA_WP_6_Annex.pdf
Year GDP
R$ billions % GDP R$ billions % GDP R$ billions % GDP R$ billions % GDP R$ billions
Appendix 4 –GDP and Activity Cycles: Case for Disaggregating?45
According to Bornhorst et al (2011), the correlation and synchrony between the gap in output and in other activity variables is an early signal for the possible presence of output-composition effects in budget results and structural balance estimates. If those cycles are highly correlated, the results from the aggregated approach should be similar to those from the disaggregated methodology. The graphs below point to a statistically significant, yet moderate correlation between the cycles of GDP and variables such as the wage bill, retail sales, bank lending, imports (in BRL), industrial production, and industrial orders
46. On average, the contemporary correlation between these cycles
is 0.48, with largely unequal pair-wise results (see graphs below). An example: the correlation between GDP and industrial output cycles is 0.82, while the correlation between GDP and wage cycles is 0.20. At first, the data suggest that output-composition effects should have a moderate impact on Brazil’s structural fiscal balance estimates. But these correlations signal that it may be worth trying the ECB method for Brazilian data. The final outcome confirmed expectations of limited composition effects.
Cycles in Real GDP and Other Activity Subcomponents
Sources: BCB, IBGE, CNI, Funcex, Itaú
45
The cycles displayed in these charts are estimated via statistical and economic filters. 46
All variables are in volume or inflation-adjusted terms.
-4.0%
-3.0%
-2.0%
-1.0%
0.0%
1.0%
2.0%
3.0%
4.0%
mar/01 mar/03 mar/05 mar/07 mar/09 mar/11
-12.0%
-9.0%
-6.0%
-3.0%
0.0%
3.0%
6.0%
9.0%
12.0%
GDP cycles (left) Retail sales cycles (right)
correlation: 0.35
-4.0%
-3.0%
-2.0%
-1.0%
0.0%
1.0%
2.0%
3.0%
4.0%
mar/01 mar/03 mar/05 mar/07 mar/09 mar/11
-12.0%
-9.0%
-6.0%
-3.0%
0.0%
3.0%
6.0%
9.0%
12.0%
GDP cycles Wage cycles
correlation: 0.20
-5.0%
-4.0%
-3.0%
-2.0%
-1.0%
0.0%
1.0%
2.0%
3.0%
4.0%
5.0%
mar/01 mar/03 mar/05 mar/07 mar/09 mar/11
-40.0%
-30.0%
-20.0%
-10.0%
0.0%
10.0%
20.0%
30.0%
40.0%
GDP cycles (left) Cycles of imports in BRL (right)
Independent variable further lagged by one quarter (relative to the original model)
*** Crosses the 1% significance level | ** Crosses the 5% significance level | * Crosses the 10% significance level
Disaggregated Approach Models (Class II) Estimation Method: Least Squares All models use Newey-West HAC Standard Errors & Covariance RESET
48 test up to 2 lags
Chow breakpoint and forecast tests applied to regressions without the correction for structural breaks by the dummies Structural breaks likely caused by the increase in job formality since 2004 #
All Independent variable further lagged by one quarter (relative to the original model)
*** Crosses the 1% significance level | ** Crosses the 5% significance level | * Crosses the 10% significance level
Chow breakpoint and forecast tests applied to regressions without the correction for structural breaks by the dummies Structural breaks likely caused by the impact of changes in the Cofins tax code from 2004 onwards #
Some independent variables with a one-quarter lead (relative to the original model)
*** Crosses the 1% significance level | ** Crosses the 5% significance level | * Crosses the 10% significance level
48
The RESET tests the null hypothesis of no specification problems. By the same token, Breusch-Godfrey and ARCH
tests verify the null hypothesis of no joint serial correlation or conditional heteroskedasticity in the residuals.
Some independent variables with a one-quarter lead (relative to the original model)
*** Crosses the 1% significance level | ** Crosses the 5% significance level | * Crosses the 10% significance level
#
All independent variables with a one-quarter lead (relative to the original model)
*** Crosses the 1% significance level | ** Crosses the 5% significance level | * Crosses the 10% significance level
Chow breakpoint and forecast tests applied to regressions without the correction for structural breaks by the dummies Structural breaks likely caused by the IPI tax exemptions in the post-Lehman Brazilian recession #
Independent variable with a one-quarter lead (relative to the original model)
*** Crosses the 1% significance level | ** Crosses the 5% significance level | * Crosses the 10% significance level
Chow breakpoint and forecast tests applied to regressions without the correction for structural breaks by the dummies Structural breaks possibly caused the global crisis (slowing global demand causing rising imports penetration) #
Independent variable further lagged by one quarter (relative to the original model)
*** Crosses the 1% significance level | ** Crosses the 5% significance level | * Crosses the 10% significance level
Chow breakpoint and forecast tests applied to regressions without the correction for structural breaks by the dummies Structural breaks possibly caused by macro-prudential measures aimed at slowing capital flows (increasing sensitiveness to financial conditions) #
Some independent variables further lagged by one quarter (relative to the original model)
*** Crosses the 1% significance level | ** Crosses the 5% significance level | * Crosses the 10% significance level
Chow breakpoint and forecast tests applied to regressions without the correction for structural breaks by the dummies Structural breaks possibly caused by Treasury loans to BNDES, which boosted the bank’s size and also the volume of dividends paid #
Independent variable with a two-quarter lead (relative to the original model)
*** Crosses the 1% significance level | ** Crosses the 5% significance level | * Crosses the 10% significance level
#
Independent variable with a one-quarter lead (relative to the original model)
*** Crosses the 1% significance level | ** Crosses the 5% significance level | * Crosses the 10% significance level
Chow breakpoint and forecast tests applied to regressions without the correction for structural breaks by the dummies Structural breaks possibly caused by a search for extra revenues in the after-Lehman period #
Independent variable with a one-quarter lead (relative to the original model)
*** Crosses the 1% significance level | ** Crosses the 5% significance level | * Crosses the 10% significance level
#
Independent variables further lagged by one quarter (relative to the original model)
*** Crosses the 1% significance level | ** Crosses the 5% significance level | * Crosses the 10% significance level
#
Independent variable further lagged by one quarter (relative to the original model)
*** Crosses the 1% significance level | ** Crosses the 5% significance level | * Crosses the 10% significance level
# Independent variable further lagged by one quarter (relative to the original model)
*** Crosses the 1% significance level | ** Crosses the 5% significance level | * Crosses the 10% significance level
# Independent variable further lagged by one quarter (relative to the original model)
*** Crosses the 1% significance level | ** Crosses the 5% significance level | * Crosses the 10% significance level
Dummy Index: find below the definition of the dummy variables used in models to control for structural breaks (in most cases, reflecting changes in the tax code):
DUMMY2003 = {―1‖ from 2003 on; ―0‖ otherwise}
DUMMY2004 = {―1‖ from 2004 on; ―0‖ otherwise}
DUMMY2004Q2 = {―1‖ from 2Q04 on; ―0‖ otherwise}
DUMMY2007 = {―1‖ from 2007 on; ―0‖ otherwise}
DUMMY2009 = {―1‖ from 2009 on; ―0‖ otherwise}
DUMMY2009Q3 = {―1‖ from 3Q09 on; ―0‖ otherwise}
DUMMY’09EXPT49
= {―1‖ from 1Q09 to 1Q10; ―0‖ otherwise}
DUMMY2010 = {―1‖ from 2010 on; ―0‖ otherwise}
49
This dummy was designed to capture the period of tax exemptions on manufactured products (especially cars) during
Appendix 10 – Assessing the Results From a Methodological Perspective
In this section, we assess the main features of our structural primary fiscal balance estimates for the Brazilian public sector. The properties of our results under each approach and method justify the decisions about the final configuration of models leading to the estimates reported in Section 4. We evaluate our framework’s performance from two standpoints: (1) the robustness of results across structural balance approaches and de-trending methods; and (2) the efficacy in stripping out cyclical components from the raw data.
We departed from a ―purer‖ framework for the IMF’s aggregated methodology, using GDP as the base for all revenue models. We also estimated the structural balance in both approaches using both statistical and economic de-trending techniques for activity and commodity cycles. To calculate a theory-based equilibrium for oil and commodity prices (in the economic filtering), we used the output from Itaú’s commodity-price models, assuming global GDP at potential for the whole period. The results from this initial setting are shown in Graph (A).
Graph (A): Structural Balance Estimates by Approach, De-trending Technique: The First Try
Source: Itaú
A first impression from the chart is that all estimates point to a similar pattern for the evolution of the structural balance across the 2000s: a tightening stance early in the decade, a lax policy drive in its latter part, a partial recovery in 2011. However, these estimates showed very large amplitude of results (i.e., gap between maximum and minimum estimates) – around 2.0%–2.5% of GDP for the period between 2003 and 2007. A good deal of this amplitude is due to the fact that estimates from the disaggregated method using economic de-trending behaved as true outliers. Two changes were made to overcome this problem. Firstly, we adapted the aggregated
approach to use retail sales as a tax base in the states’ revenue model (replacing GDP, as
proposed in the ―pure‖ IMF methodology). Second, we decided not to apply the economic filter
on commodity prices, as unintuitive cycles for raw material prices resulted from this procedure,
helping distort (to the upside) structural balance results.
The changes made produced more narrowly distributed results, as the disaggregated methodology with economic filtering posted lower structural balance estimates, more in line with economic intuition (and estimates from the other alternative methodologies and procedures). In fact, the amplitude for the period of 2003 to 2007 was reduced by about one percent, ranging around 1.0%–1.5% of GDP. Thus, the adaptations further improved the robustness of our results.
Graph (B): Structural Balance Estimates by Approach, De-trending: The Final Setting
Source: Itaú
Despite the similarity of estimates across structural balance techniques, we noted that the results from the aggregated methodology proved less dependent on the filtering techniques than the disaggregated approach. This is a natural result, as de-trending is more challenging for activity subcomponents than for GDP. There are more established theoretical procedures to estimate broad output equilibrium (e.g., the production-function approach) than methods to de-trend activity subcomponents. To further improve the robustness of our estimates, we use what we called a ―combined filter‖,
which means the average of equilibrium values estimated via statistical and economic filters51
.
As Graph (B) shows, the use of average equilibrium estimates prompted a convergence of
results: the amplitude fell sharply, to an average of 0.4% of GDP for the whole period.
Graph (C): Structural Balance Estimates: IMF and ECB Approaches with “Combined” Filtering
Source: Itaú
The relatively low discrepancy of results from aggregated and disaggregated methodologies using average trend estimates (i.e., the ―combined‖ filter) gave us the ―green light‖ to report the average of estimates from both methodologies as our baseline result. While simplicity would call for the use of the aggregated methodology only–since it is much easier to estimate – using a
51
As previously mentioned, in the case of oil prices and the CRB index, we only deploy the statistical de-trending (HP filter).
combination of both approaches allows our results to also reflect possible output-composition effects, an exclusive feature of the disaggregated methodology. Another important property to observe in our estimates is how ―cycle-free‖ our structural balance estimates have turned out, which signals the efficacy of our procedures. Graph (D) relates the official public-sector primary surplus data (as well as our own structural primary fiscal balance estimates) to real GDP growth, so as to identify cyclical influences. The linear regressions shown in the charts illustrate the relationship between these series, showing the GDP influences on the official data and our structural balance estimates. For the period spanning 2001 to 2009, the observed quarterly data point to a large correlation between activity conditions (as measured by real GDP growth) and the official primary budget results, as shown in the large slope (0.22) and R
2coefficient (0.59). Our structural balance
estimates under both approaches have successfully removed this cyclical component present in the unadjusted data, as suggest the zeroing of the R
2 coefficients and the brisk reduction of
slopes (to a range of -0.01–0.08). In all, the numbers show that not only are our structural balance results relatively robust across
methodologies, but they are also efficient in removing cyclical components of the unadjusted
series (for a period showing this influence). Overall, the final results were quite satisfactory.
Appendix 12 – Structural Fiscal Balance by Government Entities Structural Primary Fiscal Balance Estimates
Source: Itaú (A) Past twelve months to September 1/ A few disaggregated models structurally adjust revenues based on cycles of activity subcomponents with a short data series (e.g., retail sales, bank lending). In these cases, to estimate results for the year 2000, we filled the information gaps with structural balance estimates using GDP as the revenue base. The GDP elasticity used in these results follow the re-estimation of the model with GDP replacing the tax base.
2/ Public sector results account for the (non-cyclical) discrepancy between the BCB and Treasury’s estimate for central government balance.
Graph A: Relationship Between STRUCTURAL Primary Balance Estimates Across Government Entities (%GDP) – 2000 to 2011 (Quarterly Data)
Source: Itaú
Graph B: Relationship Between OBSERVED (Unadjusted) Primary Balance Estimates Across Government Entities (%GDP) – 2000 to 2011 (Quarterly Data)
Appendix 13 – Budget Balance Breakdown and Fiscal Impulse Details Using Baseline Estimates A comparison between graphs shown in Exhibit-14.A (or B) and Exhibit-14.Ax (or Bx), as well as Exhibit-15.A (or B) and Exhibit-15.Ax (or Bx), shows that the decomposition of the budget balance (between structural and cyclical components) as well as the breakdown of fiscal impulses are quite robust to the use of aggregated or baseline estimates. In sections 4-C.2 and 4-D, we use results obtained from the aggregated methodology, so as to make intuition clearer. In this section, we present the same class of results using our baseline estimates.
Exhibit-14.Ax: Structural vs. Cyclical Primary Balance Estimates (BASELINE ESTIMATE)
Exhibit-14.Bx: Contributions to the Cyclical Primary Fiscal Balance (BASELINE ESTIMATE)
REFERENCES Afonso, José Roberto R.; November 2011, ―A redução dos juros pelo Banco Central diminuirá no mesmo ritmo o custo da dívida do governo?‖, mimeo (only in Portuguese) Bevilaqua,Afonso; Werneck, Rogerio L.F.; October 1997; ―Fiscal Impulse in the Brazilian Economy, 1989-1996‖, PUC-Rio Economics Dept., Discussion Paper No.379 Bezdek, Vladimír;Dybczak, Kamil; Krejdl, Ales; 2003, ―Cyclically Adjusted Fiscal Balance – OECD and ESCB Methods‖, Czech Journal of Economics and Finance, 53 Blanchard, Olivier; April 1990, ―Suggestions for a New Set of Fiscal Indicators‖, OECD Economics and Statistics Department, Working Paper No. 79 Blanco, Fernando; Herrera, Santiago; September 2006, ―The Quality of Fiscal Adjustment and the Long-Run Growth Impact of Fiscal Policy in Brazil‖, World Bank, Policy Research Working Paper No. 4004 Bornhorst, Fabian; Dobrescu, Gabriela; Fedelino,Annalisa; Gottschalk, Jan; Nakata, Taisuke; April 2011, ―When and How to Adjust Beyond the Business Cycle? A Guide to Structural Fiscal Balances‖, IMF Technical Notes and Manuals (Washington: International Monetary Fund) Bouthevillain, Carine; Cour-Thimann, Philippine; van den Dool, Gerrit; Hernandez de Cos, Pablo; Langenus, Geert, Mohr, Matthias; Momigliano, Sandro; Tujula, Mika; 2001, ―Cyclically-Adjusted Budget Balances: An Alternative Approach,‖ ECB Working Paper No. 77 (Frankfurt: European Central Bank). Eyzaguirre, Nicolás; Kaufman, Martin; Phillips, Steven; Valdés, Rodrigo; April 2011, ―Managing Abundance to Avoid a Bust in Latin America‖, IMF Staff Discussion Note (Washington: International Monetary Fund) Fedelino, Annalisa; Ivanova, Anna; Horton, Mark; November 2009, ―Computing Cyclically Adjusted Balances and Automatic Stabilizers‖, IMF Technical Notes and Manuals (Washington: International Monetary Fund) Garcia, Márcio G.P.; August 2007, ―Dívida Pública e Reservas Cambiais‖, mimeo (only in Portuguese) Girouard, Nathalie, and Christophe André; 2005, ―Measuring Cyclically Adjusted Budget Balancesfor OECD Countries,‖ OECD Economics Department, Working Paper No. 434, (Paris: OECD) Gerin, Diretoria de PolíticaEconômica do Banco Central do Brasil; March 2010, ―IndicadoresFiscais‖ (Fiscal Indicators), Central Bank of Brazil’s Frequently Asked Series, Central Bank of Brazil, http://www.bcb.gov.br, (only in Portuguese) Gobetti, Sergio Wulff; Gouvêa, Raphael Rocha; Schettini, Bernardo Patta; December 2010, ―Resultado Fiscal Estrutural: Um Passo Para a Institucionalização de Políticas Anticíclicas no Brasil‖ (Structural Fiscal Balance: A Step Towards The Institutionalization of Anti-cyclical Policies in Brazil), Discussion Paper No. 1515, IPEA Instituto de Pesquisa Econômica Aplicada, http://www.ipea.gov.br, (only in Portuguese) Goldfajn, Ilan; Bicalho, Aurélio; February 2011; ―A Longa Travessia Para a Normalidade: Os Juros Reais no Brasil‖; Textos Para Discussão no. 2; Departamento de Pesquisa Macroeconômica do Itaú Unibanco, (only in Portuguese)
Hamilton, James D.; 1994; ―Time Series Analysis‖; Princeton University Press Hagemann, Robert; July 1999, ―The Structural Budget Balance: The IMF’s Methodology‖, IMF Working Paper No. 99/95 (Washington: International Monetary Fund) Marcel, Mario; Tokman, Marcelo; Valdés, Rodrigo; Benavides, Paula; December 2001, ―Balance Estructural: La Base de La Nueva Regla de Política Fiscal Chilena‖ (Structural Balance: The Basis of The New Chilean Fiscal Policy Rule), Economía Chilena, Vol. 4 No.3 (pages 5-27), Central Bank of Chile, http://www.bcentral.cl, (only in Spanish) Mello, Luiz de; Moccero, Diego; 2006, ―Brazil’s Fiscal Stance during 1995-2005: The Effect of Indebtedness of Fiscal Policy Over the Business Cycle‖, OECD Economics Department Working Paper No. 485, http://www.oecd.org (OECD, Paris). Noord, P. van den; January 2000; ―The size and role of automatic fiscal stabilizers in the 1990s and Beyond‖; OECD Economics Department Working Paper No. 230 (OECD, Paris). Rocha, Fabiana; Setember 2009; ―Política Fiscal Através do Ciclo e Operação dos Estabilizadores Fiscais‖, Revista Economia – Anpec, Vol.10, No.3, p.483-499, Brasília (DF), (only in Portuguese)