Political Power and Long-Run Development: Evidence from Brazil s Regime Transition

Transcription

1 Political Power and Long-Run Development: Evidence from Brazil s Regime Transition Claudio Ferraz PUC-Rio Frederico Finan UC-Berkeley April 2012 Preliminary - Please do not cite Abstract This paper examines the effects of changes in political power on long-run economic development. We use the identity of all the mayors during 1947 to 2000 from three states in Brazil to construct a measure of political concentration. To identify its effects, we exploit the fact that during Brazil s authoritarian regime ( ), military leaders implemented a series of reforms that affected the political power of traditional political families in sugar and cotton-producing areas prior to the regime change. Using this variation, we estimate large negative effects of political concentration on long-run income per capita and other measures of economic development. Key words: elites, regime transition, economic development. JEL: D72, D78, J33. Department of Economics, Pontifícia Universidade Católica do Rio de Janeiro (PUC-Rio), Rua Marquês de São Vicente, 225- Gávea Rio de Janeiro, RJ, , Brasil. cferraz@econ.puc-rio.br; and BREAD Department of Economics, Evans Hall, Berkeley, California ffinan@econ.berkeley.edu; and BREAD, IZA, NBER

2 1 Introduction Recent empirical literature has made significant advances towards estimating the effects of institutions on economic development. A common theme in many of these empirical studies has been the use of identifying variation that assumes institutions persist over time (Pande and Udry 2006). 1 Despite the recent progress, we still have a limited understanding of why institutions, particularly inefficient ones, arise or persist. One predominant view in the theoretical literature argues that institutions are selected by groups with political power based on their distributional consequences (Acemoglu and Robinson (2008), Bardhan (2005)). Accordingly, what matters for institutional persistence and economic development more generally is the distribution of de facto and de jure political power. This paper examines how changes in political power affect long-run economic and political development using data on local governments in three states of Brazil: Ceará, Minas Gerais, and Paraíba. To measure the persistence of political power, we collected data on the identity of all mayors of the municipalities from these states for the period of 1947 to Using the surnames of the mayors in power during this period, we construct indices of political persistence and concentration that measure the extent to which political families held power. The persistence of political elites and concentration of power in Brazil, as in other countries, is quite remarkable. To identify the effects of changes in political power on long-run development, we exploit Brazil s transition to a dictatorship and its implications for the de facto and de jure political power of local elites. From , Brazil was governed by a military dictatorship. When the military assumed control, they implemented a series of economic and political reforms that impacted the political power of the traditional political elites that held power prior to the dictatorship. As a result of these institutional reforms, a new class of political elites emerged which decreased local political concentration during the dictatorship. While these effects occurred in various regions throughout Brazil, they were particularly pronounced in municipalities engaged in export-led agriculture, which for our sample of states refers mainly to the production of sugar and cotton. With the emphasis on an export-driven agro-industry model of production, the traditional elites who had become reliant on government credit and subsidies to remain competitive loss key political and economic support from the state. The loss of political power greatly affected their ability to maintain their monopolistic power during the transition to the dictatorship. This interpretation of the military s influence on the power structure of traditional local elites is by no means new (e.g. Hagopian (1996), Houtzager (1998)). The study is however, quite novel in its use of municipal level data to rigorously test this hypothesis. These data, which include the population and agricultural censuses dating back to 1940, have to our knowledge never been digitalized, let alone analyzed. 1 The notable examples in the economics literature included Acemoglu and Robinson (2001), Banerjee and Iyer (2005), Engerman and Sokoloff (2002). 1

3 Our analysis begins by documenting the striking fact that conditional on post-transition levels of political concentration, places that were highly concentrated prior to the dictatorship are associated with higher levels of economic and political development, as measured in For instance, a one standard deviation increase in political concentration prior to the dictatorship is associated with a 2 percent increase in income per capita and a 6 percent increase in the effective number of candidates running for mayor. While perhaps puzzling at first glance, this positive association between political concentration prior to the dictatorship and long-run development is entirely consistent with the military s efforts to modernize the agriculture sector and centralize authority out of the hands of the traditional oligarchs. Our data allows us to document this in several steps. First, prior to the dictatorship, political concentration was much higher in places with large producers of sugar and cotton, which were the principal export crops for our sample of states during the 1940s. Yet, as we then document, it is precisely these municipalities that are better off in the long run. Based on our index of agricultural concentration, which proxies for places with large sugar and cotton producers, a one standard deviation in the index is associated with a 2.5 percent increase in long-run per capita income and a 4 percent increase in political competition. These associations in the data naturally raise the question as to why these places are faring better in the long run. Here, we present two complementary mechanisms which are again consistent the historical accounts. First, starting in the late 60s during the government of General Artur da Costa e Silva modernizing the agricultural sector became the military s top priority. As Houtzager (1998) writes The hard-liners brought in a new group of technocrats who understood the modernization of agriculture to be critical to stimulating economic growth and reestablishing economic growth. Consistent with this doctrine, we find that in places with large 1940 sugar and cotton producers, the government provided much more financial capital, which was the main policy instrument used by the military for stimulating modernization. Easy access to credit made possible the introduction of machinery into Brazilian agriculture, and we document this by showing that in these places, the share of farms with electricity and tractors grew much more during the period of the dictatorship. As historians have documented these changes in traditional agriculture reduced the agricultural labor force. We show that in areas with large 1940 sugar and cotton producers, the share employment in industry increased substantially. Second, in order to exert control over the modernization process, the military centralized authority at the expense of the traditional oligarchs. Again as Houtzager (1998) explains The military sought to centralize authority out of the hands of the oligarchies by enacting new legislation and created new bureaucratic machinery in the rural areas to circumvent existing state and local government. Consistent with this argument, we find that not only did political concentration decreased in places with large sugar and cotton producers, but that the elections were also much more competitive in these areas. Interestingly, although political concentration 2

4 decreased in these sugar and cotton producing areas, the local elites were not less likely to survive the transition, which is an argument that have been convincingly documented by Hagopian (1996). Overall our results indicate that places that were large cotton and sugar producers prior to the dictatorship have become relatively more economically and politically developed, despite the fact that these areas were initially much more politically concentrated. In-line with the historical accounts of this period, our findings suggest that this change came about because of both the military s agrarian project and centralization of authority away from the local oligarchs. While ideally, we would like to understand the relative contributions of each of these channels towards long-run development, without additional data, it is difficult to do so. The rest of the paper is organized as follows. The next section outlines our argument and provides supporting historical evidence. Section 3 describes the data and how we measure the persistence and concentration of political power. In Section 4, we describe our instrument and present our main findings. Section 5 concludes the paper. 2 The Argument and Historical Background This section provides a historical background on the distribution of political and economic power among local elites during Brazil s democratic period from 1945 to 1964, and how the authoritarian regime affected this distribution of power through a series of economic and political reforms. Rural Elites and Political Power in pre-1964 democratic period Brazil was a democratic republic between 1889 and 1930 ruled by the regional oligarchs from the states of São Paulo and Minas Gerais who rotated control of the federal executive office by freely manipulating elections using fake ballots, dead voters, and coercion (secret vote was only established in 1934). 2 Coffee barons were supported through government policies that favored their rents from exports and political power was exerted by the rural elite through the monopoly of land and the control of votes with family ties playing a crucial role in perpetuating economic and political inequality. 3 This political order was interrupted by Getulio Vargas who ruled Brazil from 1930 to He centralized power, reducing the political role of rural elites, and brought to power a new political class comprised of bureaucrats and military. He fostered industrialization with strong state participation and reformed social security to create new rights for urban workers. In addition to expanding the role of government in the economy, between 1937 and 1945 he governed Brazil under a dictatorship. He abolished elections for governor and mayors and centralized political 2 See Skidmore (1967). 3 See Hagopian (1996) and de Castro Gomes et al (2007). 3

5 power through the appointment of state and municipal chief executives. Municipalities and states lost significant autonomy they were forbidden from issuing debt and lost revenues from interstate taxes. 4 Despite the industrialization process that took place in Brazil starting in the 1930s, by 1960 Brazil remained an exporter of traditional crops. The economic activity was dependent on few crops such as coffee, cotton, cocoa and rubber that were exported in exchange for most manufacturing goods. The majority of Brazil s population still lived in rural areas and worked in agricultural related activities. 5. As in many other Latin American countries, land ownership was still closely associated with political power, especially in the poorest regions. The distribution of land was extremely unequal and landowners of export-oriented crops employed a large number of workers and controlled their votes. This system of control included the payment of low salaries combined with tenancy and sharecropping agreements that allowed the production of domestic food crops inside the big latifundias. 6 Even in municipalities that were more urban and where agriculture was not the chief economic activity, political power was nevertheless concentrated in the hands of few families who could provide patronage through public employment. Because government spending increased significantly from 1947 to 1960, public employment became an important source of clientelism (Hagopian 1996). 7. Combined with the disproportionate political power of local elites was the unequal distribution of power across Brazil s regions. Brazil s political system was reopened in 1945 after Vargas dictatorship and the political power of rural elites was strengthened by the 1946 constitution that increased the representation of smaller and poorer states in congress. 8 Because of this disproportional power, the control of local votes by deputies and mayors were rewarded with transfers from the state and federal governments. Local politicians in the postwar period acted as local brokers of targeted transfers and public service provision in exchange for votes (Hagopian 1996). Under the new democratic regime established in 1945, regional oligarchies organized national elite parties to mobilize mass support. Three main parties were formed. The PSD was founded by the elites who occupied important positions during the Getulio Vargas dictatorship pre State and municipal chief executives appointed by Vargas joined forces with old elite members that supported Vargas. Most local politicians had some relationship with families that had ruled 4 See de Castro Gomes et al (2007). 5 See de Paiva Abreu (2008) and Graham, Gauthier, and de Barros (1987) 6 See Goodman, Sorj, and Wilkinson (1985); Graham, Gauthier, and de Barros (1987). 7 For a specific example, see Murilo de Carvalho 1966 for example. 8 The 1946 constitution established a minimum number of 7 deputies per state creating a strong malapportionment that favored smaller and less developed states. See Nicolau 1997; de Oliveiro Pierucci et al (2007), Soares (1973), Skidmore (1967). 9 The expansion of government during these years allowed them to serve as prosecutors, public hospital directors, sanitary inspectors, chief of local police among other bureaucratic activities than then transitioned into political careers. 4

6 the localities in the past (et al 2007). The UDN was the rival party and represented the oligarchy that lost power under Vargas. Finally the PTB had its origins in the urban labor movement and did not include members from the regional oligarchies pre-vargas. Although the PSD was perceived as the rural-based party of agrarian elites and the UDN as a liberal-democratic party of middle class, at the local level affiliations were determined by local rivalries rather than ideological or programmatic differences. Clientelism shaped party competition and in most local elections, electoral coalitions were formed based on personal and tactical considerations. Both the PSD and the UDN had important leaders who were aligned with the rural oligarchs (de Oliveiro Pierucci et al 2007). Hence, because the same set of families could be in power at different points in time under different coalitions, the real extent of political competition in most municipalities during this period is better represented by the alternation of power across different families rather than the alternation of power across different parties. 10 In early 1960s Brazil faced a growing political polarization that had started under Vargas presidency between 1951 and 1954 and had culminated with a political gridlock induced by the resignation of Jânio Quadros who was elected in His vice-president Goulart was considered a dangerous choice due to his strong ties with the leftist movements and was not acceptable to high-ranking politicians and the representatives of the traditional elites. In the end, he took office but with reduced political power. 11 Goulart presidency brought to power a new class of political elites that fostered populist policies and threatened the survival of the traditional political elites. The traditional economic sectors were concerned with Goulart threats of radical policies such as land reform and the enfranchisement of the illiterates. This concern was supported by parts of the middle class, industrialists, and bankers (Skidmore 1967). In the countryside, rural workers were excluded from the labor laws and social security reforms put forward by Getulio Vargas in Thus, after years of land concentration and growing rural poverty, the early 1960s brought an active demand for economic rights and especially land reform (Houtzager 2000). Finally, the deterioration in economic performance is an important element in the explanation of the continuous political crisis under Goulart and the successful mobilization of political support for the 1964 military coup (de Paiva Abreu 2008). Changes in Political Power Under the Authoritarian Regime The military s coup d état, which was largely a response to a set of redistributive and populist policies that the political elites felt would undermine their political and economic interests, carried two objectives: to foster modernization and economic growth and to reorganize a political 10 Soares (1973) shows that in the Northeast of Brazil, the poorest regions, in the 1954 election 62 percent of deputies were elected under a coalition of different parties. 11 See Skidmore (1967); Cohen (1987). 5

7 system characterized by corruption and clientelistic practices, which the military viewed as a major obstacle to economic growth and political stability (Stepan 1973). To meet this political objective, and given the degree of political and fiscal autonomy state governments enjoyed prior to the dictatorship, military leaders found it imperative to concentrate political power in the hands of the federal executive. As a result, the military implemented a series of institutional changes that affected both the de jure and de facto distribution of power amongst the traditional political elites. Formally, in addition to a set of centralizing fiscal reforms that reduced state resources by approximately 25 percent, the military regime also abolished the collection of local taxes and established a block grant transferred to local governments based on population and strings attached for spending in specific public works such as roads and sanitation. 12 On the electoral front, the new regime implemented a series of institutional acts designed to extend the power of the executive and limit political opposition. For instance, when several of the military-backed candidates failed to win in the gubernatorial elections of 1965, the military responded by decreeing two institutional acts. The first of these acts, Institutional Act #2 (October 27, 1965) abolished all existing political parties and required that new parties be affiliated with at least 120 deputies and 20 senators (Skidmore 1988). This decree effectively transformed a multi-party system, which in 1964 had 13 parties, to a two-party system, forcing politicians to decide between the pro-government party (ARENA) and the opposition party (MDB). The second act, Institutional Act #3 (February 5, 1966) replaced the direct election of governors with indirect elections by state legislatures. This decree effectively allowed the military the power to name the heads of state governments. These de jure changes in political power would also lead to de facto changes in power. To control the state executive, and thus the political elites within a state, the military leaders appointed to prominent government positions technocrats with non-political backgrounds and only minimal links to the traditional political groups. For example, of the 22 governors selected in 1970, 50 percent of them were technocrats and non-political, compared to 1966 when only state governors were technocrats (Samuels and Abrucio 2000). By appointing technocrats, who not only wanted to break away from old-style politics but also build their own political base, the military was excluding traditional politicians from state patronage, which had become an important source of political power for traditional elites (Sarles 1982). As a result, factions developed within the government party, which induced intraparty competition. As Samuels and Abrucio (2000) put it ARENA began to split into two factions: one led by politicians with little popular support and few links to state elites, but with extensive links to the military high command, and another led by traditional state elites who had developed careers prior to To further accommodate these intraparty factions, the military 12 See Houtzager (2000). 6

8 government adopted a sublegenda balloting mechanism for local and senatorial elections. Parties were permitted to place up to three candidates for the same office, with each running on a separate label. The most voted-for candidate within the most voted-for party would win (Power 1997). Thus although the use of sublegendas guaranteed the ARENA party victories in almost all municipalities (for instance in 1972, ARENA won in 92 percent of all municipalities), it also fostered local competition by accommodating new elites in areas that previously wouldhave been dominated by a single family (Schmitter 1973). With Brazil s integrated political structure and the reliance of local governments on federal and state resources, changes at the national level can have important consequences for the distribution of political power both at the state and local levels. As?) states if an incumbent state group were to be defeated in state politics (possibly because the federal ins supported their adversaries), then the local elites under their protection would be left without their sponsors. The local elites place was secure only as long as the state government to whom it offered its allegiance could stay in power. Thus, when the military overthrew President Goulart in 1964, marking Brazil s ascent into 21 years of authoritarian rule, it had a profound impact on its intergovernmental relations and consequently redefined, in many cases, the traditional political elites relationship with the state (Roett 1999). On the economic front, the reforms aimed at fostering industrialization via an import substitution strategy and developing an agribusiness exporting sector. The authoritarian regime made the modernization of the rural sector a national priority. They viewed a modern agricultural sector as a pre-condition for economic growth and, at the same time, an important pillar for the industrialization process. Also, by fostering agro-industries, the government could modernize the rural areas without implementing a land reform and changing the patterns of land ownership. After an abortive attempt to modernize large farms through a complicated system of land taxes (Estatuto da Terra), in 1968 the military embarked on a three part plan: increase production and exports, integrate new frontier regions and incorporate rural workers to society ((Houtzager 1998)). As President Médici explained in 1970, The [government s] central goal is development... Here is, exactly, the greatest novelty of [our] policy... Since the 1950s, our development effort has been predominately industrial, in an unbalanced form with the agricultural sector... Our [new] development policy will target substantial growth in agricultural production and the increase of exports, which will lead to the rapid expansion of the domestic market and will induce an expansion of the industrial sector... We expect to increase our exports to strengthen the capacity to import indispensable equipment for the implementation of a current technology and to end the external imbalances already experienced, bringing to the country all the advantages of integration into an extremely dynamic world economy. 7

9 As a result, the military established a system of intense interventions in the rural areas. The mechanisms used by the government included the provision of highly subsidized agricultural credit aimed at the purchase of fertilizers and tractors, the provision of agricultural extension services, and investment in infrastructure such as roads. To appreciate how the agricultural sector changed under the military regime, particular in terms the export crops (e.g. sugar, cotton), consider the case of sugarcane production. Despite the fact that the sugar production in Brazil dates back to the sixteenth century, significant government intervention in the sector only began in the 1930 s when President Getúlio Vargas, who had formed a coalition with the sugar oligarchs, created the Instituto de Açúcar e do Alcool (IAA). The IAA was a self-financed, semi-autonomous organization that regulated all aspects of agricultural and industrial production in the sector, including the determination of prices and regional quotas, and national allocations for the export and domestic markets (Nunberg 1986). Because of Vargas support of the sugar elites, the IAA also served as an important political instrument. To compensate for their declining productivity, sugar oligarchs, particularly in the Northeast, would receive credit and subsidies to protect them from newer and more efficient producers that were beginning to emerge in other regions of the country. As Nunberg (1986) explains IAA became associated with extreme political clientelism by extending credit and subsidies to noncompetitive Northeastern producers, by hiring paternalistically, and in the second Republican period ( ), by providing political support to candidates representing the sugar oligarch... the IAA s intention was not to return the comparative advantage to the Northeast. What the IIA did manage to accomplish was the preservation of many inefficient Northeastern mills that might have failed without government intervention. Although these inefficient policies served a clear political purpose, the institute quota policies also did little to allow the market forces to eliminate these inefficiencies. In response to the world market collapse in 1930, the IAA adopted a quota policy targeted at domestic consumption in order to reduce the industry s exposure to world prices. As a result, throughout much of the 1930 s to 1950 s only a small portion of Brazil s sugar production was exported, and when overproduction did occur, it was used for manufacturing alcohol (Nunberg 1986). However, in 1960, the incentives to maintain such a policy changed dramatically when the U.S. canceled its quota for Cuban sugar and awarded Brazil in 1962 a permanent quota on the American market. Confronted with this unexpected demand shock, the military regime began a policy aimed at increasing sugar exports and modernizing the sector to augment its productive capacity. To achieve this objective, the government embarked on a policy that through both financial incentives and legislative decrees would eliminate non-profitable mills, modernize remaining physical industrial plants, and encourage investments in research and development. In addition to these modernization initiatives, the military regime also responded with a series of institutional reforms both within the IAA and the state, which greatly impacted how 8

10 policy was conducted in the sugar industry. These reforms were based on two main objectives. First, the military reorganized the institute to become more centralized, by reducing the number and power of local regional branches. As results, power was consolidate in a few branches located in the newly more productive regions of the Center-South. The second objective sought to eliminate the impact of clientelistic practices on recruitment and promotion by imposing tighter administrative regulations and by also instilling technocratic norms and values through the recruitment of technical experts for key administrative positions. These reforms in turn diminished the economic and political power of the traditional sugar elites, and instead gave rise to a new set of sugar oligarchs whose power was based on a new model of large-scale export-driven agro-industry. As Nunberg (1986) states: This centralization had important consequences for the power configuration in the sugar sector. The shift of policymaking to higher levels of government weakened the IAA, which had once been a semiautonomous autarchy. Formerly influential Northeastern elites lost access to key decision makers, and those Center-South elites promoting large-scale agro-industry found support from the new makers of state policy. In sum, the variation that we exploit in our empirical analysis is based on the idea that when the military took control in 1964, they instituted a set of reforms that diminished the de facto power of the traditional elites, particularly those engage in large export-based crops, which our sample of states, include primarily sugar and cotton production. As a result, a new class of political elites emerged, which decreased local political concentration, and ultimately led to higher levels of long-run political and economic development. 3 Data To study the long-term impacts of political persistence, we assemble a new dataset spanning the period The database contains historical electoral and census data that until now had never been digitalized. In this section, we describe these data and discuss how we measure political concentration and persistence at the local level. 3.1 Political Concentration To measure political concentration within a municipality, we collect data on the names of mayors for three states in Brazil: Ceará, Minas Gerais, and Paraíba. 13 For each state, we have a list of all the mayors who held office from 1947 to During this period, there has been an average of 13 elections across the three states, with slightly more in Paraíba (see Table 1). In Ceará and Paraíba, only 52 percent of these elections have had a different family assume power (see below 13 We selected these states based on data availability. 9

11 for how we construct this information), whereas in Minas Gerais a different family has been in power for 62 percent of its elections. The remaining rows of Table 1 compare the states along other characteristics. The principal cash crops during the pre-dictatorship period in Brazil were cotton, sugar, and coffee. Among these, cotton and sugar are the most important commodities for our sample of states. As we will document later, the concentration of political elites is on average higher among the municipalities with the highest production per farm of sugar and cotton. Coffee, which also had high production value during this period, particularly in the state of São Paulo, is only relevant in the state of Minas Gerais. For this reason the analysis mostly focuses on the places that produce sugar and cotton. In terms of other socio-economic characteristics, Minas Gerais is much larger with 833 municipalities and a population of 18 million inhabitants. Minas Gerais per capita income of R$505 is much higher than that of Ceará s R$318.24, the second highest in our sample, and its poverty rate is less than 30 percent, compared to 55 percent for Paraíba. Because of these differences across states, all specifications include use state-fixed effects to rely on within state variation. Constructing Family Links To construct our measure of political concentration within a municipality, we assume that mayors belong to the same family if they share at least one common surname. In Brazil, it is common for individuals to have at least two surnames. The first surname is typically the mother s family name, while the second is the father s family name. Table 2 provides a few examples of how we use this information to construct these family links. Our first example comes from the municipality of Barreiras do Piaui. 14 In this municipality, a member of the Barreira family has been in office since For our analysis, we code this municipality as having one political family in office throughout its history. The municipality of Carire of Ceará offers another example of a local oligarchy. In Carire, a member of the Aguiar family has been in office for 9 out of the 12 elections dating back to We still code this municipality as having had only two families in power (instead of 3) because even though a Rodrigues was in power in 1996 and 2000, members of the Aguiar and Rodrigues faimly had at some point married. An example of a slightly less political concentrated municipality is Alterosa in Minas Gerais. Here we see that 4 families have alternated power from 1947 to Figure 2 displays the share of elections that a family is in power for our sample of municipalities. Approximately 60 percent of families have been in power only once, whereas 10 percent of families have been in power for at least 25 percent of the elections. In Figure 3, we plot the proportion of municipalities that elected a mayor from a family that had been in office previously. 14 Although this example is quite illustrative of our coding procedure, we do not include the state of Piaui in our analysis because we only have electoral information dating back to

12 From 1950 s to 1970 s, mayors from previous political families were elected in approximately 25 percent of municipalities. After the 1972 elections, the percentage of municipalities jumps up to 50 percent, and remains there until There is another spike in 2000, where almost 75 percent of municipalities elected a mayor from a previous political family, but this is likely to reflect the fact that in 1997 mayors became eligible for re-election. For an interesting point of contrast, Dal Bó, Dal Bó, and Snyder (2009) compute the proportion of legislators in the U.S. Congress with previous relatives who held office. For the U.S. the proportion of legislators with previous family members is only around 15 percent during the 1850 s and declines steadily over time. We explore the ability of political families to remain in power over time more closely in Table 3. We construct a measure of political persistence by calculating the share of families in power during period t that remain in power after the regime transition in t + 1. In order to illustrate this measure, suppose four families held office during the dictatorship period. Out of these four families, two of them were elected after the democratization. The value of persistence from the dictatorship to the democratic period for this municipality would equal 0.5. Of the families that were in power prior to the dictatorship, 26 percent regained power at some point during the dictatorship. Similarly, 29 percent of political families were able to transition from dictatorship to democratization. While approximately 8 percent of families were in power during all three periods (i.e. pre-dictatorship, dictatorship, democratization), the average number of periods a family was in power is 1.7. Computing these transitions at the municipal-level, we find that on average 27 percent of families within a municipality retained power during the regime change to a dictatorship, and 64 percent of municipalities had one family survive the transition. On average, 30 percent of families within a municipality succeeded after the end of the dictatorship, and 77 percent of municipalities had one family maintain power from the dictatorship to democratization. While persistence of political power at the local level appears quite high, it is even more pronounced at higher levels of government. For instance, Fleischer (1981) finds that from , 87.5 percent of ARENA federal deputies and 63.6 percent of MDB federal deputies had relatives in politics. Whereas prior to the dictatorship from , 99 percent of federal deputies from PSD, UDN, PR (the largest parties at the time) had relatives in politics. Also using a similar definition based on family lineage, Hagopian (1996) finds that of the 80 members of the political traditional elites in Minas Gerais that occupied a state or federal post prior to the dictatorship, 55 percent survived politically after the regime. 15 The remaining rows of Table 3 present our principal measures of political concentration: a Herfindahl index for the share of terms governed by the same political family. Specifically we 15 Specifically, Hagopian (1996) defined as a traditional elite prior to the dictatorship: any politician whose previous generation was also in politics or anyone who belonged to the cliques that founded the two major elite parties, PSD and UDN, and who signed the Manifesto dos Mineiros. 11

13 compute the Herfindahl index, H mt, for municipality, m, during period t as: ( Number of elections family i has been in powerimt ) 2 H mt = i Total number of elections mt As we see in Table 3, prior to the dictatorship municipalities have on average a Herfindahl index of 0.35, which suggests that the effective number of political elites (i.e. 1/H mt ) during this period was 3.2 families. While on average political concentration remained relatively constant during the dictatorship, we do see an increase during the democratization period, as the effective number of political families decreased to Additional data Several of our outcome variables come from Brazil s 2000 population census, which are aggregated to the level of the municipality. These variables include average income per capita, average education level of the adult population and the share of households with electricity. To get a measure of economic activity in the municipality, we merge these data with the industrial and agricultural censuses of From these data, we compute the share of employment related to industry and agriculture. Our final set of outcome variables are created from data obtained from Brazil s electoral commission. With these electronic files, we compute standard measures of political competition, such as: the average number of candidates for mayor during the postdictatorship mayor elections. In the analysis, it is important to account for Brazil s vast regional differences in economic development. From Brazil s federal statistical bureau (Instituto Brasileiro de Geografia e Estatística (IBGE)), we also collect various geographical attributes of the municipality: latitude, longitude, size of the municipality, and distances to capital cities. To control for socio-economic differences across municipalities prior to 1947, we digitalized the 1940 agricultural, industrial, and population censuses, as well as the agricultural censuses. From these censuses, we construct a series of important socio-economic characteristics of the municipality, such as population size, employment rate, share of workers in the agricultural sector, average production of firms, and share of land devoted to large-scale production. One issue that arises when using data that date back to the 1940s is that municipalities will split over time. In these cases, we assign to newer municipality the 1940 value of the municipality from which it originated. We then adjust our standard errors by clustering on the municipalities that existed in In Table 4, we present basic summary statistics for the historical variables we use our analysis. The average population for our sample is 32,000 inhabitants, with 39 percent of employment belonging to agriculture. Literacy rates during the 1940 were quite low, with over 72 percent of the adult population not being able to read or write. On average, 19 percent of the population is 12

14 black, which given low mobility during this period might serve as a good proxy for areas where slave plantations existed. Approximately 39 percent of the municipalities had a mayor from UDN party (which later became a principal part of the military party) prior to the dictatorship, whereas only 13 percent of the municipalities had a mayor for the PTB party prior to the coup. Table 4 also present summary statistics for rental prices during 1940s, which should serve as a proxy for income during that period (as we see, the variable strongly predicts current day income per capita). 4 Results 4.1 Political Concentration and Economic Development We begin our analysis by documenting the associations between political concentration over the period and our main indicators of economic and political development as measured in As described in the data section, our measure of political concentration is a Herfindahl index for political families between 1947 and In Figure 3, we present three plots depicting the unadjusted relationship between political concentration and three indicators of development: log income per capita, political competition (as measured by the effective number of candidates that ran for mayor the 2000 elections), and the average years of schooling among the adult population. For each of these outcomes, we plot its average value within a 0.1 unit-wide interval of the Herfindahl index, along with vertical bars representing point-wise 95 percent confidence intervals. The patterns presented in Figure 3 are quite striking. Municipalities with higher levels of political concentration are associated with much lower levels of development along each of the indicators. For example, when measured by income per capita, the difference between municipalities with a Herfindahl index of 0.10 and 0.30 is almost 30 log points. Similarly, almost 1 year of schooling separates municipalities with effectively 3 political families versus 10 elite families, where the effective number of political families is simply the inverse of the Herfindahl index. Yet despite these impressive relationships, these correlations may simply reflect differences in geographic or socio-economic characteristics that are correlated with political concentration. To investigate the robustness of these relationships, we estimate the following regression: y ij = β 0 + β 1 pc ij + X ijθ + ν j + ε ij (1) where y ij indicates a long-run outcome (e.g. per capita income) for municipality i in state j and pc ij denotes political concentration. The vector X ij includes various geographical characteristics 16 Using the other measures of political concentration provide the very similar results. 13

15 (e.g. latitude, longitude, altitude, distance to the state and federal capital and the area of the municipality) including a series of pre-period characteristics from the 1940 census. We also rely on within state variation by controlling for state intercepts, ν j, and ε ij denotes unobserved determinants of long-run development. For a causal interpretation of the coefficient β 1, we need to assume that conditional on a rich set of 1940 characteristics of the municipality and geographical controls, E[ε ij X ij ] = 0. The results from estimating these specifications are presented in Table 5. In column 1, we present OLS estimates of the association between political concentration and log per capita income, adjusting for our set of geographical and 1940 socio-demographic controls. As suggested by the graphical analysis, there is a strong negative relationship between political concentration and per capita income. A one standard deviation increase in political concentration is associated with 2.1 percent decline in income per capita. In addition to the effects of political concentration, we also find that the share of the 1940 population employed in agriculture is negatively associated with long-term development, as is the share of the black population. Rental rates for homes during the 1940 are strongly predictive of income per capita in 2000, with the coefficient implying an elasticity of Consistent with the cross-country evidence, we also find that share of foreigners and the share of Protestants in a municipality are also strongly correlated with long-run development. Overall, political concentration, together with these additional controls, explains 73 percent of the variation in per capita income in In columns 2 and 3, we investigate how political concentration correlates with the other indicators of economic and political development. As depicted in Figure 3, municipalities where political power has been more concentrated historically are associated lower years of schooling, but are not necessarily less competitive: The coefficient will negative is not statistically significant. Our results in columns 1-3 suggest a strong negative correlation between political concentration during the 1947 to 2000 period and economic development. These correlations, however, do not capture the possibility that political competitiveness at the local level may have been quite different during Brazil s transition in and out of a military dictatorship. Moreover, political concentration during these periods may have impacted long-run development differently. To investigate this possibility, we reconstruct our Herfindahl index for the elections prior to the dictatorship, during the dictatorship, and post-dictator. Table 6 presents the results from a series of regressions that estimate the associations between political concentration during each of the three periods and our measures of long-run development. Consistent with the aggregate relationship, political concentration during democratization ( ), and to a lesser extent during the dictatorship ( ), are both negatively associated with economic and political development. For instance, a one standard deviation increase in political concentration is associated with a 3.4 percent decline in income per capita and

16 years decline in average years of schooling. However, conditional on political concentration during these two periods, we find that places that were less competitive politically prior to the dictatorship are actually faring better in the long run. A one standard deviation increase in political concentration prior to the dictatorship is associated with 2 percent increase in income per capita. These municipalities are also relatively more politically competitive (column 2) and a more educated adult population (column 3). In columns 4-6, we explore whether these results reflect the effects of political competitiveness prior to the dictatorship or the degree of political persistence within a municipality. Independently of political concentration, political persistence or entrenchment may create substantive challenges for economic development. In columns 4-6, we re-estimate our main specification including indicators for whether a family transitioned to and from the dictatorship. 17 Conditional on a municipality s level of competitiveness during the three regimes, we do not find any evidence that political persistence into the dictatorship affected long-run development. The association between long-run development and whether a family survived the transition to democracy is less clear. While the point estimates on income per capita and years of schooling are negative, and imprecisely estimated, the coefficient on political competition is positive and statistically significant, suggesting that places where a family transitioned out of the dictatorship are 4 percent more competitive. Agricultural Modernization and Political Concentration Thus far, we have documented an interesting pattern in the data: Places that were more politically concentrated prior to the dictatorship are faring better in the long run relative to places that were initially more politically competitive. This is despite the fact that political concentration as measured both during and post dictatorship is negatively correlated with income per capita, and other measures of long-run development. The question then becomes, why? As we argue in Section 2, one possible explanation for this positive association has to do with the military s decision to modernize the agricultural sector, with a particular focus on the country s main export crops at the time. Because of their technocratic approach to depoliticizing policy, the military implemented institutional reforms that affected the ability of the traditional sugar and cotton producers to compete both economically and politically. This in turn not only modernized these sectors but also engendered political competition by introducing new classes of political elites. Together these forces led these municipalities to develop relatively more in the long run. To test this theory, we begin by establishing the fact that political concentration prior to the dictatorship was higher in places with larger producers of cotton and sugar. Thus, it is presumably in these areas where the effects of the military s modernization policies on the local 17 The results are similar if we use the share of families that transition in and out of the dictatorship. 15

17 economic and political power relations would be the most pronounced. We document this relationship between political concentration and the existence of large-scale agricultural production by estimating the following regression: H m0 = α 0 + α 1 Sugar ij + α 2 Cotton ij + X ijθ + ν j + η ij (2) where H m0 is our measure of the Herfindahl prior to the dictatorship. The variables Sugar ij and Cotton ij, which measure the average amount of sugar and cotton production per farm in the municipality during 1940, proxy for municipalities that are likely to have large-scale sugar and cotton producers. 18 Given both the economic and political importance of sugar and cotton during this period, our hypothesis is that political concentration will be higher in places where the agricultural production of these crops is also more concentrated. The OLS estimates of Equation 2 are presented in Table 7. In column 1, we see that political concentration was much higher in municipalities that had larger-scale sugarcane and cotton producers. Both coefficients, which are significant at conventional levels, imply a 10 percent increase in political concentration resulting from a one-standard deviation increase in production per farm. In addition to these correlations, we also find a positive association between political concentration and whether the anti-military party (PTB) was in office prior to the dictatorship. Although not reported, political concentration is also positively correlated with agricultural employment and negatively correlated with rental prices in 1940, which as we saw in Table 5 are strong predictors of long-run income. Given that the importance of sugar and cotton for the economy differs across the states in our sample (see Table 1), in column 2 we re-estimate Equation 2 combining the production of these two crops into a single index, which we construct as an equally-weighted average of the z-scores of each measure within a state. Not surprisingly given the results in column 1, we find that a one-standard deviation increase in this agricultural concentration index is associated with a 10 percent increase in political concentration. If this measure is in fact proxying for places with large landowning elites then presumably it should also be correlated with other characteristics of the municipality that might indicate large-scale agricultural production. This is precisely what we find in column 3. Using this index as a dependent variable, we find that places with higher values of per-farm production also have higher numbers of employees per farm and have higher ratios of agricultural land devoted to large-scale versus small-scale production. In addition, although not reported, these municipalities have higher shares of employment in agriculture (coefficient = 3.06; se = 0.90), higher illiteracy rates (coefficient=3.06; standard error=0.80), a lower population (coefficient=-0.47; se=0.14), 18 Unfortunately, the 1940 census does not have information of the amount land devoted to these crops. While the obvious concern with these measures is that they are also proxying for farm productivity, as we show later, our results are robust to controlling for farm productivity in

18 and fewer shares of Protestants (coefficient=6.19; se=2.12). In column 4, we re-estimate the model presented in column 2, but also include these additional correlated of the concentration index as controls. In addition, we also add to the regression, a similarly constructed index for the other traditional cash and subsistence crops, namely: beans, corns, coffee, and cassava. We find that our coefficient on the index is robust to the inclusion of these additional variables, which suggests that our measure is not simply proxying for large-scale production or farm output in general but is capturing something specific to sugar and cotton production, which were main sources of economic and political power during this period. Another concern with this index is that it may also be identifying places that were more productive in sugar and cotton. We address this possibility, to some extent, in columns 5 and 6. Unfortunately, the 1940 agricultural census did not record for each municipality the amount of land used in the production of cotton and sugar. The 1960 agricultural census did record land usage by crop, but only if the crop represented a significant share of the state s agricultural production. As result, we only have a measure of sugar productivity in 1960 for the states of Ceará and Minas Gerais; whereas we only have a measure of cotton productivity in 1960 for the states of Ceará and Paraíba. Given these sample restriction, in columns 5 we re-estimation Equation 2 controlling productivity in sugar, and column 6 we re-estimate the model controlling productivity in cotton. In both regressions, our main results hold. Moreover, in column 6, we find that more politically concentrated places during the dictatorship were less productive in terms of cotton production. This is consistent with our broader hypothesis that it was the inefficient land oligarchs who were ultimately the most affected by the military agricultural modernization policies. Agricultural concentration and Long-run development If, as we are arguing, the military s policies of agricultural modernization disrupted the local power structure of the large landowners in sugar and cotton producing areas, then presumably these are also the areas that are better off today both politically and economically. In Table 8, we estimate the long-run relationship between our measures of economic and political development and our 1940 index of agricultural concentration in sugar and cotton. Column 1 reports a strong positive relationship between our agricultural concentration index and income per capita in The point estimate is and is statistically significant with a robust standard error of The magnitude of this coefficient is also economically meaningful: a standard deviation increase in the index led to a 2.5 percent increase in income per capita in In columns 2 and 3, we find that our results are robust to controlling for average productivity in either sugar or cotton. We also find similar patterns when using the other measures of long-run development. Places with larger producers of sugar and cotton in 1940 are much more likely to have competitive elections in 2000, as well as a more educated adult population. 17

19 4.1.1 Agricultural Concentration and Modernization The fact that these places are doing better in the long run is two-fold. First, the military introduced new policies that modernized the agriculture sector and it is likely that this would contribute to long run growth. Second, the military affected the local power relations within the municipality which in turn decreased political concentration. The increase in political competition would also contribute to long run development (Besley, Persson, and Sturm 2007). In this section, we provide evidence of both channels. In Table 9, we examine whether agricultural modernization occurred more in the municipalities with larger 1940 sugar and cotton producers. As Houtzager (1998) points out in his description of the military s agrarian project of The previous economic team had prioritized modernizing agriculture with the parameters of the Estatuto da Terra of But in practice, its attention had been focused on economic stabilization and fighting inflation. The new economic team essentially ignored the land statute. It made agricultural credit, abundant and heavily subsidized, the single most important instrument in the modernization of agriculture. In columns 1 and 2, we find evidence in support of this statement. The military provide substantially more credit, as measured both in terms of 1980 levels and in the change from 1980 to 1970, to places that had larger sugar and cotton producers in For instance, the coefficients in column 2 imply that a one standard deviation increase in agricultural concentration increased the amount government financing from 1970 to 1980 by 10 percent. To test the robustness of these results, column 3 and 4 presents the estimates of a placebo test that used as dependent variables the amount of financing from non-governmental sources, measured both in levels and in the change from 1970 to A similar estimated effect in this specification would suggest that the effects presented in columns 1 and 2 were potentially confounded by some unobserved demand-driven factor that was correlated with our index. But as reported in the columns 3 and 4, we do not find a significant association between private financing and our index of agricultural concentration. In columns 5-8, we explore other indicators of modernization. As Hagopian (1996) states Easy access to capital made possible the widespread introduction of the machinery, especially tractors, fertilizers, and pesticides into Brazilian agriculture. We find exactly this: In places with high 1940 sugar and cotton production, both the share of farms with a tractor (column 5) and the share of farms with electricity (column 6) increased during the dictatorship period. Given that the electrification of farms is unlikely to be restricted to just farms it is reassuring to see similar effects on the increase of the share of households with access to electricity as measured from the population census (see column 7). 19 Unfortunately the 1960 agricultural census did not contain information on the amount of credit farms received. 18

20 The modernization of the agricultural sector was also likely to have important implications for industry and the agricultural labor force. As Hagopian (1996) notes: These changes in traditional agriculture reduced substantially the agriculture labor force. Mechanization made many workers redundant, and the conversion of cultivated land to pasture expelled subsistence activities and lowered the overall demand for rural labor In column 8, we see some evidence of this as well. As these places modernized, they also witness a shift in employment away from agriculture and towards the industrial sector Agricultural Concentration and Changes in Political Concentration The military s policies toward agriculture did more than simply modernize the sector. It tried to rupture the ties between the traditional oligarchs and the state. For instance, as?) explains Policy toward agriculture was inevitably intertwined with national integration. It required the national state to enter rural areas where it historically had neither authority nor bureaucratic capacity to act. The agrarian project therefore represented a direct challenge to local authority-it entailed gaining a degree of control over labor, land, and capital in the countryside. At the time of the coup, the regulation of land and rural labor relations, agricultural policy, and research and extension services had been almost entirely in the hands of the regional oligarchies who ran statelevel governments. In short, the project required a period of state building not seen since the Estado Novo ( ). The military attempted just that. It sought to centralize authority out of the hands of the oligarchies by enacting new legislation and created new bureaucratic machinery in rural areas to circumvent existing state and local governments. Consistent with this argument, we should expect to see political concentration decline in areas dominated by these agricultural elites. The results in Table 10 lend further support of this hypothesis. In column 1, we that places that were engaged in large-scale sugar and cotton production in the 1940s become much less politically concentrated during the dictatorship. Interestingly, the decline in political concentration in these areas did not imply less political persistence. In column 2, we do not find any evidence that the likelihood of an elite transitioning to the dictatorship decreased in these sugar and cotton producing areas. This finding is consistent with the work of?), who documents the resilience of the local elites in their ability to maintain power during the dictatorship in Minas Gerais. In columns 3 and 4, we examine the extent to which these sugar and cotton producing areas voted for the military party (ARENA). Although these elections were in general neither entirely 19

21 fair nor contested - as the military party won over 85 percent of the elections and in many places claimed 100 percent of the votes - the opposition party did perform better in municipalities where there was more agricultural concentration. Moreover these places were much more likely to have at least one mayor be elected from the opposition party during the dictatorship (see column 4). Overall, these results suggest that the military s agrarian project did lead to modernization in these areas and induced more political competition. These two mechanisms are presumably the channels by which these places are doing better today in terms of their economic and political development. 5 Conclusions To be written... References Acemoglu, Daron, and James A. Robinson A Theory of Political Transitions. American Economic Review 91 (4): (September) Persistence of Power, Elites, and Institutions. American Economic Review 98 (1): (March). Banerjee, Ahbijit, and Lakshmi Iyer History Institutions and Economic Performance: The Legacy of Colonial Land Tenure Systems in India. American Economic Review 95 (5): Bardhan, Pranab Scarcity, Conflicts and Cooperation: Essays in Political and Institutional Economics of Development. Cambridge, MA: MIT Press. Besley, Tim, Torsten Persson, and Daniel Sturm Political Competition, Policy and Growth: Theory and Evidence from the United States. NBER Working Paper No Cohen, Yossef democracy from Above: The Political Origins of Military Dictatorship in Brazil. World Politics 40 (1): (September). Dal Bó, Ernesto, Pedro Dal Bó, and Jason Snyder Political Dynasties. Review of Economic Studies, vol. 76. de Castro Gomes et al, Angela Maria O Brasil Republicano, Vol. 10, Sociedade e Politica ( ). Rio de Janeiro: Bertrand Brasil. de Oliveiro Pierucci et al, Antonio Flavio O Brasil Republicano, Vol. 11, Economic e Cultura ( ). Rio de Janeiro: Bertrand Brasil. 20

22 de Paiva Abreu, Marcelo The Brazilian Economy, Edited by Leslie Bethell, The Cambridge History of Latin America Volume 9, Brazil since Cambridge, MA: Cambridge University Press. Engerman, Stanley L., and Kenneth L. Sokoloff Factor Endowments, Inequality, and Paths of Development Among New World Economics. NBER working papers 9259, National Bureau of Economic Research. et al, Fernando Henrique Cardoso O Brasil Republicano, Vol. 8, Estrutura de Poder e Economia ( ). Rio de Janeiro: Bertrand Brasil. Fleischer, David A Evolução do Sistema Bipartidário. Os Partidos Políticos no Brasil 1: Goodman, D.E., B. Sorj, and J. Wilkinson Agroindstria, Polticas Pblicas e Estruturais Sociais Rurais: Anlises recentes sobre a Agricultura Brasileira. Revista de Economia Poltica 5, no. 4. Graham, Douglas H, Howard Gauthier, and Jose Roberto Mendonca de Barros Thirty Years of Agricultural Growth in Brazil: Crop Performance, Regional Profile, and Recent Policy Review. Economic Development and Cultural Change 36 (1): Hagopian, Frances Traditional Politics and Regime Change in Brazil. New York and Cambridge: Cambridge University Press. Houtzager, Peter P State and Unions in the Tranformation of the Brazilian Countryside Latin American Research Review 33, no The Institutional Roots of Popular Mobilization: State Transformation and Rural Politics in Brazil and Chile, Comparative Studies in Society and History 42, no. 2. Nunberg, Barbara Structural Chnage and State Policy: The Politics of Sugar in Brazil Since Latin American Research Review 21 (2): Pande, Rohini, and Christopher Udry Institutions and Development: A View from Below. Edited by Whitney K. Newey Richard Blundell and Torsten Persson, Advances in Economics and Econometrics. Cambridge, MA: Cambridge University Press. Power, Timothy J Parties, Puppets and Paradoxes: Changing Attitudes toward Party Institutionalization in Post-Authoritarian Brazil. Party Politics 3 (2): Roett, Riordan Brazil: Politics in a Patrimonial Society. Westport, CT: Praeger Publishers. Samuels, David, and Fernando Luiz Abrucio Federalism and Democratic Transitions: The New Politics of the Governors in Brazil. Publius, Spring,

23 Sarles, Margaret J Maintaining Political Control Through Parties: The Brazilian Strategy. Comparative Politics 15 (1): (October). Schmitter, Philippe C The Portugalization of Brazi? Edited by Alfred Stepan, Authoritarian Brazil: Origins, Policies, and Future. New Haven, CT: Yale University Press. Skidmore, Thomas E Politics in Brazil : An Experiment in Democracy. Oxford: Oxford University Press The Politics of Military Rule in Brazil, Oxford: Oxford Univeristy Press. Soares, Glucio Ary Dillon Desigualdades Eleitorais no Brasil. Revista de Cincia Poltica, vol. 2. Stepan, Alfred Authoritarian Brazil: Origins, Policies, and Future. New Haven, CT: Yale Univeristy Press. 22

24 FIGURE 1: DISTRIBUTION OF THE SHARE OF TERMS A FAMILY HAS BEEN IN POWER Notes: The sample consists of all 9,058 families that held power in a municipality during