Københavns Universitet. Serving the Public Interest Markussen, Thomas; Tyran, Jean-Robert Karl. Publication date: 2010

Transcription

1 univesity of coenhagen Københavns Univesitet Seving the Public Inteest Makussen, Thomas; Tyan, Jean-Robet Kal Publication date: 2010 Document Vesion Publishe's PDF, also known as Vesion of ecod Citation fo ublished vesion (P): Makussen, T., & Tyan, J-R. (2010). Seving the Public Inteest. Deatment of Economics, Univesity of Coenhagen. Download date: 28. ma

2 Discussion Paes Deatment of Economics Univesity of Coenhagen No Seving the Public Inteest Thomas Makussen, Jean-Robet Tyan Øste Faimagsgade 5, Building 26, DK-1353 Coenhagen K., Denmak Tel.: Fax: htt:// ISSN: (online)

3 Seving the Public Inteest Thomas Makussen and Jean-Robet Tyan * il 2010 We esent a model of olitical selection in which votes elect a esident fom a set of candidates. We assume that some of the candidates ae benevolent and that all votes efe a benevolent esident, i.e. a esident who seves the ublic inteest. Yet, olitical selection may fail in ou model because votes cannot easily tell benevolent fom egoistic candidates by obseving thei e-election behavio. Egoistic tyes may stategically imitate benevolent tyes in the e-election stage to extact ents once in office. We show that stategic imitation is less likely if the olitical system is likely to oduce good govenance. That is, if benevolent candidates ae common, if the esident has little discetionay owe, and if the ublic secto is effective. We analyze the ole of institutions like investigative media and e-election and show that they can imove o futhe hame olitical selection, deending on the aametes of the olitical game. Keywods: Political selection, elections, social efeences, olitical leadeshi JEL-codes: D64, D72, D82, H0 * Both authos ae at Univesity of Coenhagen, Deatment of Economics, Øste Faimagsgade 5, DK-1353 Coenhagen. Thomas.Makussen@econ.ku.dk, Jean-Robet.Tyan@econ.ku.dk We gatefully acknowledge financial suot by the Danish Science Foundation (FSE), unde Poject Cooeation and Institutions and thank David Deye Lassen, ntonio Nicolo, Maco Piovesan and lexande Sebald fo helful comments.

4 1 Intoduction Economists have taditionally assumed that all layes in the olitical ocess, including olitical leades, ae stictly self-inteested. ccoding to this view, olitical leades ae not intinsically benevolent but thei olicy choices may o may not be in line with ublic inteest, deending on how the olitical institutions shae thei incentives. It is theefoe cucial to imose institutional ules that estain egoistic leades fom extacting ents (e.g. Bennan and Buchanan 1980). We agee that olitical leades often behave in self-seving ways and that institutions need to be set aoiately. Howeve, we disagee with the standad assumtions which, in the wods of James M. Buchanan 1, ae that all individuals must be modeled as seeking to futhe thei own naow self-inteest, naowly defined, in tems of measued net wealth osition, as edicted o exected and that thee is no suggestion that imovement lies in the selection of moally sueio agents who will use thei owes in some ublic inteest. In fact, the uose of ou ae is exactly to analyze the difficulties of selecting moally sueio candidates we call them benevolent into office. s exlained below, ou focus builds on insights fom behavioal economics showing that eole have heteogeneous social efeences, and contibutes to an emeging liteatue on olitical selection. In this ae, we assume that some of the candidates fo olitical office ae benevolent and that many ae egoistic. Once elected, the esident has much discetionay leeway ove a budget, and can use it to ovide a ublic good o fo his ivate benefit. By assumtion, egoistic esidents seek to futhe thei own naow self-inteest which, at least in the baseline vesion of ou model, means total ent aoiation by the esident. In contast, a benevolent esident sends the entie budget to ovide a ublic good which is both efficient and fai. In that sense, a benevolent esident uses his owes to seve a well-defined ublic inteest. Given these assumtions, egoists efe to goven athe than to be govened, and all votes obviously efe to be govened by a benevolent athe than an egoistic esident. Howeve, the votes oblem is that candidates tyes cannot be diectly obseved, but must be infeed fom thei behavio. nd this infeence is imefect due to stategic behavio by egoistic candidates. We analyze when stategic imitation by 1 Quoted afte Besley (2006: 29). 2

5 egoistic tyes hames the ability of the olitical system to vote benevolent leades into office. The baseline vesion of ou model has thee stages. In stage 1, candidates choose whethe to make an anti-social action which is beneficial to the candidate but modeately costly to society. Examles could be tax-evasion, cheating on an exam, o emloyment of an illegal immigant as a nanny. Choices in stage 1 can be obseved by votes with some obability which deends, fo examle, on the effectiveness of investigative media. In stage 2, a leade called esident is elected, and in stage 3, the esident is in contol of a budget. The esident chooses between oviding a ublic good, i.e. a olicy which is in the ublic inteest in the sense that it is efficient and fai, and a olicy which only benefits the esident but not the est of society. We assume two tyes of candidates and that the shaes of these tyes among candidates ae common infomation. Benevolent tyes ae non-stategic and find it otimal not to ick the anti-social action in stage 1 and to ovide the ublic good in stage 3. Egoistic tyes face a tade-off between taking the anti-social action in stage 1 to obtain immediate benefits, and abstaining fom taking it, in ode to aea as a benevolent tye in the eyes of the votes and thus to incease the chances of ent aoiation in stage 3. We say that stategic imitation evails if an egoist does not ick the anti-social action in stage 1. The incentives fo stategic imitation ae staightfowad in the baseline model. Intuitively, an egoist balances the costs of behaving o-socially in stage 1 against the exected benefits of ent aoiation in stage 3. The difficulty in this calculus is that the exected ent in stage 3 deends on the imitation behavio of the othe egoistic candidates which in tun deends on the shae of benevolent tyes. The baseline model yields two main insights. Fist, we show that stategic imitation is less likely if the olitical system is likely to oduce good govenance. On aveage, the govened fae well if the shae of benevolent candidates is high, the esident has little discetion ove the budget, and the ublic secto is effective in the sense that ublic goods ovision yields lage efficiency gains. In this case, the net gain fom govening athe than being govened is elatively low, and stategic imitation is theefoe elatively unattactive. Second, we find that moe effective investigative media which foste tansaency in the sense of imoving votes infomation about candidates e-election behavio, have 3

6 ambivalent effects on olitical selection. While highe tansaency imoves the chances of telling benevolent fom non-imitating egoistic candidates, it also ovides incentives fo egoistic candidates to ose as benevolent tyes. We find that moe tansaency imoves selection at low levels, but hames olitical selection if the level of tansaency is aleady high. We extend the baseline model by allowing incumbents to be e-elected to investigate if e-election is a emedy against the failue in olitical selection diagnosed above. In line with an extensive liteatue in olitical economy (e.g. usten-smith and Banks 1993, Besley and Case 1995), we find that incentives to seek e-election tend to disciline egoistic esidents. Howeve, the effects of e-election ae ambiguous and deend on how fa-sighted candidates ae. If candidates ae elatively atient, the incentive to seek eelection discilines egoistic esidents in the fist tem and deceases the amount of stategic imitation. If candidates ae elatively imatient, the ossibility of e-election futhe aggavates the oblem. In this case, the incentive to seek e-election does not cub ent extaction and may induce moe stategic imitation. The basic assumtions of the model ae that (i) esidential candidates ae heteogeneous with esect to social efeences, (ii) olicy choices ae in imotant ways shaed by the social efeences of esidents, and (iii) candidate s social efeences cannot be diectly obseved, and infeing them fom behavio is faught with difficulties. We now discuss these assumtions in tun by elating them to the liteatue and to emiical obsevations. ssumtion (i) is that eole, including votes and olitical candidates, ae heteogeneous with esect to social efeences. Mounting evidence fom exeimental economics suggests that o-social efeences exist and that individuals ae heteogeneous with esect to thei concen fo othes (see Camee 2003: Ch. 2 fo a suvey). Recent evidence shows that this also holds fo eesentative samles of the geneal oulation (e.g. Bellemae, Köge and van Soest 2008). While we assume that benevolent candidates exist, we conside the ossibility that thei shae is small. Paes studying selection into the olitical aena (e.g. Mattozzi and Melo 2007, Caselli and Moelli 2004, Messne and Polbon 2004) emhasize that bad tyes may select into the ool of candidates, ehas because they find a caee in the olitical aena moe attactive than one in the maket 4

7 lace. s a esult of advese selection, the shae of benevolent tyes among candidates is likely to be smalle than in the geneal oulation. ssumtion (ii) is that olicy choices deend on the tye of the esident selected. This assumtion is well in line with emiical studies showing that the identity of olicy makes shaes olicy choices and economic outcomes (e.g. Jones and Olken 2005, Lee, Moetti and Butle 2004, Chattoadhyay and Duflo 2004). ssumtion (i) that candidates ae heteogeneous togethe with assumtion (ii) that votes cae about this heteogeneity is consistent with the obsevation that esonal qualities like the integity of olitical candidates often lay an imotant ole in olitical camaigns. Fo examle, in exit olls taken in connection with the Reublican esidential imay elections in the U.S. in 2008, aound 45 ecent of votes stated that esonal qualities wee moe imotant fo thei vote than issues 2 (see also Mondak 1995). ssumtion (iii) is that olitical selection though elections is hameed because of stategic imitation. Ou ae comlements the liteatue on olitical selection though elections (e.g. Bollo et. al. 2010, Coazzini et. al. 2009, Benies and Du 2007, Besley and Smat 2007, Besley 2006: Ch. 3, Feaon 1999, Coate and Mois 1995, Banks and Sundaam 1993). We add to this liteatue in seveal ways. Fist, in most models of olitical selection though elections, the utility of a candidate who loses an election is exogenous, tyically nomalized to zeo (e.g. Feaon 1999, Besley 2006, Besley and Smat 2007, Bollo et. al. 2010). In o model, on the othe hand, the utility of losing candidates equals the utility of odinay citizens and is endogenously detemined. Concens about utility in case of electoal defeat ae an imotant deteminant of candidates imitation behavio in ou model. If egoistic candidates exect govenance to be bad, and theefoe the utility of odinay citizens to be low, they ae moe likely to make costly investments in eutation in ode to incease thei chances of holding office. This behavio, in tun, hames the olitical selection ocess. The model in Feaz and Finan (2009) shaes this featue with ou ae but focuses on the effects of oliticians wages on olitician quality and efomance. Second, most models in the liteatue study how an incumbent leade signals his o he tye befoe an election. In this context, a bad leade imitating a good one has 2 Infomation found on the webage election cente 2008 of CNN.com 5

8 ambiguous effects. On the one hand, imitation educes the obability that a good tye is elected in the next election, on the othe it leads to bette olicies in the cuent eiod, since imitation consists in choosing the olicies that a good leade would choose. In Besley (2006) and Bollo et. al. (2010) the net effect of imitation is ositive, i.e. elections disciline bad oliticians. In contast, ou model consides signaling of non-incumbent candidates in the e-election eiod. We assume that the effects on society of actions taken befoe the election ae small elative to those taken in office. Theefoe, the main effect of imitation is the negative one of weakening olitical selection. Thid, most models conside the signaling behavio of only one agent, namely the incumbent leade. In contast, we analyze the signaling decisions of seveal agents whee the exected signaling behavio of othe candidates is an imotant element in a candidate s otimization oblem. Benies and Du (2007) shae this featue with ou ae but they conside incumbents in a coalition govenment while we focus on non-incumbent candidates. The model in Coazzini et. al. (2009) also featues two candidates who can send signals to votes but in thei model signaling comes in the fom of mateially costless, nonbinding camaign omises. In contast, we focus on costly signaling. Coazini et. al. focus on benevolence in the guise of lying avesion wheeas we model it as a efeence fo efficiency and fainess. The ae is also elated to the olitical science liteatue on valence issues, i.e. issues all votes agee on. In ou model, all votes agee that it is efeable to have a benevolent leade. Most aes in this liteatue study how valence issues affect olicy. Fo examle, nsolabehee and Snyde (2000), Goseclose (2001) and agones and Palfey (2001) ovide models showing that candidates o aties with a valence advantage choose moe modeate olicies than candidates o aties with a valence disadvantage. In contast, ou model focuses not on olicy but on the selection of benevolent candidates. One imlication of ou model is that votes will often be disaointed about the chaacte of a esident in the sense that they infeed fom his e-election behavio that he was benevolent but discove that he is not when in office. Examles of such disaointment abound. Fedinand Macos, the notoiously cout esident of the Philiines, was a decoated wa heo fom Wold Wa II. Haitian Pesident Fancois Duvalie eaned his famous nickname, Paa Doc, by seving as a successful docto in a 6

9 camaign to eadicate toical diseases. s esident, he an a highly eessive and cout govenment. Robet Mugabe, the esident of Zimbabwe, was a widely admied heo of his county s libeation stuggle but gadually evealed himself as one of the most self-seving leades in fica. Stategic imitation offes a otential exlanation fo such disaointment. 2 The model The baseline model has thee stages (see figue 1): In stage 1, n candidates out of a oulation of size N un fo esident. Since the esent ae focuses on selection though voting athe than enty, the numbe of candidates is exogenous to ou model and we assume that candidates ae andomly selected fom the oulation. Candidates can make an anti-social action at some benefit b to themselves (evading tax ayment, cheating on an exam etc.) but with a cost c to society. We assume two tyes of candidates. Candidates of a given tye ae homogenous. Benevolent tyes neve choose the anti-social action because they incu a sychological cost ψ > b if they do. We can think of this as the cost of a bad conscience. Egoistic tyes suffe no sychological cost fom choosing the anti-social action. These tyes efain fom taking the anti-social action only if it maximizes thei exected ayoffs in the entie game. The shae of benevolent tyes among the candidates is θ > 0. Candidates choices in stage 1 ae obseved by votes with obability 0. We assume that θ and ae ublic infomation. We efe to as the tansaency of olitics which is shaed by, fo examle, the stength of investigative media o by legislation equiing oliticians to disclose infomation. 3 In stage 2, a esident is voted into office by N n votes. Voting is comulsoy and sincee, i.e. votes cast thei votes accoding to thei mateial efeences. The esident is elected by luality vote o by andom daw with obability 1/k if k candidates eceive the highest numbe of votes. 3 Rathe than letting candidates choose whethe to take an anti-social action in eiod 1, we may assume that they decide about taking a o-social action, such as volunteeing fo the amy o efoming chaity wok. ssume that the ivate, monetay cost of efoming the o-social act is b, and that benevolent candidates always take the osocial action, while egoists only do so only if it maximizes thei total, exected ay-off in the game. The esults following fom these assumtions ae essentially the same as the ones esented below. When agents choose whethe to take a o-social athe than an anti-social action, the aamete should be inteeted as the obability that acts can be veified, athe than the obability that they ae evealed. The eason is that all agents who take the o-social action have a incentive to voluntaily infom the ublic of thei behavio. 7

10 Figue 1: Political selection game Stage 1 (e-election) Stage 2 (election) Stage 3 (ost-election) n candidates choose anti-social action (ivate benefit b) o not ctions in stage 1 ae evealed with obability Pesident is voted into office (N-n votes) Pesident contols budget B. llocation to ublic good (benefit NB) o ivate good. In stage 3, the esident is in contol of a budget B which can be used to fund a ublic good o be aoiated fo the esident s ivate benefit. We assume that the ublic good is linea, i.e. has constant etuns to scale and benefits all votes equally. The effectiveness of the ublic secto is measued by α, the maginal e caita etun of a dolla allocated to the ublic good, whee 1/N < α < 1. llocating the entie budget to the ublic good is both efficient and fai. It is efficient in the sense that the sum of ayoffs to society is maximal (αnb > B), and it is fai in the sense that all agents, including the esident, eceive the same ayoff, αb. If an egoistic esident is elected, the entie budget is ocketed by the esident kletocacy evails. Payoffs in stage 3 ae discounted by a facto δ, 0 < δ < Equilibium We analyze the deteminants of successful olitical selection in equilibium, i.e. the equilibium amount of stategic imitation and the esulting obability of selecting a benevolent esident. We seach fo a efect Bayesian equilibium in which votes udate thei beliefs using Bayes ule and each candidate chooses an otimal stategies given the choices of othe candidates and votes. The main vaiables of inteest detemining equilibium values ae the shae of benevolent candidates θ, the level of tansaency, the effectiveness of the ublic secto, and the esident s discetionay owe ove the budget B. Given ou assumtions, egoists efe to goven athe than to be govened, and all votes efe a benevolent athe than an egoistic esident. The oblem is that votes cannot diectly obseve candidates tyes, and have to base thei decision in stage 2 on beliefs about tyes. These beliefs ae shaed by obseved behavio in stage 1. By assumtion, benevolent tyes efain fom choosing the anti-social action in stage 1 with cetainty, and egoistic tyes may o may not stategically imitate the benevolent tyes, 8

11 deending on how much imitation inceases thei exected ent. We denote the shae of all candidates who do not choose the anti-social action in stage 1 by σ θ. Uon obseving each candidate s eiod 1 action, and theefoe obseving σ, votes udate thei beliefs about the tye of each candidate. Bayes ule imlies that votes attach the obability θ/σ to a candidate being benevolent in case he has not taken the anti-social action in eiod 1, and a obability of zeo if he has. 4 In line with intuition, votes always vote fo a candidate who did not choose the anti-social act in eiod 1 (ovided that they eceive infomation about these acts). benevolent esident is selected with cetainty only if σ = θ, i.e. if thee is no stategic imitation. In contast, if σ > θ, a esident may be voted into office who does not seve the ublic inteest in stage 3. Fom the esective of an egoistic candidate, the benefits of stategic imitation deend on how much the signal inceases the obability of being elected, on the diffeence in (discounted) exected ayoffs between govening and being govened, and between good and bad govenance if not elected. Moe secifically, if egoist i is voted into office, his ayoff is B. If i is not elected, he eans 0 if the esident is an egoist and αb > 0 if the esident is a benevolent tye. We denote the conditional obability that an egoist is elected if he imitates by s and the obability if he does not by 0. We denote the obability that a benevolent esident is elected by. These obabilities deend on the tansaency of the olitical system (measued by ), i.e. on whethe infomation about eelection behavio is evealed at the election stage. The suescit indicates that infomation is evealed, and the suescit u that it is not. Fo examle, s is the obability that a candidate is elected if he imitates, and infomation about e-election behavio is evealed. 4 Fomally: ob(tye = benevolent osocial action) = [ob(osocial action tye = benevolent) * ob(tye = benevolent)] / ob(osocial action) = θ / 9

12 Figue 2: Otimization oblem fo an egoistic tye Imitate benevolent tye Not imitate s s 1 u s u u u s u 1 u 1 u 1 u δb δαb 0 δb δαb 0 δb+b δαb+b b δb+b δαb+b b Figue 2 shows the decision tee eesenting the otimization oblem of an egoist. Total discounted ayoffs ae shown at the bottom of the figue. n egoist stategically imitates if the exected ayoff is at least as big as when he does not: u u u s s s s B 1 B 0 1 B 1 B 0 u u u u B b B b b B b B b b Exession (1) can be consideably simlified by using the following facts. Imitation has no effect on the tye voted into office if no infomation about e-election behavio is available ( = 0) in the election stage. In this case, votes andomly select a candidate which u u means that 0 1/ n and s u. lso, since we assume that thee is at least one benevolent candidate (θ > 0), an egoist is neve selected if he does not imitate and infomation about e-election behavio is available, i.e Thus, (1) educes to (1) 10

13 B s 1 b, (2) whee the left-hand side is the exected discounted benefit and the ight-hand side is the ivate cost of imitation. Note that both s and ae functions of which, in tun, is a function of all aametes, including θ,,, and B. Remembe that the symbol s is the obability that a candidate is voted into office in stage 2 given that he did not chose the anti-social action in stage 1 and that infomation about eiod 1 was evealed. This obability deends invesely on n, i.e. on how many candidates choose the anti-social action in stage 1. 1 s (3) n key vaiable in ou model is the obability of selecting a benevolent esident, This obability deends on the shae of benevolent candidates and the tendency of egoists to stategically imitate. In aticula, when infomation about e-election behavio is evealed, a benevolent esident is elected fo sue (. = 1) if egoists do not imitate (i.e. if ). In contast, a benevolent esident is elected only by chance, i.e. egoists imitate (i.e. σ = 1). In geneal, it holds that = θ if all (4) Inseting (3) and (4) into (2) we get the following condition fo stategic imitation: 1 B 1 b 0 n (5) We e-wite (5) as 2 Y( ) ( bn) ( B) B 0 (6) which can easily be eesented by a aabola (see figue 3). 11

14 Figue 3: Stategic imitation equilibia Y θ σ 1 σ 2 1 (b) σ (c) (a) We denote the citical values yielding Y(σ) = 0 by σ 1 and σ 2 and distinguish 3 cases: a) No stategic imitation: Y( ) < 0 [ ;1]. In this case, the aabola in figue 3 is below 0 in the feasible ange of σ. Theefoe, stategic imitation neve ays, and σ * = θ. sufficient condition fo this no-imitation equilibium to occu is that the vetex of the aabola, Y v, is below zeo, i.e. that B 2 0 B Yv Y B 2bn 4bn. (6b) Condition (6b) is moe likely to be met and stategic imitation theefoe less likely to occu at all if, fo examle, the numbe of candidates n is high o if imitation is vey costly to the candidate (i.e. b is high). Inteestingly, insection of (6b) eveals that stategic imitation is less likely to be attactive if the ublic secto is vey effective ( is lage), if the esident has little discetionay owe of the budget (B is small), o if most candidates ae 12

15 benevolent ( is lage). In othe wods, stategic imitation is discouaged if govenance is likely to be good. * b) Unique equilibium with stategic imitation at 1. If the vetex of Y(σ) is ositive ( Y v > 0) and if Y(σ) σ = θ > 0, a unique equilibium obtains with a ositive amount of * stategic imitation of 1. In figue 3, this case is illustated fo a situation in which only some egoists imitate in equilibium (σ * < 1). Since all egoists ae identical in the model, it is natual to assume a symmetic equilibium. The only symmetic Nash equilibium imlies that each egoist lays a mixed stategy, whee he imitates with obability (σ 2 θ)/(1 θ). If Y(1) 0, imitation is dominant fo an egoist, and the equilibium is σ * = 1. In this ooling equilibium, incentives to imitate ae so stong that all egoistic candidates behave as if they wee benevolent befoe the election, but the elected esident is unlikely to seve the ublic inteest afte the election (if θ is small). Note that if Y(σ) > 0 in the entie ange, all egoists stategically imitate in equilibium, and maginal changes in aametes do not educe stategic imitation. c) Multile equilibia. If the vetex of Y(σ) is ositive ( Y v > 0) and if Y(σ) σ = θ < 0, Y(σ) intesects the hoizontal axis twice in the feasible ange of σ as illustated in gah (c) of figue 3. Multile equilibia evail in this case, namely i) σ * = θ, ii) σ * = σ 1, and iii) σ * = min (σ 2, 1). Howeve, only i) and iii) ae stable equilibia, and beliefs about σ detemine which equilibium stategic candidates coodinate on. Suose egoists hold an equilibium common io, σ 0 (see aendix fo a discussion of disequilibium beliefs). If σ 0 = σ 1, then σ * = σ 1 is an equilibium, but it is unstable because if σ 0 deviates by fom σ 1, the shae of imitating candidates conveges to eithe θ o min(σ 2,1). This means that equilibium σ 1 is unlikely to evail, and we do not conside it in the emainde of ou discussion. In sum, cases a) and ci) ovide a comlete descition of the conditions fo the absence of stategic imitation. Thus, when infomation about e-election behavio is evealed, the elected esident seves the ublic inteest with cetainty in these cases. In all othe cases, thee is a isk that an egoist is voted into office. 13

16 2.2. Comaative statics We now investigate how the extent of stategic imitation in equilibium and the obability of selecting a benevolent esident, deend on changes in aametes. We concentate on the equilibium at σ 2. The eason is that the othe equilibia ae eithe unstable o involve full o no imitation as the discussion above has shown, and ae theefoe not sensitive to maginal changes in aametes. Since (6) is a second-degee olynomial in σ, σ 2 is given by: whee B B 4bn B B D 2 (7) 2bn 2bn D B bn B. The comaative statics below ae deived by diffeentiating (7) with esect to the aamete in question. The effects of the fist thee aametes discussed below can be summaized in the statement that stategic imitation is less likely if govenance is likely to be good. We show that a highe shae of benevolent candidates, less discetionay owe of the esident ove the budget B, and a highe effectiveness of the ublic secto all imove olitical selection, because they educe the elative cost of being govened, athe than govening. We also discuss the effects of tansaency () in the olitical ocess and find that the effect of tansaency on the quality of olitical selection is ambiguous. a) Shae of benevolent candidates (θ) highe shae of benevolent tyes deceases the incentives fo stategic imitation [ B 2 0 D, see eq. (7)]. The eason is that govenance is likely to be good - a benevolent esident is moe likely to be selected - which educes the cost fo an egoist of not govening. Figue 4 illustates how the equilibium obability of selecting a benevolent esident,, deends on the shae of benevolent candidates (θ). Ou esult with stategic imitation is comaed to a benchmak without imitation in which the cost of imitation b is high enough to dete any imitation. This imlies that (1 ) (see dashed No imit. line in figue 4). The figue shows that stategic imitation has the most advese effects at low values of θ. Suose, fo examle, that 1% of all candidates ae benevolent and 99% 14

17 ae egoistic. If all egoistic tyes make anti-social choices in the e-election stage, benevolent tyes ae selected wheneve thei signal can be obseved ( is assumed to be 50% in the figue). In contast, if egoistic tyes stategically imitate, only about 1% of esidents will seve the ublic inteest. It is only fo unealistically high values of θ (of aound 60% in the figue) that stategic imitation by egoists ceases to be hamful. Figue 4: Pobability of voting a benevolent esident into office ( ) as a function of θ Note: The figue is dawn assuming equilibium σ 2 evails and falls between θ and 1. We illustate fo aametes n = 10, = 0.5, δ = 0.9, α = 0.4, b = 0.3, B = 7. b) Budget size (B) Contol of a highe budget makes govening, and theefoe stategic imitation fo 2 D B egoists moe attactive [ B 2bn 4bnB D 2 0, see eq. (7)]. highe shae of stategic imitatos educes the obability of selecting a benevolent tye. Bollo et. al. (2010) also esent a model whee an exogenous incease in the ublic budget damages 15

18 olitical selection. They find suot fo this hyothesis in an emiical study of municial govenments in Bazil. c) Efficiency of ublic goods oduction (α) The incentives fo stategic imitation fall with α, the effectiveness of the ublic secto 2 B [ 0, see eq. (7)]. The eason is again that the exected cost of being D govened athe than govening is deceasing in α. The esult is that the obability of selecting a benevolent esident inceases with the efficiency of ublic goods oduction. This esonates well with a numbe of emiical studies which have found a negative coelation between coution and the effectiveness of ublic bueaucacies. These coelations ae usually inteeted as indicating that coution educes this effectiveness (e.g. Badhan 1997, Shleife and Vishny 1993). Ou model ovides a ationale fo why the causality may un the othe way. If the ublic administation is ineffective, selfish individuals ae attacted to seeking olitical office because life without olitical owe is unleasant. This suising conclusion might hel to exlain why counties with highly ineffective ublic bueaucacies, e.g. in some fican counties, also seem to attact a highe shae of cout oliticians than counties with moe effective ublic administations. Besley and Smat (2007) also conside the effect of ublic secto effectiveness on olitical selection, but wheeas we focus on the effectiveness of ublic goods oduction, they look at technologies fo aising govenment evenue. They oint out that moe efficient means of aising evenue means that moe esouces ae available fo ent seeking. Theefoe, a moe efficient evenue aising technology may attact cout candidates and theefoe wosen olitical selection. Hence, Besley and Smat focus on anothe asect of ublic secto effectiveness than we do and each vey diffeent conclusions. d) Tansaency () Stategic imitation is moe attactive if actions in the e-election stage ae moe likely 2 B D B to be evealed [ 2bn 4bn D 2 0, see eq. (7)]. Moe stategic imitation means that the infomativeness of the signal to votes is low. On the othe hand, a highe value of also means that a benevolent candidate is moe likely to be detected at all. Theefoe, the 16

19 oveall effect of on olitical selection is non-monotonic. The obability that a benevolent esident is elected is u 1 (1 ) (1 ) 1 * * (8) If no infomation is evealed, 0, votes ick a candidate at andom which means that the obability of electing a benevolent esident is θ. Fo sufficiently low values of ( < 4bnB ), imitation is not ofitable, i.e. *. In this case, is inceasing in. Howeve, at some oint imitation becomes attactive (aound = 0.3 fo the secific * aametes in figue 5). When all egoists imitate ( 1), votes cannot extact useful infomation fom stage 1 signals, no matte how high is, and again dos to θ. Hence, inceasing tansaency has ositive effects at low levels of tansaency, but negative effects if the level is aleady high. Figue 5: Investigative media and the quality of olitical selection () σ*() Degee of tansaency () Note: Illustation is based on same aamete values as Figue 4, and θ = 0.4. The intuition fo this esult is that highe tansaency inceases the cost of acting selfishly in the e-election stage and theefoe inceases incentives fo egoists to behave as 17

20 if they wee benevolent. This, in tun, deceases the quality of infomation available to votes about the tye of the candidates. The finding that highe tansaency of olitics does not necessaily imove olitical economic outcomes esonates well with Mattozzi and Melo (2007) who find that highe tansaency can lead to less cometent oliticians, because it inceases the ootunity cost of olitical activity fo highly cometent agents. Besley (2006) also finds that the effect of tansaency on olitical selection is ambiguous. He assumes that votes eceive an exogenous signal about an incumbent leade s tye and shows that a stonge signal might lead to wose outcomes because it weakens the incentive fo a bad incumbent to mimic a good one by choosing good olicies. Emiical studies mostly find a negative effect of media feedom on coution (e.g. Bunetti and Wede 2003, Feille et. al. 2007), but tyically do not test fo non-linea effects. 3 Re-election: cue fo coution? We now extend ou baseline model to investigate e-election as a constitutional constaint on ent extaction by egoistic (cout) esidents. While the analysis of incentives fom eelection has been a ecuent theme in the ublic choice liteatue (e.g. Bao 1973, Feejohn 1986, Smat and Stum 2006), ou analysis comlements that liteatue by allowing fo a (small) shae of benevolent candidates fo olitical office. s in Coate and Mois (1995), incentives fom e-election seve two oles: to cutail ent-extaction by egoistic incumbents a moal hazad asect and to select a benevolent candidate an advese selection asect. We analyze if e-election is an effective constitutional constaint to ent extaction by adding a stage 4 and 5 which essentially elicate stages 2 and 3 in the baseline game of figue 1. key diffeence is that at the beginning of stage 4, votes lean how much of the budget the incumbent allocated to ublic goods in stage 3. The main esults of ou analysis deend on egoistic candidates atience, i.e. by how much they discount ayoffs accuing in late eiods, and ae as follows. If candidates ae sufficiently atient, incentives to seek e-election disciline egoistic esidents and imove olitical selection. Moe secifically, we show that if e-election incentives disciline 18

21 egoistic esidents in stage 3, they also tend to educe stategic imitation, because the benefits fom imitation ae eaed only with a delay, i.e. afte the second election. Howeve, if agents ae imatient, the incentive to seek e-election might not be stong enough to disciline incumbents and the intoduction of a second election might lead to futhe deteioation of olitical selection. Stategic imitation is now moe attactive because imitatos get two, athe than one, chances of winning the esidency and catuing the budget. We now deduce these esults by solving the game backwads. In stage 5, egoistic esidents aoiate the entie budget while benevolent esidents allocate B to ublic goods. In stage 4, votes choose between the incumbent and a challenge. We assume that votes efe the incumbent ove a challenge if the incumbent is equally o moe likely to be benevolent than the challenge. This is the case if the incumbent allocates B to ublic goods in stage 3. If not, it is clea that the incumbent is an egoist and he would not be eelected in stage 4. Stage 3 outcomes can be undestood as follows. We denote by 5 the obability that a benevolent tye is elected in the stage 4, given that an egoist was elected in stage 2. We assume that 5 = θ when no infomation is evealed, and 5 = when it is (fo simlicity, we do not take into account that the aticula egoist who won the fist election is not a viable candidate in the second election. Unless n is small, this changes the esults only slightly). If an egoistic esident allocates B to ublic goods, his ayoff is αb in stage 3, and δb in stage 5. If instead he ockets the budget B in stage 3, his exected ayoff in stage 5 is 5 δαb. n egoistic esident theefoe chooses the benevolent olicy in stage 3 if and only if B B B 5 B, i.e. if: 1 * 1 5 (9) Note that the theshold value fo imatience * is deceasing in α and inceasing in 5. moe efficient ublic secto inceases the esident s incentive to act in the ublic inteest because he himself benefits fom the goods oduced by the ublic secto. highe obability of a benevolent esident taking office in eiod 5, on the othe hand, deceases 19

22 the exected loss fom giving u the chance of e-election. Since, moe atience is equied to event an egoistic esident fom catuing the ent when infomation about stage 1 actions is evealed than when it is not. Note that 5, and theefoe on the amount of stategic imitation. * is endogenous. It deends on In stage 2, a candidate who chose the benevolent action in stage 1 is elected, wheneve infomation about e-election behavio is evealed. Even if votes ealize that an egoistic esident is going to allocate B to the ublic good in stage 3, they also know that he will be e-elected and kee the entie budget in stage 5. Theefoe, votes still have an incentive to favo the candidates they believe ae most likely to be benevolent. In stage 1, an egoist s incentive to choose the anti-social action deends on his olicy lans if elected esident, and on his beliefs about the olicy choices of othe egoistic esidents. 5 * (i). If an egoist is sufficiently atient, he allocates B to ublic goods in stage 3 accoding to eq. (10) (below we check if the assumtion * is consistent with the equilibium level of stategic imitation). Stategic imitation affects the obability of winning the election in stage 2 in exactly the same way as in the baseline game, but the discounted gains fom winning the esidency ae smalle. In stage 3, all agents ean αb, and thee is no benefit fom holding office. Howeve, the esident is e-elected with cetainty, and eans B in stage 5. The obability that a benevolent esident holds office in stage 5 is the same as in the baseline game. Hence, the exected gain fom imitation is the same as in the baseline case, excet that it is discounted by δ 2 athe than δ. In this case, theefoe, the intoduction of a second election leads to (weakly) less stategic imitation than in the baseline case. Denote by the equilibium value of σ in the case whee the egoist intends to allocate B to the ublic good, and exects othe egoists to do the same. The esulting value of 5 is denoted by 5 /, and the esulting value of * by * (ii). The oblem of a not sufficiently atient egoist is moe comlicated and analyzed in detail in aendix B. Egoistic esidents now gab the ublic budget in stage 3. 5 endix C shows that the otimal stategies deived hee ae consistent with the assumtions about * used in thei deivation. 20

23 The gains fom imitation which ae catued in stage 3 ae theefoe the same as in the oiginal game, fo a given value of σ (see aendix B fo a oof). The additional benefit catued in stage 5 can be ositive o negative. On the one hand, imitation ovides a chance to win the second election, even if infomation about stage 1 actions is evealed. In this sense, imitation buys the egoist an additional chance of catuing the ublic budget. On the othe hand, imitation slightly deceases the obability that a benevolent tye holds office in stage 5, which has a negative effect on the egoist s exected ayoff in that eiod. 6 The aendix shows that the net effect might be ositive o negative. Denote the equilibium value of σ in this case by. The esulting value of 5 is denoted *, and the esulting value of by 5 / If imitation inceases exected eanings in stage 5, i.e. if the effect fom two athe than one chance of winning the esidency dominates, the total incentive fo stategic imitation is stonge than in the case with only one election. Hence, the intoduction of a second election leads to moe stategic imitation and theefoe less effective olitical selection comaed to the situation with only a single election. Figue B2 in aendix B illustates that this is the case. 6 The obability that a benevolent tye holds office in the last stage is highe when the egoist chooses not to imitate than when he does. To see that this is the case, denote the obability that a aticula egoistic agent is elected esident in stage 2 when infomation about e-election behavio is evealed by, j=s,0. The j obability that a benevolent tye holds office in eiod 5 is (1 ) (1 )(1 ) j 5 j 5 j 5 5 whee the fist tem is the obability that the egoistic agent is elected in the fist election, and a benevolent tye elaces him. The second tem is the obability that a benevolent tye is elected in the fist election and becomes e-elected. The thid tem is the obability that an egoist othe than the one we ae consideing is elected in the fist election, and that he is elaced by a benevolent tye in the second election. Since 0 0, it follows: (1 ) (1 )(1 ) (1 ) (1 ) (1 )(1 ) (1 )(1 ) s 5 s 5 s s

24 4 Concluding emaks This ae illustates that olitical selection may fail even if benevolent candidates exist and all votes efe to have benevolent esidents. We show that whethe selection succeeds o fails deends on institutional and social factos like the discetionay leeway of the esident, ossibilities fo e-election and the stength of investigative media. Stategic imitation by cout oliticians has clealy advese welfae effects in ou model because imitation hames olitical selection. 7 Ou model sheds new light on the elation between good govenance and the selection of cout oliticians ( egoistic esidents in ou teminology). While it is natual to assume that cout oliticians ae a cause of bad govenance, ou esults suggest that causality may also un the othe way. If govenance is oo, life as a citizen is unleasant and egoists have a stong incentive to seek a caee in olitics to ea the ents available to office holdes. Thus, oo govenance beeds cout oliticians. If, on the othe hand, govenance is good, ent seeking is elatively less attactive, and only individuals with intinsic efeences fo seving the ublic inteest stive fo ublic office. Thus, good govenance enables votes to select vituous oliticians. The aametes of good govenance, i.e. the high efficiency of the ublic secto, the lack of discetionay owe of the esident, and a high shae of benevolent tyes in the ool of candidates, detemine which outcome obtains. In ou simle model, these factos ae exogenous. Howeve, it seems lausible that the causation uns both ways because these factos would be affected by the chaacte of the esident in owe. Pesidents with a flawed chaacte may geneate ineffective administation, aise taxes to incease the size of the budget they contol, and destoy social caital in the ool of candidates. vicious cicle is set off whee bad govenance beeds bad oliticians who, in tun, beed moe bad govenance. Ou assumtion that some candidates ae benevolent elates ou model to the issue of how social caital and olitical cultue (e.g. Hillman and Swank 2000) shae olicy outcomes. If benevolence among candidates is a ositive function of benevolence in the oulation at lage, the shae of benevolent candidates can be inteeted to eflect social caital in a society. Putnam (1993) agues that diffeences in the quality of govenance 7 Howeve, the welfae effects of stategic imitation would be ambiguous if the social cost of the anti-social act in stage 1 (c) was assumed to be sufficiently high (see Cugno and Feeo 2004 fo a elated case). 22

25 between Nothen and Southen Italy mio vaiations in civic cultue between the two egions. Ou model offes a diect and an indiect exlanation fo such data attens. Not only ae thee fewe benevolent candidates aound in laces with weak civic cultue (diect effect); ou model also shows that stategic imitation is moe common when is low, and stategic imitation makes it moe difficult fo votes to ick the benevolent candidates out (indiect effect). While ou simle model of olitical selection yields a numbe of inteesting esults, we believe that the model could be fuitfully extended to catue additional asects of selection. Fist, while we focus on olitical selection of chaacte, selection of cometence may be equally imotant (e.g. cemoglu, Egoov and Sonin 2009, Besley 2005, Messne and Polbon 2004, Mattozzi and Melo 2007). The model may be extended to allow candidates to diffe in cometence as well as chaacte, as in Gesbach (2004) and Benies and Du (2007). Second, we investigate olitical selection though elections fom a given ool of candidates but the ocess of ecuiting candidates, i.e. who selects into the olitical aena, is clealy also inteesting to study. These two selection ocesses may inteact in imotant ways as demonstated in the models of Besley (2006), Feaz and Finan (2009) and Bollo et al. (2010). Fo examle, if egoistic tyes ae moe likely than benevolent tyes to ente the olitical game, as agued by Caselli and Moelli (2004), the shae of candidates with benevolent efeences (the aamete θ in ou model) is small, and stategic imitation is common. 23

26 Refeences cemoglu, D., Egoov, G. and Sonin, K. (2009): Political Selection and Pesistence of Bad Govenments. NBER Woking Pae nsolabehee, S. and Snyde, J.M. (2000): Valence Politics and Equilibium in Satial Election Models. Public Choice 103(3-4): agones, E. and Palfey, T. (2001): Mixed Equilibium in a Downsian Model with a Favoed Candidate. Jounal of Economic Theoy 103(1): usten-smith, D. and Banks, J. (1993): Electoal ccountability and Incumbency. In Odeshook, P.C. (ed.): Models of Stategic Choice in Politics. nn bo: The Univesity of Michigan Pess. Banks, J. and Sundaam, R. (1993): dvese Selection and Moal Hazad in a Reeated Elections Model. In W. Banett, M. Hinich and N. Schofield (eds.): Political Economy: Institutions, Infomation, Cometition and Reesentation. Cambidge: Cambidge Univesity Pess. Badhan, P. (1997): Coution and Develoment. Review of Issues. Jounal of Economic Liteatue 35(3): Bao, R. (1973): The Contol of Politicians. Public Choice 14(1): Bellemae, C., Köge, S. and van Soest,. (2008): Measuing Inequity vesion in a Heteogenous Poulation using Exeimental Decisions and Subjective Pobabilities. Econometica 76(4): Benies, K.J. and Du, R. (2007): Politicians Motivation, Political Cultue and Electoal Cometition. Intenational Tax and Public Finance 14(1): Besley, T. (2005): Political Selection. Jounal of Economic Pesectives 19(3): Besley, T. (2006): Pinciled gents? The Political Economy of Good Govenment. Oxfod: Oxfod Univesity Pess. Besley, T. and Case,. (1995): Does Political ccountability ffect Economic Policy Choices? Evidence fom Gubenatoial Tem Limits. Quately Jounal of Economics 110(3): Besley, T. and Smat, M. (2007): Fiscal Restaints and Vote Welfae. Jounal of Public Economics 91(3):

27 Bennan, G. and Buchanan, J.M. (1980): The Powe to Tax. nalytical Foundations of the Fiscal Constitution. Cambidge: Cambidge Univ. Pess. Bollo, F., Nannicini, T., Peotti, R. and Tabellini, G. (2010): The Political Resouce Cuse. IGIER Woking Pae Seies, Bocconi Univesity. Bunetti,. and Wede, B. (2003): Fee Pess is Bad News fo Coution. Jounal of Public Economics 87: Camee, C.F. (2003): Behavioal Game Theoy. Pinceton, NJ: Pinceton Univesity Pess. Caselli, F. and Moelli, M. (2004): Bad Politicians. Jounal of Public Economics 88(3): Chattoadhyay, R. and Duflo, E. (2004): Women as Policy Makes: Evidence fom an India- Wide Randomized Policy Exeiment. Econometica 72(5): Coate, S. and Mois, S. (1995): On the Fom of Tansfes to Secial Inteests. Jounal of Political Economy 103(6): Coazini, L., Kube, S., Maechal, M.. and Nicolo,. (2009): Elections and Decetions. Woking Pae. Univesity of Padua. Cugno, F. and Feeo, M. (2004): Cometition among Voluntees. Euoean Jounal of Political Economy 20(3): Feaon, J.D. (1999): Electoal ccountability and the Contol of Politicians: Selecting Good Tyes vesus Sanctioning Poo Pefomance. In. Pzewoski, S.C. Stokes and B. Manin (eds.): Democacy, ccountability and Reesentation. Cambidge: Cambidge Univesity Pess. Feejohn, J. (1986): Incumbent Pefomance and Electoal Contol. Public Choice 50(1): Feaz, C. and Finan, F. (2009): Motivating Politicians. The Imacts of Monetay Incentives on Quality and Pefomance. NBER Woking Pae Feille, S., Haque, E.M. and Knelle, R. (2007): Contibution to the Emiics of Pess Feedom and Coution. Euoean Jounal of Political Economy 23: Gesbach, H. (2004): The Paadox of Cometence. CEPR Discussion Pae no Goseclose, T. (2001): Model of Candidate Location when one Candidate has a Valence dvantage. meican Jounal of Political Science 45(4): Hillman,.L. and Swank, O. (2000): Why Political Cultue should be in the Lexicon of Economics. Euoean Jounal of Political Economy 16(1):

28 Jones, B. and Olken, B. (2005): Do Leades Matte? National Leadeshi and Gowth since Wold Wa II. Quately Jounal of Economics 120(3): Lee, D.S., Moetti, E. and Butle, M. (2004): Do Votes ffect o Elect Policies? Evidence fom the U.S. House. Quately Jounal of Economics 119(3): Mattozzi,. and Melo,. (2007): The Tansaency of Politics and the Quality of Politicians. meican Economic Review 97(2): Mauo, P. (1995): Coution and Gowth. Quately Jounal of Economics 110(3): Messne, M. and Polbon, M.K. (2004): Paying Politicians. Jounal of Public Economics 88(12): Mondak, J.J. (1995): Cometence, Integity, and the Electoal Success of Congessional Incumbents. Jounal of Politics 57(4): Putnam, R. (1993): Making Democacy Wok: Civic Taditions in Moden Italy. Pinceton, NJ: Pinceton Univesity Pess. Shleife,. and Vishny, R.W. (1993): Coution. Quately Jounal of Economics 108(3): Smat, M. and Stum, D.M. (2006): Tem Limits and Electoal ccountability. CEPR Discussion Pae no