Factions in Nondemocracies: Theory and Evidence from the Chinese Communist Party

Transcription

1 Factions in Nondemocracies: Theory and Evidence from the Chinese Communist Party Patrick Francois, Francesco Trebbi, and Kairong Xiao December 16, 2017 Abstract This aer investigates, theoretically and emirically, factional arrangements within the Chinese Communist Party CCP), the governing olitical arty of the Peole s Reublic of China. Our emirical analysis ranges from the end of the Deng Xiaoing era to the current Xi Jining residency and it covers the aointments of both national and rovincial officials using detailed biograhical information. We resent a set of new emirical regularities within the CCP, including substantial leadershi remia in the Politburo and Central Committee, intra-faction cometition for romotions, and systematic atterns of cross-factional mixing at different levels of the olitical hierarchy. An organizational economic model suited to characterizing factional olitics within single-arty nondemocratic regimes rationalizes the data in-samle and dislays excellent out-of-samle erformance. JEL Classification codes: P3, P48. Francois: University of British Columbia, 6000 ona Dr, Vancouver, BC Canada V6T1L4, and CFAR atrick.francois@ubc.ca). Trebbi: University of British Columbia, 6000 ona Dr, Vancouver, BC Canada V6T1L4, CFAR and NBER francesco.trebbi@ubc.ca). Xiao: Columbia Business School, 3022 Broadway, New York, NY United States kairong.xiao@gsb.columbia.edu). The authors would like to thank Matilde Bombardini, Wei Cui, Hulya Eraslan, Chang-Tai Hsieh, Ruixue Jia, Hao Li, Peter Lorentzen, Debraj Ray, Thorsten Rogall, Guoguang Wu and seminar articiants at various institutions for their comments and suggestions. Francesco Trebbi gratefully acknowledges suort by the Canadian nstitute For Advanced Research and the Social Sciences and Humanites Research Council of Canada. 1

2 1 ntroduction This aer resents a theoretical and emirical analysis of the internal organization of China s olitical linchin: the Chinese Communist Party CCP). As the regime arty of the Peole s Reublic of China PRC), the CCP is, de jure and de facto, the be-all and end-all of olitical activity in the second largest economy and the most oulous country in the world today. 1 This motivates the interest of olitical economists in the CCP. The oaque and often informal nature of elite interaction within a country lacking cometitive elections and with a rich history of informal olitics raises formidable obstacles to a rigorous olitico-economic analysis. The economic literature on the internal organization at the highest levels of the Chinese government is limited. 2 Political scientists focused on China studies have been more attentive, but also often more qualitative and descritive, at least until recently. 3 The CCP remains today a secretive, selective organization of about 65 million members who have ositions of influence in all sectors of Chinese society... Nathan and Gilley, ). 4 Oerations of the Politburo and the highest echelons of the CCP have been often described as oaque at best Pye, 1980; Dittmer, 1995; Shih, 2008). Within this context, intra-elite cometition is extremely hard to assess. The CCP officially rejects factional elite olitics 5, but scholars since Nathan 1973) have emhasized how the faction intended as a atron-client cluster of mutually linked officials reresents the correct unit of analysis of elite olitics in China. Since Nathan 1973), evidence suorting this interretation has steadily accumulated Pye, 1981; Dittmer and Wu, 1995; Nathan and Gilley, 2003; Shih, 2004; Li, 2012; Li, 2013; Shih, 2016; Meyer, Shih, and Lee 2016). Existing literature often ortrays a faction as a single fictional decision maker. While this unitary action assumtion may be useful in some scenarios, such as formal olitical arties in Western arliamentary democracies, it is less so in the case of Chinese olitics. Factions in China are rather informal and loosely organized grous of olitical entrereneurs Bo 2004; 1 And lays a crucial role in steering economic activity in the country. See Bai, Hsieh, and Song 2016). 2 The study of the olitical economy of China has several imortant excetions, but often not recisely focused on national elite cometition. Persico, Pueblita, and Silverman 2011) in their analysis of factional olitics focus on the CCP, among their various case studies. Less relatedly, work such as Li and Zhou 2005) and Jia, Kudamatsu, and Seim 2015) focuses on the role of ersonal connections in romotion of rovincial leaders. Work by Lau, Qian, and Roland 2000) models the rocess of reform under Deng Xiaoing and the reform era. 3 Descritive discussion most ertinent to this aer includes Li 2012, 2013). Several quantitative excetions are discussed in Shih 2016) with resect to scholarshi in East Asian studies and olitical science, while less recent examles include Huang 2000), Shih 2004, 2007); Shih, Adolh and Liu 2012). 4 At the time of writing. By 2016 the CCP membershi has grown to million. 5 BBC, Monday January 5, 2015: An editorial in Monday s flagshi newsaer, The Peole s Daily, says cliques are akin to arasites and are harmful for both the country and the eole. htt:// 2

3 2008), linked by recirocal social ties guanxi). CCP factions dislay very limited olicy differentiation Dittmer, 1995). They also exhibit rich internal dynamics. For instance, the career trajectories of junior members aear to move in lockste with the fortunes of their atrons, and cometition often arises among members of equal seniority within a faction. Recently, for instance, Lungu 2017) describes how two owerful equal ranking members of the CYLC faction 6, Wang Yang and Li Yuanchao, cometed for one osition in the Politburo Standing Committee of the most recent 19th Central Committee. Eisodes of within-faction internal cometition are reorted in Qiu 2005) and Li 2007). To understand the internal dynamics in Chinese olitics, we resent a organizational economic model of factional interaction. While caturing cross-faction interaction and factional evolution, we ay secial attention to individual incentives, sulying a career concerns model of individual romotion in the CCP hierarchy. n our model lower-level officials [...] join factions in order to secure romotions and other regime goods from owerful atrons Shih, 2016,.1) and romotion dynamics throughout the arty organization are microfounded and characterized. This framework catures the olitical tradition in China, where the government landscae is shaed by the gradual rogression of individual oliticians, rather than by sweeing electoral shifts as in Western democracies. n our model, factions oerate within a given arty hierarchy, contesting oen ositions. The seniority of a member within the faction is determined by his level in the arty. A senior faction member can decide whether to give suort to romotions of junior members within his jurisdiction. On one hand, a successful romotion of a cofactional enhances the faction s overall ower, which can benefit the senior member. On the other hand, the ascendancy of a junior faction member may dilute the chance of romotion for the senior member himself, esecially when the junior asires to an equal rank as the senior. The trade-off faced by the senior member is between enhancing the faction s overall ower and reserving his own career rosects. We show that when an oening occurs far down the hierarchy, the senior member is likely to suort such romotion, as the concern of internal cometition is minor. However, when an oening occurs at the same level of the senior member, he will ask the junior to wait until the senior himself is romoted. The faction seniority is reserved endogenously by the allocation of factional suort and a junior member s ascendancy through the hierarchy is tethered to the rise of the relevant seniors above him. n the unique Markov Perfect) equilibrium that we characterize, the factional comosition of each level of the hierarchy at every oint in time determines the romotion oortunities at lower levels of the hierarchy, which ultimately affect the costs and benefits of joining factions. Unaffiliated neutral) oliticians emerge in equilibrium because, although they do not enjoy factional suort, they are also not restricted in their caacity to contest oenings higher u. This is a relevant theoretical result, as in the data a substantial share of oliticians aear 6 We will rovide information on this faction, and others, below. 3

4 neutral. Using a large biograhical database which covers more than 5,000 Chinese olitical elites, we document a set of new emirical regularities within the CCP which anchors our model. Absent hard and verifiable information, we rely on the extant discussion of Chinese elite olitics to identify a minimal set of factions within the CCP. Factions have historically emerged within the CCP through close ersonal connections with rominent atrons e.g. in the cases of former General Secretary Jiang Zemin and his successor, Hu Jintao) to mutually foster the career rosects of affiliated cadres, and do not necessarily reresent secific territorial or economic interest grous Dittmer, 1995). As we discuss in Sections 2 and 3, this aer will lever only the most obvious factional links identified within the CCP, links based on affiliation to the Communist Youth League of China CYLC, related to General Secretary Hu Jintao) or to the so-called Shanghai Gang affiliated most rominently with Jiang Zemin and bolstered by the secial status of Shanghai in Chinese olitics). A necessary condition for our model s coherence is that factions may deliver advantages to their members. n the data, we show systematic evidence that faction affiliation, on average, increases one s chance of romotion comaring to an unaffiliated olitician. Furthermore, the benefit of being a faction member is time-varying: when a faction leader becomes the sureme leader of the arty, faction members enjoy additional remia in romotion. The existence of recisely estimated leadershi remia oints in the direction of factions both being reasonably identified within our analysis and of oerative relevance within the arty. n line with the organizational economics aroach that we follow, we find emirical evidence that cometition arises internally within a faction. Secifically, at each level of the arty hierarchy, if there are more candidates from the same faction cometing for the same oenings, each individual faction member s chance of romotion is diluted. n contrast, the resence of faction members at other levels of the hierarchy has no such effects. These emirical results are not driven by time-varying romotion oortunities at each level of hierarchy or by changes in a faction s bargaining ower. Consistently with the model, we find systematic atterns of cross-factional mixing at different levels of the olitical hierarchy. Two oliticians from the same faction in the same office and area rarely share ositions with equivalent rank. For instance, in the rovincial government the to two ositions are Provincial Party Secretary and Provincial Governor. We find that if the Party Secretary is from one faction, the Governor is very unlikely to be from the same faction. This attern endogenously arises in our model, as a faction senior asks juniors who asire to an equal osition as the senior to bide their time until they are themselves romoted. To further examine the quantitative relevance of our aroach, we estimate our model structurally, obtaining estimates of the rimitive arameters such as leadershi remia and arameters governing the contest functions for romotion), and formally test our mechanism against alternatives, including ones based on ure seniority or meritocracy. We show 4

5 how the estimated leadershi remia in the CCP are quantitatively substantial, but quite far from winner-take-all levels, and show that intra-faction cometition among members oerates as a de facto endogenous damening mechanism in slowing down factional growth. Our model generates an excellent in-samle fit in matching emirical moments in terms of factional comosition, romotion rates, and effects of changes in the factional identity of the to leadershi in China. n an out-of-samle exercise, our model redicted very closely the factional arrangements in Politburo romotions for the most recent 19th) Party Congress of October n our referred secification, an out-of-samle rediction made in October 2016 matched the emirically realized factional arrangement almost exactly. 7 We redicted Shanghai Gang, Neutral, and CYLC romotion shares of 24%, 66%, 9%, resectively, versus realized shares of 25%, 65%, 10%. We believe this, at the very least, rules out gross misecifications in our analysis. The aer includes several counterfactuals, imortant to assess the olitico-economic equilibria of China. We model ossible institutional changes within the CCP, including the effect of increased leadershi remia, which may indicate a break away from the collective leadershi design envisioned by Deng. We also study the role of the identity of the to leadershi, secifically, General Secretary Xi Jining s factional affiliation. A final emirical contribution is a formal test whether certain olitical grous constitute olitical factions. Political observers and oular media often use attributes such as college, hometown, or family background to grou similar oliticians into factions. This leads to a lethora of utative factions, such as Tsinghua Clique, Shaanxi Clique, and Princelings a rominent set of cadres with family ties to CCP veterans). t is an ongoing debate in the literature whether some of these grous are relevant olitical units, yet no formal statistical tests on the identity of factions have been roosed. We show how our structural aroach allows us to fill this void. mortantly, we reject at standard confidence levels the hyothesis that the often mentioned Princeling grou to which both Xi Jining and the disgraced former governor Bo Xilai are said to belong) is a faction. Our aer contributes to the olitico-economic literature on Chinese elite olitics. Scholars such as Shih, Shan, and Liu 2010), Shih, Adolh, and Liu 2012), Jia et al. 2015) have exlored methodologies for the imutation of factional linkages based on lace of birth, university ties, and shared career rofiles. 8 While we also focus on systematic biograhical information, we remain wary of otential mismeasurement in the identification of factional ties, as is likely for factional affiliation based urely on lace of birth or shared career aths. An imortant reason for this wariness is evident in our statistical analysis. Based on a minimal set of factions which are well established in the extant discussion of Chinese elite olitics, Shih 2008,.66) discusses issues of measurement with the remise that Desite the centrality of factions in Chinese olitics, they are extremely difficult to observe in a systematic manner, esecially in such an oaque olitical system. 5

6 we find that oliticians working in the same deartment or arty branches are not necessarily from the same faction. nstead, if they have similar ranking, they are actually more likely to belong to different factions. Thus, simly sharing art of their career aths may not be informative of factional affiliation for CCP elite officials. n fact, our evidence shows it may mislead comletely. 9 The statistical analysis of these systematic factional distributions in to CCP ositions is new to the literature. Besides the olitico-economic literature on Chinese elite olitics mentioned above, this aer seaks to the literature on the internal organization of autocratic regimes. Francois, Rainer, and Trebbi 2015, 2016) discuss at length the imortance of its connection to the exanding literature on the olitical economy of develoment. Most related to our work and one of the first rigorous analyses of factional olitics within the economic literature) is Persico, Rodriguez-Pueblita, and Silverman 2011), who resent a theoretical model of endogenous factional growth in a democracy with cometitive elections and link it qualitatively to evidence from factional local olitics in Mexico within the nstitutional Revolutionary Party. 10 n Persico et al. 2011), romotions of members from the same faction are bounded together, deending on the outcome of local elections. n contrast, we allow each individual faction member to have its own career ath, which gives rise to a richer set of redictions on the internal dynamics of factions. Dewan and Squintani 2015) model endogenous faction formation an issue we address in our setting as well, when characterizing the decision of arty members to join a faction). The authors develo a model where incentives for faction formation are ideological rather than economic as in our setting and in Persico, Rodriguez-Pueblita and Silverman, 2011) and show how within their framework factions may serve welfare-enhancing uroses, limiting extremists within the arty by tying them to moderate faction leaders. Factions are also shown to facilitate information sharing and arty effectiveness in their model. The remainder of this aer is organized as follows. n Section 2 we rovide a brief institutional overview of the CCP. n Section 3 we discuss our data, oerationalize factions, and rovide a descritive analysis of our samles. Section 4 roduces a set of stylized facts, some novel, useful to frame and guide the theoretical analysis. n Section 5 we discuss our theoretical setu and Section 6 develos our estimator. Our main emirical results are reorted in Section 7. Section 8 resents our counterfactual exercises. Section 9 concludes. 9 Fisman et al. 2017) shows that using hometown and college ties may also lead to mismeasurement since leaders may want to avoid the aearance of favoritism. n fact, hometown and college ties with leaders are associated with lower romotion robability. 10 See also Belloni and Beller 1978). Persico et al. 2011) also oint to the relevance of factional olitics well beyond Mexico s camarillas or the CCP, with references to studies of factionalism within the Jaanese legislature Cox et al., 1999, 2000) and the talian arliament Zuckerman, 1975; Kato and Mershon, 2006; Ceron, 2015; and Laver and Giannetti 2004). Factions in Australian olitics are discussed in McAllister 1991). The US urban arty machine factional structure, such as in the case of Tammany Hall, are subject of an entire and even earlier literature. See Myers 1917). 6

7 2 nstitutional Background: the CCP This section resents a brief institutional overview of the internal organization of the CCP in the reform era. t is in no way exhaustive, but only of assistance to the reader unfamiliar with Chinese olitics in framing the analysis that follows. 11 n 2016 the Chinese Communist Party, with its 88.8 million members, is one of the largest olitical arties worldwide and one of the most enduring founded in 1921). The CCP organization is strongly hierarchical in nature and the arty reflects one-to-one the organization of the Chinese state, as tyical in the architecture of Leninist regimes. The to of the CCP hierarchy is shared by the figures of the General Secretary of the CCP and the second ranked member of the CCP, which resectively assume the roles of President and Premier of the State Council of the PRC. Both leaders belong in turn to the Politburo Standing Committee PBSC), formed by the other 5 members and which reresents the set of the highest ranked oliticians in China. The PBSC is an exression of the 25-member Politburo PB), the executive body of the Central Committee of the Chinese Communist Party. The Central Committee CC) is de jure the highest olitical body in the CCP and currently consists of 205 full members and a set of 171 Alternate Central Committee AC) members in junior standing relative to the full members and without voting rights). All members of the CC and AC are ranked hierarchically. The CC and AC are elected during National Congresses of the CCP and the interim lenary sessions fill retirements or deaths, granting romotions and occasionally administers demotions). Tyically, CC members include ministerial-level officials and rovincial ranking officials, including Provincial Party Secretaries the highest CCP ost in a Province) and Governors the second ranked). t is imortant to notice that Provinces tend to dislay a olitical architecture that mimics the national government and the national arty structure. Provincial leaders oerate in the context of local arty committees and local arty congresses are held tyically every five years. The CCP maintains a yramidal structure, branching all the way down to the village level and the Village Party Branch Secretary. While not all layers of the Chinese olitical hierarchy resent nodes maing into a diarchic structure, most do, tyically searating arty roles and administrative roles. Examles of diarchic arrangements include the residency and remiershi as the two highest ranking members of the Politburo Standing Committee; the PRC Presidency President and Vice President); the State Council Premier and Executive Vice Premier); and the to dyads at the rovincial level Provincial Party Secretary and Governor). 12 We will occasionally refer 11 See also Chater 1 in Nathan and Gilley 2003) for a less brief overview. For a comrehensive discussion of elite olitics in China see references in Shih 2016). 12 See Li 2014) for a discussion and examles. Other instances include the CMC chairman and executive vice chairman), the CCP Secretariat, the NPC and CPPCC chairman and executive vice chairman), the Sureme Peole s Court. Assuming the resence of such dyads across the whole hierarchy should be simly 7

8 to such airs of ositions as osition 1 and 2. The oortunity of entering the ranks of the CCP is closely guarded and arty membershi tyically guarantees access and career oortunities beyond those available to common citizens. 13 For this reason, an elaborate recruitment rocess tyically oerates through the selection of successful university students and through family and work connections. Membershi of the Communist Youth League of China CYLC), an ancillary organization to the CCP resonsible for the youth members are tyically between 4 and 28 years of age), has traditionally oerated as an entry oint in the CCP. As discussed in Li 2012, 2013), individuals with a background in the CYLC are often referred to as members of the tuanai i.e. Youth League [faction]) and tend to originate, although by no means exclusively, from the less roserous red ) regions. 14 Li 2012) associates with the CYLC oulist olicies close to the rural oor and recent migrants to cities, as oosed to the olicies referred by more elitist grous comrised by CCP cadres close to former General Secretary Jiang Zemin and a grou of arty officials connected to the Shanghai municial administration. ndeed, the economic and olitical role of Shanghai cannot be emhasized enough in CCP internal interactions, to the oint that the term Shanghai Bang Gang) has often been emloyed to identify the atronage cluster close to Jiang and to the economic interests of the coastal blue) rovinces Li, 2002). Whether additional factional grous besides the CYLC and the Shanghai Gang may be resent within the CCP is unclear and disuted even among scholars of Chinese elite olitics. For instance, some observers oint at the anomaly of the excetionally raid careers of sons and daughters of rominent arty officials and revolutionary veterans under Mao, often referred to as Princelings. The analysis below will discuss this secific grou of CCP members in detail. 3 Data We combine two biograhical databases of Chinese oliticians. The first data source is China Vitae, which collects biograhical information on more than 4, 494 Chinese elites in government, olitics, the military, education, business, and the media since nformation rovided by China Vitae includes gender, year of birth, lace of birth, ethnicity, colleges attended, and career trajectory. nformation in China Vitae comes from Chinese and English language web sites in China that are suorted by or affiliated with the Chinese government. read as allowing for the resence of a close substitute in the arty hierarchy for any member. 13 The Organization Deartment of the CCP Central Committee on June 30th, 2016 in an official release indicated that 22 million Chinese residents had alied in 2015 and less than 4.5% of the alications were acceted. htt://news.xinhuanet.com/english/ /30/c htm 14 Prominent members include current Premier Li Keqiang and former General Secretary and President of the PRC Hu Jintao. 8

9 Our second data source is a biograhical database of CC members develoed by Shih, Shan, and Liu 2008), and further udated by Lu and Ma 2015). This database contains all CC and AC members from the first Party Congress in 1921 to the eighteenth Party Congress in This data also rovides biograhical information and career trajectories similar to China Vitae. We focus our analysis on the eriod of 1956 to 2014, which starts from the first Party Congress since the founding of Peole s Reublic of China 8th Party Congress in 1956) and ends with the most recent Central Committee 18th Party Congress in 2012), covering a total number of 1, 853 individuals. Data for the 2017 nineteenth Party Congress is still being comleted and only available at this time for the Politburo and higher levels. We combine these two data sources to construct our estimation samles. Whenever there is inconsistency between the two data sources, e.g. multile oliticians in the same osition in the same year), we manually check with a third source, tyically official websites affiliated with the Chinese government e.g. cc.eole.com.cn). We also collect rovincial oulation and GDP data from China Data Online. The anti-corrution data originates from ChinaFile and China s Central Commission for Disciline nsection CCD) website. Following the literature on Chinese olitics Bo, 2008; Li, 2013a; Li, 2013b), we construct four affiliation indicators for the full samle of oliticians: CYLC, Shanghai Gang, but also Military and Princeling status. A olitician is classified as from the CYLC if he/she has held rovincial and national level ositions in CYLC. A olitician is classified as from the Shanghai Gang if he/she has held official ositions in the Shanghai municial arty aaratus, municial government, municial Peole s Congress, and municial Peole s Political Consultative Conference. This again underlies the excetionality of the Shanghai olitical machine. A olitician is classified as from the Military if he/she served as military ersonnel in the Revolutionary Era ), or has articiated in the volunteer armies to Korea or Vietnam, or served as military ersonnel for more than half of its career after the founding of Peole s Reublic of China. The restriction on the minimum time of military exerience is to rule out civilian officials who work as the arty secretary of a military region for a short eriod of time e.g. Hu Jintao as the First Secretary of Guizhou Military District from 1985 to 1988), or civilian officials who chair the Central Military Commission e.g. Jiang Zemin as the chairman of the Central Military Commission from 1990 to 2005). A olitician is classified as a Princeling if he/she is from a rominent olitical family, the so called red aristocracy rominent examles include General Secretary Xi Jining and disgraced former governor of Liaoning Bo Xilai). These four affiliations are not mutually exclusive for examle, Xi Jining is both a Princeling and an affiliated of the Shanghai Gang according to our definition) and not all arty members in our samle are affiliated. n fact, we allow for oliticians in our samle to also be unaffiliated neutral, indicated as N). Theoretically one could consider CYLC, Shanghai Gang, military, and Princelings alternative olitical factions. n Section 4 we show however than only two of these grous, CYLC 9

10 and Shanghai Gang, truly exhibit the features of olitical factions within the CCP. Formal statistical tests will also be develoed and brought in as suort of this assertion. To distinguish, we will refer to Princelings and military as grous and CYLC and Shanghai Gang as factions. The military is virtually a arallel structure with limited olitical control, while the Princelings as a grou are extremely heterogeneous and aear to oerate as a set of neutral and indeendently owerful actors in fact, often times in dee rivalry among themselves, such is the case of Bo Xilai and Xi Jining). While we will kee track of all tyes of affiliations in the analysis that follows, we emhasize here that our theoretical and emirical design will searate CYLC and Shanghai Gang faction members from all other olitical actors, including the military and Princelings, which we will deem neutral. Because of the traditional coloring associated with these two established factions, we will also occasionally refer to the CYLC as the Red faction, R, and to the Shanghai Gang as the Blue faction, B. Table 1 rovides summary statistics of demograhics and careers of 4, 494 elites who held imortant ositions in government, olitics, the military, education, business, and the media in China since The unit of observation is a osition-individual air. We classify the organizations into 12 categories: arty aaratus, government, military, Peole s Congress, Chinese Peole s Political Consultative Conference CPPCC), court, rocuratorate, CYLC, business, media, education, and an unclassified category. The average duration of each osition is about 4 years, and the age of starting each osition varies from the early 30s CYLC) to the late 50s Peole s Congress). ndividuals who hold these ositions are redominately male, which reflects the large gender imbalance at the to levels of government and business in China. 15 Ethnicity is redominately Han, reflective of the ethnic comosition in the Chinese oulation. The last four columns rovide the frequency of the various affiliations in each tye of organization. CYLC members tend to work in the arty aaratus and media instead of the government system. 16 The Shanghai Gang is more evenly distributed across all tyes of organizations. Princelings are more likely to have exerience in the military, but are less likely to work in the legal system court and rocuratorate), otentially due to the fact that the ower of the judiciary is relatively muted in China. We then turn our focus to a subset of elites, the members of Central Committees of the CCP. This is a grou of around 400 eole who comrise the CCP to leaders. Table 2 rovides the demograhics and the factional affiliation by sessions of the Central Committees. Similarly to the larger samle of elites, the CC members are redominantly male, in their mid-50s and mostly Han. Over the ast 60 years, more members hold college or even ostgraduate degrees. However, only 10 ercent of them studied or worked abroad. More than 10 ercent of them have worked as ersonal secretaries Mishu) of rominent oliticians, % of CCP members were women in This is consistent with the anecdotal discussion of Hoffmann and Enright 2008) that CYLC leaders often have exerience in non-economic fields, such as arty organization and roaganda 10

11 illustrating the imortance of ersonal ties in Chinese olitics. Conditioning on entering the Central Committee, around 20 ercent of them are romoted to a higher level in the four levels of the Central Committee, and around 50 ercent will retire in the next CC session. n terms of factional affiliation, CYLC, Shanghai Gang, and Princelings each account for around 5 ercent to 10 ercent of members. The military has exerienced a large downward trend, droing from 56 ercent in the 8th Central Committee to less than 20 ercent in recent years. 4 CCP Factional Politics: Reduced Form Results This section resents a set of facts on factional olitics in China, the most imortant of which are novel, to the best of our knowledge. These stylized facts are going to inform and motivate the theoretical analysis that follows. i) National Political Actors. We begin by arguing qualitatively that the factional affiliations we osit CYLC and Shanghai Gang) share roerties that make them bona fide large national layers within the CCP and are not merely olitical actors reresenting local constituencies. n Figure 1 and Table 3 we describe the geograhic distribution of members affiliated with the CYLC and the Shanghai Gang in rovincial roles. As is evident, the reresentation across rovinces is fairly broad and not limited to a articular local area, desite a small ositive correlation between the resence of Shanghai Gang and the average GDP er caita of the rovince. On the other hand, individuals associated with Princelings and the military grou are distributed more unevenly: Princelings are more likely to hold ositions in rich coastal areas ossibly due to their rivileged status while military members are more concentrated in oorer western rovinces and laces with strategic imortance e.g. Fujian, which neighbors Taiwan). ii) Leadershi Premia. A crucial feature of any theoretical model of factional olitics is the ability of factions to deliver resources to their members. This seems a necessary condition that our factional definition should satisfy, a concetual underinning that we must be able to verify in the CCP data in order to justify our aroach. We will do this in what is ossibly the starkest way: estimating remia in factional seat assignment and romotion rates of cofactionals of the country leader i.e. the PRC President and General Secretary of the CCP). Again, we are not aware of any systematic analysis of this tye for the CYLC and Shanghai Gang. Table 4 shows a anel regression of romotion and retirement dummies on the factional affiliation of Central Committee members interacted with the faction of the General Secretary. The samle includes all members of the 8th to the 18th Central Committees Politburo Standing Committee members are excluded from the romotion regression). Promotion is 11

12 equal to 1 if a Central Committee member moves u in the rank defined by the four levels of Central Committee 1 PBSC, 2 PB, 3 CC, and 4 AC). As is clear from the reduced form regressions, an R resectively, a B) olitician has substantially higher likelihood of romotion when an R resectively, a B) leader is in ower. On average CYLC and Shanghai Gang members exhibit romotion rates higher by 10 ercentage oints relative to neutral members excluding military and Princelings), as reorted in Aendix Table 3. However, this result masks substantial heterogeneity. While CYLC and Shanghai Gang members romotion rates hover around 4 ercentage oints higher than neutrals in times where the leadershi is not from an individual faction, having a cofactional leader adds 20.6 ercentage oints to CYLC and 19.3 to Shanghai Gang, inducing a substantial, highly significant, leadershi remium to the seed at which leader s cofactionals are romoted. Figure 2 rovides a visualization of the leadershi remia in romotion rates. We also erform an analysis looking at allocations of crucial osts to factional members. The deendent variables include: the share of official ositions allocated to a faction constructed following the scheme of Bo 2010) and weighted by value we will refer to it as ower score ); the share of seats of Alternate Central Committee members AC); of the full Central Committee CC); of the Politburo members PB); and of the Politburo Standing Committee members PBSC). These effects are reorted in Table 5. Leadershi remia are statistically significant, between 4 ercentage oints higher in terms of ower score shares for the CYLC and around 2 ercentage oints for the Shanghai Gang. These estimates are not trivial, but quite far from winner-take-all levels. The leadershi remia in the ower score can be easily observed in the simle time series lots of Figure 3. iii) nternal cometition between faction members Given the total resources of a faction are often limited, cometition among faction members may arise. Anecdotal evidence suggests faction members indeed comete internally for romotion. 17 We formally test whether whether the chance of romotion of a faction member is affected 17 For instance, an article in the Dilomat notes that two members of the CYLC faction, Wang Yang and Li Yuanchao, cometed for one osition in the Politburo Standing Committee of the 19th Central Committee: n the case of Wang Yang, his main cometitor is Li Yuanchao. Both are considered art of the Youth League faction, being allies of Hu Jintao. Both served on the Politburo starting in Both hold state ositions vice resident and vice remier). As a former head of the organization deartment, Li Yuanchao should eventually reach the PSC and this is his last chance. On the other hand, Wang Yang could still remain second-ranked vice remier and reach the PSC in 2022, as vice remier. But it s ossible that Li Yuanchao, who is 67, might be forced to retire or to remain on the Politburo in order to facilitate Wang s romotion Lungu, 2017). As it turns out, Wang Yang managed to enter the Standing Committee but Li Yuanchao did not. Similar cometition also arose when two high rank CYLC members, Li Keqiang and Li Yuanchao, cometed to succeed Hu Jintao Li, 2007). Another story is documented by Qiu 2005): when Li Keqiang was the head of the Deartment of School Affairs of the CYLC Central Committee, he faced fierce cometition from the head of another deartment for romotions. Such bitter rivalry even led Li s cometitor to worry that Li would use his ower to seek revenge against him after Li was romoted to be first secretary of the CYLC. 12

13 by the share of cofactional members in the same level. We focus on the two well-established factions, CYLC and Shanghai Gang. The result is reorted in Table 6. We find that a higher share of cofaction members reduces a olitician s chance of romotion. This suggests that indeed there is intra-faction cannibalization in the romotion cometition. We also include the overall faction share in the whole hierarchy as a control variable. nterestingly, the overall faction share has insignificant effects on the romotion, suggesting that intra-faction cometition only occurs among oliticians with similar ranking. The result is not driven by the time-varying romotion oortunities, as adding rank-year fixed effects in column 4 barely changes the result. The result is not driven by time-varying faction bargaining ower, as adding year-faction fixed effects in column 5 does not change the result either. The above finding suggests that factions oerate in a more comlex way than the unitary action assumtion would suggest and there is substantial internal cometition among faction members. n a sense, a faction is a team of rivals. n some circumstances, faction members hel each other, while in other situations they may comete with each other. A deeer understanding of the incentives of individual faction members seems to be a necessary ste to understand how factions actually work. iv) Cross-Factional Mix. Useful to the understanding of factional dynamics within the CCP is the study of the eculiar factional mix which we observe when samling the diarchic nodes ervading Chinese institutional design. These are airs of ositions of similar rank and oerating in close institutional roximity to each other. Table 7 shows the raw frequency of the factional comosition of virtually all to two leadershi osts in ost-deng China. Table 8 reorts formal statistical tests. n articular, we ask: given the factional affiliation of a olitician sitting in one of the to two leadershi ositions of a national or rovincial organ, what is the likelihood that the other osition will be held by a cofactional member? t turns out it is extremely low. Table 8 shows anel regressions of the factional affiliation of the number 1 official on the number 2 official s affiliation at the same node. The variables CY LC1 and Shanghai1 resectively, CY LC2 and Shanghai2) are dummies which equal 1 if the number 1 official resectively, number 2) is from that faction and 0 otherwise. We will also refer to such factions through the abbreviations R, B. The samle eriod is from 1992 to Columns 1-4 include all ositions, and Columns 5-6 break down to rovincial and national level ositions. The rovincial ositions include 31 rovincial and municial units secretary and governor) 18. The national ositions include the Politburo Standing Committee two highest ranking members), PRC residency President and Vice President), the State Council Premier and Executive Vice Premier), Central Military Committee Chairman and Executive Vice Chairman), CCP Secretariat two highest ranking secretaries), NPC Chairman and Executive Vice Chairman), CPPCC Chairman and Executive Vice Chairman), the Sureme 18 Shanghai Municiality is excluded in the regression samle of Shanghai Gang. 13

14 Peole s Court President and Executive Vice President). Taking the to two leadershi ositions in any CCP or PRC) organ, osition 2 being filled by a R resectively, a B) olitician redicts negatively and significantly the likelihood of osition 1 being filled by an R resectively, a B) olitician. The estimated negative coefficients indicate a statistically robust lower likelihood of same-faction airs R, R) or B, B) relative to what would haen in case of airings forming randomly between B, R, N. nterestingly, the evidence for Princelings is much weaker, in line with further evidence below showing their lack of behavior as an organized faction. To the best of our knowledge these facts on systematic cross-matching within Chinese elite olitics are new. An imlication of this evidence is that methodologies imuting factional affiliation based solely on shared rofessional aths may be roblematic, as discussed in the ntroduction. iv) Anti-Corrution Camaign. As in the allocation of rewards to cofactionals through leadershi remia, we would also exect evidence of factional bias in the administration of unishment. We have limited systematic evidence in this resect, but it interestingly oints in a direction consistent with the limited leadershi remia discussed in oint iii). This novel evidence comes from the factional analysis of the CCP members hit by President Xi Jining s anti-corrution camaign initiated in 2012 and still ongoing as of 2017). A remarkable factional balance seems to be resent in the administration of unishment, when looking at the detailed resumes of the so-called tigers, a code name for high-ranking arty members affected by the urge 19. Table 9 shows that both CYLC 20 and Shanghai Gang aear reresented in the urged samle 21 and, imortantly, both factions are reresented in shares roortional to their overall reresentation in the uer echelons of the CCP, and not statistically significantly higher or lower. The reader may however notice a lower, but not significant, reresentation of Shanghai Gang members, the faction most likely to be associated with Xi if at all see Section 8). These results aear also comletely consistent with an indeendent analysis of the anti-corrution camaign resented in Lu and Lorentzen 2016). v) Post-Deng era. Finally, we rovide brief emirical justification for our focus on the ost-deng era. Mao Zedong and Deng Xiaoing have often been characterized as olitical 19 As oosed to low-level oliticians, flies, involved in etty corrution. Tigers directly hit by the anticorrution urge have included retired PBSC member Zhou Yongkang and retired PB member Xu Caihou. 20 Links to the CYLC were evident in official news releases by The Peole s Daily which exlicitly singled out secific subsets of this faction, articularly The Shanxi Gang, officials linked to Ling Jihua, a disgraced rotégé of Hu Jintao. htt:// 21 We build a corrution dummy indicator for whether a olitical/military official is listed in the ublic anticorrution database of the Central Commission for Disciline nsection and from ChinaFile. Table 9 shows the cross-section regression of corrution dummy on faction affiliation of an official. The samle includes all individuals covered by China Vitae who have not retired by the year 2007, the year of 17th arty Congress. 14

15 strong men by many observers, as their legendary careers in the revolutionary era won them ultimate control over the military. n contrast, subsequent leaders, Jiang Zemin, Hu Jintao, and Xi Jining, aear categorically different: civilian officers who rose through the arty hierarchy relying on their ability and connections. This structural break is evident in the data. Underlying the symbolic retirement of Deng in 1989, we document structural changes in the whole sectrum of olitical elites. Figure 4 shows the share of ower score by factions or grous in the Central Committees of the CCP. Post-Deng China witnesses a significant decline in the influence of the military grou, and a rise in factions such as CYLC and Shanghai Gang. Figure 5 breaks down the ower score by four constituencies of the Central Committee: state organs, arty aaratus, military, and regional governments. The re- Deng era was ridden with volatile shifts across constituencies, with the most salient examle being the Cultural Revolution between 1966 and 1976, during which state organs and arty aaratus were virtually aralyzed. n contrast, the ost-deng era witnessed the stabilization of ower shares for each constituency. Desite the lack of olitical reform often alleged by outside observers, the above evidence suggests that Chinese olitics evolved to a new hase in which olitical strongmen became relaced by factional olitics after Deng 22. This is the eriod we focus on. 5 Model Having roduced a series of statistical regularities ointing in the direction of a systematic role for factional affiliation in the CCP, we roceed with the construction of an economic model useful to understanding the incentive structure driving the data in the ost-deng era. 5.1 The Hierarchy of Positions There is a L level hierarchy of leadershi ositions, ordered from the highest level 1, to the bottom, L. Each level, l, of the hierarchy has a M l /2 leadershi nodes. Each leadershi node has a air of leadershi ositions. The two ositions at each node are ordered osition 1 and osition 2). The hierarchy is broken u into regions, each of which nests a higher number of smaller regions below it. Level 1, the to level, has one node and hence two ositions; M 1 = 2. t is the aramount leadershi node for the country as a whole currently, President Xi Jining and Premier Li Keqiang). Level 2, the second layer in the hierarchy, has M 2 > M 1 ositions divided u into M 2 /2, and so on, with the number of ositions 22 Aendix Figure 3 shows additional evidence that age limits on Politburo members are strictly and systematically enforced in the ost-deng era, again another sign of a break toward institutional regularization. 15

16 strictly increasing down to level L. The nodes at the lowest level are the entry leadershi ositions, corresonding to the first ste in a olitical life that we model. Time is continuous. Each individual olitician dies or exogenously retires) with an instantaneous robability, δ, which also acts as the instantaneous discount rate. Uon a olitician s demise, his or her osition oens u for relacement. A olitician s osition also oens u when romoted to a osition above, freeing the current sot. We assume that the flow utility from being in office is increasing in the osition within the hierarchy. Denote by u l the instantaneous utility generated at any osition at level l, with l {1,..., L}, so that ul) > ul + 1). Positions within a level are ranked, but the utility flow difference is negligible and ignored; each osition at level l identically generates flow utility of ul) Factions There are two factions, denoted B Blue) and R Red), and the remaining individuals are neutrals, denoted N. 24 Factions exist to foster career rosects of affiliated cadres through a quid-ro-quo relationshi. On the way u the hierarchy, a faction member receives suort from senior faction members. n return, the faction member rovides suort to junior members for romotions. More faction members at higher levels enhances the overall faction ower, which benefits all faction members. This characterization of a faction aims at caturing, in a stylized fashion, the essential atron-client nature of such an organization, as emhasized in Nathan 1973). Factions are organized geograhically Li, 2013), in a way that mimics the allocation of ower ositions within the country. Faction members are designated by their region. A member at the to of the national government has a jurisdiction over the whole country, while a member at the to of a rovincial government has a jurisdiction over the corresonding rovince but not other rovinces. For examle, if a faction member is a rovincial leader in rovince P 1, he has jurisdiction over all the city- and village-level ositions in rovince P 1, but he does not have jurisdiction over any city- or village-level ositions in rovince P Promotions and Faction Suort When an oening arises in level l, the set of eligible candidates for a faction is first drawn from the faction members immediately junior to the oen osition, l + 1. f a faction has no members at the level l + 1, then they can ut forward a member from l + 2, and so on if l + 2 is also vacant. Such jums are occasionally referred as helicoter careers in China. The ordering of ositions at a node is not an imediment. A olitician at osition 2 in level 23 Formally: osition 1 generates µ 0 + extra utility relative to a osition 2 at all l in the same node. 24 The resence of more than two factions is easily incororated. Here, we maintain this assumtion only for exositional uroses and in line with the emirical analysis that follows. 16

17 l + 1 can be considered for romotion to osition 1 in level l equivalently to a olitician at 1 in level l + 1. When an oening arises, faction seniors who have jurisdiction over this osition must decide which, if any, faction candidates will receive their suort. For examle, if there is an oening in rovince P 1, only faction seniors in the same rovince and in the national government will be consulted, while faction seniors in other rovinces will not. Suort decisions are made sequentially by relevant faction seniors, roceeding through the hierarchy from the osition of the oening uwards. The decisions of higher ranked members suersede those of lower ones. This rocess continues u to the highest ranked member, each one being ratified or overturned) by successive stes on the way u. At the to, the final irrevocable, decision of factional suort is made. Note that faction seniors cannot choose who wins the romotion contest as the faction candidates comete with candidates from other factions. 25 However, a faction member not suorted by his faction seniors cannot win romotion to the osition rovided there are other suorted candidates, or Neutrals contesting.. 26 Ceteris aribus, each faction member receives a small utility benefit when a member of their faction is suorted for an oening. We denote this benefit by ɛ and assume that ɛ 0. The faction s restige is enhanced by romotions, and the members are accordingly leased, but this is assumed to be of no material or career benefit to members er se; unlike the substantial benefits that arise when leadershi falls to their faction modeled below). 27 We assume that at each decision level, overturning the decision received from a level below in the faction imoses a cost c > 0, on the decision maker. This cost can but need not) be small and always strictly exceeds ɛ; c > ɛ. ntuitively, we model here the fact that it is always easier to rubber stam a subordinate s decision and ush it u the hierarchy than to submit a reversed one. Otherwise, decision making, er se, imoses no costs. So, for an oening at level l agents at lower levels, l > l of the hierarchy including those eligible for the oen osition) have no inut into the faction s suort decision. We denote their inut by. The first inut into the factional suort decision arises for the co-node member at l, if there is one at this node. 28 This agent s decision is a vector of choices from within the interval [0, 1] alicable to each eligible candidate from the faction from the level below. So, for an oening at l there are l+1 t members of faction eligible. f ) the conode member is from faction then he chooses an l+1 t vector of values d l 1, d l 2,...d l where l+1 t 25 n Aendix, we rovide a descrition of the formal selection and aointment of arty and government leading CCP cadres that concurs with our setu. 26 f a faction does not suort a member, he could, in rincile, quit the faction and contest ositions as a neutral. We do not allow this, imlicitly assuming that the costs of doing this are rohibitive factions are like the mafia: able to severely unish eole who do not fulfill commitments. 27 We could alternatively assume that ɛ is realized only when a suorted member is actually successful in a romotion contest. This alternative assumtion has no consequence for the results. 28 Note that the co-node member of the oening is senior than cometing candidates. 17

18 d l i = x denotes suort romotion i with robability x and d l i = 0 denotes a request to stay i, i.e. a request to bide one s time until the senior cofactional himself is romoted. 29 The further restriction is that the sum of the suort allocations across all eligible members cannot exceed 1, though it can sum to less than 1. Denote this vector of choices by an agent d l 1, d l 2,...d l t l+1 ). The second factional inut regarding an oening at l in the at l by d ll faction s sequence of decisions falls to a faction member if there is one) in the lowest osition at level l 1. This agent observes the decision handed u to him from the decision maker at level l and similarly has an l+1 t dimensional vector of suort choices to make. Denote this decision by d l 1)l. f there is a faction member in the to osition at the node, his decision occurs next and is denoted d l 1)l. f there is no decision maker from the faction at a node oening, we shall denote the inut at that node by. From now on, we comactly use the notation d l j)l to denote the decision inutted into the factional decision over the oening at l by the faction members at level l j. That is, d l j)l d l j)l d l j)l if d l j)l if d l j)l if d l j)l =, and d l j)l =, and d l j)l Thus, for an oening at l the maximal set of inuts into factional decision making is the set of vectors from each of the L levels of the hierarchy ordered from lowest to highest: { },,...d ll, d l 1)l...d 1l. The last listed vector of decisions corresonding to the highest ranked member in the faction, determines the faction s actual suort action. This is denoted d 1l for faction in this examle, and is in general the final non-null entry in this sequence. f this decision involves a robability distribution over multile eligible candidates, the realized selection is then made by the faction according to that distribution. Thus, when the actual romotion contest takes lace within the arty described in the next section), only a single eligible candidate is ut forward to contest the oening by each faction. 29 Note that, in the event of there being no faction members eligible at level l + 1, all faction members would instead choose a vector of values over the eligible candidates that are drawn from level l + 2, t l+2, if there are none at this level, then at l + 3 and so on. For simlicity, we do not reflect this ossibility in the notation used for the analysis here as it has no substantive effect on the analysis. Moreover, it turns out that the factions are widesread enough in the data to ensure that the ossibility of no faction members at levels 3 onward does not arise in the data, and therefore does not require such additional notation. = 1) 18

19 5.4 Paramount Leadershi and Contests Once factions have decided on their suorted candidates, the arty s internal machinery determines the contest for each osition as follows. f both factions suort a candidate, and at least one neutral stands, then the robability of winning romotion for a faction candidate is given by the following contest function, W ): W ) = i β + ρ + η, 2) where i = β, if = B; i = ρ, if = R; i = η, if = N. β, ρ, and η are arameters determining the strength of faction members in the contest function. Additionally, having the aramount leadershi osition in a faction hels a romotion win for the faction s suorted candidate. Secifically, assume that if the aramount leader comes from faction B, we allow β l > β, and if from faction R we allow ρ l > ρ, thus incororating leadershi remia in the model. 30 Note that neutrals here are treated tantamount to a disorganized faction. The overall likelihood of a osition going to a neutral is unaffected by the number of neutrals contesting a osition, rovided there is at least one. Their total contest weight function is η. This treats neutrals symmetrically to factions and can be thought of as a roortional diluting of the neutral suort in the same way a faction s suort would be diluted were they to forward multile candidates instead of one. The contest function can be interreted as an endogenous outcome of inter-faction bargaining in similar sirit as Francois, Rainer, and Trebbi 2015). Essentially our current model can be nested in a more general setting in which factions first bargain for the weights in the contest function, then each olitician decides on its suort in its jurisdiction. n the inter-faction bargaining stage, the equilibrium outcome would favor the faction holding the aramount leadershi osition, but the threat of revolt gives other factions some weights in the romotion contest. The bargaining ower of each faction could otentially deend on the overall ower of the faction, which can exlain why faction members enjoy ɛ utility gain when they suort their junior in the second stage. 30 We allow for the ossibility of no factional advantage, which might be esecially likely at low levels of the hierarchy where the reach of the aramount leader could be muted. Note that it is also the case that a neutral s ascension to the aramount osition does not advantage neutrals down the hierarchy. 19

20 5.5 The State Variable n rincile, the suort decisions made by all faction members, and hence each olitician in the hierarchy s romotion robability at each oint in time, will deend on both their faction identity and the distribution of faction members across all other ositions in the hierarchy. Hence, the state variable of this system is the full distribution of ositions by faction. Denote this by S t at instant t. The state sace is thus a L l=1 M ll dimensional sace, with each dimension taking one of three values B, R, N. The state does not change if no osition oens u. However, each time an oening haens at a level l, then one individual will be romoted from l+1 to l to fill the oen osition, creating an oening at l+1 leading to one romotion from l + 2, and so on, until the bottom of the hierarchy L. Thus a single oening will lead to a cascade or, what we call, a chain of romotions. We assume that these chains occur immediately, and if at least one individual moving in a chain relaces an individual from a different faction, then S t changes Promotions and Factional Distributions The hierarchical structure of leadershi ositions needing to be filled within the CCP is taken as given and held constant over time. Thus romotions arise only to fill oenings occasioned by a death or retirement, or to fill a vacancy caused by a romotion that was itself initiated by a death or retirement. As already exlained, a single death can have many knock-on effects. At level 1, the instantaneous robability of an oening arising at any osition is δ. Since this is the highest level we observe, only death/retirement removes the to leader. However, the instantaneous robability of an oening arising at a ost at level 2 comrises the death hazard δ, lus the robability that there was an oening at level 1 and the individual at that level 2 ost ascended to level 1 to fill it. This robability of romotion can, in rincile, deend on both the factional affiliation of the individual at the ost at level 2 and the faction of the individual at the ost artnering the oening at level 1. Similarly, the instantaneous robability of an oening at a ost at level 3 is δ lus the robability that the individual at the ost at level 3 ascended to an oening at level 2 in the hierarchy, and so on. n the estimation Section 6 that follows, these knock on romotions, or romotion chains, will be exlicitly comuted. Let δ l S t ) denote the instantaneous arrival rate of romotions for an olitician at level l in state S t. Let J l St ) denote the robability that an faction member at level l is romoted to an oening aired with a J faction member at level l 1 in state S t, for, J = B, R, or N. Let l t denote the number of ositions held by faction at level l, at time t for = B, R, or N. Since oenings are not left vacant M l Rl t + Bt l + N l t. Since the instantaneous arrival rate of death is δ at any osition, there are, in exectation, l tδ deaths arriving at a osition aired with an at level l each instant, and M l δ at level l in general at 20

21 each instant. Let δ l St ) denote the instantaneous arrival rate of a romotion for a olitician sharing a node with an olitician in level l in state S t. Consider first the simlest case, which is a romotion from level 2 to the to of the hierarchy l = 1. Since there are, in exectation, t 1)δ oenings arriving for a osition aired with an due to a death, and since at level 1 there is no other way for an oening to arise, the instantaneous arrival of romotion for a from level 2 is: 31 δ 2 S t ) 3) = R t 1δ R 2S t ) + N t 1δ N 2S t ) + B t 1δ B 2S t ). We can now similarly comute the arrival of romotions from level 3 to level 2. As already noted, these arise when either a olitician at level 2 dies/retires, or is himself romoted to level 1, which in turn deends on a death/retirement at level 1, as secified in equation 3). Using these, we can comute the instantaneous arrival of romotions for an from level 3 at t as deending on the robability of a osition aired with an being romoted or dying. The instantaneous death arrival of such an individual is δ, the robability of the aired artner being romoted is δ 2 St ) for each of the factions at level 2 at time t, hence: δ 3 S t ) = R2 t R2 St ) ) R 3S t ) + N2 t +B2 t B2 St ) ) B 3S t ). N2 St ) ) N 3S t ) Similarly, continuing down the hierarchy, we have for any level l > 2: δ l S t ) 4) = Rl 1 t Rl 1 St ) ) R ls t ) + Nl 1 t Nl 1 St ) ) N ls t ) +Bl 1 t Bl 1 St ) ) B ls t ). 4) exlicitly shows that the arrival rate of romotions at level l deends not only on the distribution at level l 1, i.e. on Bl 1 t, Rt l 1, N l 1 t, but also, through each of the δ l 1, on Bl 2 t, Rt l 2, N l 2 t, and otentially via the faction member s decision strategies on other asects of the hierarchy as well. 5.6 Markov Strategies We take as given the entry decisions so that the factions are set. 32 Markov strategies at t are conditioned on the state at t, and ayoff relevant actions of layers receding a decision 31 This exression uses the fact that in continuous time simultaneous hazards do not arrive. That is, we ut zero weight on the robability of a death oening occurring at the same instant in two ositions. 32 We consider individual choices of faction entry subsequently. 21

22 maker s actions at t. The ayoff relevant actions are decisions members of factions have made regarding factional suort. Given the factional hierarchy sequence of decisions, the first decision arises for a faction member at the level of the oening. 33 With an oening at level l a Markov strategy for the co-node member at level l mas from the state S t to the l t dimensional dll vector of choices indicating stay at that level / suort romotion, i.e. a number in the interval [0, 1]) for all l t eligible co-faction members; with the additional restriction that the sum of these cannot exceed 1. Recall that the first suerscrit in this notation denotes the osition of the agent, and the second the oening of the osition that is being decided on. Denote the suort strategy of an faction member at l for an oening at l by σ ll St ) d ll. The suort strategy for a member at node l 1 mas from St and the observed decision of a faction member in the level below if there is one) d ll, to the same t l dimensional vector, and is denoted σ l 1)l ) S t, d ll l 1)l d. Similarly, the suort strategy for a member at node l 2 mas from S t and the decision of faction members in the levels below if there are such) d ll, ) dl 1)l to the l t dimensional vector of binary choices, and is denoted σ l 2)l S, d ll, dl 1)l d l 2)l. n general, for the jth removed factional decision maker, ) the Markov strategy is σ l j)l S, d ll,...dl j+1)l d l j)l. 34 Such Markov strategies are defined for all faction members at the level of the oening and above, and can be denoted for a faction member of at level l deciding on suort at levels l to l + h by σ l σ ll where l + h L, and σ ll+h), σl)l+1 σ ll+2) S, d l+1)l+h)...σ ll+h) σ ll+1) S, d l+1)l+2), d l+2)l+h) σ ll S, d l+1)l+1), d l+2)l+2) )...d l+h)l+h) S) ) d ll )., d ll+1), d ll+2), d ll+h). 33 Though there is an ordering of decisions within the faction inversely corresonding to seniority we assume that these are all undertaken instantaneously, so that the factional suort decision for an oening at t arrives immediately. 34 Note that the strategies do not condition a member s suort on the suort decisions of members of the other faction, which are assumed to be unobserved. Note also that the notation is slightly abused in the case of a decision maker from a faction at osition 1 in level l j who is following on from a decision handed u from a co-faction member at osition 2 in level l j. n such a case, the decision maker s strategy mas from the decision of his nodal subordinate s decision d l j)l ) and should be more correctly written as σ l j)l S, d ll,...dl j)l d l j)l, which again, for simlicity we do not bother exlicitly stating in the text. 22

23 5.7 Definition of Equilibrium We characterize a Markov Perfect Equilibrium. This is an equilibrium in which all layers are laying best resonses to strategies that satisfy the Markov restrictions, and these best resonses are themselves Markov Strategies. n equilibrium, layers maximize exected utility by choice of suort strategy for any oening that may arise. For each layer in faction at level l these strategies secify a suort decision for all ositions l, l L, subordinate to the layer. Define σ l+h)+ as the suort decision sent u to the individual at l concerning a osition at l + h by the immediate factional decision maker below l; that is the decision sent u from below over which the member at l must decide to ratify or overturn. ɛ is already defined as the utility gain of suorting a co-factional, c as the utility loss of overturning the suort decision made by a co-faction member received from the level below. ndividuals choose σ l to maximize net resent value of exected discounted flow utility from olitical office. The value function can be written as: V l, S) 5) = u l) + max σ l ) c 1 {σ l+h)+ &σ ll+h) σ l+h)+ } + ɛ 1 {σ ll+h) =1} [ L l h=1 +1 δ)dte [ δ l S σ l)) V l 1, S σ l)) + 1 δ l S σ l))) V l, S σ l))] The value function is interreted as follows. The agent discounts by the death robability over the interval dt, robability δ. A romotion, robability δ l S), leads to a level transition. The exectations oerator is taken over the state that will revail after the interval, dt, S. The state will otentially affect both romotions from a given l and value functions at the level. The suort strategy, σ l has no contemoraneous effect on ul) but can otentially affect the realization of the state, and so in turn can affect value functions via effects on romotions and future valuations. PROPOSTON 1 : The Unique, Symmetric, Markov Perfect Equilibrium features: 1) All members of both factions ratify the suort decisions of co-faction subordinates below them in the hierarchy. 2) For members making the first suort decision within their faction for an oening there are two ossibilities. 2.a) The oening is aired with them at their node of the hierarchy. n this case, they ask any co-faction members below them to stay at their levels and wait for them to be romoted first. 2.b) The oening is at a node below theirs. n this case they allocate factional suort for the romotion of one randomly selected) candidate from the set of eligible candidates within their faction for the oen osition. ] 23

24 3) The leader always suorts co-faction members in a romotion contest for a osition at the to node. A formal reresentation of the strategies is reorted in Aendix. Proosition 1 shows that factions differentiate strongly in how they allocate suort, deending on the location of their faction members. f a faction member is not resent at the node to which a junior member is asiring, then the senior faction member unanimously suorts a junior member s contest for the osition if there are multile juniors, they suort one, allocating suort randomly). However, senior faction members issue staying orders to subordinates from their own faction who wish to contest an oening at their own node. Moreover, faction members above them ratify this decision. This attern of equilibrium decisions leads to a faction effectively asking junior members to wait for his/her immediate suerior to be romoted first, before allowing them to move u a level. This attern of romotions, which tethers a junior s rise to the one of his immediate suerior, is the unique Markov erfect equilibrium of the decentralized, hierarchical factional decision making. t is chosen indeendently by all faction members, as it maximizes each member s exected flow utility. t is worth noting that the uniqueness of the symmetric equilibrium alies only because of the restriction to Markov strategies. t is, in rincile ossible to construct non-markovian subgame erfect equilibria which may instead feature faction members conditioning their romotion suort decisions on the revious suort decisions of other agents within the hierarchy; secifically on whether agents suorted them in the ast. Such non-markovian SGPE allow for agents within a faction to otentially make mutually imroving side deals suorted via, for examle, trigger strategies. Though such strategies may be ongoing within sub-grous of factions, it is reasonable for the more formalized faction oerations to have relatively simle non history-deendent) rules. A factor making these mutual suort contracts difficult to imlement in ractice, and favoring history indeendent Markov strategies at the level of the faction as a whole as we have roceeded), is that romotion decisions are made behind the scenes and not subject to ublic deliberation or comment. Given this, the suort decision of a suerior is virtually unobservable to a articular candidate and, since the final outcome observed success or failure at romotion) deends on more than the suort decision of any one faction or member, it may be difficult to sustain these side deals. The result in Proosition 1 generates a large amount of structure to the attern of oenings imlying that no two cofactional members will ever be aired at the same node. We have already verified in Section 4 that this is, in fact, a systematic feature of the data. Moreover, the rosects of romotion at any node deend not only on the distribution of oenings at the level immediately above, but also on the distribution of oenings further u the hierarchy, as these determine the chances that a olitician immediately above will himself be romoted. Promotion chances at all levels are affected by the full distribution of ositions 24

25 above. We can comute this exlicitly using the recursive structure of the δ l S) terms and our results on factional suort. When each faction suorts one of their members for an oen osition, the exlicit form of the δ l S) using the contest function 2) is: δ l 1 t l i if K i+k+k l t, K t l > 0, i if K i+k l t > 0, K t l = 0, i if K i+k t l = 0, K t l > 0, 1 if Kl t, K t l = 0, where { t, i} = {B t, β}, {R t, ρ} or {N t, η}; {K t, k} = { t, i}, {K t, k }; {K t, k } = { t, i}. Since the roosition allows us to determine when a faction will suort one of their members for vacant oenings and not, we are now able to comute exlicitly the instantaneous arrival rate of romotions at all levels based on the numbers of faction members resent at each level. PROPOSTON 2 The instantaneous arrival rate of romotions at each level of the hierarchy is as follows. Let t B = 1, iff Bt l > 0 and t B = 0, otherwise; t R = 1, iff Rt l > 0 and t R = 0, otherwise; t N = 1, iff N l t > 0 and t N = 0, otherwise. For an N member: ) δ Nl = η R t Rl 1 Nl t l 1 t B β + η + N l 1 t Nl 1 t B β + t R ρ + η + t Bt Bl 1 l 1 t R ρ + η. 6) For a B member and R member, resectively: δ Bl = β R t Rl 1 Bl t l 1 β + t N η + N l 1 t ) Nl 1 β + t R ρ + t N η, δ Rl = ρ R t l Bl 1 t Bl 1 ρ + t N η + N l 1 t Nl 1 t B β + ρ + t N η ). 25

26 For each one of these exressions we can see the negative deendence on the revalence of one s own faction members. 35 Take for examle the last exression for R. The greater the number of other R s at level l at t, the more diluted is an R s suort i.e. the lower the robability that any given R member will be chosen by the faction as the one to be suorted), as er Rl t in the denominator. Further, the more frequent the R s at level l 1 the harder it is to get an oening for which an R at l will be suorted by his faction e.g. at the extreme if Rl 1 t = M l 1 t, then δt Rl = 0). This is true for all levels of the hierarchy from the recursion of these equations. The roosition highlights the ossible down side of factional affiliation. Though factions have the otential to rovide suort for romotions such suort is essentially decided by cofactional members sitting above one in the hierarchy, if there is one. The rise of a junior is tethered to, and thus deends uon, the rise of his cofactional seniors. f they do not rise, then, not only do they not generate the extra suort that comes from the aramount leadershi, they imose a delay on their own juniors, as factions ensure subordinates do not overtake seniors. f a junior asires to a osition with a similar rank to his factional senior, the junior will be asked to bide his time until the senior gets romoted. This theoretical feature will be useful to match slow-moving dynamics of factional evolution of the CCP hierarchy over time, consistently with data where factional reresentation changes slowly and olitical cohorts move u in lockste. n addition, the roosition formalizes a natural ecking order in the factional structure, a feature that aears realistic in large organizations and is generated as an equilibrium outcome of decentralized individual decision making within the faction due to the career concerns of sueriors. The Confucian ethos and the secific norms of guanxi remarked uon within CCP olitics aear culturally consistent with this result. 5.8 Entry Entry into the hierarchy of olitical ositions occurs only at the lowest level, L. An entering olitician at instant t decides which faction to join when starting his olitician career, or to contest as a neutral, and bases this decision on the discounted exected utility he will receive via each one of the otions. He maximizes his discounted exected utility stream: V t = t e δs υ s ds where υ τ is the instantaneous utility at τ. We formally consider this decision here. Recall that ul) denotes the olitician er instant ayoff to holding a osition at level l {1, L} in 35 Note that each statement of δ l in Proosition 2 ignores the effect of a faction s holding of the aramount leadershi on romotion. Effectively δ l is stated for the case of an N in aramount leadershi. n the Aendix we state the full set of δ l conditional uon aramount leadershi affiliation. 26

27 the hierarchy. So that if a olitician holds a osition at l at instant t then υ t = ul). 36 We have already defined the corresonding value function for a olitician of tye = B, R, N at level l given state S in exression 5). By substituting for the equilibrium strategies σ l these can be written simly as: V l, S) = u l) + 1 δ)dte [δ l S ) V l 1, S ) + 1 δ l S )) V l, S )] 7) The exectations oerator aears in the exression because the value of being a tye olitician at l deends on the instantaneous robability of being romoted to the next level and getting V l 1, S ). Though this is known at the current state, via δ l, the value of being at this higher level in turn deends on the evolution of the state, S, which is changing continuously in a stochastic manner due to deaths, oenings, and romotions occurring through time via the contest function 2). The entering olitician at state S chooses the faction with the highest valuation at the entry level, L, i.e.: su {V B L, S), V R L, S), V N L, S)}. 8) After entry, since a olitician is fixed in his faction from then on, his choices are simle. He will aly for all romotions to which he is eligible, and he will choose suort decisions according to the Proosition above. We consider the more difficult roblem of the initial entry decision 8) now. 5.9 Equilibrium Entry Entering oliticians will choose to enter the faction or remain neutral) yielding the highest exected utility, which imlies choosing the faction guaranteeing in exectation the fastest rogression through the hierarchy. The most immediately relevant information for the agent will be the arrival of romotions if he/she registers as a olitician from level L to L 1, but one cannot secify, a riori, the relative weight an entering olitician uts on the chances of being romoted at higher levels of the hierarchy comared to lower levels. Perhas oliticians care little about regional romotions, that occur early in their career, but greatly about romotions from the rovince to the central government. Conversely, oliticians may ut substantial value on their immediate entry rosects. Note that, indirectly at least, the relative erformance of factions at higher levels already enters into a olitician s evaluation of romotion at the lowest level, L, since oenings immediately above deend negatively on the frequency of cofactional oliticians all the way u the hierarchy; as discussed above after Proosition 2. At any oint in time this valuation will deend on the full distribution of ositions higher than the olitician, that is on S t, the high-dimensionality state sace of the 36 Since c is never realized on equilibrium aths it is ignored, as is ɛwhich though realized is arbitrarily small. 27

28 system. Without maing the full form of exected hierarchy evolution, it is not ossible to comute value functions exlicitly. However, it is ossible to establish a sufficient condition under which otimal entry ensures that along any time ath all factional tyes and neutrals will be observed in equilibrium: PROPOSTON 3 With M l large enough for all l, any equilibrium necessarily involves oliticians in factions B, R, and N. ntuitively, with sufficiently many oenings at all levels of the hierarchy, the value of entering via a faction or as a neutral) that is not already resent will eventually outweigh even the largest arametric disadvantages of that faction or being a neutral). That is, for examle, even if β ρ so ceteris aribus it is better to enter as an R than a B), if there are sufficiently many ositions in the hierarchy, a large number of R members and Proosition 1 will imly that the exected romotion rate will be faster if entering as a rare) B member over entering as one of the many) R. Thus, though we are not able to fully characterize otimal entry in an equilibrium, the sufficient condition of the roosition ensures that any equilibrium distribution of ositions that we do observe will feature both factions and neutrals From Model to Data Oenings in the hierarchy occur at any oint in time via the functions in Proosition 2. Other than through the effect of time on the changing distribution of factions across the hierarchy S t, which the model exlicitly accounts for, the rocess leading to oenings occurs indeendently of time conditional on S t ) 37. Treating oenings this way amounts to assuming that oenings are indeendent events caused by exogenous factors, each triggering a chain of knock on effects. This assumtion may be violated at the time of Chinese Communist Party Congresses, when there aears to be a large number of shuffles at different levels of the hierarchy observed in a way that seems simultaneous, not sequential. ndeed, for the most art, the data is observed at low frequency, i.e. at each CCP Congress T, T + 1,... This imlies that the romotion chains that our model ostulates are not fully observable, so simulation methods will be necessary to link two subsequent S T,S T +1. To oerationalize the model in our secific emirical setting, we will assume that the simultaneity observed in exits and romotions reflects a articular structure, as follows. First, we urge all individuals from all ositions that we observe leaving the data in between snashots T, T + 1,... That is, all individuals who are no longer resent between times T and T + 1 are assumed to have retired at some oint between two Congresses. 37 n what follows below we will disense with the time index t for the emirical analysis. 28

29 Second, oenings are filled through a sequence of romotion chains. Each chain starts with the highest ranked exit in the samle and selects oliticians to fill in the knock-on oenings sequentially. This continues until all the exits and romotions between S T and S T +1 are accounted for and all ositions have been filled. Because there are many sets of romotion chains that can rationalize the observed oenings in the data, Section 6 shows how simulation methods can be used to transarently address this issue in ractice. n addition, it is difficult to observe the exlicit dyadic structure for many ositions in the data. Therefore, we draw at random a aired olitician from the set of otential matches at the level at which the romotion occurs. 6 Maximum Simulated Likelihood Estimation This Section describes our estimation methodology. Define Y the observed data on career outcomes i.e. romotions, exits, etc.) between two Congresses T and T + 1 and X the observed data on the hierarchy lus a set of individual characteristics i.e. X includes factions and osition within the hierarchy/level S T, lus individual covariates). We define k as a set of romotion chains, so that k = {k1), k2),...}, where each chain kc) of oliticians say, s 0, s, s,and s ) is simly a set of oliticians each belonging to different, but adjacent hierarchical levels l, whose romotions were triggered by the exit of the highest ranking one of the chain e.g. when s 0 dies or retires at l = 1, s is romoted from l = 2 to l = 1, then s relaces s at level l = 2, and then s relaces s at l = 3). A chain starts from an oening at level l 1 and involves romotions from l all the way down to L. 38 We imose that each olitician romoted in the data belongs to exactly one chain and that each change between S T and S T +1 is art of at least one chain kc). A olitician romoted by two levels between T and T + 1 will need to belong to two searate romotion chains.) Let C = # k) be the number of romotion chains in set k. The unconditional likelihood of observing Y given X is: fy X) = E k [fy X, k)]. Define Y kc) as the set of career outcomes ertinent to the individuals involved in romotion chain number c of k. Because the structure of the olitical hierarchy will change once a romotion chain is realized i.e. the interim S will change), ositions within the hierarchy/level and factional affiliations at all levels X kc) need to be modified after each chain kc) is realized. 38 Plus a new entry at the lowest level, which we do not model, as er our discussion of Proosition 3. The entry choice is not necessary for estimation and all arameters are identified without its addition. 29

30 The conditional likelihood uon the realization of a set of romotion chains k haening over time is given by: fy X, k) = C fy kc) X kc), k). c=1 The likelihood contribution fy kc) X kc), k) of a chain kc) of romotions initiated at l 1 involves comuting the conditional romotion robabilities of all individuals involved in kc) at the various levels, down to L. A romotion from level l to level l 1 to be aired to a olitician K = R, B, N is a random event distributed over a discrete suort formed of M l oints individual oliticians), B l of which occur with robability K Bl, R l occur with robability K Rl, and N l occur with robability K Nl. We omit state indexes as they are unnecessary here.) Given the indeendence of the romotion events across levels, the construction of this likelihood is straightforward. Let l be the faction of the individual belonging to kc) at level l and J l 1 be the faction of the individual with which s/he is aired when romoted to level l 1: fy kc) X kc), k) = δ L l=l J l 1 l l). Going back to the examle above of a chain of oliticians s 0, s, s,and s belonging to factions N, R, and B resectively, and assuming they all haen to get aired with N-tye oliticians, the likelihood contribution of this chain is: fy kc) X kc), k) = δ N N 2) N R 3) N B 4) where each robability J l 1 l) is comuted based on X l kc), ordered from the to romotion to the level L romotion, as imosed by the sequential nature of the romotions comrised in each chain. The Maximum Simulated Likelihood MSL), for given number of simulated sets of romotion chains R K, 39 is: fy X) = 1 R K This is the estimator that we emloy. R K r=1 c=1 C fy krc) X krc), k r ). 39 We emloy 100 simulated chains sets for each CCP National Party Congress. 30

31 7 CCP Factional Politics: Structural Results This section resents MSL estimates of the model and samle fit assessments. The samle includes all the members of the 14th-18th Central Committees in the ost-deng era. The simulation rocedure in Section 6 was first imlemented in a series of Monte Carlo simulations and successfully robed for: i) identification of the structural arameters; ii) sensitivity to missecification in the number of factions; and iii) sensitivity to missecification in the contest function we use 40. We begin our analysis with the most arametrically arsimonious model ossible, one where we normalize η = 1 and the two faction arameters {β, ρ} are estimated on to of a single leadershi remium λ, defined as λ = β l /β = ρ l /ρ. The MSL results for this model are reorted in Column 1 of Table 10. The estimated contest function arameters are and for CYLC and Shanghai Gang resectively, which are close to the average share of seats in the Central Committee. The estimated leadershi remium λ is 2.553, imlying that a faction candidate is more than twice as likely to be romoted when the aramount leader is from the same faction. The magnitude of the leadershi remium is consistent with the reduced form evidence in Table 4. All arameters driving the romotion rocess across factions are recisely estimated. Because it may seem restrictive to assume a common contest function across all levels of the CCP to echelons which include heterogeneous layers in both size and jurisdiction, such as the to CCP ositions and the PBSC, PB, CC, AC), Column 2 in Table 10 allows for level-secific arameters {β k, ρ k } k=h,l for the PB and higher versus CC and lower. The arameter estimates show that faction affiliation hels significantly more at higher levels than that at lower levels within the CCP: the estimated contest function arameters reach and at the PB and higher for CYLC and Shanghai Gang relative to CC and AC levels of and One may also wonder whether the leadershi remium differs across factions. Column 3 of Table 10 exlores this ossibility by allowing for faction-secific leadershi remia {λ R, λ B }. The arameter estimates show that two factions have very similar remia both are between 2 and 3). The imrovement of log-likelihood is negligible, indicating that the two factions oerate in a similar fashion. This result is also consistent with the reduced-form evidence in Table 4. Column 4 in Table 10 combines both level-secific arameters {β k, ρ k } k=h,l and factionsecific leadershi remia {λ R, λ B }. We conduct Likelihood Ratio LR) tests for model 1, 2 and 3 against model 4 numbering indicates the Column of reference). LR tests reject model 1 and 3, which imose a constant contest function across levels, against model 4, but do not reject model 2, which imoses a constant leadershi remium across factions. n the following analysis, we will thus use the more arsimonious model 2 as our benchmark and 40 All simulation results are available uon request. 31

32 refer to it as the baseline faction model. Figure 6 rovides a visual reresentation of the factions seat shares by level as redicted by the model. The five bars reresent the five levels of the Central Committee the to two CCP ositions, PBSC, PB, CC, and AC). The blue, white, and red arts reresent the seat shares of the Shanghai Gang, Neutral, and CYLC resectively. The left anel is the data, while the right are the redictions of our baseline faction model. Our baseline faction model successfully relicates the distribution of factions across different levels of the hierarchy: faction members are relatively scarce in the lower levels, but become increasingly concentrated in the higher ones. This is related to the increasing contest function arameters estimated above, which imly an increasing advantage of factional affiliation as one rogresses u the hierarchy. Notice that our model also catures the inertia of the factional comosition of the various levels over time evident in the data thanks to the slow ercolation of factional members u the hierarchy. The intuition is that romotions and retirements occur gradually over time. t takes time for a faction leader to grow his inner circle from the bottom of the hierarchy u. nterestingly, such dynamics can function as checks and balances on an incoming aramount leader. When a new leader first assumes ower, he is likely to be surrounded by members from rival factions. There is also anecdotal evidence in line with this finding: Jiang himself once described his first few years as the General Secretary as standing on the brink of a dee ravine, or walking on thin ice 41. Bo 2004) also suggests that the Shanghai Gang continued to exert strong influence in the first term of Hu Jintao. This finding is articularly useful in understanding the ucoming second term of General Secretary Xi, exected by many observers to gain greater clout relative to his first term in office 42. Our faction model also rovides insights for the dynamics of ower transition between factions. Figure 7 lots the aggregate share of romotions of each faction over time 43. The share of romotions is defined as the ratio between the number of romotions for a faction and the total number of romotions. Again, the fit of the model is good. Figure 7 oints also to a more subtle imlication of our model: there are no discontinuous dros in the share of romotions of the aramount leader s faction right after he retires. When Jiang Zemin retired after the 15th Party Congress, a large share of the Shanghai Gang continued to be romoted to the 16th Central Committee. The attern was reeated at Hu Jintao s transition to Xi Jining at the 18th Party Congress. We will show below that the same holds true for the 19th Congress Politburo. n reality there is uncertainty over the recise oint at which the influence of the incoming aramount leader eclises that of the dearting incumbent and this influences romotion rates. Scholars have suggested that Deng retained considerable influence well after formal retirement in 1989; Jiang maintained informal and 41 See Kuhn 2005). 42 The 19th Congress occurred between October 18th and 25th, A more detailed breakdown by level of the Central Committee can be found in Table

33 formal military oversight after steing down as General Secretary. A retiring aramount leader may continue to shae the comosition of the next Central Committee. Such intricate dynamics are catured by our simulation aroach that draws different aramount leader transition dates across multile simulations, smoothing out shar discontinuities around the official ower transition date. 7.1 Adding ndividual Covariates So far we have assumed that faction members are selected to challenge a ost randomly within a faction and level modulo stay orders, of course). We can easily add individual characteristics, Z, to the within-faction selection rocess as well. Consider each row of the matrix Z s to be a vector of characteristics for olitician s. Define q,s l) as the robability that s of faction is selected as the candidate of this faction at level l, also define A l) as the set of the members of faction at level l. We assume a within-faction selection robability of the logistic form 44 : q,s l) ex γz s ) s A l) ex γz s ). Therefore, the robability of winning romotion can be rewritten as q,s l) W ). Notice that our baseline faction model is nested in this formula by setting coefficients of individual characteristics, γ, to 0. n this case we get back our random within-faction selection robability, q,s l) 1. We refer to the above model as the faction model with individual l) characteristics. The arameter estimates are reorted in the Column 4 of Table 12. Comaring with the baseline faction model in Column 1, we see a reasonable imrovement in model fit measured by log-likelihood. At the same time, however, we observe little change in the estimates of the arameters for the contest function and the leadershi remium, suggesting that these arameters are indeed more related to the technology of factions than to individual covariates omitted in the baseline model. Examining the estimated coefficients of individual characteristics, we find that being a Princeling or a male increases the robability of romotion, while having a graduate degree or being an ethnic minority hurts. The effect of age is non-linear: it has a ositive effect at first, but eventually negatively affects romotion chances, in line with reviously observed hard age limits enforced within the CCP. 44 Since our data only includes the to 5 levels of the arty hierarchy President/Premier, PBSC, PB, CC, AC), individual characteristics of the otential candidates eligible for romotion to AC are not always observable to us. As a result, we assume within-faction selection is random below the AC level. 33

34 7.2 Alternative Models Given our main secifications, we are equied for both in-samle and out-of-samle fit analysis of our structural model. t is useful in this resect also to resent some alternative benchmarks to which we can comare our model s erformance. First, we can use as the simlest alternative a model based on random romotion. This is done by setting: l 1 M l. Second, we imlement a ure seniority-based romotion mechanism, setting for olitician s: sl ϕage s) M l, with ϕ.) a third order) olynomial in age 45. Figure 8 rovides the scatter lots of model redicted shares of romotions by Party Congress and by level of the CCP against the data 46. Our models baseline faction and faction with individual characteristics) handily outerform both the random and the seniority models: the redicted shares by the faction models line u with the data nicely along the 45 degree line, whereas the shares redicted by random and ure seniority-based romotion models aear comletely flat. This result is indeendent of which Party Congress we consider. More concretely, with only five structural arameters our baseline faction model reduces the mean squared errors of the redicted shares of romotion by more than 80 ercent comaring to the random model, and more than 70 comaring to the seniority model. The right anel of Table 12 conducts formal secification tests. The Vuong statistics reject the random and the ure seniority-based romotion models against the faction model with individual characteristics 47. What about olitical meritocracy? Because the CCP romotion model is by many referred to as a strictly meritocratic mechanism Li and Zhou, 2005; Bell, 2015) and there is substantial debate as to whether systematic assessment of cadres based on economic erformance lays a role in the CCP, we test our model against this third urely meritocratic mechanism. First, in order to find a suitable measure of erformance, we need to restrict our analysis of romotions to rovincial leaders in the Central Committee. We associate these rominent rovincial CCP cadres with the economic erformance in terms of real GDP 45 For the seniority and random model, we calibrate the robability of entering AC using the average share of each faction in the Central Committee. 46 We combine level 1, 2 and 3 because there are two few observations in the first two levels. 47 The ure seniority-based model outerforms the baseline faction model in terms of log-likelihood. However, this is driven by the fact that only 10% of the oliticians have factional affiliation. After we include individual characteristics in the factional model, the ure seniority-based model is easily rejected by the Vuong test. 34

35 growth over their tenure) of their Province of service recisely as in Li and Zhou 2005) and Jia et al. 2015) and use this as a admittedly rough) roxy for overall erformance. Grahical evidence of the erformance of our model is reorted in Figure n the aroriate subset of romotions i.e. those for which erformance metrics are available), our model erforms better than the urely meritocratic model in terms of sum of mean squared errors, which is reduced by 35 ercent. We further examine the out-of-samle fit of our model. Secifically, we re-estimate the model using only the 14th-17th Central Committees and redict the shares of romotion of each faction at 18th Central Committee. We comare the model redicted share of romotion with the actual data in the scatter lot in Figure 10. Our faction model again outerforms random and seniority-based models in terms of out-of-samle fit. The reduction in mean squared errors is 77 ercent and 69 ercent comaring our baseline faction model to the random model and seniority model resectively, reassuring us of its robustness. 8 Counterfactuals and Model Analysis Within our econometric framework we can exlore a set of counterfactual exercises and resent an additional quantitative analysis of several questions relevant to the study of Chinese olitical economy. 8.1 Forgoing Collective Leadershi We begin by exloring an historical counterfactual on leadershi remia in the ost-deng era. Our model exlicitly recognizes such remia, but a wealth of anecdotal discussion in Chinese olitics and the emirical evidence of Section 4) suggests them to have been curtailed in the ost-deng era. This eculiarity of the ost-deng Chinese system, the emergence of socalled collective leadershi, has been frequently recognized in the literature. t is often indicated as the main structural break from the strongman olitical equilibria thought to have revailed under Mao Zedong 49 and the aramount leadershi of Deng Xiaoing Tsou, 1995; Fewsmith, 2001; Shambaugh, 2008). This exercise is also useful in ersective of the resent, latent changes in Chinese olitics. Scholars like Nathan 2016) suggest President Xi may be overturning Deng s system, as he has taken the chairmanshi of the most imortant seven of the twenty-two leading small grous that guide olicy in secific areas and tightened direct control over the military. 48 n this scatter lot, we do not break down the share of romotion by level because of the small number of observations in the subset of rovincial leaders. 49 During the Maoist era, factions were ideologically as well as ersonally defined, and remained fiercely loyal in what could become a winner-take-all game. Dittmer 2004,.18) 35

36 Here, we will ram u the limited role layed by leadershi remia in factional reresentation in China and resent a counterfactual of what would have haened under heightened winner-take-all tye factional cometition. We run the model with twice as high a leadershi remium λ. Results are reorted in Figure 10. A more detailed breakdown by level can be found in the third anel of Table 11. The counterfactual is imlemented by simulating for each Congress T the share of romotion of each faction to the following Congress T + 1. Under the Jiang Zemin Shanghai Gang) residency, oenings in the Politburo and the Central Committee are filled with more of the to leader s cofactionals. Under the CYLC leadershi of Hu Jintao, numbers would have been comarable, swinging in the oosite direction with more tuanai members romoted. The magnitude of the increase in the shares of romotions, however, is less than the increase in the leadershi remium. The damening effect emerges from the nature of intra-factional cometition, and the factions requiring members to resect factional seniority in romotion, as er Proosition 1. As members of a faction become crowded at a certain level l, junior members from the same faction are more likely to be required by their cofactionals to hold off on ascending to that level until incumbents suerior to them have moved uwards. ndeed, as shown in Table 11, the damening effect is stronger in higher levels of the hierarchy where faction members are more concentrated Figure 6). Therefore, individual incentives in intra-faction cometition counterweigh the caacity of the aramount leader to grow his own faction without bound. 8.2 Li Keqiang Presidency A second counterfactual we erform involves the choice of leadershi ratified by the th Party Congress. This is the event that brought Xi Jining to the PRC Presidency. Nathan and Gilley 2003) resent comelling documentary evidence that already ten years before the formal ower transition to Xi Jining and Li Keqiang, the current PRC Premier, belonged to a select few with otential accreditation to the aramount ost. Bo Xilai was also art of this highly selected grou. t is ossible for us to study a counterfactual Li residency. Figure 11 reorts the aggregate share of romotion, and a more detailed breakdown by level can be found in the fourth anel of Table 11. nterestingly, given the estimated leadershi remia, the romotion at PB level would have had a very limited increase in CYLC reresentation Li s faction). More radical shifts would have been recorded in the romotion at the CC and AC though. Again this is a result of the slow ercolation of factional reresentation induced by our model, comounded with the already high CYLC reresentation at the uer levels of the CCP at the end of Hu s last term in office. 36

37 8.3 Are Princelings a Faction? The reader will notice that the analysis above osits factional affiliation of resident Xi Jining as a member of Shanghai Gang. This is in itself a matter of debate among scholars interested in Chinese elite olitics. For instance, Li 2013) in his bi-factional reresentation of the Chinese to tiers defines Xi as a Princeling associated with Jiang s cam Shanghai Gang). n fact, Xi sent only seven months in any official role in Shanghai, but Jiang s substantial influence on Xi has been noted by many. Other researchers have ointed to President Xi as the leader of a new faction of his own, mostly with roots in Shaanxi, where Xi was born, and in Zhejiang Province, where he served as Party Secretary from 2002 to Our model allows a formal statistical analysis of some of these questions. We begin by investigating whether our ostulate of the Princelings not behaving as a unified faction is warranted by the data. To assess this formally we imlement Vuong secification tests between our baseline model and one where Princeling status is coded as membershi in faction P, with a secific arameter π regulating an exanded contest function of the tye 2): π W P ) = β + π + ρ + η. We also secify a faction-secific leadershi remium, λ = π l /π, which regulates the differential romotion robability when the aramount leader is from the Princelings e.g. Xi in the 18th Party Congress). Results are reorted in Table 13. The Vuong test indicates that the model where Princelings are considered to be neutrals is referred over one where Princelings are treated as a searate faction. More imortantly, the estimated leadershi remium within the model imosing Princelings as a faction, λ, is estimated to be less than 1. This means that if we osit that Xi ascended to the aramount osition as a member of a Princeling faction, other Princelings did not aear to enjoy a higher remium in romotions. This finding rima facie violates one of the crucial features of factional olitics delivering resources to members of the faction once the faction leader is in ower and aears in stark contrast to what we have already observed for the broadly acceted factions, CYLC and Shanghai Gang, where we estimate λ well in excess of 1. n brief, the evidence rejects the hyothesis that Princelings oerate as a unified faction. 50 Some recurring affiliated oliticians include current PBSC member and anti-corrution czar Wang Qishan, and otential PBSC future members such as Li Zhanshu, director of the CC General Office, and Politburo member Zhao Leji. Shih 2016) estimates, based on shared career exerience, that less than 6 ercent of current CC members have ast ties with President Xi. This should however not be confused with a truly factional organization of the President s inner circle for which hard evidence is not available. 37

38 8.4 s President Xi Jining Affiliated to the Shanghai Gang? Our structural aroach also allows us to undertake formal tests for the analysis of factional affiliation of the to leadershi. The case of Xi Jining is emblematic because of both his strong ties to the CCP elite through family connections and his reeated rejection of intraarty factional olitics e.g. cabals and cliques mentioned in official transcrits on Peole s Daily, May 3rd, ). To this goal, we re-estimate the model assuming that Xi is an unaffiliated neutral, and comare the alternative model against our baseline secification where Xi is a Shanghai Gang member. Results are reorted in Table 13. The Vuong test shows that Xi is slightly more likely to be a Shanghai Gang member, although the statistical evidence is inconclusive. Our tests do not have enough ower in this secific instance. Fortunately, such ambiguity is likely to be resolved after the 19th Party Congress data collection is comleted. 8.5 An Out-of-samle Forecast for the th Party Congress To conclude our quantitative exercises we emloy our model to forecast the 19th Party Congress in Although admittedly seculative, to the best of our knowledge this is robably one of the very few rigorous quantitative environments allowing for an exercise of this kind. The model incororates individual characteristics in this analysis to obtain more accurate forecasts 52. The to anel of Table 14 shows that share of romotions by level of the Central Committee. Under the assumtion that Xi is in fact a Shanghai Gang member, the Shanghai faction is exected to enjoy a higher share of romotions in the Politburo than the CYLC faction due to leadershi remia. n contrast, romotions at lower levels are exected to be more comarable between the two factions due to the damening effects stemming from stay orders. Since there is still unresolved ambiguity regarding Xi s factional affiliation, we also conduct a forecast assuming Xi is a neutral in the bottom anel of Table 14. n this case, the Shanghai Gang would aear to lose its advantage in romotion for all the levels of the Central Committee. Using comlete biograhical information for the 19th Congress Politburo we can assess the recision in these forecasts at this olitical level. The redicted B; N; R) romotion shares allocated by the model under the assumtion of Xi as a Shanghai member are 0.24; 0.66; 0.10). Under the assumtion of Xi as a neutral, the model redicted shares are 0.14; 0.76; 0.10). The emirically observed romotion shares are 0.25; 0.65; 0.10) under the 51 Available at htt://en.eole.cn/n3/2016/0503/c html 52 For individuals who newly enter AC at the 19th Party Congress whose characteristics are not readily available, we randomly draw the characteristics from the samle of the new entries of 18th Party Congress. 38

39 first scenario and 0.21; 0.69; 0.10) under the second. Such redictions were run in advance of the 19th Congress and resented a year earlier, in October An analysis of our forecasting erformance for the CC and AC levels in the 19th Congress will be ossible as soon as comlete data is made available on China Vitae. For now, the emirical erformance on the 19th Politburo, jointly with our revious, more standard out-of-samle exercises, aears reassuring of the validity of the model s secification. 9 Conclusions This aer contributes to an emerging literature on the olitical economy of economic develoment by focusing on elite organization in a nondemocracy. We secifically focus on modern China and on the internal organization of the Chinese Communist Party. The CCP, much like historical Leninist arties in Socialist countries, reresents the linchin of national olitics and understanding its inner workings is central to any olitical economic analysis of the PRC. We resent an economic model of internal organization of this single-arty regime, where exlicit factional dynamics within the arty enrich a roblem of career concerns of olitical cadres. The model offers a series of novel insights on the role of factions in these regimes in a fully microfounded setting. Alternative modeling choices are also discussed. The model is validated emirically emloying a rich data set on the career rofiles of to CCP members. n reduced form, a set of reviously unexlored systematic emirical regularities in Chinese elite olitics are robed and discussed. The extent of the anti-corrution urge in shaing Chinese factional olitics is also analyzed. n our structural estimation, we exlore imortant counterfactuals ertinent to the Chinese historical case and use the model to answer a series of questions relevant to the olitical economy of the CCP. We hoe that this framework may also rove useful to the understanding of the latent institutional shifts occurring within the CCP under General Secretary Xi. n future research we hoe to extend our analysis to similarly comlex nondemocratic environments the examle of Russia comes to mind where our model of hierarchical arty organization may be transosable. 53 Please refer to the NBER working aer version of this article, available at 39

40 REFERENCES Bell, Daniel A The China Model: Political Meritocracy and the Limits of Democracy. Princeton, NJ: Princeton University Press Bai, Chong-En, Chang-Tai Hsieh, Zheng Michael Song Crony caitalism with Chinese characteristics. Mimeo Chicago Booth. Belloni, Frank P., and Beller, Dennis C Faction Politics: Political Parties and Factionalism in Comarative Persective. Santa Barbara, Clio Press. Bo, Zhiyue The 16th Central Committee of the Chinese Communist Party: formal institutions and factional grous. Journal of Contemorary China, 1339), Bo, Zhiyue The nstitutionalization of Elite Management in China. Holding China together: Diversity and national integration in the ost-deng era 2004): Bo, Zhiyue China s Elite Politics: Political Transition and Power Balancing. Hackensack, NJ: World Scientific Publishing Ceron, Andrea The Politics of Fission: An Analysis of Faction Breakaways Among talian Parties ). British Journal of Political Science, 451), Chen, Ye, Hongbin Li, and Li-An Zhou Relative Performance Evaluation and the Turnover of Provincial Leaders in China. Economics Letters, 88, Cox, Gary W.; Frances McCall Rosenbluth; Michael F. Thies Electoral Reform and the Fate of Factions: The Case of Jaan s Liberal Democratic Party. British Journal of Political Science, Vol. 29, No. 1. Jan., 1999), Cox, Gary W.; Frances M. Rosenbluth; Michael F. Thies Electoral Rules, Career Ambitions, and Party Structure: Comaring Factions in Jaan s Uer and Lower Houses. American Journal of Political Science, Vol. 44, No. 1. Jan., 2000), Dewan, Turun, Francesco Squintani n Defense of Factions. American Journal of Political Science, in ress. Dittmer, Lowell Chinese informal Politics. The China Journal 34: Dittmer, Lowell, Leadershi Change and Chinese Political Develoment. in The New Chinese Leadershi: Challenges and oortunities after the 16th Party Congress. The China Quarterly Secial ssues No. 4 Cambridge University Press. Fewsmith, Joseh The New Shae of Elite Politics, The China Journal, 45: Francois, Patrick, lia Rainer, Francesco Trebbi How s Power Shared in Africa?, Econometrica, March 2015, 832): Francois, Patrick, lia Rainer, Francesco Trebbi The Dictator s nner Circle, mimeo UBC. Huang, Jing Factionalism in Chinese Communist Politics. Cambridge University Press. Jia, Ruixue, Masayuki Kudamatsu and David Seim Political Selection in China: Comlementary Roles of Connections and Performance. Journal of the Euroean Economics 40

41 Association, 13 4): Kato, Junko and Carol Mershon, nternal Party Organization in the talian Christian Democrats and Jaanese Liberal Democrats: Factional Cometition for Office, Clienteles, and Corrut Exchange. n Comaring Political Corrution and Clientelism, J. Kawata, ed., Ashgate Publishing Comany, Burlington VT. Kuhn, Robert Lawrence The Man Who Changed China: The Life and Legacy of Jiang Zemin. Random House. Laver, Michael, Daniela Giannetti Party cohesion, arty disciline, arty factions in taly, in ntra-arty olitics and coalition governments, ed. by K. Benoit, and D. Giannetti, Routledge, London. Li, Cheng The Shanghai Gang: Force for Stability or Cause for Conflict? China Leadershi Monitor 2: Li, Cheng China s Two Li s: Frontrunners in the Race to Succeed Hu Jintao China Leadershi Monitor 22:1-22 Li, Cheng The Battle for China s To Nine Leadershi Posts. The Washington Quarterly 35:1, Li, Cheng A Biograhical and Factional Analysis of the Post-2012 Politburo. China Leadershi Monitor 41: Li, Hongbin, and Li-An Zhou Political Turnover and Economic Performance: the ncentive Role of Personnel Control in China. Journal of Public Economics 89 9): Lu, Xi and Lorentzen, Peter L Rescuing Autocracy from tself: China s Anti- Corrution Camaign. Mimeo. UC Berkeley Lu,Fengming and Ma, Xiao Kee Silent and Make a Big Fortune: Partially Free Media and an Authoritarian ntra-elite Election. Mimeo Lungu, Andrei China s 19th Party Congress: Projecting the Next Politburo Standing Committee. The Dilomat McAllister an Party Adatation and Factionalism within the Australian Party System. American Journal of Political Science, 353), Myers, Gustavus The History of Tammany Hall. Boni & Liverlight, NY Nathan, Andrew A Factionalism Model of CCP Politics. China Quarterly 53 January), Nathan, Andrew and Bruce Gilley China s New Rulers: The Secret Files. New York, NY: The New York Review Books Nathan, Andrew Who s Xi? The New York Review of Books, May 12th ssue. Persico, Nicola, José C. Rodriguez-Pueblita, and Dan Silverman Factions and Political Cometition. Journal of Political Economy, 119 2): Pye, Lucian The Dynamics of Factions and Consensus in Chinese Politics: A Model and Some Proositions. RAND Air Force Reort. R-2566-AF 41

42 Qiu, Ping The Fifth Generation of CCP Leaders. Hong Kong: Xiafeier Publishing Comany Limited, 102 Shambaugh, David L China s Communist Party: Atrohy and Adatation. UC Press. Shih, Victor Factions Matter: Personal Networks and Distribution of Bank Loans in China. Journal of Contemorary China 13: Shih, Victor Factions and Finance in China: Elite Conflict and nflation. Cambridge University Press Shih, Victor, Christoher Adolh, and Mingxing Liu Getting Ahead in the Communist Party: Exlaining the Advancement of Central Committee Members in China. American Political Science Review 106 1): Shih, Victor Efforts at exterminating factionalism under Xi Jining: Will Xi Jining dominate Chinese olitics after the 19th Party Congress? Mercator nstitute for China Studies Paers, 1 June): Tsou, Tang Chinese Politics at the To: Factionalism or nformal Politics? Balance-of-Power Politics or a Game to Win All? The China Journal 34: Zhengxu Wang, Jinghan Zeng Xi Jining: the game changer of Chinese elite olitics? Contemorary Politics 1, 1-18 Zuckerman, Alan Political Clienteles in Power: Party Factions and Cabinet Coalitions in taly. Sage Professional Paers in Comarative Politics, 1-55, Beverly Hills and London: Sage Publications 42

43 Aendix This aendix briefly describes the formal rocedure of romotion based on the nterim Regulations on Selection and Aointment of Party and Government Leading Cadres issued in A detailed account can be found in Bo 2004). There are two subsequent udates issued in 2002 and 2014 but the main rocedure remains the same. According to the regulations, the aointment rocess consists four stes: democratic recommendations, screening, deliberation, and discussions and decision. First, the arty committee of the same level of the oening or the organization deartment of a next higher level delimit a ool of otential candidates. Second, the organization deartment screens candidates by having rivate meetings with relevant individuals, conducting ublic oinion olls, and interview the candidates. Third, the list of candidates are vetted through a rocess of deliberation. The articiants of the deliberation include the leaders of the arty committee, the legislature, and the government at the same level of the oening. Forth, the list of candidates is resented to the next higher-u arty committee where the selection decision is made. The arty committee of this level may make suggestions regarding the selection. n our model, faction seniors of the same level of the oening have the first inut in deciding the candidate ool. This is consistent with the formal arty regulation that arty leaders at the same level of oening has significant influence in the rocess of democratic recommendations, screening, and deliberation, which essentially ins down the list of the candidates. Notice that in our model the faction seniors at the same level of the oening do not make the final decision on which candidate is chosen. The final decision is modeled as a contest function which is outside of the control of any individual faction senior. This is also the case in reality. We further allow faction seniors at next higher level to overturn faction suort decisions downstream, following the rincile of one-level down control in CCP. 43

44 Aendix Equilibrium strategies in Proosition 1. Formally for each faction at any oening l 1: ) S t = {0, 0...0} d ll σ ll σ l 1)l ) {d S t, d ll ll = if l t { > 0 1/ t l+1, 1/l+1 t,..., } 1/t l+1 else ) d l 1)l σ l 2)l S t, d l 1)l if l 1 t > 0, d ll = d ll else if l t { > 0 9) 1/ t l+1, 1/l+1 t,..., } 1/t l+1 else ) σ l h)l S t, d l h+1)l..., d l 1)l, d ll =. d l h+1)l if l h+1 t > 0 d l h+2)l else if l h+2 t > { 1/ t l+1, 1/l+1 t,..., } 1/t l+1. else For an oening at l = 1: σ ) 11 S t = { 1/2, t 1/2, t..., 1/2} t d 11 Proof of Proosition 1: Consider a faction member at level l when an oening arises somewhere in the hierarchy over which he can exercise a suort decision. Define S as the current state variable. Define S as the new state variable after the romotion has been made for the oening, which occurs in interval dt. The value function of a faction member at level l given S has already been defined, V l, S). Recall that σ l is the set of suort decisions of the faction member at level l. V l, S) can be written as: V l, S) = u l) + max σ l ) L l h=1 c 1 {σ l+h)+ and σ ll+h) σ l+h)+ } + ɛ 1 {σ ll+h) =1} +1 δ)dte [δ l S σ l)) V l 1, S σ l)) + 1 δ l S σ l))) V l, S σ l))] Ste 1. We first show that it is always a dominant strategy for any faction member at level 1 to suort the romotion of a co-faction member at 2 to level 1. Consider then the value function for an individual at level 1 which now becomes: 44

45 V 1, S) = u 1) + max σ l [ L l h=1 ) c 1 {σ l+h)+ and σ ll+h) σ l+h)+ } + ɛ 1 {σ ll+h) =1} +1 δ)dte [ V 1, S σ l))] Since an individual at 1 no longer has romotion rosects, δ 1 0, and further, for an oening at level 1 there are no subordinates that make decisions on factional suort, the first indicator function 1 {σ l+h)+ and σ ll+h) σ l+h)+ } 0. By direct insection, the value function will be maximized by suorting a co-faction candidate at 2, since this yields an ɛ utility gain via the second indicator function, 1 {σ ll+h) =1}. Moreover, this makes the value function indeendent of the state: V 1, S) V 1, S ), S, S. Since we focus on symmetric equilibria, suort is allocated equally over all co-faction members: σ 11 S t ) = {1/2, t 1/2, t..., 1/2} t. Ste 2. Given that a leader immediately suorts co-faction members at level 2, in a romotion race to level 1, irresective of S, it then follows that, under the Markov restriction, for all values of S, a faction member at level 2 will choose σ l for levels strictly below his own by aying attention only to the term 1 {σ l+h)+ and σ ll+h) σ l+h)+ }. Promotions to levels strictly below a member s level do not affect the robability of romotion at the member s level and therefore cannot affect his valuation; V l, S) V 1, S σ l)), rovided that S is identical to S at all levels k l. This is true for all levels. To see why note that the distribution of co-faction members at l + k cannot affect value functions at level l or above. Adding an additional co-faction member at any l + k has no effect if co-faction members are already resent since the faction can only romote a single member for any oening. f the faction has no members at l + k, then the faction would have turned to the highest l + jth, where j > k, level at which there is a faction member to ut forward a candidate. This will now not be necessary with a member romoted to l + k but it is, in either case, irrelevant for an individual at l, since it affects neither his romotion robability at resent, or the set of otential candidates he will comete with for romotion from his own faction in future. ntuitively, only co-faction members at one s level via direct cometition) and above otentially via suort decisions) can affect romotion decisions from one s own level. When a suort decision does not affect a faction member s own robability of romotion, i.e., when it concerns romotion to level 3 or below for a member at 2, his otimal decision deends only on the indicator functions secified in the value function. He will ratify the decision of his immediate subordinate, that is σ ll+h) = σ l+h)+. This is clearly otimal if the immediate candidate below suorted at least one candidate from the faction since this also maximizes 1 {σ ll+h) =1}. f the suort decision of the immediate candidate below involves suorting none for the oen osition then ratifying this decision is again otimal since c > ɛ. Note again, that this is true indeendent of S. Since he knows suort will 45 ]

46 be given for his subsequent romotion oortunity, from Ste 1 above, he avoids costs c by ratifying subordinate decisions. The same follows at all levels of the hierarchy. Namely, members at any level l when considering factional suort for oenings at level l+1 and below, find it a dominant strategy to ratify the decisions of the subordinate immediately below them in the hierarchy, irresective of S. That is σ l h)l S t, d l h+1)l )..., d l 1)l, d ll = d l h+k)l for the smallest k at which l h+k > 0. Ste 3. Consider again the case considered in ste 2, but now suose that the decision maker is the first faction member suerior to the oening osition. That is a member at l k deciding on suort for some osition strictly below, denote it level l, for which there are no other co-faction members in ositions l to l k. Since this member s suort decision does not affect his own robability of romotion from his current level, for the same reason as in ste 2 above, he only affects his utility by changing the robability of a co-faction member s romotion. So he thus chooses σ l to suort any candidate rising to a osition strictly below l k in order to maximize 1 {σ ll+h) =1}. n the symmetric case this is set to: σ l k)l S t, ) = { 1/l+1 t, 1/t l+1,..., l+1} 1/t, where it again follows that he will choose the unique robability weights over his co-factional subordinate candidates that sum to 1. Ste 4. Now consider oenings at a member s own node for non-leaders: i.e., σ ll St ) at l 1. We now establish that such members will issue a stay order to a co-faction member for an oening at a osition at his own node, asking them to wait until he himself is romoted. Consider the decision by a faction- olitician in node l of whether to suort a cofactional s romotion to his node. f he chooses to suort some co-factional from level l + 1 then we know from ste 2 above that the faction members suerior to him will ratify his decision. This imlies that, with factional suort, via the romotion contest function 6) there is a strictly ositive robability that a co-faction member will ascend to node l. Let l denote the total number of faction members that would be resent at level l after this romotion. Then, using equation 4) and 6), the romotion hazard arameter for this olitician at l if the other faction K also issues a stay order to co-faction members that could ascend to their level before their own romotion) will become: δ l = i l ) Jl 1 J l 1 i + η Nl 1 + N l 1 j + i + η with j = β and i = ρ, or vice versa. f instead, the olitician asks a co-faction member to wait, i.e., σ ll St ) = {0, 0...0} it then again follows from ste 2 above that, necessarily, either a member of the other faction or a neutral) ascends to the oening at node l. His romotion )), 46

47 hazard then becomes: δ l = i ) Jl 1 l 1 J l 1 i + η Nl 1 + N l 1 j + i + η The latter exression is strictly greater and thus maximizes the instantaneous arrival of romotion rates. Since there are no costs to taking such a decision since no subordinate decisions exist for a decision maker at the node of the oening, then 1 {σj and σ j σ + j } 0 and σ ll S) = {0, 0...0} is otimal. f the other faction does suort its members when oenings arise, the relevant exressions become, resectively: δ l = i l δ l = ) Jl 1 J l 1 j + i + η i ) Jl 1 l 1 J l 1 j + i + η )) )) + N Nl 1 l 1, j + i + η )) + N Nl 1 l 1. j + i + η And the latter hazard is clearly higher again so that σ ll St ) = {0, 0...0} is otimal. To summarize: We have established from ste 2 above that all members ratify suort decisions of subordinates for oenings below ) their current osition, if such subordinates exist. That is σ l h)l S t, d l h+1)l..., d l 1)l, d ll = d l k+1)l if there exists some k [1, h] such that l k+1 t > 0. We have established from ste 3 above that when there are no subordinates between a member and an oening strictly below the member s level, he suorts all candidates symmetrically. That is: σ l h)l,,..,, ) = { 1/l+1 t, 1/t l+1,..., l+1} 1/t. Ste 4 establishes that all members issue stay orders to their own node from within the faction. That is: σ ll St ) = {0, 0...0}. These stes thus corresond to the strategies listed in the statement of the roosition. Since these have been shown to be dominant strategies under the Markov restriction, that is they are otimal irresective of S and irresective of the strategies followed by other layers, it then necessarily follows that these strategies are also best resonses to each other, and hence constitute a Markov Perfect Equilibrium. Moreover these strategies are symmetric, and though there also exist non-symmetric Markov strategies that are not strictly dominated, these are the unique symmetric strategies satisfying these criteria. Since these are dominant strategies, and they are the unique symmetric strategies that are dominant, they also constitute the Unique Symmetric Markov Perfect Equilibrium. Proof of Proosition 2. We dro time suerscrits and suress state notation from now on unless it is confusing to do so. Let us define the indicator functions B = 1, iff B l > 0 and. 47

48 B = 0, otherwise; N = 1, iff N l > 0 and N = 0, otherwise; R = 1, iff R l > 0 and R = 0, otherwise. Start with a neutral N), who is at level l in the hierarchy. δ l 1 is determined from the hierarchy above: δ Nl = R l 1 Rl 1) R Nl + N l 1 Nl 1) N Nl + B l 1 Bl 1) B Nl. Consider further that, differently from 6) where R Nl = η/ Bβ + η + R ρ), now R Nl = η/ B β + η) because in Proosition 1 it has been shown that faction R will ask any member from the set of faction members R l 1 rising to level l and artnering an R already there to wait. For a similar reason, it holds that B Nl = η/ Rρ + η). We then have: δ Nl = η N l ) Rl 1 R l 1 B β + η Similarly, for a faction B member this is given by: δ Bl = R l 1 Rl 1) R Bl + N l 1 Nl 1) N Bl + B l 1 Bl 1) B Bl = β ) Rl 1 R l 1 B l β + N η ) Nl 1 + N l 1 B β + R ρ + η + B l 1 Bl 1 R ρ + η )) ) ) Nl 1 + N l 1, 10) β + R ρ + N η where the last line uses the fact that otimal factional actions from Proosition 1 imly B Bl = 0, while R Bl = β/ β + Nη) and N Bl = β/ β + Rρ + N η). Finally, for a faction R member this is: δ Rl = R l 1 Rl 1) R Rl + N l 1 Nl 1) N Rl + B l 1 Bl 1) B Rl = ρ ) Bl 1 B l 1 R l ρ + N η ) ) Nl 1 + N l 1, 11) B β + ρ + N η. 48

49 where the last line uses the fact that otimal factional actions from Proosition 1 imly R Rl = 0, while B Rl = ρ/ ρ + Nη) and N Rl = ρ/ ρ + Bβ + N η). Full Listing of δ l) conditional on aramount leadershi For an N. f an N is aramount leader: δ Nl = η ) ) )) Rl 1 Nl 1 R l 1 + N l 1 N l B β + η B β + R ρ + η + B Bl 1 l 1. R ρ + η f an R is aramount leader: δ Nl = η ) Rl 1 R l 1 N l B β + η f a B is aramount leader: δ Nl = η ) Rl 1 R l 1 N l B β l + η ) Nl 1 + N l 1 B β + R ρ l + η + B l 1 ) + N Nl 1 l 1 B β l + R ρ + η + B l 1 Bl 1 R ρ l + η Bl 1 R ρ + η where B = 1, iff B l > 0 and B = 0, otherwise; R = 1, iff R l > 0 and R = 0, otherwise. For faction B member. f an N is aramount leader: δ Bl = β ) ) ) Rl 1 Nl 1 R l 1 + N l 1, B l β + N η β + R ρ + N η f an R is aramount leader: δ Bl = β B l ) Rl 1 R l 1 β + N η ) ) Nl 1 + N l 1, β + R ρ l + N η f a B is aramount leader: δ Bl = β ) ) ) Rl 1 R l 1 B l β l + N η + N Nl 1 l 1, β l + R ρ + N η where N = 1, iff N l > 0 and N = 0, otherwise; R = 1, iff R l > 0 and R = 0, otherwise. For a faction R member. f an N is aramount leader: δ Rl = ρ ) ) ) Bl 1 Nl 1 B l 1 + N l 1, R l ρ + N η B β + ρ + N η 49 )) ))..

50 f an R is aramount leader: δ Rl = ρ ) ) ) Bl 1 B l 1 R l ρ l + N η + N Nl 1 l 1, B β + ρ l + N η f a B is aramount leader: δ Rl = ρ ) ) ) Bl 1 Nl 1 B l 1 + N l 1, R l ρ + N η B β l + ρ + N η where N = 1, iff N l > 0 and N = 0, otherwise; B = 1, iff B l > 0 and B = 0, otherwise. Proof of Proosition 3. Since flow ayoffs are higher the higher the olitician is in the hierarchy, i.e. u l 1 > u l l, then it is immediate by insection of exression 7) that increasing the rate of romotion > 0 l. This imlies that δ l > δ Jl ensures V l, S) > V J l, S). The roof roceeds by establishing sufficient conditions for three arts. i) The existence of neutrals given factions exist; ii). The existence of a single faction given neutrals exist; iii) The existence of a second faction, given neutrals and a first faction already exist. n each art, a sufficient condition is rovided for δ l > δ J l and δ K l at a single level, l. The sufficient condition established in each case is thus required to hold at all l in order to ensure that an entering olitician refers entry as a tye. Part i). We establish a sufficient condition for there to be neutrals, i.e. l : N l 0. raises valuations, dv l,s) dδ l Suose, on the contrary, that N l = 0 l; there exist no N members in the hierarchy. Necessarily, due to Proosition 1, without N s, all nodes will be filled by both a B and an R. Assume, without loss of generality, that the aramount leadershi osition is held by a B. Consider level l in the hierarchy. Necessarily the romotion hazard for an N at level l, if one were to exist, would be given by: δ Nl = η ) Rl 1 R l 1 B β l + η )) + B Bl 1 l 1. R ρ + η Due to otimal faction decisions at each node, it must be the case that R l 1 = B l 1 = M l 1 /2 and B = R = 1. The relationshi between δ Rl 1 and δ Bl 1 is ambiguous. So consider both cases searately. First, assume that δ Rl 1 δ Bl 1, which will imly, due to the symmetry 50

51 of the osited hierarchy, that δ Rl δ Bl too. Then, substituting for B, R, R l 1 and B l 1 yields: ) )) Rl 1 Bl 1 δ Nl = η M l 1 /2 + M β l l 1 /2. + η ρ + η Since δ Rl 1 δ Bl 1 then: δ Nl ηm l 1 /2 Rl 1 ) 1 β l + η + 1 ), ρ + η and assuming, for now, that β l > ρ imlies: δ Nl ηm l 1 ) ) 1 Rl 1. 12) β l + η Now consider δ Bl : δ Bl = 2βl M l 1 /2 M l Rl 1 β l )) Then δ Nl > δ Bl if: ηm l 1 Rl 1 = 1 M l Ml 1 Rl 1)). ) ) 1 > 1 Ml β L + η M Rl 1)), l which rearranges to: Now consider δ Rl : η β l + η > 1 M l. 13) δ Rl = 2ρ M l 1 /2 M l Bl 1 ρ )) = 1 M l Ml 1 Bl 1)). Since, by suosition, δ Rl 1 δ Bl 1 it is ossible to define Z 1 such that δ+δ Bl Z. Note δ+δ Rl that Z is invariant with resect to M l. To see why, note that with a symmetric hierarchy in 51

52 which each node is filled by a B and R air we have: δ Rl = δ Bl and δ Bl = δ Rl. Thus, using equations 10) and 11) and the fact that in such a hierarchy R l = B l = M l /2, we have δ Rl = M l 1 ) Rl 1) M l and So the ratio δ Bl δ Rl = δ Rl δ Bl the notation Z we then have: Then δ Nl > δ Rl if: which rearranges to: δ Bl = = M δ+δ Rl 1 ) l 1 M l M l 1 δ+δ Bl 1) M l ηm l 1 Rl 1 M l 1 ) Bl 1). M l = δ+δrl 1 δ+δ Bl 1, which is clearly indeendent of M l. Using δ Rl = 1 M l Ml 1 Rl 1) Z ). ) ) 1 > 1 ) Ml 1 β L + η M Rl 1) Z, l η β l + η > Z M l. 14) which again holds for M l large enough at all l. So for sufficiently large M l, neutrals will be the referred entering tye, thus contradicting the maintained assumtion that neutrals are not in the hierarchy. Assuming, alternatively, that β l ρ, instead of using the inequality in 12) we now have: δ Nl ηm l 1 ) ) 1 Rl 1, ρ + η which, by following the same rocedure as above, yields the analog to 13) as a sufficient condition for δ Nl > δ Bl, namely: η ρ + η > 1 M l. 15) This again holds for sufficiently high M l, and again will also hold for sufficiently high M l for the R entrants subject to the scaling by factor Z. Again, entering oliticians will choose to be neutral. Now suose the alternative relationshi between δ Rl 1 and δ Bl 1, that is: δ Rl 1 > δ Bl 1, and again first osit that β l > ρ. Then using these two inequalities and substituting for 52

53 B, R, R l 1 and B l 1 exactly as we did above. Equation 12) now yields: ) )) Rl 1 Bl 1 δ Nl = η M l 1 /2 + M β l l 1 /2 + η ρ + η > ηm l 1 /2 ) 1 Bl 1 β l + η + 1 ) ρ + η Now δ Rl is given by: > ηm l 1 Bl 1 ) 1 β l + η δ Rl = 2ρ M l 1 /2 M l ). Bl 1 ρ = 1 M l Ml 1 Bl 1)). Then δ Nl > δ Rl if: ηm l 1 ) ) 1 Bl 1 β L + η A sufficient condition for this is: This again holds for M l high enough. Now δ Bl is given by: η β l + η > 1 M l. δ Bl = 2β M l 1 /2 M l )) > 1 M l Ml 1 Bl 1)). Rl 1 β = 1 M l Ml 1 Rl 1)). Since, by suosition it is now the case that, δ Bl 1 δ Rl 1 it is ossible to define K 1 such that δ+δ Rl K. Similarly to the above, K is invariant with resect to M δ+δ l. Substituting Bl for K we have: δ Bl = 1 Ml 1 ) M Bl 1) K. l Then δ Nl > δ Bl if: ηm l 1 ) ) 1 Bl 1 β L + η 53 )) > 1 M l Ml 1 Bl 1) K ).

54 A sufficient condition for this is: η β l + η > K. M l This again holds for M l high enough. So new entrants will refer to enter as neutrals over either faction for M l large enough. The analogous rocedure under the alternative assumtion β l ρ yields a sufficient condition exactly as in 15): η ρ + η > 1. M l Part ii). We now establish a sufficient condition for there to exist at least a single faction. Suose that all ositions in the hierarchy are held by a neutral. Consider an entrant choosing to also be a neutral. He would then have: δ Nl = N l 1 M l Nl 1). But by entering as a B member the entrant would have: δ Bl = βn l 1 Nl 1 β + η These rearrange to imly that δ Bl > δ Nl rovided that M l > β+η. The analogous sufficient β condition for an R entrant is M l > ρ+η. This roves art ii). ρ Part iii). We establish a sufficient condition for two factions to exist. We roceed as above, by demonstrating a contradiction. f there is only one faction resent, without loss of generality let it be B, and the other oliticians are N, for sufficiently high M l, δ Rl > δ Bl or δ Nl, so that an entering olitician will choose to enter as an R. As reviously, with only N and B in the hierarchy we have: δ Nl = ). ) η Nl 1 N l 1 M l B l β + η δ Bl = β )) Nl 1 N l 1. B l β + η Bl 1 + B l 1 η )), 16) Either δ Nl > δ Bl, so that a new entrant would refer to enter as an N over a B, or the converse, in which case he would choose to enter as a B over an N. Suose first that δ Nl > δ Bl and consider the romotion hazard for a single entering R: δ Rl = ρ ) Bl 1 B l 1 ρ + η 54 Nl 1 + N l 1 ρ + β + η )).

55 f δ Nl > δ Bl for an increase in M l, then necessarily the term M l B l in exression 16) increases with M l, since an extra olitician would enter as an N instead of a B. But since δ Rl above is indeendent of M l, there exists an M l sufficiently large so that δ Rl > δ Nl, and an entering olitician would instead choose to be an R over being an N, contradicting the osited non-existence of R members in equilibrium. Alternatively, suose that δ Nl δ Bl, then, for an increase in M l necessarily the term M l N l increases, as a olitician would choose to enter as a B over being an N. Now consider the romotion hazard for a B: )) β Nl 1 δ Bl = N l 1. M l N l β + η Again, since δ Rl is indeendent of M l, there exists an M l great enough so that δ Rl > δ Bl, which imlies that a new entrant will choose to enter as an R member, again contradicting the osited non-existence of R members. 55

56 Figure 1: Geograhic Distribution of Factions or Grous ) This grah shows the geograhic distribution of factions or grous across rovinces municialities) over the eriod of 1956 to The color scale reresents the average share of faction or grou in a rovince municiality). 56