Public Spending and Economic Activity: Quasi- Experimental Evidence from Brazilian Municipalities

Transcription

1 Public Spending and Economic Activity: Quasi- Experimental Evidence from Brazilian Municipalities Raphael Corbi Elias Papaioannou Paolo Surico August 2013 Abstract According to Brazilian law, federal transfers to municipal governments vary discontinuously at numerous pre-determined population thresholds. Using changes in local population around these thresholds as an instrument for actual public spending in a fuzzy regression discontinuity set-up, we identify the causal effect of fiscal policy on economic activity. Our baseline estimates point to an average local multiplier between 1.6 and 2 across a range of empirical specifications that control for fixed municipal characteristics, national business cycle, monetary policy, and federal fiscal policy. The average effect, however, masks significant heterogeneity. An exogenous increase in government spending of 1% of municipal GDP tends to generate an insignificant boost in local activity, with point estimates of 0.9% to 1.3%. In contrast, the multiplier associated with a public spending cut appears highly significant, with central values between 2.5 and 3. Contractions in government spending on transportation, housing and, to a lesser extent, health and education appear the most detrimental for the local economy. JEL codes: E62, H72, C26. Keywords: local government spending multiplier, quasi-experiment, fuzzy RD. London Business School: rcorbi.phd2008@london.edu London Business School, NBER and CEPR: epapaioannou@london.edu London Business School and CEPR: psurico@london.edu 1

2 1 Introduction Six years have passed since the outbreak of the Great Recession and the public debate on the effects of government interventions is still far from settled. The on-going crisis in the countries of the European periphery and the implementation of large fiscal adjustment programs has given new poignancy to the opponents of austerity plans and major policy institutions, most recently the IMF, have revised upward their estimates of the size of the fiscal multiplier (see Blanchard and Leigh, 2013). Amainreasonbehindthelackofconsensusamongeconomistsandpolicymakersliesinthe fact that fiscal policy is highly endogenous to the business cycle. Fiscal policy simultaneously shapes and is shaped by economic conditions. Furthermore, its effects are often anticipated and tend to peak with delays of several quarters. As such, it seems hard to identify the causal impact of government spending on economic activity, as recently emphasized by Perotti (2007), Hall (2009), Ramey (2011a) and Auerbach and Gale (2010). This paper contributes to the debate on the effects of government interventions using quasi-experimental variation in fiscal policy from a large emerging country, Brazil. We isolate the impact of exogenous shifts in municipal government spending on local economic conditions exploiting a peculiar feature of the Brazilian law: federal transfers to the local authorities change discontinuously (and thus locally exogenously) at fixed pre-determined population thresholds. While the official number of inhabitants in any particular municipality is measured every ten years by the Brazilian Census, an independent federal agency the Instituto Brasileiro de Geografia e Estatistica (IBGE) provides annual estimates of local population in-between Census years. As these estimates are used by the central government for the allocation of federal transfers to municipalities, their year on year variation across the legislated thresholds represents an ideal source of exogenous changes in municipal spending to quantify their effects on local economic activity. Besides the attractiveness of a ( fuzzy ) Regression Discontinuity (RD) approach to iden- 2

3 tification, the richness of the allocation mechanism of federal transfers in Brazil brings into our analysis three additional benefits. First, the large number of municipalities that jump from one population bracket to another during the 2000s (about 2, 500 episodes or 10% of the sample) implies that one can exploit within-municipality variation and thus account for time-invariant factors that may shape local economic growth and local expenditure or that may be the target of other federal interventions (e.g. specific transfers for economically disadvantaged areas). Second, our sample includes many episodes of municipalities moving to a higher population bracket hence receiving a higher amount of federal funds in per capita terms and, on average, spending more as well as many cases of municipalities moving to a lower population bracket thus receiving a smaller amount of federal transfers in per capita terms. This allows us to investigate the asymmetric response of output growth to both increases and decreases in public spending. Third, since the Brazilian law governing federal transfers to municipalities exhibits discontinuities at numerous population thresholds, the local (at the discontinuity-) effects identified by the RD estimates are more likely to capture the overall effect of fiscal policy on local output growth. In addition, the Brazilian database we use reports government expenditure not only at the aggregate municipal level but also disaggregated by its main categories. As such, we can explore potentially interesting heterogeneity based on the type of public spending. Results Preview. Our regression discontinuity analysis highlights three major features of the data. First, the average effect on local economic activity of an exogenous increase in public spending of 1% of municipal GDP varies from 1.6% to 2% across a range of empirical specifications which, over and above municipality fixed factors, also control for variation in the national business cycle, monetary policy and federal fiscal policy. Second, the local multiplier appears highly asymmetric. When we focus on episodes of fiscal expansions (triggered by an increase in population estimates), the point estimates for the multiplier range from 0.9 to 1.3 which, in most specifications, are statistically indistinguishable from zero. In contrast, the 3

4 effects associated with fiscal contractions (triggered by a decrease in population estimates) are highly significant and considerably larger, with point estimates between 2.5 and 3. Third, the most detrimental effect on local activity comes from cuts in spending on transportation and housing, followed by health and education. Over and above the novel identification of local fiscal multipliers, the focus on Brazil is of independent interest for at least two reasons. First, the empirical literature appears dominated by studies on the United States and the evidence on emerging markets is very scant, at best. Our study thus provides fresh evidence on the causal effect of fiscal policy on a large developing country that, in many respects, differs considerably from the U.S. and other advanced economies. Second, Brazil is currently the most prominent economy in South America and the sixth largest economy in the world (ranking already above Canada, Italy and the United Kingdom), thereby making the evidence presented here of high policy relevance. Related literature. Our work is mostly related to the voluminous literature that assesses the role of fiscal policy on macroeconomic performance. One strand of contributions has used time series at the national level, exploiting for identification either aggregate sources of exogenous variation, mostly wars, or imposing zero and sign restrictions in structural vector autoregressions. 1 Given the identification challenges of these approaches, it comes at no surprise that this body of research has produced a wide range of estimates for the aggregate multiplier. Thus, a second (and growing) strand of the empirical literature has turned to more disaggregated analyses, creatively exploiting variation in public spending across regions within a country. Our paper fits into this line of research, which has mostly looked at states or counties within the United States. Following the approach introduced by Barro (1981), Nakamura and Steinsson (2013) interact state-level military procurement and spending with country-level exogenous changes in military buildups to identify the impact of fiscal shocks 1 See Parker (2011) and the references inthere, including Blanchard and Perotti (2002), Perotti (2007), Mountford and Uhlig (2009), Barro and Redlick (2011), Ramey (2011b), Auerbach and Gorodnichenko (2012), Caldara and Klemps (2012) and Ilzetzki, Mendoza and Vegh (2013), among many others. As for taxes, Romer and Romer (2010) have pioneered the approach based on narrative records of legislated policy changes. 4

5 on state output. Shoag (2010) uses variation in the idiosyncratic component of US states portfolio of defined-benefit pension plan asset returns as an instrument for local spending. Serrato and Wingender (2011) exploit fund reallocations driven by revisions to unanticipated local population revisions. Overall these studies report point estimates for the local multiplier in the neighborhood of two. As for recent government interventions, Feyrer and Sacerdote (2012) use state variation in the seniority of the U.S. Congress representatives as instrument for local government expenditure, reporting average multipliers below one and some heterogeneity across spending categories. Chodorow-Reich, Feiveson, Liscow and Woolston (2012) exploit pre-crisis variation on medicare/medicaid allocations and find larger effects of the 2009 American Recovery and Reinvestment Act on employment. It would seem hard to extrapolate the findings from the earlier contributions above to other countries and estimates of the (local) fiscal multipliers outside the U.S. are rare. There are, however, a handful of notable exceptions. Acconcia, Corsetti and Simonelli (2013) exploit exogenous cuts in public spending triggered by the dismissal of Italian province governments suspected of mafia infiltration and find government spending multipliers just below two. Kraay (2012) focuses on a sample of twenty-nine low-income countries whose borrowing from the World Bank financed a substantial fraction of government spending. Using the fact that fluctuations in World Bank financed spending in a given year are largely determined by fluctuations in project approval decisions made in previous years, he identifies spending multipliers in the proximity of 0.5. Brückner and Tuladhar (2013) exploit geographical variation within Japanese prefectures and report average estimates that are not statistically different from one. They also show that firms financial distress (economic slack) has a significantly negative (positive) effect on local public spending multipliers. While our identification strategy is new in the fiscal policy literature, a growing body of empirical research in political economy has exploited the allocation mechanism of funds from the central government to local authorities in Brazil. Recent work focuses, among others, on 5

6 the role of federal transfers on poverty and education (Litsching and Morrisonn, 2013), test scores (Corbi, 2013), and corruption (Brollo, Nannicini, Perotti and Tabellini, 2013). Using a strategy similar to ours, though based on the different population thresholds that determine local politicians pay, Ferez and Finan (2011a) study the role of monetary incentives on politicians quality (see also Ferez and Finan, 2008 and 2011b). Caselli and Michael (2012) assess the impact of oil-related revenues and find that government spending does increase with oil discoveries, thought social welfare does not improve. 2 Institutional design and data Institutional framework. The Federative Republic of Brazil is organized at three different levels of government: the Union (federal level), 26 states (the regional level), and 5, 565 municipalities (the local level). The executive and legislative powers are organized independently at all three spheres of government, while the judiciary system is organized only at the federal and regional levels. Municipal governments are managed by an elected mayor (Prefeito) andanelectedcitycouncil(camara dos Vereadores), which are in charge of asignificantshareoftheprovisionofpublicgoodsandservices,relatedtoeducation,health and infrastructure. Brazilian municipalities have limited ability to raise taxes, which on average correspond to only 6% of total local revenues in our sample. This makes their funding highly dependent on fiscal transfers from the states and the federal government. 2 Amajorroleisplayedby the automatic federal fiscal transfers from the Fundo de Participacao dos Municipios (FPM), which accounts for about 33% of total revenues and amounts to almost 80% of all federal transfers. Municipal governments are required by law to spend at least 15% of these transfers on education and at least another 15% on health care, while the remaining share is employed at the discretion of the local authorities. At the federal level, the pool of resources for the 2 In an interesting recent study, for instance, Brollo and Nannicini (2013) show that local governments receive relatively higher federal transfers in pre-election years when the local mayor is aligned with the party of the federal government and the President. 6

7 FPM fund comes from 22.5% of total revenues raised by the federal income (IR) tax and the industrial products (IPI) tax. Established in its current form by the Federal Constitution of Brazil of 1988 (Art. 159 Ib), the FPM is allocated to each municipality according to a predetermined mechanism that rests on local population and the state which the municipality belongs to. First, each of the 26 states is assigned with a fixed share of total funds. Second, each municipality in a given state is assigned a coefficient that depends on its population in a highly discontinuous fashion. Let FPM k i be the FPM transfers received by municipality i in state k. The allocation mechanism is described by: FPMi k = FPM k i P i k where FPM k is the amount of resources allocated to state k and i i is the FPM coefficient of municipality i based on its population size. Table 1 reports the population brackets and the associated FPM coefficients. The Tribunal de Contas da Uniao (TCU, Union Budget Court) assigns a coefficient to each municipality on the basis of the population estimates for the previous year. These are calculated annually by an independent statistical agency, the IBGE - Instituto Brasileiro de Geografia e Estatistica (Brazilian Institute of Geography and Statistics). The population estimates are made public by October 31 st and the municipal budget is typically approved by the second half of December. The IBGE method to estimate municipal population takes into consideration past censuses, regional birth and death rates as well as migration trends so as to make local population estimates consistent with state-level and federal-level estimates. 3 Examples. Acoupleofexamplescanhelpillustratethediscontinuousnatureofthe allocation mechanism that we exploit for identification. Consider two municipalities of similar size in the state of Sao Paulo, Queluz and Vera Cruz. The municipality of Queluz had an 3 Data on FPM transfers come from the Brazilian National Treasury (Tesouro Nacional). Population estimates, municipal output-gdp (and its sectoral components) and municipal public expenditure (and its subcategories) are retrieved from IPEA (Instituto Pesquisas Economica Aplicada). The Brazilian Institute of Statistics (Instituto Brasileiro de Geografia e Estatístic) provides a detailed description of the FPM program. 7

8 estimated population of 10, 148 in 2006, just below the first threshold of 10, 189: hence,queluz received a total of BRL 1, 882, 048 in the following year. In 2007, the local population estimate increased by 175 inhabitants to 10, 323 and thus Queluz moved to the second population bracket: hence, total transfers increased considerably to BRL 3, 004, 393 (a rise of 60% in just one year). The municipality of Vera Cruz, on the other hand, had a population estimate of 11, 117 in 2006: hence, Vera Cruz was within the second population bracket and received transfers for BRL 2, 513, 691 in the following year. In 2007, its population decreased to 10, 020, and thus Vera Cruz just fell below the first threshold (of 10, 189): this led to a 10% fall of federal transfers to BRL 2, 253, 295. Sample. Our yearly data cover the period Because of sample size limitations, following Brollo, Nannicini, Perotti and Tabellini (2013), we focus on the seven lower thresholds in Table 1 and restrict the analysis to municipalities with population below 47, 545. These account for about 34% of Brazilian population, which was close to 190 million in Given the wider brackets at higher cutoffs, we consider municipalities whose population is between 1, 698 inhabitants below the first threshold (i.e. with total population at or above 8, 491)and3, 396 inhabitants above the seventh threshold (i.e. with total population at or below 47, 545). Our working dataset comprises an unbalanced panel of 25, 800 municipality-year observations, covering 2, 791 distinct municipalities. Technical Issues. It should be noted that the population estimates from IBGE in a given year do not always predict perfectly the FPM transfers that each municipality receives in the following year. This mis-assignment of funds can have many causes, including the fact that throughout the 1990s many municipalities split into two but managed to keep their former FPM coefficient (through court disputes), although they should have been automatically reduced because of the smaller population. As discussed in Brollo, Nannicini, Perotti and Tabellini (2013), in order to correct such distortions, the federal government established that by 2008 all municipalities should be framed in the correct population brackets with their 8

9 relevant coefficients. To avoid immediate disruption to the public finances of the involved municipalities, however, the law established a transition period to the new regime, so that in the period some municipalities still received FPM transfers that were not consistent with their population. Finally, a new population census was organized in 2007 so that the relevant municipalities could have new population counts. This led to the number of municipalities that changed population brackets in 2008 being higher than the average number over the rest of the sample. We have thus verified that our findings are not sensitive to year 2008 (and also the period ), whose population changes, while unrelated to the local business cycle, may have been anticipated. The results of this exercise are reported in the penultimate table of the Appendix. Summary statistics. Table 2 illustrates the richness of the experiment. Panel A shows that during the period, there have been 671 episodes of municipalities moving to alowerpopulationbracketand1, 856 episodes of municipalities moving to a higher bracket. 4 Panel B shows that most of the upward or downward jumps regarded relatively smaller cities, falling within the first three thresholds (i.e. with population up to 22, 000). There have been also many discontinuous jumps around threshold 4 (around 30, 000 inhabitants). Finally, given the smaller number of larger cities, there are fewer jumps around thresholds 5, 6, and 7. According to Panel C, only 92 of all 2, 791 municipalities in the sample do not change population bracket at any point in the period, and 1, 022 experienced at least one positive and one negative jump. In Table 3, wereportthenumberofobservationsoneither side of the seven lower thresholds per year. Table 4 reports descriptive statistics on the main variables (at the municipal level) used in the empirical analysis. As shown in Panel A, average municipal GDP growth for the period was strong (around 6.3%). During this period, the average GDP growth was 3.5% at the national level and was 2.9% and 2.0% in the two largest cities, Sao Paulo and 4 The larger number of positive jumps should not come as a surprise as during the period under examination Brazil experienced strong population growth (see Table 4). 9

10 Rio de Janeiro respectively, which account for almost 20% of the population in the country. On average, Brazilian municipalities received around 3, 647 thousand Reais at 2000 prices (i.e. BRL 8, 267, 227 or USD 3, 720, 252 at 2013 prices), which roughly correspond to around one third of average public expenditure. For a typical municipality, services was the largest economic sector, accounting for around 57% of local GDP. Smaller municipalities exhibit agriculture participation of above 20% of GDP and industry around 15%. The pattern is reversed for municipalities with more than 30, 000 inhabitants, where the agriculture (industry) share of GDP is about 15% (20%). In Table 5, we report total expenditure broken down by economic nature (current vs capital spending) and by categories. The composition of expenditure does not seem to vary systematically with population size. Current expenditure accounts for some 85% of total, leaving 15% to capital investment. The main categories of government spending are education (42%), health (30%) andpublicadministration(20%). 3 Econometric strategy In this section we describe the (fuzzy) regression discontinuity (RD) design that allows us to isolate the causal effect of municipality spending on local activity. We begin by describing the source of exogenous variation and then present the empirical specifications. 3.1 Identification The allocation of federal transfers to municipal governments is a function of population (and since we exploit within municipality variation also of population growth). While population (and population growth) is likely to depend on local economic conditions, transfers change discontinuously at some arbitrary population thresholds which are pre-determined by law. This implies that movements across the cutoffs governingthesizeoftheper-capitafiscal transfers can be used as a source of (locally) exogenous variation to estimate the causal effects of public spending on local economic activity at the discontinuity. The main idea 10

11 behind our identification design is that public spending increases (decreases) sharply when population in a given municipality goes above (below) the cut-off level. On the one hand, transfers are a highly non-linear discontinuous function of population with known thresholds. On the other hand, the exact allocation of federal resources may also depend on other factors, including the state a municipality is located and manipulative sorting, among others. Following Angrist and Lavy (1999), Hann, Todd and van der Klaauw (2001), van der Klaauw (2002) and numerous subsequent works (e.g., Ferez and Finan, 2011, and Brollo, Nannicini, Perotti and Tabellini, 2013), we therefore employ a fuzzy regression discontinuity approach that accounts for the fact that population thresholds may not be the only determinant of federal transfers. Specifically, we use the legislated cutoffs drivingthe discontinuous allocation of federal funds as an instrument for local spending, so as to identify the impact of municipal spending on local output growth. The regression discontinuity design relies on two identifying assumptions. First, as typical in RD designs, federal transfers and local spending should be the only factors that change discontinuously around the various population thresholds. Second, the local authorities should not manipulate systematically municipal population estimates so as to receive more transfers from the federal government. Along the lines suggested by Imbens and Lemeiux (2008), previous works exploiting the discontinuities of federal transfers across Brazilian municipalities provide strong support for both assumptions (see Brollo, Nannicini, Perotti and Tabellini, 2013). In addition, as our identification strategy relies on comparing the effect of transfers within the same municipality at different points in time, there is little scope for other (non-policy) municipal characteristics to change abruptly in the neighborhood of the population cut-off. 5 On the second assumption, we have examined whether there is evidence of clustering at 5 Exploring an alternative constitutional threshold at 10, 000 inhabitants that restricts mayor s pay, Ferez and Finan (2011a) find a discontinuity on mayor s characteristics at this threshold. The last table in the Appendix shows that our results are not driven by the first FPM threshold at 10, 189, which is quite close to the one shaping local politicians wages. 11

12 levels of population just above the discontinuity. Figure A in the Appendix suggests that there is little bunching of municipalities just above each of the seven thresholds, where the per-capita transfers become discontinuously larger. This is not surprising, as the population estimates come from an independent federal agency and not from local (or state) administrators. 6 Furthermore, in the spirit of McCrary (2008), we formally test for the presence of adensitydiscontinuityattheseseventhresholdsbyrunningkernellocallinearregressionsof the log of the density separately on both sides of each threshold and for each year. Figure B shows there is no evidence that municipalities sort just above - as opposed to just below the thresholds - except for year 2008, when the new population census took place. 3.2 Empirical specifications The effects of fiscal transfers on public spending and local economic activity are estimated using variants of the following empirical specifications which link public spending and local economic activity to changes in population across the thresholds that determine the amount of transfers from the federal government to each municipality: G i,t = f(p i,t )+g( P i,t )+ FS JUMP i,t + i + t + " i,t (1) Y i,t = f(p i,t )+g(. P i,t )+ RF JUMP i,t + i + t + u i,t (2) The dependent variable in the first equation, G i,t,representsthechangeinrealper-capita public expenditure for municipality i in year t over per-capita municipal GDP in year t-1. The second specification focuses on real per-capita output growth (Y i,t ). The function f(.) is a high-order RD polynomial in population (the forcing variable in the RD), which accounts flexibly for movements around each cut-off point. Since we exploit within-municipality variation, we also include a third-order polynomial in population growth to account for dynamics of population changes (g(. P i,t )). We estimate two versions of the specifications (1) and (2): 6 Litschig (2012) detects some evidence of manipulative sorting around the FPM thresholds in the TCU population figures for the years 1989 and

13 one with a common across all thresholds RD polynomial and another with threshold-specific RD polynomials. In the baseline estimation, JUMP i,t is a trichotomous variable that takes the value of +1 when the population of municipality i goes above a legislated threshold between year t-1 and year t; the value of 1 when the local population falls below a given threshold between year t-1 and year t, and the value of 0 when a municipality remains in the same population bracket of the previous year. Both specifications (1) and (2) include a set of municipality fixed effects, i, and year fixed effects, t. The former capture all time-invariant factors that determine changes in local economic activity and possibly local spending which are related to geography, ecology, culture, and other pre-existing conditions; the latter account flexibly for time shocks and trends that are common to all municipalities, such as changes in monetary policy, federal fiscal policy, and national business cycle. To further account for unobserved heterogeneity, we also estimate specifications including state-specific time trends or allowing for state-specific time fixed-effects. Given the inclusion of the RD polynomial (that accounts flexibly for changes in municipal population around cutoffs), the estimated FS on the trichotomous JUMP i,t variable in equation (1) identifies the (First Stage) effect of locally exogenous changes in fiscal transfers on changes in local spending relative to GDP. Likewise, the coefficient RF in equation (2) reflects the (Reduced-Form) effect on output growth of changes in population close to the locally exogenous thresholds. It should be noted that, unlike earlier contributions exploiting the allocation of federal resources across municipalities in Brazil to study other political outcomes, our (fuzzy) RD design makes the identification strategy in this paper closer to a difference-in-difference/discontinuity estimation. 7 Specifically, we compare average change in local spending and economic performance associated with municipalities that stayed in the same population bracket over two consecutive years (the control group ) with the average change associated with municipalities 7 More precisely, we exploit a difference-discontinuity. Grembi, Nannicini, and Troiano (2013) adopt a similar approach to study the role of relaxing fiscal restraints on municipality debt levels in Italy. 13

14 that moved to a higher (or lower) population bracket (the treatment group ). Following Angrist and Levy (1999) and van der Klaauw (2002), equations (1) and (2) can be jointly estimated in an Instrumental Variable (IV) set-up, which isolates the effects on local activity of a change in public spending stemming from a change in transfers triggered by a jump across population thresholds (fuzzy RD). More specifically, the IV model reads: Y i,t = f(p i,t )+g(. P i,t )+ IV Ĝ i,t + i + t + v i,t (3) where Ĝi,t denotes the predicted public spending based on the first-stage specification (1). The estimated IV measures the change in local activity caused by a change in public spending as large as 1% of municipal GDP. This is our estimate of the local government spending multiplier. Following Brollo, Nannicini, Perotti, and Tabellini (2013), we estimate all three specifications (first-stage, reduced-form, and IV) either by pooling data across all (seven) thresholds or by allowing the coefficients on the polynomial terms and the jump index to differ at each (of the seven) threshold. We do so by interacting the polynomial terms of the function f(p i,t ) with the index JUMP,usingindicatorsthatcaptureeachofthesevendiscontinuities. To this end, we define the midpoints separating each pair of thresholds and then we assign each municipality in a year to a single discontinuity. Finally, to investigate asymmetries in the effects of fiscal expansions and fiscal contractions, we also estimate variants of the empirical models above using two separate dummies for upward and downward changes in population across the exogenous thresholds. 4 The average local multiplier This section presents the results of the fuzzy RD approach that identifies the local government spending multiplier. We begin by reporting the results of the first stage estimation linking federal transfers and public spending to the discontinuous movements of municipalities across the population brackets that shape federal transfers. Then, we present the reduced- 14

15 form specifications that associate output growth with changes in local population across the legislated thresholds. Finally, we quantify the average effect of public spending on local economic activity by combining the reduced-form estimates with the first stage estimates in an instrumental variables setting. Structure of the presentation. The first row of Tables 6 8 reports the results from variants of specification (1), pooling together observations across all seven population brackets. The estimates displayed in the second and third rows allow for a different coefficient on the JUMP variable for smaller municipalities (i.e. thresholds 1 to 3, with population range 8, , 377) andlargermunicipalities(i.e. thresholds4 to 7, with population range 20, , 454). The remaining rows record results based on threshold-specific coefficients on the JUMP index. While it is interesting to examine heterogeneity across the various population thresholds, one should bear in mind that the more flexible specifications in the tables trade-off generality with efficiency as the number of municipalities decreases rapidly at higher population brackets. Following the recommendation in Imbens and Lemieux (2008), we experiment with various RD polynomial functions. In columns (I)-(III), we include a third, forth, and fifth order RD polynomial in the distance of local population from the closest discontinuity. In column (IV), we add a threshold-specific third-order RD polynomial which effectively allows the impact of population to differ in the neighborhood of each discontinuity. For comparability, in column (V) we report a parsimonious specification without any RD polynomial. And to account further for state-specific dynamics, columns (VI) and (VII) include state-year fixed effects and state-specific linear time trends (over and above year fixed effects), respectively. 4.1 First-stage estimates Identification requires that federal transfers and (most importantly) public spending increase (decrease) when municipalities move up (down) across different population brackets. Table 15

16 6 (7) reportstheprojectionsoftheannualchangeinrealper-capitafederaltransfers(real per-capital government spending) relative to municipal per-capita GDP on the trichotomous JUMP variable according to equation (1). Federal transfers at the discontinuities. Across all model permutations reported in Table 6, thecoefficient on the JUMP index is positive and highly significant. In line with the federal transfer mechanism, the estimates imply that whenever local population jumps to a higher population threshold transfers increase substantially (in per capita terms). The point estimate (around 0.009) impliesthatapositive(negative)moveinpopulationbracketleads to an increase (decrease) of approximately 16% of the FPM transfer/gdp ratio observed over the period. As shown in the second and third rows, and further detailed in the remaining rows, this effect is somewhat stronger for smaller municipalities which depend more heavily on federal transfers (as opposed to local tax revenues). Figures 1a 1b provide a graphical illustration of this relationship. The seven vertical lines represent the FPM population thresholds. Small light-colored dots represent municipalityyear observations and larger dark dots represent federal transfers averaged over population bins of 200 inhabitants. There are clear jumps whenever local population moves across one of the seven population brackets shaping the resource allocation. Figure 1a shows clearly that total transfers increase with population; most importantly for our identification strategy, there are clear jumps when a city s population moves across the seven population brackets that shape allocations. Figure 1b illustrates the relationship between transfers per capita and population using all municipality observations across all years. The discontinuities on the amount of transfers per capita when a municipality jumps (or falls) a particular threshold are evident and stark. 8 Figure 2 gives a RD-style graph that illustrates the estimates shown in Table 6. We pool 8 Some noise around each threshold is visible in Figure 1, pointingtopossiblecasesofmis-assignment. As discussed in Section 2, this may originate from municipalities that broke up during the 1990s. The discontinuities of federal transfers around the legislated population cut-offs arerobusttorestrictingthe sample to either or

17 all observations (municipality-years) across the seven cutoffs, where FPM transfers change discontinuously. Each dot gives the average value of federal transfer growth in the year after a particular municipality has moved to a higher population bracket for municipalities grouped in bins of 200 inhabitants. 9 On the left of the vertical line that illustrates the discontinuity, we plot the averages of observations below the cutoff (where population takes negative values), while on the right we plot the averages of observations above the cutoff (where population takes on positive values). The figure also plots predicted values of federal transfer growth from a third order RD polynomial on distance (in population terms) from the discontinuity fitted separately for observations above and below the cutoff. The jump on federal transfer growth at the discontinuity is larger, around 0.07, andhighlysignificant. Public spending at the discontinuities. Identification requires that municipal spending varies with the change in federal transfers generated by crossing a legislated threshold. In Table 7, we test this hypothesis. The coefficient on the JUMP index in the first row that summarizes the average effect across all thresholds is positive (around 0.065) and highly significant across all specifications. This implies that a positive (negative) jump leads to an increase (decrease) of approximately 6.5% of the average municipal expenditure / GDP ratio. In analogy to the findings for transfers, the increase (decrease) in local spending associated with a higher (lower) population bracket tend to be somewhat larger for smaller municipalities, around 0.08, which is 50% higher than the coefficient for larger municipalities, where the estimate is around 0.05 in the third row. Figure 3 gives the RD-style graph that illustrates the discontinuous jump of government spending when a municipality population jumps to a higher FPM bracket. Each point gives the average of public spending growth for cities averaged over bins of 200 inhabitants away from any of the seven discontinuities. The figure also graphs predicted values (and associated 95% confidence intervals) from a third-order polynomial on the distance of each municipal 9 Since we explore within municipality variation in all RD figures we plot average transfer growth after partialling out municipality fixed effects as well as year fixed effects. 17

18 population from the closest cutoff. There is a sharp and evident jump on local spending growth at the discontinuity. The magnitude (0.07) is also quite similar to the formal RD estimates in Table 7. Summary. The RD results reported in Tables 6 and 7 and the corresponding graphical illustrations in Figures 2 and 3 suggest that the JUMP index, that captures the discontinuous allocation of federal funds across municipalities, represents a source of locally exogenous variation which can be used to identify the causal effect of public spending on output growth. 4.2 Reduced-form estimates In Table 8, we present estimates from the reduced-form specification (2), which associates real per-capita GDP growth at the municipality level with changes across the legislated populated thresholds. In the first row, we report the composite effect of changes in population brackets on output growth, as summarized by the coefficient on the JUMP index. The estimate is positive and highly significant across all models. The coefficient magnitude of implies that, conditional on a flexible functional form for population movements around the discontinuities, a jump (fall) to a higher (lower) population bracket increases municipality GDP growth by more than one percentage point. This effect corresponds to approximately 15% of the annual growth in local activity over the sample period. Furthermore, the estimates in the second and third rows suggest that the local effect of transiting to a higher (lower) population bracket is somewhat higher for smaller municipalities (which move across thresholds 1-3) as compared to larger cities (which move across thresholds 4-7). Figure 4 provides a visualization of the discontinuity in output growth at the cutoffs governing federal transfers to Brazilian municipalities. For the RD graph, we pool the seven thresholds together by normalizing population size according to the distance of each municipality in each year from the above or below threshold. Each point captures average output growth for municipalities above and below the cutoff grouped in intervals (bins) of 200 in- 18

19 habitants. The figure also graphs predicted values of output growth fitted separately for observations below and observations above the discontinuity. In line with the evidence in Table 8, there is a clear jump of output growth as a municipality crosses the cutoff value and receives a higher amount of federal transfers per capita (see Figure 2). 4.3 Instrumental variable estimates The first-stage regressions have established that population movements around the thresholds that shape discontinuously federal transfers affect government spending at the municipal level. The reduced-form estimates have demonstrated that local activity tends to increase significantly whenever municipalities move across population thresholds. In Table 9, we combine these two sets of results in an instrumental variables (fuzzy RD) set-up, which allows us to infer the causal effect of fiscal policy on output growth. The first row of Table 9 records the estimates of the local fiscal multiplier ( IV in equation (3)), that captures the impact of government spending on local activity averaged across all municipalities and thresholds. The coefficient is positive and highly significant across all perturbations, suggesting that (locally exogenous) increases in municipal spending do tend to increase output growth. The point estimates of the local fiscal multiplier range from 1.6 to 1.9. This finding appears to hold for smaller municipalities (i.e. thresholds 1-3) aswellas larger municipalities (i.e. thresholds 4-7). As we observe municipalities moving both to a higher and to a lower population bracket, we can decompose the JUMP index into two dummy variables that separately identify years when a municipality raises above a higher threshold and years when a municipality falls below a lower population threshold. This is useful not only because we can study the asymmetric effects of fiscal expansions and fiscal contractions, but also because we can estimate the IV model employing two instruments. Besides the efficiency gain, this allows us to perform the standard over-identification test of instrument validity to examine whether the two dummy variables for positive and negative jumps have an independent effect on output 19

20 growth that works over and above their effect on government spending (i.e. whether the exclusion restriction is satisfied; see Wooldridge (2002)). Table 10 reports the IV results for the two instruments case. The IV slopes are positive and significant at the 1% level across all permutations. The estimated coefficients appear quite stable and range between 1.7 and 2.1, values somewhat higher than the analogous estimates in Table 9. The statistics of the Hansen J-test for the over-identifying assumption imply that we cannot reject the null hypothesis of instrument validity at standard confidence levels. This further shows that the impact of movements to a higher (lower) population threshold on output growth revealed in the reduced-form estimates (in Table 8) works only via affecting municipal spending (as shown in Table 7). 5 Expansions versus contractions in public spending An advantage of our data is that we can look at episodes of fiscal expansions and fiscal contractions separately, exploiting the balanced number of observations for positive and negative jumps (Table 2). We thus distinguish between upward and downward jumps in government expenditure at the municipal level to assess potential asymmetry in the impact of fiscal expansions and fiscal contractions on output growth. 10 Table 11 reports the results. For brevity, the table displays only the estimates from specifications in which municipalities are pooled together across the first seven thresholds. Let us start with the first-stage estimates, reported in Panel A. Both indicators enter with positive and highly significant estimates, suggesting that positive and negative jumps are good predictors for an increase and a decrease in local government spending respectively. The estimated coefficient on the indicator that identifies upward movements is around 0.058, while the estimate on the dummy that identifies downward movements is larger, This 10 As discussed in the introduction, Brazil experienced an average growth rate of about 3.5% between 2000 and Excluding municipalities with more than 47, 545 inhabitants brings the national average to about 6% (see Table 4). While not all municipalities witnessed such a strong performance, the number of negative growth episodes appears so small as to prevent an accurate analysis of asymmetry in the fiscal multiplier across output recessions and expansions along the lines of Auerbach and Gorodnichenko (2012). 20

21 suggests that the boost in spending associated with a positive jump is somewhat smaller than the cut associated with a negative jump. Panel B reports the reduced-form estimates. An interesting finding emerges. Positive jumps to a higher municipality threshold lead to a small and statistically indistinguishable from zero increase in municipal output growth (of approximately 0.7%). In contrast, downward jumps to a lower population threshold that are associated with significant cuts in local spending (Panel A) are followed by sizable falls in municipal output growth. The estimate implies that such movements to a lower population threshold are associated with a fall in municipal output growth of approximately 2.2% 2.5%, a sizable drop(of approximately 40% of average output growth during this period). Figures 5a and 5b give a graphical illustration of the reduced form estimates. Local per capita GDP growth drops considerably more when a municipality moves to a lower population bracket (and thus receive a substantially smaller amount of federal transfers per capita) as compared to the case when a municipality population grows and the city moves to a higher population bracket (receiving thus more federal transfers in per capita terms). Panel C reports the IV estimates. In line with the reduced-form estimates, the local fiscal multipliers appear highly asymmetric. The IV point estimates that capture increases in public spending associated with jumps across population thresholds are between 0.9% and 1.3%. More importantly, in virtually all specifications, the estimates are statistically indistinguishable from zero. This implies that an exogenous increase in government spending (of 1% of municipal GDP) typically brings about an insignificant boost in local economic activity (of about the same magnitude). In contrast, the multipliers associated with a spending contraction triggered by a municipality falling into a lower population threshold are much larger around Furthermore, thecoefficients are more precisely estimated and always significant at standard confidence levels In the Appendix (Table A), we explore the possibility that the asymmetry in Table 11 is driven by time-to-build considerations. To this end, we add lagged values of the JUMP variable (at time t 1). The coefficients for the lagged values are both very small and statistically insignificant across all specifications. 21

22 To the best of our knowledge, the estimates in Table 11 are the first set of empirical findings documenting highly asymmetric effects on output growth of exogenous cuts and exogenous increases in government spending. The results suggest that fiscal contractions have sizable and significantly negative repercussions on economic activity while increases in public spending bear little significant stimulus for local output. 6 The heterogeneous effects of government spending In this section, we decompose the estimates of the previous section by sectors of the local economy and by spending categories. 6.1 Sectors of the economy In the previous sections, we have showed that the average local multiplier ranges from 1.6 to 2.1, stemmingfromacombinationofpointestimatesbetween0.9 and 1.3 for fiscal expansions and between 2.5 and 3 for fiscal restrictions. In this part of the paper, we ask whether these findings are dominated by a particular sector of the economy. To this end, Table 12 reports the average multipliers for agriculture, industry, and services. As the change in public expenditure is normalized by municipal GDP (at time t 1), the multiplier estimates should now be interpreted as the effect on a particular sector of a change in government spending that increases value added in that sector by 1% of municipal GDP. We have chosen to report the point estimates associated with this transformation because, arguably, makes it easier and more transparent the comparison across sectors. Alternatively, one can compute the product between the sector-specific multipliers and the average sector share reported in the last column to obtain an estimate of the change in sectoral value added associated with apublicspendingchangeequalto1% of the value added in that sector. Under the latter scenario, however, the size of the additional monetary transfers would vary across sectors There is some evidence that the asymmetry is slightly reversed over the following years but the statistical significance of such a difference seems very weak. Similar findings, not reported but available upon request, are obtained using lags up to two years. 22

23 according to their value added shares. The evidence in Table 12 suggests that public spending has a significant impact on agriculture and, to a lesser extent, services. In contrast, the multipliers are statistically indistinguishable from zero for industry/manufacturing. Furthermore, a change in transfers of a given amount would boost agricultural value added by more than a change of the same magnitude would do to the value added in the other sectors. 12 However, Table 13 reveals that, once more, looking at the average effect may be misleading. The estimated multipliers associated with a government spending reduction are, in fact, large and statistically significant for all sectors. This holds true for all model perturbations. Agriculture appears to bear the most detrimental consequences of fiscal contractions, followed by industry, and services. In line with the aggregate evidence (in Table 11), no sector seems to benefit significantly by fiscal expansions. Reduced-form estimates associated with the asymmetric sectoral multipliers are reported in Table C of the Appendix. The sectoral analysis reveals that the costs of public spending cuts are sizable for all sectors of the economy, while the benefits of increased public spending are limited. Moreover, the negative impacts associated with fiscal cuts on sectoral value added appears the largest for agriculture. 6.2 Spending categories Since the data report also spending by category, we also assess the extent to which the effects of public spending on local activity are driven by some specific categories of government purchases. As a preliminary step toward this, Table 14 reports the findings from (first stage) regressions in which the different sub-components of municipal spending are projected on the trichotomous JUMP variable. The top two rows look at expenditure by economic nature (current vs capital), while the remaining rows focus on the categories (e.g., administration, health, education, etc.). Among the nineteen categories, we find that changes in spending 12 Table B in the Appendix shows that these findings are robust to using positive and negative jumps as separate instruments. 23

24 on public administration, education, housing and urbanism, transportation, and health and sanitation are systematically related to the JUMP index that identifies years when a municipality moves to a higher or lower population bracket. This implies that in response to an increased (or lower) amount of federal transfers local administrations respond by increasing (or lowering) expenditure in this categories (which also tend to be the main types of municipal spending in the last column). Table 15 displays the estimated average multipliers across both types of government expenditure (current and capital) as well as across the five categories where there is a strong (first stage) link between the instrument and public spending. In all specifications, the relevant explanatory variable becomes the change on a specific type/category of expenditure over GDP in the previous period, thereby implying that the multiplier estimates now reflect the percentage increase in local economic activity driven by an increase in public spending on a particular type/category that is as large as 1% of municipal GDP. InanalogytoTable 12, bycomputingtheproductbetweenthereportedestimatesandtheexpendituresharesin the last column, one can obtain the multiplier relative to a spending increase that is 1% of that particular type/category of spending. Two main findings emerge from Table 15. First, for the same amount of resources (i.e. 1% of municipal GDP), government capital expenditure tends to affect local activity by more than government current expenditure. Second, among the spending categories, the largest and most significant point estimates are associated with transportation and housing, followed -in order- by health, education and public administration. This ranking is not overturned in Table 16 when we look at the asymmetric effects of fiscal policy by spending type/category. In analogy to the result on total municipal government spending, the multipliers for fiscal contractions are highly significant and much larger than the average effect. In contrast, the estimated effects associated with fiscal expansions are small and statistically indistinguishable from zero in all specifications Results for the average multipliers and for the asymmetric first stage regressions based on multiple 24

25 7 Conclusions A fuzzy regressiondiscontinuitydesignthatexploitsthehighlynon-linearallocationmechanism of federal transfers to local municipalities in Brazil allows us to identify the causal effects of fiscal policy on economic activity. The analysis uncovers three main regularities. First, a change in public spending of 1% of municipal GDP is associated with an average change in local output between 1.6% and 2%. The estimates of the average local multiplier appear robust across a range of empirical specifications that control for time-invariant municipal characteristics as well as aggregate time-varying factors reflecting federal fiscal policy, monetary policy and national business cycles. Second, the average effect masks considerable asymmetry in the impact of fiscal policy. An exogenous increase in government spending of 1% of municipal GDP tends to generate an insignificant boost in local activity of 0.9% to 1.3%. In contrast, the multiplier associated with local public spending cuts appears larger and highly significant, at values between 2.5 and 3. Third, using disaggregated local government data, we explore heterogeneity across various spending categories. We find that contractions in public spending on transportation, housing and, to a lesser extent, health and education appear the most detrimental for the local economy. In a figure of speech often attributed to Keynes, monetary policy is like pushing on a string : it can more likely stop an expansion than it can end a contraction. Our empirical results suggest that this metaphor applies to government policy in Brazilian municipalities, where locally exogenous changes in public spending allows us to identify separately the output effects of fiscal expansions and the output effects of fiscal contractions. Future research is need to explore the mechanism(s) behind the asymmetric impact of government spending as well as to investigate whether a similar asymmetric pattern emerges in other countries. instruments are recorded in Tables D and E of the Appendix, respectively. 25

26 References Acconcia, A., G. Corsetti and S. Simonelli, 2013, Mafia and Public Spending: Evidence on the Fiscal Multiplier from a Quasi-experiment, mimeo, University of Naples Federico II and University of Cambridge. Angrist, J. and V. Lavy, 1999, Using Maimonides Rule to Estimate the Effect of Class Size on Scholastic Achievement, Quarterly Journal of Economics 114(2), pp Auerbach, Alan and William Gale, 2010, Activist Fiscal Policy to Stabilize Economic Activity. In "Financial Stability and Macroeconomic Policy", pp Kansas City: Federal Reserve Bank of Kansas City. Auerbach, A. and Y. Gorodnichenko, 2012, Measuring the Output Responses to Fiscal Policy, American Economic Journal: Economic Policy, 4(1), pp Blanchard, O. and R. Perotti, 2002, An Empirical Characterization of the Dynamic Effects of Changes in Government Spending and Taxes on Output, Quarterly Journal of Economics 117(4), pp Barro, R. and C. Redlick, 2011, Macroeconomic Effects from Government Purchases and Taxes, Quarterly Journal of Economics 126(1), pp Barro, R., 1981, Output effects of government purchases, Journal of Political Economy Vol. 89, No. 6 (Dec., 1981), pp Blanchard, O. and D. Leigh, 2013, Growth Forecast Errors and Fiscal Multipliers, IMF working paper 13/1. Brollo, F. and T. Nannicini, 2012, Tying Your Enemy s Hands in Close Races: The Politics of Federal Transfers in Brazil, American Political Science Review 106, pp

27 Brollo, F., Nannicini, T., Perotti, R. and G. Tabellini, 2013, The political resource curse, American Economic Review, 103(5): Brückner, M. and A. Tuladhar, 2013, Local Government Spending Multipliers and Financial Distress: Evidence from Japanese Prefectures, Economic Journal, forthcoming. Caselli, F. and G. Michaels, 2013, Do oil windfalls improve living standards? Evidence from Brazil, American Economic Journal: Applied Economics 5, pp Caldara, D. and C. Kamps, 2012, The analytics of SVAR: a unified framework to measure fiscal multipliers, Finance and Economics Discussion Series Chodorow-Reich, G., Feiveson, L. and Z. Liscow, and William Woolston, 2012, Does State Fiscal Relief During Recessions Increase Employment? Evidence from the American Recovery and Reinvestment Act, American Economic Journal: Economic Policy 4, pp Corbi, R., 2013, Fiscal Transfers and the Quantity and Quality of Education, mimeo, London Business School. Ferraz, C. and F. Finan, 2011a, Motivating politicians: the impacts of monetary incentives on quality and performance, mimeo, PUC-Rio and University of California, Berkeley. Ferraz, C. and F. Finan, 2011b, Electoral accountability and corruption in local governments: Evidence from audit reports, American Economic Review 101, pp Ferraz, C. and F. Finan, 2008, Exposing corrupt politicians: The effect of Brazil s publicly released audits on electoral outcomes, Quarterly Journal of Economics 123, pp Feyrer, J. and B. Sacerdote, 2012, Did the stimulus stimulate? Effects of the American Recovery and Reinvestment Act, mimeo, Dartmouth University. 27

28 Grembi, V., Nannicini, T. and U. Troiano, 2013, Policy Responses to Fiscal Restraints: A Difference-in-Discontinuities Design, mimeo, Harvard University and Bocconi University. Hahn, J., Todd, P. and W. van der Klaauw, 2002, Identification and Estimation of Treatment Effects with a Regression-Discontinuity Design, Econometrica 69, pp Hall, R., 2009, By How Much Does GDP Rise If the Government Buys More Output?, Brookings Papers on Economic Activity 2, pp van der Klaauw, W., 2002, Estimating the Effect of Financial Aid Offers on College Enrollment: A Regression-Discontinuity Approach, International Economic Review 43, pp Ilzetzki, E., E. Mendoza and C. Vegh, 2013, How Big (Small?) are Fiscal Multipliers?, Journal of Monetary Economics 60, pp Imbens, G., and T. Lemieux, 2008, Regression Discontinuity Designs: A Guide to Practice, Journal of Econometrics 142, pp Kraay, A., 2012, How large is the Government Spending Multiplier? The Quarterly Journal of Economics first published online March 28, 2012 doi: /qje/qjs008 Litschig, S., 2012, Are rules-based government programs shielded from special-interest politics? Evidence from revenue-sharing transfers in Brazil, Journal of Public Economics, forthcoming. Litschig, S. and K. Morrison, 2013, The impact of intergovernmental transfers on education outcomes and poverty reduction, American Economic Journal: Applied Economics, forthcoming. McCrary, J., 2008, Manipulation of the Running Variable in the Regression Discontinuity Design: A Density Test, Journal of Econometrics 142, pp

29 Mountford, A. and H. Uhlig, 2009, What are the effects of fiscal policy shocks?, Journal of Applied Econometrics 24, pp Nakamura, E. and J. Steinsson, 2013, Fiscal stimulus in a monetary Union: Evidence from U.S. Regions, American Economic Review, forthcoming. Parker, Jonathan, 2011, On Measuring the Effects of Fiscal Policy in Recessions, Journal of Economic Literature 49, pp Perotti, R., 2007, In search of the transmission mechanism of fiscal policy. In D. Acemoglu, K. Rogoff, and M. Woodford (eds.), NBER Macroeconomics Annual 2007, pp Ramey, V., 2011a, Can Government Purchases Stimulate the Economy?, Journal of Economic Literature 49, pp Ramey, V., 2011b, Identifying government spending shocks: It s all in the timing, Quarterly Journal of Economics 126, pp Romer, C., and D. Romer, 2010, The Macroeconomic Effects of Tax Changes: Estimates Based on a New Measure of Fiscal Shocks, American Economic Review 100, pp Serrato, J., and P. Wingender, 2011, Estimating local fiscal multipliers, mimeo, University of California, Berkeley. Shoag, D., 2010, The impact of government spending shocks: Evidence on the Multiplier from State Pension Plan Returns, mimeo, Harvard University. Wooldridge, J., 2002, Econometric Analysis of Cross Section and Panel Data, MIT Press. 29

30 Table 1: FPM transfers coefficients Population brackets. FPM coefficients Below 10, ,189 13, ,585 16, ,981 23, ,773 30, ,565 37, ,357 44, ,149 50,940 2 Above 50, Note: The FPM (Fundo de Participacao dos Municipios) is a federal fund designed to transfer funds from the federal government to municipalities, as described in section 2. A FPM coefficient is the weight assigned to a municipality in a particular state in the FPM allocation mechanism. In a given state, the higher the coefficient the larger the fiscal transfer a municipality receives. 30

31 Table 2: Distribution of jumps across time, population and unique municipality PANEL A: Number of jumps across years jump total ,779 2,274 1,705 2,344 2,316 2,066 2,247 2,324 1,699 2,146 2,373 23, , Total 1,896 2,394 2,441 2,464 2,433 2,307 2,371 2,424 2,261 2,354 2,455 25,800 PANEL B: Distribution of jumps per population bracket total Below 10, , ,336 10,189 13, , ,693 13,585 16, , ,251 16,981 23, , ,647 23,773 30, , ,232 30,565 37, , ,958 37,357 44, , ,299 Above 44, Total ,273 1, ,800 PANEL C: Distribution of jumps for each unique municipality jump 0 positive Total negative 1,019 1,022 2,041 Total 1,111 1,680 2,791 Note: Each observation corresponds to a municipality-yeay. A JUMP is defined by a change in the population bracket from one year to the next. At a given year, municipality that remains in the same bracket as in the previous year has JUMP = 0. An movement upwards (downwards) is represented by a positive (negative). Jumps higher than 1 in absolute terms imply a JUMP of two or more brackets. PANEL A, B and C show the distribution of JUMP across time, population brackets and for each unique municipality. 31

32 Table 3: Distribution of observations around threshold Population bin (8,491-11,887) (11,887-15,283) (13,585-20,377) (20,377-27,169) (27,169-33,961) (33,961-40,753) (40,753-47,545) threshold 1 threshold 2 threshold 3 threshold 4 threshold 5 threshold 6 threshold 7 year below above below above below above below above below above below above below above Total 3,134 3,184 2,483 2,482 1,795 3,423 2,242 1,862 1,386 1, Note: Each observation corresponds to a municipality-year. This table shows the number of observations around each of the seven thresholds (10,189; 13,585; 16,981; 23,773; 30,564; 37,356; and 44,148). Each observation is assigned to the nearest threshold (above or below), effectively creating 7 population-intervals centered on each thresholds limited by the midpoints between adjacent thresholds. The bins are 8,491-11,887; 11,887-15,283; 13,585-20,377; 20,377-27,169; 27,169-33,961; 33,961-40,753, 40,753-47,

33 Table 4: Municipal GDP, Population, Expenditure and Fiscal Transfers (FPM) Population Share in total per capita Std. Dev. Transfers / Expenditure / Number Population growth population GDP growth growth GDP GDP observations Below 10, % 1.30% 6.36% % 13.4% 3,134 10,189 13, % 3.29% 6.59% % 14.4% 5,667 13,585 16, % 3.28% 6.54% % 14.1% 4,277 16,981 23, % 5.65% 6.47% % 13.1% 5,665 23,773 30, % 4.25% 6.12% % 11.9% 3,248 30,565 37, % 3.78% 6.55% % 11.1% 2,007 37,357 44, % 2.84% 5.78% % 10.1% 1,333 44,149 50, % 1.30% 5.97% % 10.2% 469 Total 1.19% 25.7% 6.30% % 13.1% 25,800 GDP Agriculture Industry Services Agriculture Industry Services Population (thousands) (% of GDP) (% of GDP) (% of GDP) growth growth growth Below 10,188 48, % 15.0% 54.5% 8.5% 11.8% 5.7% 10,189 13,584 53, % 14.4% 56.7% 8.8% 10.0% 6.1% 13,585 16,980 77, % 15.3% 57.3% 8.9% 12.8% 5.9% 16,981 23, , % 16.3% 57.2% 9.0% 9.4% 6.2% 23,773 30, , % 17.7% 57.5% 8.5% 9.5% 5.7% 30,565 37, , % 19.0% 57.9% 8.7% 10.3% 5.9% 37,357 44, , % 19.5% 58.5% 7.2% 9.7% 5.4% 44,149 50, , % 20.4% 58.2% 8.1% 10.0% 5.6% Total 1,113, % 17.2% 57.2% 8.4% 10.4% 5.8% Note: All value are in Brazilian reais (BRL) at 2000 prices. Sample: (yearly data with 25,800 observations). Fiscal Transfers are reported values transferred to municipalities from the Federal Government through the Fundo de Participacao dos Municipios (FPM). Each observation corresponds to a municipality-year. Agriculture, Industry and Services correspond to the value-added component of each sector. 33

34 Table 5: Municipal Expenditure according to economic nature and categories Nature of Expenditure Categories of Expenditure Population Current Capital Administration Assistance Communication Science & Tech Education Below 10, % 14.8% 21.9% 7.4% 0.1% 0.0% 40.7% 10,189 13, % 14.7% 22.3% 7.3% 0.1% 0.0% 42.2% 13,585 16, % 14.6% 21.0% 7.2% 0.1% 0.0% 43.0% 16,981 23, % 14.5% 21.6% 7.5% 0.1% 0.0% 43.2% 23,773 30, % 14.5% 21.6% 7.8% 0.1% 0.0% 44.5% 30,565 37, % 14.6% 21.7% 7.6% 0.1% 0.0% 44.1% 37,357 44, % 15.2% 21.7% 7.8% 0.1% 0.0% 43.4% 44,149 50, % 14.7% 20.8% 7.6% 0.1% 0.0% 45.6% Total 1.19% 25.7% 6.30% 5,273 3,647 11,084 25,801 Categories of Expenditure Energy & Housing & Industry, Retail Judiciary Legislative Sports & Citizenship & Population Resources Urbanism & Services Leisure Justice Below 10, % 11.3% 0.6% 0.3% 4.7% 1.3% 0.1% 10,189 13, % 11.7% 0.5% 0.2% 4.6% 1.3% 0.1% 13,585 16, % 12.4% 0.6% 0.2% 4.4% 1.3% 0.1% 16,981 23, % 12.4% 0.7% 0.3% 4.5% 1.2% 0.1% 23,773 30, % 12.9% 0.8% 0.3% 4.3% 1.2% 0.1% 30,565 37, % 13.0% 0.8% 0.2% 4.2% 1.2% 0.2% 37,357 44, % 14.4% 1.0% 0.3% 4.1% 1.3% 0.3% 44,149 50, % 13.9% 0.8% 0.4% 4.2% 1.2% 0.1% Categories of Expenditure Financing & National Foreign Health & Labor Transportation Agriculture Population Social Security Defense Relations Sanitation Below 10, % 0.2% 0.004% 29.0% 0.2% 7.0% 2.9% 10,189 13, % 0.2% 0.003% 28.4% 0.2% 5.9% 2.7% 13,585 16, % 0.2% 0.001% 29.1% 0.3% 5.6% 2.6% 16,981 23, % 0.2% 0.005% 28.9% 0.2% 4.9% 2.5% 23,773 30, % 0.3% 0.005% 29.3% 0.1% 4.2% 2.5% 30,565 37, % 0.4% 0.007% 29.6% 0.2% 3.9% 2.1% 37,357 44, % 0.4% 0.001% 31.3% 0.3% 3.8% 2.2% 44,149 50, % 0.5% 0.000% 31.1% 0.1% 3.3% 2.0% Note: Sample: (yearly data with 25,800 observations). Fiscal Transfers are reported values transferred to municipalities from the Federal Government through the Fundo de Participacao dos Municipios (FPM). Total Expenditure can be classified according to functions (categories) or to economic nature (current or capital). Within the economic nature classification, Current expenditure comprises expenses such as personnel, consumption, external services and estate maintenance, as well as social and economic transfers to citizens or public or private organizations. Capital expenditure comprises investment expenses (building/purchases of real estate, capital goods) and financial investments. The classification according to functions divides total expenditure into the categories, as seen on the table. 34

35 Table 6: First Stage estimates: FPM transfers (I) (II) (III) (IV) (V) (VI) (VII) Cross Cross Cross Threshold Parsimonious Cross Cross threshold threshold threshold specific (no threshold threshold 3rd order pol 4th order pol 5th order pol 3rd order pol polynomial) 3rd order pol 3rd order pol Dependent variable: DTransfers per capita/gdp per capita t-1 state-year FE with state time-trend Thresholds *** *** 0.009*** *** *** 0.009*** *** (0.0003) (0.0003) (0.0003) (0.0003) (0.0003) (0.0003) (0.0003) Thresholds *** *** *** *** *** *** *** (0.0004) (0.0004) (0.0004) (0.0004) (0.0004) (0.0004) (0.0004) Thresholds *** 0.005*** 0.005*** *** *** *** 0.005*** (0.0002) (0.0002) (0.0002) (0.0002) (0.0002) (0.0002) (0.0002) Threshold *** *** 0.014*** *** *** *** *** (0.0008) (0.0008) (0.0008) (0.0008) (0.0008) (0.0008) (0.0008) Threshold *** *** *** *** *** *** *** (0.0006) (0.0006) (0.0006) (0.0006) (0.0006) (0.0006) (0.0006) Threshold *** *** *** *** *** *** *** (0.0005) (0.0005) (0.0005) (0.0005) (0.0005) (0.0005) (0.0005) Threshold *** *** *** 0.007*** 0.007*** *** *** (0.0005) (0.0005) (0.0005) (0.0005) (0.0005) (0.0004) (0.0005) Threshold *** *** *** *** *** *** *** (0.0004) (0.0004) (0.0004) (0.0004) (0.0004) (0.0004) (0.0004) Threshold *** *** 0.003*** *** *** 0.003*** *** (0.0004) (0.0005) (0.0005) (0.0005) (0.0004) (0.0004) (0.0004) Threshold *** *** *** *** *** 0.002*** *** (0.0005) (0.0005) (0.0005) (0.0005) (0.0005) (0.0005) (0.0005) Municipal FE X X X X X X X Year dummies X X X X X X X Note: First-stage estimates of changes in population brackets from one year to the next (JUMP) on FPM current fiscal transfers. For the definition of JUMP, refer to table 3.The coefficients in the row Threshold 1-7 are obtained by estimating the main regression in the entire sample; the heterogeneity coefficients in the other rows are obtained by interacting JUMP with population-interval dummies (from the midpoint below to the midpoint above each threshold, as explained on table 3) for Thresholds 1-3, Thresholds 4-7, and each individual threshold, respectively. Each regression discontinuity polynomial as specified. Sample: (yearly data with 25,800 observations). Robust standard errors clustered at the municipality level in parentheses. Significance at the 10% level is represented by *, at the 5% level by **, and at the 1% level by ***. 35

36 Table 7: First Stage estimates: municipal expenditure (I) (II) (III) (IV) (V) (VI) (VII) Cross Cross Cross Threshold Parsimonious Cross Cross threshold threshold threshold specific (no threshold threshold 3rd order pol 4th order pol 5th order pol 3rd order pol polynomial) 3rd order pol 3rd order pol Dependent variable: DExpenditure per capita/gdp per capita t-1 state-year FE with state time-trend Thresholds *** *** *** *** *** *** *** (0.0007) (0.0007) (0.0007) (0.0007) (0.0007) (0.0007) (0.0007) Thresholds *** *** *** *** 0.008*** *** *** (0.0009) (0.001) (0.001) (0.001) (0.0009) (0.0009) (0.0009) Thresholds *** *** *** *** *** *** *** (0.001) (0.001) (0.001) (0.001) (0.0009) (0.0009) (0.0009) Threshold *** *** *** *** *** *** *** (0.0016) (0.0015) (0.0015) (0.0016) (0.0015) (0.0014) (0.0016) Threshold *** *** *** *** *** 0.007*** *** (0.0016) (0.0016) (0.0016) (0.0017) (0.0016) (0.0015) (0.0016) Threshold *** *** *** 0.007*** *** *** *** (0.0016) (0.0016) (0.0016) (0.0016) (0.0016) (0.0015) (0.0016) Threshold *** *** *** *** *** 0.006*** *** (0.0015) (0.0015) (0.0015) (0.0015) (0.0015) (0.0014) (0.0015) Threshold ** ** ** ** ** * ** (0.0018) (0.0018) (0.0018) (0.0018) (0.0017) (0.0017) (0.0018) Threshold *** *** *** *** *** *** *** (0.0017) (0.0018) (0.0018) (0.0019) (0.0017) (0.0015) (0.0017) Threshold * ** ** * * (0.002) (0.002) (0.002) (0.0022) (0.002) (0.002) (0.002) Municipal FE X X X X X X X Year dummies X X X X X X X Note: First-stage estimates of changes in population brackets from one year to the next (JUMP) on current total expenditure. For the definition of JUMP, refer to table 3.The coefficients in the row Threshold 1-7 are obtained by estimating the main regression in the entire sample; the heterogeneity coefficients in the other rows are obtained by interacting JUMP with population-interval dummies (from the midpoint below to the midpoint above each threshold, as explained on table 3) for Thresholds 1-3, Thresholds 4-7, and each individual threshold, respectively. Each regression discontinuity polynomial as specified. Sample: (yearly data with 25,800 observations). Robust standard errors clustered at the municipality level in parentheses. Significance at the 10% level is represented by *, at the 5% level by **, and at the 1% level by ***. 36

37 Table 8: Reduced-form estimates (I) (II) (III) (IV) (V) (VI) (VII) Cross Cross Cross Threshold Parsimonious Cross Cross threshold threshold threshold specific (no threshold threshold 3rd order pol 4th order pol 5th order pol 3rd order pol polynomial) 3rd order pol 3rd order pol state-year FE with state per capita GDP growth time-trend Thresholds *** *** *** *** *** *** *** (0.0033) (0.0033) (0.0034) (0.0033) (0.0031) (0.003) (0.0033) Thresholds *** *** *** *** *** *** 0.013*** (0.0043) (0.0043) (0.0044) (0.0045) (0.0042) (0.004) (0.0043) Thresholds ** ** ** * ** (0.0046) (0.0047) (0.0047) (0.005) (0.0044) (0.0042) (0.0046) Threshold (0.0061) (0.0061) (0.0062) (0.0062) (0.0061) (0.0056) (0.0062) Threshold ** *** ** ** ** *** ** (0.0083) (0.0083) (0.0083) (0.0087) (0.0082) (0.0078) (0.0083) Threshold (0.007) (0.007) (0.0071) (0.0073) (0.0069) (0.0063) (0.0069) Threshold (0.0078) (0.0079) (0.0079) (0.0084) (0.0077) (0.0068) (0.0078) Threshold (0.0081) (0.0081) (0.0082) (0.0086) (0.008) (0.0075) (0.008) Threshold (0.0094) (0.0095) (0.0095) (0.0112) (0.0093) (0.0097) (0.0093) Threshold ** ** *** ** ** (0.0094) (0.0094) (0.0094) (0.0108) (0.0093) (0.0092) (0.0093) Municipal FE X X X X X X X Year dummies X X X X X X X Note: Reduced-form estimates of changes in population brackets from one year to the next (JUMP) on FPM current fiscal transfers and current total expenditure. For a better description of definition of JUMPs, refer to table 3 The coefficients in the row Threshold 1-7 are obtained by estimating the main regression in the entire sample; the heterogeneity coefficients in the other rows are obtained by interacting JUMP with population-interval dummies (from the midpoint below to the midpoint above each threshold, as explained on table 3) for Thresholds 1-3, Thresholds 4-7, and each individual threshold, respectively. Each regression discontinuity polynomial as specified. Sample: (yearly data with 25,800 observations). Robust standard errors clustered at the municipality level in parentheses. Significance at the 10% level is represented by *, at the 5% level by **, and at the 1% level by ***. 37

38 Table 9: Fiscal Multiplier - Instrumental variables estimates (I) (II) (III) (IV) (V) (VI) (VII) Cross Cross Cross Threshold Parsimonious Cross Cross threshold threshold threshold specific (no threshold threshold 3rd order pol 4th order pol 5th order pol 3rd order pol polynomial) 3rd order pol 3rd order pol state-year FE with state IV Estimates time-trend Thresholds *** 1.932*** 1.928*** 1.564*** 1.576*** 1.601*** 1.887*** (0.512) (0.525) (0.542) (0.509) (0.469) (0.487) (0.523) Thresholds *** 1.746*** 1.747*** 1.585*** 1.413*** 1.768*** 1.697*** (0.572) (0.579) (0.595) (0.582) (0.54) (0.538) (0.584) Thresholds ** 2.396** 2.378** ** ** (0.998) (1.021) (1.046) (1.046) (0.936) (0.989) (1.018) Municipal FE X X X X X X X Year dummies X X X X X X X Note: Multipliers estimates using a two-stage least squares estimator using JUMPs in population brackets as instruments. For the definition of JUMP, refer to table 3. The coefficients in the row Threshold 1-7 are obtained by estimating the regression in the entire sample; the heterogeneity coefficients in the other rows are obtained by interacting the relecant specification with population-interval dummies (from the midpoint below to the midpoint above FPM thresholds) for Thresholds 1-3, Thresholds 4-7, and each individual threshold, respectively. Each regression discontinuity polynomial as specified. Sample: (yearly data with 25,800 observations). Robust standard errors clustered at the municipality level in parentheses. Significance at the 10% level is represented by *, at the 5% level by **, and at the 1% level by ***. 38

39 Table 10: Fiscal Multiplier estimates with multiple instruments (I) (II) (III) (IV) (V) (VI) (VII) Cross Cross Cross Threshold Parsimonious Cross Cross threshold threshold threshold specific (no threshold threshold 3rd order pol 4th order pol 5th order pol 3rd order pol polynomial) 3rd order pol 3rd order pol state-year FE with state IV Estimates time-trend Thresholds *** 2.064*** 2.076*** 1.684*** 1.709*** 1.691*** 2.015*** (0.499) (0.512) (0.525) (0.495) (0.456) (0.481) (0.513) Hansen J (p-value) (0.132) (0.144) (0.167) (0.131) (0.101) (0.108) (0.103) Thresholds *** 1.772*** 1.783*** 1.599*** 1.471*** 1.777*** 1.728*** (0.56) (0.568) (0.58) (0.567) (0.528) (0.536) (0.574) Hansen J (p-value) (0.384) (0.395) (0.421) (0.669) (0.309) (0.299) (0.321) Thresholds *** 2.737*** 2.733*** 1.998** 2.327*** *** (0.987) (1.009) (1.032) (1.024) (0.91) (0.959) (1.014) Hansen J (p-value) (0.106) (0.117) (0.134) (0.027) (0.112) (0.143) (0.095) Municipal FE X X X X X X X Year dummies X X X X X X X Note: Multipliers estimates using a two-stage least squares estimator using positive and negative JUMPs in population brackets as two different instruments in the same regressions. Over identification test p-values reported. For the definition of JUMP, refer to table 3. Each regression discontinuity polynomial as specified. Sample: (yearly data with 25,800 observations). Robust standard errors clustered at the municipality level in parentheses. Significance at the 10% level is represented by *, at the 5% level by **, and at the 1% level by ***. 39

40 Table 11: Asymmetry in multipliers: positive vs negative shocks to expenditure (I) (II) (III) (IV) (V) (VI) (VII) Cross Cross Cross Threshold Parsimonious Cross Cross threshold threshold threshold specific (no threshold threshold 3rd order pol 4th order pol 5th order pol 3rd order pol polynomial) 3rd order pol 3rd order pol state-year FE with state First-Stage time-trend jump> *** *** *** *** *** *** *** (0.0008) (0.0008) (0.0009) (0.0008) (0.0008) (0.0008) (0.0008) jump< *** *** *** *** *** *** *** (0.0014) (0.0014) (0.0014) (0.0014) (0.0014) (0.0013) (0.0014) coeff. equality test (p-value) Reduced-Form jump> (0.0043) (0.0044) (0.0044) (0.0044) (0.004) (0.004) (0.0043) jump< *** 0.025*** *** *** *** *** *** (0.0057) (0.0057) (0.0057) (0.0057) (0.0056) (0.0055) (0.0057) coeff. equality test (p-value) IV Estimates - Asymmetric Multipliers jump> * (0.75) (0.763) (0.808) (0.753) (0.679) (0.685) (0.746) jump<0 2.82*** 2.873*** 2.817*** 2.506*** 2.595*** 2.689*** 2.968*** (0.758) (0.769) (0.758) (0.738) (0.717) (0.807) (0.803) Municipal FE X X X X X X X Year dummies X X X X X X X Note: Reduced-form, First stage and multiplier IV estimates using positive and negative JUMPs as two separate instruments. For the definition of JUMP, refer to table 3. Coefficient equality test p-values are reported for the reduced-form and first-stage estimates. Each regression discontinuity polynomial as specified. Sample: (yearly data with 25,800 observations). Robust standard errors clustered at the municipality level in parentheses. Significance at the 10% level is represented by *, at the 5% level by **, and at the 1% level by ***. 40

41 Table 12: Fiscal Multipliers and Reduced-form estimates by component of GDP (I) (II) (III) (IV) (V) (VI) (VII) Cross Cross Cross Threshold Parsimonious Cross Cross share in threshold threshold threshold specific (no threshold threshold GDP 3rd order pol 4th order pol 5th order pol 3rd order pol polynomial) 3rd order pol 3rd order pol (std. dev) state-year FE with state Reduced-form time-trend Agriculture *** *** 0.028*** *** *** *** ** 20.9% (0.0079) (0.008) (0.0081) (0.0079) (0.0074) (0.0064) (0.0079) (0.14) Industry % (0.0053) (0.0053) (0.0054) (0.0052) (0.005) (0.005) (0.0053) (0.14) Services *** *** *** 0.004** *** ** *** 57.0% (0.0021) (0.0021) (0.0021) (0.0021) (0.002) (0.002) (0.002) (0.14) IV Estimates - Multipliers Agriculture 3.877*** 4.231*** 4.279*** 3.825*** 3.258*** 2.508*** 3.873** 20.9% (1.253) (1.287) (1.331) (1.237) (1.149) (1.042) (1.267) (0.14) Industry % (0.797) (0.802) (0.823) (0.774) (0.739) (0.778) (0.804) (0.14) Services 1.078*** 1.041*** 0.998*** 0.601*** 1.014*** 0.893*** 1.103*** 57.0% (0.314) (0.323) (0.327) (0.308) (0.297) (0.309) (0.319) (0.14) Municipal FE X X X X X X X Year dummies X X X X X X X Note: Reduced-form and Multipliers estimates using a two-stage least squares estimator using JUMPs in population brackets as instruments. For the definition of JUMP, refer to table 3. Each regression discontinuity polynomial as specified. Share in GDP of each component and their standard deviation are also reported. Sample: (yearly data with 25,800 observations). Robust standard errors clustered at the municipality level in parentheses. Significance at the 10% level is represented by *, at the 5% level by **, and at the 1% level by ***. 41

42 Table 13: Asymmetry in fiscal multipliers per GDP component: positive vs negative jumps (I) (II) (III) (IV) (V) (VI) (VII) Cross Cross Cross Threshold Parsimonious Cross Cross threshold threshold threshold specific (no threshold threshold 3rd order pol 4th order pol 5th order pol 3rd order pol polynomial) 3rd order pol 3rd order pol state-year FE with state Agriculture time-trend jump> (1.726) (1.774) (1.876) (1.718) (1.541) (1.411) (1.709) jump< *** 6.59*** 6.578*** 6.513*** 6.104*** 5.135*** 6.614*** (1.952) (1.985) (1.965) (1.964) (1.884) (1.79) (2.048) Industry jump> (1.125) (1.142) (1.185) (1.12) (1.042) (1.034) (1.106) jump< ** 2.762** 2.73** 1.947* 2.409** 1.492** 2.734** (1.231) (1.239) (1.225) (1.188) (1.186) (1.286) (1.298) Services jump> (0.468) (0.48) (0.5) (0.472) (0.43) (0.431) (0.463) jump< *** 1.526*** 1.498*** 1.087** 1.545*** 1.287** 1.562*** (0.47) (0.472) (0.467) (0.446) (0.458) (0.505) (0.492) Municipal FE X X X X X X X Year dummies X X X X X X X Note: Multipliers estimates using positive and negative JUMPs as two separate instruments. For the definition of JUMP, refer to table 3. Each regression discontinuity polynomial as specified. Sample: (yearly data with 25,800 observations). Robust standard errors clustered at the municipality level in parentheses. Significance at the 10% level is represented by *, at the 5% level by **, and at the 1% level by ***. 42

43 Table 14: First-Stage estimates: public expenditure by economic nature and category (I) (II) (III) (IV) (V) (VI) (VII) Cross Cross Cross Threshold Parsimonious Cross Cross share in threshold threshold threshold specific (no threshold threshold total 3rd order pol 4th order pol 5th order pol 3rd order pol polynomial) 3rd order pol 3rd order pol expend state-year FE with state iture Nature of Expenditure time-trend Current Expenditure 0.441*** 0.437*** 0.43*** 0.454*** 0.44*** 0.438*** 0.429*** 85.3% (0.057) (0.057) (0.058) (0.058) (0.054) (0.052) (0.056) (0.09) Capital Expenditure 0.232*** 0.234*** 0.23*** 0.225*** 0.234*** 0.204*** 0.231*** 14.7% (0.043) (0.043) (0.044) (0.043) (0.041) (0.041) (0.043) (0.09) Category of Expenditure Administration 0.171*** 0.179*** 0.177*** 0.189*** 0.164*** 0.179*** 0.167*** 21.7% (0.038) (0.039) (0.04) (0.037) (0.036) (0.038) (0.039) (0.11) Assistance 0.024* 0.027* 0.025* 0.024* 0.028** % (0.014) (0.014) (0.015) (0.014) (0.013) (0.014) (0.014) (0.05) Communications % (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0) Science & Technology % (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0) Education 0.269*** 0.271*** 0.273*** 0.273*** 0.258*** 0.23*** 0.263*** 42.9% (0.033) (0.033) (0.034) (0.034) (0.031) (0.033) (0.033) (0.12) Energy & Resources 0.012*** 0.013*** 0.012*** 0.012*** 0.012*** 0.013*** 0.013*** 0.6% (0.005) (0.005) (0.005) (0.005) (0.004) (0.005) (0.005) (0.01) Housing & Urbanism 0.163*** 0.166*** 0.167*** 0.164*** 0.156*** 0.147*** 0.163*** 12.4% (0.08) Industry, retail, services % (0.006) (0.005) (0.006) (0.006) (0.006) (0.006) (0.006) (0.02) Judiciary % (0.004) (0.004) (0.005) (0.005) (0.004) (0.004) (0.004) (0.01) Legislative *** *** -0.02*** ** ** *** *** 4.4% (0.006) (0.006) (0.006) (0.006) (0.006) (0.006) (0.006) (0.02) Sports & Leisure 0.015** 0.016** 0.016** 0.017** 0.019** 0.016** 0.015** 1.3% (0.02) Citizenship & Justice * ** * % (0.005) (0.005) (0.005) (0.005) (0.005) (0.005) (0.005) (0.01) Financing & Social Sec * % (0.042) (0.043) (0.042) (0.043) (0.038) (0.043) (0.042) (0.04) 43

44 Table 14: First-Stage estimates: public expenditure by economic nature and category (cont ed) (I) (II) (III) (IV) (V) (VI) (VII) Cross Cross Cross Threshold Parsimonious Cross Cross share in threshold threshold threshold specific (no threshold threshold total 3rd order pol 4th order pol 5th order pol 3rd order pol polynomial) 3rd order pol 3rd order pol expend state-year FE with state iture Category of Expenditure time-trend National Defense % (0.004) (0.004) (0.004) (0.004) (0.004) (0.004) (0.004) (0.01) Foreign Relations % (0.0002) (0.0001) (0.0002) (0.0002) (0.0001) (0.0002) (0.0002) (0) Health and Sanitation 0.175*** 0.175*** 0.177*** 0.166*** 0.185*** 0.159*** 0.173*** 29.1% (0.029) (0.029) (0.03) (0.029) (0.027) (0.028) (0.029) (0.1) Labour % (0.009) (0.009) (0.009) (0.01) (0.008) (0.009) (0.009) (0.02) Transportation 0.086*** 0.089*** 0.083*** 0.081*** 0.087*** 0.066*** 0.083*** 5.3% (0.018) (0.018) (0.018) (0.018) (0.016) (0.018) (0.018) (0.06) Agriculture % (0.014) (0.014) (0.014) (0.013) (0.012) (0.014) (0.014) 3.0% (0) Municipal FE X X X X X X X Year dummies X X X X X X X Note: First-stage estimates of changes in population brackets from one year to the next (JUMP) on expenditure, both by functions (categories) and by economic nature. Within the economic nature classification, current expenditure comprises expenses such as personnel, consumption, external services and estate maintenance, as well as social and economic transfers to citizens or public or private organizations. Capital expenditure comprises investment expenses (building/purchases of real estate, capital goods) and financial investments. The classification according to functions divides total expenditure into categories, as seen on the table. Each regression discontinuity polynomial as specified. All coefficients and standard deviations are multiplied by 100 for exposition purposes. Share in total expenditure and their standard deviation are also reported. Sample: (yearly data with 25,800 observations). Robust standard errors clustered at the municipality level in parentheses. Significance at the 10% level is represented by *, at the 5% level by **, and at the 1% level by ***. 44

45 Table 15: Fiscal Multiplier - IV estimates by nature and category of expenditure (I) (II) (III) (IV) (V) (VI) (VII) Cross Cross Cross Threshold Parsimonious Cross Cross share in threshold threshold threshold specific (no threshold threshold total 3rd order pol 4th order pol 5th order pol 3rd order pol polynomial) 3rd order pol 3rd order pol expend state-year FE with state iture Current Expenditure time-trend 2.829*** 2.968*** 2.959*** 2.337*** 2.415*** 2.348*** 2.903*** 85.3% (0.811) (0.84) (0.867) (0.783) (0.739) (0.725) (0.837) (0.09) Capital Expenditure 5.388*** 5.538*** 5.534*** 4.724*** 4.541*** 5.034*** 5.393*** 14.7% (1.701) (1.725) (1.786) (1.703) (1.502) (1.769) (1.706) (0.09) Public Administration 7.097*** 6.916*** 6.657** 4.145* 6.733** 4.388* 7.172** 21.7% (2.778) (2.658) (2.709) (2.236) (2.711) (2.31) (2.857) (0.11) Education 4.666*** 4.826*** 4.7*** 3.914*** 4.13*** 4.47*** 4.76*** 42.9% (1.354) (1.37) (1.38) (1.325) (1.291) (1.456) (1.388) (0.12) Housing & Urbanism 7.714*** 7.906*** 7.694*** 6.499*** 6.846*** 6.993*** 7.685*** 12.4% (2.406) (2.42) (2.424) (2.315) (2.282) (2.49) (2.398) (0.08) Health and Sanitation 7.19*** 7.501*** 7.257*** 6.426*** 5.75*** 6.445*** 7.251*** 29.1% (2.15) (2.209) (2.209) (2.231) (1.797) (2.155) (2.176) (0.1) Transportation *** *** *** *** *** ** *** 5.3% (4.995) (4.85) (5.362) (5.098) (4.242) (6.266) (5.225) (0.06) Municipal FE X X X X X X X Year dummies X X X X X X X Note: Multipliers estimates using a two-stage least squares estimator using JUMPs in population brackets as instruments. For the definition of JUMP, refer to table 3. Expenditure is classified both by functions (categories) and by economic nature. Within the economic nature classification, current expenditure comprises expenses such as personnel, consumption, external services and estate maintenance, as well as social and economic transfers to citizens or public or private organizations. Capital expenditure comprises investment expenses (building/purchases of real estate, capital goods) and financial investments. The classification according to functions divides total expenditure into categories, as seen on the table. Each regression discontinuity polynomial as specified. Share in total expenditure and their standard deviation are also reported. Sample: (yearly data with 25,800 observations). Robust standard errors clustered at the municipality level in parentheses. Significance at the 10% level is represented by *, at the 5% level by **, and at the 1% level by ***. 45

46 Table 16: Asymmetry in fiscal multipliers: positive vs negative jumps by nature and category (I) (II) (III) (IV) (V) (VI) (VII) Cross Cross Cross Threshold Parsimonious Cross Cross share in threshold threshold threshold specific (no threshold threshold total 3rd order pol 4th order pol 5th order pol 3rd order pol polynomial) 3rd order pol 3rd order pol expend state-year FE with state iture Current Expenditure time-trend jump> % (1.266) (1.312) (1.4) (1.239) (1.133) (0.975) (1.266) (0.09) jump< *** 3.94*** 3.856*** 3.365** 3.563*** 4.19** 4.097*** (1.101) (1.122) (1.1) (1.04) (1.037) (1.347) (1.178) Capital Expenditure jump> % (1.97) (1.971) (2.064) (1.998) (1.77) (2.484) (1.94) (0.09) jump< ** ** ** 9.826** 9.551** 7.507*** ** (4.422) (4.498) (4.443) (4.511) (3.992) (2.933) (4.643) Public Administration jump> % (2.862) (2.736) (2.804) (2.527) (2.84) (2.526) (2.884) (0.11) jump< * * * * * * (10.3) (9.794) (9.658) (7.248) (10.449) (7.054) (11.244) Education jump> % (1.804) (1.809) (1.854) (1.787) (1.743) (1.815) (1.789) (0.12) jump< *** 7.824*** 7.573*** 6.843*** 7.168*** 8.466*** 8.269*** (2.527) (2.525) (2.426) (2.402) (2.399) (3.377) (2.765) Housing & Urbanism jump> % (3.664) (3.675) (3.734) (3.593) (3.537) (3.869) (3.576) (0.08) jump< *** *** *** 9.127*** 9.447*** 9.273*** *** (3.54) (3.508) (3.419) (3.313) (3.237) (3.566) (3.618) 46

47 Table 16: Asymmetry in fiscal multipliers: positive vs negative jumps (cont ed) (I) (II) (III) (IV) (V) (VI) (VII) Cross Cross Cross Threshold Parsimonious Cross Cross share in threshold threshold threshold specific (no threshold threshold total 3rd order pol 4th order pol 5th order pol 3rd order pol polynomial) 3rd order pol 3rd order pol expend state-year FE with state iture Health and Sanitation time-trend jump> % (4.026) (4.196) (4.353) (4.11) (3.008) (4.514) (4.064) (0.1) jump< *** 8.354*** 8.012*** 7.871*** 7.598*** 6.917*** 8.331*** (2.339) (2.36) (2.25) (2.502) (2.209) (2.209) (2.359) Transportation jump> % (4.43) (4.276) (4.736) (4.607) (3.984) (5.752) (4.522) jump< * * * (39.466) (38.833) (38.729) (41.631) (35.909) (35.477) (44.514) Municipal FE X X X X X X X Year dummies X X X X X X X Note: Multipliers estimates using positive and negative JUMPs as two separate instruments. For the definition of JUMP, refer to table 3. Expenditure is classified both by functions (categories) and by economic nature. Within the economic nature classification, current expenditure comprises expenses such as personnel, consumption, external services and estate maintenance, as well as social and economic transfers to citizens or public or private organizations. Capital expenditure comprises investment expenses (building/purchases of real estate, capital goods) and financial investments. The classification according to functions divides total expenditure into categories, as seen on the table. Each regression discontinuity polynomial as specified. Share in total expenditure and their standard deviation are also reported. Sample: (yearly data with 25,800 observations). Robust standard errors clustered at the municipality level in parentheses. Each regression discontinuity polynomial as specified. Sample: (yearly data with 25,800 observations). Robust standard errors clustered at the municipality level in parentheses. Significance at the 10% level is represented by *, at the 5% level by **, and at the 1% level by ***. 47

48 Figure 1a,1b: FPM Transfer to municipalities (all years) Note: Scatterplot of FPM transfers (left panel) and per capita FPM transfers (right panel) versus population size. Larger dots represent averaged over 200-inhabitant bins. Vertical lines refer to the lower seven legislated thresholds. 48

49 Figure 2: Change in Transfer/GDP around thresholds Note: The solid line is a split third-order polynomial in population size, fitted separately on each side of the pooled FPM thresholds at zero (population size is normalized as the distance from the above or below threshold; symmetric intervals with no municipality in more than one interval). The light-colored lines are the 95% confidence interval of the polynomial. Scatter points are averaged over 220-unit intervals. 49

50 Figure 3: Change in Expenditure/GDP around thresholds Note: The solid line is a split third-order polynomial in population size, fitted separately on each side of the pooled FPM thresholds at zero (population size is normalized as the distance from the above or below threshold; symmetric intervals with no municipality in more than one interval). The light-colored lines are the 95% confidence interval of the polynomial. Scatter points are averaged over 220-unit intervals. 50

51 Figure 4: Per capita GDP growth around thresholds Note: The solid line is a split third-order polynomial in population size, fitted separately on each side of the pooled FPM thresholds at zero (population size is normalized as the distance from the above or below threshold; symmetric intervals with no municipality in more than one interval). The light-colored lines are the 95% confidence interval of the polynomial. Scatter points are averaged over 220-unit intervals. 51

52 Figure 5a,5b: Per capita GDP growth around thresholds Note: The solid line is a split third-order polynomial in population size, fitted separately on each side of the pooled FPM thresholds at zero (population size is normalized as the distance from the above or below threshold; symmetric intervals with no municipality in more than one interval). The light-colored lines are the 95% confidence interval of the polynomial. Scatter points are averaged over 220-unit intervals. The upper (bottom) panel refers to fiscal contractions (expansions) as defined by values of JUMP < 0 (JUMP > 0). 52

53 Appendix: Further results Table A: Long-run Estimates: Reduced-form with current and lagged shock to expenditure (I) (II) (III) (IV) (V) (VI) (VII) Cross Cross Cross Threshold Parsimonious Cross Cross threshold threshold threshold specific (no threshold threshold 3rd order pol 4th order pol 5th order pol 3rd order pol polynomial) 3rd order pol 3rd order pol state-year FE with state Reduced-form time-trend jump_t *** *** *** *** *** *** *** (0.0037) (0.0038) (0.0038) (0.0037) (0.0034) (0.0035) (0.0037) jump_t (0.0036) (0.0036) (0.0036) (0.0035) (0.0034) (0.0034) (0.0036) Reduced-Form with asymmetries jump_t> ** ** ** ** * 0.009** ** (0.0047) (0.0048) (0.0048) (0.0048) (0.0042) (0.0044) (0.0046) jump_t< *** *** *** *** *** *** 0.023*** (0.0062) (0.0062) (0.0062) (0.0061) (0.0061) (0.0059) (0.0061) jump_t-1> (0.0043) (0.0044) (0.0044) (0.0043) (0.0041) (0.0041) (0.0043) jump_t-1< Municipal FE X X X X X X X Year dummies X X X X X X X Note: Reduced-form estimates using JUMPs and lagged JUMPs, allowing for asymmetry for positive and negative JUMPs in the lower panel. For the definition of JUMP, refer to table 3. Each regression discontinuity polynomial as specified. Sample: (yearly data with 25,800 observations). Robust standard errors clustered at the municipality level in parentheses. Significance at the 10% level is represented by *, at the 5% level by **, and at the 1% level by ***. 53

54 Table B: Fiscal Multipliers by component of GDP with multiple instruments (I) (II) (III) (IV) (V) (VI) (VII) Cross Cross Cross Threshold Parsimonious Cross Cross share in threshold threshold threshold specific (no threshold threshold GDP 3rd order pol 4th order pol 5th order pol 3rd order pol polynomial) 3rd order pol 3rd order pol (std. dev) state-year FE with state time-trend Agriculture 4.186*** 4.538*** 4.635*** 4.139*** 3.586*** 2.705*** 4.162*** 20.9% (1.242) (1.276) (1.314) (1.229) (1.142) (1.036) (1.261) (0.14) Hansen J p-value (0.105) (0.128) (0.136) (0.076) (0.054) (0.067) (0.083) Industry % (0.789) (0.795) (0.814) (0.766) (0.729) (0.777) (0.801) (0.14) Hansen J p-value (0.07) (0.062) (0.072) (0.073) (0.071) (0.616) (0.069) Services 1.135*** 1.104*** 1.075*** 0.658** 1.075*** 0.922*** 1.151*** 57.0% (0.306) (0.314) (0.316) (0.299) (0.289) (0.306) (0.313) (0.14) Hansen J p-value (0.268) (0.241) (0.225) (0.229) (0.196) (0.363) (0.285) Municipal FE X X X X X X X Year dummies X X X X X X X Note: Multipliers estimates using a two-stage least squares estimator using positive and negative JUMPs in population brackets as two different instruments in the same regressions. Over-identification statistics and p-values reported. For the definition of JUMP, refer to table 3. Each regression discontinuity polynomial as specified. Share in GDP of each component and their standard deviation are also reported. Sample: (yearly data with 25,800 observations). Robust standard errors clustered at the municipality level in parentheses. Significance at the 10% level is represented by *, at the 5% level by **, and at the 1% level by ***. 54

55 Table C: Reduced-form for components of GDP with positive vs negative shocks to expenditure (I) (II) (III) (IV) (V) (VI) (VII) Cross Cross Cross Threshold Parsimonious Cross Cross threshold threshold threshold specific (no threshold threshold 3rd order pol 4th order pol 5th order pol 3rd order pol polynomial) 3rd order pol 3rd order pol state-year FE with state Agriculture time-trend jump> (0.01) (0.0101) (0.0103) (0.0101) (0.0091) (0.0083) (0.0099) jump< *** *** *** 0.055*** *** *** *** (0.0142) (0.0142) (0.0142) (0.014) (0.014) (0.012) (0.0141) joint significance Industry jump> (0.0065) (0.0066) (0.0066) (0.0065) (0.0061) (0.0061) (0.0064) jump< ** ** ** ** ** (0.0099) (0.0099) (0.0099) (0.0097) (0.0098) (0.0096) (0.0099) joint significance Services jump> (0.0027) (0.0028) (0.0028) (0.0028) (0.0025) (0.0025) (0.0027) jump< *** *** *** ** *** *** *** (0.0037) (0.0038) (0.0038) (0.0037) (0.0037) (0.0036) (0.0037) joint significance Municipal FE X X X X X X X Year dummies X X X X X X X Note: Reduced-form estimates using positive and negative JUMPs as two separate instruments. For the definition of JUMP, refer to table 3. Joint significance test p-values are reported for the reduced-form and first-stage estimates. Each regression discontinuity polynomial as specified. Sample: (yearly data with 25,800 observations). Robust standard errors clustered at the municipality level in parentheses. Significance at the 10% level is represented by *, at the 5% level by **, and at the 1% level by ***. 55

56 Table D: Fiscal Multipliers with multiple instruments by nature and category of expenditure (I) (II) (III) (IV) (V) (VI) (VII) Cross Cross Cross Threshold Parsimonious Cross Cross share in threshold threshold threshold specific (no threshold threshold total 3rd order pol 4th order pol 5th order pol 3rd order pol polynomial) 3rd order pol 3rd order pol expend state-year FE with state iture Current Expenditure time-trend 3.053*** 3.191*** 3.197*** 2.553** 2.651*** 2.44*** 3.137*** 85.3% (0.777) (0.805) (0.822) (0.055) (0.707) (0.721) (0.809) (0.09) Hansen J (0.275) (0.31) (0.35) (0.25) (0.206) (0.083) (0.227) p-value Capital Expenditure 5.424*** 5.551*** 5.594*** 4.692*** 4.599*** 5.428*** 5.388*** 14.7% (1.703) (1.726) (1.79) ( ) (1.504) (1.751) (1.706) (0.09) Hansen J (0.06) (0.061) (0.068) (0.058) (0.044) (0.283) (0.049) p-value Public Administration 6.635** 6.405** 6.162** 3.737* 6.328** 4.059** 6.615** 21.7% (2.732) (2.608) (2.66) (2.219) (2.68) (2.292) (2.796) (0.11) Hansen J (0.022) (0.028) (0) (0) (0.041) (0.152) (0.017) p-value Education 4.906*** 5.066*** 4.973*** 4.119*** 4.417*** 4.54*** 4.96*** 42.9% (1.351) (1.369) (1.376) (1.32) (1.283) (1.458) (1.389) (0.12) Hansen J (0.099) (0.103) (0.109) (0.098) (0.095) (0.082) (0.073) p-value Housing & Urbanism 8.313*** 8.486*** 8.308*** 7.093*** 7.506*** 7.547*** 8.299*** 12.4% (2.339) (2.356) (2.347) (2.244) (2.188) (2.412) (2.343) (0.08) Hansen J (0.361) (0.382) (0.379) (0.329) (0.346) (0.445) (0.324) p-value 56

57 Table D: Fiscal Multipliers with multiple instruments by nature and category of expenditure (cont ed) (I) (II) (III) (IV) (V) (VI) (VII) Cross Cross Cross Threshold Parsimonious Cross Cross share in threshold threshold threshold specific (no threshold threshold total 3rd order pol 4th order pol 5th order pol 3rd order pol polynomial) 3rd order pol 3rd order pol expend state-year FE with state iture Health and Sanitation time-trend 7.579*** 7.834*** 7.578*** 6.936*** 6.286*** 6.655*** 7.658*** 29.1% (1.915) (1.964) (1.925) (2.019) (1.651) (1.865) (1.939) (0.1) Hansen J (0.65) (0.712) (0.737) (0.544) (0.381) (0.835) (0.64) p-value Transportation *** 11.29*** *** 9.81** 9.462** ** ** 5.3% (4.521) (4.368) (4.847) (4.626) (3.937) (5.864) (4.669) (0.06) Hansen J (0.004) (0.003) (0.005) (0.006) (0.003) (0.035) (0.003) p-value Municipal FE X X X X X X X Year dummies X X X X X X X Note: Multipliers estimates using a two-stage least squares estimator using positive and negative JUMPs in population brackets as two different instruments in the same regressions. Over-identification statistics and p-values reported. For the definition of JUMP, refer to table 3. Expenditure is classified both by functions (categories) and by economic nature. Within the economic nature classification, current expenditure comprises expenses such as personnel, consumption, external services and estate maintenance, as well as social and economic transfers to citizens or public or private organizations. Capital expenditure comprises investment expenses (building/purchases of real estate, capital goods) and financial investments. The classification according to functions divides total expenditure into categories, as seen on the table. Each regression discontinuity polynomial as specified. Share in total expenditure and their standard deviation are also reported. Sample: (yearly data with 25,800 observations). Robust standard errors clustered at the municipality level in parentheses. Significance at the 10% level is represented by *, at the 5% level by **, and at the 1% level by ***. 57

58 Table E: First-stage with multiple instruments: positive vs negative jumps by nature and category (I) (II) (III) (IV) (V) (VI) (VII) Cross Cross Cross Threshold Parsimonious Cross Cross share in threshold threshold threshold specific (no threshold threshold total 3rd order pol 4th order pol 5th order pol 3rd order pol polynomial) 3rd order pol 3rd order pol expend state-year FE with state iture Current Expenditure time-trend jump> *** *** *** *** *** *** *** 85.3% (0.0007) (0.0007) (0.0007) (0.0007) (0.0006) (0.0006) (0.0007) (0.09) jump< *** *** *** *** *** *** *** (0.0012) (0.0012) (0.0012) (0.0012) (0.0012) (0.001) (0.0012) joint significance Capital Expenditure jump> *** *** *** *** *** *** *** 14.7% (0.0005) (0.0005) (0.0006) (0.0006) (0.0005) (0.0005) (0.0005) (0.09) jump< *** *** *** *** *** *** *** (0.0009) (0.0009) (0.0009) (0.0009) (0.0009) (0.0008) (0.0009) joint significance Public Administration jump> *** *** *** 0.002*** *** *** *** 21.7% (0.0005) (0.0005) (0.0005) (0.0005) (0.0005) (0.0005) (0.0005) (0.11) jump< ** ** ** ** ** ** ** (0.0007) (0.0007) (0.0007) (0.0007) (0.0007) (0.0007) (0.0007) joint significance Education jump> *** *** *** *** *** *** *** 42.9% (0.0004) (0.0004) (0.0004) (0.0004) (0.0004) (0.0004) (0.0004) (0.12) jump< *** *** *** *** *** *** 0.003*** (0.0007) (0.0007) (0.0007) (0.0007) (0.0007) (0.0007) (0.0007) joint significance 58

59 Table E: First-stage with multiple instruments: positive vs negative jumps (cont ed) (I) (II) (III) (IV) (V) (VI) (VII) Cross Cross Cross Threshold Parsimonious Cross Cross share in threshold threshold threshold specific (no threshold threshold total 3rd order pol 4th order pol 5th order pol 3rd order pol polynomial) 3rd order pol 3rd order pol expend state-year FE with state iture Housing & Urbanism time-trend jump> *** *** *** *** *** *** *** 12.4% (0.0003) (0.0004) (0.0004) (0.0004) (0.0003) (0.0003) (0.0004) (0.08) jump< *** *** *** *** *** *** *** (0.0006) (0.0006) (0.0006) (0.0006) (0.0006) (0.0006) (0.0006) joint significance Health and Sanitation jump> *** *** *** *** *** 0.001*** *** 29.1% (0.0004) (0.0004) (0.0004) (0.0004) (0.0003) (0.0004) (0.0004) (0.1) jump< *** 0.003*** *** *** 0.003*** *** 0.003*** (0.0006) (0.0006) (0.0006) (0.0006) (0.0005) (0.0006) (0.0006) joint significance Transportation jump>0 5.3% 0.001*** *** 0.001*** 0.001*** 0.001*** *** 0.001*** jump<0 (0.0002) (0.0002) (0.0002) (0.0002) (0.0002) (0.0002) (0.0002) * * * * * * (0.0003) (0.0003) (0.0003) (0.0003) (0.0003) (0.0003) (0.0003) joint significance Municipal FE X X X X X X X Year dummies X X X X X X X Note: First-Stage estimates using positive and negative JUMPs as two separate instruments. For the definition of JUMP, refer to table 2. Joint significance test p-values are reported. Expenditure is classified both by functions (categories) and by economic nature. Within the economic nature classification, current expenditure comprises expenses such as personnel, consumption, external services and estate maintenance, as well as social and economic transfers to citizens or public or private organizations. Capital expenditure comprises investment expenses (building/purchases of real estate, capital goods) and financial investments. The classification according to functions divides total expenditure into categories, as seen on the table. Each regression discontinuity polynomial as specified. Share in total expenditure and their standard deviation are also reported. Sample: (yearly data with 25,800 observations). Robust standard errors clustered at the municipality level in parentheses. Significance at the 10% level is represented by *, at the 5% level by **, and at the 1% level by ***. 59

60 Table F: First stage, Reduced form and IV estimates: sample 2000 to 2007 (I) (II) (III) (IV) (V) (VI) (VII) Cross Cross Cross Threshold Parsimonious Cross Cross threshold threshold threshold specific (no threshold threshold 3rd order pol 4th order pol 5th order pol 3rd order pol polynomial) 3rd order pol 3rd order pol state-year FE with state First-stage time-trend jump *** *** *** *** *** *** *** (0.001) (0.001) (0.001) (0.001) (0.0009) (0.0009) (0.0009) jump> *** *** *** *** *** *** *** (0.0011) (0.0011) (0.0011) (0.0011) (0.001) (0.001) (0.0011) jump< *** *** *** *** *** *** *** (0.0021) (0.0021) (0.0021) (0.0022) (0.0021) (0.0019) (0.0021) Reduced-form jump ** 0.009** ** *** ** (0.0041) (0.0042) (0.0043) (0.0043) (0.0039) (0.0038) (0.0041) jump> (0.0053) (0.0054) (0.0055) (0.0055) (0.0048) (0.0049) (0.0053) jump< *** *** *** ** 0.021*** *** *** (0.008) (0.0081) (0.0081) (0.008) (0.0079) (0.0077) (0.0081) IV Estimates jump 1.53** 1.6** 1.58* *** 1.6** (0.78) (0.79) (0.83) (0.76) (0.71) (0.79) (0.79) jump> (1.16) (1.17) (1.29) (1.17) (1.02) (1.06) (1.15) jump<0 2.66** 2.65** 2.59** 2.23** 2.66** 3.98*** 2.8** (1.19) (1.18) (1.17) (1.13) (1.16) (1.58) (1.27) Obs 18,707 18,707 18,707 18,707 18,707 18,707 18,707 Note: Reduced-form, First stage and multiplier IV estimates are reported for both JUMP (overall effect) and "JUMP>0 and JUMP<0" (asymmetric effect) as instruments. For the definition of JUMP, refer to table 3. Each regression discontinuity polynomial as specificed. Data from 2000 to Robust standard errors clustered at the municipality level in parentheses. Significance at the 10% level is represented by *, at the 5% level by **, and at the 1% level by ***. 60

61 Table G: First stage, Reduced form and IV estimates: sample 11,887-47,575 inhabitants (I) (II) (III) (IV) (V) (VI) (VII) Cross Cross Cross Threshold Parsimonious Cross Cross threshold threshold threshold specific (no threshold threshold 3rd order pol 4th order pol 5th order pol 3rd order pol polynomial) 3rd order pol 3rd order pol state-year FE with state First-stage time-trend jump *** *** *** *** *** *** *** (0.0008) (0.0008) (0.0008) (0.0008) (0.0008) (0.0007) (0.0008) jump> *** *** *** *** *** *** *** (0.0009) (0.0009) (0.0009) (0.0009) (0.0009) (0.0008) (0.0009) jump< *** *** *** *** *** *** *** (0.0016) (0.0015) (0.0015) (0.0016) (0.0015) (0.0014) (0.0015) Reduced-form jump *** *** *** *** *** *** *** (0.0038) (0.0039) (0.004) (0.0039) (0.0035) (0.0035) (0.0038) jump> ** ** * * * ** * (0.0048) (0.0049) (0.005) (0.0049) (0.0044) (0.0045) (0.0048) jump< *** *** 0.024*** *** *** *** *** (0.0068) (0.0068) (0.0068) (0.0067) (0.0066) (0.0063) (0.0068) IV Estimates jump 2.54*** 2.57*** 2.52*** 2.27*** 2.01*** 2.12*** 2.58*** (0.73) (0.74) (0.78) (0.77) (0.65) (0.67) (0.75) jump>0 1.89** 1.9** 1.86** 1.59* 1.38* 1.6* 1.87* (0.91) (0.91) (0.99) (0.95) (0.81) (0.84) (0.91) jump<0 3.88*** 3.98*** 3.79*** 3.67*** 3.46*** 3.29*** 4.13*** (1.33) (1.38) (1.33) (1.4) (1.22) (1.26) (1.46) Obs 19,425 19,425 19,425 19,425 19,425 19,425 19,425 Note: Reduced-form, First stage and multiplier IV estimates are reported for both JUMP (overall effect) and "JUMP>0 and JUMP<0" (asymmetric effect) as instruments. For the definition of JUMP, refer to table 3. Each regression discontinuity polynomial as specificed. Data from 2000 to 2010, excluding observations around the first threshold (10,189 inhabitants). Robust standard errors clustered at the municipality level in parentheses. Significance at the 10% level is represented by *, at the 5% level by **, and at the 1% level by ***. 61

62 Figure A: Population distribution Note: Frequency of cities according to population size. Cities below 44,545 inhabitants only. The vertical lines identify the first seven FPM revenue-sharing thresholds. Sample: (yearly data with 25,800 observations). 62

63 Figure B: McCrary Density Tests: Pooled Threshold Year by Year Notes: Weighted kernel estimation of the log density (according to population size), performed separately on either side of the pooled FPM revenue-sharing threshold (1 7) for each year in the sample period. Optimal bin-width and bin size as in McCrary (2008). 63