John F. Kennedy School of Government Harvard University Faculty Research Working Papers Series

Transcription

1 Jon F. Kennedy Scool of Government Harvard University Faculty Researc Working Papers Series In Searc of te Holy Grail: Policy Convergence, Experimentation and Economic Performance Sarun Mukand and Dani Rodrik July 2002 RWP Te views expressed in te KSG Faculty Researc Working Paper Series are tose of te autor(s) and do not necessarily reflect tose of te Jon F. Kennedy Scool of Government or Harvard University. All works posted ere are owned and copyrigted by te autor(s). Papers may be downloaded for personal use only.

2 IN SEARCH OF THE HOLY GRAIL: POLICY CONVERGENCE, EXPERIMENTATION, AND ECONOMIC PERFORMANCE* Sarun Mukand and Dani Rodrik Revised July 2002 Department of Economics Tufts University Braker Hall: Economics Medford, MA (617) Jon F. Kennedy Scool of Government Harvard University 79 Kennedy Street Cambridge, MA (617) ABSTRACT We consider a model of policy coice in wic appropriate policies depend on a country s own circumstances, but te presence of a successful leader generates an informational externality and results in too little policy experimentation. Corrupt governments are reined in wile onest governments are disciplined inefficiently. Our model yields distinct predictions about te patterns of policy imitation, corruption, and economic performance as a function of a country s location vis-à-vis successful leaders. In particular, it predicts a U-saped pattern in economic performance as we move away from te leader in te relevant space of caracteristics: close neigbors sould do very well, distant countries moderately well on average wit considerable variance, and intermediate countries worst of all. An empirical test wit te experience of postsocialist countries provides supportive results. * We tank Daron Acemoglu, Abiit Baneree, Georges de Menil, Jeffry Frieden, Peter Gourevitc, Ricardo Hausmann, Yannis Ioannides, Sanay Jain, Simon Jonson, Doug Nort, Jim Robinson, Gerard Roland, and seminar participants at Harvard, te NBER, and te Board of Governors of te Fed for very elpful comments. Te Ford and Rockefeller Foundations ave provided financial support for tis researc, wic we gratefully acknowledge.

3 IN SEARCH OF THE HOLY GRAIL: POLICY CONVERGENCE, EXPERIMENTATION, AND ECONOMIC PERFORMANCE Sarun Mukand and Dani Rodrik I. Introduction Tere as been a remarkable convergence of views over te last two decades among bot scolars and policy makers on wat constitutes desirable economic policies to promote growt and development. Yet economic performance is more eterogeneous across te world tan it as ever been. Growt rates in te developing world ave been lower on average in te past two decades tan tey were in te 1960s and 1970s. And te spread in growt rates across countries as been wider (see Table 1). Wy tis disappointing performance if we ave apparently learned so muc about wat we need to do foster economic convergence? One possible answer to tis question is tat te divergence in performance is due to te fact tat not all governments around te world ave adopted te requisite policies. Peraps wat is going on is tat countries tat ave latced on to te consensus view ave increased teir growt rates, wile oters are lagging. Tere are reasons to view tis explanation wit great suspicion, owever. One bit of telling evidence comes from Latin America, were tere as been greater entusiasm for Wasington Consensus-style reforms tan in any oter corner of te world. Morley et al. (1999) ave put togeter an index of structural reform for Latin America tat quantifies te policy canges in te areas of trade liberalization, tax reform, financial liberalization, privatization, and opening up to capital flows. Te average value of tis index for te continent rises from around 0.47 in te early 1970s (out of a maximum of 1) to around 0.55 in te early 1980s, and ten umps up to 0.82 by Yet economic performance in Latin

4 2 America as been quite disappointing during tis period. Even if we leave aside te debt-crisis decade of te 1980s (and ignore te natural rebound from it subsequently), one can count on te fingers of one and tose countries tat ave done better in te 1990s tan tey did during And of tose, only Cile is a genuine success. 1 Of course, te relationsip between reform and growt is a complex one, and tere is only a limited conclusion tat one can draw from suc a broad-brus look at te evidence. But te Latin American experience does suggest tat te growt payoff reaped by te most ambitious reformers as been meager at best. Anoter important strand of evidence comes from te last two decades success cases. Cina and India stand out ere. Tese two large countries ave experienced significant increases in teir growt rates Cina since te late 1970s, and India since te early 1980s in a period wen most oter developing countries ave gone te opposite way. 2 Indeed, Cina s and India s performance oversadows te disappointing outcomes elsewere, making te last two decades a developmental success on a population-weigted basis (see Table 1). Te policies tat ave enabled tis performance present a very awkward fit wit te usual list tat as been on te agenda of reformers in Wasington and elsewere. Cina s reforms ave been marked by partial liberalization, two-track pricing, limited deregulation, financial restraint, an unortodox legal regime, and te absence of clear private property rigts. India s reforms ave been less distinctive, but still marked by significant departures from te rulebook. Even after te trade reforms of te early 1990s, for example, India remained one of te world s most protected economies. One could argue tat tese economies would ave grown even faster ad tey 1 Argentina as since collapsed, and Uruguay and Bolivia (te oter successes ) ave anemic growt rates tat look good only in relation to an even worse performance in te earlier period. Mexico, Brazil, and te oter Latin American countries ave yet to attain te growt rates tey experienced during India was te world 10 t fastest growing economy in te 1980s, and te 12 t fastest in te 1990s. Cina was second (beind Korea) in te 1980s and te first in te 1990s.

5 3 embarked on a more ortodox reform agenda. But te difficulty wit tis argument is tat tose countries tat adopted te ortodox policies suc as Latin American countries for te most part did worse tan Cina and India, not better. 3 We entertain a different possibility in tis paper. We consider a world in wic appropriate policies and institutional arrangements ave a large element of specificity, and experimentation is required to discover wat works locally. Reforms tat succeed in one setting may perform poorly or fail completely in oter settings. Two-track reform may work well in Deng s Cina but not in Gorbacev s Soviet Union. Gradualism may be appropriate to India, but not Cile. Import-substitution may foster competitive industries in Brazil, but not in Argentina. Industrial policy may produce results in Sout Korea, but not in muc of Africa. Export processing zones may work wonders in Mauritius, but not in most oter countries tat ave created tem. Privatization of utilities may be necessary in Latin America, but not Asia. Openness to capital flows may enance efficiency and growt in ric countries but wreak avoc in poor countries. Centralized wage bargaining may be feasible and desirable in Europe but not in Latin America. Suc specificity could arise from differences in istorical traectories, geograpy, political economy, institutional settings, or oter initial conditions. It could elp explain wy successful countries Cina, India, Sout Korea, and Taiwan among oters ave almost always combined unortodox elements wit ortodox policies. It could also account for wy important institutional differences persist among te advanced countries of Nort America, Western Europe, and Japan in te role of te public sector, te nature of te legal systems, 3 One migt also point to te growt miracles of earlier decades, suc as Sout Korea and Taiwan countries tat also employed a mix of ortodox and unortodox policies. On te ortodox side, tese countries empasized fiscal conservatism, exports, and uman resources. On te unortodox side, tey protected teir domestic markets and made extensive use of industrial policies.

6 4 corporate governance, financial markets, labor markets, and social insurance mecanisms, among oters. We do not mean to suggest tat economic principles work differently in different places, or tat economics itself needs to be tailored to local conditions. We make a distinction instead between economic principles and teir institutional embodiment. Most first-order economic principles come institution-free. Incentives, competition, ard-budget constraints, sound money, fiscal sustainability, property rigts are central to te way tat economists tink about policy and its reform. But tese principles do not map directly into institutional solutions. Property rigts can be implemented troug common law, civil law, or, for tat matter, Cinese-type socialism. Competition can be maintained troug a combination of free entry and laissez-faire, or troug a well-functioning regulatory autority. Macroeconomic stability can be acieved under a variety of fiscal institutions. Institutional solutions tat perform well in one setting may be inappropriate in oter setting witout te supporting norms and complementary institutions. In te words of Nort: economies tat adopt te formal rules of anoter economy will ave very different performance caracteristics tan te first economy because of different informal norms and enforcement. Te implication is tat transferring te formal political and economic rules of successful Western economies to tird-world and Eastern European economies is not a sufficient condition for good economic performance. (Nort 1994, 8) 4 In addition, since policy makers always operate in second-best environments, optimal reform traectories even in apparently straigtforward cases suc as price reform cannot be designed 4 See for example Kanna, Kogan, and Palepu (2001) on convergence in corporate governance practices around te world in te last two decades. Tese autors find some evidence of convergence in form, but no evidence of convergence in practice, indicating tat implementation as been problematic in varying local conditions. Pistor (2000) provides a general treatment of te issue of legal transplantation, and sows ow importation of laws can backfire. On te role of te WTO in policy convergence in te area of trade, see Morissey and Nelson (2001).

7 5 witout due regard to prevailing conditions and witout weigting te consequences for multiple distorted margins. A paper tat takes tese issues seriously as a triple burden. It must explain wy countries do converge on consensus policies even wen teir circumstances call for different arrangements. Second, to te extent tat some countries coose to experiment rater tan imitate, it must provide a reason wy tey do so. Finally, it must provide a plausible accounting of te pattern of economic performance tat emerges wen tere is a tendency for countries to converge on similar policies. Tis paper contains some first steps in all tree directions. We consider a model in wic policies (or equivalently, institutional arrangements) can vary on two dimensions. One, wic we refer as appropriateness, corresponds to te fit between policies and a country s circumstances. Policies tat are well matced wit circumstances produce iger output tan policies tat are not. Te second dimension, wic we refer to as transparency, relates to te conduciveness of policies to skimming and corruption by politicians. Some policies are transparent, and terefore are relatively immune to manipulation for corrupt purposes, wile oters are not. We take transparency to be independent of a country s circumstances, in te sense tat a policy tat is transparent in one country is transparent in all oter countries. We also assume tat all appropriate policies are transparent, wile all transparent policies need not be appropriate. Te ustification is easy to provide. It is ard to see wy non-transparent policies, in te sense of policies tat are easily corruptible, could ever really be appropriate. On te oter side, one could always limit corruption by legislating a rule tat proibits all policy making; tis would root out corruption by eliminating discretion, but it is ard to see ow it could be appropriate in general.

8 6 Governments, in turn, can be of two types: onest and corrupt. Tey face electorates tat are more likely to trow tem out of office if tey are perceived as being of te corrupt type. Governments receive a private signal about teir country s underlying state of te world tat is not observed by teir electorates. We focus on te stage game were te governments and teir electorates in a follower country ave already observed a successful policy in a neigboring country (called te leader ). We assume tat te leader s policy is so successful tat te electorate in te follower can deduce te policy is of te transparent type. Te follower government ten faces tis dilemma. It can imitate te leader s policy, and tereby communicate tat tere will be no corruption, but at te cost of adopting a policy tat may not be appropriate. Or it can coose a policy tat targets as closely as possible its private signal, in wic case it incurs te costs of experimentation (along wit te cost of being perceived as corrupt). Our model captures several elements tat we believe are plausible. First, and key to our argument, is te idea tat tere is considerable context-specificity to desirable policies. Second, we also incorporate a universal element to policies, wic we model wit te notion of transparency. Tird, we assume tere is an element of uncertainty inerent in a country's searc for te ideal policy. Governments eiter lack perfect knowledge on te rigt course of action, or face uncertainty in carrying out te actual implementation of a policy. We capture tis aspect of policy making by assuming tat tere is uncertainty in implementing an untested policy. Fourt, our framework incorporates te obvious fact tat some governments are more committed tan oters to searcing for policies tat are in te long-term interest of a country. Te sadow of corrupt governments imposes a cost on onest governments. Finally, our framework captures an important aspect of policy coices in a globalizing world, namely te existence of informational externalities. In particular, countries observe te success or failure of alternative

9 7 pats of development and ave te option of mimicking tese policies in te ope of matcing teir success. Our framework is ric enoug to capture all of tese tensions inerent in policy making. We sow tat te informational externality created by successful leaders results in bot an upside and a downside. Countries wose underlying state is close to te leader i.e., te leader s neigbors by te relevant metric coose to mimic te leader s policies, even wen teir governments are of te corrupt type. Tis yields a double benefit to te neigbors: tey can forego te costs of experimentation, wile reaping te benefits of te discipline tat is exercised on potentially corrupt governments. Countries in te far peripery are unaffected by te leader s example, as governments of bot types coose to experiment. Te cost is borne by onest governments in te near peripery i.e., by countries tat are too far from te leader for mimicking to be welfare improving, but close enoug for te informational externality to generate an incentive for onest governments to mimic. Te last case entails inefficient disciplining of government policies. Terefore our model yields distinct predictions about te patterns of policy imitation, corruption, and economic performance as a function of a country s position vis-à-vis successful leaders. In particular, it predicts a U-saped pattern in economic performance as we move away from te leader in te relevant space of caracteristics: close neigbors sould do very well, distant countries moderately well on average wit considerable variance, and intermediate countries worst of all. We believe tis framework elps account for some of te salient features of te economic landscape of te last few decades. Countries in te vicinity of growt poles suc as Japan and later te East Asian tigers ave tended to do very well, in part by imitating many of te policies

10 8 followed by tese leaders. Countries wose underlying caracteristics or geograpic distance place tem very far from te leaders ave experienced igly variable fortunes. Superstars like Cina and Mauritius, wic ave acieved success on te back of a great degree of policy experimentation, ave occasionally emerged from tis group. Some of tese experimenters in turn ave become examples for oters to follow, suc as Cina in te case of Vietnam. And we would claim tat many countries in between ave been inefficiently disciplined, adopting policies tat are surely less corrupt and more transparent tan in te past, but also peraps less appropriate to teir circumstances. Te Latin American economies of te 1990s may constitute te cief examples of tis last group. We also undertake a simple empirical test of our framework in tis paper, by focusing on te experience of post-socialist countries. Tese countries were forced to searc for alternative policies once tey abandoned socialism, so tey constitute a useful sample for our purposes. For most of tem, te model to emulate, if any, was te Western European example. We ypotesize tat te geograpic distance between eac of tese countries and Western Europe is an adequate (inverse) proxy for te suitability of European-style institutions to teir circumstances. We ten demonstrate tat tere is a robust U-saped relationsip between tese countries distance from Brussels and teir post-transition growt rates. Strikingly, te U-saped relationsip survives wen we control for a variety of oter determinants, including te extent of structural reform undertaken. Wile some previous studies ave noted te negative gradient in performance in te vicinity of Western Europe, we are not aware of any tat ave picked up te upwards-sloping part of te relationsip, nor of any teory (oter tan ours) tat would account for te latter. Measures of corruption and policy imitation also beave broadly in te way te teory predicts.

11 9 Wile te sample of countries covered is necessarily small, we are greatly encouraged by te strengt of tese results. Tere are many antecedents in te literature to tis line of tougt. Economic istorians ave long empasized tat te fragmentation and diversity of early modern Europe was a source of economic strengt tat allowed Europe to eventually overtake te centralized empires in Cina, te Middle East, and te Indian sub-continent. Central ere is te idea tat decentralization enabled competition, not only in te market for goods and services, but also in market for institutional arrangements. As David Landes puts it, only societies wit room for multiple initiatives, from below more tan from above, could tink in terms of a growing pie (1998, 32). Wit reference to te failure of Cina and Islam, Douglass Nort points out: Centralized political control limits te options, te alternatives tat will be pursued in te context of uncertainty about te long run consequences of political and economic decisions. In [te] competitive decentralized environment [of Europe] lots of alternatives were pursued; some worked, as in te Neterlands and England; some failed, as in te case of Spain and Portugal; and some, suc as France, fell in between tese two extremes. But te key to te story is te variety of te options pursued and te increased likeliood (as compared to a single unified policy) tat some would turn out to produce economic growt. (Nort 1993) Jared Diamond (2001) as empasized te geograpic sources of tese different istorical traectories. 5 Ironically, decentralized experimentation appears to ave taken root in Cina in te last tree decades, and is often pointed to as te source of tat country s recent success (Naugton 1995; Jefferson and Rawski 1994). 5 Cina as a smoot coastline. Europe as an indented coastline, and eac big indentation is a peninsula tat became an independent country, independent etnic group and independent experiment in building a society: te Greek peninsula, Italy, te Iberian peninsula Europe is transected by mountain ranges tat split up Europe into different principalities: te Alps, te Pyrenees, Carpatians Cina does not ave mountain ranges tat transect Cina. In Europe big rivers flow radially.and tey don t unify Europe. In Cina te two big rivers flow parallel to eac oter, are separated by low lying land, and were quickly connected by canals. For tose geograpic reasons, Cina was unified in 221 B.C. and stayed unified most of te time since ten, wereas fore geograpic reasons, Europe wasn t. Augustus couldn t do it, Carlemagne couldn t do it and Napolean and Hitler couldn t unify Europe (Diamond, 2001).

12 10 In te recent literature on development strategies, Evans (2001) and Rodrik (2000) ave argued for te importance of local deliberation in ensuring a good fit between institutions being adopted and local conditions. Evans dubs te 1990s convergence on Anglo American institutions te institutional monocropping approac. Besley (2000) as also empasized te importance of policy experimentation, noting tat adopting a policy is analogous to adopting new tecnologies wit uncertain benefits. Dewatripont and Roland (1995) and Roland (2000) underscore te importance of uncertainty and experimentation in te context of transition economies. Unger (1998) as forcefully argued tat te institutional arrangements we observe presently among te successful advanced industrial countries constitute only a subset of te full range of institutional possibilities. Peraps most directly relevant to our researc is a recent paper by Berkowitz, Pistor, and Ricard (2001). Tis paper analyzes te istorical process of legal evolution in forty-nine countries to ascertain wy some legal systems are more effective tan oters. Te autors dependent variable is te quality of legal institutions and enforcement (wic tey call legality ). Te paper s central argument is tat te manner in wic te legal order develops makes a significant difference to legality. In particular, te autors make a distinction between two types of countries: tose tat developed teir formal legal orders internally ( origins ), adapted imported codes to local conditions, or ad familiarity wit foreign codes ( receptive transplants ), and tose tat acquired teir formal legal order from abroad witout muc adaptation and/or familiarity ( unreceptive transplants ). Teir main finding is tat te second group of countries as ended up wit significantly lower levels of legality. Equally important, tey find tat te unreceptive transplant effect is a more important determinant of legality, and ultimately of economic growt, tan te family from wic te legal regime was drawn. Wat

13 11 seems to matter is not common versus civil law, but te suitability of te laws to local conditions. Our paper is also related to Kremer, Onatski, and Stock (2001), wo analyze te statistical properties of te distribution of per-capita incomes around te world and present a simple framework to account for it. Tey too consider te role of experimentation in driving te pattern of incomes around te world. In particular, in order to explain wy some poor countries grow ric wile ric countries ardly ever get poor, tey develop a model in wic countries searc among policies until tey reac a satisfactory income level, at wic point tey stop experimenting. Teir framework is based on te assumption tat te requisite policies are country-specific, but tey also discuss informally te likeliood tat countries may learn from eac oter. Our framework puts more structure on tese issues, and also, more importantly, endogenizes te coice between experimentation and imitation. Te plan of te paper is as follows. Section II presents a stripped-down version of te model to drive ome te central message of te paper as directly as possible. Section III describes te full model, providing political-economy underpinnings to te coice of policy by a follower government. Section IV presents an empirical test wit te post-socialist sample of countries. Section V presents some concluding remarks. II. Te Bencmark Model We describe te complete model in two stages. We present in tis section te essential features of a minimal bencmark model, in order to illustrate te dilemma at te eart of our argument in a transparent way. In te next section we furter elaborate on tis model and provide political underpinnings for te government s decision making.

14 12 We assume tat te governments in all countries are divided into two coorts a leader and a follower coort. A government in a follower country as one of two policy coices: it can coose to imitate or mimic te policy cosen by a leader country, or it may prefer to experiment wit its coice of policy. Tis decision is a function of several factors, wic we describe below. Countries: Location and Policy Specificity Te world is made up of a large, discrete number of countries. Countries differ from eac oter in terms of teir geograpy, istorical traectory, culture and oter local conditions, wic make any given policy more or less appropriate across countries. We capture tese country specific local conditions by assuming tat eac country i as a unique state of te world z i. We assume tat countries are distributed uniformly on a circle, wose circumference we normalize to two units. Terefore eac country i as a location z i, on tis circle (Figure 1). Reducing all differences in underlying caracteristics across countries to a single dimension makes it easy to assess weter countries are proximate or distant from eac oter. In particular, if two countries are located at z i and z, ten te distance between tese countries is defined as i = z i - z. Te government in any given country gets to coose a policy a tat affects national income. Te important feature of a policy tat we empasize is its specificity or statecontingent nature, wit te impact of a policy on national output depending on te country s location. If a country wose underlying location is z i cooses te policy a, ten national income is given by y i 2 = θ ( a zi ). As is clear from tis formulation, te closer or more appropriate is a policy to a country's underlying state; te iger is national income. Terefore,

15 13 if two countries and k are neigbors (i.e. k is small), ten it is less costly to adopt te policy of a neigbor. Te Government, Information and Policy Coices We are interested in capturing te uncertainty inerent in te policy making process. Tere are two ways of modeling tis: eiter we assume tat te government is imperfectly informed or we can assume (as we do in tis paper) tat te government as imperfect control over te policy implementation process. For simplicity, we follow te latter route and assume tat te government receives a perfectly reliable private signal, z ˆ = z. Tis would seem to i i suggest tat a government's ideal policy coice is a simple matter: matc te policy to te appropriate state. However, even if government i knows te ideal policy, it can determine te actual policy implemented only up to a random error term, a i = zi + η, wit η distributed wit 2 mean zero and variance σ. Tis is a relatively simple way to represent te costs as well as te benefits of experimenting wit a policy tat no oter country as successfully tried before. On te oter and, if country i imitates and picks up an off-te-self policy tat as been tried and successfully tested in some oter country, ten we assume tere is no uncertainty associated wit its implementation. 6 Given our assumptions so far, and assuming tat te follower government maximizes (expected) national income, te coice of experimentation versus imitation is a simple function of two factors: te distance between te follower country and te successful leader and te 6 We ave made tese assumptions in order to capture te gains and costs of experimentation in policy making in te simplest possible way. An alternative more intuitive (toug more complicated to work wit) assumption is tat governments receive noisy signals about z i. In tis case in order to learn about te true z i, te government will ave to experiment and will gradually converge to te ideal policy a i = z i. Te crucial aspect of te assumption is tat te government as more reliable information about its true location, tan te citizen does.

16 14 uncertainty associated wit policy experimentation. If te distance is large relative to te uncertainty, ten te government will prefer to experiment. We now introduce an additional complication, wic drives a wedge between te government s utility function and overall social welfare. We assume tat te government s preferences are given by v i = y λ K, were K>0 i is a private (not social) fixed cost borne by te government and λ is a dummy variable tat equals zero or one, depending on weter te government imitates (λ = 0) or experiments (λ = 1). Hence te government incurs tis private cost only wen it cooses to experiment. For te moment we simply take as given te existence of tis private cost. In te next section we will provide microfoundations for tis assumption, based on political-economy considerations. To Experiment or to Imitate: A Heuristic Analysis Suppose tat country 1 from te first coort becomes a successful leader by implementing policy a 1, suc tat a 1 =z 1 and y 1 = 0, tereby acieving te igest possible output. We now analyze te policy dilemma confronting a follower country tat is located at z 2, Te government of country 2 observes policy coices of te first coort including tose made by te successful leader. In addition te government receives a private signal about its country s location z ˆ2 = z2. If te government imitates te successful leader s policy coice a 1, its payoff equals v 2 (a 1 =a 2 )= y 2 = -θ( a 1 - z 2 ) 2. Tis payoff is solely a function of te distance between te leader and te follower country. In contrast, if it cooses to experiment and follow a policy in accordance wit its own private signal, te payoff to te government is given by ' v a = a = ˆ ) = E(y) - K = E{-θ( z 2 +η - z 2 ) 2 } K = -θσ 2 - K. Here te government incurs 2 ( 2 2 z2 bot te cost of an uncertain tecnology as well as te private fixed cost of experimenting. Follower country 2 will prefer to imitate te successful leader rater tan experiment if te

17 15 ' following inequality olds: v a = a ) v ( a = a = ˆ ). Tis implies tat te follower 2 ( z2 government will coose to imitate if z 2 lies in te interval [a 1 -(σ 2 +K/θ) 1/2, a 1 +(σ 2 + K/θ) 1/2 ], and to experiment oterwise. In te absence of te private fixed cost, te interval of imitation would ave been [a 1 -σ, a 1 + σ] instead. Clearly, te larger te government s private fixed cost, te larger te zone of inefficient imitation te zone witin wic a country ends up imitating even toug it would ave been better off experimenting. Despite te very simple structure of our model, we are able to generate some fairly striking implications. Te pattern of economic performance tat results can be summarized wit te elp of Figure 3, wic we will refer to again wen we lay out te full model in te next section. In particular, our framework yields a U-saped relationsip between economic performance and distance from te leader. Specifically: (i) In te immediate neigborood of te successful leading country, countries prefer to imitate te leader s policies, and acieve ig economic performance. Tere is a growt pole around te successful leader. (ii) Countries tat are very far from te leader located in te zone we call te farperipery coose to experiment, rater tan imitate. Teir economic performance is on average worse tan tat of countries in te close neigborood of te leader. But tese also exibit muc greater variance in performance (due to te uncertainty in te policy implementation process). So some experimenters could acieve better performance tan imitators in te neigborood of te leader. (iii) Countries located in te intermediate zone between tese groups countries in te nearperipery are strictly worse off tan bot of tose groups (compared to a situation

18 16 were tere was no successful leader). Tese are countries wose governments coose to mimic, even toug tey are too distant from te leader to benefit. (iv) Extending te model dynamically (if informally), we can see tat growt poles are likely to develop sometime down te line in te far peripery, but not in te center or te nearperipery. Tat is because experimentation takes place only in te far peripery. Wile te above analysis is somewat euristic, it as te merit of delineating most of te key results of te full model in a transparent manner. Te key to te U-saped pattern is te inefficiency tat arises from te government s private cost, K. In te next section, we dispense wit K, and extend te model to provide a direct explanation as to wy governments may incur a private cost if tey coose to experiment rater tan imitate. III. Political Underpinnings of te Searc for Prosperity: te Full Model Transparency, Corruption and Policy Universality It is useful to keep in mind tat in our analysis a policy as very broad connotations and is not necessarily ust related to te economic policy coice per se. A policy may consist of any institutional arrangement tat affects a country s national income and welfare. Tis broader interpretation of policies becomes particularly relevant in understanding te additional aspect of a policy tat we now introduce. We assume tat some policies or institutional coices are conducive to te inefficient siponing away of national income (i.e. troug corruption and wastage), wile oters encourage discipline. We call tis element of policy "transparency," to distinguis it from te notion of "appropriateness" developed in te preceding section. We assume tat all appropriate policies are transparent in tat tey encourage discipline and prevent wastage and corruption. Hence any transparent policy tat disciplines one country is assumed to

19 17 discipline te government of any oter country tat adopts it. However, all transparent policies are not appropriate in tat tey need not be well matced wit te underlying state of te country in question. 7 Our framework terefore encapsulates in a relatively parsimonious form two aspects of policies. First, our formulation captures te notion of specificity tat policies need to be suited to country-specific local conditions. Second, we also capture aspects of policies tat are universal, in tat greater transparency, te rule of law and accountability are useful under all conditions regardless of local context. Government and te Citizen Everyting else being te same, all governments would normally prefer to boost national income and enance te well being of teir citizens. However, governments ave additional motives as well. Tey prefer remaining in power, and, furtermore, some are corrupt and beolden to special interests. We capture tese additional motivations for a government of type t k from country i by using te following utility function: v y + λ R + ε, were ε > 0 ik = i k represents te ego rents from being in office, and were governments can be onest (t ) wit probability p or corrupt (t c ) wit probability 1-p. Te economic rents from aving a nontransparent policy tat allows corruption is given by R k, were R c > R 0. For reasons tat will become clear sortly, we sall assume tat eac period is made up two stages. Te economic and ego rents bot accrue to te government at te end of te second stage of te period. 7 Tis assumption is muc stronger tan wat is required for te analysis tat follows. All tat is required is tat te citizens ave te perception tat (in expected terms), any 'appropriate' policy is more likely to be 'transparent'. For example, all tat is required for is tat te citizen in Hungary perceives tat EU style institutional arrangements, if adopted by is country, will reduce corruption.

20 18 Similarly, te impact on national income of aving a corrupt government becomes visible to te citizen only at te end of te second stage. 8 Finally, eac country i as a single representative citizen, wo lives one two-stage period t = 1 2 t 1 and as preferences given by = [ y R ] u δ, were δ is te discount factor and t is te i it stage. However, tis representative citizen does more tan ust consume national income. He is in a position to organize a coup, a revolt or to force mid-term elections at te end of te first stage, if e is dissatisfied wit te incumbent government. Te probability tat a citizen will successfully replace an incumbent is a linear function of is perception tat resources will be siponed away. We furter elaborate on tis subsequently. kt Te Timing in te Complete Model Te entire set of countries is divided into two coorts, were countries tat make policy coices at te beginning of te first period constitute te first period coort. Te governments in tese countries receive private signals and coose policies. Once selected, policies remain in place during bot stages of te period, regardless of weter a government remains in office trougout. At te end of te first stage, citizens in tese countries observe te government's policy coice and coose weter to retain or attempt an ouster of te government. At te end of tis second stage output is publicly observed and payoffs are realized to te government and te citizen. Citizens and governments in te second follower coort observe policy coices of te first coort and teir national income. Te government receives a private signal about te state 8 Tis assumption simplifies te analysis of te political game since it ensures tat te citizen s attempt to replace te government is a function only of observed policy coice. Te qualitative direction of te results will remain

21 19 and ten makes its policy coice at te beginning of tis second period. Once again citizens observe te policy coices and make a decision of weter to retain te incumbent or replace it at te end of te first stage of tis second period. Te preceding sequence is repeated. Tis timeline is illustrated in Figure 2. We now examine te equilibrium policy coices of a government. As will soon be evident, te results are similar to te euristic analysis in te last section. Discipline and Corruption in te Searc for Prosperity: Equilibrium Analysis We analyze a government s coices in te following scenario. We consider a situation were at te end of te first period, tere as emerged a sole successful leader tat as successfully implemented policy a 1. Tis perfect matc of te transparent policy a 1 to te appropriate state z 1, as resulted in te maximum possible national income of 2 ( a ) 0 y θ, were we assume a 1 = z 1 = 1 (see Figure 1). 9 1 = 1 z1 = We ten focus on te dilemma faced by te follower government. Sould it minimize uncertainty by picking an off-te-self policy a 1 or sould it experiment in its coice of policy? In order to facilitate our analysis, we propose te following equilibrium to te sub-game involving te follower governments. All onest follower governments wose private signal ẑ i lies in te interval [ z 2 ], will be disciplined into imitating te policy cosen by te z successful leader. Similarly, corrupt governments wo receive private signals in te range uncanged even if we allow for observability of interim ouput by te citizen, so long as its realization(or its observability to te citizen) is sufficiently noisy. 9 Allowing for te process of policy experimentation in te first period, to result in more tan one successful leader, does not qualitatively alter any of te implications of te analysis tat follows.

22 [ z 2 ] c z c 20,, will also be disciplined into mimicking te successful leader, were z < z c. 10 All oter countries wit onest governments (e.g. tose wit z [0,z )), will pursue a policy in accordance wit teir private signal, wile corrupt governments in te range z [0, z c ) will pursue policies wic toug positively correlated wit teir private signal, do allow te siponing away of resources. Citizens will keep in power tose governments tat are disciplined into pursuing incorruptible policies. Governments tat pursue any oter policy will, wit positive probability, be successfully replaced at te end of te first stage by a randomly drawn callenger. Given tat we ave a finite two-stage game, we analyze our proposed equilibrium backwards, from te beginning of te second stage of te second period. At tis stage te citizen as to decide weter to attempt to replace or retain te government. Depending on te political structure specific to te country e may try to force a mid-term election or carry out a revolt. Te citizen will attempt to replace te government if e believes tat is second stage utility will be iger wit a randomly drawn government tan wit te incumbent. Remember tat, once implemented, a policy is in place for bot stages of a period. Terefore, if te policy in place is not transparent, ten in te second stage a corrupt government can sipon away resources and lower national income. In tis case, if te incumbent is perceived to be corrupt, ten te citizen may prefer to replace te incumbent wit a randomly drawn callenger. On te oter and, if te policy in place is transparent and guarantees discipline, ten te citizen is no better off by replacing te incumbent, wo will terefore be retained Since in our proposed equilibrium, tere are always a non-empty set of governments tat prefer to not imitate te leader, we do not worry about out-of-equilibrium beliefs. 11 Tis is always true since under te proposed equilibrium, te difference in te citizen s second stage payoff from aving a government tat pursues its own signal and replacing it wit a randomly drawn incumbent equals: [1- z p /(z p + z c (1-p ) ) ]R c - (1-p )R c, wic is strictly negative for all p >0.

23 21 Consider te payoff from mimicking obtained by an onest follower government tat is on te margin of indifference between mimicking and pursuing its own private signal ẑ, were 2 ˆ. Tis is given by θ ( z a ) R + P t a ) ε z = z 1 ( = a1, were P(t a =a 1 ) is te probability tat an onest government tat imitates te leader will remain in power. Given tat te countries are uniformly distributed on te circle, 12 te above simplifies to: v ( z, a 2 2 ( z a ) R + ε = θ ( z ). = a1) = θ 1 a1 + ε Te first expression on te rigt and side is te national income tat results from te government's policy coice. Since we assume tat R equals zero, tere is no loss of economic rents if te incumbent is onest. Furtermore, since te government as been disciplined into enacting a transparent policy (in place for bot stages), te citizen-voter as no incentive to replace te incumbent wit a randomly drawn callenger. Hence te incumbent retains power wit probability one (i.e. P(t a =a 1 ) =1 ) and earns ego rents of ε. Similarly, te (expected) payoff to an onest government from experimenting and pursuing a policy a, in accordance wit its private signal ẑ is given by: v ( z, a 2 2 [ θ Ε( z z ) ] + P( t a a ) ε = [ θσ ] z p a1 ) = 1 + ε. z p + zc ( 1 p ) A government faces uncertainty wen it pursues an untried policy. Terefore, if te tecnology of policy making is poorly understood, tis uncertainty results (in expected terms) in a loss of output, wic gives rise to te term in te first square bracket. 13 Te second term is te expected 12 Witout loss of generality, te analysis tat follows only considers te policy coices of countries uniformly distributed on alf te diameter of te circle i.e. tose located in te range [0,1]. By symmetry, including te complementary set of countries located on te oter alf, will not alter our results. 13 Since policies are cosen from a continuum, we assume tat even if a particular policy is incorruptible, tere exists a policy tat is arbitrarily close wic a corrupt government may prefer to take, since it allows diversion of resources. Terefore a government of type k does not ave to lose economic rents, R k wen it cooses to experiment.

24 22 payoff from remaining in power. As is evident, te probability of remaining in power is a linear function of te probability of being perceived onest. Tis implies tat, given uniformity of te underlying distributions, te probability of being re-elected is given by te expression in te second square bracket. Terefore, te net payoff to an onest government tat prefers to imitate rater tan pursue its private signal z ˆ = z is given by: Z( t; a1, a ) = v ( z, a = a1) v ( z, a a1), wic simplifies to: Z 2 [ R + ε ] ( t ; a, a ) = ( z a ) 2 z pε θ 1 θ σ + 0. (1) z p + zc ( 1 p ) 1 = Similarly, te net payoff to a corrupt government from mimicking rater tan pursuing a policy in accordance wit its private signal z ˆ = z is given by, c c Z 2 [ R + ε ] ( t ; a, a ) = ( z a ) c 2 z pε θ c 1 c θ σ + 0. (2) z p + zc ( 1 p ) 1 c = Te above expressions capture te payoffs to a follower government from pursuing alternative courses of action. We can now summarize our basic results on policy coice. Proposition I. Tere exists an equilibrium z *< z c *, suc tat te policy coices made by a follower country are a function of its location z, and weter its government is corrupt or onest, suc tat if te successful leader is located at z 1 = 1, te following is true: (a) all countries run by onest governments wit z [ z *,1 ] are disciplined into imitating policy * a 1, wile tose wit z [0, ) experiment and pursue policy a in accordance wit teir private signal, z However, since voters perfectly observe te policy coice of te government, te government will be vulnerable to overtrow and losing ego rents, even if te policy coice of te government is arbitrarily close, but not equal to a 1.

25 23 (b) all countries run by a corrupt government wit z z,1], are disciplined into imitating a 1, * wile tose wit z [0, ), experiment and pursue a policy a in accordance wit teir private z c signal and earn rents R c. Proof: See Appendix. [ * c Tis proposition establises te existence of an equilibrium were some countries mimic and oters pursue teir own course of action. Te mecanism is a simple one. Wile governments are interested in enacting policies tat enance national income, tey would also like to remain in power. All governments would prefer to signal troug teir policy coice tat tey are relatively onest, and ence increase teir cances of remaining in power. Tis gives rise to te possibility of governments deliberately coosing to imitate a successful leader, even toug suc a policy migt result in lower national income. It is important to keep in mind tat te precise mecanism tat generates tis incentive to imitate is not important. In reality, alternative mecanisms may be at work so long as not imitating results in a private cost of some kind (as in te previous section). For instance, suppose international capital markets expect tat policies tat constitute te Wasington consensus are more likely to work for developing countries. If a government as information tat suggests an alternative independent course of action migt be superior, it is likely to be punised in te form of low capital inflows. In turn tese lower capital flows, by lowering employment, may ave a negative impact on te political survival of te government. Terefore, fear of lower capital flows (and its adverse political and economic consequences) may prevent governments from pursuing policies tey know are likely to work best See Mukand (1999) for an elaboration of tis argument.

26 24 We are particularly interested in evaluating te welfare impact of tis disciplining of nations. To facilitate our analysis, we first caracterize as a bencmark te socially optimal policy coice, equivalent to te policy coice made by an onest government in te absence of any global informational externality. If te citizens of a country cannot observe te policy coices (or te economic outcomes) in te rest of te world, ten tey are not in a position to make any assessment of te onesty or corruptibility of te incumbent merely by looking at its policy coice. Terefore, in te absence of an informational externality, a onest government, wic as received te private signal z ˆ = z, will prefer to imitate te successful leader, if 2 Z, wic is true iff θ ( 1 z ) θ σ t ( ) 0 2. Terefore, te socially efficient policy coice implies tat a government prefers to imitate te successful leader so long as: z eff 2 z = a σ R θ = 1 σ, since R equals zero and a 1 =1. As is evident, te set of 1 onest governments tat will imitate te policy coice of a successful leader is increasing in te uncertainty associated wit te implementation of te new policy. In wat follows, we label all countries, wit underlying states of te world eff z z, as being neigbors of te leader country. Te country-specific socio-cultural or geograpical conditions of suc a neigbor are relatively similar to tat of te leader. In contrast countries tat are in te region eff z < z, are labeled as being located in te peripery of te leader country. Terefore, in te absence of te global informational externality an onest government beaves in a socially efficient manner wile a corrupt government engages in too little imitation. In te next proposition, we demonstrate tat te global informational externality exerts a disciplinary influence on all governments and results in te following: z decreases for onest

27 25 governments, wic get inefficiently disciplined, wile z c also decreases for corrupt government, wic get efficiently disciplined. Proposition II. Te relationsip between national income and a follower country's location is given by te following: (i) all governments (weter corrupt or onest) in te range z * [ eff,1), tat are c z neigbors of te successful leader, get efficiently disciplined into imitating te successful leader and ave average incomes tat are iger tan te oter countries; (ii) te set of onest governments located in te near-peripery i.e. te range z [ z *, z eff ), are inefficiently disciplined into imitating te leader and enact policies tat lower national income, wile te set of corrupt governments in tis range experiment; * (iii) all governments in te far-peripery i.e. z [ 0, ), experiment wit teir coice of z policies and ave iger average incomes tan tose in te near-peripery. Proof: See Appendix. We examine facets of te above proposition by studying te associated Figures 3 and 4, were we observe te following: Region A. Neigbors: Location is z [ eff,1]. Tere exists a growt pole in te close z proximity of te successful leader. Governments in te close proximity of te leader, weter corrupt or onest, mimic its policies. Here, we sould expect tat national income gradually comes down wit distance, as te ideal policy of a mimicking country moves furter away from te leader s policy coice a 1. Moreover, in tis group of follower countries, te variance in