Economic Performance, Government Size, and Institutional Quality *

Transcription

1 Economc Performance, Governmen Sze, and Insuonal Qualy * Anóno Afonso $ and João Tovar Jalles 2015 Absrac Ths paper sudes he emprcal lnk beween governmen sze, nsuons and economc acvy usng a panel of 140 counres over 40 years. Our resuls, robus under dfferen economerc echnques, show mosly a negave effec of governmen sze on oupu, whle nsuonal qualy has generally a posve mpac. Moreover, he dermenal effec of governmen sze on economc acvy s sronger he lower nsuonal qualy, and he posve effec of nsuonal qualy on oupu ncreases wh smaller governmen szes. JEL: C10, C23, H11, H30, O40 Keywords: producon funcon, nsuons, globalzaon, GMM, pooled mean group, common correlaed effecs * The opnons expressed heren are hose of he auhors and do no necessarly reflec hose of her employers. $ $ ISEG/ULsbon Unversy of Lsbon, Deparmen of Economcs; UECE Research Un on Complexy and Economcs, R. Mguel Lup 20, Lsbon, Porugal, emal: aafonso@seg.ul.p. UECE s suppored by Fundacão para a Cênca e a Tecnologa (Poruguese Foundaon for Scence and Technology) hrough he projec PEs-OE/EGE/UI0436/2011. IMF, Fscal Affars Deparmen, h sree NW, Washngon DC 20431, USA. emal: jjalles@mf.org. Cener for Globalzaon and Governance, Nova School of Busness and Economcs, Lsboa, Porugal. emal : joaojalles@gmal.com. 1

2 1. Inroducon Governmens end o absorb a szeable share of socey s resources and, herefore, hey affec economc developmen and growh n many counres. 1 Throughou hsory hgh levels of economc developmen have been aaned wh governmen nervenon. Where governmen dd no exs, lle wealh was accumulaed. However, despe necessary, governmen nervenon s no a suffcen condon for prospery, f leads o he monopolzaon of he allocaon of resources and oher mporan economc decsons, and socees dd no succeed n aanng hgher levels of ncome. 2 Theorecally here can be a pon for a posve effec of governmen spendng on growh, essenally va a Keynesan vew upon whch he governmen mpulse o aggregae demand plays an upward role. Usually, one would hen call upon fscal polcy o expand n order o foser growh, ceers parbus. On he oher hand, he negave effec of governmen spendng on growh can sem from he deas of he fscal heory of he prce level, where a afer a conracon n governmen spendng, he creaon of expecaons on consumers pon o fuure lower axes and an upward push n growh, also ceers parbus. In addon, economc progress s lmed when governmen s zero percen of he economy (absence of rule of law, propery rghs, ec.), bu also when s closer o 100 percen (he law of dmnshng reurns operaes n addon o, e.g., ncreased axaon requred o fnance he governmen s growng burden whch has adverse effecs on human economc behavour, namely on consumpon decsons). Ths dea s relaed o he so-called Armey Curve (Armey and Armey, 1995), who borrowed a graphcal echnque popularzed by Arhur Laffer. Fredman (1997) suggesed ha he hreshold a whch governmen affecs economc growh s beween 15-50% of he correspondng naonal ncome. The exsng leraure, noably he emprcal work spannng from Barro (1990) o Gemmell e al. (2011), assess denfcaon ssues relaed wh esablshng he growh effecs of dfferen ypes of publc spendng and axaon, and also presens mxed resuls as o he relaonshp beween governmen sze and economc developmen (for a recen survey see Bergh and Henrekson, 2011). Imporan dfferences n exsng research concern he measuremen of governmen sze, he ype of counres suded (rch vs. poor) and he me span consdered. On he one hand, he former may mpac economc growh negavely due o governmen neffcences, crowdng-ou effecs, excess burden of axaon, dsoron of he ncenves 1 Accordng o he Wagner s Law he scope of he governmen usually ncreases wh he level of ncome because governmen has o manan s admnsrave and proecve funcons, s aemps o ensure he proper operaon of marke forces and provson of socal and culural (publc) goods. 2 Publc choce explanaons of governmen growh are dscussed n Holcombe (2005). 2

3 sysems and nervenons o free markes (Barro, 1991; Bajo-Rubo, 2000). Indeed, several sudes repor ha he effcency of governmen spendng can ncrease, eher by delverng he same amoun of servces wh fewer resources or by usng more effcenly exsng spendng levels (see Afonso e al., 2005; Angelopoulos e al., 2008). Moreover, Slemrod (1995) and Tanz and Zee (1997) fnd a negave mpac f he sze of governmen exceeds a ceran hreshold. The raonale behnd hs argumen s ha n counres wh bg governmens he share of publc expendures desgned o promoe prvae secor producvy s ypcally smaller han n counres wh small governmens (Folser and Henrekson, 2001). On he oher hand, governmen acves may also have posve effecs due o benefcal exernales, he developmen of a legal, admnsrave and economc nfrasrucure and nervenons o offse marke falures (Ghal, 1998). On he debae beween he posve vs. negave effecs of governmen growh, Grossman (1988) suggesed ha a non-lnear model was preferred n explanng s mpac on oal economc oupu. Addonal assessmens are provded noably by Dar and Amrkhalkhal (2002) who repor for OECD counres ha Toal Facor Producvy growh and producvy of capal are weaker n counres wh larger governmens. In he same ven, Afonso and Jalles (2013b), who repor a negave effec of he deb rao on growh, and ha fscal consoldaon promoes growh, noably for OECD counres. Our movaon also comes from Guseh (1997) who presens a model ha dfferenaes he effecs of governmen sze on economc growh across polcal sysems n developng counres. Growh n governmen sze has negave effecs on economc growh, bu he negave effecs are hree mes as grea n non-democrac sysems as n democrac sysems. Undersandng he lnk beween governmen sze, nsuons and growh ress s mporan because, on he one hand, larger governmens ncrease he poenal for ren-seekng ha adversely affecs nnovaon. On he oher hand, beer nsuons can mgae ren-seekng acves and hereby nduce he reverse effecs on growh (see, e.g., Chaudhry e al., 2007). Our paper ncludes several conrbuons: ) we analyse a wde se of 140 counres composed of boh advanced and emergng and low ncome counres, usng a long me span runnng from , and employng dfferen proxes for governmen sze and nsuonal qualy o ncrease robusness; ) we buld new measures of exreme-ype polcal regmes whch are hen neraced wh approprae governmen sze proxes n non-lnear economerc specfcaons; ) we make use of recen panel daa echnques ha allow for he possbly of heerogeneous dynamc adjusmen around he long-run equlbrum relaonshp as well as heerogeneous unobserved parameers and cross-seconal dependence (e.g. Pooled Mean Group, Mean Group, 3

4 Common Correlaed Effecs esmaors, ner ala); and v) we also deal wh poenally relevan endogeney ssues. Our resuls sugges he exsence of a negave effec gong from large governmens o growh. Ineresngly, governmen consumpon s conssenly dermenal o oupu growh rrespecve of he counry sample consdered (OECD, Emergng and Low Income Counres). On he oher hand, nsuonal qualy has a sgnfcan posve mpac on he level of real GDP per capa. Moreover, ) he negave effec of governmen sze on GDP per capa s sronger a lower levels of nsuonal qualy, and ) he posve effec of nsuonal qualy on GDP per capa s sronger a smaller levels of governmen sze. Furhermore, he sably of he coeffcen esmaes seems o have been relavely consan over me. Fnally, our resuls prove hemselves robus o several robusness analyzes. The remander of he paper s organsed as follows. Secon 2 presens he analycal framework. Secon 3 presens he daa. Secon 4 elaboraes on he economerc mehodology and dscusses our man resuls. Secon fve concludes. 2. Analycal framework In hs secon we presen a growh model ha relaes oupu and governmen sze and wll provde he heorecal movaon for our emprcal (panel) analyss n Secon 3. Our model fs whn a broader leraure ha expands a Barro (1991)-ype model where governmen plays an acve role. 3 We consder a ypcal economy wh a consan elascy of subsuon uly funcon of he represenave agen gven by: U 0 e 1 c 1 d 1 (1) where c s per capa consumpon, s he neremporal subsuon and s he (subjecve) me dscoun rae or rae of me preference (a hgher mples a smaller desrably of fuure consumpon n erms of uly compared o uly obaned by curren consumpon. Populaon (whch we assume dencal o labour force, L) grows a he consan rae n, ha s, L L 0 e n producon funcon:. Oupu n each counry a me s deermned by he followng Cobb-Douglas 3 Peden and Bradley (1989) employ a heorecal model of oupu growh o derve an equaon ha conrols for cyclcal nfluencces and dsngushes he effecs of governmen growh on he economc base from he effecs on he economc growh rae. Lee (1992) and Devarajan e al. (1996) expand Barro s model, allowng dfferen knds of governmen expendures o have dfferen mpacs on growh. A a more dsaggregaed level, dsngushng beween producve and non-producve spendng, Glomm and Ravkumar (1997) and Kneller e al. (1999) are able o deermne he opmal composon of dfferen knds of expendure, based on her relave elasces. Smlarly, Chen (2006) nvesgaes he opmal composon of publc spendng and s relaonshp o economc growh. 4

5 Y K G A L. (2) 1 ( ),0 1, 0 1, 0 1 Y s he fnal good, used for prvae consumpon, G s publc consumpon expendure, whch proxes for governmen sze, and K s nvesmen n physcal capal. We consder he case of no deprecaon of capal. The oupu used o produce G equals qg (whch one can hnk of as beng equvalen o a crowdng-ou effec n prvae secor s resources). A s he level of echnology and grows a he exogenous consan rae, ha s, we have A I 0 (3) A e wh I beng a vecor of nsuonal qualy, polcal regme and oher relaed facors ha may affec he level of echnology and effcency n counry a me, and s a vecor of (unknown) coeffcens relaed o hese varables. In hs framework, he sae of labouraugmenng echnology (A) depends no only on exogenous echnologcal mprovemens deermned by, bu also on he level of nsuonal qualy (such as he rule of law), he degree of democrac polcal foundaons, ec. Insuons may be crcal n faclang echnologcal breakhroughs, whch may no occur whou approprae sound nsuonal envronmens. The presence of effcen and effecve nsuons ensures ha labour can be used for producve purposes, nsead of beng wased wh red ape or ren seekng acves (Norh, 1990; Nelson and Sampa, 2001). We begn by wrng down he resource consran for hs economy n per worker erms, gven by: where K Y C qg k y c qg nk (4) K s he me dervave of physcal capal and small leers represen per worker erms (afer scalng down by L). We now wre he condons ha characerze he opmal pah for he economy and deermne he seady-sae soluon for prvae and publc consumpon and ncome per worker. The opmal pah s he soluon of: max c, g 0 s..: k e k 1 c 1 d 1 g A 1 c qg nk. (5) Solvng he Hamlonan s correspondng frs order condons and afer some manpulaons yelds: 5

6 k g y c * * * * A A k y 1 1 * * g n q * A k * 1 * * ( n ) k qg q 1 1. (6) A specal case occurs when 1 and n 0 n whch here s no ranson dynamcs and he economy s always n he balanced growh pah. We refran from makng full consderaons on he model s soluon, bu one, n parcular, s worh makng: 4 an ncrease n q (whch mplcly proxes he overall sze of he publc secor ranslang he fac ha more resources are needed/requred o fnance G) reduces boh he opmal level of prvae consumpon per worker (and physcal capal per worker) and, more mporanly, he opmal level of oupu per worker n hs model economy. Turnng o economerc specfcaon, n he seady sae, oupu per effecve worker ˆ ) s consan whle oupu per worker ( y Y / L ) grows a he exogenous ( y Y / A L rae. In general, oupu n effecve worker erms evolves as yˆ ( k ) ( g ) and n (raw) worker erms, oupu evolves accordng o y A ( k ) ( g ). Takng logs on boh sdes we ge ln y ln A ln k ln g, and usng (3) and he fac ha n (2) we have ( A L ) 1 enerng he uly funcon, we oban, ln y A0 (1 ) (1 ) I ln k ln g. (7) Equaon (7) descrbes he evoluon of oupu per worker, as a funcon of a vecor of nsuonal and polcal relaed varables, whch may change over me, he sze of he publc secor or governmen, he level of physcal capal and he exogenous growh rae of oupu. Gven he producon funcon relaonshp, (7) s vald boh whn and ousde he seady-sae and hs s mporan, parcularly, f one makes use of sac panel daa echnques for esmaon purposes. Moreover, s no dependen on assumpons on he behavour of savngs, hence offerng a reasonable bass for esmaon. Takng (7) as he baselne we wll hen augmen by splng k no physcal (k) and human capal (h) 5, and use boh a lnear and non-lnear specfcaon (n whch neracon erms are ncluded), as follows: 4 In an alernave seng n whch he governmen nroduces a ax over oal ncome (or producon) o fnance publc consumpon, he overall concluson (wh respec o he effec of governmen sze) does no change. 5 We hank an anonymous referee for rasng he ssue ha one should nclude an educaon proxy. 6

7 ln y b b b I b ln k b ln h b ln g (8) ln y b b b I b ln k b ln h b ln g b ( I g ) (9) where he b s are (unknown) parameers o be esmaed, I and g denoe he proxes for nsuonal qualy and governmen sze, respecvely, and and are model specfc error erms sasfyng he usual assumpons of zero mean and consan varance. Equaons (8) and (9) provde he bass for he emprcal models o be esmaed n Secon 4. Fnally, he varaon of causaly beween governmen sze and growh deeced n crosssecon and me-seres papers suggess ha here are mporan dfferences n he way n whch governmens nfluence economc performance across counres. We argue ha may reflec nsuonal dfferences across counres and, whle hs s a plausble conjecure, here s as ye lle drec evdence o confrm ha nsuons and polcal regmes make a dfference o he way n whch governmens affec economc oucomes. 3. Daa The daase consss of an unbalanced heerogeneous panel of 140 counres over he perod n 5-year non-overlappng averages (o overcome shor-run busness cycle flucuaons as s common pracce n he emprcal growh leraure). 6 Counres are grouped no advanced (OECD), emergng marke economes (EME) and low ncome counres (LIC) based on he World Bank classfcaon. Annual daa on real GDP per capa (y) and gross fxed capal formaon (nv) are rereved from he World Bank World Developmen Indcaors. We esmae he capal sock (Ky) usng he perpeual nvenory mehod, ha s, Ky Inv, where Inv s he nvesmen and s he deprecaon rae. Daa on ( 1 ) Ky 1 Inv comes from Summers and Heson s PWT 7.1 as real aggregae nvesmen n PPP. We esmae he nal value of he capal sock ( Ky 0 ), n year 1950 as Inv /( g ) where g s 1950 he average compound growh rae beween 1950 and 1960, and s he deprecaon rae (se o 7% for all counres and years). In addon we use average years of schoolng from Barro- Lee daase as our proxy for human capal. The proxes for governmen sze (our G) are governmen consumpon (govcons_gdp), oal governmen expendures (ogovexp_gdp), oal governmen revenues (ogovrev_gdp), and oal governmen deb (govdeb_gdp) all expressed n percenage of GDP. They come from he IMF s Inernaonal Fnancal Sascs (IFS). 6 Summary sascs and correlaon marces are omed for economy of space bu hey are avalable upon reques. 7

8 For nsuonal-relaed varables (our I) we rely on: ) Freedom House s Polcal Rghs, Cvl Lberes, ) he Poly 2 ndex, a democracy ndex and he regme durably (n years) from Marshall and Jaegger s Poly s 4 daabase, ) an ndex of democrazaon due o Vanhanen (2005), v) a governance ndex and s 6 sub-componens 7 from Kaufman e al. (2009) (World Bank Governance Indcaors) 8 ; v) governmen fraconalzaon from he Qualy of Governmen Daase 9 ; v) he KOF ndces of economc, polcal and socal dmensons of globalzaon 10 ; years n offce, ndex of legslave compeveness, ndex of execuve compeveness, checks and balances and dummy for legslave elecon held all from he World Bank s Daabase on Polcal Insuons Mehodology and Resuls 4.1 Baselne Resuls Equaons (8) and (9) can be esmaed drecly usng panel daa echnques, whch allow for boh cross-secon and me-seres varaon n all varables. Table 1 presens our frs se of resuls for he fxed-effecs specfcaons coverng dfferen proxes for nsuonal qualy (egh n oal). We use governmen consumpon for our baselne, as he more obvous spendng polcy measure avalable o he fscal auhores, and dscuss s ndvdual ncluson n our regresson of neres as well as s neracon wh a varable I. [Table 1] We fnd a posve effec of he capal sock on he level of real GDP per capa hroughou he dfferen specfcaons regardless of he nsuonal varable employed. The posve role of human capal s also evden from hese regressons n accordance o oher sudes. One also fnds a sascally sgnfcan negave coeffcen on he governmen sze proxy n 10 ou of 16 specfcaons. Is coeffcen vares beween and -0.11, meanng ha an ncrease n governmen sze by 10 percenage pons, s assocaed wh a 1.1% o 0.8% lower annual growh per capa. Ths order of magnude s conssen wh prevous leraure. Insuonal qualy, generally speakng, has equally a conssen and sascally sgnfcan posve mpac 7 Ths s he resul of averagng sx varables: voce and accouably, polcal sably, governmen effecveness, regulaory qualy, rule of law and conrol of corrupon. 8 The neresed reader should refer o he orgnal sources for he full defnon of he varables used. 9 hp:// 10 hp://globalzaon.kof.ehz.ch/. 11 hp://econ.worldbank.org/wbsite/external/extdec/extresearch/0,,conenmdk: ~pagep K: ~pPK: ~heSePK:469382,00.hml. 8

9 on economc acvy, hough s order of magnude dffers beween he seleced proxy under scruny. When sascally sgnfcan he neracon erm s negave, meanng ha ) he negave effec of governmen sze on GDP per capa s sronger a lower levels of nsuonal qualy, and ) he posve effec of nsuonal qualy on GDP per capa s sronger a smaller levels of governmen sze. The neracon erm means ha he margnal effec of governmen sze wll dffer a dfferen levels of nsuonal qualy. However, hs resul depends on he proxy used for I. Neverheless, we oban n mos regressons consderably hgh R-squares. Moreover, when regonal dummes are ncluded, coeffcens keep her sascal sgnfcance and sgn (no shown). We go on n furher explorng he robusness of our resuls o oher nsuonal proxes and he resuls are generally n lne wh ones already repored. In Table 2, we observe a sascally sgnfcan negave coeffcen for he governmen sze proxy n 12 ou of 16 specfcaons, and he coeffcen s range compares broadly n magnude wh one repored before. [Table 2] Moreover, f one uses also he World Bank s Governance (WBG) ndcaors, as a robusness check and despe he fac ha hese are only avalable from 1996 onwards, he conclusons sll hold (see Table 3). Lkewse, and complemenng he aggregae full sample esmaon we presen a bar char (Fgure 1) by counry group (Advanced Economes, Emergng Marke Economes and Low Income Counres) where we dsplay only he sascally sgnfcan coeffcens, for he respecve WBG ndces. Voce and accounably as well as rule of law are he wo mos mporan governance aspecs o mpac GDP per capa n OECD counres; n emergng marke economes more mporance s arbued o governmen effecveness; n low ncome counres regulaory qualy and rule of law sand ou as he mos mporan ndcaors. [Table 3] [Fgure 1] In relaon o he evoluon over me of he relevance of boh governmen sze and nsuonal qualy, Fgure 2 and Fgure 3 summarse respecvely he relevan sascal esmae coeffcens comng from he esmaon of Equaon (8). Checkng he me dmenson consues an addonal robusness check and we do no observe any swchng sgns over me; more mporanly for mos proxes esmaes have been relavely sable hroughou he perod under consderaon. [Fgure 2] 9

10 well behaved. 13 [Table 4] [Fgure 3] 4.2 Dynamc Panel Esmaon: Accounng for Endogeney In he analyss of emprcal producon funcons, he ssue of varable endogeney s generally of concern. Tha sad, nsead of esmang sac equaons, we now allow for dynamcs o play an acve role. A negave correlaon beween governmen sze and economc acvy does no mply causaly. In fac, he mos obvous reason (among many) o suspec reverse causaly a problem s ha welfare saes socal nsurance schemes ac as auomac sablzers. We reformulae our regresson equaon(s) o nclude addonally he lagged real GDP per capa (he remander of he varables s unchanged). We esmae by means of he Arellano-Bover sysem-gmm esmaor whch jonly esmaes he equaons n frs dfferences, usng as nsrumens lagged levels of he dependen and ndependen varables, and n levels, usng as nsrumens he frs dfferences of he regressors. 12 Inuvely, he sysem- GMM esmaor does no rely exclusvely on he frs-dfferenced equaons, bu explos also nformaon conaned n he orgnal equaons n levels. The resuls n Table 4 for a seleced number of nsuonal qualy proxes suppor he prevous fndngs. Moreover, dagnosc sascs, n parcular he valdy of he nsrumen se used (assessed va he Hansen es), are 4.3 Usng New Polcal Sysem s Measures Followng Rodrk and Waczarg s (2005) approach we consruc new (and more meanngful) democracy measures based on he varable Poly. The role of polcal sysems and democracy n parcular, on he governmen sze-growh relaonshp s assessed by regressng hree srucural aspecs of democracy (o be defned below) on 5-year averages of real GDP per capa growh raes. 14 Indeed, poly does no capure wo mporan dmensons of polcal 12 As far as nformaon on he choce of lagged levels (dfferences) used as nsrumens n he dfferences (levels) equaon, as work by Bowsher (2002) has ndcaed, when comes o momen condons (as hus o nsrumens) more s no always beer. The GMM esmaors are lkely o suffer from overfng bas once he number of nsrumens approaches (or exceeds) he number of counres (as a smple rule of humb). In he presen case, he choce of lags was dreced by checkng he valdy of dfferen ses of nsrumens. 13 In he grea majory of our sysem-gmm regressons he Hansen-J sasc s assocaed wh p-values larger han 10%. Ths sasc ess he null hypohess of correc model specfcaon and vald overndenfyng resrcons,.e., valdy of nsrumens. 14 An equaon wh real GDP per capa growh as he dependen varable s movaed by (sandard) augmenaon of Solow-Swan ype models wh a governmen sze proxy and followng Barro and Sala--Marn s (1992) approach. 10

11 regmes - eher her newness (followng, for example, democrazaon or a reurn o auhoraran rule) or her more esablshed (consoldaed) naure. Rodrk and Waczarg (2005) defne a major polcal regme change o have occurred when here s a shf of a leas hree pons n a counry s score on poly over hree years or less. Usng hs creron we defne new democraces (ND=1) n he nal year (and subsequen four years) n whch a counry s Poly score s posve and ncreases by a leas hree pons and s susaned, ND=0 oherwse. Esablshed democraces (ED=1) are hose new democrac regmes ha have been susaned followng he 5 years of a new democracy (ND). In any subsequen year, f esablshed democraces (ED) fal o susan he saus of ND, ED=0. Usng hese crera, hey defne susaned democrac ransons (SDT) as he sum of ND and ED. They use he same procedure, muas muands, o defne new auocraces (NA), esablshed auocraces (ES) and susaned auocrac ranson (SAT). Ths yelds sx dsnc bnary-ype measures of he characer of polcal regmes - ND, ED, NA, EA, SDT, and SAT - for mos years durng Fnally, Rodrk and Waczarg (2005) defne small regme changes (SM) as changes n poly from one year o he nex ha are less han hree pons. 15 There are several advanages from creang hese new measures, whch allow us o dsngush he mpac of new and esablshed elecoral democraces and auocraces on economc developmen. 16 Endogeney 17 beween rgh-hand sde proxes of democracy and auocracy and a sandard se of conrol varables s correced for by akng a sysem-gmm (SYS-GMM) approach as dealed above. As suggesed n Mauro (1995), Hall and Jones (1999) and Acemoglu e al. (2001), he democracy measures are nsrumened by: 1. he durably (age n years) of he polcal regme ype (durable) rereved from Marshall and Jaeggers daabase laude: Hall and Jones (1999) launched he general dea ha socees are more lkely o pursue growh-promong polces, he more srongly hey have been exposed o Wesern European nfluence, for hsorcal or geographcal reasons. Two oher possble nsrumens could be common and cvl law, ranslang he ype of legal orgn of each counry. 3. ehnc fragmenaon (ehnc): on a broad level, he role of ehnc fragmenaon n explanng he (possble) growh effec of democracy can be derved from he leraure on he economc consequences of ehnc conflc. I has been shown ha he level of rus s 15 Thus SM = 1 for a small regme change and SM = 0 oherwse. 16 An emprcal applcaon of hese measures o explan he mpac of exreme-ype polcal regmes on economc performance can be found n Jalles (2010). 17 And also he exsence of possble measuremen errors when accounng for democracy. 18 The average age of he pary sysem s also used n Przeworsk e al. (2000). 11

12 low n an ehncally dvded socey (Alesna and La Ferrara, 2000). Moreover, he lack of co-operave behavour beween dverse ehnc groups, leads o he ragedy of he commons as each group fghs o dver common resources o non-producve acves (Mauro, 1995). 19 Table 5 repors he resuls wh he four proxes for governmen sze. We observe ha governmen sze appear mosly wh sascally sgnfcan negave coeffcens. When neraced wh SDT has a posve and sascally sgnfcan coeffcen, meanng ha n democrac counres he negave mpac of governmen sze s mgaed bu remans mosly negave. The remanng proxes keep he sascally negave coeffcen, bu neracon erms lose economc and sascal relevance. For he OECD sample he ndvdual effecs of he dfferen proxes of governmen sze are smlar bu neracon erms are never sascally sgnfcan. Emergng marke and low ncome counres repor a sascally negave coeffcen on publc consumpon, oal governmen expendure and deb-o-gdp rao, wh he laer havng a lesser dermenal effec n democrac counres (see Fgure 4 for a graphcal summary). All n all, governmen consumpon s he proxy ha s more conssenly and more dermenal o oupu growh and he one ha should be used for he remander of he paper. [Table 5] [Fgure 4] As suggesed by Ram (1986) anoher possble specfcaon s he use of he growh rae of he governmen sze proxy. We also esed hs specfcaon o deermne s mpac on growh across dfferen levels of nsuonal qualy. All varables were reaned excep replaced by dg / G G ha s now ogeher wh he correspondng neracon erms (resuls avalable upon reques). Comparng wh our prevous resuls he coeffcens of he lnear erm of governmen sze proxes were posve and sascally sgnfcan n wo ou of fve specfcaons. However, accordng o Cone and Darra (1988) Ram s specfcaon s suable for esng shor-erm effecs, whle he specfcaon used n hs paper assesses he effecs of governmen sze on he underlyng longer-run economc acvy. 4.4 Robusness Checks One concern when workng wh me-seres daa s he possbly of spurous correlaon beween he varables of neres (Granger and Newbold, 1974). Ths suaon arses when 19 Oher smlarly possble nsrumens are he hsorcal seler moraly or populaon densy n 1500, as n Acemoglu and Robnson (2005), he share of populaon ha speaks any major European language - Eurfrac -, ner ala. For he hree nsrumens chosen he excluson resrcon s ha durably, laude and ehnc fragmenaon do no have any mpac on presen economc growh oher han her mpac on democracy. 12

13 seres are no saonary, ha s, hey conan sochasc rends as s largely he case wh GDP and nvesmen seres. Resuls of frs (Im-Pesaran-Shn, 1997) and second generaon (Pesaran, 2007) panel negraon ess (no shown) sugges ha we can accep mos conservavely ha non-saonary canno be ruled ou n our daase. I seems ha he me-seres properes of he daa play an mporan role: we sugges ha he bas n our models s he resul of nonsaonary errors, whch are nroduced no he fxedeffecs and GMM equaons by he mposon of parameer homogeney. Hence, careful modellng of shor-run dynamcs requres a slghly dfferen economerc approach. We assume ha (8), or (9), represens he equlbrum whch holds n he long-run, bu ha he dependen varable may devae from s pah n he shor-run (due, e.g., o shocks ha may be perssen). There are ofen good reasons o expec he long-run equlbrum relaonshps beween varables o be smlar across groups of counres, due e.g. o budge consrans or common echnologes (unobserved TFP) nfluencng hem n a smlar way. In lne wh dscussons n he emprcal growh leraure we shall assume ha he long-run relaonshp s composed of a counryspecfc level and a se of common facors wh counry-specfc facor loadngs. The parameers of (8) and (9) can be obaned alernavely va recen panel daa mehods. We resor o he Mean Group (MG) esmaor (Pesaran and Smh, 1995) and he Pooled Mean Group (PMG) esmaor, whch nvolves boh poolng and averagng (Pesaran e al., 1999). These esmaors are approprae for he analyss of dynamc panels wh boh large me and cross-secon dmensons, and hey have he advanage of accommodang boh he long-run equlbrum and he possbly heerogeneous dynamc adjusmen process. A second sep s o make use of he Common Correlaed Effecs Mean Group (CCEMG) esmaor ha accouns for he presence of unobserved common facors whose presence facors s acheved by consrucon and he esmaes are obaned as averages of he ndvdual counry-specfc esmaes (Pesaran, 2006). A relaed and recenly developed approach due o Eberhard and Teal (2010) was ermed Augmened Mean Group (AMG) esmaor and accouns for cross-seconal dependence by ncluson of a common dynamc process. The panel analyss ha follows s based on hs unresrced error correcon ARDL(p,q) represenaon: y p1 q1 y ' x y ' x u, 1,2,..., N; 1,2,..., T (10) 1 1 j j j j j1 q1 where y s a scalar dependen varable, x s he k 1 vecor of regressors for group, represens he fxed effecs, s a scalar coeffcen on he lagged dependen varable. ' s s he k 1vecor of coeffcens on explanaory varables, j s are scalar coeffcens on lagged 13

14 frs-dfferences of dependen varables, and j s are k 1 coeffcen vecors on frsdfferences of explanaory varables and her lagged values. We assume dsurbances u o be ndependenly dsrbued across and, wh zero mean and consan varance. Assumng ha 0 for all, here exss a long-run relaonshp beween y and x defned as: y ' x 1, 1,2,..., N; 1,2,..., T (11) where ' '/ s he k 1 vecor of he long-run coeffcens, and s are saonary wh possble non-zero means (ncludng fxed effecs). Equaon (10) can be rewren as: y p1 q1 y ' x u, 1,2,..., N; 1,2,..., T (12) 1 j j j j j1 q1 where 1s he error correcon erm gven by (5), hence s he error correcon coeffcen measurng he speed of adjusmen owards he long-run equlbrum. Table 6 presens our frs se of robusness resuls, and ncludes for each sub-sample boh he PMG and MG esmaes usng Poly as our chosen proxy for nsuonal qualy whch eners n lnear form ogeher wh publc consumpon (Panel A) and well as n mulplcave form (Panel B). 20 For he PMG esmaes, he OECD sub-group obans a posve and sascally sgnfcan coeffcen on nsuons and sascally negave coeffcens on governmen sze when usng he boh he PMG and MG esmaors. One should expec rch counres o ge a negave correlaon beween governmen sze and oupu f hough n erms of he Olson s (1982) mechansm: organzed neres groups end o evolve, and sruggle o ge advanages for hemselves n he form of ransfers or legslaon, whch have a sde effec, delayng he regular funconng and growh of economy. The scope for neres group acon s lkely o be greaer n counres wh larger governmens, where here s ncreased poenal for profs from ren-seekng acves, leadng o a greaer dverson of resources o unproducve ends (Buchanan, 1980). Recenly, Bergh and Karlsson (2010) also uncovered a dermenal oupu effec of larger governmens n a panel of rch counres usng he Bayesan Average over Classcal Esmaes approach. For boh emergng and low ncome counres sascal sgnfcance of governmen sze s weaker bu he nsuonal proxy s sll sascally sgnfcan for emergng counres and for low ncome counres. [Table 6] The MG esmaor provdes conssen esmaes of he mean of he long-run coeffcens, hough hese wll be neffcen f slope homogeney holds. Under long-run slope homogeney, 20 Ths secon reles on presenng resuls from ncreasngly sophscaed economerc echnques, raher han, for nsance, allowng for a wder se of conrol varables o play a role, smply because our nenon s o solae he effecs of our man varables of neres and addng more conrols could conamnae hose esmaes. 14

15 he pooled esmaors are conssen and effcen. The hypohess of homogeney s esed emprcally n all specfcaons usng a Hausman-ype es appled o he dfference beween MG and PMG. 21 The p-value of such a es s also presen n Table 6, and only for he OECD he null s rejeced, beng he MG esmaor more effcen, and he long-run slope homogeney rejeced. An equvalen se of resuls wh he neracon erm beween publc consumpon and our nsuonal proxy s presened n Panel B. I shows ha n he case of he OECD he neracon erm s negave and sascally sgnfcan. However, n he case of emergng and low ncome counres he neracon erm s no longer sascally sgnfcan. Fnally, n Table 7 when ones allows for heerogeneous echnology parameers and facor loadngs, we sll ge negave and sascally sgnfcan coeffcens on our governmen sze proxy whch s n lne wh Romero-Avla and Srauch (2008) who found a negave a sgnfcan effec from governmen consumpon (and ransfers) on oupu. However, n he case of emergng and low ncome counres such effec s no longer as srong and somemes even posve (even f no sascally sgnfcan). 22 We compare dfferen economerc specfcaons for robusness and compleeness. Sll, for he OECD sample here are negave neracon erms: ) he negave effec of publc consumpon on real GDP per capa s sronger a lower levels of nsuonal qualy, and ) he posve effec of nsuonal qualy on real GDP per capa ncreases a smaller levels of publc consumpon. [Table 7] 5. Concluson We provde new evdence on wheher oo much governmen s good or bad for economc progress and macroeconomc performance, parcularly when assocaed wh dfferenaed levels of (underlyng) nsuonal qualy and alernave polcal regmes. We conduc an emprcal panel exercse akng 140 counres from and employng dfferen proxes for governmen sze and nsuonal qualy. Moreover, we make use of recen panel daa echnques ha allow for he possbly of heerogeneous dynamc adjusmen around he longrun equlbrum relaonshp as well as heerogeneous unobserved parameers and crossseconal dependence; we also deal wh poenally relevan endogeney ssues. 21 Under he null hypohess he dfference n he esmaed coeffcens beween he MG and he PMG esmaors s no sgnfcan and he PMG s more effcen. 22 In poor counres publc secors are ypcally small, and he relaonshp beween governmen sze and oupu can even be posve (because a sae ypcally succeeds n collecng axes when successful a provdng he sably necessary for economc acvy sound nsuons o sar growh) see Besley and Persson (2009). 15

16 Our fndngs can be summarzed as follows: ) generally speakng, our resuls seem o sugges ha bgger governmens end o hamper economc acvy and hs fndng s nvaran o he seleced governmen sze proxy under scruny, counry groupng analysed or esmaon mehod employed; ) nsuonal qualy has a posve mpac on he level of real GDP per capa, hough resuls are dependen on he proxy consdered; ) he negave effec of governmen sze on real GDP per capa s sronger a lower levels of nsuonal qualy, and he posve effec of nsuonal qualy on GDP per capa s sronger a smaller levels of governmen sze. A possble nerpreaon for our resuls ress on acknowledgng ha mos counres may already be a he slppery sde of he Armey curve, where spendng s no producve enough and/or governmen fnancng s oo dsoronary. A leas one cavea should be menoned and hs relaes o he fac ha we do no dfferenae beween dfferen publc spendng ems. Ths s dffcul a he aggregaon level consdered and would go beyond he scope of hs paper. Neverheless, Afonso and Jalles (2013a) sudy he relaonshp of educaon, socal secury and healh publc spendng and he economc busness cycle n more deal. Fnally, he deermnaon of he opmal level of governmen and oher ssues relaed o publc secor effcency, hough neresng quesons per se, were beyond he scope of hs paper and reman canddaes for fuure research. 16

17 Aknowledgmens We are graeful o Jula Lendva, Nklas Porafke, o parcpans a he 14 h Banca d'iala Publc Fnance Workshop (Peruga), and a he 68 h Annual Conference of he Inernaonal Insue of Publc Fnance (Dresden), for useful commens on prevous versons. We also hank wo anonymous referees for useful suggesons ha sgnfcanly mproved he paper. The opnons expressed heren are hose of he auhors and do no necessarly reflec hose of IMF or s member counres. References 1. Acemoglu, D. and Robnson, J. (2005), "Economc Orgns of Dcaorshp and Democracy", Cambrdge Unversy Press. 2. Acemoglu, D., Johnson, S. and Robnson, J.A. (2001),"The colonal orgns of comparave developmen: an emprcal nvesgaon", Amercan Economc Revew, 91, Afonso, A., and Jalles, J. T. (2013a), The cyclcaly of educaon, healh, and socal secury governmen spendng, Appled Economcs Leers, 20 (7), Afonso, A., Jalles, J. T. (2013b). Growh and Producvy: he role of Governmen Deb, Inernaonal Revew of Economcs and Fnance, 25, Afonso, A.; Schuknech, L., Tanz, V. (2005). Publc Secor Effcency: an Inernaonal Comparson, Publc Choce, 123 (3-4), Alesna, A. and La Ferrara, E. (2000), "The Deermnans of Trus", NBER Workng Papers Angelopoulos, K., Phlppopoulos, A., Tsonas, E. (2008), Does publc secor effcency maer? Revsng he relaon beween fscal sze and economc growh n a world sample, Publc Choce, 137 (1-2), Armey, D. and Armey, R. (1995), The freedom revoluon: he new republcan house majory leader ells why bg governmens faled, why freedom works and how we wll rebuld Amerca, Washngon, DC, Regnery Publshng Inc. 9. Bajo-Rubo, O. (2000), A furher generalzaon of he Solow growh model: he role of he publc secor, Economc Leers, 68, Barro, R. (1990), Governmen spendng n a smple model of endogenous growh, Journal of Polcal Economy, 98 (5), s Barro, R. (1991), Economc growh n a cross secon of counres, Quarerly Journal of Economcs, 106, Barro, R. and Sala--Marn, X. (1992), Convergence, Journal of Polcal Economy, 100, Bergh, A., Karlsson, M. (2010), Governmen sze and growh: accounng for economc freedom and globalzaon, Publc Choce, 142, Bergh, A., Henrekson, M. (2011), Governmen Sze and Growh: a survey and nerpreaon of he evdence, Journal of Economc Surveys, 25(5), Besley, T., Persson, T. (2009), The orgns of sae capacy: propery rghs, axaon, and polcy, Amercan Economc Revew, 99(4), Bowsher, C. G. (2002), On esng overdenfyng resrcons n dynamc panel daa models, CEPR Dscusson Paper No. 3048, CEPR London. 17. Buchanan, J. (1980), Ren-seekng and prof-seekng, n J. M. Buchanan and G. Tullock (eds), Towards a Theory of he Ren-Seekng Socey. College Saon, TX. 17

18 18. Chaudhry, A. And Garner, P. (2007), Do governmens suppress growh?: nsuons, ren-seekng, and nnovaon blockng n a model of Schumpeeran growh, Economcs and Polcs, 19, 1, Chen, B. (2006), Economc Growh wh opmal publc spendng composon, Oxford Economc Papers, 58, Cone, M.A., and A.F. Darra (1988), Economc Growh and he Expandng Publc Secor, Revew of Economcs and Sascs, Dar, A., and Amrkhalkhal, S. (2002). Governmen sze, facor accumulaon and economc growh: Evdence from OECD counres, Journal of Polcy Modellng, 24, Easerly, W. R. (2001), The Los Decades: Developng Counres Sagnaon n Spe of Polcy Reform , Journal of Economc Growh, 6(2), Eberhard, M., and Teal, F. (2010), Producvy Analyss n Global Manufacurng Producon, Oxford Unversy, Dscusson Paper Seres #515, November Fölser, Sefan; and Magnus Henrekson "Growh effecs of governmen expendure and axaon n rch counres", European Economc Revew 45, Fredman, M. (1997), If only he US were as free as Hong Kong, Wall Sree Journal, July 8, A Gemmell, N., Kneller, R., Sanz, I. (2011), The mng and perssence of fscal polcy mpacs on growh: Evdence from OECD counres, Economc Journal, 121 (550), F33 F Ghal, K. H. (1998), Governmen sze and economc growh: evdence from a mulvarae conegraon analyss, Appled Economcs, 31, Glomm, G., Ravkumar, B. (1997), Producve governmen expendures and long-run growh, Journal of Economc Dynamcs and Conrol, 21, Goel, R., Nelson, M. (1998), Corrupon and governmen sze: a dsaggregaed analyss, Publc Choce, 97 (1-2), Granger, C. W. J. And Newbold, P. (1974), Spurous regressons n economercs, Journal of Economercs, 2, Grossman, P. (1988), Governmen and economc growh: a non-lnear relaonshp, Publc Choce, 56 (2), Guseh, J. S. (1997), Governmen sze and economc growh n developng counres: a polcaleconomy framework, Journal of Macroeconomcs, 19, Hall, R.E. and Jones, C. (1999), Why do some counres produce so much more oupu per worker han ohers? Quarerly Journal of Economcs, 114, Holcombe, R. (2005), Governmen growh n he weny-frs cenury, Publc Choce, 124 (1-2), Im, K. S., Pesaran, M. H. and Shn, Y. (2003), Tesng for un roos n heerogeneous panels, Journal of Economercs 115, IMF (2001), A Manual on Governmen Fnance Sascs 2001, (GFSM 2001), hp:// Washngon DC: Inernaonal Moneary Fund. 37. Jalles, J. T. (2010), Does democracy foser or hnder growh? Exreme-ype polcal regmes n a large panel," Economcs Bullen, 30(2), Kaufmann, D., Kraay, A. and Masruzz, M. (2009), Governance Maers VIII: Aggregae and Indvdual Governance Indcaors for , World Bank Polcy Research Paper No hp://ssrn.com/absrac= La Pora, R., López-de-Slanes, F., Shlefer, A. and Vshny, R. (1999), The Qualy of Governmen, Journal of Law, Economcs and Organzaon, 15(1), Lee, J. (1992), Opmal sze and composon o governmen spendng, Journal of he Japanese and Inernaonal Economes, 6,

19 41. Mauro, P. (1995), "Corrupon and growh", Quarerly Journal of Economcs, 110, Nelson, R. R. and B.N. Sampa (2001), Makng sense of nsuons as a facor shapng economc performance, Journal of Economc Behavor and Organzaon, 44 (2001) Norh, D. (1990), Insuons, Insuonal Change and Economc Performance, Cambrdge: Cambrdge Unversy Press. 44. Olson, M. (1982), The rse and declne of Naons: economc growh, sagflaon, and socal rgdes, New Haven, CN: Yale Unversy Press. 45. Peden, E., Bradley, M. (1989), Governmen sze, producvy and economc growh: he pos-war experence, Publc Choce, 61 (3), Pesaran, M. H. (2006), Esmaon and nference n large heerogeneous panels wh a mulfacor error srucure, Economerca, 74 (4), Pesaran, M. H. (2007), A smple panel un roo es n he presence of cross secon dependence, Journal of Appled Economercs, 22(2), Pesaran, M. H. and Smh, R. P. (1995), Esmang long-run relaonshp from dynamc heerogeneous panels, Journal of Economercs, 68, Pesaran, M. H., Shn, Y. and Smh, R. P. (1999), Pooled mean group esmaon of dynamc heerogeneous panels, Journal of Amercan Sascal Assocaon, 94, Przeworsk, A., Alvarez, M., Chebub, J., Lmong, F. (2000), "Democracy and Developmen: Polcal regmes and economc well-beng n he World, ", New York: Cambrdge Unversy Press. 51. Ram, R. (1986), Governmen Sze and Economc Growh: a New Framework and Some Evdence from Cross-secon and Tme Seres Daa, Amercan Economc Revew, 76, Rodrk, D., and R. Waczarg (2005), Do Democrac Transons Produce Bad Economc Oucomes?", Amercan Economc Revew, 95(3), Romero-Avla, D., Sruch, R. (2008), Publc fnances and long-erm growh n Europe: evdence from a panel daa analyss, European Journal of Polcal Economy, 24(1), Slemrod, J. (1995), Wha do cross-counry sudes each abou governmen nvolvemen, prospery, and economc growh?, Brookngs Papers on Economc Acvy, 2, Tanz, V., Zee, H. (1997), Fscal polcy and long-run growh, IMF Saff Papers, 44, Vanhanen, T. (2005), Measures of Democracy [compuer fle]. FSD1289, verson 2.0 ( ). Tampere: Fnnsh Socal Scence Daa Archve [dsrbuor]. 19

20 Table 1: Baselne resuls, full sample: Fxed Effecs esmaon wh neracon erms Esmaor Fxed Effecs Spec. (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16) Insuonal Proxy Cvl Lberes Polcal Rghs Poly Democracy Index Regme Durably Vanhanen Democracy Governmen Fraconalzaon Governance Physcal capal 0.601*** 0.601*** 0.610*** 0.608*** 0.768*** 0.800*** 0.670*** 0.687*** 0.671*** 0.569*** 1.598*** 1.556*** 0.870*** 0.869*** *** *** (0.145) (0.145) (0.144) (0.144) (0.150) (0.151) (0.150) (0.152) (0.147) (0.143) (0.268) (0.270) (0.156) (0.156) (0.159) (0.162) Human capal 0.271*** 0.267*** 0.250*** 0.246*** 0.258*** 0.246*** 0.314*** 0.308*** 0.335*** 0.311*** 0.225** 0.226** 0.266*** 0.267*** 0.640** 0.717** (0.083) (0.083) (0.083) (0.083) (0.082) (0.083) (0.085) (0.086) (0.080) (0.078) (0.111) (0.111) (0.083) (0.083) (0.308) (0.307) Publc Consumpon *** *** *** *** *** * *** *** *** ** (0.022) (0.056) (0.022) (0.049) (0.022) (0.040) (0.022) (0.046) (0.021) (0.027) (0.038) (0.054) (0.025) (0.028) (0.036) (0.038) Insuons 0.191* 0.421** 0.312*** 0.465*** 0.119*** 0.181*** 0.164*** 0.220*** 0.654*** 1.761*** 0.369** *** (0.101) (0.194) (0.081) (0.156) (0.023) (0.043) (0.043) (0.083) (0.079) (0.181) (0.176) (0.424) (0.045) (0.110) (0.047) (0.074) Ineracon * *** ** (0.113) (0.096) (0.027) (0.054) (0.101) (0.310) (0.065) (0.043) Observaons R-squared Number of counres Noe: The models are esmaed by Fxed-Effecs. The dependen varable s he logarhm of real GDP per capa. A consan erm and me effecs have been ncluded bu are no repored for reasons of parsmony. Robus heeroskedasc-conssen sandard errors are repored n parenhess below each coeffcen esmae. *, **, *** denoe sgnfcance a 10, 5 and 1% levels. Table 2: Baselne resuls, full sample: Fxed Effecs esmaon wh neracon erms Esmaor Fxed Effecs Spec. (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16) Insuonal Proxy KOF Socal KOF Polcal KOF Economc Years n Offce Legslave elecon Legslave compeveness Execuve compeveness Checks and Balances Physcal capal *** 0.502*** *** 0.749*** 0.807*** 0.812*** 0.807*** 0.812*** 0.805*** 0.831*** 0.747*** 0.756*** (0.153) (0.154) (0.163) (0.159) (0.167) (0.170) (0.152) (0.151) (0.152) (0.154) (0.154) (0.156) (0.153) (0.154) (0.151) (0.152) Human capal 0.210*** 0.210*** 0.283*** 0.269*** 0.179** 0.186** 0.233*** 0.235*** 0.248*** 0.248*** 0.242*** 0.241*** 0.225*** 0.216*** 0.211** 0.208** (0.074) (0.074) (0.080) (0.079) (0.085) (0.085) (0.081) (0.081) (0.082) (0.082) (0.082) (0.083) (0.082) (0.083) (0.083) (0.083) Publc Consumpon *** *** *** *** *** *** *** * *** *** *** *** (0.025) (0.070) (0.026) (0.055) (0.026) (0.048) (0.022) (0.031) (0.022) (0.028) (0.022) (0.044) (0.022) (0.042) (0.022) (0.030) Insuons 0.696*** 0.797*** 0.251** 0.666*** *** 0.013*** 0.098* ** *** 0.373*** (0.104) (0.238) (0.124) (0.226) (0.089) (0.191) (0.001) (0.004) (0.052) (0.140) (0.070) (0.135) (0.072) (0.138) (0.081) (0.182) Ineracon * *** (0.132) (0.122) (0.105) (0.002) (0.083) (0.008) (0.009) (0.012) Observaons R-squared Number of counres Noe: The models are esmaed by Fxed-Effecs. The dependen varable s he logarhm of real GDP per capa. A consan erm and me effecs have been ncluded bu are no repored for reasons of parsmony. Robus heeroskedasc-conssen sandard errors are repored n parenhess below each coeffcen esmae. *, **, *** denoe sgnfcance a 10, 5 and 1% levels. 20

21 Table 3: World Bank Governance Indcaors, full sample: Fxed Effecs esmaon wh neracon erms Esmaor Fxed Effecs Spec. (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) Insuonal Proxy Voce & Accounably Polcal Sably Governmen Effecveness Regulaory Qualy Rule of Law Conrol of Corrupon Physcal capal 0.397** *** 0.550*** 0.478*** 0.333** 0.534*** 0.487*** 0.518*** 0.440*** 0.431*** 0.279* (0.162) (0.174) (0.156) (0.157) (0.159) (0.163) (0.157) (0.163) (0.159) (0.161) (0.162) (0.163) Human capal 0.551* 0.661** 0.532* 0.522* 0.604* 0.709** 0.758** 0.800** 0.573* 0.646** 0.558* 0.639** (0.321) (0.319) (0.306) (0.307) (0.314) (0.310) (0.310) (0.312) (0.312) (0.311) (0.320) (0.313) Publc Consumpon *** ** *** *** *** ** *** *** *** ** *** ** (0.037) (0.040) (0.035) (0.037) (0.036) (0.037) (0.035) (0.037) (0.036) (0.038) (0.037) (0.037) Insuons *** 0.119*** 0.137*** 0.140*** *** *** ** (0.037) (0.068) (0.024) (0.045) (0.042) (0.073) (0.035) (0.068) (0.043) (0.071) (0.037) (0.064) Ineracon *** *** ** *** (0.043) (0.033) (0.043) (0.040) (0.041) (0.038) Observaons R-squared Number of counres Noe: The models are esmaed by Fxed-Effecs. The dependen varable s he logarhm of real GDP per capa. A consan erm and me effecs have been ncluded bu are no repored for reasons of parsmony. Robus heeroskedasc-conssen sandard errors are repored n parenhess below each coeffcen esmae. *, **, *** denoe sgnfcance a 10, 5 and 1% levels. 21

22 Table 4: Accounng for Endogeney: Sysem GMM esmaon, full sample Esmaor SYS-GMM Spec. (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) Insuonal Proxy Cvl Lberes Polcal Rghs Poly Democracy Index Regme Durably Vanhanen Democracy Governmen Fraconalzaon Governance KOF Socal KOF Polcal KOF Economc Physcal capal Human capal 1.467*** 1.448*** 1.570*** 1.539*** 1.628*** 1.620*** 1.712*** 1.461*** 1.825*** 1.736*** 1.714*** (0.208) (0.207) (0.201) (0.192) (0.188) (0.149) (0.252) (0.218) (0.195) (0.235) (0.213) * 0.185* 0.190** 0.600* 0.295*** 0.178** 0.313*** (0.082) (0.080) (0.076) (0.070) (0.072) (0.096) (0.088) (0.370) (0.095) (0.090) (0.109) Publc *** *** *** *** *** ** Consumpon 0.113*** 0.117*** 0.083*** 0.085*** 0.113*** (0.024) (0.025) (0.023) (0.026) (0.024) (0.017) (0.027) (0.058) (0.031) (0.040) (0.038) Insuons 0.328*** 0.181*** 0.068*** 0.083** 0.177*** 0.197* ** *** (0.093) (0.068) (0.019) (0.033) (0.046) (0.109) (0.036) (0.043) (0.105) (0.090) (0.114) Observaons Number of counres Hansen (pvalue) AB AR(1) (pvalue) AB AR(2) (pvalue) Noe: The models are esmaed by sysem GMM (SYS-GMM). The dependen varable s real GDP per capa growh. Robus heeroskedascconssen sandard errors are repored n parenhess below each coeffcen esmae. The Hansen es evaluaes he valdy of he nsrumen se,.e., ess for over-denfyng resrcons. AR(1) and AR(2) are he Arellano-Bond auocorrelaon ess of frs and second order (he null s no auocorrelaon), respecvely. Also a consan erm, lagged dependen varable and me effecs have been ncluded bu are no repored for reasons of parsmony. *, **, *** denoe sgnfcance a 10, 5 and 1% levels. Table 5: New Polcal Sysem s Measures: Sysem GMM esmaon wh neracon erms, full sample Esmaor SYS-GMM Spec. (1) (2) (3) (4) (5) (6) (7) (8) Governmen Proxy Publc Consumpon Governmen Expendure Governmen Revenue Publc Deb Physcal capal 0.758*** 0.767*** 1.726*** 1.722*** 1.078*** 1.091*** 0.518*** 0.525*** (0.153) (0.153) (0.185) (0.184) (0.173) (0.173) (0.156) (0.155) Human capal 0.260*** 0.252*** 0.272*** 0.260*** 0.274*** 0.266*** (0.085) (0.085) (0.088) (0.087) (0.082) (0.082) (0.089) (0.088) Governmen proxy *** ** *** *** * *** *** *** (0.026) (0.023) (0.017) (0.015) (0.019) (0.018) (0.002) (0.002) G*SAT *** ** *** ** (0.019) (0.011) (0.012) (0.003) G*SDT 0.055*** 0.031*** 0.035*** 0.011*** (0.019) (0.011) (0.011) (0.003) Observaons Number of counres Hansen (p-value) AB AR(1) (p-value) AB AR(2) (p-value) Noe: The models are esmaed by sysem GMM (SYS-GMM). The dependen varable s real GDP per capa growh. SDT and SAT sand for susaned democrac ranson and susaned auocrac ranson for more deals refer o he man ex. Robus heeroskedascconssen sandard errors are repored n parenhess below each coeffcen esmae. The Hansen es evaluaes he valdy of he nsrumen se,.e., ess for over-denfyng resrcons. AR(1) and AR(2) are he Arellano-Bond auocorrelaon ess of frs and second order (he null s no auocorrelaon), respecvely. Also a consan erm, lagged dependen varable and me effecs have been ncluded bu are no repored for reasons of parsmony. *, **, *** denoe sgnfcance a 10, 5 and 1% levels. 22

23 Table 6: Esmaons allowng for heerogeneous echnology parameers bu homogeneous facor loadngs (wh and whou neracon erms) PANEL A Esmaor PMG MG Spec. (1) (2) (3) (4) (5) (6) Sample OECD Emergng Low Income OECD Emergng Low Income Physcal capal 1.656*** 1.246*** 0.805* 1.125*** 0.956* 0.727* (0.135) (0.196) (0.434) (0.255) (0.616) (0.421) Human capal 0.181* 0.124* 0.159* 0.217* 0.179* 0.249** (0.105) (0.072) (0.096) (0.131) (0.081) (0.080) Publc Consumpon ** * ** (0.059) (0.144) (0.199) (0.062) (0.227) (0.213) Poly 0.032** 0.572* 0.457* 0.124* 0.407** (0.010) (0.352) (0.261) (0.081) (0.204) (0.611) Error Correcon -0.76*** -0.64*** -0.71* -0.86*** Publc Consumpon *** * *** (0.044) (0.074) (0.081) (0.078) (0.072) (0.188) Poly 0.660* ** (0.386) (0.574) (0.886) (0.386) (0.418) (0.276) Ineracon * (0.549) (0.648) (0.549) (0.781) (0.801) (0.594) Error Correcon -0.81*** -0.74** -0.57* -0.64*** -0.78* Hausman es for homogeney (pvalue) PANEL B Wh neracon erm Physcal capal 1.303*** 1.001** 0.903* 1.001*** 1.223* 0.628* (0.227) (0.727) (0.627) (0.371) (0.905) (0.402) Human capal 0.174* 0.120* 0.141* 0.139** 0.201* 0.214** (0.124) (0.08) (0.08) (0.060) (0.161) (0.108) Hausman es for homogeney (pvalue) Noe: The models are esmaed by eher PMG or MG esmaors. The dependen varable s he logarhm of real GDP per capa. Hausman es for homogeney: under he null hypohess he dfference n he esmaed coeffcens beween he MG and PMG esmaors, s no sgnfcan and PMG s more effcen. *, **, *** denoe sgnfcance a 10, 5 and 1% levels. Table 7: Esmaons allowng for heerogeneous echnology parameers and facor loadngs (wh and whou neracon erms) Sample OECD Emergng Low Income Esmaor CCEMG AMG CCEMG AMG CCEMG AMG Spec. (1) (2) (3) (4) (5) (6) Physcal capal 0.256*** 0.433*** 0.959*** 0.538*** 0.447** 0.655*** (0.019) (0.115) (0.278) (0.181) (0.186) (0.062) Human capal 0.210*** 0.181** 0.110* * (0.071) (0.909) (0.082) (0.017) (0.027) (0.155) Publc Consumpon * * * (0.101) (0.011) (0.127) (0.109) (0.221) (0.105) Poly 0.105* * 0.090* 0.325* 0.346** (0.068) (0.007) (0.161) (0.064) (0.211) (0.138) Wh neracon erm Physcal capal 0.300*** 0.461*** 0.919*** 0.405* 0.567** 0.731* (0.125) (0.112) (0.273) (0.223) (0.292) (0.402) Human capal 0.233** 0.197* 0.076** 0.038* 0.144* 0.239* (0.101) (0.147) (0.037) (0.021) (0.088) (0.159) Publc Consumpon ** * * (0.134) (0.107) (0.192) (0.451) (0.248) (0.216) Poly 0.148* ** * (0.091) (0.477) (0.039) (0.520) (0.703) (0.930) Ineracon * (0.082) (0.827) (0.153) (0.399) (0.237) (0.089) Noe: The models are esmaed by CCEP or AMG esmaors. The dependen varable s he logarhm of real GDP per capa. *, **, *** denoe sgnfcance a 10, 5 and 1% levels. 23

24 Fgure 1: Growh regressons esmaed coeffcens usng World Bank s Governance Indcaors by counry group (only sascally sgnfcan) Noe: auhor s compuaons. Fgure 2: Esmaed coeffcens of governmen consumpon n regressons usng dfferen nsuonal proxes, over me (only sascally sgnfcan) Cvl Lberes Polcal Rghs Poly Democracy Durably KOF legal Kof Polcal KOF Economc Noe: auhor s compuaons. 24

25 Fgure 3: Esmaed coeffcens of dfferen nsuonal proxes, over me (only sascally sgnfcan) Noe: auhor s compuaons. Fgure 4: Esmaed coeffcens of he neracon erms of dfferen governmen sze proxes wh new polcal sysem s measures, by counry group Noe: auhor s compuaons. 25