To the ones in need or the ones you need?

preliminary version: please do not cite or circulate without the authors permission To the ones in need or the ones you need? The Political Economy of Central Discretionary Grants: Empirical Evidence from Indonesia Gerrit J. Gonschorek, Günther G. Schulze and Bambang Suharnoko Sjahrir Institute of Economics Department of International Economic Policy University of Freiburg, Germany 30. June 2017 Abstract: We analyze the allocation of discretionary grants from the central government to local governments in Indonesia. Using OLS and Fixed Effects models on an unbalanced panel data set for more than 400 Indonesian districts covering the period 2004-2013, we investigate whether the allocation of the grants are determined by the need of a district, by political alignment of the central government with the local district heads, or by reelection motives of the incumbent president. We find that grants are hardly determined by need characteristics and that political considerations matter significantly. Districts with low support for the president received significantly more than the core supporting districts. This effect is limited to the first term of the president. In the second term, after which reelection is impossible, political considerations were largely absent. This pattern is consistent with the view that the incumbent president considers discretionary grants as an instrument to increase reelection probabilities. Unlike the evidence for most countries, we find no effect for political party alignment with local district heads. Our results are robust to the inclusion of a number of other variables capturing competing motives. Acknowledgments: We would like to thank Krisztina Kis-Katos, Jürgen Rüland, Christian von Lübke, Volker Lindenthal, Nikita Zakharov and Antonio Farfán Vallespin for valuable comments. We thank Marcus Mietzner, Kevin Evans, Titi Anggraini (PERLUDEM), as well as the people at SMERU, KPPOD in Jakarta for their support and valuable expertise as well as many more friends and colleagues in Indonesia. We thank all participants of the Doctoral Workshop on Development Economics of the German Economic Association (University of Zürich, Switzerland, 2016), at the Annual Meeting of the European Public Choice Society (EPCS) (University of Groening, Netherlands, 2015) plus researchers of Gadjah Madah University (UGM), University of Indonesia (UI), Australian National University (ANU), KITLV at Leiden University, the Arnold Bergstraesser Institute, the Freiburg Institute of Advanced Studies (FRIAS) and the Southeast Asian Studies Group at Freiburg University for their helpful comments. We would also like to thank the German Federal Ministry of Education and Research (BMBF) and the Wissenschaftliche Gesellschaft Freiburg for partially funding this project and Fabrizio Kuruc for his excellent research assistance. All errors are our own. 1

1. Introduction This paper analyzes the allocation determinants of a particular grant program, 'Tugas pembantuan (TP), from the central government to the regions in Indonesia. The program is designed to fund physical infrastructure, including agricultural infrastructure, and is allocated to the regions at the discretion of the central government. Every funded project needs the presidential approval. This raises the question whether the president approves allocations according to a specific rationale. Allocations could be made on a needs basis, given that Indonesia's 510 (in 2014) districts vary tremendously in prosperity and economic activity and that many districts are lagging behind. Poverty rates span from 1.75 percent to 47.52 percent with a mean of 13.59 and a standard deviation of 8.47. 1 A second prime candidate for such a rationale is to support politically aligned district heads in order to secure their reelection and to broaden the party basis in the regions and thus indirectly also at the national level. There is substantial evidence for developed countries that politically aligned regions receive preferential treatment from the central government (inter alia Berry et al. (2010) and Larcinese et al. (2006) for the US, Bracco et al. (2015) for Italy, Kauder et al. (2016) for Germany, Solé-Ollé and Sorribas-Navarro (2008) for Spain, Veiga and Pinho (2007) and Veiga and Veiga (2013) for Portugal). There is less evidence for developing and threshold countries (Arulampalam et al. (2009) for India, Brollo and Nannicini (2012) for Brazil and Francken et al. (2012) for Madagaskar). Lastly, the transfers could be distributed with the objective of enhancing the reelection probabilities of the incumbent president directly. There is ample empirical evidence that incumbents spend public funds in order to enhance reelection probabilities, in particular before election years, which gives rise to political business cycles (PBC, Nordhaus, 1975). These cycles depend on the characteristics of the political system, notably they are strong in young democracies while they tend to be less pronounced or even disappear in mature democracies (Brender and Drazen (2005), DeHaan and Klomp (2013), Shi and Svensson (2006)). Pork barrel politics may be less pronounced also in the presence of term limits (Aidt and Shvets, 2012). While PBCs have been intensely researched, 2 it has not been analyzed whether intergovernmental fiscal transfers could serve the purpose of supporting the reelection bid of 1 Data refer to 2013, source World Bank (DAPOER), http://databank.worldbank.org/data/reports.aspx?source=indo~dapoer-(indonesiadatabase-for-policy-and-economic-research) 2 Cf. Sjahrir et al. (2013) for an analysis of PBCs in Indonesia at the local level. 2

the central government s CEO. We provide such an analysis for Indonesia. The transfer system Tugas pembantuan is ideally suited for this purpose as it is allocated at the discretion of the central government and, being earmarked for capital goods, has a great visibility by the public. Our data allow to test different hypotheses on a possible rationale behind the allocation of the discretionary intergovernmental fiscal transfers. If transfers were allocated based on neediness, needs parameters such as poverty rate, HDI, or the status of the infrastructure in a district should be significantly correlated with the transfers received. If political alignment played a significant role, the party affiliation of the district head (or the composition of the local parliament) should turn out significant. If the reelection motivation of the president guided the allocation, the transfers should go to the districts where they would mobilize most additional votes; this would be the districts that did not strongly vote for the incumbent in the first elections, as voters might be persuaded to support the president by improved infrastructure services. Yet, this should be limited to the first term as the president can be reelected only once. Our panel data set of Indonesian districts covers the period 2004-2013, which coincides with the two terms of former president Susilo Bambang Yudhoyono. Indonesia is not only an important case study as it is the fourth most populous nation and the third largest democracy in the world and the dominating country in Southeast Asia. Its particular features make the analysis of Indonesia interesting beyond the country as such and sets it apart from the cases analyzed so far. Indonesia is a strongly decentralized, very diverse presidential democracy with little party loyalty and strong clientelistic elements. It has turned into a democracy only in 1998, after long-time dictator Suharto stepped down; in 2001, it decentralized in a big bang, delegating major government functions such as health, primary and secondary education, infrastructure, and environmental protection to the district level, while provinces received mostly backstopping and supervisory functions. As a consequence, almost 40 percent of the budgetary responsibilities were devolved to the regions (World Bank, 2008). Whereas the expenditure authority has been devolved to a large extent, regions still heavily rely on fiscal transfers from the center (Lewis 2014). The fiscal and administrative decentralization was complemented by a political decentralization with free elections of local parliaments in 2001 and asynchronous elections of the district heads first by local parliaments and, beginning in 2005, directly by local electorates. Indonesia is a presidential democracy with a multiparty system. Since 2004, the president has been elected directly by the voters; the lower house (Dewan Perwakilan Rakyat, DPR) is 3

elected by proportional representation and is quite fragmented, which requires coalition building on part of the president. Ideological differences between parties are relatively small and party loyalty of voters, but also of candidates for political offices, is comparatively weak (Mietzner, 2013). The political system is often characterized as highly clientelistic with voters expecting tangible material benefits in exchange for their votes (Mietzner (2011), Aspinall and van Klinken (2011)). While the lacking ideological divisions and the clientelistic nature of politics is not uncommon in many developing countries, studies on the political-economic determinants of transfers to regions have focused on advanced countries with mature political systems. The two democratic developing countries that have been studied Brazil and India are characterized by strong ideological divides ((Arulampalam et al. (2009) and Brollo and Nannicini (2012)). Hence Indonesia is an ideal case for analyzing politically motivated transfers in a developing country s context also because of its large number of districts, its great diversity in economic development, its political system, political constellations at the local level, and high data quality. Moreover elections of district heads are asynchronous and exogenously timed which allows to clearly identify a party alignment effect. Our paper is the first to study central government transfers to the regions in Indonesia from a political-economic perspective. Previous analyses have either exclusively focused on the formula-based transfers that disallow decisions at the discretion of the central government (Brodjonegoro and Martinez-Vazquez (2004), McLeod and Fadliya, 2010)) or tested for a possible needs orientation of discretionary transfers and their effects on service delivery but disregarded potential political-economic motivations (Lewis, 2016). We use a pooled OLS and fixed effects models on a unique unbalanced panel data set for more than 400 Indonesian districts from 2005-2013, representing around 90 percent of Indonesia s population in 2013. Our results show that central discretionary grants are hardly influenced by needs characteristics and that political considerations matter significantly. Unlike the evidence for most other countries studied, we find no effect for political party alignment between the central and the district government on grant allocation which is explained by the political system. In contrast the allocations are determined by political support considerations of the president facing reelection: Districts with low support for the president received significantly more per capita grants than the core supporting districts as they offer the biggest opportunities for increasing the vote share for the incumbent. This effect is limited to the president s first term in office because after the first term reelection is not possible, and 4

therefore political considerations should be absent (Aidt and Shvets, 2012). Indeed there is no evidence for politically motivated grant allocations in the second term. This pattern is consistent with the view that discretionary grants are used as an instrument to increase the reelection probabilities of the president (Cox and McCubbins, (1986), Lindbeck and Weibull (1987), Dixit and Londregan (1996)). Our results are robust to the inclusion of potential local political budget cycles, the political concentration at the local parliaments, regional favoritism by the incumbent president, the existing service infrastructure, the importance of the agricultural sector and the cost of providing services at the district level. The paper proceeds as follows: Section 2 provides a brief background on Indonesia s political system, especially on the presidential elections and the fiscal structure relevant for our analysis. Section 3 discusses the theoretical argument for the determinants of centralgovernment transfers in a decentralized system, relates them to the Indonesian context, and presents four testable hypotheses. In Section 4, we describe the data, our empirical approach, and the main results. Section 5 presents additional robustness checks and discusses the limitations of our approach. Section 6 concludes. 2. Institutional Background 2.1 Elections Indonesia is a presidential democracy. The president has veto power over all legislation; by constitution, laws must be made by joint agreement with the House of Representatives (DPR) (Sherlock,2010). The president appoints his cabinet members and is directly elected for a maximum of two 5-year terms. Since 2004 the president has been elected directly by popular vote. 3 Presidential elections are held in two rounds if no ticket receives the absolute majority in the first round. Since a political party must have a minimum of 20 percent of the votes or 15 percent of the seats in the House of Representatives (DPR) to nominate a presidential candidate (Ufen, 2009) presidential elections take place after the legislative elections. 4 The DPR consists of 560 members and is elected by proportional representation at the district 3 Constitutional Amendment in 2002. Before 2004, the Indonesian President was appointed by Indonesia s House of Representatives (DPR). Since he/she is directly elected, the power of the DPR has decreased significantly (Schmidt (2010). According to Schmidt (2010), the appointed members of the DPR elected Suharto in engineered elections to be president seven consecutive times (1966-1998). 4 In 2004, parties that received at least 5 percent of the total vote or 3 percent of the parliamentary seats at the DPR were allowed to nominate a presidential or vice-presidential candidate (Law 23/2003 Article 101). For 2009, this minimum was changed to 15 percent of the seats or 20 percent of the votes. 5

level. Since 2008, seats are allocated through an open party list system, all candidates must run for a certain political party, but voters can vote for a party or an individual candidate (Sherlock, 2010). The second chamber of parliament is the Regional Representative Council (Dewan Perwakilan Daerah, DPD). It consist of 132 members, four from each of Indonesia`s provinces and is elected at the provincial level. It has only a participating and advisory function to the DPR. The MPR (Majelis Permusyawaratan Rakyat) consists of the DPR and DPD members and oversees presidential decisions. Elections for the national, provincial, and district parliaments are held simultaneously every five years (2004 and 2009 in our observation period). The local parliaments at the district level, the DPRDs (Dewan Perwaklian Rakyat Daerah), are also elected by proportional representation with an open party list (Ananta et al., 2005). Provincial governors, mayors (walikotas), and regents (bupatis) have been elected directly by popular vote since 2005 (Law No.32/2004). The timing of the district heads elections were determined exogenously and independently from the timing of the national executive and legislative elections. They took place asynchronously after the 5-year term of a local official had ended, which is a legacy from the Suharto era as district heads were appointed at different points in time and were allowed to serve out their term after Suharto's demise and the democratization that followed. The asynchronous nature of district head elections allow us to clearly identify any effect of political party alignment or political business cycle on transfer decisions. 5 Only political parties or coalitions with at least 15 percent of seats in the local DPRD are allowed to nominate candidates; the candidates need at least 25 percent of the votes to win the election in the first round, otherwise they face a run-off election against the candidate with the second largest vote share in the first round (Choi and Nankyung, 2009). 6 Local government heads, in particular district mayors, are the ones defining local policy making in contrast to the rather weakly influential local parliaments (Buehler, 2016). In the 2004 presidential elections, Susilo Bambang Yudhoyono and Yusuf Kalla, nominated by Partai Demokrat (PD), were elected as President and Vice-president of Indonesia (60.6 percent). They defeated Megawati Sukarnoputri (the incumbent president) and her running- 5 Only since December 2015, district head elections have been held simultaneously across all district experiencing a term limit of their local government head in that year. 6 Since 2008, independent candidates are allowed to run for office at local executive elections (Law 12/2008). However, due to the high costs of an electoral campaign and costly requirements to withdraw a nomination (Law 12/2008, Article 62), only a few independent candidates have contested in local direct elections (Buehler, 2010). 6

mate Hasyim Muzadi (39.4 percent), nominated by PDI-P (Ananta et al., 2005). These were the first direct elections in Indonesia s history. 7 Yudhoyono and Kalla, both former ministers in President Megawati s cabinet were supported by the People s Coalition (Koalisi Kerakyatan) formed by PD, PBB, PKIP, and PKS. Megawati Soekarnoputri, daughter of Indonesia s first President Soekarno (in office 1945-1967), and Hasyim Muzadi, a leader of the Al-Hikam Islamic Boarding School, were supported by the National Coalition (Koalisi Kebangsaan), consisting of Golkar, PDI-P, and PPP (Ananta et al., 2005). 8 In the 2009 presidential elections, Yudhoyono was reelected for his second (and final) term as president with the Indonesian central bank's governor Boediono as his running mate. Supported by his own party PD as well as by PKS, PAN, PPP, and PKB (Sukma, 2009), he won in a landslide victory (60.8 percent) in the first round. 9 Megawati Soekarnoputri (PDI-P) and her running mate Prabowo (26.8 percent) as well as Yudhoyono s first-term vice president Jusuf Kalla (head of the Golkar Party) and his running mate Wiranto (Partai Hanura) (12.4 percent) were defeated. Figures B1-B5 in Appendix B show the regional distribution of votes for Yudhoyono in the 2004 and 2009 presidential elections across Indonesian districts, showing stark regional differences in vote support for Yudhoyono in 2004 and 2009 (see Appendix B). 10 2.2 Fiscal Transfers Indonesia has three levels of government, the central, the provincial level and the district level. Indonesia s central government is responsible for judiciary, law enforcement, monetary and macroeconomic policies, religious affairs, defense, foreign relations and security policy. Sub-national governments are responsible for all remaining functions, especially for decentralized service sectors like education (excluding tertiary education), health, and infrastructure. Provinces have limited responsibilities compared to district governments, they are responsible for supervision and cross-district cooperation. Districts provide core services; they receive the lion s share of their overall revenue from transfers from the center. In 2010 7 In the DPR elections, Golkar had received the highest share of votes (21.6 percent), followed by PDI-P (18.5 percent), PKB (10.57 percent), PPP (8.15 percent), and the presidential party PD (7.45 percent).for more information on each of the political parties in Indonesia see Mietzner (2013), Ananta et al. (2005), Aspinall and Mietzner (2010) or Bünte and Ufen (2009). 8 The fact, that Yusuf Kalla a former Golkar official was not supported by Golkar in the 2004 presidential elections, illustrates how blurry party lines in Indonesia sometimes are. 9 In the 2009 national DPR elections a few month before, the presidential party PD won in a landslide receiving 20.9 percent of the votes. Followed by Golkar (14.5 percent), PDI-P (14.0 percent), PKS (7.9 percent), PAN (6 percent), PPP (5.3 percent), PKB (4.9 percent), Gerindra (4.5 percent) and Hanura 3.8 percent (Sukma, 2009). 10 For more information on the 2009 presidential elections see Sukma, (2009); Tomsa, (2010); Fealy, (2011) and Aspinall et al., (2015). 7

central government transfers accounted for 54 percent of provincial and 93 percent of district expenditures (Shah, 2012). Indonesia s major government transfers (DAU, DBH, DAK) are non-discretionary; they are either determined by tax revenue generated at the sub-national level, specific criteria, or by a formula. 11 The general allocation grant DAU (Dana Alokasi Umum) is a non-earmarked, formula-based general purpose grant and the most important sources of sub-national government revenue in Indonesia. DAU transfers accounted for 52 percent of total subnational government revenue between 2001 and 2009 (Agustina et al. (2012)). The DAU formula takes into account fiscal capacity and fiscal needs of a district (for details see Agustina et al. (2012), McLeod and Fadliya (2010)). The specific allocation grant DAK (Dana Alokasi Khusus) is earmarked for national priorities. Its allocation is determined by general criteria such as the financial capacity of a subnational government, technical criteria such as guidelines established by the responsible line ministry) and special criteria (e.g. specific characteristics of a region). The DBH (Dana Bagi Hasil) is Indonesia s tax and natural resource revenue-sharing system and is allocated according to a formula (Agustina et al. (2012)). 12 It is motivated by political-economic considerations, to give national resource rich regions a larger share of resource rents (Harjowiryono, 2011). Compared to these non-discretionary transfers, TP (Tugas pempantuan, Co-Administration and Assistance Tasks Funding) is a discretionary central-government grant under authority of the president. 13 Since 2001, line ministries of the central government which are not responsible for the five core responsibilities foreign affairs, defense, justice, fiscal/monetary policy and religion have to delegate implementation of their tasks to sub-national governments, which act as a representative of the central government. 14 TP grants are not part of the local budget, but they are spent locally and are co-administered with a provincial governor and/or district head. TP grants should be allocated to tasks of `physical nature` (Government Regulation 7/2008). Even though the allocations should be in accordance to some general principles such as 11 Indonesia s Law No. 33/2004 on fiscal decentralization, Law No. 32/2004 on sub-national governance, and Law No. 25/1999 on fiscal balance between central government and regions. Besides these transfer mechanisms, there are: Special Autonomy Funds (for Papua and Aceh) based on Law No. 35/2008, Law No. 11/2006, Law No. 21/2001, Adjustment Funds for financial ad hoc assistance, Hibah-transfers for assistance in the infrastructure sector, a special Incentive grant (DID) as well as the some smaller performance based grant schemes. 12 Since the beginning of 2014 the property tax (Territory and Building Tax, PBB and Property Title Transfer Fees, BPHTB) has been devolved to the subnational level. This excludes mining, plantation and the forestry sector. 13 Government Regulation 7/2008 Article 6 paragraph 2. 14 Government Regulation No. 52/2001, Government Regulation No. 7./2008 and Government Regulation No.106/2000. 8

`harmonious national and regional development` (Government Regulation 7/2008) there are no clear allocation criteria in place leaving the allocation at the discretion of the central government. The line ministries planning to establish a TP funded project have to consult with the National Development Planning Agency (BAPPENAS), which is responsible for the coordination and planning of projects. 15 However, every project needs the approval of the president. 16 After presidential approval the Ministry of Finance disburses the funds, and the Ministry of Home Affairs has a supervisory function. Figure 1 shows the allocation of TP per capita by district for the year 2007. While districts on the economically most prosperous Island Java receive lower grants per capita, there are distinct differences in the allocation to neighboring and economically similar districts. This suggests that political factors may play an important role in explaining the allocation of central discretionary government grants across Indonesia. The fact that TP grants should be allocated to tasks of `physical nature` (Government Regulation 7/2008) would ensure good visibility of the results of TP transfers to the voters. 15 The National Development Planning Agency meets with local governments once a year in order to discuss priorities of local development for the following year. There is no other official channel for a proposal of grants by the local governments. 16 Government Regulation 7/2008 Article 6 paragraph 2 9

Figure 2: Geographical distribution of Tp per capita in 2007 source: authors illustration 10

3. Theoretical Considerations From an efficiency perspective central government transfers to the region are regarded as an instrument that allows benefitting from the advantages of decentralization while minimizing the costs of decentralization in terms of fiscal inequity or negative external effects (Boadway, 2007). Decentralization is advantageous for governance and public service delivery because it allows for better preference matching of geographically heterogeneous population as local governments have an informational advantages over the central government, their actions are more transparent to the local electorate and thus participation of local constituency is higher (Oates, 1972). Moreover, in decentralized states mobile individuals can sort themselves into local jurisdictions that offer the best mix of public services and taxes (Tiebout, 1956). Decentralization can also be seen as creating a laboratory for policy solutions (Hayek, 1945) and giving rise to competition between districts, which increases their performances (Besely and Case, 1995). Yet this competition may create negative externalities as local governments do not take into considerations the effects of their policies on other local jurisdictions, for instance the erosion of local tax bases (Wilson, 1999) or may bias public expenditure composition (Keen and Marchand, 1997). Oates (1999) argues that central government transfers have the normative objectives of fiscal equalization, securing national public service standards and the internalization of spillover benefits of local public services (see also Boadway, 2007). As a consequence, central government transfers should account for differences in fiscal capacity, in the costs of providing public services of a jurisdiction as well as for different development levels. Indeed Government Regulation 7/2008 on Deconcentration and Co-Administration Funds stipulates that the allocation of TP should be determined by the district s financial capacity, the need of providing public services, and the district s development. We thus formulate our first hypothesis: Hypothesis 1 ("Needs orientation"): Central discretionary grants are lower for districts with higher fiscal capacity and better socioeconomic development. This hypothesis is testable as data for development levels, such as human development index, GDP per capita, price indexes and fiscal variables are available at the local level (and differ 11

very substantially between districts). In the Indonesian context only Lewis (2016) has tested for a possible needs orientation of TP transfers; he finds very mixed results. In contrast to this normative point of view, government transfers from a positive point of view are motivated by maximizing the probability of reelection. The distribution of transfers is based on tactical and strategic considerations in order to maximize votes, seen as political capital (Grossman, 1994). The incumbent may target swing voters (Lindbeck and Weibull, 1987), core supporters (Cox and McCubbins, 1986) or either swing or core supporters depending on the relative effectiveness of this political investment (Dixit and Londregan, 1996). 17 The allocation depends also on an incumbent s attitude towards risk (Cox and McCubbins, 1986), whether there are systematic (observable) party preferences in the electorate (Lindbeck and Weibull, 1987), and on the ability to target transfers to any certain group of the electorate (Dixit and Londregan, 1996). Additionally, Arulampalam et al. (2009) assume voters to vote along party lines whereas Dixit and Londregan (1998) distinguish voting behavior between different government levels. For Arulampalam et al. (2009) transfers are used to finance public projects at the state level, increasing the popularity (electoral benefit) of the central incumbent party. Through these projects, the incumbent of the state receives some electoral benefit as well. If both the central and state incumbent are from the same party, a party receives the total additional electoral benefit from such additional expenditure at the state level. If, however, they are from different parties, some of the electoral benefit at the state level might leak to the opposition. Here the incumbent focuses on swing states if the transfers provide information on their source, known as targeted grants. Otherwise the electorate, by mistake, accounts any electoral benefit to the opposition in charge at the local level. Indonesia s district government level however is characterized by local district heads that often only have a loose (institutional) attachment to a political party. Local candidates often see political parties just as a provider for nomination (Qodari, 2010). They are often no official party members or switch their party background to a party with a stronger local network and better financial opportunities (Ufen, 2009). Political parties in turn tend to sell nominations to 17 According to Drazen (2000) a swing voter can be defined as a voter with weak ideological party preferences or a voter likely to change its voting behavior due to received economic benefits, since he is close to the margin of indifference between the political parties policies. A core voter in contrast with strong ideological political party preferences that was loyal to a party in past elections. 12

promising, well-endowed independent local candidates that do not have any backing of a political party yet (Buehler (2010), Mietzner (2009)). As Mietzner (2013, p: 224) puts it While formally nominated by parties, these local officials [ ] have no or only weak institutional party connections. 18 Thus one could think that a president might support a local district head as long as he/she runs under his political party banner. Due to the rather low political party attachment however, unlike most empirical evidence on other countries, we do not expect to find a significant influence of political alignment between the president and a local district head for Indonesia. Hypothesis 2 ( Political alignment ): Central discretionary grants are higher for districts that are politically aligned with the president. Indonesia also does not have a two-party system and strict politico-ideological lines. Instead it is characterized by a multiparty system with changing rainbow coalitions at the central government level (Aspinall et al., 2015), decreasing politico-ideological party lines (Mietzner, 2013), political party platforms hardly distinguishable (Mujani and Liddle, 2010) and an increasing importance of candidate-popularity (Mietzner, 2013). According to Ufen (2009) the two ticket, two round system of the direct presidential elections implemented in 2004 (Section 2.1) has decreased ideological party differences. The candidates often just represent different levels of religiosity or a geographic area. According to Ananta et al. (2005) e.g. the mixture of Javanese/non-Javanese, secular/islamic, as well as Kalla s civilian and SBY s military background made them an promising pair for the presidency from the start. Here personal popularity of a candidate often matters more for electoral success than political party strengthen or a candidates political ideology (Mujani and Liddle (2010), Qodari, (2010), Mietzner (2009)). 19 In the past parties aligned based on ideology, today parties start to align themselves more based on their electoral interests (Basewedan, 2007). As his predecessor SBY also created 18 At the central level, the political party Partai Demkorat (PD) is also seen solely as a presidential party, used as an electoral vehicle to win presidential office by the Yudhoyono in 2004 (Aspinall et al., 2015). Lacking a real political platform and an organizational structure at the local level (Ufen, 2009). A part from the at least branches in 60 percent of all provinces every local political party in Indonesian needs by law (Palmer,2010) 19 However, at the same time the backing of Yusuf Kalla and its Golkar party in the run up to 2004 presidential elections helped SBY to gain office the first time, since PD had just received 7.5 percent in the DPR elections (Ufen, 2009). 13

large rainbow coalitions, political party coalitions across ideologies and in changing combinations, with the only aim to stay or get in power. 20 SBY s first cabinet (2004-2009) and second cabinet (2009-2014) included among technocrats, members of seven and six parties respectively (Aspinall et al., 2015). Accounting for 73 (2004-2009) and 75 percent (2009-2014) of the seats in parliament. Due these pragmatic rainbow coalitions across all politicoideological lines, often no clear distinction between the opposition and the government was possible (Ufen, 2009). In contrast to most other emerging democracies there is also no classical left-right divide between Indonesian parties at the regional level, ideological lines rather run along the role of Islam in state affairs (Mietzner, 2013). 21 Hence Mietzner (2013) argues that politicoideological party differences have decreased, leading to some dealignment of electoral politics as Indonesia moves from a party-based to a candidate-driven system. 22 However this would not lead to the full abolishment of established parties and their embedded roots in specific constituencies (Mietzner, 2013). Here Mietzner (2013) argues that although party identification of the electorate has decreased and party networks have been weakened, most parties still have deep roots in specific constituencies and most party leaders still represent long-standing politico-ideological camps. 23 Most Indonesian parties still seem to have core constituencies providing them with electoral support and networks and political supporters embedded in particular milieus aliran (streams) (Ufen, 2009). Although Mujani and Liddle (2010) find no evidence for a polarization of voters based on aliran and religiosity in the legislative or presidential elections of 2009. Pratikno (2009) argues, that although the importance of ideological lines for the candidacy in local government elections has declined (see above), members of each political party still use their ideological lines and networks to mobilize votes. Meaning parties have a well-defined support bases along particular religious, regional or socioeconomic characteristics (Sherlock, 2004). However, although political parties have well-defined supporters with respect to religion, region or socio-economic background there often is no focus on well-defined policies to cater the specific needs of those 20 Sherlock argued in 2004 already, that due to Indonesia s strong fragmentation of the electorate, presidential candidates must have to build a coalition to win the first past the post system leading to backroom deals over other important positions e.g. in the cabinet (Sherlock, 2004). 21 Mietzner (2013, p. 219): To be sure, Indonesia does not fit into the left-right scheme that political theorists often view as a precondition for solidly structured parties and party systems. Instead the main cleavage in Indonesia s party landscape separates advocates from opponents of politico-social Islamization. This cleavage has instilled an ideological sense of purpose into party politics, and sets it apart from most other new democracies. 22 Mietzner (2013, p. 220): assessments diagnosing the disappearances of aliran-or politico-ideological streams-have been premature. There is no doubt that some form of dealignment in electoral politics has taken place, as the rise of presidential parties has demonstrated (Mujani and Liddle (2010)). But this has not led to the exodus of established parties with deep roots in specific societal constituencies. 23 Especially if you compare Indonesia to other East Asian countries like Thailand and the Philippines (Mietzner, 2013). 14

communities (Sherlock, 2004). In the 2004 elections the political differences between the president and the challengers in terms of their political agenda was blurry and most voters were not aware of any party differences (Mujani and Liddle, 2010). A survey on Indonesia s electorate, done by the Asia Foundation, showed that 66 percent of the eligible voters in 2003 saw no differences between the contesting parties, 9 percent said there were, but were not able to name any differences. In 2009 however, Yudhoyono and its PD were clearly challenged by Megawati and the PDI-P (Mujani and Liddle, 2010). In contrast however, political patronage and clientelism in Indonesia is very prevalent (Mietzner (2013), Aspinall and Sukmajati (2016)), votes are often given to a candidate or party as a reward for services ( service-based candidate-voter relationship) (Palmer, 2010), the introduction of direct elections in 2004 have increased the need to maximize votes (Ufen (2009), Qodari (2010)) and voters show an increasing importance of their perception of the economic situation. Referring to a study by the Indonesian Survey Institute (Lembaga Survey Indonesia, LSI) Mujani and Liddle (2010) argue that leadership, candidate appeal (especially in the media), self-identification with a political party, and demographics determined voting behavior in Indonesia s 2004 election rather than the perception of the economy, religion, region ethnicity, social class, or location of residency (Java/Non-Java). In contrast political economic factors, rather than leadership appeal and campaign advertising, were identified as one factor explaining the 2009 parliamentary vote in Indonesia (Mujani and Liddle, 2010). Indonesians with a positive view of the national economy or government performance were more likely to vote for PD in the parliamentary elections compared to for example PDI-P, enabling Yudhoyono to get re-elected in 2009 (Mujani and Liddle, 2010). Hence in 2009 political economy factors played an increasing role in determining electoral preferences. All this makes Indonesia a particular interesting case for testing the classical determinants of central discretionary grants prevalent in the literature. Biasing grants towards districts showing a larger vote support for the president could secure past electoral support for the future. Alternatively, the president could also have an interest to bias his grants to weaker supporters giving him larger potential returns of his investment in votes. Here this bias towards weaker supports is more likely than a bias towards swing districts. Since in an incumbent president in Indonesia needs to maximize votes across all districts (does not have to win a district majority like e.g. in the US). Due to the increasing importance of political 15

economic factors and Indonesia s service-based candidate-voter relationship this could secure voting support in the next elections. Potentially also leading to better voting results at the local (DPRD) and national parliamentary elections (DPR). 24 This would facilitate a potential renomination of the incumbent president as a presidential candidate by his party and allow PD to govern more effectively once SBY gets reelected (e.g. having higher support in passing legislations in the DPR and maintaining a stable cabinet). 25 We are aware that Indonesia s central government has the possibility to distribute strategic favors using implicit transfers (subsidies, licensing) as well of course. However, given Indonesia s large scale of nondiscretionary transfer schemes the question remains why such a discretionary grant mechanism scheme actually exits, if not for political reasons. Hypothesis 3 ( Reward hypothesis ): Central discretionary grants are biased towards electoral core supporters of the president in the past. Additionally, if any type of this vote maximization calculus exits, it would most likely to be detected in the first term of the president, when reelection was still possible (2004-2009) not during his second term (2010-2013) once the president was facing his term limit. Hypothesis 4 ( Political investment ): (a) Central discretionary grants are biased towards districts with low support for the president. (b) This bias exists only in the first term, when reelection is possible. 24 Based on empirical evidence by Basewedan (2007), Indonesians voted for the same political party across different levels of government in the 2004 elections. For 2004, an increase in votes for PD at the national level, increased votes at the district level by 0.922, significant at the 1 percent level 24. From those voters voting for the PD in the 2004 parliamentary elections, almost 100 percent voted for candidates running for PD (Yudhoyono-Kalla) in the second round of presidential elections (Ananta et al., 2005)) 25 According to Sukma (2010) and Sherlock (2004) the president of Indonesia is still interested in the composition of the parliament (DPR) in order to achieve his legislative goals, despite him being elected directly by popular vote and not by the DPR. 16

4. Empirical Evidence for Indonesia 4.1 Data To test these 4 Hypotheses, we collected a unique unbalanced panel dataset including 497 Indonesian districts from 2004-2013, representing more than 90 percent of Indonesia s population in 2013. Our main sources of data are the Indonesian Database for Policy and Economic Research (DAPOER) of the World Bank Indonesia, Indonesia s Election Commission (KPU), the Ministry of Home Affairs, Ministry of Finance and the Statistical Office (BPS). For descriptive statistics, variable definitions and all corresponding data sources see Appendix A. All districts of the special autonomy provinces Aceh Darussalam are not included. 26 Province DKI Jakarta is excluded since as the capital region it has a special legal status and its districts are not autonomous. Papua had to be excluded due to data restrictions. Hence our main restricted sample includes 428 districts. 4.2 Empirical Model We use a pooled ordinary least squares model pooled OLS (1), clustering standard errors at the district level to account for serial correlation of unobserved variables within a district. Additionally, we use a fixed effects (FE) model (2) to control for unobservable time-invariant district characteristics. The dependent variable is defined as the total natural logarithm of the central discretionary government grant Tp per capita in district d at time t, ( ). Our Baseline controls for a districts fiscal capacity from other (non-discretionary) revenue sources and its overall socioeconomic development. All time-variant variables are lagged by one year, since budgetary decisions are usually made one year in advance. measures the fiscal capacity of a district as the natural logarithm of the total revenue per capita by other (nondiscretionary) revenue sources (own source revenue, DAU, DBH and DAK) in the year before. controls for a district s overall socioeconomic development. It consists of the natural logarithm of real GRDP per capita (excluding oil and gas), and the Human Development Index (HDI) at the district level. Plus we control for district size using population density (district population/area size) and for year fixed effects ( ) to account for common macroeconomic shocks. In the pooled-ols estimation (1) we include provincial dummies ( ) to control for unobservable time-constant factors at the province 26 According to Law No. 35 (2008), Law No. 11 (2006), Law No. 21 (2001) districts in these regions receive special autonomy funds. Our main results however do not change once we include Aceh Darussalam. 17

level, our omitted category being Bali Province. We add a dummy equal to one if a district is a city-district (kota), since more rural regencies (kabupatens) are likely to differ in need characteristics compared to urban areas. Plus we add a dummy for coastal regions ( ) to account for the difference between more landlocked districts and the once with a direct coastline. Due to the common district splits in Indonesia (permekaran) we add a Dummy equal one if a district lost parts of it administrative area within our observation period ( ). All these time-invariant district fixed effects are automatically controlled under a fixed effects model, thus they are excluded in specification (2).! is the error term. Base Specifications: (1) = # +% + + + + + +! (2) ''''''''''' = #( ''''''''')+%( '''''''''''''+ ) +!! ''' To test the effect of political alignment between the president and district heads we add a dummy variable (( ) equal to one for all the years the head of a district was formally affiliated to the presidential party (PD), meaning nominated by the presidential party (PD) in the last local district head elections, zero otherwise. Since local direct district head elections are independently conducted from national elections and at different points in time across districts, we have a within-district time varying measure of political alignment with the president. Thus we are able to run all our specifications in a fixed effects model, automatically controlling for all unobservable time-invariant district fixed effects (Table 2). To test for the effect of presidential vote support we define a dummy (()* ) equal to one, if a district supports the president and its party by more than 70 percent in the final round of the last direct presidential elections for all the years until the next presidential election (including the election year), zero otherwise 27. We also include various dummies for different vote share groups and a swing dummy, being equal to one if the incumbent president lost or won a district by less than 5 percent of the votes in the last presidential elections (Table 3-4). 27 Mean support for president in 2004 and 2009: 58.25 percent, standard deviation 16.12 percent. Minimum: 8.37 percent Maximum: 97.51 percent. In 2004: Mean 60.41 percent, standard deviation 17.12 percent. Minimum: 18.51 percent Maximum: 98.43 percent In 2009: Mean 56.35 percent, standard deviation 14.96 percent. Minimum: 8.37 percent Maximum: 94.97 percent 18

4.3 Empirical Results 4.3.1 Baseline Our Baseline specification (Table 1) show the overall socioeconomic development of a district is not significantly correlated with the actual grant allocation. Larger areas with a more dense population receive less per capita grants, most likely explained by economies of scale. In contrast to the according government regulation (Government Regulation 7/2008) however, the fiscal capacity of a district by other non-discretionary revenue sources is not correlated with discretionary grant allocations. Same accounts for the Human Development Index and and the GRDP of a district in the past. These results on the weak influence of these main need characteristics on Tp grant allocations are in line with Lewis (2016). In a detailed analysis for the need determinants of Tp grants for a different time period Lewis (2016) shows: transfers, local government spending and local service access are all positively associated with Tp grant allocations. Hence also not in the way predicted by normative theory and/or the according government regulation. This could be an indication for potential political factors at play, for which we are going to test for in the following. 28 4.3.2 Political Alignment Political Alignment between a district head and the president does not have any significant effect on the distribution of grants. According to our results testing Hypothesis 2 (Table 2), if a district head was nominated by the presidential party is not significantly correlate to the amount of central discretionary grants towards this district (1-10). 29 However districts holding local direct elections receive between 7-9 percent less grants per capita compared to two or more years before. This could be explained by uncertainty about the election outcome at the local level or limited lobbying capacity from the local level due to upcoming elections, both reducing the amount of grants allocated to districts. However there is no indication of a local political budget cycle or a preferential treatment of politically aligned district heads facing a local direct election. Grants one year before local direct elections are not significantly higher (3-4), and mayors from the presidential party do not receive more grants in a local election year (5-6). Plus district heads from one of Indonesia s highly entrenched political parties or 28 The fact that in all years before and after the national presidential elections in 2009 per capita grants towards districts are significantly smaller could be a first indication for this. 29 The amount of district heads between 2005-2013 only nominated and elected into office by the presidential party is low (10 districts). Hence the overall amount of observations with a district head affiliated with the presidential party only accounts for 49 observations. However we also find no bias towards districts heads nominated by a coalition including the presidential party (PD). 19

the main opposition party to the president do not receive significantly more or less grants (7-10). Golkar is one of Indonesia s most established parties, in power for many years under Suharto, hence most likely still well connected within Indonesia s government structure and according to Mietzner (2013), often described as having a large patronage capacity. Our results show that district heads from Golkar also do not receive more grants. If there is no preferential treatment of the district heads close to the central government or well entrenched in the government system, one could suspect a potential discrimination of the opposition party. Our results show however that district heads from PDIP, strongest opposition and competitor to the presidential party over our observation period, do not receive significantly less grants (9-10) either. To control for a potential political alignment at the provincial level we also controlled for districts under authority of a governor (province level) nominated by a coalition including the presidential party PD, showing no significant influence. 30 As expected (Hypothesis 2), we find no evidence for the political party background or political alignment of a district head with the president influencing the distribution of central discretionary grants in Indonesia. 4.3.3 Past Presidential Support Core supporting districts of the president in the last presidential elections receive significantly less grants per capita, in particular in a national election year. Instead the weak supporters of the president are targeted, not the swing districts. Core districts are defined as showing a vote support for the incumbent president of larger than 70 percent in the last local direct presidential elections. Here on top of our baseline controls all specifications are controlled for a possible change in the political composition of a district due to a district split (permekaran). We add dummies equal to one if a district split between 2004-2008 (incl. 04 and 08) and between 2009-2013 (incl. 09). If for example a district split in 2007, this dummy accounts for a potential change in the district s political composition up until 2009 (next presidential elections). After 2009 the district gets treated as its own new observation, categorized based on its own according presidential vote support in the 2009 elections. Since an incumbent president in Indonesia does not have to maximize district majorities but the total amount of votes we also control for the amount of eligible voters in the 2009 and 2014 presidential elections. 30 There was no governor only nominated by PD in our whole observation period from 2005-2013. 20

According to our results testing Hypothesis 3 (Table 3), being a core supporting district in the last presidential elections decreases grants per capita by around 19 percent per year (1-2), an additional 25-27 percent per capita in the year of the national presidential election in 2009 (3-4). Instead weaker supporters, showing an electoral support of less than 70 percent for the president receive significantly more grants. Districts having a vote support of 0-30 percent in the last presidential elections receive 30-50 percent more discretionary grants per capita per year than a core supporting district (5-6), and even more in the national election year 2009 (7-8). As expected, districts hardly won or hardly lost in the last presidential elections are not targeted significantly, also not in the national election year 2009 (9-12). Swing districts are defined as districts were the president lost or won by a vote margin of 5 percent. In summary, central grants are significantly targeted towards districts supporting the opposition candidate in the last presidential elections. Table 3 shows grants are invested in districts that showed the weakest vote support, having a higher return in the investment of votes. This is especially the case in the year of the direct national presidential elections 2009, were the incumbent president was running for reelection. If this vote maximization behavior by the incumbent government exists however, it should only be prevalent once the incumbent can actually run for reelection (Hypothesis 4). Splitting our observation period into a period where reelection of the president was possible (2005-2009) and one were not (2010-2013) gives us the results presented in Table 4a and 4b. Our results show, in the period where reelection was possible districts showing lower support in the 2004 national presidential elections receive significantly more Tp per capita grants (Table4a), in particular in the national election year of 2009 (2-4). In contrast, once the president cannot can for reelection anymore (Table4b), weaker supporters do not receive significantly more grants anymore (3). In summary, once the incumbent president still can be reelected central grants focus on his weaker supporters of past presidential elections, at the cost of the top core supporters. Once the president faces his term limit however, past electoral vote support does not seem to have this influence anymore. If any, moderate supporters are rewarded (Table 4b, Specification (2)). This is consistent with the view that discretionary grants are seen as an instrument by the president to increase his reelection probabilities. 21

Table 1: Baseline regression Discretionary Grants per capita (2005-2013) Natural Logarithm of Tp per capita (1) OLS (2) FE Log of GRDP per capita (ex. Oil/Gas) -0.036-0.163 [0.063] [0.253] Log of Fiscal Capacity pc. (non-discretionary) 0.158 0.091 [0.124] [0.093] Human Development Index 0.009 0.003 [0.010] [0.009] Log of Population Density -0.381*** -0.811*** [0.093] [0.311] Log of Area Size -0.277*** -0.815*** [0.095] [0.290] Dummy Any District Split -0.026 [0.060] Dummy Urban District -0.261* [0.134] Dummy Coastline 0.078 [0.054] 2005-0.600*** -0.678*** [0.128] [0.124] 2006-0.576*** -0.631*** [0.100] [0.096] 2007-0.215*** -0.238*** [0.060] [0.061] 2008-0.265*** -0.286*** [0.038] [0.041] 2010-0.851*** -0.811*** [0.049] [0.049] 2011-0.264*** -0.211*** [0.052] [0.056] 2012-0.317*** -0.217*** [0.061] [0.073] 2013-0.253*** -0.150* [0.078] [0.084] Province Fixed Effects Yes No _cons 12.125*** 20.056*** [2.845] [4.269] R 2 0.56 0.14 N 3331 3408 Note: Pooled OLS (1) and Fixed Effect Model (2). Robust standard errors clustered at the district level are reported in brackets. DKI Jakarta and Special autonomy regions (Aceh) are excluded from the analysis. Papua is excluded. The base category for this Province Dummy is Bali, the base year for the time fixed effects is 2009. Al time-variant controls are lagged by one year. * p<0.1; ** p<0.05; *** p<0.01 22

Table 2: Political Alignment and Discretionary Grants per capita (2005-2013) Natural Logarithm of Tp per capita (1)OLS (2)FE (3)OLS (4)FE (5)OLS (6)FE (7)OLS (8)FE (9)OLS (10)FE Dmy PD Mayor -0.107-0.376-0.105-0.374-0.081-0.356-0.102-0.374-0.096-0.372 [0.190] [0.254] [0.189] [0.251] [0.208] [0.275] [0.190] [0.251] [0.190] [0.250] DmyLocalDirectElection -0.093*** -0.074** -0.092*** -0.073** -0.092*** -0.073** -0.091*** -0.076** [0.031] [0.032] [0.031] [0.032] [0.031] [0.032] [0.030] [0.032] DmyLocalDirectElection in t+1 0.009 0.011 0.009 0.012 0.008 0.011 0.006 0.014 [0.042] [0.040] [0.042] [0.040] [0.042] [0.040] [0.042] [0.040] Dmy PD Mayor X DmyLocalElec. -0.115-0.085 [0.201] [0.201] Dmy Golkar Mayor 0.034 0.007 0.041 0.002 [0.062] [0.066] [0.064] [0.066] Dmy PDI-P Mayor 0.058-0.156 [0.098] [0.107] Year Fixed Effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Province Fixed Effects Yes No Yes No Yes No Yes No Yes No _cons 12.797*** 12.929*** 12.928*** 12.777*** 12.926*** 12.779*** 12.998*** 12.758*** 13.028*** 12.492*** [2.878] [3.865] [2.884] [3.917] [2.884] [3.918] [2.859] [3.916] [2.832] [3.924] R 2 0.58 0.15 0.58 0.15 0.58 0.15 0.58 0.15 0.58 0.15 N 2849 2912 2844 2907 2844 2907 2844 2907 2844 2907 Note: Pooled OLS (OLS) and Fixed Effects models (FE). Robust standard errors clustered at the district level are reported in brackets. DKI Jakarta and Special autonomy regions (Aceh) are excluded from the analysis. Papua is excluded. All Baseline controls are included. * p<0.1; ** p<0.05; *** p<0.01 23

Table 3: Past Presidential Vote Support and Discretionary Grants per capita (2005-2013) Natural Logarithm of Tp per capita (1)OLS (2)FE (3)OLS (4)FE (5)OLS (6)FE (7)OLS (8)FE (9)OLS (10)FE (11)OLS (12)FE DmyCorePresident (>70%) -0.193*** -0.190*** -0.161** -0.155** [0.068] [0.066] [0.070] [0.068] DmyCorePresidentXNational Election -0.270*** -0.253*** [0.087] [0.088] DmyElectoralSupportPresident1(0-30%) 0.335*** 0.476*** 0.262*** 0.410*** [0.092] [0.085] [0.090] [0.091] DmyElectoralSupportPresident2(30-40%) 0.222*** 0.221** 0.187** 0.196* [0.081] [0.100] [0.084] [0.101] DmyElectoralSupportPresident3(40-50%) 0.114 0.133* 0.085 0.108 [0.073] [0.079] [0.074] [0.080] DmyElectoralSupportPresident4(50-60%) 0.173** 0.168** 0.135* 0.133* [0.077] [0.074] [0.079] [0.076] DmyElectoralSupportPresident5(60-70%) 0.211** 0.115 0.207** 0.110 [0.083] [0.079] [0.084] [0.080] DmyElectoralSupportPresident1XNationalElection 0.572*** 0.470** [0.181] [0.185] DmyElectoralSupportPresident2XNationalElection 0.323** 0.242* [0.127] [0.130] DmyElectoralSupportPresident3XNationalElection 0.252** 0.202* [0.128] [0.122] DmyElectoralSupportPresident4XNationalElection 0.339*** 0.319*** [0.111] [0.112] DmyElectoralSupportPresident5XNationalElection 0.028 0.044 [0.133] [0.132] DmySwingPresident (+/-5% votemargin) -0.047-0.019-0.052-0.022 [0.071] [0.089] [0.071] [0.087] DmySwingPresidentXNationalElection 0.044 0.026 [0.155] [0.147] Year Fixed Effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Province Fixed Effects Yes No Yes No Yes No Yes No Yes No Yes No _cons 14.666*** 23.740*** 14.723*** 23.753*** 14.264*** 24.762*** 14.103*** 24.561*** 14.313*** 21.582*** 14.301*** 21.563*** [3.138] [5.269] [3.141] [5.245] [3.120] [5.017] [3.121] [4.879] [3.112] [4.974] [3.115] [4.978] R 2 0.56 0.14 0.56 0.14 0.56 0.14 0.57 0.15 0.56 0.14 0.56 0.14 N 3284 3284 3284 3284 3284 3284 3284 3284 3284 3284 3284 3284 Note: Pooled OLS and Fixed Effect Model (FE). Robust standard errors clustered at the district level are reported in brackets. DKI Jakarta and Special autonomy regions (Aceh) are excluded from the analysis. Papua is excluded. All specifications include Province Dummies (omitted category: Prov. Bali). A district is defined as a core supporter of the president if it showed an electoral support of larger than 70.0 percent for SBY (mean+ sd) in the last direct presidential elections of 2004 and/or 2009. All Baseline controls are included. * p<0.1; ** p<0.05; *** p<0.01 24

Table 4a: Past Presidential Support in 2004 and Discretionary Grants per capita (2005-2009), reelection possible Natural Logarithm of Tp per capita (1) OLS (2) OLS (3) OLS (4) OLS DmyCoreSupporterPresident (>70%) -0.218* [0.118] DmyElectoralSupportPresident1(0-30%) 0.554*** 0.439** [0.187] [0.176] DmyElectoralSupportPresident2(40-50%) 0.381*** 0.321** [0.147] [0.151] DmyElectoralSupportPresident3(50-60%) 0.216 0.176 [0.131] [0.130] DmyElectoralSupportPresident4(50-60%) 0.264** 0.203 [0.128] [0.133] DmyElectoralSupportPresident5(60-70%) 0.161 0.139 [0.126] [0.130] DmyElectoralSupportPresident1XNationalElection 0.509*** [0.185] DmyElectoralSupportPresident2XNationalElection 0.281** [0.138] DmyElectoralSupportPresident3XNationalElection 0.182 [0.128] DmyElectoralSupportPresident4XNationalElection 0.284** [0.123] DmyElectoralSupportPresident5XNationalElection 0.097 [0.141] DmySwingPresident (+/-5% votemargin) -0.008 [0.085] _cons 12.347*** 11.642*** 11.553*** 11.740*** [3.771] [3.728] [3.742] [3.731] R 2 0.54 0.54 0.54 0.53 N 1718 1718 1718 1697 Note: Pooled OLS. Robust standard errors clustered at the district level are reported in brackets. DKI Jakarta and Special autonomy regions (Aceh) are excluded from the analysis. Papua is excluded. All specifications include Province and Year Fixed Effects. All Baseline controls are included. * p<0.1; ** p<0.05; *** p<0.01 25

Table 4b: Presidential Support in 2009 and Discretionary Grants per capita (2010-2013), no reelection possible Natural Logarithm of Tp per capita (1)OLS (2)OLS (3)OLS DmyCoreSupporterPresident (>70%) -0.084 [0.088] DmyElectoralSupportPresident1(0-30%) -0.220 [0.143] DmyElectoralSupportPresident2(40-50%) -0.063 [0.109] DmyElectoralSupportPresident3(50-60%) -0.113 [0.105] DmyElectoralSupportPresident4(50-60%) 0.008 [0.097] DmyElectoralSupportPresident5(60-70%) 0.208** [0.098] DmySwingPresident (+/-5% votemargin) -0.106 [0.110] _cons 15.668*** 15.457*** 15.574*** [3.584] [3.531] [3.620] R 2 0.61 0.62 0.61 N 1566 1566 1566 Note: Pooled OLS. Robust standard errors clustered at the district level are reported in brackets. DKI Jakarta and Special autonomy regions (Aceh) are excluded from the analysis. Papua is excluded. All specification include Province and Year fixed effects. All Baseline controls are lagged by one year. * p<0.1; ** p<0.05; *** p<0.01 5. Robustness Checks Ministry of Finance Recommendation The absence of an investment in weak supporters once the president faces his term limit (Table 4b) could be driven by a change in the discretionary scope for the incumbent government. Since 2010 the Ministry of Finance gives a nonbinding allocation recommendation which districts are of higher priority for the allocation of grants. According to the Ministry of Finance this recommendation was implemented for grants to become more effective, transparent, proportionate, efficient and most importantly not concentrated in a particular (geographic) area. 31 Here districts are separated into 4 categories, the highest priority districts in time t are those that have a fiscal per capita capacity (based on nondiscretionary transfers and own source revenue) below the national average in t-2 and below the national average of the Human Development Index (HDI) in t-1. Second priority are those districts that have a fiscal capacity per capita larger the national average but a HDI below the 31 Recommendations on the Regional Balance of Funding, Ministry of Finance Indonesia (published in June 2017) 26

national average. Districts above the national average in HDI are of non-priority, independent of their fiscal capacity per capita. 32 Controlling for this (although rather limited) theoretical decrease in the discretionary scope by an unbinding recommendation of the Ministry of Finance does not change our results. District that are by definition of the Ministry of Finance top priority for the allocation of grants (below the national mean of HDI) do receive around 20 percent more Tp per capita grants compared to districts in the two lowest priority groups (below national HDI) (Table RB1). However, we also find this effect for the period 2005-2009, when the policy was not yet implemented. Here districts within this group receive significantly more and even by larger magnitude. Plus as we can see in Figure 3-4 below, although the priority group (priority 1 and 2) receives significantly more (Table RB1), there are still districts within this group receiving very different TP allocations. For the years after 2010 districts in the priority group 1 and 2 are only in the 4 th quartile of Tp per capita grant allocations. Plus some districts that should be non-priority cases are among the top recipients. Our main results do not change once we control for the presence of this allocation recommendation by the Ministry of Finance. 32 If their fiscal capacity per capita is above/below national average they are accounted as non-priority group 4/group 3. 27

Table RB1: Ministry of Finance Recommendation Natural Logarithm of Tp per capita (1) (2) (3) (4) (5) 2010-2013 2005-2009 DmyCoreSupporterPresident (>70%) -0.075-0.247** [0.086] [0.111] DmyElectoralSupportPresident1 (0-30%) -0.245* 0.588*** 0.469*** [0.139] [0.185] [0.174] DmyElectoralSupportPresident2 (30-40%) -0.092 0.368** 0.292** [0.105] [0.142] [0.145] DmyElectoralSupportPresident3 (40-50%) -0.131 0.197 0.149 [0.103] [0.136] [0.132] DmyElectoralSupportPresident4 (50-60%) 0.000 0.303** 0.246** [0.095] [0.119] [0.124] DmyElectoralSupportPresident6 (60-70%) 0.201** 0.184 0.151 [0.095] [0.119] [0.119] DmyElectoralSupportPresident1XNationaElection 0.419** [0.177] DmyElectoralSupportPresident2XNationaElection 0.273** [0.118] DmyElectoralSupportPresident3XNationaElection 0.173 [0.113] DmyElectoralSupportPresident4XNationaElection 0.205* [0.115] DmyElectoralSupportPresident5XNationaElection 0.119 [0.124] DmyPriority (Q1,Q2) 0.182** 0.199** 0.242*** 0.239*** 0.238*** [0.083] [0.081] [0.077] [0.078] [0.078] _cons 15.485*** 15.159*** 14.903*** 14.090*** 13.970*** [3.686] [3.598] [4.481] [4.446] [4.460] R 2 0.62 0.63 0.63 0.64 0.64 N 1553 1553 1342 1342 1342 Note: Pooled OLS. Robust standard errors clustered at the district level are reported in brackets. DKI Jakarta and Special autonomy regions (Aceh) are excluded from the analysis. Papua is excluded. All specifications included Province Dummies, omitted category Prov. Bali. The base year for the time fixed effects is 2005. All Baseline Controls are included. Non-Priority districts are the omitted category. p<0.1; ** p<0.05; *** p<0.01 28

Figure 3 source: authors illustration Note: Marked in red the top 25 percent recipients of Tp per capita allocations. In orange the third quartile, in blue and green the second and first quartile. Figure 4 source: authors illustration Note: Marked in red the top 25 percent recipients of Tp per capita allocations. In orange the third quartile, in blue and green the second and first quartile. 29