Texto para Discussão. Série Economia

Transcription

1 Texto para Dscussão Sére Economa TD-E / Muncpalty secesson, voter s preference and persstence of power Paulo Arvate Vladmr Ponczek Av. Banderantes, Monte Alegre - CEP: Rberão Preto-SP Fone (16) /Fax (16) e-mal: cebelma@usp.br ste:

2 Unversdade de São Paulo Faculdade de Economa, Admnstração e Contabldade de Rberão Preto Retora da Unversdade de São Paulo Suely Vlela Dretor da FEA-RP/USP Rudne Toneto Junor Chefe do Departamento de Admnstração Marco Mattos Borges de Olvera Chefe do Departamento de Contabldade Maísa de Souza Rbero Chefe do Departamento de Economa Mara Chrstna Squera de Souza Campos CONSELHO EDITORIAL Comssão de Pesqusa da FEA-RP/USP Faculdade de Economa, Admnstração e Contabldade de Rberão Preto Avenda dos Banderantes, Rberão Preto SP A sére TEXTO PARA DISCUSSÃO tem como objetvo dvulgar: ) resultados de trabalhos em desenvolvmento na FEA-RP/USP; ) outros trabalhos consderados de relevânca dadas as lnhas de pesqusa da nsttução. A sére fo subdvdda em função das prncpas áreas de atuação da FEA-RP/USP: Economa, Admnstração e Contabldade. Veja o ste da CPq na Home Page da FEA-RP: Informações: e-mal: cpq@fearp.usp.br Av. Banderantes, Monte Alegre - CEP: Rberão Preto-SP Fone (16) /Fax (16) e-mal: cebelma@usp.br ste:

3 Muncpalty secesson, voter s preference and persstence of power Abstract Ths paper attempts to evaluate muncpalty secesson n terms of the medan voter response on the executve local electons. Our results mply that ncumbents n the new muncpaltes have a hgher chance of reelecton whch suggests that the medan voter approved the secesson process. Moreover, that effect s enhanced because those partes that managed the muncpalty secesson also had the probablty of persstence n power (n more than one term) ncreased. JEL: H77, D72, H72 Keywords: muncpalty secesson, reelecton, voter s preference and persstence of power

4 1. Introducton Ths paper attempts to evaluate muncpalty secesson n terms of the ex-post medan voter response on the executve local electons. Accordng to the lterature on muncpaltes secesson, the decson undertaken by the local governments nvarably nvolves ex-ante economc arguments. These arguments are concerned wth expected net welfare gans to the medan voter related to the muncpalty or country dvson. Recent studes ncorporate ex-ante economc arguments assocated to effcency of publc provson for large jursdctons (scale gans) and to the costs of havng dverse populatons wthn muncpalty (targetng performance), for secesson or ntegraton (see Alesna and Spolaore,1997, Bolton and Roland, 1997, Person and Tabelln, 2000) 1. For nstance, Brnk (2004) places a framework to analyze the break-up of muncpaltes n Sweden. She fnds that dfferent tax bases, poltcal preferences and populaton sze n the muncpaltes mght justfy muncpaltes` dvson once she dentfes dfferent groups after the secesson. 2 These papers clarfy the role of economc reasons to secede, however the lterature lacks nvestgaton about the ex-post evaluaton of that secesson n terms of the percepton of the ndvduals. The man contrbuton of ths paper s to evaluate the approval of muncpalty secessons n Brazl after ts occurrence. In other words, we test whether that decson to secede n fact was successful to match the preferences of the medan voter and the publc polcy executed n the new muncpalty. Ths test s mplemented usng reelecton results. To the best of our knowledge, ths s the frst paper that attempts to document that relaton. The use of reelecton results as an evaluaton tool on the part of voters has been used extensvely n the poltcal economy lterature (Peltzman, 1992, Brender, 2003, Brender and Drazen, 2005a and Arvate et al, 2007). We use a smlar approach to evaluate muncpalty secesson. The dea supportng ths strategy s that f the secesson was benefcal for the medan voter the ncumbent of new muncpalty can capture these welfare gans through reelecton. 3 The emprcal challenge s to dsentangle the effects of secesson gans from the effects of a hgh-qualty admnstraton. Nevertheless, f we beleve that mayors n seceded muncpaltes condtonal on observed 1 Oates (1972) presents the trade-off arguments necessary for takng the decentralzaton decson. On one hand, a decentralzed government has a better perspectve of ther ctzens, matchng the publc goods provson accordngly. On the other hand, that government mght not consder spllover actvtes nether coordnate redstrbutve and macroeconomc polces takng nto consderaton all the muncpaltes. 2 See Blanchard and Shlefer, (2001), Bradhan and Mookherjee (2000) and Chekbossan (2008) for the exposton of trade-off arguments relatng to bureaucracy appropraton versus targetng when publc polces are desgned. 3 Powell (2000) consders the electons as a tool to control polcymakers n a majortaran electoral system. Ths clam renforces the use of reelecton as a devce to evaluate not only the polcymaker but also the polces mplemented under hs term.

5 fscal varables are as good admnstrators as the ones n the rest of muncpaltes, dfferences n reelecton probabltes capture the publc percepton regardng the secesson. We are also mplctly assumng that the medan voter cannot perfectly dstngush the gans from secesson from those comng from the qualty of the admnstraton. In other words, we beleve that the dfference between benefts and costs comng drectly from the secesson s not observable (a latent varable). The observable varable reelecton takes value one or zero dependng on the value of ths dfference: one ndcates that the medan voter has more benefts wth the decentralzaton than costs and zero otherwse. Therefore, the decson to reelect the ncumbent s the devce used by the medan voter to valdate whether the proposed publc polcy (mplemented throughout the secesson) was successfully put n place. The Brazlan decentralzaton process has two specal features that turn out to be extremely useful for ths analyss 4. Frst, t matters for the voters. The Brazlan muncpal decentralzaton, occurred mostly n the 90 s, was compulsorly voted n a local referendum. Second, the reelecton of the local poltcan or local party after the muncpal secesson through a compulsory electoral process can be consdered a reasonable evaluaton tool of the decentralzaton. 5 We assume that the frst mayor elected after the secesson was a man player n the decentralzaton process and s hold responsble for the decentralzaton s success or falure (Acemouglu, Jonhson and Robnson, 2004). The New 1988 Brazlan Consttuton establshes that the secedng muncpalty process depends on the approval of consttuent poltcal system on the states, the State Legslatve. That characterzes the fact that ths process s controlled by state or federal deputes or local leaders who eventually become mayors n those secedng muncpaltes. Ths decentralzaton or secesson mght happen as a result of cooperatve strategy between local and central governments (Cox and MaCubbns, 1986, Lndbeck and Webull, 1993, Dxt and Londregan, 1996) or as a consequence of local poltcans concerned wth ther career (Leon, Perera and Rennó, 2004, Dermeer, Keane and Merlo, 2005). In order to document the voters response to the decentralzaton process, we analyze the mpact of secesson on two dfferent electoral outcomes: the probablty of party reelecton n the next perod and ts persstence of power (number of electons won consecutvely by that party) after 4 Brazl s a developng country wth per capta GDP about US$ 3,460 n It s the fourth largest democracy n the world, after the Unted States, Inda, and Indonesa. Its 1988 New Consttuton defnes as major admnstratve dvson s muncpaltes whose mayors and legslatve representatves are elected every four years n a legally bndng calendar. It s the frst Latn Amercan country to adopt unversal suffrage, however votng s compulsory for all electons. Ths last aspect s n contnuous debate n terms of the mpacts of mandatory votng on less sophstcated voter s decson (Lma Jr., 1990, Power and Roberts, 1995 and Elkns, 2000). In addton, Brazl presents a great varaton n soco-economc varables across states. For nstance, n 2003 the GDP per capta of the two rchest states, Sao Paulo (Southeast regon n Brazl) and the Federal Dstrct Brasla (MdWest regon), was around US$ 5,500, smlar to Chle and the Czech Republc, whle the two poorest states, Maranhão and Pauí (Northwest regon), dsplayed a per capta ncome of US$ 850, close to Cameroon and Gunea-Bssau. 5 Indvdual reelecton was prohbted untl 1997.

6 the break-up 6. An mportant challenge of ths type of study s to truly dentfy a causal relatonshp from the muncpalty secesson decson to reelecton results. It could be the case that muncpaltes wth popular local leaders are more lkely to secede snce charmastc leaders may nfluence over the voter helpng breakng-up approval. Ther popularty could also be assocated to success n future electons. In ths case, our results would be capturng a spurous relaton between secesson and reelecton chances. Another source of endogenety arses from the fact that postve economc shocks n sub regons of a muncpalty could deflagrate a secesson movement nsde the regon n order to capture the entre gans of the shock. If the shock ncreases the probablty of reelecton, our results could be based. As a robustness check, we also estmate a system of smultaneous equatons usng the fact that the muncpalty s assessment to secede s affected by the number of seceded muncpaltes to create an nstrument for the secesson varable 7. Our results suggest that the secesson process was approved by the medan voter. That s, ncumbents n the new muncpaltes have a hgher chance of reelecton. Moreover, that effect s enhanced because those partes that managed the muncpalty secesson also had the probablty of persstence n power (n more than one term) ncreased. Ths result s contrary to Tebout (1956) model where ndvduals vote wth ther feet, snce we present evdences that ndvduals vote for (approve) the change n the sze of the jursdctons and consequently the publc sector s budget by secesson and not movng away from that muncpalty 8. Next secton descrbes the data. It also addresses the nsttutonal change occurred n 1988 wth the new Brazlan Consttuton. Secton 3 lays out the emprcal model and results. Secton 4 concludes. 2. Data The data come from three sources. Frst, nformaton on 1992 muncpalty electon s obtaned n the Regonal Electoral Courts (State Courts). Second, the data that lnk mayors and consttuent state authorty (governor and the majorty party n the State Legslatve) s found on the Insttuto Unverstáro de Pesqusas do Ro de Janero IUPERJ ( de dados.php). Thrd, IPEADATA ( provdes the other varables (see Table A1 n the appendx). We create two ndcator varables to characterze the mpact of muncpalty secesson: Reelecton and Poltcal Persstence. The former s equal to one f the mayor party between See Acemoglou and Robnson (2007) for detals on persstence of power. 7 We use the dea proposed n Arvate et all (2008) whch clams that the medan voter s not fully nformed ex-ante on the benefts of secesson and use the nformaton on hs counterpart s neghbor muncpalty. 8 See also Gramlch and Rubnfeld (1982) and Brueckner (2004) for emprcal applcaton of Tebout s model.

7 and 1992 (the frst term after the New Consttuton) s reelected to a second term ( ) and zero otherwse. The latter can have four values: zero, one, two and three. It ndcates a poltcal persstence of a party on power. It s equal to zero f the mayor party does not get reelected to a second term ( ), t equals to one f the mayor party goes to a second term ( ), t assumes value of two f the mayor party obtans a sequence of two terms ( and ) and fnally, t s three f the frst mayor party elected after the New Consttuton accomplshes a sequence of three terms n power ( , and ). Table 1 presents the persstence of partes n power data. Table 1: Persstence of partes n power Number of All sample Non-seceded Muncpaltes Seceded Muncpaltes reelecton Frequency Percentage Frequency Percentage Frequency Percentage None One , Two contnuous and consecutves , Three contnuous and consecutves , Average number of reelecton 0,37 0,36 0,42 Total , ,6 Note that about four percent (4%) of the muncpaltes mantan the same party n power after four electons (three reelectons). Ths evdence renforces the use of tme length of the party n power as a resultant process after the muncpalty secesson 9,10. We also observe that seceded muncpaltes contnuously reelect the mayor s party more tmes on average. The largest dfference between seceded and non-seceded muncpaltes occurs n the second reelecton. Seceded muncpaltes are 16% (14,84%/12,78%) more lkely to reelect the party one tme, 46% more lkely to reelect t two tmes and 16% less lkely to reelect t three tmes compared to non-seceded muncpaltes. We should keep n mnd that those fgures n Table 1 are uncondtonal to any characterstc of the muncpaltes and could possble be capturng dfference n the muncpaltes other than the secesson. In order to deal wth ths problem, we use two types of control varables: at the muncpal 9 The lterature focused on the persstence n power s related to two aspects: () famly dynasty (Dal Bó, Dal Bó and Snyder, 2007, Nartom, Soares and Assunção, 2007) and () legslatve careers (Dermer, Keane and Merlo, 2005; Merlo and Mattozz, 2005). Even though the structure of partes n Brazl mght not be fully organzed (Perera and Muller, 2003 and Leon, Perera and Rennó, 2004), we want to test whether the preferences of the medan voter are manfested n the partes electon and f these partes manage to take advantage, n terms of reelecton, of well succeed polces such as muncpalty secesson. 10 Two reasons justfy the choce of partes as the unt of analyss. Frst, the poltcans may be lookng for prospectve careers n that party (Perera and Muller, 2003 and Merlo and Mattozz, 2005). Second, f the party n Brazl were really weak, one would observe the majorty of projects rejected n the Congress. However Rcc (2003) presents that out of 1430 projects, only 30 (15 out of 30 partally) were rejected. Ths ndcates that, at least some organzaton n the Congress s acheved through partes assocaton. Moreover, a sgnfcant postve effect of secesson n partes reelecton only strengthens our argument

8 and poltcal levels. Concernng the characterstcs of muncpalty: the average tax revenue per capta, the average ntergovernmental grants receved from both state and federal governments per capta, the average educaton expendture per capta, the average health expendture per capta, the average nvestment per capta, number of houses n the muncpalty wth sewage treatment (census data, 2000), number of houses n the muncpalty wth electrcty (census data, 2000), Gn ndex (census data, 2000), the percentage of houses n the muncpalty wth trash servce (census data, 2000), the (Ln) human captal on the muncpalty 11 (census data, 2000) and the total populaton on the muncpalty n (census data, 2000). The publc fnance varables are centered on muncpalty term for two reasons. Frst, t permts us to control for the characterstcs of the secedng muncpalty once we do not have data pror ts creaton. Second, the perod of s the frst term after the Consttutonal change occurred n 1996 whch made more dffcult to approve the muncpaltes secesson process see Table 2, column (3)). The poltcal characterstcs of each muncpalty are: the average muncpal legslatve expendture per capta ( ), the percentage of houses wth TV (census data, 2000), the percentage of rural populaton (census data, 2000), the lteracy rate (census data, 2000) and the relatonshp between the mayor and the consttuent power (governor and the state Legslatve: a dummy wth value equal one f the mayor elected n 1996 belongs to the same party of the governor elected n 1994 and zero otherwse; and a dummy wth value equal one f the mayor elected n 1996 belongs to the same party of the one wth the majorty seats n the state Legslatve n 1994 and zero otherwse). The legslatve expendture can be seen as the cost of ther admnstratve autonomy. The percentage of rural populaton and lteracy rate are voters characterstcs. Gven that the vote s compulsory n Brazl, a low lteracy rate (medan of 0.78 on sample) and a consderable number of rural populaton n muncpaltes (medan of 0.41 on sample), suggests a more naïve voter n Brazl. For ths reason, we ncorporate TV (medan of 0.75 of houses have TV) as a poltcal control. That nfluences the means that the nformaton reaches the voters. Last, the relatonshp between the mayor and consttuent power s used to control for other poltcal forces that could nfluence the secedng process. From the perspectve of the poltcan, the local government (mayor) electon depend on supportng poltcal (ncluded allances wth partes n the state legslatve) at the tme of ther (personal, party or coalton) reelecton, fnancal supportng of party to fght over the electon and the possblty of promoton on career passes by one hgher level of government (Enkolpov and Zhuravskaya, 2007, Leon, Perera and Rennó, 2004). Therefore, t s possble to assume that a mayor or a state legslatve offce has also a perspectve n terms of regressve and 11 The human captal s the expected present value of the annual earnngs (dscounted at 10% per year) assocated wth the schoolng level and experence (age) of the economcally actve populaton (15 to 65 years old). It s annually calculated by IPEA.

9 state 13 As consequence, we have a large number of muncpaltes created n the perod between the progressve ambton. 12 Ths s corroborated by the Brazlan 1988 Consttuton. It was bult after the end of Authortaran regme and establshed new rules of muncpaltes break up. These rules defne a set of ncentves as key actors to determne the fnal number of muncpaltes wthn a 1988 Consttuton as shown n Table 3. Ths table presents the muncpaltes emancpaton by regons. There s no systematc dfference among regons n terms of number of secedng muncpaltes even though there s a sgnfcant dfference among them n terms of per capta ncome. 14 The regons North and Northeast are the poorest n Brazl. The regons South East and South are the rchest. Independent of ths, the regons Northeast and South were the regons wth hgher number of breakng up muncpaltes n Brazl between 1988 and Table 3: Number of break up muncpaltes n Brazl by Regon Brazl North Northeast South East South Centre West Source: Bremaeker (2001) 3. Emprcal Strategy We analyze the mpact of muncpalty secesson on two poltcal outcomes: the probablty of party reelecton and the probablty persstence of partes n power for two or more terms 15. In order to nvestgate the frst relatonshp, we used the followng Probt model: Pr ob(re electon = 1 Secesson, X ) = Φ( α. Secesson + γx ) [1] where Reelecton s a dummy varable that ndcates f the mayor s party on muncpalty was reelected to the term; Secesson s another dummy varable that ndcates f the muncpalty was seceded durng the perod; and X s a vector of control varables. 12 Dermeer, Keane and Merlo (2005) show the poltcal economy of the US Congressonal careers. 13 See Appendx 2 for detals about the consttutonal rules of muncpaltes s secesson. 14 The regons South East and South are the rchest ones. The regons Northeast and South were the regons wth hgher number of break up muncpaltes n Brazl n the perod. 15 We also estmated a model usng party deology nstead of party per se (Coppedge, 1997 and Arvate, Avelno and Lucnda, 2007). The dea of ths strategy conssts of elmnatng the fact that a poltcan can move from one party to another but t s harder to change hs deology. For nstance, t s easer for a poltcan to move from a left party (PT) to a center-left (PDT) than to a center-rght one (PPB). All the results are smlar.

10 The followng Posson regresson model examnes the relatonshp between secesson and persstence n power: e Pr ob( Persstence = y λ ) = [2] y where ln(λ )=α. Secesson +γx and Poltcal Persstence s a count varable (0,1,2 and 3) λ λ y whch ndcates the number of terms whch the mayor party on muncpalty perssts nnterruptly n power. More specfcally, y =0 when there was no reelecton to the term; y =1 when there was only one reelecton to the term ; y =2 when there was two reelectons to and terms; y =3 when there was three reelectons to , and terms. A caveat of ths strategy s the possble endogenety of the muncpalty secesson varable. As dscussed before, muncpaltes wth charsmatc local poltcal leaders may be more lkely to secede snce these leaders may nfluence the decson of breakng-up, or exstence of postve economc shocks n sub regons of a muncpalty could deflagrate a secesson movement nsde the regon n order to capture the entre gans of the shock. As a robustness strategy, we also estmate the followng smultaneous model by Full Informaton Maxmum Lkelhood (FIML) 16 : Re electon = 1( α. Secesson + γ1 X + ε1 > 0 ) [3] Secesson = ( βsecesson _ State + γ X + ε 0) [4] > ( ε1, ε 2 X ) ~ N(0,0,1,1, ρ) where 1( ) s the ndcator functon, Secesson_State s number of muncpalty that seceded wthn the state of muncpalty n the perod; and ρ s the correlaton between the error terms (ε 1 and ε 2 ). A smlar strategy s pursued n the case of persstence n power. A FIML s also used 17. However, n ths case, the model s characterzed by a Posson model equaton and a Probt model equaton: Pr ob( Persstence α Secesson _ State + γ1 X + ε1 αsecesson _ State + γ1x + ε1 exp( e ) e = y Secesson, X ) = [5] y Secesson = ( βsecesson _ State + γ X + ε 0) [6] > ε1, ε X ) ~ N(0,0,1,1, ρ) ( 2 16 See Wooldrdge (2002) p. 478 for more detals 17 See Terza (1998) and Romeu and Vera-Hernandez (2005) for more detals.

11 The dea behnd the use of Secesson_State as an nstrument of the muncpalty secesson has to do to the fact that the muncpalty s not fully nformed about the effcency gans of the secesson n the provson of publc goods. On one hand, breakng-up a muncpalty may decrease the amount of publc good provded to each ctzen snce there could be gans of scale on ts producton (Oates,1972). On the other hand, breakng-up can be a good soluton for the ctzens snce t elmnates the omtted preferences problem exstent n a communty when there s publc good provson (Tebout, 1956) and ncreases the per-capta amount of transfers from the federal government to the muncpalty ncreases because the law whch rules the transfers benefts smaller ctes. The number of neghbor muncpaltes that seceded wthn the state s valuable for the muncpalty because t brngs addtonal nformaton about ths trade-off 18. It s mportant to observe that a fundamental assumpton of our approach s that, condtonal on X, there s no correlaton between the number of muncpaltes secesson wthn the state n the prevous term and ε 2 (consequently no correlaton wth ε 1 ) Reelecton results The results for both Probt and FIML regressons are reported n Table 4. The estmatons suggest that those muncpaltes created after secesson between 1989 and 1992 have a hgher probablty to reelect the mayor s party to the term. The results ndcate that the secedng process s successful to match the preferences of the medan voter and the publc polcy executed n the new muncpalty. The ncumbents of new muncpalty benefted from these net welfare gans through a hgher probablty of beng reelected. The results are robust wth respect to dfferent specfcatons. All estmated coeffcents of the Secesson are sgnfcant at the 1% level. For nstance, n the Probt sngle equaton model wth both poltcal and muncpaltes controls, t s 0.421, whch mples a margnal effect of Therefore, a mayor of a seceded muncpalty s 13.5 percentage ponts more lkely to be reelected, the coeffcent estmated coeffcent. After takng nto account the possble endogenety of the Secesson varable, ts mpact on reelecton becomes even stronger. The coeffcent estmated n the system of equatons model s 1.370, mplyng a margnal effect of Table 4: Reelecton (Probt) Independent varable Dependent Varable: Mayor Party Reelecton Probt FIML Probt FIML Probt FIML Secesson 0.280*** 1.804*** 0.268*** 1.352*** 0.421*** 1.370*** 18 A formal model of the relatonshp between muncpalty secesson and number of secedng muncpaltes wthn state s developed by Arvate et all (2008). 19 Snce the varable Secesson s bnary the margnal effect s ndeed an effect of a dscrete change of dummy varable from 0 to 1. The tables wth the margnal effects are avalable upon request.

12 (0.063) (0.286) (0.066) (0.258) (0.070) (0.294) Muncpaltes Controls NO YES YES Poltcal Controls NO NO YES Observatons Note: *** Sgnfcant at the 1% level. Standard errors n parentheses. Table 5 depcts the results of the Probt regresson [4]. They suggest that the number of secedng muncpaltes wthn state between 1988 and 1992 (Secesson_State) postvely affected the burnng of new muncpalty. (at 1% level). Ths result corroborates the model proposed n Arvate et all (2008): Table 5: Full Informaton Maxmum Lkelhood (Probt) Independent varable Dependent Varable: Secesson [1] [2] [3] Secesson_State 0.006*** 0.006*** 0.003*** (0.001) (0.001) (0.001) Muncpaltes Controls NO YES YES Poltcal Controls NO NO YES Observatons Note: *** Sgnfcant at the 1% level. Standard errors n parentheses Persstence of power Turnng the results to the effect of muncpaltes secesson on the poltcal persstence of mayor party n the power, estmatons on Table 6 suggest that muncpaltes created n nfluence not only the frst reelecton but the mantenance of frst mayor party n the power (more than one term). The secesson of the muncpaltes ( ) sgnfcant ncreases the probablty of havng the frst (the one that won the frst electon after the break up) mayor s party perssted n the power. The result s robust to all dfferent specfcaton, except the FIML estmator wthout ncludng poltcal controls. The ncdence rate rato estmated n the sngle equaton Posson model wth both set of control varables s e = For nstance, these estmates mply that beng a mayor of a seceded muncpalty ncreases the expected number of consecutve terms n power by 72.6% (100*( )) n the sngle equaton model. The FIML wth both set of control varables estmates a much hgher coeffcent of secesson o persstence n power (2.288), whch renforces the sngle-equaton result.. Table 6: Persstence n the Power (Posson) Independent varable Dependent Varable: Number of tmes whch the Mayor Party was Reelected after the succeedng Posson FIML Posson FIML Posson FIML Secesson 0.367*** 1.29*** 0.369*** *** 2.288***

13 (0.068) (0.137) (0.072) (0.266) (0.074) (0.525) Muncpaltes Controls NO YES YES Poltcal Controls NO NO YES Observatons Note: *** Sgnfcant at the 1% level. Standard errors n parentheses. Fnally, table 7 shows the regresson results of the Probt model [6]. As expected, the results are vrtually the same of table 5, confrmng the postve mpact of neghbor s secessons on the muncpalty break up. Table 7: Full Informaton Maxmum Lkelhood (Probt) Independent varable Dependent Varable: Secesson [1] [2] [3] Secesson_State 0.006*** 0.006*** 0.004*** (0.001) (0.001) (0.001) Muncpaltes Controls NO YES YES Poltcal Controls NO NO YES Observatons Note: *** Sgnfcant at the 1% level. Standard errors n parentheses. 4. Concluson Ths paper attempts to estmate ex-post evaluaton of muncpalty secessons occurred n Brazl usng the medan voter response on the executve local electons. We test whether seceded muncpaltes enhance the chances to have the actual mayor s party reelected. That s the same tool used n Peltzman (1992), Brender (2003), Brender and Drazen (2005a) and Arvate et al (2007) to capture the voter s evaluaton on ncumbent s polcy. We also check ths party persstence n the power and fnd those muncpaltes that seceded n the prevous term have a hgher probablty to reelected party s mayor. In addton, the secesson of the muncpaltes ncreases the probablty of havng the frst (the one that won the frst electon after the break up) mayor s party perssted n the power of the next three electons. Although the lterature (Oates, 1972, Gramlch and Rubnfeld, 1982 and Brueckner, 2004) presents the trade-offs arguments for decentralzaton that can mpose gans or losses to the voters, ths paper attempts to capture the correct evaluaton on the part of the voters. If the secedng process and the publc polcy executed n the new muncpalty s successful to match the preferences of the medan voter, that party s reelected. Ths result suggests that Brazlan ndvduals can vote for a change n the sze of ther jursdcton and ther publc sector budget. Ths corroborates Brnk s (2004) pont that secesson s also preference-matchng mechansm as opposed to the mgraton perspectve (Tebout (1956). In addton, our conclusons also put n perspectve the ones found n Acemoglu and Robnson (2007).

14 They obtan that economc results favor the elte, beng nvarant to poltcal nsttutons. In partcular, Nartom, Soares and Assunção (2007) fnd the persstence of famly n power as consequence of the colonal hertage (sugar cane and gold cycles). We, on the other hand, show that the partes n Brazl mght capture the benefts of a well executed publc polcy.

15 References Acemoglu, D.; Johnson, S. and Robnson, J. (2004) Insttutons as the fundamental cause of longrun growth. NBER Workng Paper 10481, Natonal Bureau of Economc Research, Cambrdge, Massachusetts. Acemoglu D. and Robson, J. (2007). Persstence of power, eltes and nsttutons. Workng paper 12108, Natonal Bureau of Economc Research, Cambrdge, Massachusetts. Alesna, A. and Spolaore, E. (1997) On the number and sze of natons. The Quarterly Journal of Economcs, vol 112(4), pages Arvate, P; Avelno, G. and Tavares, J. (2007) Fscal Conservatsm n a New Democracy: Sophstcated versus Naïve Voters. CEPESP Workng Paper, Centro de Polítca e Economa do Setor Públco da Fundação Getulo Vargas, São Paulo. Arvate, P; Avelno, G. and Lucnda, C. (2007) Exste nfluênca da deologa sobre o Resultado fscal dos Governos Estaduas Brasleros? CEPESP Workng Paper, Centro de Polítca e Economa do Setor Públco da Fundação Getulo Vargas, São Paulo. Arvate, P. Mattos, E. and Ponczek, V. (2008) Muncpaltes Secesson and uncertanty on publc goods provson. CEPESP Workng Paper, Centro de Polítca e Economa do Setor Públco da Fundação Getulo Vargas, São Paulo. Arzagh, M. and Henderson, J.V. (2005) Why countres are fscally decentralzng. Journal of Publc Economcs, vol 87, ssue 7, pages Blanchard, O. and Shlefer A. (2001) Federalsm wth and wthout poltcal centralzaton: Chna versus Russa. IMF Staff Papers, Internatonal Monetary Fund, vol 48(4), page 8. Bolton, P. and Roland, G. (1997) The breakup of natons: a poltcal economy analyss. The Quarterly Journal of Economcs, vol 112(4), pages Bardhan, P.; Mookherjee, D. (2000) Capture and governance at local and natonal levels. Amercan Economc Revew, vol.90, pages Bremaeker, F.E.J. (2001) Evolução do quadro muncpal braslero no período entre 1980 e Insttuto Braslero de Admnstração Muncpal. Ro de Janero. Brender, A. (2003) The Effect of Fscal Performance n Local Government Electon Results n Israel: , Journal of Publc Economcs, vol. 87, pages Brender, A. and Drazen, A. (2005a). Poltcal Budget Cycles n New versus Establshed Democraces, Journal of Monetary Economcs, vol. 52, pages Brender, A. and Drazen, A. (2005b).How Do Budget Defcts and Economc Growth Affect Reelecton Prospects? Evdence from a Large Cross-Secton of Countres, NBER Workng Paper 11862, Natonal Bureau of Economc Research, Cambrdge, Massachusetts. Brnk, A. (2004) The Break-up of muncpaltes: votng behavor n local referenda. Economcs of Governance, vol 5, number 2, pages

16 Brueckner, J. K. (2004) A Tebout/tax-competton model. Journal of Publc Economcs, vol. 77, pages Brueckner, J. K. (2006) Fscal federalsm and economc growth. Journal of Publc Economcs, vol. 90, pages Chekbossan, G. (2008) Rent-seekng, spllovers and the benefts of decentralzaton Journal of Urban Economcs, vol 63, ssue 1, pages Cox, G. W. and MaCubbns, M. D. (1986) Electoral Poltcs as a Redstrbutve Game. Journal of Poltcs, vol.48, pages Coopedge, M. (1997). A Classfcaton of Latn Amercan Poltcal Partes. The Hellen Kellog Insttute. Workng Paper # 244. November of 1997 Dal Bó, E., Dal Bó, P. and Snyder, J. (2007) Poltcal Dynastes. Workng paper 13122, Natonal Bureau of Economc Research, Cambrdge, Massachusetts. Dermeer, D.;Keane, M. and Merlo, A. A Poltcal Economy Model of Congressonal Carrers. The Amercan Economc Revew, vol. 95, number 1, pages Dxt, A. and Londregan, J. (1996) The determnant of success of specal nterests n redstrbutve poltcs. Journal of Poltcs, vol 58, number, pages Elkns, Z.(2000). Quem ra votar? Conhecendo as conseqüêncas do voto obrgatóro no Brasl. Opnão Públca, 6(1): Enkolopov, R. and Zhuravskaya, E. (2007) Decentralzaton and poltcal nsttutons. Journal of Publc Economcs, vol. 91, ssue 11-12, pages Gramlch, E.M.; Rubnfeld, D.L. (1982) Mcro estmates of Publc Spendng Demand Functons and Tests of the Tebout and Medan Voter Hypotheses. Journal of Poltcal Economy, vol. 90, number 3, pages Leon, E.; Perera, C. and Rennó, L. (2004) Poltcal Survval strateges: poltcal career decsons n the Brazlan Chamber of Deputes. Journal of Latn Amercan Studes, vol. 36, pages Lndbeck, A. and Webull, J.(1993) Balanced-Budget redstrbuton as the outcome of poltcal competton. Publc Choce, number 52, pages Lma Jr., O. B. (1990). Alenação Eletoral e seus Determnantes; notas de pesqusa. Revsta Braslera de Cêncas Socas, 14: Merlo, A. and Mattozz, A. (2005) Potcal Careers or Career Poltcans? NBER Workng Paper 12921, Natonal Bureau of Economc Research, Cambrdge, Massachusetts. Nartom, J.; Soares, R.R. and Assunção, J.J. (2007) Rent seekng and the unvelng of de facto nsttutons: development and colonal hertage wthn Brazl. NBER Workng Paper 13545, Natonal Bureau of Economc Research, Cambrdge, Massachusetts.

17 North, D. (1991) Insttutons. Journal of Economc Perspectves, volume 5, number 1, pages Oates, W. (1972) Fscal Federalsm, Harcourt, New York Peltzman, S. (1992).Voters as Fscal Conservatves. The Quarterly Journal of Economcs, CVII, pages Perera, C. and Muller, B. (2003) Partdos fracos na arena eletoral e partdos fortes na arena legslatva no Brasl: Conexão Eletoral no Brasl. Revsta Dados, vol 45, pages Pessach, A. R. (2005) Can decentralzaton be benefcal? Journal of Publc Economcs, vol. 89, ssue 7, pages Person, T. and Tabelln, G. (2000), Poltcal Economcs. The MIT Press. Cambrdge, Massachusetts. Powell, G.B. (2000) Electons as nstruments of Democracy: majortaran and proportonal vsons. New Haven. Yale Unversty Press. Power, T. J., and Roberts, J. T. (1995). Compulsory Votng, Invald Ballots, and Abstenton n Brazl. Poltcal Research Quarterly, 48(4): Rcc, P. (2003) O conteúdo da produção legslatva braslera: les naconas ou polítcas paroquas? Revsta Dados, vol 46, numero 4. Romeu, A. and Vera-Hernandez, M. (2005) Counts wth an endogenous bnary regressor: a seres expanson approach. Econometrcs Journal, vol 81, pages Rubnchk-Pessach, A. (2005) Can decentralzaton be benefcal? Journal of Publc Economcs, vol 87, ssue 7, pages Terza, J.V. (1998) Estmatng count data models wth endogenous swtchng: sample selecton and endogenous treatment effects. Journal of Econometrcs, vol 84, pages Tebout, C. (1956) A pure theory of local expendtures. Journal of Poltcal Economy, vol. 64, number 5, pages Wooldrdge, J. W. (2002) Econometrc Analyss of cross secton and panel data. Cambrdge, MA. MIT Press.

18 Appendx 1 Table A1: Descrptve Statstcs Varable Observatons Mean Std. Devaton Mnmum Maxmum Reelecton Secesson (Born Muncpalty between 1993 and 1996) Muncpaltes Controls Average Tax Revenue per capta ( ) Average Intergovernmental Grants receved from state government per capta ( ) Average Intergovernmental Grants receved from federal government per capta ( ) Average educaton expendture per capta ( ) Average health expendture per capta ( ) Average nvestment per capta ( ) Total populaton on the muncpalty (2000) E+07 Percentage of houses on the muncpalty wth trash servce (2000) Number of houses n the muncpalty wth sewage treatment (2000) , Number of houses n the muncpalty wth electrcty (2000) , Gn Index (2000) (Ln) Human Captal on the muncpalty (2000) Poltcal Controls Average muncpal Legslatve Expendture per capta ( ) Percentage of houses wth TV (2000) Mayor (1996) s the same Governor Party Elected n (1994) Mayor (1996) s the same Majorty State Legslatve Party Elected n (1994) Percentage of rural populaton (2000) Lteracy rate (2000)

19 Appendx 2 Table A2: Insttutonal ncentves to muncpaltes secesson Rules Secesson Intergovernmental grants to muncpaltes (economc ncentves) Dstrbuton of grants among muncpaltes The Authortaran Consttuton (1967) Three regulatons through the perod: 1. Responsble: Federal government (1967 Consttuton). 2. Responsble: populaton and ncome are the crterons to the dvde (Consttutonal Emend, number 1). 3. From 1969 untl 1979 only the Presdent should authorze the dvson of muncpaltes (Consttutonal Emend, number 11). 1. The federal government shares wth the muncpaltes 10% of two federal tax revenue ( IPI and IR, 1967 Consttuton, artcles 86 and 88.) 2. The state government shares wth the muncpaltes 20% of the largest tax collected, a sales tax - ICMS. 1. Federal A.10% to the state captal; B. 90% to the other muncpaltes. The crtera of dstrbuton depends on per capta ncome (nverse) of the state and populaton (Complementary Law number 5172, artcle 91) 2. State A. The dstrbuton among muncpaltes depends on the valued added of each on the State s GDP (The 1968 Decree, Law 380). The New Consttuton (1988) Responsble: populaton drectly affected (votng consderng only the populaton of the future muncpalty). However both ntal (votng) and fnal (recognton) processes must be approved by the state Legslatve (988 Consttuton, artcle 18, and paragraph 4). After Economc study of economc vablty (Consttutonal Emend, number 15) 2. After ths study a votng occurs wth total populaton affected (both the future muncpalty and orgnal muncpalty s nhabtants). As before, the ntal (votng) and fnal (recognton) processes must be approved by the state Legslatve. 1. The federal government shares wth the muncpaltes 22,5% of two the federal tax revenue ( IPI and IR) 2. The state government shares wth the muncpaltes 25% of the largest tax collected, a sales tax -ICMS. 1. Federal A. 10% to the state captal; B. 90% to other muncpaltes (the crtera of dstrbuton depends on populaton) 2. State A. The dstrbuton among muncpaltes depends on 75% of ther contrbuton on the state s GDP and 25% s dstrbuted accordng to each state s law (The 1988 Consttuton, artcle 157).