Centralized decision making against informed lobbying

Transcription

1 Centralized decision making against informed lobbying Rafael Costa Lima Humberto Moreira Thierry Verdier USP FGV PSE March 9, 01 Abstract We re-address the trade-off between centralized and decentralized decision making of local olicies when olicy makers are subject to cature. In articular, we consider the case where lobbies have rivate information about their ability to influence. We find a new informational effect in the olitical game under centralized structures that gives the olicy maker additional bargaining ower against lobbies. Thus, when comared to decentralization, centralization reduces cature, and is more likely to be welfare enhancing in the resence of information asymmetries. Then, we aly the model to the classical roblem of local ublic good rovision and to identify the incentives towards customs unions agreements. 1 Introduction The roblem of allocation of decision rights inside governments is a central question in Public Economics. This is erhas best illustrated by the classical debate on the costs and benefits of centralization versus decentralization of ublic decision making. In his seminal work on the rovision of local ublic goods, Oates (197) considered this roblem and emhasized the tradeoff between the relative imortance of inter district externalities (favoring centralized systems) and heterogeneity of references across districts (favoring decentralized systems). A subsequent literature embedded the discussion within a olitical economy framework reflecting the fact that centralized and decentralized systems face different olitical constraints or incentives in terms of ublic sendings (Inman and Rubinfeld 1997, Lockwood 00, Besley and Coate 003). An imortant dimension in the discussion is the fact that olicy makers may be subject to olitical influence by secific interest grous, and that the structure of ublic decision making may affect the likelihood of such olitical cature (Bardhan and Mokherjee 000). It is indeed recognized that interest grous may influence the olicy rocess through two secific channels. The first one is knowledge of rivate information that can be strategically disclosed according to one s own agenda. Alternatively, interest grous may also act as cometing rent-seeking actors, influencing olicy makers through bribes or financial contributions conditional on the olicies that Preliminary version! 1

2 they favor. Whatever the mechanism, the degree of effectiveness of ressure grous is likely to deend on how the structure of decision rights in the government frames cometition for olitical influence. In such a context, what are the costs and benefits of centralized and decentralized systems? What should be the otimal structure of government from the oint of view of society? The urose of this aer is to consider these issues in a simle context in which the two sources of influence (asymmetry of information and contributions) are closely connected. Secifically, we consider situations where secial interest grous have rivate information on how olicies affect their ayoffs and at the same time, may influence the olitical rocess through the use of financial transfers to oliticians and olicy makers. In such a setting, we comare informed lobbying influence in centralized and decentralized decisions structures. Our starting oint is to recognize the fact that centralized systems concentrate the focus of olitical cometition for influence at ahighercommonlevelofgovernment,andalsogeneratemoreolicyuniformityacrosslocalities than decentralized structures. In such a context, we identify two effects that hel reduce olitical cature in centralized systems. The first effect is a reference dilution effect that occurs indeendently from the structure of information asymmetries. In centralized systems, ublic decision making takes more encomassing olicy views across districts. As a consequence, centralized olicies tend to accommodate more reference tradeoffs across locations and the scoe for influence by lobby grous located in different areas is reduced comared to decentralized systems. This reference dilution effect in turn reduces lobbies incentives to send money for olitical cature and therefore romotes olicies that more aligned with the general ublic interest. Our main contribution is to show that in a context of asymmetric information, centralization also induces another effect: an information transmission effect that tends to reduce the degree of olitical cature by rivately informed lobbying grous. This effect arises from the fact that in centralized systems, olicies integrate cross-districts secificities and therefore may create strategic informational interdeendences for rivately informed lobbies located in these districts. As a matter of fact, a centralized olicy maker s willingness to grant a favorable olicy to a lobby located in one area deends on how much that olicy is also serving a "rival" lobby in another area. When lobbies have rivate information, this imlies that each interest grou s otimal influence strategy deends not only on his own rivate information but also on the rivate information ossessed by the other rival lobby. Under centralization however, each lobby rooses to the common olicy maker financial contributions that may reveal art of their rivate information characteristics. In equilibrium; this feature allows the common olicy maker to learn something about each lobby s rivate information. As this is relevant for the design of his otimal influence strategy, a given lobby has then an incentive to screen from the olicy maker what the latter has learned from other rival lobbies. Screening however is costly and therefore induces lobbies to exert less influence. Additionally, information transmission increases the olicy maker s bargaining ower, so the latter enjoys olitical (informational) rents. As a consequence, olicies get closer to the society s otimum and the level of cature decreases with information asymmetry. Our second contribution is to discuss the conditions under which centralized systems are referable (or not) to decentralized systems. In our simle setting, the tradeoff weights the reference dilution and information transmission effects against the standard costs of uniformization of centralized olicies. Secifically, our analysis indicates that the larger the degree of lobbies rivate information and the wider the extent of lobbies references, the stronger the information trans-

3 mission effect and dilution effect and therefore the more likely the centralized regime dominates the decentralized regime from a normative ersective. While an analytical characterization of the olicy regimes is not ossible for general reference functional forms, we rovide a fully exlicit solution of the olitical game under centralized and decentralized systems for simle linear and linear-quadratic arametric secifications. Finally, we use the two revious examles to illustrate our model s imlications in two classical roblems of joint olicy making: the rovision of local ublic goods and the setting of an imort tariff in customs unions agreements. The first alication is quite direct, the amount of ublic good is the local olicy decision. This decision can be undertaken by one olicy maker on each district, or by a unique olicy maker for both districts. Centralization imlies that the unique olicy maker has to set a uniform olicy for both districts. Thus, a centralized olicy cannot satisfy districts with different olicy references. Yet, in centralized decision making, information transmission increases the bargaining ower of the olicy maker. Thus the trade-off between centralization and decentralization weights the gains from designing a olicy tailored to the districts influenceadjusted references in decentralization and the gains from decreased cature due to information transmission in centralization. The second alication considers the effects of a custom union agreement. In this agreement, countries remove all imort tariffs between them and coordinate their trade olicy to the same external imort tariff for countries outside the union. The tye of agreement therefore corresonds to a centralized decision structure. Conversely, when countries do not sign the agreement, the choice of the imort tariff is decentralized. Abstracting from standard terms of trade effects associated with custom unions; our model highlights again the effects of information on the olitical game of trade rotection. Setting a uniform tariff decided by a common olicymaker changes the incentives of rotectionist lobbies in each country articiating in the union. This examle illustrates how the information transmission effect can be a driving force towards a customs union agreement. The lan of the aer is the following. In the next section we discuss the related literature. Section 3 introduces a basic model of olicy making under lobbying influence with two social entities and one lobby grou associated to each entity. Section 4 solves for the equilibrium of the olitical game under centralization and decentralization when there is erfect information. Section 5 then considers the case with lobby secific rivate information. In articular, we rovide the exlicit characterization of the equilibrium olicies and contributions for the linear and linear-quadratic arametric examles. Section 6 discusses the otimality of the centralized and the decentralized regimes under both erfect and asymmetric information. Section 7 and 8 rovide the alication of our simle arametric examles to the case of local ublic good rovision and custom union agreements. Finally section 9 concludes and discuss avenues for future research. Related literature This aer investigates how cature affects olicy decision according to the structure of decision making. Bardhan and Mookherjee (000) and Bordingnon, Colombo and Galmarini (008) have also aroached this issue. In Bardhan and Mookherjee (000) centralization is better when lobbies are less well-organized at the national level while decentralization dominates when local 3

4 districts have a strong references for one arty. Centralization and decentralization regimes have different imacts because voters have different levels of awareness and lobbies have different levels of cohesion in the two decision structures 1. In our case, the centralized and decentralized structures,affect differentially olitical cometition because of information screening and its imlications for lobbying incentives. Bourdignon, Colombo and Galamarini (008) also studied the effects of lobbying under centralization and decentralization in a setting similar to ours. They find that centralization is better when the lobbies references are conflicting while decentralization is better when these references are aligned. In our aer, we restrict lobbies references to be aligned (aart from a difference in intensity) but we allow for information asymmetries between lobbies and the olicy maker. This is what crucially generates our "information transmission" effect under centralization. As already mentioned, our aer is obviously related to the classical work of Oates (197). As in Oates (197), we allow for heterogeneity in districts references. However, in order to resent in the simlest ossible way the effects of lobbying and information transmission, we do not allow for inter-district sillovers. In our setting, decentralization is always welfare suerior in the absence of lobbying. The benefits from centralization comes therefore uniquely from the dilution of lobby influence and from the information transmission effect. As is well known in a erfect information context, introducing district sillovers would make the case for centralization even stronger. Moreover under asymmetric information, our basic argument for the existence of a beneficial information transmission effect would also be reinforced. Indeed with cross districts sillovers, a central olicymaker would design his olicies to correct for such cross-district externalities. These olicies would therefore naturally deend on cross district characteristics. Under lobbies rivate information about these characteristics, this feature would create the informational strategic interdeendence in the olitical game between lobbies that is at the heart of the our information transmission effect. In the context of local ublic good rovision, other aers address different olitical economy asects of the trade-off between centralization and decentralization (Seabright (1996), Lookwood (00), (Besley and Coate (003), and Redoano and Scharf (004)). For instance Seabright (1996) focuses on the effect of greater accountability of oliticians in decentralized decisions versus the increased coordination in centralized decisions. Lockwood (00) and Besley and Coate (003) breaks down the uniformity of olicies in centralized decisions, but considers a common ool system of financing the local ublic good, so one district could end u financing the ublic good for the other district. As a result, centralization can lead to oversending on local ublic goods. Redoano and Scharf (004) investigates the incentives for olicy centralization in direct and indirect democracies. Our second alication to customs unions is related to the large literature on the olitical economy of trade agreements. Close to our work, De Melo, Panagariya and Rodrik (1993) identifies also a reference dilution effect and finds that a trade agreement (not only a customs union) reduces the relative weight of lobbies in the objective function of decision makers when such olicymakers take into consideration the imlications on artner countries. Richardson (1993) comares free trade areas (FTAs) and customs unions, and finds the second tye of agreement to 1 Also, olicies are uniform in centralization. However, they allow lobbies to influence olicy makers of other districts even in decentralization and there are interdistrict externalities, which, for simlification, we do not include in our model. 4

5 be welfare suerior because tariffs become ublic good for lobbies on the same sector but from different countries. Hence, in customs unions the lobbies free ride on the contributions of each other and the overall rotection falls. Grossman and Helman (1995) and Krishna (1998) also consider the role of olitics on the incentives to sign referential trade agreements (PTAs). In a context where tariffs are endogenously defined by lobbying, they find that trade diverting FTAs were more likely to be suorted. Krishna (1998) also finds that the incentives for engaging in multilateral liberalization decrease after joining a FTA. More recently, Ornelas (005) and Maggi and Rodrígues-Clare (007 discuss the role for lobbying before and after an agreement is signed. The first aer shows that the rents that lobbies can cature decrease in a FTA, which makes welfare decreasing agreements less likely to be imlemented. The second aer considers the role of trade agreements as a commitment against future lobbying and also finds that trade agreements result in deeer liberalization when countries are more olitically motivated. Comared to the revious literature, our main contribution is to address the trade-off between centralization and decentralization with rivately informed lobbying. In this sense, our aer is therefore also connected to the large literature on the role of lobbies as roviders of information, such as Austen-Smith (1995), Austen-Smith and Wright (199), Potters and Van-Winden (199), Bennedsen and Feldman (006). In that literature, the lobby grou owns some information that is relevant for the decision maker and it may disclosure this information, according to its interests. Therefore, lobbying may otentially imrove efficiency. Our work follows though a different aroach, closer to Costa Lima and Moreira (01) which treats lobbies as rent-seekers with rivate information about their own references or technologies. From a technical ersective, our analysis borrows from the literature on informed rincial roblems (Maskin and Tirole (199)), and the recent theoretical literature on common agency with rivately informed rincials (Martimort and Moreira 010, Costa Lima and Moreira 01). The first literature rovides the aroriate framework to analyze our olitical game under decentralization, while the second one allows us to characterize the olitical game under centralization. We aly the technics develoed therein to contrast how centralized versus decentralized olicy structures differentially affect olitical cometition between rivately informed interest grous. For linear and quadratic linear secifications, this allows us to recisely uncover how the information transmission effect contributes ositively to the benefit of centralized systems. 3 The model We consider an economy with two distinct entities (grous, districts, communities, countries,...) A and B. Ineachentityn {A, B}, aolicymakerisneeded toimlementalocalolicy n 3. Each entity is comosed of two tyes of agents with different references over the imlementation of the olicy n. First, there is a continuum of identical individuals (the size of which is normalized to 1) having the following references: W n ( n )= 1 ( αn ), 3 The olicy n can be for examle an amount of a local ublic good, a secific local tax or a regulation when the entities are geograhic districts within the same national territory. It can be a "border" olicy like trade, immigration or international caital flows regulations when the entities are themselves national governments. 5

6 .where α n reflects the individual s referred olicy level in entity n. Second,thereisalsoaolitically organized lobby grou. That lobby grou reflects the interests of a small fraction of agents in entity n that have references: different from the first grou above. On to this lobby can disburse some money to influence the olicymaker resonsible for the imlementation of the olicy. More recisely, we assume that the lobby grou objective function can be described as: V (θ n, n,c n )=v (θ n, n ) C n where θ n is a secific arameter of the grou and C n the amount of money contributions that can be sent to influence olicymaking. Under decentralization of decision-making, each entity n is endowed with one olicymaker. As is common in the influence lobbying literature (Bernstein and Whinston (1986), Grossman and Helman 1994, 1996), we assume that this olicy maker cares about the society s welfare but likes to receive money contributions, C n.secifically,hisreferencesaregivenby U n ( n,c n )=C n λ (n α n ), where λ is the relative reference between contributions and the society s welfare function. Under centralization, the two entities can delegate the olicy decision to a joint olicy maker. In that case, this agent cares about aggregate society welfare and can be influenced by both lobbies. Additionally, we assume that the olicy maker has to set a common olicy to both entities. This assumtion of olicy uniformity is natural when centralization imoses by definition a common olicy instrument between the two entities. This is for instance the case with a custom union or a regional economic union that regulates uniformly "border" olicies of different national entities. In the case of fiscal federalism, this feature is however not necessarily satisfied and may demand secific assumtions (see Besley and Coates (003), Lockwood (00), Loeer (011)). Still, as a first ass it may be useful to cature the idea that centralized decision making is less sensitive to local secificities than decentralized decision making. Moreover, as will be clear in the sequel, this assumtion is not crucial for our basic conclusions. What is imortant for the information transmission effect that we identify is the fact that the olicy of one entity generates externalities (any kind of externality) on the other entity. In this resect, uniformization of olicies induce a ublic good comonent of centralized olicymaking, which is then the simlest case of externalities that we need for our olitical game. Secifically, the references of the joint olicy maker under centralization can be reresented as : U, C A,C B =Σ n C n λ ( αn ), reflecting therefore just the sum of the references of the decentralized case over the two entities. We make then the following assumtions. Assumtion 1 1. v 0, thatis,foragivenθ, v ( ) is concave in. 6

7 . v (θn,α n ) > 0, that is, the interest grou s referred olicy is always greater than the society s referred olicy. 4 Assumtion 1.i is made to ensure interior solution of the lobbies utility maximization roblem. Assumtion 1.ii imlies that the lobbies do not have oosing references for the olicy. That is, aart from the differences in intensity, their references are aligned. This is reasonable for a situation when the main conflict of interests is between lobbies and the rest of society, and not between lobbies from different entities. In order to get exlicit analytical solutions for the basic trade-offs of the model, we will consider two secific functional forms for the lobby s reference function v ( ). As it turns out, these two functional forms will be useful to illustrate stylized versions of our roblem in two interesting examles of olicy centralization: the rovision of local ublic good within a federation (section 7) and trade olicy harmonization within a custom union (section 8). Examle. (Quadratic Examle) The lobbies referred olicies are different from those of the society and catured by the following function: v (θ n, n )= 1 (n θ n ). Assumtion 1.ii imlies that θ n >α n for both lobbies. Examle. (Linear Examle) The lobby references are not satiated on. This can be reresented by the following function: v (θ n, n )=θ n n. In this case, the lobbies referred olicy is infinity. The timing of the game is then as follows: (0) In each entity n {A, B} Nature draws the lobby tyes θ n ; (1) Lobbies offer contributions to the olicy maker(s); () The olicy maker(s) accets or rejects the contributions; (3) The olicies are set, and if contributions were acceted, ayments are made accordingly. The benchmarks To understand the effects of lobbying and olitical influence, it is first useful to resent some benchmark results without lobbying. With decentralized olicies, each olicy maker chooses the olicy that maximizes the society s references. In this simle setu, that is exactly the society s referred olicy α n. Therefore, in a decentralized olicy making n = α n, for all n,where n is the decentralized olicy of district n without lobbying influence. 4 The model could be set u with v (θn,α n ) < 0 without significant differences in the results. However, severe technical comlications arise when there is no definite sign for this derivative. 7

8 When the olicy maker has to set a uniform olicy for both entities, he solves the following roblem max 1 α A + α B. which rovides the otimal olicy ˆ = αa + α B. as simly the average of the districts otimal olicies. Social welfare under decentralization is given by W n ( n )=0, for all n, while centralization rovides W A (ˆ) +W B α (ˆ) = A α B < 0. Obviously, decentralization yields higher ayoffs since olicies are tailored to meet the entities social references. Under centralization, neither of the entity gets its referred olicy. By construction, the model has therefore a decentralization bias, since we do not introduce any of the usual cross-entity externality that is art of the usual argument for olicy centralization. 4 Political influence Consider now the situation where olicy makers can be influenced the interest grous. We follow the standard influence lobby grou literature (Bernheim and Whinston (1986), Grossman and Helman (1994, 1996) that views the determination of olicymaking as the outcome of a common agency game with different rincials (the lobbies) that use lobbying contributions as an incentive device to induce the olicymaker (the agent) to take secific olicy choices. Comared to that literature however, we introduce the ossibility of asymmetry of information between the informed rincials and the uniformed agents, (the olicymakers) and focus on the interlay between lobbying and information asymmetries under centralized and decentralized structures. To do this, we first start to resent the olitical game with erfect information under both decentralization and centralization.. Decentralization As by assumtion there is only one lobby in each entity n, inadecentralizedsystemtheolitical game of influence collases to a simle rincial-agent model where each lobby incentivizes his local olicymaker to imlement his favored olicy n. More recisely, given the realization of his secific arameter θ n, the lobby of each entity n solves the following rogram: max v (θn, n ) C n n subject to the olicy maker s articiation constraint C n λ ( αn ) 0. For quasi-linear references, the roblem simlifies to max n v (θn, n ) λ (n α n ). 8

9 The olicy that maximizes this function, ˇ(θ n ), is the solution of the following first-order condition v n (θn, ˇ n )=λ (ˇ n α n ). (1) From Assumtion 1, it naturally follows that ˇ(θ n ) >α n,namelythattheimlementedolicy ˇ(θ n ) is above the olicy level that maximizes the entity s social welfare. For our secific functional forms, we get the following exressions. Examle. (Quadratic Examle) ˇ(θ n )= θn + λα n 1+λ. () That is, the olicy is a weighted average between the lobby s referred olicy θ n and the society s otimal olicy α n.itisincreasinginthelobby styeθ n and as θ n α n this olicy tends to the welfare otimum olicy, α n. Examle. (Linear Examle) ˇ(θ n )= θn λ + αn. (3) In this examle, the olicy is given by the society s target lus the lobby s relative strength weighted by λ. It is increasing in the lobby s tye θ n, while as θ n 0 it tends to the welfare otimal olicy, α n. Centralization In a centralized structure, the olicy is common to both entities. As a consequence, there is a ublic good comonent for both lobbies who offer contributions to the joint olicy maker. While that olicy maker now cares about the welfare of both districts, he is also subject to the influence of the two lobbies. After the realization of the secific arameters θ A and θ B,theoliticalgame becomes therefore a standard common agency game in which each lobby i rooses a contribution schedule C (, θ i ) to influence the choice of. We follow Bernheim and Whinston (1996), and as usual assume that lobbies lay truthful strategies. Thus, the equilibrium of the olitical game is equivalent to the solution of a centralized roblem max v θ A, + v θ B, λ The olicy that solves this roblem is θ A,θ B such that α A + α B. v θ A, + v θ B, λ α A α B =0. (4) Again, from Assumtion 1.ii, we have that θ A,θ B > (α A +α B )/. Equation (4) shows that under centralized decision making, the equilibrium olicy will reflect both the society s average reference and the lobbies references. Again, for our secific functional forms, we get the following exressions: 9

10 Examle. (Quadratic Examle). Under centralization, the olicy that solves (4) is given by θ A,θ B = θa + θ B + λ α A + α B. (5) (1+λ) Hence, the olicy is the average of the decentralized olicies under influence. It is increasing in the lobbies tyes and as θ A + θ B α A + α B it tends to the welfare otimal uniform olicy. Examle. (Linear Examle) Under centralization, the olicy that solves (4) is given by θ A,θ B = θa + θ B λ + αa + α B. (6) The olicy is the average of the decentralized olicies under influence. It is increasing in the lobbies tyes, while as both θ stendtozero,ittendstothewelfareotimaluniformolicy. 5 Lobbying with rivate information We consider now the situation where the lobbies are rivately informed about the arameter θ of their references for the olicy. As a result, the influence level is unknown ex-ante by the society and the olicy maker. For simlicity, we restrict ourselves to the case where the lobby s rivate information is not olicy relevant, that is, does not enter the society s welfare function directly. We assume that in each entity n,thelobby styeθ n is drawn from a i.i.d. uniform distribution f(θ) =1/ θ θ on the interval θ, θ with 3θ > 5 θ. We begin first with the analysis of the decentralized structure. Decentralization In a decentralized structure, each lobby offers contributions to the olicy maker of her entity. The olitical game is thus an informed rincial roblem. From Maskin and Tirole (1990), we know that in informed rincial roblems with quasi-linear references, the rincial reveals her information directly to the agent. Therefore, there are no distortions due to information asymmetry. As a consequence, the equilibrium olicies are the same as in the erfect information decentralized structure, namely ˇ(θ n ) as given by (1). Centralization In a centralized structure, the lobbies offer contributions to the same olicy maker. Each lobby is rivately informed about the realization of his tye θ n and does not know his rival s tye. Therefore, the utility maximization roblem of each lobby as an informed rincial roblem with the olicymaker. Several remarks are in order however. First, in this informed rincial roblem, each lobby has rivate information about his own tye while the olicymaker has no direct rivate information. However, the olicymaker simultaneously receives the contribution from the other rival 5 Hence the distribution of tyes is the same in the two entities A and B. 10

11 lobby. When the latter s contributions is searating, the olicy maker learns in equilibrium the rival s tye. Given that this iece of information is relevant for the lobbies ayoff, each lobby s roblem becomes then a rincial-agent roblem where the olicy maker is rivately informed about the rival s tye. Second, from Maskin and Tirole (1990), we know that informed rincials do not gain by ostoning information revelation. This justifies therefore our focus on informative equilibria with searating differentiable contribution schedules 6.Asaconsequence,ouroliticalcommonagency game with exogenous asymmetric information between informed rincials and an uninformed agent, becomes endogenously from the ersective of each rincial, a rincial-agent roblem with an asymmetrically informed agent about the characteristics of the other rincial. In solving that game, we follow closely Martimort and Moreira (010). As said, we restrict ourselves to searating equilibrium strategies reflecting the fact that a given lobby i chooses different contributions schedule [C i (., θ i )] as his tye θ i changes. We first consider one lobby i s best resonse contribution schedules to the other rival lobby j sstrategy,assumingthatthelatteruses asearatingstrategy[c j (., θ j )] with θ j θ, θ. Because of this, before choosing the level of the joint olicy in the second stage of the game, the olicymaker has some endogenous rivate information on θ j by simly observing the contribution schedule [C j (., θ j )] roosed by the rival j. From this it follows that lobby i s own otimal contribution schedule has to take into account the information rent that the olicymaker obtains from his endogenous knowledge about θ j.one may then characterize the otimal contribution schedule of lobby i, assuming that the olicymaker is erfectly informed on lobby i styeθ i. As noticed by Martimort and Moreira (01), the fact that the two lobbies tyes do not enter directly into the olicymaker s objective function ensures that the corresonding rofile of contribution schedules is also a best resonse in the more general asymmetric information game where lobby i has asymmetric information on θ i 7.Doingthisway, it turns out that lobby i s best resonse is itself searating and therefore conveys information on his tye to the olicymaker. This observation justifies then the fact that the same techniques can be used to comute in a symmetric way the rival lobby j s best resonse. The aroach holds consistently to characterize the informative equilibria we are looking for. More secifically, we denote by θ i the realization of the tye of district s n lobby and by θ j the realization of rival lobby j. Solving backwards,given that we are in a searative equilibrium, the olicy maker s roblem has full knowledge about θ i and θ j when deciding his olicy. Giventhe informative contribution schedules C i (, θ i ) and C j (, θ j ),hethensolves max C i (, θ i )+C j (, θ j )+λw (), (7) where we denote by W () =W A ()+W B () the utilitarian welfare of both entities. This roblem 6 This is also in the sirit of equilibrium allocations that are informative as in Sence (1973) and Riley (1979). 7 The reason is that the incentive and articiation constraints of the olicymaker do not deend on his beliefs on lobby i s tye but only on the schedule that this lobby offers to him. Therefore it follows that the olicymaker decisions to enter into the bilateral coalition with lobby i and to imlement the olicy accordingly are also indeendent on his beliefs on the lobby s tye. Any deviation away from the contribution that lobby i would otimally offer had the olicy maker being informed on his tye is thus dominated for any out-of equilibrium beliefs. 11

12 has the following first-order condition C i (, θ i)+ C j (, θ j)+λw () =0 (8) It is imortant to note that the equilibrium olicy deends on the sloe C i / and C j / of the contribution schedules which in turn, deend on the lobbies tyes θ i and θ j. It follows that the equilibrium olicy (θ i,θ j ) satisfying (8) deends as well on the lobbies tyes. Moreover when the necessary second order conditions of (7) hold and the contribution schedules C i (, θ i ) satisfy the Sence Mirlees roerty C i / θ i 0 and C B / θ B 0 8, simle differentiation of (8) rovides that the equilibrium olicy (θ i,θ j ) is increasing in the lobbies tyes θ i and θ j. Now, consider each lobby s utility maximization roblem. Since equilibrium olicies are assumed to be increasing in lobbies s tye, the roblem of choosing a contribution schedule and a rice, can be reduced for each lobby i to the roblem of choosing a value ˆθ i that defines the sloe of the contributions, given (8) and given the lobby s true tye θ i. Moreover, lobby i chooses his contribution non-cooeratively, uninformed about her rival s tye θ j. Therefore, he solves the following roblem: ˆ θ max v θ i, ˆθi,θ j C ˆθi,θ j, ˆθ i f (θ j ) dθ j (9) ˆθ i θ subject to (8). The fact that we focus on informative (truthful) strategies imlies that the solution of (9) should be ˆθ i = θ i for all θ i θ, θ. Following Martimort and Moreira (010) and focusing on oint-wise otimization, we obtain the following roosition characterizing the otimality conditions of each lobby, given his tye. Proosition 1. The otimality conditions of (9) for lobby i are given by the first-order condition: v (θ i,(θ i,θ j )) + C j ( (θ i,θ j ),θ j ) λw ( (θ i,θ j )) = θ θj C j θ j, (10) and the second order condition θ i (θ i,θ j ) 0. for (i, j) {A, B}, i = j and all (θ i,θ j ) θ, θ. Proof: In the aendix This first-order condition is a standard condition in screening models. It states that the marginal surlus of the bilateral coalition between lobby i and the olicymaker on the left-hand side of (10) is equal to the marginal cost of the latter s information rent (the right-hand side of (10). It looks similar to the first order condition obtained under erfect information, excet for 8 Something that we assume and that can be checked ex-ost after comuting the equilibrium contributions C A (, θ) and C B (, θ). 1

13 the fact that there is now a new term due to the information distortion. Since lobby i does not know his rival s tye θ j,hehastogiveincentivestotheolicymakertoreortthistyecorrectly. This means that he has to screen the rival s information from the olicy maker. As in most screening roblems, informational rents have to be given to induce the olicy maker reveal this iece information and choose accordingly a olicy according to the true tye of the rival. To save on such rents enjoyed by the high tye rivals, lobby i distorts the olicy it demands when facing low tye rivals, reducing therefore the sloe of his contribution schedule with resect to the olicy. The second order condition requires only that olicies are increasing in the lobby s own tye. This will be obtained when the second order conditions of (7) are satisfied and the equilibrium contribution schedules C A (, θ) and C B (, θ) satisfy a Sence Mirlees roerty C A / θ A 0 and C B / θ B 0. To comute the equilibrium olicy and the equilibrium informative contribution schedules C A (, θ) and C B (, θ), we solve the system of first order conditions (10), together with (8), the olicy maker s first-order condition. The second order conditions can be then checked ex-ost in the comuted equilibrium. The system of equations (10) and (8),define a system of artial differential equations in the contribution schedules C A (, θ) and C B (, θ) with boundary conditions given by the fact that the olicy maker s articiation constraints should be binding (no informational rent) for low tyes θ i = θ j = θ. Martimort and Moreira (010) show that there exists a solution to this system. Moreover they show that the equilibrium olicy (θ i,θ j ) is such that (θ i,θ j ) (θ i,θ j ) where ( ) is the centralized olicy under erfect information with the equality holding only when both lobbies are of the high tye (i.e.. θ i = θ j = θ). Hence asymmetry of information on the lobbies side reduces the joint olicy imlemented by the olicy maker. The intuition for this result comes from the fact that at a best resonse, each lobby induces a lower olicy level from the common olicy maker than what would be ex ost efficient for their bilateral coalition. This downward distortion reduces the information rent that the olicymaker gets from his endogenous rivate knowledge on the other lobby s tye. As both lobbies frame their contribution schedules in a way that induces the olicy maker to reduce his chosen olicy level, the actual equilibrium olicy will be reduced comared to the one obtained under erfect information. Under centralization, there exists an information transmission effect between the two lobbies through the joint olicy maker. This effect creates endogenously some informational advantage that the olicymaker can exloit, increasing therefore the cost of influence of the lobbies. As the latter reduce consequently the intensity of their contributions, olicy distortions are reduced. The information transmission effect sheds a new ersective with resect to the design of decision making under olitical influence. In a context of asymmetric information, centralization through delegation to a common olicymaker creates a mechanism that rovides some informational leverage of the olicymaker against interest grous. The result of this is less influence and reduced olicy distortions. This is readily illustrated with our quadratic and linear examles that rovide exlicit characterizations of the equilibrium.olicies and contribution schedules in an informative equilibrium. Proosition. When the lobbies references are given by v (θ, ) = 1 ( θ), 13

14 i) The equilibrium olicy is given by (θ i,θ j )= 3 / (θ i + θ j ) θ (1+λ) + λ α A + α B (1+λ) = (θ i,θ j ) θ 1 / (θ i + θ j ), for all (θ i,θ j ) θ, (1+λ) θ (11) where ( ) is the centralized olicy under erfect information. ii) The searating equilibrium contribution schedules Ci (, θ) for i {A, B} are given by Ci (, θ) =C (λ 1) (, θ) = θ i + 6 θ 6 λ α A + α B + C 0 (θ i ) for θ θ, 3 θ where C 0 (θ) is an increasing quadratic function of θ (whose exact shae is rovided in the aendix) Proof: seetheaendix Two things can be noticed. First we can see in exlicit terms that the olicy level under asymmetric information is smaller than the erfect information centralized olicy, since it has an additional negative term θ 1/(θ i +θ j ). This term is due to information transmission that makes (1+λ) lobbies less aggressive and consequently diminishes their influence. Second, the informative equilibrium contribution schedules are quadratic both in the olicy level and the lobby arameter θ. Moreover one can immediately check that this equilibrium contribution satisfies the Sence- Mirrlees condition, since C θ = 1 > 0. Also the second order conditions of the roblem (7) of the olicy maker are satisfied 9. Hence the second order conditions of roosition 1 are also satisfied. Second, one can also see that this contribution schedule is increasing in the lobby own arameter θ. Hence it is informative and allows the olicy maker to learn the arameter of each lobby for the second olicy imlementation stage. Proosition 3. When the lobbies references are given by v (θ, ) =θ i) The equilibrium olicy is given by (θ i,θ j )= 1 3 λ (θ i + θ j ) θ + αa + α B where ( ) is the centralized olicy under erfect information. 9 Indeed the second order condition of (7) is given by : C i (, θ i)+ C j = (θ i,θ j ) 1 θ θ i + θ j, (1) λ (, θ j)+λ W () < 0 which after substitution of the equilibrium schedule C (, θ) and welfare function W () = α A + α B rovides 1 (λ 1) 3 λ <0 14

15 ii) The searating equilibrium contribution schedules Ci (, θ) for i {A, B} are given by Ci (, θ) =C (, θ) = λ θ 3 + θ 6 λ α A + α B + C 0 (θ) for θ θ, 3 θ where C 0 (θ) is an increasing quadratic function of θ (whose exact shae is rovided in the aendix) The olicy level in a centralized structure is smaller under asymmetric information than under erfect information. Also the remarks discussed under the quadratic case equally aly to the linear secification. Hence the second order conditions of roosition 1 are also satisfied. and the equilibrium contribution schedules are informative in the sense that.the olicy maker is fully informed about each lobby s characteristic after receiving his contribution and olicy offer 6 Comaring centralization and decentralization In this section we comare the welfare of the two structures. We define welfare as the sum of the district s welfare functions, W ( ) =W A ( )+W B ( ). This criterion excludes the ayoff of the layers of the olitical game. It is a reasonable criterion if the lobbies and the olicy maker s sizes are negligible comared to the society, as will be the case of the examles we consider in the following sections. Given that we wish to highlight in the most transarent way the role of lobbies information asymmetries and the imortance of the information transmission effect in the comarison between centralized and decentralized structures, we simlify drastically the way the two entities A and B interact. Indeed we only include the fact that centralized decision making tends to roduce olicies less resonsive to the local environment than decentralized decision making (i.e. our "uniformity" assumtion). We should however kee in mind that this setting avoids imortant dimensions that are generally discussed in the literature on centralization and decentralization. In articular, our framework does not include features such as direct budgetary or environmental externalities, or strategic delegations across entities. Those elements are known to be imortant determinants of the comarison between centralized and decentralized structures. Nevertheless, under erfect information, our setting will first reroduce two effects that have been already emhasized under different forms in the literature. The first one is a standard "uniformization effect". As emhasized by the traditional literature on centralization (Oates 197), this effect tends to favor decentralization. The second one is an "influence dilution" effect that has been first illustrated in the olitical economy of centralization and regional agreements (De Melo, Panagariya and Rodrik (1993)). This effect er se tends to favor centralized decision making. The introduction of asymmetric information allows us then to highlight a third new effect into the tradeoff: the "information transmission" effect that rovides informational leverage to centralized decision making subject to the lobbies influence. We start the discussion of our basic tradeoffs in the context of our general framework. We then roceed to our two quadratic and linear arametrizations that allow us to get exlicit analytical conditions for the different dimensions of the tradeoffs. Without loss of generality, we assume α = α A α B > 0. Wealsodenote θ θ by θ. 15

16 6.1 General model discussion The Perfect Information Case Consider first the erfect information case. There are two main differences in the olitical game between centralization and decentralization. The first one is that under centralization, given that the olicy is uniform, the olicy maker cannot adjust its level according to the secificities of the entity s reference. This effect is a standard "uniformization effect" generally emhasized by the traditional literature on centralization (Oates 197). The second difference is the fact that the lobbies offer contributions to the same (unique) olicy maker. As a consequence, been subject to different sources of olitical influence, the olicy maker cannot fully adjust his olicy to reflect the reference of one secific lobby. He has to set the olicy according to the "mix of olitical references" of the interest grous he faces. We refer to this effect as a "influence dilution" effect. As is well known, the uniformization of olicies decreases social welfare. The size of the welfare loss is directly related to the extent of differences between the entities references, α. Onthe other hand, the "dilution influence" effect tends to increase social welfare. Indeed because the welfare function W ( ) is concave in the olicy level, welfareassociatedtotheaverageofthe two distinct olicies is greater than the average welfare of these two olicies. Hence centralized olicymaking that is subject to some "average olitical influence" of two lobbies generate higher social welfare over the two entities than decentralized olicymaking where each entity s olicy maker is subject to the influence of one secific lobby Moreover this effect increases with the range of lobbies tyes, θ, which determines the robability of having distinct lobbies across entities. In this erfect information set-u, the tradeoff between centralization and decentralization comes therefore from the comarison between these two effects: the "influence dilution" effect favors centralization, whereas the "uniformization" effect favors decentralization. Given that the "uniformization" effect (resectively the "influence dilution") effect is ositively related to α (resectively θ), the trade-off between centralization and decentralization will deend on the relative size of θ and α. This is most clearly illustrated by looking at two limit cases, where one of the two effects disaear. Consider first the case when when α 0 (i.e. entities have similar references) while θ >0,. In such a case, there is no "uniformization" effect. Simle insection of (1) and (4) when α =0 rovides that the exression for the welfare loss of the olicy is the same under centralization and decentralization. On the other hand, (4) shows that the olicy obtained under centralization reflects the marginal reference of the average of both lobbies, while in (1) the olicy obtained under decentralization reflects in each entity the marginal reference of the entity s own lobby. Hence, only the "influence dilution" effect revails leading to centralization to dominate decentralization from a welfare oint of view. Consider next the other limit case when θ =0, (with α >0): the two lobbies have the same olicy references. In such a case, there is no "influence dilution" effect. Indeed comaring (1) and (4), the lobbies marginal contribution v (θ, ) for a given is the same across entities. Hence (1) sets decentralized olicies according to the lobby s marginal reference and each entity s reference while (4) sets centralized olicies according to the lobbies (similar) references and the average of the entities references. The result of this that there is no "influence dilution" effect. The uniformization effect then imlies that decentralization is welfare suerior to centralization. 16

17 The Asymmetric Information Case We now allow for rivate information on the lobbies side. Under centralization, section 4 tells us that the lobbying game between the two interest grous and the joint olicy maker generates an "information transmission" effect; This effect tends to reduce the equilibrium level of the centralized olicy; On the other hand, such effect does not arise under decentralized decision making. Given that lobbies have intrinsically references biased to excessively large olicy levels, the "information transmission" effect contributes ositively to social welfare under centralization, while there is no such effect under decentralization. Consequently, for arameter configurations that under erfect information make the two decision making structures socially equivalent, the "information transmission" effect under asymmetric information shifts the trade off in favor of centralization. Moreover, the "information transmission" effect is directly related to the degree of asymmetric information that exists between the lobbies and the olicymaker. Therefore it deends ositively on the range of lobbies tyes, θ, and has no imact on the model when θ =0. Together with the two receding "uniformization" and "influence dilution" effects already identified under erfect information, the "information transmission" effect rovides an additional comonent of the tradeoff between centralization and decentralization that weights in favor of centralization. As under erfect information, we may again exect the tradeoff to deend on the relative sizes of θ and α. Clearly there will be a configuration of these two arameters such that social welfare under centralized and decentralized systems will be the same. Dearting from this situation, a larger value of θ strengthens the "influence dilution" and the "information transmission" effects, and therefore makes centralization suerior. On the other hand, a larger value of α reinforces the "uniformization" effect and therefore make decentralization suerior. Moreover, when lobbies have rivate information, the same configuration of arameters is more likely to induce centralization. With general functional forms v ( ) for the lobbies olicy references, one cannot exlicitly characterize the value of the various thresholds that characterize the receding tradeoff. The following sections consider the quadratic and linear arametrizations and rovide analytical conditions for the comarisons between centralized and decentralized decision making Hence, from this oint on, we work only with the two running examles. 6. Quadratic and Linear Examles The Perfect Information Case Proosition 4. a) If v (θ, ) = 1 ( θ) and θ sareerfectinformation,theexectedsocial welfare under centralization is greater than under decentralization if and only if ( θ) 6 >λ( + λ)( α). (13) b) If v (θ, ) =θ,andtheθ sareerfectinformation,theexectedsocialwelfareundercentralization is greater than under decentralization if and only if ( θ) 6 >λ ( α). (14) 17