Logrolling under Fragmented Authoritarianism: Theory and Evidence from China

Transcription

1 Logrolling under Fragmented Authoritarianism: Theory and Evidence from China Mario Gilli a, Yuan Li b, Jiwei Qian c a Deartment of Economics, University of Milan-Bicocca. Piazza dell Ateneo Nuovo,, Milan, Italy. mario.gilli@unimib.it b Mercator School of Management and Institute of East Asian Studies, University of Duisburg-Essen. Forsthausweg 2, Duisburg, Germany. yuan.li@uni-due.de c East Asian Institute, National University of Singaore. 69 Bukit Timah Road, Singaore. jiwei.qian@nus.edu.sg Preliminary Version Setember 0, 20 ABSTRACT Although olitical economy has achieved signi cantly rogress in understanding authoratarian systems, the inner working of olicy-making in authoratarian olities is still obscure. In articular, logrolling or trading of favors among vertical bureaucratic systems in China has been extremely under researched. An imortant rogress in the analytical methodology has been rovided by the Fragmented Authoritarianism framework develoed by China study scholars to analyze the olicy making roblems in a vertically segmented olitical system. However, although this framework hels illuminate the olicy rocess in China, it lacks formal modeling and a full econometric testing. This aer develos a formal model to study the e ect of logrolling on olicy making. We nd that, olicies under logrolling tend to be ine ciently high and olicies excluded from logrolling tend to be ine ciently low. We rovide emirical evidence to suort our argument by studying the logrolling between Ministry of Civil A airs and Ministry of Health. The consequence of logrolling is there is too high bene t associated with "Dibao" such that it even crowds out the unemloyment insurance. Similarly, there is over enrollment in rural health insurance. However, there is insu cient inut in infrastructure building in mental health care. Key Words: Authoritarianism, Logrolling, Fragmented Authoritarianism, Policy Making, China.

2 2 "The Chinese government makes olicy according to a decision rule of delegation by consensus If the agents reach consensus, the decision is automatically rati ed by the higher level; if the agents cannot agree, then the authorities ste in to make the decision, or the matter is droed or tabled until consensus can be achieved". Shirk 99,.6 "Instead of a small, stable grou of central o cials who are willing to trade votes on one issue because they trust others to ay them back on the next issue, there are larger, ad hoc grous that nd logrolling and agreement much more roblematic". Shirk 99,.27. Introduction In the ast few decades, olitical economics has achieved quite a lot of rogress towards analyzing the olicy making rocess in democratic olitical systems. Nonetheless, the formation of olicies in autocracies remains a di cult question for nowadays olitical economists. Conventional wisdom sees olicy making in autocracy as a black-box where decisions are made in smoke- lled room. Therefore, there has been little attemt to build a general analytical framework on autocratic olicy rocess. Based on the study of olicy making in China, this aer tries to exlain olicy making in autocracy by develoing an analytical framework with a seci c focus on "logrolling" among grous have arochial interests, both theoretically and emirically. Simly seaking, logrolling means two (or more) interest grous trade their mutual suort, so both can get bene t in the end. It is one way to reach consensus among the grous of roughly equal authority and have mutual veto ower. An imortant thesis we intend to forward in this aer is that, because of logrolling, autocratic governments tend to adot overreaching olicies but lose the ability to ull back even when the costs outweigh its bene ts. We will rst study the bench-mark models where olicies are made under dictatorshi or bargaining searately, and then we will set u a simle logrolling model to comare the results with the bench-mark cases. As an examle to illustrate the concet of logrolling, consider the decision making rocess in nowadays China, which has long been treated as a consensus-making rocess, as the leaders reluctant to challenge one another. In China, bureaucratic interest grous are stove-ied - meaning they are searate vertical organizations, reaching down from Beijing to the rovinces and cities. Each bureaucracy ushes for olicies in their own interests - to increase its own budgets, sta ng, etc.. The leaders make decisions by logrolling (or trading favors) with each other. They follow the rule, "I ll go along with what you want to do in your domain, if you let me call the shots in my domain"(interview with Susan Shirk). In this aer, we will illustrate the logrolling among two functional vertical organizations in China, say, the Ministry of Civil A airs (MCA) and the Ministry of Health (MOH). Ministry of Civil A airs(mca) treats the "Minimum Livelihood Guarantee Scheme" (or "Dibao") initiated in urban areas as its core resonsibility. MCA s objective is to rovide as much nancial and other assistances for households tagged as Dibao households as ossible. The medical assistance, one imortant in-kind transfer determining the overall bene ts of Dibao, is under the control of the MOH. MOH identi es who are

3 eligible to receive the medical assistance and sets the assistance standards. Therefore, MCA wants the suort of MOH on its agendas concerning the Dibao rogram, and has a strong incentive to trade favor with MOH. On the other hand, rural health insurance ( New Cooerative Medical Scheme ), one of the most imortant health insurance rograms in China, is the core interest of Ministry of Health. MOH s objective is to exand the coverage of rural health insurance, and the ideal osition for MOH is to have universal coverage. However, a lot of oor households in the rural areas do not have the nancial ability to ay for the insurance remium. They deend on the medical assistance fund controlled by MCA to hel them cover the cost. Therefore, in order to increase the number of enrollees registered with the rural health insurance lan, MOH needs the cooeration of MCA on assistance concerning rural health insurance, and has a strong incentive to ask MCA to follow what MOH wants. Logrolling in this context between MOH and MCA is that MCA uses its assistance fund to ay for the insurance remium for oor households in rural areas to register with rural health insurance lan. In exchange, MOH simly egs the health care subsidy to "Dibao", i.e. the eole who receive "Dibao" are automatically entitled to the subsidy of medical assistance. The result of logrolling between MOH and MCA is over-reaching and ine ciency in both Dibao and rural health insurance. We nd the overall bene t associated with "Dibao" is too high such that it even crowds out the unemloyment insurance. Note that the cash bene t of Dibao (claimed from MCA) is not so high and does not crowd out unemloyment insurance, but once other transfer-in-kind rograms managed by other ministries (including medical assistance managed by MOH) are included. The overall bene ts from Dibao may exceed the bene ts o ered from the unemloyment insurance and consequently crowd out some unemloyment insurance enrollees. Local o cials who manage Dibao (under Ministry of civil a airs) set the line accordingly to make sure Dibao bene t itself does not crowd out unemloyment insurance. However, o cials in other ministries (including MOH) managing di erent social assistance rograms simly use the Dibao as a tag to target the oor. In this case, overall bene ts of Dibao including cash bene t (claimed from MCA managed Dibao ) and other in-kind-transfer (e.g. medical assistance) may surass the bene ts from unemloyment insurance (author s interview). Similarly, we nd there is ine ciency for enrollees of rural health insurance in utilization of medical services. It is estimated that there are over 00 million eole are being covered by more than one social health insurance rograms in China. 2 Peole can only claim bene t from one of these social health insurance rograms if they registered with more than one social health insurances. It is not e cient that enrollees ay remium and government ay subsidy for an insurance lan for which those enrollees may never claim bene t from. Mental healthcare, which is one of the most imortant ublic health issues in China, is another olicy arena for which both MOH and MCA are resonsible. Both MOH and MCA are oerating mental hositals for treating atients. However, mental healthcare, which is not the riority issue of both MOH and MCA received de ciency in government investment. For examle, regarding to sychiatric care, the current bed utilization rate See htt:// 2 See, htt://news.xinhuanet.com/olitics/20-0//c_267.htm

4 Figure. Share of enrollees of unemloyment insurance and basic ension scheme in urban labor forces is over 96% and average length of stay for inatients is about days in China, which is much longer than the national average of.9 days for all diseases and it is estimated that over 70% of atients with serious mental disease did not receive roer treatment in China (Qian 202). The following descritive gures rovide the overall glimse of the ine ciencies caused by logrolling between MCA and MOH. One stylized fact is that there are much more urban labor forces enrolling with another social insurance rogram the Basic Pension Scheme (BPS) than with the unemloyment insurance (UI). From Figure, we can see there are only 0% of urban labor forces registered with unemloyment insurance while over 60% of urban labor forces registered with the Basic Pension Scheme. Both BPS and UI are comulsory for urban labor forces and in rincial the enrollment rate should be similar. However, the enrollment of UI is much lower, so we can infer that many urban labor forces chose to quit UI. And most of those who quit UI, join Dibao because of its higher bene t. From gure 2, we can see most of the eole who enrolled in Dibao are actually unemloyed or exibly emloyed. Figure shows that the coverage of health insurance is universal. In 20, the total

5 Figure 2. Comosition of Bene ciaries under Dibao in urban areas

6 6 Figure. The number of enrollees under di erent social health insurances number of enrolees under three social health insurance lans is over.7 billion which exceed the number of total oulation in China (.6 billion, National Bureau of Statistics 20). A large amount of eole are covered by more than one social insurance lans. However, oor family may still have di culties to ay for healthcare exenditure. The out-of-ocket exenditure is over RMB trillion in 20 (total health exenditure is about RMB trillion). Figure shows that the share of government inut in mental hositals in total government health exenditure is decreasing in both tyes of hositals under ministry of Health (MOH) and Ministry of Civil A airs(mca). World Health Organization (WHO) s threshold level is at least 2% of total health exenditure should be allocated for mental healthcare, whereas the share of mental hosital exenditure in China is less than this threshold. The aer roceeds as follows. Section 2 reviews the existing literature on logrolling and olicy making in China in order to illustrate the contributions of our aer. Section is a theoretically resentation of the analytical framework of logrolling and our nding. In section, we resent the emirical model, the data and the emirical results. Conclusions are drawn in the nal section.

7 Figure. Government inuts in mental hositals as a share of total government health exenditure 7

8 2. Literature Review 2.. Literature on logrolling Originated from the old custom in the lumber regions of Maine, where lumberjacks assisted one another in rolling the logs to the river after they were felled and trimmed (as in hrase you roll my log and I ll roll yours), the word logrolling later was widely used in olitical science to refer to the henomenon of trading of favors. The academic study of logrolling received substantial attentions since the discussion of the issue in The Calculus of Consent, in which logrolling is described as the rocess of vote-trading in democratic olitics. If the individual articiant recognizes the economic value of his own vote to others on certain issues and, in turn, recognizes the economic value of others votes to him on searate issues, he will be motivated to engage in trade.... The individual may e ectively, but imerfectly, sell his vote on a articular issue, securing in return the votes of other individuals on issues of more direct interest (Buchanan and Tullock 962,. 92). To some degree, logrolling occurs in all olitical systems. The ruling coalition in Germany before the First World War was the nationalist "marriage of iron and rye", in which aristocratic landowners suorted a eet-building rogram that industrial interests desired; in exchange, big business suorted high agriculture tari s (Snyder 99). In democracies, logrolling occurs in the form of vote trading by oliticians, who trade suort for one issue or iece of legislation in exchange for another olitician s suort to the issue of their own interest. For instance, a vote on behalf of a tari of textile may be traded by a congressman for a vote from another congressman on behalf of a subsidy to the steel interests to ensure that both acts will gain a majority and ass through the legislature (Stratmann 99). Being viewed as deviation from the orderly working of the democratic rocess, vote trading of this kind was believed would lead to too much government sending and socially ine cient olicies even under majority rule (Tullock, 99). After the early studies of Buchanan and Tullock, the henomenon of logrolling in olicy making became a focus toic during the 960s to 970s, thereby economists and olitical scientist started to construct formal theories on it. For examle, many thought logrolling would lead to ine ciency, emhasizing the welfare loss associated with vote trading and view trading votes as a negative-sum game (Wilson 969; Riker and Brahms, 97; Tullock 970; Haefele 97; Browning and Browning, 979; Koford, 92). But the e ciency issue was highly disutable as some argued that vote trading might be socially e cient, because it allowed legislators to exress di erent intensities of references, making every legislator better o than he would have been in the absence of vote trading (Coleman 966; Schwartz 97). More recent works (Miller 977; Enelow, 96; Carrubba and Volden 2000) studied the conditions for the existence of logrolling arrangements under di erent voting rules. The limitation on most of the theoretical researches of logrolling is that they rely too heavily on satial modeling aroach which has been commonly used to study electoral cometition or social choice, and to a large extend, ignoring the crucial strategic asect of how agents interact in olitical and market environments. Focusing on htt:// Online Etymology Dictionary, accessed on Setember 0, 20. See also the lm htts://

9 the strategic interaction ersective, our aer takes a di erent aroach from the existing literature. We study the logrolling rocess between di erent layers by incororating the rich literature on modern bargaining theory develoed after the 90s (see Rubinstein, 92; Canning, 990; Osborne and Rubinstein, 990). In its broad sense, logrolling can be treated as a secial form of bargaining game. For examle, in organizational analysis, logrolling is described as one of the ways of bargaining that negotiators can reach integrative agreements (Pruitt, 9, 9). Logrolling is a bargaining rocess that allows the arties to trade o their low-riority concerns to achieve high-riority concerns (ibdl.). But in its narrow sense, logrolling is di erent from the tyical bargaining models described in the existing bargaining literature. In the classic bilateral bargaining (or Rubinstein bargaining) models (Rubinstein, 92; Binmore 96), a unique equilibrium exists where the rst roosal is acceted when information is comlete and the layers are imatient. In the legislative bargaining models (Austen-Smith and Banks 9; Baron and Ferejohn 99; Alesina and Rosenthal 996; Merlo 997), the bargaining rocess is multilateral where members of legislature act non-cooeratively, and the bargaining outcomes is a function of rules of bargaining, the legislative structural, and the reference arameters (Banks and Duggan, 2000). Logrolling di ers from these bargaining models in at least two asects. First, both these bargaining models focus on the question of how to divide bene ts among the members from one xed ie, no matter the rocess is bilateral or multilateral. The main roblem is that members have con icting reference towards the allocation of bene ts and there is no majority rule equilibrium in the standard social choice framework. However, these models do not involve the trading of favors among the members which is considered to be the essence of logrolling. In logrolling model considered here, there are at least two issues under consideration, and the layers have di erent riorities on these issues so they can trade favors with each other. Second, in the Rubinstein bargaining and legislative bargaining literature, the agenda setter gets more bene ts than the others in the sense that there exists rst-mover advantage. By letting di erent layers making roosals on di erent tye of olicies sequentially where the roosal ower is symmetrically distributed, in the logrolling model considered here the rst-mover advantage disaears. In order to see the di erences in the olicy outcomes, we searate the logrolling game from the bargaining game, and we comare the equilibrium outcomes under both games. Another tye of activities adoted by organized interest grous to romote their olitical objectives that have been widely discussed in the literature is lobbying. Logrolling is also di erent from lobbying activities. While lobbying takes lace between interest grous and olicy makers, logrolling takes lace between arallel interest grou. In the jargon of the economics literature, lobbying refers to meetings between reresentatives of interest grous and olicymakers in which the former try to ersuade the latter to romote the grous objectives (Grossman and Helman, 200.0). Such lobbying activities may involve the transfer of information by verbal argument which is regard as chea talk, or the transfer of monetary contributions to olicy maker (Persson, 99; Grossman and Helman, 200). The common-agency model (Grossman and Helman, 99, 99) has been widely used in the analysis of the allocation of grou seci c goods under lobbying activities. Generally, the organized grous get more than the social otimum, and the unorganized grous get less (Persson 99; Persson and Tabellini, 2000). 9

10 0 A more imortant and relative question that has been under studied by the existing theoretical literature is whether logrolling will lead to oversending and overreaching in state olicies. Through dee studies on the causes of the two world wars and the cold war, international relations scholar Jack Snyder warns logrolling by domestic interest grous might lead to the lack of constraint in military exansion. He oints out that the danger of logrolling is some well-organized arochial interest grous can hijack national olicy in directions that bene t themselves but are detrimental for the national good. Such olicies may be ine cient and undercutting roductivity which is also harmful to the ruling elite. But sometimes the state is unable to ull back from the detrimental olicies even if the cost of over-reaching has exceeded its bene t to the ruling elite (Snyder 99). Using a simle analytical framework, our aer shows that logrolling leads to oversending comared to the bench mark of benevolent social lanner and bargaining. We nd that through logrolling, rojects that bene t articular interest grous can get the aroval, even though the costs are disersed throughout the entire oulace. We may thus exect higher government sending on certain rograms when logrolling is allowed to ourish, and the citizens seen as a common ool end u receiving less ublic goods than those who are not in a logroll system. Most of the studies of logrolling are theoretical, because to testify the existence of logrolling deends on the record of trading of favors, which may not exist when the trade is imlicit (Evans, 99), or may di cult to acquire as such trade is usually considered morally rerehensible behavior (Buchanan and Tullock 962). Stratmann (992, 99) made some rogress in emirically test the henomenon of logrolling in the US congressional voting in the 960s and 90s. Later, and Irwin and Kroszner (996) rovide evidence on how interest grous traded favors with each other in the assage of Smoot- Hawley Tari Act of 90 by calculating the votes. The most recent literature tries to aly the theory of logrolling to exlain the equilibrium EU olicies (Crombez 2000) and olitical economy of IMF lending (Coelovitch, 200). However, in the context of Chinese olitics, it is almost imossible to nd such voting records. The existing emirical aroach seems to be not feasible in this aer, due to lack of data. Instead, we take an indirect aroach by rst looking at the overreaching olicy outcomes and then, use them to identify who are involved in logrolling and how they log roll with each other Literature on olicy making in China Making olicy is the core function of all nation-states. Understanding the olicy making rocess hels to oen u the black box of China s domestic olitics. During Mao Zedong s era, China has been treated as a near-totalitarian system (Richard Walker, 9). Like all autocratic regimes, the ruler, not accountable to anyone else for what he does, makes decisions and enforces the imlementation by using heavy doses of roaganda and coercion (Feridrich and Brzezinski, 96). However, even under such a olitically hierarchical system, in China economic decision-making involved a lot of negotiation and bargaining. Bargaining mainly took lace between the sueriors and subordinates, and concentrated on the level of commands coming down from above and the tye of information going u. Those at lower levels, for examle, have an interest in maniulating and distorting information in order to advance their own goals. It is observed in the literature that centrally lanned economy in China was signi cant di erent from other lanned economy

11 in that the Center s control over the economic system was not as strict as its control over the olitical system (Yingyi Qian and Xu 99; Xu, 20). Therefore, the decision making before the economic reform can be referred to as "command-bureaucratic" instead of "command", in order to cature both the authoritarian avor and the sense of unresonsiveness due to the imoverished information ow (Naughton, 992). There have been dramatic changes in China s olity in the ost-mao reform era. The Chinese leaders, who have become rogressively less dominant, have transformed toward more constrained gures who are rimus inter ares within a collective grou (Lamton, 20,.9). The to leaders need the suort of the selectorate to remain in ower (Shirk 99). There was an increase in the size and ower of the selectorate after Deng s reform, which created selectorate accountability in the olitical system (Gilli and Li, 20). When the olicy making system was dominated by strong leaders, the system had low friction, as subordinates did what they were told. But when the leaders became less dominant, the rocess of making olicy became more comlex and arduous. Beginning from the end of the 90s, a grou of in uential China Study scholars develoed the "Fragmented Authoritarianism" framework. In their view, the authority below the very eak of the Chinese olitical system is vertically fragmented (stove-ied), reaching down from Beijing to various levels near the bottom. These searate functional vertical organizations, such as various ministries, have equal rank according to China s bureaucratic ranking. Therefore, they cannot command each other. Besides, these vertical systems intersect with multitude of horizontal territorial systems from rovinces to townshis. So at every intersection, several organizations are involved for a single issue. The ossible ways to make decision and coordinate behavior include command by hierarchy, voting, and bargaining. As no single organization is suerior over another and voting has been avoided, the system falls back on bargain where decisions are made by rule of consensus or mutual accommodation (Lieberthal and Lamton 992; Shirk 99,.6; Lamton 20,.6). The ositive side of the fragmentation of authority is it revents overconcentration of ower; and the negative side is it makes achieving consensus very di cult (Shirk 99,.27). Each vertical organization is suosed to reresent its constituents and ushes for olicies in their own interests, but there are inadequate horizontal mechanisms of coordination, so they often nd themselves at loggerheads or gridlock. If some organizations refuse to comromise and agreement cannot be reached, the issue is either droed or is referred to a higher level for resolution (Shirk 99,.6). This is decision rule is also call "management by excetion" by management secialists (Lawler 976). Usually, the excessive amounts of issues to be solved are over the higher level authorities caacity limit. Therefore, in Chinese ublic organs, too many roblems remain unsolved for a long time simly because of the objection from a minority (Chen 97, Lamton 992,.7). Although the "Fragmented Authoritarianism" framework changes the simly image of the to-down olicy rocess, the roblem is it s mainly a descritive narration of the situation of domestic olitics in China. It is still far from a theoretically and emirically grounded analytical framework, based on which we can make redictions. There do exist lenty of theoretical studies of China s governance, but most of these The tyical volums on this include: Policy Imlementation in Post-Mao China, ed. David Lamtom 97; Policy Making in China: Leaders, Structures, and Processes, Lieberthal and Oksenberg, 9; Bureaucracy, Politics, and Decision Making in Post-Mao China, eds., Liebertal and Lamton, 992.

12 2 studies focus on the vertical relation between the uer level governments and the lower level governments. Based on the theories of multidivisional structure of large cororations, the governance structure of China is modeled as a multiregional governance form (M-form) (e.g. Qian and Xu 99; Eric Maskin, Qian, and Xu 2000; and Qian, Roland, and Xu 2006, 2007). In the M-form hierarchy, every region is controlled by the central government olitically, whereas each region not only enjoys a certain degree of autonomy but also is self-contained in its functions (Xu 20). Such structural arrangements generate high-owered incentives associated with regional cometition which contributes to the imroved quality of ublic scal olicy in China, as China is often seen as a major examle of market-reserving scal federalism (Gabriella Montinola, Qian, and Weingast 99; Qian and Weingast 997; Hehui Jin, Qian, and Weingast 200). But the moral hazard roblem is still revalent in the rincile-agent relationshi between the center and the local. Such roblems may range from the soft budget constraint syndrome (Qian and Roland 99), to the local governments acting in ways that corrode the caacities of the central state (Hongbin Cai and Daniel Treisman 200, 200). The most e ective tool the central leaders can use to control the lower level o cials is the ower of ersonnel aointment. For vertical control in Chinese bureaucracy system, a erformance evaluation system is alied to coordinate governmental deartments. Aointment, romotion and demotion of lower level bureaucrats are decided by whether they have ful lled the uer level government s requirements for various olicy targets. It is also observed in the literature that under the erformance evaluation system, local o cials in China are likely to be romoted on the basis of growth rate of GDP and scal revenue (Landry 200, Li and Zhou 200, Shih, et al. 202). Di erent from the above studies which focus on the vertical central-local relation in China, our research addresses the horizontal coordination (logrolling) among the units within Chinese government, which has been under-researched in the existing theoretical literature. However, logrolling agreement may be also roblematic if too many grous are involved and if there is no external institution to guarantee the enforcement of the agreement, because the willingness to trade votes on one issue deends on the trust of others to ay back on the next issue (Shirk 99,.27). The earlier literature on logrolling doesn t consider the issue of enforcement, neither the time dimension in the imlementation; all trades occur within one legislative session. Weingast and Marshal (99) bring intertemoral enforcement into the icture and, demonstrate that the committee system in congress also serves as enforcement. We think, in China, the inter-temoral dimension is also relevant in enforcing the logrolling deals. After studying olicy coordination in China during the 90s, Halern (992) oints out the long-term lanning e orts altered the olicy rocess in ways that e ectively laced the articiants in a reeated-game situation, incentives were thereby created to cooerate now in exchange for future cooeration. Moreover, some of that future cooeration could be institutionalized in the resent when olicy documents were formulated that lanned ahead many years (Halern, 992). This is similar to transaction in an economic market, in which each ministry may use olicy documents as a kind of contract to establish and rotect their olitical roerty rights away from the discretion of bureaucrats (Moe 990). For examle, in the guideline for social assistance released by Chinese government in May 20, the role and resonsibility of di erent government deartment are stated exlicitly in this guideline for the forth-

13 coming exansion of social assistance rograms. However, the enforcement of logrolling agreement is not the core issue of the current aer, the discussion here tries to justify our assumtion in the following theoretical model that all the layers commit to the logrolling deal.. The Theoretical Model Consider a society with distinct but homogeneous interest grous I 2 f; g, each interest grou can be seen as a single layer and. Grou has references, U (h; x) = h + C(x); () and grou has references, U (h; y) = h + C(y); (2) where h is general ublic good. The increasing and concave function C(:), with C(0) = 0, is de ned over the sending on grou seci c good x 2 [0; ] or y 2 [0; ]. As a secial examle, suose C() = (). x + y + h :.. Decision Made By Benevolent Social Planner First, we derive the e cient benchmark. Suose the allocation decision is made by a utilitarian social lanner whose goal is to maximize social welfare W, then it maximizes the utilitarian social welfare function subject to the resource constraint x + y + h : max (h;x;y) U (h; x) + U (h; y) = max (h;x;y) 2h + x + y () s:t: x + y + h ; x 0; y 0; h 0: () Since both utility functions are strictly increasing in constraint is binding which imlies h = x y: Hence, the social welfare roblem is x; y and h, the ublic budget max (x;y) 2 2x 2y + x + y (6) s:t: x + y ; x 0; y 0: (7) The objective function is concave and the constraint is linear, therefore the Kuhn-Tucker conditions are both necessary and su cient. The solution of the roblem is, 7 (h ; x ; y ) = ; 6 ; : 6 This allocation is e cient as it maximizes the utilitarian social welfare. The full set of the Kuhn-Tucker conditions is given in the aendix. ()

14 .2. Decision Made By One Interest Grou Then suose the olicy decision is made by either one of the interest grous. If grou has the decision making ower, it will simly maximize its grou welfare, max (h;x;y) U = h + x () s:t: h + x + y ; h 0; y 0; x 0: (9) As s objective function does not deend on y; then y = 0: Moreover, since s utility functions is strictly increasing in x and h, the ublic budget constraint is binding which imlies h = x: (0) Hence, the maximization roblem is max x x + x s:t: x 0; x : () The objective function is concave and the constraint is linear, therefore the Kuhn-Tucker conditions are both necessary and su cient 6 : The solution of the roblem is, if has full ower to choose its referred allocation, then it would imlement (h ; x ; y ) = ; ; 0 : Symmetrically, if has the decision making ower, it will choose h ; x ; y = ; 0; : Comare the result with the olicy made by the benevolent social lanner, we can see h = < h ; x > x ; y < y ; x < x ; y > y : It means there is subotimal amount of ublic good because of the excessive sending on its referred issue by the grou who has the decision making ower... Decision Made By Bargaining Next we focus on the bargaining rocess, adating the simlest legislative bargaining rocess discussed in the seminal work by Baron and Ferejohn (99), Persson (99), Persson and Tabellini (2000). The bargaining follows the following sequence of events:. one of the interest grous, say (we can also assume, but the analysis is totally symmetric), is chosen to be the agenda setter; 6 The full set of the Kuhn-Tucker conditions is given in the aendix.

15 2. the agenda setter makes a olicy roosal (h; x; y);. the other grou chooses to accet the roosal or not. If it accets the roosal made by the agenda setter then the roosal is imlemented; if not, a default outcome (h; x; y) = (; 0; 0) gets imlemented, which makes it to receive a reserve utility of U S =. From a formal oint of view, this is equivalent to an ultimatum game, with usual standard roerties.we can work out this simle game backward. The other grou will reject the roosal made by the agenda setter if it not getting at least as high a ayo from the olicy roosal as from the default olicy (we assume throughout that the other grou will accet a roosal when indi erent between it and the alternative). Hence, the other grou will accet the roosal if and only if U U S = x y + y 0 () x y y: (2) Knowing this, the agenda setter will maximize its own utility () subject to the incentive constraint (2). Thus, the agenda setter chooses to make a olicy roosal according to the following rogramming. max (x;y) U = x y + x () s:t: x y y; x + y ; x 0; y 0: () The objective function is concave and the constraints are convex, therefore the Kuhn- Tucker conditions are both necessary and su cient 7. The solution of the roblem is, if is rst-mover, then she would imlement h B ; x B ; y B = r 2 + ; ; r! : Symmetrically, in a bargaining setting if is the rst-mover, then he will choose h B ; x B ; y B = r 2 + ; r! ; : Therefore, when decision is made by bargaining, the agenda setter has the rst-mover advantage. Comare the result with the olicy made by the benevolent social lanner and the interest grou itself, we can see h = < h =B < h ; x > x B > x ; y < y B < y ; x < x B < x ; y > y B > y : It imlies a subotimal allocation of x; y and h, because of an excessive sending on its referred issue by the grou having the agenda setting ower. However, the distortion and overreaching are reduced with resect to the case when the interest grou has the monooly ower on olicy making. 7 The full set of the Kuhn-Tucker conditions is given in the aendix.

16 6 α Y αl αl αl x y + x, x y + y αl αl αl N, α x αl [ 0,] β Y N β β y y β L α L α [ 0 x ] L, α [ 0, x ] L α Y N,, Figure. The game of logrolling.. Decision Made By Logrolling Next, we consider the simlest logrolling rocess:. rooses a motion x L 2 [0; ]; 2. chooses whether to suort or not 0 s roosal, i.e. c 2 fy; Ng;. rooses a motion y L 2 [0; x L ];. chooses whether to suort or not 0 s roosal, i.e. c 2 fy; Ng;. If the two layers suorted each other s roosal, i.e. c = c = Y; we say a logroll is forged, and the roosals x L y L ; x L ; y L in the logroll are imlemented; 6. otherwise, if any layer rejected the other, we say the logrolling failed. Then a default outcome (h; x; y) = (; 0; 0) gets imlemented, hence both layers will get the reserve utility U S = : () The following icture reresents the game tree: We can work out this game from backward. We can let moves rst, but the analysis is totally symmetric.

17 7... STEP : In the last decision ste, chooses between to accet the olicy roosed by and the default olicy. We should distinguish two di erent set of subgames, whether in ste 2 rejected or acceted s roosal. If c = N, then s choice is irrelevant as both layers will get the default outcome. If c = Y, then c = Y if and only if U Y jh = (x; Y; y) U S () x y + x () y x x (6) The condition de ned by inequality (6) is s incentive logrolling (IL) constraint. Hence, we get the following sequential best rely for : < 2 fs; Ng; if h = (x; N; y) SBR (h ) = N; if h = (x; Y; y) s:t: y x x : Y; if h = (x; Y; y) s:t: y x x : (7)..2. STEP : In ste, will choose y L to maximize his utility subject to SBR = Y. Again, we should distinguish two di erent set of subgames, i.e. whether in ste 2 rejected or acceted s roosal. If c = N, then s choice is irrelevant as both layers will get the default outcome. Hence. if h 2 = (x; N), then SBR (h 2 ) 2 [0; ]; 2. if h 2 = (x; Y ), then SBR (h 2 ) 2 argmax y U (yjh 2 = (x; Y ); SBR (h )). If y L x x, then will not choose to logroll with, and can only get the reserve utility, i.e. U y L jh 2 = (x; Y ); SBR (h ) = U S =. Otherwise, will solve following maximization roblem: max y ( x y + y) () s:t: y x x (IL constraint) y x (Resource constraint) y 0 (9) Note that in the above roblem IL constraint is more restrictive than the resource constraint, hence we can omit it. The objective function x y + y is great than if and only if x. Moreover, 0 s objective function has a maximum when y =, which is great or equal than x x for any x. Therefore, the solution of the sequential best resonse of is < SBR (h 2 ) = : x x if h 2 = (x; Y ) & x 2 0; 2 [ x x; ] if h 2 = (x; Y ) & x 2 ; 2 [0; ] if h 2 = (x; N):

18 ... STEP 2 In ste 2, will choose whether to suort or not. In articular, c = Y if and only if U (Y jx; SBR (h 2 ); SBR (h )) U S (20) which imlies x SBR (h 2 ) SBR (h 2 ): Hence, we need to distinguish two cases, x 2 0; and x 2 ;. Case x 2 0; In this case will choose c = Y if and only if x SBR (h 2 ) SBR (h 2 ) (2) which imlies x q x x x x : (22) (22) is always satis ed for any x 2 0;. Case x 2 ; In this case U (Y jx; SBR (h 2 ); SBR (h )) = U S, hence c 2 fy; Ng : From both cases we get the following sequential best rely for : SBR Y if x 2 0; (x) = 2 fy; Ng if x 2 ; :... STEP Moving backward, in ste, will choose x 2 [0; ] to maximize U (xjsbr (x); SBR (h 2 ); SBR (h )) which is x x + x + x = if x 2 0; U (xjsbr (x); SBR (h 2 ); SBR (h )) = if x 2 ; : Thus U (xjsbr (x); SBR (h 2 ); SBR (h )) is constant and equal to ;hence SBR 2 [0; ] : To nd a solution, we assume that there is a lexicograhic reference for forging an agreement, hence SBR = : h L ; x L ; y L = h L ; x L ; y L = 2 ; ; :... Comments: We can ut olicy outcomes from di erence decision-making rules together in order to make comarision. The e cient allocation is (h ; x ; y ) = 7 ; 6 ; 6 ; When decision making is by only one grou (when one grou dominates the other), the allocation is (h ; x ; y ) = ; ; 0 or h ; x ; y = ; 0; ; (2)

19 9 When decision making is by bargaining, h B ; x B ; y B q = + ; ; 2 h B ; x B ; y B q q = + ; ;. There is rst-mover advantage; 2 q or When decision making is by logrolling, h L ; x L ; y L = h L ; x L ; y L = 2 ; ;. There is no rst-mover advantage: Comare the allocation of di erent issues under logrolling with the revious allocations, we can see h L = h L < h = < h =B < h ; x = x L > x B > x ; y < y B < y < y L ; x < x B < x < x L ; y = y L > y B > y : There are several interest oints worth to mention. First, the allocation of x; y and h is subotimal, but the distortion caused by logrolling is the most serious, as ublic good rovision is the smallest under logrolling. The direction of olicy distortion is di erent from other cases. Under logrolling, there is excessive sending on both x and y, the two issues through which the interest grous trading favor with each other. It means the roblem of olicy overreaching is more serious under logrolling. Second, the rst-mover advantage we observed in the case of bargaining no longer exists in the case of logrolling. The rst-mover advantage in the exisiting legislative bargaining literature is treated as the agenda setting ower (Persson, 99). But when decision making is by logrolling, there is no rst-mover advantage. This nding hels to understand the fundamental di erence between logrolling and bargain. The game under bargaining is a strictly cometitive game, in which one layer s gain is at the cost of the other layer s welfare. Therefore, the layer who is chosen to move rst will exloit all the advantages. However, the game under logrolling is not a strictly cometitive game, where the sum of the layers utility changes with their strategy. Coordinating well, they can increase their joint bene t at the cost of society s welfare, as there is under-rovision of ublic goods. This is the element cooeration in the logrolling which does not exist in bargaining. Therefore, under logrolling, the each layer s welfare is internalized, and hence, it doesn t matter which layer moves rst... Robustness check In the above calculations, we assume when logrolling fails a default outcome (h; x; y) = (; 0; 0) will be imlemented. In order to check whether the outcome of logrolling deends on the default olicy, we will generalize the default outcome. We can assume when logrolling failed both layers will get a reserve utility U D = 2 [0; ]. The incentive logrolling (IL) constraint for then becomes U Y jh = (x; Y; y) U D (2) which imlies x y + x (2)

20 20 i.e. the other grou will reject the logrolling roosal if it not getting at least as high a ayo from the default olicy. In equilibrium the layer s ayo from logrolling is ever greater or equal than the default olicy, therefore the layers always have incentive to forge a logrolling deal and the value of the default olicy is irrelavant from the outcome ath 9. Proosition In the logrolling game, assuming that both agents utility when the logrolling fails is U D = 2 [0; ] ; (26) the outcome ath is indeendent from and is h L ; x L ; y L = h L ; x L ; y L = 2 ; ; :. The Emirical Evidence In this section, we illustrate emirical evidences of logrolling among the functional vertical organizations in China, which results in over-reaching and ine cient olicy outcomes. The logrolling is taking lace between the Ministry of Civil A airs (MCA) and the Ministry of Health (MOH) in the olicy arena concerning social assistance. In our examle, the two negotiators, MCA and MOH, must choose government olicies with resect to three di erent issues, i.e. Dibao, Rural health insurance, and Mental Health Care. For logrolling to be ossible, the negotiators must have di erent riorities on di erent issues. The two ministries reference on di erent kind of goods is shown in Figure 6. MCA s to riority is Dibao, and MOH s to riority is rural health insurance. If each ministry sticks to its own references in these issues, they cannot reach consensus. According to "rule of delegation by consensus" described in the citation at the beginning of aer, these issues will be tabled or are referred to a higher level for resolution. Then the authorities ste in to make the decision, so that the default olicy is imlemented. If the two ministries trade favor, they could form a logrolling coalition and secure the olicy issue of their high riority. The MOH suorts the increase of the bene ts of Dibao and bears extra cost, but it anticiates MCA will suort its core interest in rural health insurance. In exchange, the MCA, will ay for the insurance remium for oor households in rural areas in order to suort MOH s core interest in exanding coverage of health insurance. Mental health care, which is not the riority for neither of the two ministries, will be sacri ced. Figire 7 shows how ministries can exchange interests via various olicies... Ine ciency in the high bene t associated with "Dibao" Because of logrolling, there is overreaching in the bene t associated with "Dibao" such that it even crowds out social insurance. Here we show the evidence of crowding out between Dibao and social insurance. We use two city-level datasets with data of social assistance rograms over 20 cities in China between year 200 and We nd 9 The detailed rove is given in the aendix.

21 2 Figure 6. Preference of the ministries High Priority Low Priority Public Good MCA MOH Dibao Providing as much financial and other assistances for households tagged as Dibao households as ossible. Rural health insurance Exanding coverage of health insurance. Increasing the number of enrollees. Rural Health Insurance* Regarding to oor rural households, MCA has incentive to only insure high risk eole with rural health insurance but not the low risk eole. Dibao MOH sets the standard for receiving medical assistance. Whether a reciient under Dibao or not is not very imortant for MOH Mental Health Care Treating atients in MCA managed hositals Mental Health Care Treating atients in MOH managed hositals *Note:Different from social health insurance in develoed counties, rural health insurance in China is a voluntary health insurance. Figure 7. Logrolling via various olicies Dibao Rural health insurance Mental Health Care MCA MOH Dibao households automatically are included as reciients for medical assistance. MCA uses the assistance fund to ay for the insurance remium for oor households in rural areas to register with rural health insurance lan. Insufficient inut for infrastructure building Insufficient inut for infrastructure building

22 22 that increasing social assistance exenditure are likely to reduce the number of registered unemloyed, which imlies some urban residents claiming social assistance are either leaving the labor market or had evaded from contribution for the unemloyment insurance. Many urban residents do not enroll with unemloyment insurance because the bene ts from Dibao and other social assistance rograms are higher than the bene t from unemloyment insurance. Our nding suggests that the crowding out e ect is a result of logrolling.... Hyotheses While we do not have the data for the number of enrollees of unemloyment insurance in each city, we use the number of registered unemloyed as a roxy for the number of enrollees of unemloyment insurance in the informal sector. While the number of enrollees has increased in the formal sector, workers in the informal sector may choose to enroll with unemloyment insurance or not. Since only eole who have registered as unemloyed can claim the unemloyment insurance, we can infer the number of unregistered unemloyed in the following year is ositively correlated with the number of enrollees of unemloyment insurance in the current year. We have following two hyotheses regarding to whether the regional equity has been addressed in the urban social assistance rograms: Hyothesis : increasing government exenditure on "Di Bao" is associated with a smaller number of registered unemloyed in a city. Hyothesis 2: increasing government exenditure on social assistance rograms in general is associated with a smaller number of registered unemloyed in a city...2. Data We have two data sources. The rst data set is collected from China City Statistical Yearbook (NBS: various years). This dataset includes observations of 22 cities (refecture level) covering 26 out of 27 rovinces between 200 and 2006 (i.e. Tibet is not included). These four years (i.e. from 200 to 2006) are only years for which the data for sending on social assistance is available. The second dataset is data reorted by MCA about Dibao data at the refecture level between 2007 and 2009, which was collected from the website of MCA. Both of our datasets include data for 22 refecture cities out of 2 refecture cities in total in China. City statistical yearbook only reorts data in 22 cities. A refecture city usually has both urban (i.e. city district) and rural areas (i.e. county). Since we are interested in urban social assistance rograms, most variables in our dataset are de ned in the scale of city district of these refecture cities only.... Methodology Research on crowding-out e ect uses individual level data. We do not have the luxury to access such data in China. However, we use city level data to treat a city rather than individuals as the basic research unit. City level data may reveal eole s resonse to social rograms since city level government is the level of government managing the social rograms and olicies regarding to the bene t and eligibility for social rograms varies with city. However, omitted variable bias is a concern for this kind of research. We have aly

23 2 (city) xed e ect model to address this issue. Simultaneity is another concern. We use redetermined values for control and indeendent variables to address this issue. Nevertheless, we acknowledge that city level data cannot tell us what individual characteristics are more likely to lead to ot out from unemloyment insurance. Also, some concerns about the measurement of unemloyment. For a long time, China only reorts the registered unemloyed and actual unemloyed number is absent. However, for this aer, registered unemloyed number is a su cient indicator for the crowding-e ect for the unemloyment insurance. Only registered unemloyed can be eligible to claim unemloyment insurance. It may be argued that the number of unemloyed may be reduced via other avenues such as exogenous shocks of government olicies or new olicy initiatives which have imact on urban unemloyment. In this case, we use local education exenditure as a regressor in the lacebo test. If the number of unemloyed is not correlated with education exenditure, it con rms the existence of crowding-out e ect.... Basic Model We estimate the following model: Unemloyed i;t = overty_assistance i;t + X i;t + i +! t + e i;t (27) where Unemloyed is the number of registered unemloyed er 000 eole in city i during year t. and are arameters for the corresonding variable(s) in the model. X i;t corresonds to covariates including average income, scal exenditure er caita, size of service economy, local gross roduct as well as unemloyed in the revious year. i denotes rovince-seci c e ects, whereas! t corresonds to year dummy variables, with e i;t de ned as the error term. We use standard anel data analysis to estimate above model. If there is a crowing out e ect between social assistance rograms and unemloyment insurance, urban residents may have not enough incentive to enroll with unemloyment insurance. In consequence, urban unemloyed have not enough incentive to register as unemloyed.... Variables The deendent variable is Unemloyment: unemloyment refers to the number of registered unemloyed in the city district of a city (In,000 eole). Many of the eole claiming Dibao are registered unemloyed. Figure 2 shows the number of unemloyed who are under Dibao rogram, which amounts to about 20% of bene ciaries who are under social assistance rograms. Control and indeendent variables are listed as following: Poverty_assistance denotes the city level sending on social assistance rograms (in Million RMB). The amount of overty assistance exenditure includes all exenditure including Dibao and other comlementary social assistance (Figure ). Dibao_exenditure denotes the city level sending on Dibao rograms (in Million RMB). Variable "oulation" denotes the number of residents in the city district of a city (Million oulation).