Growth, inequality and poverty : a robust relationship?

Empi ical Economics (2022) 63:725–791 h ps://doi.o g/10.1007/s00181-021-02152-x G ow h, inequali y and po e y: a obus ela ionship? Gus a o A. Ma e o1,2 ·Luis Se én3 Recei ed: 12 Oc obe 2020 / Accep ed: 30 Sep embe 2021 / Published online: 23 No embe 2021 © The Au ho (s) 2021 Abs ac The consequences o po e y and inequali y o g ow h ha e long p eoccupied academics and policy-make s. This pape e isi s he inequali y-g ow h and po e y- g ow h links. Using a panel o 158 coun ies be ween 1960 and 2010, we ind ha he co ela ion o g ow h wi h po e y is consis en ly nega i e: A 10 p.p. dec ease in he headcoun po e y a e is associa ed wi h a subsequen inc ease in pe capi a GDP be ween 0.5 and 1.2% pe yea . In con as , he co ela ion o g ow h wi h inequali y is empi ically agile—i can be posi i e o nega i e, depending on he empi ical speci i- ca ion and econome ic app oach employed. Howe e , he indi ec e ec o inequali y on g ow h h ough i s co ela ion wi h po e y is obus ly nega i e. Close inspec ion shows ha hese esul s a e d i en by he sample obse a ions ea u ing high po e y a es. Keywo ds G ow h ·Inequali y ·Po e y ·Indi ec impac s JEL Classi ica ion O40 ·O11 ·O15 ·E25 1 In oduc ion Wha is hee ec o po e yonagg ega eincomeg ow h?And hee ec o inequali y? Academicsandpolicy-make sha elongbeenconce nedwi h heseques ions.Bu hey ha e ypically been explo ed as sepa a e issues. Ye p ope ly answe ing hem equi es BGus a o A. Ma e o [email p o ec ed] Luis Se én [email p o ec ed] 1Depa amen o de Economía, Con abilidad y Finanzas, CEDESOG, Uni e sidad de La Laguna, San C is óbal de La Laguna, Spain 2EQUALITAS, Mad id, Spain 3CEMFI, Mad id, Spain 123 726 G. A. Ma e o, L. Se én aking hem up join ly, because po e y and inequali y a e in e ela ed ea u es o he same income dis ibu ion (Bou guignon 2004). This pape a emp s o ill ha gap by p o iding an empi ical explo a ion o he g ow h e ec s o bo h po e y and inequali y and, in pa icula , o hei espec- i e obus ness. The e ec s o po e y ha e been analyzed by nume ous heo e ical pape s highligh ing a a ie y o mechanisms h ough which po e y may become sel - pe pe ua ing. Bu empi ical wo k has been mo e limi ed and la gely inconclusi e. Indeed, a basic implica ion o he heo e ical models o po e y aps—namely, ha coun ies su e ing om highe le els o po e y should g ow less apidly han com- pa able coun ies wi h lowe po e y—has been la gely o e looked. This is he key hypo hesis pu sued in his pape . I can be iewed as a weak e sion o he po e y ap hypo hesis, in ha o suppo i we do no need o ind e idence o mul iple equilib ia o income s agna ion, bu jus empi ical p oo ha , o he hings equal, po e y ends o hold back g ow h. In con as , he e ec s o inequali y ha e a ac ed massi e empi ical li e a u e, albei wi h sha ply con lic ing esul s. The p esen pape adds o exis ing wo k by highligh ing a no el angle, namely he indi ec e ec o inequali y on g ow h acc uing h ough he impac o inequali y on po e y: gi en he po e y line and he o e all popula ion’s mean income, an inc ease in inequali y will ypically aise po e y, by pushing mo e indi iduals below he po e y line.1I po e y a ec s g ow h, so will inequali y h ough his indi ec channel—in addi ion o any di ec e ec s ha inequal- i y migh exe on g ow h. To assess he espec i e g ow h impac s o po e y and inequali y, we es ima e a educed- o m g ow h equa ion wi h inequali y and po e y added sepa a ely and join ly o an o he wise s anda d se o g ow h de e minan s (educa ional a ainmen , in es men p ices, go e nmen size, deg ee o openness, public in as uc u es, e c.). Fo he es ima ion, we assemble a la ge panel da a se o non-o e lapping i e-yea obse a ions comp ising 158 coun ies o e he pe iod 1960–2010. The sample is hea ily unbalanced, and i s size exceeds by a ha ound in ea lie s udies o he po e y-g ow h link. Ou econome ic app oach is based on GMM es ima ion employing in e nal ins u- men s (A ellano and Bo e 1995; Blundell and Bond 1998; Roodman 2009). In ou se ing, he choice o his app oach is dic a ed by he sho ime dimension and la ge c oss-sec ional dimension o ou panel da ase —which makes panel ime-se ies me h- ods unsui able—and by he po en ial endogenei y o he eg esso s—which demands an ins umen al a iable app oach. These issues a ec also much o he empi ical li e - a u e on he links be ween po e y, inequali y and g ow h, which—like ou pape —has o con end wi h he po en ial p oblem o wo-way causali y be ween he a iables a he co e o he analysis. In his con ex , GMM ep esen s a na u al me hodological choice, which we also sha e wi h much o he ela ed empi ical li e a u e.2Mo eo e , his common empi ical 1The consequences o inequali y o po e y a e highligh ed o example by Bou guignon (2003,2004)o Ra allion (2005). Ma e o and Se én (2018) p o ide nume ical simula ions illus a ing he magni ude o he e ec o inequali y on po e y, o gi en a e age income. 2Empi ical analyses o he links be ween agg ega e g ow h, po e y and inequali y commonly use an ins umen al a iable app oach. A ew pape s ea u e ex e nal ins umen s—e.g., B ueckne e al. (2015), 123 G ow h, inequali y and po e y: a obus ela ionship? 727 me hodologyalsomakesou pape mo eeasilycompa ablewi hexis ingwo k.Finally, while ou use o GMM o g ow h empi ics is no no el, ou pape is among he i s o examine igo ously, in a sys em GMM se ing, he po en ial p oblem o weak ins umen s plaguing much o he empi ical g ow h li e a u e, as i s aised by K aay (2015) in he con ex o he empi ical ela ionship be ween inequali y and g ow h. Ou main inding is ha po e y has a obus nega i e and signi ican e ec on g ow h. As o inequali y, we ind ha he sign and signi icance o i s di ec e ec on g ow h a e agile. Howe e , i s indi ec e ec ( h ough po e y) is obus ly nega i e. Fu he inspec ion e eals he p esence o nonlinea i ies, in ha hese esul s a e d i en by he sample obse a ions ea u ing high po e y: when po e y is low, i s impac on g ow h is no signi ican , and he indi ec e ec o inequali y on g ow h is he e o e absen . We each a simila conclusion when we le he g ow h impac o po e y di e be ween de eloped and de eloping coun ies: I is nega i e and signi ican o he la e , bu no o he o me . Ou esul s su i e a ba e y o obus ness checks, including he use o al e na i e se s o ins umen s and speci ica ions in he GMM es ima ion, di e en po e y lines and po e y measu es, al e na i e po e y da a, nonlinea and nonpa ame ic spec- i ica ions, o he use o al e na i e se s o con ol a iables. We also ind ha ou p e e ed GMM speci ica ion can add ess in a sa is ac o y manne he endogenei y, unde -iden i ica ion and weak ins umen s p oblems o en encoun e ed in mac oeco- nomic applica ions o dynamic panel models (Bazzi and Clemens 2013). Ou pape is embedded in an ex ensi e li e a u e ( ecen ly su eyed by Ce a e al. 2021a) analyzing he mul idi ec ional links among g ow h, inequali y and po e y. Th ee s ands a e especially ele an in ou con ex . They, espec i ely, ocus on he impac o po e yong ow h, heimpac o inequali yong ow h,and hecon ibu iono inequali y and income g ow h o po e y. We p o ide a b ie e iew o hese li e a u e wo ks in he nex sec ion. The es o he pape is s uc u ed as ollows. As jus no ed, Sec . 2is de o ed o a selec i e summa y o he li e a u e on he g ow h-inequali y-po e y nexus. In Sec . 3, we desc ibe he da a and we lay ou he empi ical s a egy o es o he e ec s o po e y and inequali y on g ow h. In Sec . 4, we epo he main empi ical esul s o ou baseline speci ica ion. Sec ion 5 epo s ex ensi e obus ness checks on ou empi ical esul s. Sec ion 6analyzes how he links o po e y and inequali y wi h g ow hmigh depend on he p e ailing deg eeso po e y and/o inequali yand gauges he di ec and indi ec e ec s o inequali y on g ow h. Finally, Sec . 7concludes. 2 The g ow h-inequali y-po e y nexus: a e iew The seminal wo k o Kuzne s (1955) is he s a ing poin o an ex ensi e li e a u e analyzing he g ow h-inequali y-po e y nexus (see Bou guignon 2004, and he ecen su eys by Ce a e al. 2021a,b). Ou pape ela es o se e al s ands o his li e a u e. Foo no e 2 con inued assessing he e ec o GDP g ow h on inequali y—bu GMM using in e nal ins umen s (gi en by sui ably lagged and ans o med eg esso s) is much mo e commonly used: o example, by Pa idge (1997), Fo bes (2000), Panizza (2002), o Be g e al. (2018), all o which a e conce ned wi h he opposi e di ec ion o causali y, om inequali y o g ow h. 123 728 G. A. Ma e o, L. Se én Fi s , a long-s anding heo e ical li e a u e has s udied a a ie y o mechanisms h ough which po e y may de e economic g ow h. I s a gumen s a e mos ly based on he exis ence o po e y aps, i.e., mechanisms h ough which po e y p e en s a signi ican sha eo hepopula ion omhelpingigni e heg ow hengine(Aza iadisand S achu ski2005;Bowlese al.2006;Haide e al. 2018). Unde app op ia e condi ions, hose mechanisms may lead o mul iple equilib ia and make he nega i e impac o po e y on g ow h sel - ein o cing. In gene al, he mechanisms highligh ed in he li e a u e ope a e by educing he incen i es and/o abili ies o he poo o unde ake isky en ep eneu ial ac i i ies, and/o o accumula e physical and human capi al. A p ominen mechanism in ol es ‘ h eshold e ec s’ (Aza iadis and D azen 1990), esul ing, o example, om indi isibili ies o inc easing e u ns o scale.3Fo exam- ple, i po e y is coupled wi h c edi cons ain s, he esul is ha below a ce ain le el o income o weal h economic agen s may be oo poo o a o d he in es men s (in human o physical capi al) o he echnologies necessa y o aise hei income (Galo and Zei a 1993; Bane jee and Newman 1993). Malnu i ion p o ides ano he example. In de eloping coun ies, po e y is associa ed wi h high a es o malnu i ion (Das- gup a and Ray 1986), which impac s cogni i e abili ies and school absen eeism and is ansmi ed o he child en’s capaci y o lea n. The esul ing educa ional inequali y is also g ow h-de e ing (Galo and Moa 2004). Ins i u ional a angemen s ha place economic oppo uni ies beyond he each o he poo can likewise esul in educed income g ow h (Mookhe jee and Ray 2002; Enge man and Sokolo 2006). Ano he po e y-pe pe ua ing mechanism is ela ed o isk a e sion (Bane jee 2000): Because poo e indi iduals a e ypically mo e isk a e se, in he absence o well- unc ioning insu ance and c edi ma ke s, hey will skip p o i able in es men oppo uni ies ha hey deem oo isky.4Po e y can also al e he decision-making p ocess o indi iduals owa d less g ow h-enhancing ac i i ies. Fo ins ance, he poo de o e a signi ican ac ion o hei income o sa is ying basic needs (Shah e al. 2012) and o “ emp a ion” goods (Bane jee and Mullaina han 2010) and educe he esou ces de o ed o educa ion, heal h and in es men . Poo indi iduals show also lowe aspi a ions, as hey an icipa e ha hei cu en s a us will impede hei u u e success (La Fe a a 2019). In spi e o he di e si y o hese analy ical models, e idence on hei empi ical el- e ance emains la gely inconclusi e. A ew pape s (see Du lau 2006, o a e iew) ha e sea ched o a ious empi ical egula i ies consis en wi h hose models, such as agg ega e non-con exi ies (Aza iadis and S achu ski 2005) and con e gence clubs (Quah 1993). A b oade empi ical e iew o di e en mechanisms ad anced in he li e a u e inds li le e idence ha hey may be a wo k, excep pe haps in emo e o disad an aged a eas (K aay and McKenzie 2014). Mo e ecen ly, la ge-scale andom- ized e alua ions, such as he one de eloped by Bandie a e al. (2017) in Bangladesh, 3Po e y aps a ising om h eshold e ec s ha e o en been o e ed as a a ionale o a ‘big push’ app oach o policy. In pa icula , when la ge aid p og ams a e coo dina ed in a mul i- ace ed way, a ‘big push’ can be e ec i e o enginee g ow h akeo s (Bane jee e al., 2015). Howe e , in a c oss-coun y da ase , Eas e ly (2006) inds ha akeo s a e a e and, in gene al, hey a e no associa ed wi h ‘big push’ s a egies. 4The a gumen ha isk a e sion leads o unde in es men goes back o S igli z (1969). See also Ageno and Aizenman (2011), who a gue ha aid ola ili y could induce po e y aps in poo coun ies h ough a simila mechanism. 123 G ow h, inequali y and po e y: a obus ela ionship? 729 yield s ong e idence ha he poo ace impe ec ions in capi al ma ke s ha keep hem in a low asse -low employmen po e y ap. Somewha su p isingly, jus a ew pape s ha e aken up he undamen al agg ega e implica ion o he po e y ap li e a u e— ha , ce e is pa ibus, coun ies wi h highe po e y should g ow mo e slowly. The lis is limi ed o ou wo king pape e sion, Ma e o and Se én (2018), plus López and Se én (2015) and Ra allion (2012), all o which conclude ha po e y is g ow h-de e ing5;Eas e ly(2006) shows a non- signi ican impac o po e y on g ow h. The second s and o li e a u e o which ou pape is ela ed is conce ned wi h he impac o inequali y on g ow h. I includes a la ge numbe o empi ical con ibu ions eaching con lic ing conclusions; o o e iews, see Voi cho sky (2011), Be g e al. (2018), and Ce a e al. (2021a). Fo example, Alesina and Rod ik (1994) and Pe o i (1996) ound a nega i e ela ionship be ween inequali y and g ow h in c oss sec ion da a, bu subsequen ly, Li and Zou (1998) and Fo bes (2000) ob ained he opposi e esul using panel da a. Ba o (2000) ound ha inequali y migh a ec g ow h in di e en di ec ions depending on he coun y’s le el o income, while Panizza (2002) ound ha esul s migh depend on he model speci ica ion and he quali y and ype o da a (see also Deininge and Squi e 1998). In u n, Bane jee and Du lo (2003) concluded ha he esponse o g ow h o inequali y changes has an in e ed U-shape. Themul iplici yo ac o sa ec ingbo hinequali yandg ow hmigh explains hese con adic o y esul s. Fo example, ising inequali y could be he esul o g ow h- enhancing echnological change whose e u ns a e cap u ed by alen ed indi iduals a he op o he dis ibu ion (Goldin and Ka z 2008). In con as , i en -seeking is he undamen al o ce behind g owing incomes o he ich, he inc ease in inequali y could come along wi h declining g ow h (S igli z 2012). In his line o enqui y, Galo and Moa (2004) a gue ha he eplacemen o physical capi al accumula ion by human capi al accumula ion as a p ime engine o economic g ow h has changed he quali a i e impac o inequali y on g ow h. Ma e o and Rod íguez (2013) emphasize ha he sign o he e ec o inequali y on g ow h depends on he ype o inequali y conside ed (i.e., inequali y o oppo uni y o o e o ). Voi cho sky (2005) and, mo e ecen ly, an de Weide and Milano ic (2018) a gue ha he e ec o inequali y is nega i e o he income g ow h o he poo bu posi i e o he income g ow h o he ich—i.e., inequali y ends o be sel - ein o cing. The e ec s o inequali y on g ow h migh also depend on he sec o al s uc u e o he economy (E man and e Kaa 2019) and on he deg ee o in e gene a ional mobili y (Aiya and Ebeke 2020).6 In gene al, di e en mechanisms a ec ing g ow h in opposi e di ec ions h ough di e en channels ac all simul aneously, leading o con lic ing in e ences. In he empi ical li e a u e, an eme ging consensus iew is ha he long- un e ec o inequal- i yong ow hissigni ican lynega i e,andonlywhenlookinga ela i elysho pe iods 5Eas e ly (2006) in es iga es (and ejec s) a mo e ex eme hypo hesis, namely ha high po e y coun ies should show no g ow h. 6E man and e Kaa (2019) show ha highe inequali y inc eases g ow h in physical capi al-in ensi e indus ies, while i ha ms g ow in indus ies using skilled labo in ensi ely. 123 730 G. A. Ma e o, L. Se én o ime, he ela ionship may u n posi i e (Hal e e al. 2014; B ueckne e al. 2015; Be g e al. 2018; B ueckne and Lede man 2018).7 A hi d s and o he li e a u e explo es he links be ween g ow h and inequali y, on he one hand, and po e y, on he o he . The bulk o his li e a u e, which is qui e ex ensi e (Ce a e al. 2021a), ocuses on he po e y- educing e ec o g ow h and he ac o s ha shape i (Dolla and K aay 2002; Bou guignon 2003; Ra allion 2004). This angle o he po e y-g ow h link is he opposi e o ha pu sued in his pape . Empi ically, he e is ample consensus ha g ow h educes po e y—i.e., i is “good o he poo .” Dolla and K aay (2002), and he subsequen upda es using al e na i e da abases and empi ical app oaches (K aay 2006, Dolla e al. 2016) ind ha he income o he poo es deciles a ies in he same p opo ion as a e age income, hence os e ing agg ega e g ow h is p o-poo (see also Fe ei a e al. 2010, o Loayza and Radda z 2010). Recen wo k con i ms his esul (Fosu 2017; Bluhm e al. 2018; Be gs om 2020). Fo example, Be gs om (2020) inds ha , in a la ge c oss-coun y sample, 90% o he a ia ion in po e y is explained by a ia ion in pe capi a GDP. Howe e , he eason is ha he sample a ia ion in pe capi a income is much la ge han ha o inequali y;indeed,inmos o he sample coun ies, hees ima edinequali y elas ici y o po e y exceeds he income elas ici y o po e y—which sugges s ha declines in inequali y o e a la ge po en ial (as ye un ealized) o educe po e y a es. Compa a i ely, he li e a u e has paid less a en ion o he impac o inequali y on po e y (Bou guignon 2003; Ra allion 2005; Fe ei a e al. 2010; Kalwij and Ve - schoo 2007).Thisis p ecisely hemechanismbehind heindi ec inequali y- o-g ow h channel analyzed in his pape , and no co e ed in ea lie li e a u e. Mo e ecen ly, Seh awa and Gi i (2018), he a o emen ioned Be gs om (2020) and Lakne e al. (2020) ind e idence suppo ing he ole o declining inequali y o po e y educ ion. 3 G ow h, inequali y and po e y: da a and empi ical implemen a ion We u n o he desc ip ion o ou empi ical s a egy. Fi s we desc ibe he da a and hen he econome ic app oach employed in he es ima ion. 3.1 Da a Sinceou ocusisno oncyclicalg ow h luc ua ions,we ollow heempi icalli e a u e on inequali y and g ow h and cons uc a panel da a se o non-o e lapping 5-yea obse a ions on he h ee a iables o in e es : inequali y, g ow h and po e y. We ocus on he 1960–2010 pe iod, as done by he ecen empi ical li e a u e on inequali y and g ow h. G ow h is measu ed as he log di e ence o eal pe capi a income o e he en i e 5-yea in e al, while po e y and inequali y a e measu ed a he beginning 7Amo elimi edli e a u e hasexamined heinequali y-g ow hlink om heopposi e pe spec i e,assessing how income g ow h a ec s inequali y. I s esul s a e mos ly inconclusi e, howe e . Fo ins ance, while B ueckne e al. (2015) and Blau (2018) ind ha GDP g ow h educes inequali y, K usell e al. (2000)and Aghion e al. (2019) each he opposi e conclusion. 123 G ow h, inequali y and po e y: a obus ela ionship? 731 o he in e al. This means we only need o collec po e y and inequali y da a up o 2005. We use he Gini index o measu e inequali y and ake he UN-WIID2 (2008) da abase as ou p ima y sou ce o da a on income inequali y. I includes 5313 su eys o 154 coun ies om 1950 o 2006. We comple e he WIID2 da a wi h in o ma ion omPo calNe ,whichadds ano he 122coun y-yea (16coun ies)obse a ionso e he 1960–2010 pe iod. In a numbe o ins ances, he e a e mul iple su eys e e ing o he same coun y-yea , bu hey o e di e en co e age o use di e en concep s o income. We es ic ou sample o Gini indexes based on na ionally ep esen a i e su eys. Mo eo e , da a a e some imes based on income and o he imes on expendi- u e igu es; income is ne o ans e s and axes in some cases and no in o he s; he uni o analysis may be he indi idual o he household, e c. To co ec a leas in pa o his he e ogenei y, we adjus he o iginal Gini da a ollowing Dolla and K aay (2002).8 Fo economic g ow h, we use na ional accoun s pu chasing-powe -pa i y (PPP)- adjus ed pe capi a GDP da a om he Penn Wo ld Tables 7.1, he same sou ce used by Be g e al. (2018) and many o he s udies o inequali y and g ow h, which acili a es compa abili y wi h hem. Sala-i-Ma in (2006) and Dolla and K aay (2002), among many o he s, emphasize he ad an ages o using pe capi a GDP ins ead o he mean le el o income ob ained di ec ly om household su eys. The su ey mean usually does no ma ch pe capi a income om he na ional accoun s, because o di e ences in concep s and me hodology, inconsis en da a collec ion me hods, mis epo ing, e c. Addi ionally, o manyo hecoun y-yea obse a ions o whichweha ein o ma ion on inequali y, we do no ha e ma ching in o ma ion on mean income om he same sou ce, which hampe s he cons uc ion o a la ge panel da ase . In con as , na ional accoun s da a a e epo ed yea ly o all coun ies, using a homogenous me hodology, which, in addi ion, allows us o compa e ou empi ical esul s wi h hose o he ample mac oeconomic li e a u e on income inequali y and g ow h. Rega ding po e y da a, we ollow he s a egy p oposed by Dolla and K aay (2002), López and Se én (2015), Sala-i-Ma in (2006) and Pinko skiy and Sala-i- Ma in (2013). These au ho s poin ou ha combining po e y and income g ow h da a om household su eys and na ional accoun s may lead o misleading conclu- sions, because o he inconsis encies be ween he wo sou ces jus no ed. To a oid his p oblem, hey use PWT da a o cons uc bo h income g ow h and po e y mea- su es, wi h he la e compu ed assuming ha household income ollows a logno mal dis ibu ion. Thus, we cons uc a se o po e y measu es ( he headcoun a io P0, he po e y gap P1 and he squa ed po e y gap P2) using a logno mal app oxima ion on he basis o he obse ed pe capi a GDP le els and Gini coe icien s.9We also 8Speci ically, we pool he sample and eg ess he Gini coe icien on a cons an , egional dummies and dummy a iables indica ing whe he he su ey is s a ed in e ms o g oss income o consump ion ( he omi ed ca ego y is income ne o axes and ans e s). We hen sub ac he es ima ed mean di e ence be ween hese wo al e na i es and he omi ed ca ego y o a i e a a se o Gini indices ha no ionally co espond o hedis ibu iono incomene o axesand ans e s.The esul so heseadjus men eg essions a e a ailable upon eques , bu hey show simila conclusions as in Dolla and K aay (2002). 9The UN-WIID2 Gini index is no always a ailable o he i s yea o each 5-yea in e al. In such cases, we alloca e he a ailable obse a ion(s) o he closes s a ing yea o a 5-yea in e al, wi h a limi o 2 yea s 123 732 G. A. Ma e o, L. Se én expe imen wi h al e na i e, widely used po e y lines: US$ 1.25, US$ 2 and US$ 4 pe pe son pe day, in 2005 PPP US$ (see Appendix 1 o de ails). This app oach allows a conside able inc ease in sample size. Despi e he p og ess made in ecen yea s, mainly h ough he Po calNe p ojec , su ey-based po e y da a a e s ill ela i ely sca ce, a leas in compa ison wi h he size o he s anda d c oss- coun y ime-se ies g ow h da ase . Using he logno mal app oxima ion, we assemble 746obse a ionsonpo e yo e non-o e lapping5-yea in e als,co e ing156coun- ies be ween 1960 and 2005 (an a e age o almos i e obse a ions pe coun y).10 In con as , using he Janua y 2020 e sion o Po calNe o e he same 1960–2005 ime span, we can cons uc a da ase o 383 po e y obse a ions o e non-o e lapping 5-yea in e als o 144 coun ies, oughly hal he size o ou sample—i.e., an a e age o less han 3 obse a ions pe coun y, wi h da a o he as majo i y o coun ies s a ing in 1990 o la e .11 As a as we a e awa e, ou s is he la ges sample used o da e o s udy he impac o po e y on g ow h. I exceeds by a he samples used by he wo ea lie pape s analyz- ing he po e y-g ow h nexus in a panel eg ession se ing: López and Se én (2015) assemble a sample comp ising 325 obse a ions om 85 coun ies o e 1960–2000, while Ra allion (2012) uses unbalanced panel da a om Po calNe co e ing up o 97 de eloping coun ies o e a sho e ime span, 1981–2005. Table 1p esen s summa y s a is ics on annual g ow h, mean income, inequali y and po e y o he common sample o hese a iables in he unbalanced 1960–2010 panel. The able shows he wide ange o pe capi a income le els (exp essed in 2005 US dolla s in PPP e ms) in he sample— om jus o e $200 ( he Democ a ic Republic o Congo in he mid-2000s) o abou $73,000 (Luxembou g in 2005). The median obse a ion co esponds o B azil in he mid-1970s, wi h pe capi a income abou $5500. The o e all sample mean is abou $9800, much la ge han he median, which e lec s a wo ld income dis ibu ion skewed o he igh . Rega dinginequali y,bo h hemedianand hemeano heGinicoe icien equal0.4, which ma ches he alues ound o he U.S. (in 2000), Bu kina Faso (in 1995), Tu key (in 2010) o Singapo e (in 1970). The maximum alue (abo e 0.74) co esponds o Foo no e 9 con inued o di e ence. When mo e han one obse a ion is a ailable wi hin he 2-yea limi , we ake he a e age. Because o he s ong ine ia o inequali y and po e y ime se ies, using a 1-yea limi ins ead o 2 yea s, o no using means, yields e y simila esul s (Dolla and K aay 2002). 10 Ou da a comp ise 121 da a poin s co esponding o 32 low-income coun ies, 180 o 41 lowe -middle income coun ies, 240 o 44 uppe -middle income, 57 o 11 high-income non-OECD, and 206 o 30 high- income OCDE coun ies. The sample includes 18 obse a ions (2 coun ies) om No h Ame ica, 248 (48 coun ies) om Eu ope and Cen al Asia, 159 (28 coun ies) om La in Ame ican and he Ca ibbean, 53 (12 coun ies) om Middle Eas and No h A ica, 144 (40 coun ies) om Sub-Saha an A ica, 56 (9 coun ies) om Sou h Asia and 126 (19 coun ies) om Eas Asia and he Paci ic. 11 This sample size would be oo small o many o ou exe cises, and hus o he obus ness es s using Po calNe da a epo ed in Sec ion V below, we eso o he in e pola ed Po calNe se ies, which allows inc easing he sample size o 556 obse a ions. These in e pola ed in o ma ion s a in 1981 and a e epo ed e e y h ee yea s. Thus, o cons uc a non-o e lapping 5-yea panel da a simila o he one used in ou baseline speci ica ion and ma ch he iming o po e y da a wi h ha o he o he a iables (g ow h and o he con ols), we use a “closes ” c i e ion o ake he a e age i wo po e y obse a ions a e one yea abo e and one below he assigned yea . We should also no e ha he cu en Po calNe se ies uses a po e y line o 1.90 2011 US$, which eplaces i s p e ious line o 1.25 2005 US$ (see Fe ei a e al. 2016, o mo e de ails). 123 G ow h, inequali y and po e y: a obus ela ionship? 733 Table 1 G ow h, inequali y and po e y da a: summa y s a is ics Median Mean S d P10 P90 Min Max GDP pe capi a g ow h 0.025 0.025 0.030 −0.012 0.061 −0.086 0.201 Real pe capi a income 5651.1 9792.5 10,462.5 816.4 26,053.7 207.5 73,243.0 Gini coe - icien 0.394 0.402 0.100 0.280 0.543 0.157 0.742 P0 (US$ 1.25) 0.005 0.096 0.177 0.000 0.364 0.000 0.906 P0 (US$ 2) 0.023 0.162 0.247 0.000 0.610 0.000 0.969 P0 (US$ 4) 0.130 0.287 0.336 0.000 0.881 0.000 0.999 P1 (US$ 1.25) 0.001 0.040 0.086 0.000 0.146 0.000 0.602 P1 (US$ 2) 0.006 0.073 0.132 0.000 0.270 0.000 0.722 P1 (US$ 4) 0.038 0.151 0.212 0.000 0.523 0.000 0.855 P2 (US$ 1.25) 0.000 0.023 0.056 0.000 0.077 0.000 0.497 P2 (US$ 2) 0.002 0.044 0.089 0.000 0.158 0.000 0.594 P2 (US$ 4) 0.016 0.100 0.156 0.000 0.356 0.000 0.750 Headcoun po e y a e (P0); po e y gap (P1); squa ed po e y gap (P2); al e na i e po e y lines: US$ 1.25, US$ 2 and US$ 4, pe pe son pe day (2005 PPP). Po e y is ob ained om a logno mal app oxima ion on he basis o he obse ed pe capi a GDP (PWT 7.1) le els and Gini coe icien s (UNU-WIDER 2008). See Appendix 1 o de ails Zimbabwe in 1995, and he minimum (below 0.16) co esponds o Bulga ia in 1975. A ound 80% o he obse a ions all in he ange be ween 0.28, a alue ound among Wes e n Eu opean coun ies, and 0.54, a alue ound among La in Ame ican and Sub-Saha an A ican coun ies. Po e y ises by cons uc ion wi h he po e y line and declines as he po e y mea- su e changes om P0 o P2 (i.e., as one conside s mo e bo om-sensi i e measu es). Fo ou logno mal po e y es ima es, he able shows ha median headcoun po e y P0 is 0.6% using US$ 1.25 pe day as po e y line, bu i aises o 2.3% wi h a US$ 2 po e y line, and o 13% wi h US$ 4. Likewise, he median P1 anges om less han 0.1% o US$ 1.25 o abou 4% o US$ 4, while he median P2 anges om less han 0.1% o US$ 1.25 o almos 2% o US$ 4. Al hough he mean and he median o hese po e y measu es a e ela i ely small, he he e ogenei y in he sample is qui e high, since he anges o he a ious po e y measu es un om a minimum o ze o ( e lec ing he p esence o high-income coun ies in he sample) o a maximum whose 123 740 G. A. Ma e o, L. Se én Table 2 G ow h, po e y and inequali y: panel OLS es ima es M1. Skele on model M2. Ex ended wi h educa ion and in . p ices M3. Ex ended wi h policy a iables M4. Ex ended wi h policy and in as uc u es P0, lag − 0.0450*** (−5.70) − 0.0440*** (−5.71) − 0.0328*** (−3.88) − 0.0334*** (−4.02) − 0.0382*** (−4.58) − 0.0380*** (−4.70) − 0.0424*** (−3.84) −0.0430*** (−4.03) Gini, lag − 0.0415*** (− 3.63) − 0.0393*** (−3.50) −0.0252** (−2.12) −0.0266** (−2.31) − 0.0399*** (−3.35) − 0.0396*** (−3.36) −0.0354** (−2.44) −0.0366*** (−2.60) log y,lag − 0.00781*** (−5.43) −0.00143 (− 1.61) − 0.00873*** (−5.88) − 0.00892*** (−4.80) − 0.00381*** (−3.08) − 0.00952*** (−5.05) − 0.00803*** (−5.43) − 0.00303*** (−3.20) − 0.00920*** (−5.97) − 0.0213*** (−6.13) − 0.0143*** (−4.18) −0.0215*** (−6.31) In . de la o , lag − 0.00482** (−2.23) − 0.00629** (−2.20) −0.00453* (−1.91) Female educ., lag −0.00299 (−1.16) −0.00300 (−1.09) −0.00176 (−0.66) 0.00560*** (3.79) 0.00370** (2.44) 0.00509*** (3.48) Male educ., lag 0.00747*** (2.89) 0.00671** (2.40) 0.00567** (2.13) In la ion −0.00728 (−1.41) −0.00409 (−0.84) −0.00728 (−1.41) − 0.0232*** (−3.66) −0.0165** (−2.58) −0.0217*** (−3.31) T ade openness (log) 0.0113*** (4.25) 0.0136*** (5.05) 0.0115*** (4.34) Go . size (log) −0.00110 (−0.40) −0.00191 (−0.67) −0.00135 (−0.48) In as uc u e, lag 0.00847*** (2.94) 0.00830*** (2.78) 0.00754*** (2.70) 123 G ow h, inequali y and po e y: a obus ela ionship? 741 Table 2 (con inued) M1. Skele on model M2. Ex ended wi h educa ion and in . p ices M3. Ex ended wi h policy a iables M4. Ex ended wi h policy and in as uc u es Num. obs 745 745 745 676 676 676 656 656 656 477 477 477 R2-adjus ed 0.096 0.072 0.112 0.124 0.108 0.130 0.120 0.107 0.135 0.149 0.125 0.161 Unbalanced panel wi h da a a 5-yea in e als o e 1960–2010. The dependen a iable is he annual g ow h a e o pe capi a GDP. The explana o y a iables a e eal pe capi a GDP (in logs), he headcoun po e y a e (P0) using US$ 2 as po e y line, he Gini coe icien , and al e na i e se s o addi ional con ols ha a y ac oss models M1 (skele on model), M2 (educa ion and in es men p ices), M3 (policy a iables) and M4 (policy a iables and in as uc u es). Explana o y a iables a e all lagged one pe iod (5 yea s), wi h he excep ion o he policy a iables in models M3 and M4, which a e aken as con empo aneous 5-yea a e ages. A cons an e m and ime dummies a e included in all models. Robus s a is ics in pa en heses: ***deno es signi icance a 1%, **a 5%, *a 10% 123 742 G. A. Ma e o, L. Se én Table 3 G ow h, po e y and inequali y: wi hin-g oup es ima es M1. Skele on model M2. Ex ended wi h educa ion and in . p ices M3. Ex ended wi h policy a iables M4. Ex ended wi h policy and in as uc u es P0, lag − 0.0665*** (−3.79) − 0.0764*** (−4.45) − 0.0664*** (−3.89) − 0.0792*** (−4.65) − 0.0716*** (−3.86) − 0.0861*** (−4.53) − 0.0514** (−2.46) − 0.0633*** (−2.82) Gini, lag 0.0378 (1.27) 0.0643** (2.27) 0.0454 (1.46) 0.0742** (2.46) 0.0576** (2.00) 0.0865*** (2.98) 0.0457 (1.48) 0.0641** (2.05) log y,lag − 0.0409*** (−6.28) − 0.0303*** (−4.91) − 0.0429*** (−6.18) − 0.0394*** (−5.51) − 0.0256*** (−4.02) − 0.0426*** (−5.86) − 0.0667*** (−8.31) − 0.0557*** (−6.62) − 0.0709*** (−8.38) − 0.0643*** (−8.11) − 0.0580*** (−8.04) − 0.0674*** (−8.18) In . de la o , lag − 0.00916** (−2.33) − 0.0116** (−2.33) − 0.00899** (−2.31) Female educ., lag −0.00122 (−0.15) −0.0124 (−1.63) −0.00196 (−0.26) 0.00471* (1.79) 0.00261 (1.04) 0.00603** (2.39) Male educ., lag 0.00637 (0.75) 0.0148* (1.85) 0.00935 (1.15) In la ion − 0.0224*** (−3.87) − 0.0221*** (−3.87) − 0.0219*** (−3.72) − 0.0368*** (−4.93) − 0.0369*** (−5.45) − 0.0366*** (−5.10) T ade openness (log) 0.0258*** (3.50) 0.0312*** (3.57) 0.0238*** (3.49) Go . size (log) − 0.0237*** (−2.94) − 0.0196** (−2.57) − 0.0251*** (−3.26) In as uc u e, lag 0.0177*** (3.75) 0.0236*** (5.90) 0.0170*** (3.44) Num. obs 745 745 745 676 676 676 656 656 656 477 477 477 123 G ow h, inequali y and po e y: a obus ela ionship? 743 Table 3 (con inued) M1. Skele on model M2. Ex ended wi h educa ion and in . p ices M3. Ex ended wi h policy a iables M4. Ex ended wi h policy and in as uc u es R2-adjus ed 0.202 0.171 0.216 0.218 0.192 0.236 0.325 0.302 0.348 0.336 0.329 0.350 Num. coun ies 156 156 156 131 131 131 147 147 147 88 88 88 See no e in Table 2 123 744 G. A. Ma e o, L. Se én (inM4)ca yposi i eandsigni ican coe icien s(Calde óne al.2015).Incon as , he e ec so maleand emaleseconda yeduca iondependonmodelspeci ica ion.Female educa ion ca ies a posi i e and signi ican coe icien in M4, bu u ns insigni ican in M2, while he coe icien o male educa ion is gene ally posi i e. Simila ly, among he policy a iables, he coe icien o go e nmen size is gene ally nega i e, bu i is signi ican only o he WG es ima es. Table 4shows es ima ion esul s o i s -di e ence GMM, while Table 5shows he esul s o he baseline sys em GMM speci ica ion (limi ing he ins umen ma ix o wo lags). In Appendix 3(Tables 15 and 16), we epo esul s unde al e na i e app oaches o educing he dimension o he sys em GMM ins umen se : collapsing he ma ix o ins umen s while using all lags as ins umen s (Table 15), and limi ing hem o wo lags and collapsing he ins umen s a he same ime (Table 16). Fo i s -di e ence GMM (Table 4), we use h ee lags in he ma ix o ins umen s so as o ha e he same numbe o o hogonali y condi ions as in he baseline sys em GMM es ima ion, hus making he esul s mo e easily compa able.18 The p alues o he Hansen es s sugges ha in i ually e e y case, he null o join alidi y o all ins umen s canno be ejec ed. Mo eo e , he Di e ence-in-Hansen es esul s, whose p alues always exceed 0.10, poin owa d he supe io i y o sys em GMM o e i s -di e ence GMM. The pa ame e es ima es o he a iables o in e es ollow he same pa e n ound ea lie . The coe icien on he po e y headcoun is consis en ly nega i e and highly signi ican , ega dless o he choice o model and speci ica ion. In con as , he coe i- cien o he inequali y a iable a ies in sign and signi icance depending on he GMM app oach and he con ols used in he es ima ion. I is always posi i e and in one case signi ican o i s -di e ence GMM, consis en wi h ou esul s o he WG es ima es in Table 3and pa o he ea lie li e a u e (e.g., Fo bes 2000). Howe e , i is nega- i e and, in some cases, signi ican o sys em GMM, consis en wi h ou esul s o pooled-OLS and ano he s and o he li e a u e (e.g., Be g e al. 2018, and e e ences he ein). The nega i e e ec o po e y on g ow h is obus o changes in model spec- i ica ion and es ima ion me hod, while he e ec o inequali y on g ow h, which has been he ocus o a massi e li e a u e, is no . The heo e ical model ou lined in López and Se én (2015) and explo ed in Ma e o and Se én (2018) helps a ionalize ou empi ical esul s. In ha model, poo indi idu- als—i.e., hose whose ini ial endowmen is below a minimum consump ion le el—do no sa e and do no con ibu e o he economy’s agg ega e g ow h. In he absence o inancial ma ke s, he model shows ha po e y is unambiguously g ow h-de e ing, while inequali y can a ec g ow h di ec ly, h ough he sa ings o he non-poo , and indi ec ly, h ough i s e ec on po e y. While he indi ec e ec is nega i e, he di ec e ec is ambiguous (as ound by he empi ical li e a u e), and so is he o e all impac o inequali y on g ow h. As a u he diagnos ic check on he GMM es ima es o Tables 4,5,15,16,we inspec ed he esiduals o c oss-sec ionaldependence,usingPesa an’s(2021)CD es , 18 Da a o he in as uc u e index included in M4 a e a ailable o only 88 coun ies unde sys em GMM and 79 unde he i s -di e ence GMM speci ica ion. Using wo lags as ins umen s o es ima e his model would esul in he numbe o ins umen s exceeding he c oss sec ion dimension o he da a. Thus, we limi he numbe o ins umen s o jus one lag. 123 G ow h, inequali y and po e y: a obus ela ionship? 745 Table 4 G ow h, po e y and inequali y: i s -di e ence GMM es ima es M1. Skele on model M2. Ex ended wi h educa ion and in . p ices M3. Ex ended wi h policy a iables M4. Ex ended wi h policy and in as uc u es P0, lag − 0.0941*** (−2.59) − 0.0981*** (−2.63) − 0.150*** (− 3.35) − 0.150*** (−5.16) − 0.0997* (− 1.79) − 0.0947** (− 2.19) − 0.117** (− 2.25) − 0.103*** (−2.75) Gini, lag 0.0253 (0.37) 0.113 (1.10) 0.0330 (0.47) 0.131** (2.01) 0.0690 (0.73) 0.112 (1.56) 0.0434 (0.43) 0.115 (1.32) log y,lag − 0.106*** (−4.15) − 0.0876*** (−3.89) − 0.119*** (−5.45) − 0.100*** (− 3.48) − 0.0756*** (−4.60) − 0.0948*** (−4.47) − 0.154*** (− 3.71) − 0.105*** (− 4.50) − 0.149*** (− 4.47) − 0.103*** (− 3.24) − 0.0883*** (−4.01) − 0.106*** (−5.25) In . de la o , lag −0.0117 (− 1.19) −0.0177 (−1.47) −0.00942 (−0.90) Female educ., lag 0.0616** (2.39) 0.0148 (0.82) 0.0465*** (2.70) 0.00688 (0.71) −0.00271 (−0.20) 0.00611 (0.57) Male educ., lag − 0.0525* (− 1.92) −0.0140 (−0.59) −0.0319 (−1.51) In la ion 0.0005** (2.51) 0.0005** (2.26) 0.0004* (1.91) 0.0001 (0.06) 0.0017 (0.98) 0.0009 (0.83) T ade openness (log) 0.0396 (1.56) 0.0532** (2.24) 0.0457* (1.95) 123 746 G. A. Ma e o, L. Se én Table 4 (con inued) M1. Skele on model M2. Ex ended wi h educa ion and in . p ices M3. Ex ended wi h policy a iables M4. Ex ended wi h policy and in as uc u es Go . size (log) −0.0265 (− 1.50) −0.0350 (− 1.51) − 0.0407** (− 2.05) In as uc u e, lag 0.00750 (0.37) 0.0245* (1.78) 0.00778 (0.42) m2- es (p alue) 0.854 0.924 0.568 0.565 0.503 0.336 0.505 0.833 0.750 0.658 0.641 0.622 AR(3) (p alue) 0.0261 0.00635 0.0238 0.117 0.102 0.416 0.186 0.155 0.264 0.142 0.146 0.187 Hansen (p alue) 0.167 0.199 0.0571 0.412 0.269 0.330 0.581 0.431 0.759 0.397 0.575 0.424 Num. obs 502 503 502 467 468 467 248 249 248 345 346 345 Num. coun ies 130 130 130 113 113 113 84 84 84 79 79 79 Num. ins umen s 39 39 54 84 84 99 59 59 70 47 47 55 SeeNo einTable2. Es ima ions a e done using 2-s ep i s -di e ence GMM educing he numbe o ins umen lags o h ee. The ins umen se s a s a −3, and he a iance co a iance ma ix is compu ed using he small sample co ec ion o Windmeije (2005). Robus s a is ics in pa en heses. ***deno es signi icance a 1%, **a 5%, *a 10% 123 G ow h, inequali y and po e y: a obus ela ionship? 747 Table 5 G ow h, po e y and inequali y: sys em GMM es ima es M1. Skele on model M2. Ex ended wi h educa ion and in . p ices M3. Ex ended wi h policy a iables M4. Ex ended wi h policy and in as uc u es P0, lag − 0.117*** (−3.81) − 0.121*** (−5.34) − 0.0883*** (−3.71) − 0.0846*** (−4.12) − 0.0666*** (−2.96) − 0.0697*** (−3.85) −0.0506* (−1.88) −0.0503** (−2.32) Gini, lag − 0.107*** (− 2.92) − 0.0955** (−2.12) −0.0553 (− 1.44) −0.0587 (−1.57) − 0.0908*** (−3.30) − 0.106*** (−4.10) − 0.00339 (− 0.08) −0.0346 (−1.25) log y,lag − 0.0200*** (−4.60) − 0.00156 (− 0.61) − 0.0228*** (−6.19) − 0.0166*** (−4.61) − 0.00210 (− 0.92) − 0.0173*** (−5.30) − 0.0160*** (−4.72) − 0.00653** (−2.50) − 0.0184*** (−5.95) − 0.0362*** (−4.72) −0.0168 (− 1.53) − 0.0299*** (−3.29) In . de la o , lag −0.00137 (−1.38) − 0.00339* (− 1.93) − 0.000830 (−0.70) Female educ., lag 0.00122 (0.20) 0.00267 (0.44) 0.00385 (0.57) 0.00714** (2.47) 0.000557 (0.15) 0.00557* (1.83) Male educ., lag 0.00241 (0.36) − 0.00158 (− 0.23) −0.00144 (−0.19) In la ion 0.001* (1.93) 0.0012*** (2.91) 0.001*** (2.60) 0.0004 (0.68) 0.001** (2.02) 0.001* (1.83) T ade openness (log) 0.0255*** (3.26) 0.0264*** (2.83) 0.0217*** (2.58) 123 748 G. A. Ma e o, L. Se én Table 5 (con inued) M1. Skele on model M2. Ex ended wi h educa ion and in . p ices M3. Ex ended wi h policy a iables M4. Ex ended wi h policy and in as uc u es Go . size (log) −0.00306 (−0.42) −0.00861 (−0.91) −0.00349 (−0.44) In as uc u e, lag 0.0222*** (3.07) 0.0189** (2.03) 0.0166** (2.01) m2- es (p alue) 0.108 0.223 0.215 0.0574 0.117 0.117 0.279 0.434 0.493 0.227 0.203 0.313 AR(3) (p alue) 0.942 0.659 0.671 0.735 0.729 0.785 0.622 0.609 0.474 0.845 0.773 0.746 Hansen (p alue) 0.138 0.160 0.279 0.225 0.242 0.572 0.224 0.295 0.616 0.163 0.269 0.477 Di -Hansen o le els (p alue) 0.220 0.640 0.248 0.382 0.490 0.848 0.375 0.662 0.783 0.372 0.617 0.850 Num. obs 745 745 745 676 676 676 656 656 656 477 477 477 Num. coun ies 156 156 156 131 131 131 147 147 147 88 88 88 Num. ins umen s 54 54 76 120 120 142 116 116 138 82 82 97 See no e Table 4Es ima ions a e done using 2-s ep sys em GMM educing he numbe o ins umen lags o wo. The ins umen se s a s a −3, and he a iance co a iance ma ix is compu ed using he small sample co ec ion o Windmeije (2005). The di e ence Hansen es assesses he alidi y o he ins umen s o he le el equa ion in sys em GMM. Robus s a is ics in pa en heses. *** deno es signi icance a 1%, ** a 5%, * a 10% 123 G ow h, inequali y and po e y: a obus ela ionship? 749 and ocusing on he model e sions including bo h po e y and inequali y. Resul s a e shown in Table 17 (Appendix 4). In he majo i y o cases, he es esul s a e suppo i e o he empi ical speci ica ion. This is pa icula ly he case o he models including policy a iables (models M3 and M4 in he a o emen ioned ables), o which he es ails in all cases o ejec he null o c oss-sec ional independence. Fo he s ipped- down model M1, which omi s all con ols, esul s a e mo e mixed, as he es ails o ejec he null a he con en ional 5% le el in some exe cises ( hose in Tables 4and 5) bu ejec s i in o he s ( hose in Tables 15,16). The excep ion is model M2, o which he es consis en ly inds signi ican e idence o c oss-sec ional dependence.19 O e all, we ake hese esul s as suppo ing he iew ha models M3 and M4 a e co ec ly speci ied. Howe e , he p esence o esidual c oss-sec ional co ela ion in model M2— i s explo ed by Pe o i (1996) and Fo bes (2000), sugges s ha he model’s es ima ed s anda d e o s may be inco ec .20 4.1 Weak ins umen s analysis Bazzi and Clemens (2013) ha e aised he po en ial p oblem o weak ins umen s when using sys em GMM es ima ion in g ow h eg essions. Weak iden i ica ion a ises when he ins umen s a e only weakly co ela ed wi h he endogenous eg esso s, and i s consequence is ha es ima o s pe o m poo ly (Nelson and S a z 1990). To assess he s eng h o he ins umen s employed in ou sys em GMM es ima ions—in pa icula , he iden i ica ion o he po e y and inequali y pa ame e s—we use ools designed o se ings ea u ing mul iple endogenous eg esso s. We ollow Sande son and Windmeije (2016) (SW he ea e ), who p opose a condi ional Fs a is ic based on Ang is and Pischke (2009) o es whe he , in a mul i a ia e se ing, a pa icula endogenous eg esso is weakly ins umen ed. Fo each such eg esso , a condi ional es is cons uc ed by “pa ialing-ou ” linea p ojec ions o he emaining endogenous eg esso s. SW show ha he condi ional Fs a is ic can be assessed agains he S ock and Yogo c i ical alues, and he weakness can hen be exp essed in e ms o he size o he bias o he IV (o 2SLS) es ima o ela i e o ha o he OLS es ima o . The null hypo hesis is ha he ins umen s a e weak. I is ejec ed i he condi ional Fs a is ic exceeds he co esponding c i ical alue, and we use a c i ical alue allowing o a 30 pe cen maximal ela i e bias. We also pe o m a Chi-squa e unde -iden i ica ion es sepa a ely o each eg esso . He e, he null hypo hesis is ha he ma ix o coe i- cien s om he i s -s age condi ional eg essions is no ull ank, signaling a comple e 19 The obus ness exe cises in sec ion V ollow he same pa e n ega ding c oss-sec ional dependence es s: The esiduals o models M3 and M4 show no e idence o dependence, while in mos cases, hose o model M3 yield he opposi e conclusion. Model M1 again yields mixed esul s. 20 The absence o c oss-sec ional dependence in models M3 and M4 (and, o a lesse ex en , M1) may seem su p ising gi en ha sho - e m g ow h luc ua ions ypically display signi ican in e na ional como emen . Howe e , ou use o 5-yea a e ages g ea ly mi iga es he como emen usually ound a annual (o highe ) equency. In addi ion, he inclusion o ime dummies in ou empi ical speci ica ions also helps soak up common ac o s a ec ing g ow h in mul iple coun ies. Las ly, he p esence o s a is ically signi ican policy a iables in models M3 and M4 likely helps soak up any emaining c oss-sec ional co ela ion in hese speci ica ions, unlike in models M1 and M2. 123 756 G. A. Ma e o, L. Se én Table 7 (con inued) M1. Skele on model M2. Ex ended wi h educa ion and in . p ices M3. Ex ended wi h policy a iables M4. Ex ended wi h policy and in as uc u es P0, lag − 0.142*** (− 4.86) − 0.152*** (− 6.15) − 0.101*** (−4.93) − 0.0944*** (−4.78) − 0.0652*** (−3.12) − 0.0765*** (−4.41) −0.0557 (− 1.54) − 0.0629*** (−2.62) Gini, lag − 0.107*** (− 2.92) −0.0689 (− 1.55) − 0.0553 (− 1.44) −0.0412 (−1.24) − 0.0908*** (−3.30) − 0.107*** (−3.74) − 0.00339 (− 0.08) −0.0206 (−0.58) Hansen (p alue) 0.124 0.160 0.207 0.292 0.242 0.688 0.195 0.295 0.618 0.181 0.269 0.543 Po e y Gap, P1, Po e y line US$ 1.25 P1, lag − 0.263*** (− 3.74) − 0.227*** (− 4.57) − 0.145*** (−3.35) − 0.138*** (−2.90) − 0.200*** (−4.89) − 0.158*** (−4.49) − 0.144** (− 2.28) −0.121** (−2.45) Gini, lag − 0.107*** (− 2.92) − 0.0658* (− 1.66) − 0.0553 (− 1.44) −0.0486 (−1.21) − 0.0908*** (−3.30) − 0.0789*** (−2.63) − 0.00339 (− 0.08) −0.0182 (−0.57) Hansen (p alue) 0.319 0.160 0.346 0.184 0.242 0.620 0.194 0.295 0.520 0.129 0.269 0.630 Po e y gap, P1, po e y line US$ 2.0 123 G ow h, inequali y and po e y: a obus ela ionship? 757 Table 7 (con inued) M1. Skele on model M2. Ex ended wi h educa ion and in . p ices M3. Ex ended wi h policy a iables M4. Ex ended wi h policy and in as uc u es P1, lag − 0.191*** (− 3.65) − 0.176*** (− 5.08) − 0.118*** (−3.46) − 0.111*** (−3.66) − 0.141*** (−4.17) − 0.119*** (−4.20) − 0.0895** (− 1.98) −0.0854** (−2.57) Gini, lag − 0.107*** (− 2.92) − 0.0756* (− 1.93) − 0.0553 (− 1.44) −0.0524* (−1.72) − 0.0908*** (−3.30) − 0.0925*** (−3.45) − 0.00339 (− 0.08) −0.0356 (−1.10) Hansen (p alue) 0.210 0.160 0.290 0.253 0.242 0.491 0.217 0.295 0.493 0.134 0.269 0.470 Po e y gap, P1, po e y line US$ 4.0 P1, lag − 0.169*** (− 3.57) − 0.183*** (− 5.61) − 0.125*** (−3.41) − 0.120*** (−4.23) − 0.0993*** (−3.76) − 0.104*** (−4.71) − 0.0706* (− 1.86) −0.0709** (−2.33) Gini, lag − 0.107*** (− 2.92) − 0.0871** (− 2.03) − 0.0553 (− 1.44) −0.0451 (−1.15) − 0.0908*** (−3.30) − 0.102*** (−3.78) − 0.00339 (− 0.08) −0.0326 (−1.08) Hansen (p alue) 0.0985 0.160 0.267 0.208 0.242 0.581 0.208 0.295 0.683 0.191 0.269 0.540 Squa ed po e y gap, P2, po e y line US$ 1.25 123 758 G. A. Ma e o, L. Se én Table 7 (con inued) M1. Skele on model M2. Ex ended wi h educa ion and in . p ices M3. Ex ended wi h policy a iables M4. Ex ended wi h policy and in as uc u es P2, lag − 0.413*** (− 4.15) − 0.347*** (− 4.18) − 0.197*** (−2.85) −0.177** (−2.36) − 0.300*** (−4.54) − 0.237*** (−4.22) − 0.230** (− 2.35) −0.197** (−2.56) Gini, lag − 0.107*** (− 2.92) −0.0518 (− 1.07) − 0.0553 (− 1.44) −0.0370 (−1.15) − 0.0908*** (−3.30) − 0.0709** (−2.52) − 0.00339 (− 0.08) −0.00956 (−0.30) Hansen (p alue) 0.431 0.160 0.420 0.172 0.242 0.567 0.212 0.295 0.528 0.153 0.269 0.662 Squa ed po e y gap, P2, po e y line US$ 2.0 P2, lag − 0.266*** (− 3.70) − 0.239*** (− 4.81) − 0.150*** (−3.63) − 0.149*** (−3.77) − 0.203*** (−4.90) − 0.164*** (−4.45) − 0.139** (− 2.19) −0.116** (−2.41) Gini, lag − 0.107*** (− 2.92) − 0.0652* (− 1.68) − 0.0553 (− 1.44) −0.0442 (−1.31) − 0.0908*** (−3.30) − 0.0820*** (−2.79) − 0.00339 (− 0.08) −0.0241 (−0.81) Hansen (p alue) 0.270 0.160 0.318 0.210 0.242 0.587 0.211 0.295 0.521 0.127 0.269 0.622 Squa ed po e y gap, P2, po e y line US$ 4.0 123 G ow h, inequali y and po e y: a obus ela ionship? 759 Table 7 (con inued) M1. Skele on model M2. Ex ended wi h educa ion and in . p ices M3. Ex ended wi h policy a iables M4. Ex ended wi h policy and in as uc u es P2, lag − 0.201*** (− 3.58) − 0.196*** (− 5.44) − 0.131*** (−3.06) − 0.127*** (−3.78) − 0.132*** (−3.98) − 0.123*** (−4.59) − 0.0850* (− 1.85) −0.0833** (−2.26) Gini, lag − 0.107*** (− 2.92) − 0.0801** (− 2.03) − 0.0553 (− 1.44) −0.0493 (−1.53) − 0.0908*** (−3.30) − 0.0980*** (−3.89) − 0.00339 (− 0.08) −0.0327 (−1.55) Hansen (p alue) 0.136 0.160 0.285 0.248 0.242 0.499 0.188 0.295 0.487 0.160 0.269 0.603 See no e in Table 4 123 760 G. A. Ma e o, L. Se én 14). Fu he inspec ion e eals ha he co ela ion is highe o he mo e ecen da a, eaching 0.93 in 2005 and 0.96 in 2010. Table 8shows es ima ion esul s o models M1, M2, M3 and M4 using he Po calNe in e pola ed po e y se ies and ou p e e ed sys em GMM speci ica ion. Compa ison wi h Table 4 e eals ha he esul s a e obus o he use o his al e na i e sou ce o po e y da a: Po e y consis en ly ca ies a nega i e coe icien , signi ican in all cases bu one. In u n, he coe icien on inequali y is also nega i e in mos ins ances, bu insigni ican in h ee ou o eigh cases. 5.4 Addi ional con ols Nex , we assess he obus ness o ou esul s o he use o al e na i e con ols. We ocus on wo ex ensions. Fi s , we conside al e na i e measu es o educa ion o p oxy o human capi al. Second, we conside a se o ins i u ional quali y a iables. Resul s a e shown in Table 19 in he Appendix 6. In model M2, we added male and emale educa ion sepa a ely, ollowing Pe o i (1996) and Owen e al. (2002). He e, we es ima e se e al a ian s o model M2, using a e age yea s o schooling, on he one hand, and he pe cen age o he popula ion wi h a leas p ima y o seconda y educa ion, on he o he hand ( i s and second columns in Table 19). In u n, we conside wo o he mos widely used measu es o he quali y o ins i- u ions (see also Table 13 in Appendix 2): an index o democ a ic accoun abili y (“democ acy”), and an index o go e nmen s abili y (“s abili y”), in o ma ion aken om he poli ical isk module o he In e na ional Coun y Risk Da abase.22 Columns 3, 4 and 5 o Table 19 ex end models M2, M3 and M4 wi h hese ins i u ional a iables; column 6 epo s he es ima ion esul s when join ly including all he a iables om M2, M3 and M4. Finally, and jus o illus a i e pu poses, we epo (in he las column o he able) es ima es o a model including all he con ols. They should be aken wi h cau ion, howe e , gi en he sha p educ ion in sample size (by almos hal ela i e o columns 1–2) and he high deg ee o collinea i y among he eg esso s. Es ima ed coe icien s o he pe cen age o popula ion wi h p ima y and seconda y educa ion a e posi i e and signi ican . In he ex ended speci ica ions wi h ins i u ional a iables, he coe icien s o bo h he quali y o democ acy and go e nmen s abili y a e posi i e and, in mos cases, signi ican , con i ming ha he quali y o ins i u ions is posi i ely co ela ed wi h g ow h. Mo e impo an ly, he baseline es ima ion esul s o po e y (consis en ly nega i e) and inequali y (i s sign and signi icance depends on he pa icula speci ica ion) a e obus o he inclusion o all hese addi ional con ols. 22 The e a e o he ins i u ional dimensions, such as he con ol o co up ion, he mili a y in powe , he deg ee o in e na ional con lic s, o he Poli y2 a iable ( om he Poli y IV p ojec ). Including all hese dimensions/ a iables simul aneously would in oduce se ious p oblems o collinea i y in he es ima ed model. 123 G ow h, inequali y and po e y: a obus ela ionship? 761 Table 8 Sys em GMM es ima es: obus ness o he use o Po calNe da a M1. Skele on model M2. Ex ended wi h educa ion and in . p ices M3. Ex ended wi h policy a iables M4. Ex ended wi h policy and in as uc u es P0, lag − 0.0959*** (0.0244) − 0.103*** (0.0325) − 0.0675** (0.0272) −0.0537* (0.0323) − 0.0637** (0.0285) − 0.0635** (0.0323) −0.0327* (0.0183) −0.0206 (0.0448) Gini, lag − 0.0794* (0.0407) −0.0271 (0.0571) − 0.0783** (0.0366) − 0.0777** (0.0383) − 0.105*** (0.0363) − 0.0865** (0.0412) 0.0218 (0.0491) 0.000979 (0.0694) log y,lag − 0.0134*** (0.00496) − 0.000522 (0.00322) − 0.0166*** (0.00498) − 0.00978** (0.00465) 0.00155 (0.00333) − 0.00822* (0.00498) − 0.0141** (0.00594) − 0.00678** (0.00290) − 0.0160*** (0.00576) − 0.0246** (0.0103) − 0.0289** (0.0123) −0.0351*** (0.0121) In . de la o , lag − 0.0256*** (0.00749) − 0.0355*** (0.0119) − 0.0270*** (0.00792) Female educ., lag 0.000286 (0.00584) 0.00111 (0.00629) 0.000747 (0.00695) 0.000651 (0.00376) 0.00286 (0.00440) 0.00464 (0.00476) Male educ., lag 0.00145 (0.00634) −0.00156 (0.00668) − 0.0000402 (0.00700) In la ion − 0.000204 (0.000193) 0.000531* (0.000297) 0.000430 (0.000353) − 0.000114 (0.000248) 0.000823 (0.000520) 0.000887 (0.000602) T ade openness (log) 0.0284** (0.0117) 0.0320** (0.0144) 0.0280* (0.0144) Go . size (log) −0.00351 (0.0125) 0.00171 (0.0102) 0.00139 (0.0118) In as uc u e, lag 0.0208*** (0.00773) 0.0284*** (0.0105) 0.0286* (0.0151) 123 762 G. A. Ma e o, L. Se én Table 8 (con inued) M1. Skele on model M2. Ex ended wi h educa ion and in . p ices M3. Ex ended wi h policy a iables M4. Ex ended wi h policy and in as uc u es m2- es (p alue) 0.125 0.345 0.140 0.00886 0.0692 0.0129 0.115 0.990 0.383 0.0668 0.219 0.109 AR(3) (p alue) 0.0812 0.231 0.383 0.152 0.255 0.475 0.247 0.478 0.470 0.136 0.0734 0.466 Hansen (p alue) 0.0304 0.102 0.0159 0.0934 0.153 0.173 0.114 0.318 0.206 0.246 0.182 0.196 Di -Hansen o le els (p alue) 0.422 0.459 0.087 0.710 0.586 0.556 0.432 0.932 0.816 0.618 0.317 0.205 Num. obs 522 522 522 474 474 474 491 491 491 360 360 360 Num. coun ies 136 136 136 116 116 116 130 130 130 81 81 81 Num. ins umen s 32 43 48 102 100 96 102 97 96 80 67 68 See no e in Table 5. F om Po calNe , he po e y line is 1.90 US$ 2011, which upda es he p e ious line o 1.25 US$ 2005 (Fe ei a e al. 2016). We use he in e pola ed po e y se ies p o ided in Po calNe , which s a in 1981 and a e epo ed e e y 3 yea s. To cons uc a non-o e lapping 5-yea s panel da a simila o he one used in ou baseline speci ica ion, and ma ch po e y da a wi h all o he a iables (g ow h and o he con ols), we use a “closes ” c i e ia o ake he a e age i wo po e y obse a ions a e 1 yea abo e and one below he assigned yea 123 G ow h, inequali y and po e y: a obus ela ionship? 763 5.5 Al e na i e econome ic speci ica ions We also pe o med a numbe o o he obus ness checks conce ning he empi ical speci ica ion and es ima ion app oach. To sa e space, we jus p o ide a b ie sum- ma y he e ( esul s a e a ailable upon eques ). Fi s , we modi ied he sys em GMM es ima ion employing di e en lag s uc u es—e.g., using yi −s,pi −s,gi −sand xi −s, o s≥4 o he i s -di e ence equa ion and yi −4,pi −4,gi −4and xi −4 o he le el equa ion—o using 1-s ep ins ead o 2-s ep es ima es. We also expe imen ed wi h a modi ied e sion o he basic empi ical equa ion including a quad a ic e m in he Gini coe icien . The main conclusion is ha he signi ican ly nega i e e ec o po e y on g ow h is qui e obus o all hese a ia ions in speci ica ion and es ima ion app oach, while he inequali y-g ow h ela ionship is highly agile. Finally, we also e-es ima ed he models in a pu e c oss sec ion o coun ies, wi h he a iables exp essed as a e ages o e he en i e sample pe iod, cap u ing wha could be iewed as he long- un ela ionship be ween hem. The es ima ed po e y coe icien emains uni o mly nega i e and signi ican , al hough i s p ecision declines somewha ela i e o he panel es ima es. In u n, inequali y ends o show a nega i e and signi ican coe icien , mo e equen ly han in he panel es ima es, consis en wi h ecen e idence (e.g., Hal e e al 2014;Be ge al.2018) ha inequali y exe s a nega i e long- un impac on g ow h. 6 Po e y egimes 6.1 The e ec o po e y and inequali y on g ow h Thenonpa ame icanalysisin hep ecedingsec ionhin eda possiblenonlinea e ec s o po e y and inequali y on g ow h. To ake a deepe look, we es ima e al e na i e e - sions o Eqs. (1)–(3) allowing o di e en coe icien s on lagged po e y and lagged inequali y depending on whe he he lagged alue o P0 lies abo e o below he sample median (2.7% o ou baseline P0, see Table 1). We ollow he same s a egy condi- ioning ins ead on he lagged le el o inequali y, and es ima e Eqs. (1)–(3) allowing o di e en coe icien s on po e y and inequali y depending on whe he he lagged Gini coe icien lies abo e o below i s sample median (39.8%, see Table 1). Table 9 epo s es ima es dis inguishing whe he po e y is abo e o below he median—wha we shall label he ‘high po e y egime’ and ‘low po e y egime,’ espec i ely. In u n, Table 10 epo s he es ima es dis inguishing whe he inequali y is abo e o below he median— he ‘high inequali y egime’ and ‘low inequali y egime,’ espec i ely. In bo h cases, we use he baseline sys em GMM speci ica ion (Table 5). Table 9shows ha , unde he low po e y egime, he impac o po e y on g ow h is nega i e bu s a is ically insigni ican . Howe e , i is nega i e and highly signi ican unde he high po e y egime. In u n, he es ima ed coe icien on he Gini index is in mos cases nega i e, bu i u ns signi ican only o high po e y a es and o he M1 and M3 model speci ica ions. Thus, like wi h he uncondi ional es ima es, while he esul o po e y is obus , he esul o inequali y is no . In con as , Table 10 shows ha , when we condi ion on he lagged le el o inequali y, he es ima ed coe icien s on 123 764 G. A. Ma e o, L. Se én Table 9 Es ima ion esul s by po e y egimes: baseline sys em GMM M1. Skele on model M2. Ex ended wi h educa ion and in . p ices M3. Ex ended wi h policy a iables M4. Ex ended wi h policy and in as uc u es P0, lag (P0 ≤ Median) 0.427 (0.49) −0.101 (− 0.16) −0.729 (− 0.92) −0.944 (− 1.29) −0.293 (− 0.44) −0.430 (− 0.61) 1.230 (1.44) −0.320 (− 0.41) −1.019 (−0.84) P0, lag (P0 > Median) − 0.125*** (− 4.30) − 0.124*** (− 5.04) − 0.0881*** (− 4.08) − 0.0928*** (− 5.25) − 0.0733*** (− 3.25) − 0.0784*** (− 3.65) − 0.0868*** (− 2.84) − 0.0586*** (− 2.66) −0.0927** (−2.31) Gini, lag (P0 ≤ Median) −0.0553 (− 0.81) −0.0589 (− 1.24) − 0.00103 (− 0.02) −0.00504 (− 0.10) −0.0689 (− 1.52) −0.0759 (− 1.50) −0.0285 (− 0.46) −0.0224 (− 0.55) −0.0114 (−0.14) Gini, lag (P0 > Median) − 0.138*** (− 2.83) − 0.0980*** (− 2.63) − 0.0544 (− 1.23) −0.0547 (− 1.46) − 0.0969*** (− 2.85) − 0.107*** (− 3.17) −0.0546 (− 1.00) −0.0409 (− 1.40) −0.0759 (−0.95) log y,lag − 0.0208*** (− 4.56) − 0.0153*** (− 2.73) − 0.0275*** (− 5.31) − 0.0160*** (− 4.45) − 0.0115* (− 1.89) − 0.0236*** (− 6.43) − 0.0179*** (− 4.86) − 0.0134*** (− 2.73) − 0.0243*** (− 5.32) − 0.0420*** (− 2.98) − 0.0374*** (− 3.29) − 0.0344*** (− 4.65) − 0.0504*** (−6.72) m2 (p alue) 0.097 0.294 0.205 0.106 0.138 0.168 0.379 0.527 0.539 0.195 0.456 0.389 0.652 Hansen (p alue) 0.0990 0.0476 0.256 0.211 0.111 0.354 0.408 0.238 0.476 0.163 0.425 0.990 0.122 Num. obs 745 745 745 676 676 676 655 655 655 477 477 477 477 Num. coun ies 156 156 156 131 131 131 147 147 147 88 88 88 88 Num. Ins umen s 55 55 85 100 100 130 97 97 127 77 77 127 41 Baseline sys em GMM es ima es: 1 lag in he ins umen ma ix, s a ing a −3. See also he no e o Table 4. In he las column o he able, o u he educe he numbe o ins umen s in model M4, we conside he case wi h 2 lags, s a ing a −3, and using he collapse op ion. The sample is di ided acco ding wi h he sample median o P0, which is 2.7% o ou baseline P0 wi h po e y line o 2US$ 123 G ow h, inequali y and po e y: a obus ela ionship? 765 Table 10 Es ima ion esul s by inequali y egimes: baseline sys em GMM M1. Skele on model M2. Ex ended wi h educa ion and in . p ices M3. Ex ended wi h policy a iables M4. Ex ended wi h policy and in as uc u es P0, lag (Gini ≤ Median) − 0.109*** (− 3.33) − 0.0957*** (− 4.04) − 0.0789*** (− 4.54) − 0.0924*** (− 4.64) − 0.0561** (− 2.51) − 0.0681*** (− 3.09) − 0.0529** (− 2.06) −0.0527* (− 1.68) −0.0700** (−1.97) P0, lag (Gini > Median) − 0.146*** (− 4.87) − 0.115*** (− 3.93) − 0.0976*** (− 4.15) − 0.0696*** (− 2.93) − 0.105*** (− 4.03) − 0.0821*** (− 3.64) − 0.102*** (− 3.50) − 0.0530** (− 2.13) −0.0935** (−2.57) Gini, lag (Gini ≤ Median) 0.0343 (0.32) −0.0374 (− 0.48) 0.119* (1.86) 0.0885 (1.60) −0.0408 (−0.52) −0.0461 (− 0.73) 0.0405 (0.36) 0.00487 (0.08) 0.0931 (0.71) Gini, lag (Gini > Median) −0.0397 (−0.53) −0.0484 (− 0.91) 0.0345 (0.71) 0.0223 (0.55) −0.0683 (−1.22) −0.0647 (− 1.51) −0.00858 (− 0.10) −0.00988 (− 0.22) 0.0139 (0.17) log y,lag − 0.0230*** (− 5.98) − 0.00713*** (−2.91) − 0.0203*** (− 6.25) − 0.0152*** (− 5.05) − 0.00585* (− 1.82) − 0.0163*** (− 5.55) − 0.0186*** (− 4.63) − 0.00851*** (−3.26) − 0.0190*** (− 6.22) − 0.0262*** (− 2.73) − 0.0301*** (− 3.68) − 0.0353*** (− 4.53) − 0.0368*** (−3.86) m2 (p alue) 0.1000 0.130 0.142 0.0681 0.0628 0.0770 0.352 0.475 0.441 0.203 0.359 0.277 0.285 Hansen (p alue) 0.116 0.0375 0.168 0.130 0.0793 0.430 0.242 0.137 0.377 0.0907 0.232 0.988 0.254 Num. obs 745 745 745 676 676 676 655 655 655 477 477 477 477 Num. coun ies 156 156 156 131 131 131 147 147 147 88 88 88 88 Num. Ins umen s 55 55 85 100 100 130 97 97 127 77 77 127 41 Baseline sys em GMM es ima es: 1 lag in he ins umen ma ix, s a ing a −3. See also he no e o Table 4. In he las column o he able, o u he educe he numbe o ins umen s in model M4, we conside he case wi h 2 lags, s a ing a −3, and using he collapse op ion. The sample is di ided acco ding wi h he sample median o he Gini index, which is 39.8% 123 772 G. A. Ma e o, L. Se én is no pe mi ed by s a u o y egula ion o exceeds he pe mi ed use, you will need o ob ain pe mission di ec ly om he copy igh holde . To iew a copy o his licence, isi h p://c ea i ecommons.o g/licenses/ by/4.0/. Appendix 1: Logno mal app oxima ion o al e na i e po e y measu es Following Dolla and K aay (2002), López and Se én (2015) o Pinko skiy and Sala-i-Ma in (2013), we cons uc a se o po e y igu es ( he headcoun a io, P0, he po e y gap, P1 and he squa ed po e y gap, P2) using a logno mal app oxima ion on he basis o he obse ed pe capi a income le els and Gini coe icien s, which a e a ailable much mo e widely han su ey-based po e y da a. The use o he logno mal app oxima ion o he dis ibu ion o income da es back o Gib a (1931). The li e a u e employs also o he unc ional o ms, such as he Pa e o, he gamma o he Weibull dis ibu ion, bu he logno mal is he mo e widely used. Indeed, López and Se én (2006) compa e he quin ile income sha es gene a ed by a logno mal dis ibu ion wi h hei obse ed coun e pa s using da a om o e 1000 household su eys and ind he logno mal app oxima ion i s he da a ex emely well, so ha hey a e unable o ejec he null hypo hesis ha pe capi a income ollows a logno mal dis ibu ion. Unde logno mali y, gi en he Gini coe icien (g), he s anda d de ia ion (σ)o he log o income is gi en by σ−11+g 2, whe e (·) is he s anda d no mal cumula i edis ibu ion unc ion.Using hisexp essionand helogo pe capi aincome (y), we can compu e he FGT amily o po e y measu es o a gi en po e y line zas: P0log(z)−y σ+σ 2 P1log(z)−y σ+σ 2−ey zlog(z)−y σ−σ 2 P2log(z)−y σ+σ 2−2ey zlog(z)−y σ−σ 2+ey z2eσ2log(z)−ν σ−3σ 2. Appendix 2: Da a desc ip ion and c oss-co ela ions See Tables 13,14. 123 G ow h, inequali y and po e y: a obus ela ionship? 773 Table 13 Desc ip ion o he a iables Name Desc ip ion Sou ce Num.Obs. ( es ic ed o P0 and Gini sample) Sample a e age S anda d de ia ion Pe capi a eal GDP Le el o ac i i y and deg ee o de elopmen : PPP Con e ed GDP Pe Capi a (Chain Se ies), a 2005 cons an p ices Penn Wo ld Tables 7.1 749 9793 US$ (PPP- 2005) 10,462 (PPP-2005) Po e y The headcoun a io P0 (le el o po e y), he po e y gap P1 (in ensi y), and he squa ed po e y gap P2 (se e i y). Fo he log-logis ic measu e, he baseline po e y line is US$ 2; o Po calNe , we use US$ 1.90 as po e y line Own calcula ion based on logno mal app oxima- ion; Po calNe 749 (logno - mal) 556 (Po cal.) 16.18% (P0) 7.34 (P1) 4.40% (P2) 18.56 (Po cal.) 24.75% (P0) 13.22% (P1) 8.92% (P2) 21.14% (Po cal.) Gini coe icien Measu e o income inequali y (be ween 0 and 1). Based only on na ionally ep esen a i e su eys (a ea, popula ion and age), and based on income (ne o ans e s and axes) and expendi u e igu es UN-WIID2 (2008); Po calNe 749 40.20% 9.98% Yea s o seconda y educa ion ( o al, male and emale) A e age yea s o seconda y educa ion o he male popula ion and he a e age yea s o seconda y educa ion o he emale popula ion Ba o and Lee (2013) Educa ional A ainmen Da a 684 1.95 ( o al) 1.77 ( emale) 2.15 (male) 1.42 ( o al) 1.44 ( emale) 1.44 (male) 123 774 G. A. Ma e o, L. Se én Table 13 (con inued) Name Desc ip ion Sou ce Num.Obs. ( es ic ed o P0 and Gini sample) Sample a e age S anda d de ia ion A ained educa ion (p ima y and seconda y) Pe cen age o popula ion ( o al) wi h a leas p ima y o seconda y educa ion Ba o and Lee (2013) Educa ional A ainmen Da a 684 19.5 (p ima y) 16.2 (sec- onda y) 12.8 (p ima y) 13.3 (seconda y) In es men p ices Domes ic p ice o in es men goods ela i e o ha o he U.S. as a measu e o ma ke dis o ions Penn Wo ld Tables 7.1 745 0.65 ( ela i e o US) 0.31 ( ela i e o US) In la ion GDP de la o , as an indica o o mac oeconomic s abili y Wo ld De elopmen Indica o s, Wo ld Bank 667 16.35% 32.05% Deg ee o openness Volume o ade wi h espec o i s GDP Penn Wo ld Table 7.1 749 75.8% 49.7% Go e nmen size The a io o public consump ion o GDP: as an indica o o he bu den imposed by he go e nmen on he economy Penn Wo ld Table 7.1 749 9.65% 5.41% 123 G ow h, inequali y and po e y: a obus ela ionship? 775 Table 13 (con inued) Name Desc ip ion Sou ce Num.Obs. ( es ic ed o P0 and Gini sample) Sample a e age S anda d de ia ion In as uc . Index Composi e index o public in as uc u e including: elecommunica- ion sec o (numbe o main elephone lines pe 1000 wo ke s), he powe sec o ( he elec ici y gene a ing capaci y in MW pe 1000 wo ke s), he anspo a ion sec o ( he leng h o he oad ne wo k—in km. pe sq. km. o land a ea) Wo ld De elopmen Indica o s, Wo ld Bank. Based on Calde ón e al. (2015) 528 0.39 1.33 Democ acy Deg ee o Democ acy: whe he he e a e ee and ai elec ions and he deg ee o go e nmen ’s accoun abili y. Range o alues be ween 0—minimum democ acy—and 6—maximum democ acy) In e na ional Coun y Risk Da abase 474 4.15 1.46 123 776 G. A. Ma e o, L. Se én Table 13 (con inued) Name Desc ip ion Sou ce Num.Obs. ( es ic ed o P0 and Gini sample) Sample a e age S anda d de ia ion Go e nmen s abili y Deg ee o Go e nmen s abili y: measu es he go e nmen ’s abili y o ca y ou i s decla ed p og am(s) and i s abili y o s ay in o ice. Range o alues be ween 1—minimum s abili y—and 12—maximum s abili y In e na ional Coun y Risk Da abase 474 7.71 2.06 123 G ow h, inequali y and po e y: a obus ela ionship? 777 Table 14 Co ela ion ma ix G ow h pcGDP (log) P0 (US$ 2) logno mal P0 (Po calne ) Gini Second.y o al Second.y emale Second.y male P ima y a ain. (%) G ow h 1.000 pcGDP(log) 0.152 1.000 P0 (US$ 2) −0.176 −0.828 1.000 P0 (Po cal.) −0.162 −0.843 0.892 1.000 Gini −0.184 −0.484 0.357 0.363 1.000 Sec.y o al 0.198 0.766 −0.592 −0.651 −0.447 1.000 Sec.y emale 0.195 0.777 −0.595 −0.650 −0.409 0.989 1.000 Sec.y male 0.196 0.738 −0.575 −0.638 −0.477 0.988 0.956 1.000 P im.a 0.007 0.202 −0.154 −0.193 −0.088 −0.151 −0.146 −0.150 1.000 Sec.a 0.214 0.641 −0.529 −0.579 −0.404 0.877 0.865 0.871 −0.184 In .P ice −0.008 0.280 −0.033 −0.139 −0.220 0.238 0.235 0.235 0.012 In la ion −0.079 −0.058 −0.022 0.000 0.071 −0.031 −0.033 −0.029 −0.072 Open 0.298 0.143 −0.168 −0.192 −0.087 0.266 0.285 0.240 −0.120 Go .Size −0.115 −0.332 0.381 0.453 0.080 −0.267 −0.248 −0.282 −0.071 In as 0.197 0.922 −0.781 −0.812 −0.499 0.741 0.743 0.722 0.218 Democ 0.126 0.682 −0.443 −0.516 −0.377 0.503 0.539 0.456 0.219 Go .S ab 0.354 0.164 −0.094 −0.150 −0.099 0.225 0.221 0.223 0.020 123 778 G. A. Ma e o, L. Se én Table 14 (con inued) Second. a ain. (%) In . p ice ( ela i . US) In la Open, adjus . (log) Go . size (log) In as (index) Democ (0–6 index) Go . s ab. (0–12 index) G ow h pcGDP(log) P0 (US$ 2) P0 (Po cal.) Gini Sec.y o al Sec.y emale Sec.y male P im.a Sec.a 1.000 In .P ice 0.173 1.000 In la ion −0.017 −0.051 1.000 Open 0.308 −0.014 −0.138 1.000 Go .Size −0.182 −0.104 −0.041 −0.129 1.000 In as 0.657 0.251 −0.120 0.228 −0.307 1.000 Democ 0.432 0.306 −0.133 0.173 −0.214 0.704 1.000 Go .S ab 0.186 −0.040 −0.098 0.267 −0.110 0.259 0.158 1.000 Va iables a e ans o med in he same way as o he eg ession analysis. Fo example, pe capi a GDP in logs; po e y and he Gini coe icien in le els; openness in logs and adjus ed by popula ion, kilome e s, o be oil expo e s and landlock; go e nmen size in logs, e c. We conside wo al e na i e measu es o he headcoun po e y a e: i s , using a logno mal app oxima ion (Dolla and K aay 2002; Sala-i-Ma in 2006; López and Se én 2015), and, o a educed sample, using he Po calNe da abase 123 G ow h, inequali y and po e y: a obus ela ionship? 779 Appendix 3: Al e na i e sys em GMM es ima ion esul s See Tables 15,16. Table 15 G ow h, po e y and inequali y: sys em GMM es ima es (collapse, all lags) M1. Skele on model M2. Ex ended wi h educa ion and in . p ices M3. Ex ended wi h policy a iables M4. Ex ended wi h policy and in as uc u es P0, lag − 0.157*** (−4.14) − 0.138*** (−4.06) − 0.104*** (−3.85) − 0.0860*** (−3.77) − 0.0963*** (−4.17) − 0.0918*** (−4.61) − 0.0626** (−2.06) −0.0581** (−2.08) Gini, lag − 0.165** (− 2.53) − 0.153*** (−3.01) − 0.0798* (− 1.77) −0.0444 −0.0289 (−0.87) − 0.0766** (−2.30) 0.0105 (0.32) −0.0215 (−0.60) (−1.05) log y,lag − 0.0264*** (−4.83) − 0.00454 (− 1.35) − 0.0271*** (−5.36) − 0.0216*** (−4.22) − 0.00310 (− 1.18) − 0.0181*** (−4.37) − 0.0172*** (−5.09) − 0.0057** (−2.20) − 0.0194*** (−5.80) − 0.0324*** (−3.87) − 0.0174** (−2.13) − 0.0303*** (−3.58) In . de la o , lag − 0.00251* (−1.66) − 0.0032 (− 1.42) −0.0021 (−0.91) Female educ., lag 0.00284 (0.60) 0.00890 (1.54) 0.00805 (1.40) 0.0068*** (2.61) 0.0022 (0.59) 0.0059** (2.20) Male educ., lag 0.00123 (0.27) − 0.00899 (− 1.35) −0.00664 (−1.03) In la ion 0.0008** (2.13) 0.0011*** (2.81) 0.0006 (1.49) 0.0009*** (3.57) 0.0009*** (2.79) 0.0007** (2.14) 123 780 G. A. Ma e o, L. Se én Table 15 (con inued) M1. Skele on model M2. Ex ended wi h educa ion and in . p ices M3. Ex ended wi h policy a iables M4. Ex ended wi h policy and in as uc u es T ade openness (log) 0.0202* (1.79) 0.0372*** (4.17) 0.0159* (1.65) Go . size (log) 0.0199 (1.52) 0.00859 (0.88) 0.00910 (0.97) In as uc u e, lag 0.0165** (2.32) 0.0175*** (2.61) 0.0143* (1.90) m2- es (p alue) 0.147 0.313 0.332 0.0710 0.179 0.115 0.236 0.427 0.397 0.226 0.179 0.262 AR(3) (p alue) 0.832 0.527 0.575 0.817 0.837 0.955 0.631 0.544 0.532 0.802 0.887 0.804 Hansen (p alue) 0.00333 0.000907 0.0517 0.0972 0.0428 0.114 0.118 0.0331 0.293 0.266 0.286 0.499 Di -Hansen, le els (p alue) 0.135 0.108 0.636 0.463 0.413 0.470 0.746 0.462 0.845 0.569 0.442 0.756 Num. obs 745 745 745 676 676 676 656 656 656 477 477 477 Num. coun ies 156 156 156 131 131 131 147 147 147 88 88 88 Num. ins umen s 40 40 55 85 85 100 82 82 97 82 82 97 See No e Table 4. Es ima ions a e done using 2-s ep sys em GMM (all lags s a ing a −3), bu collapsing he ma ix o ins umen s. The ins umen se s a s a −3. The di e ence Hansen es assesses he alidi y o he ins umen s o he le el equa ion in sys em GMM. Robus s a is ics in pa en heses. ***deno es signi icance a 1%, **a 5%, *a 10% 123 G ow h, inequali y and po e y: a obus ela ionship? 781 Table 16 G ow h, po e y and inequali y: sys em GMM es ima es (collapse, educe) M1. Skele on model M2. Ex ended wi h educa ion and in . p ices M3. Ex ended wi h policy a iables M4. Ex ended wi h policy and in as uc u es P0, lag − 0.183*** (−5.21) − 0.167*** (−4.98) − 0.143*** (−5.55) − 0.141*** (−5.96) − 0.131*** (−2.82) − 0.126*** (−3.59) − 0.0888*** (−3.31) − 0.0849*** (−3.40) Gini, lag − 0.161* (− 1.65) −0.0798 (−0.96) −0.136 (− 1.54) −0.0411 (−0.70) −0.0575 (−1.50) − 0.0972** (−2.19) 0.0447 (0.86) −0.0322 (−0.91) log y,lag − 0.0306*** (−5.34) 0.00134 (0.29) − 0.0299*** (−5.03) − 0.0296*** (−5.08) 0.00413 (0.78) − 0.0280*** (−5.10) − 0.0255*** (−3.80) − 0.00586* (−1.82) − 0.0260*** (−4.04) − 0.0449*** −0.0218 (−1.63) − 0.0431*** (−5.02) (−4.50) In . de la o , lag 0.000917 (0.26) 0.00168 (0.36) 0.00159 (0.39) Female educ., lag 0.00563 (0.60) 0.00250 (0.29) 0.00329 (0.37) 0.00757** (2.06) −0.00199 (−0.39) 0.00654* (1.85) Male educ., lag −0.00274 (−0.30) − 0.00947 (− 1.00) −0.00197 (−0.22) In la ion 0.0009** (1.96) 0.0007 (0.89) 0.0003 (0.34) 0.0013** (2.33) 0.0017* (1.90) 0.0014** (2.44) T ade openness (log) 0.0199 (1.29) 0.0385*** (3.86) 0.0179 (1.43) 123 788 G. A. Ma e o, L. Se én Re e ences Acemoglu D, Johnson S, Robinson JA, Ya ed P (2008) Income and democ acy. Am Econ Re 98:808–842 Ageno R, Aizenman J (2011) Aid ola ili y and po e y aps. J De Econ 91:1–7 Aghion P, Akcigi U, Be geaud A, Blundell R, Hemous D (2019) Inno a ion and op income inequali y. Re Econ S ud 86(1):1–45 Aiya S, Ebeke C (2020) Inequali y o oppo uni y, inequali y o income, and economic g ow h. Wo ld De 136:105–115 Alesina A, Rod ik D (1994) Dis ibu i e policies and economic g ow h. Q J Econ 109:465–490 Ang is JD, Pischke JS (2009) Mos ly ha mless econome ics: an empi icis ’s companion. P ince on Uni- e si y P ess, P ince on (US) A ellano M, Bond S (1991) Some es s o speci ica ion o panel da a: mone ca lo e idence and an appli- ca ion o employmen equa ions. Re Econ S ud 58:277–297 A ellano M, Bo e O (1995) Ano he look a he ins umen al a iable es ima ion o e o -componen s models. J Econom 68:29–52 Aza iadis C, D azen A (1990) Th eshold ex e nali ies in economic de elopmen . Q J Econ 105:501–526 Aza iadisC, S achu ski J (2005) Po e y aps. In: Aghion P, Du lau S (eds) Handbook o economic g ow h. Else ie , Ams e dam Bal agi BH, Li D (2002) Se ies es ima ion o pa ially linea panel da a models wi h ixed e ec s. Ann Econ Finance 3:103–116 Bandie a O, Bu gess R, Das N, Gulesci S, Rasul I, Sulaiman M (2017) Labo ma ke s and po e y in illage economies. Q J Econ 20:1–60 Bane jee A (2000) The wo po e ies. Massachuse s Ins i u e o Technology, Depa men o Economics WP 01-16 Bane jee A, Du lo E (2003) Inequali y and g ow h: wha can he da a say? J Econ G ow h 8:267–299 Bane jee A, Mullaina han S (2010) The shape o emp a ion: implica ions o he economic li es o he poo . NBER Wo king Pape 15973, Na ional Bu eau o Economic Resea ch, Camb idge, MA Bane jee A, Newman A (1993) Occupa ional choice and he p ocess o de elopmen . J Poli Econ 101(2):274–298 Bane jee A, Du lo E, Goldbe g N, Ka lan D, Osei R, Pa ien e W, Shapi o J, Thuysbae B, Ud y C (2015) A mul i ace ed p og am causes las ing p og ess o he e y poo : e idence om six coun ies. Science 348(6236):1260799–1260799 Ba o R (2000) Inequali y and g ow h in a panel o coun ies. J Econ G ow h 5:5–32 Ba o R, Lee JW (2013) A new da a se o educa ional a ainmen in he wo ld, 1950–2010. J De Econ 104:184–198 Bazzi S, Clemens MA (2013) Blun ins umen s: a oiding common pi alls in iden i ying he causes o economic g ow h. Am Econ J Mac oecon 5(2):152–186 Be g A, Os y JD, Tsanga ides CG, Yakhshiliko Y (2018) Redis ibu ion, inequali y, and g ow h: new e idence. J Econ G ow h 23:259–305 Be gs om KA (2020) The ole o inequali y o po e y educ ion. Policy Resea ch Wo king Pape 9409, Wo ld Bank, Washing on, DC Blau BM (2018) Income inequali y, po e y, and he liquidi y o s ock ma ke s. J De Econ 130:113–126 Bluhm R, de C omb ugghe D, Szi mai A (2018) Po e y accoun ing. Eu Econ Re 104(C):237–55 Blundell R, Bond S (1998) Ini ial condi ions and momen es ic ions in dynamic panel da a models. J Econom 87:115–143 Bou guignon F (2003) The g ow h elas ici y o po e y educ ion: explaining he e ogenei y ac oss coun ies and ime pe iods. Inequali y and g ow h: heo y and policy implica ions. Wo king Pape 28104, WB, Washing on, DC Bou guignon F (2004) The po e y-g ow h-inequali y iangle. Wo king pape 125, Indian Council o Resea ch on In e na ional Economic Rela ions (ICRIER), New Delhi Bowles S, Du lau SN, Ho KR (2006) Po e y aps. P ince on Uni e si y P ess, P ince on B ueckne M, Lede man D (2018) Inequali y and economic g ow h: he ole o ini ial income. J Econ G ow h 23(3):341–366 B ueckne M, Dabla No is E, G ads ein M (2015) Na ional income and i s dis ibu ion. J Econ G ow h 20(2):149–175 Bun M, Sa a idis V (2015) Dynamic panel da a models. In: Bal agi BH (ed) The Ox o d handbook o panel da a. Ox o d Uni e si y P ess, Ox o d, pp 76–110 123 G ow h, inequali y and po e y: a obus ela ionship? 789 Bun M, Windmeije F (2010) The weak ins umen p oblem o he sys em GMM es ima o in dynamic panel da a models. Econom J 13(1):95–126 Calde ón C, Mo al de Beni o E, Se én L (2015) Is in as uc u e capi al p oduc i e? A dynamic he e oge- neous app oach. J Appl Econom 30(2):177–198 Ce a V, Lama R, Loayza NV (2021) Links be ween g ow h, inequali y, and po e y: a su ey. Policy esea ch wo king pape 9603. Wo ld Bank G oup Ce a V, Eicheng een B, El-Ganainy A, Schindle M (2021) How o achie e inclusi e g ow h. Ox o d Uni e si y P ess and IMF ( o hcoming) Dasgup a P, Ray D (1986) Inequali y as a de e minan o malnu i ion and unemploymen : heo y. Econ J 96(384):1011–1034 Deininge K, Squi e L (1998) New ways o looking a old issues: inequali y and g ow h. J De Econ 57(2):259–287 Dolla D, K aay A (2002) G ow h is good o he poo . J Econ G ow h 7:195–225 Dolla D, Kleinebe g T, K aay A (2016) G ow h s ill is good o he poo . Eu Econ Re 81:68–85 Du lau S (2006) G oups, social in luences, and inequali y. In: Bowless S, Du lau S, Ho K (eds) Po e y aps. P ince on Uni e si y P ess, P ince on Eas e ly W (2006) Reli ing he 1950s: he big push, po e y aps, and akeo s in economic de elopmen . J Econ G ow h 11:289–318 Enge man S, Sokolo K (2006) The pe sis ence o po e y in he Ame icas: he ole o ins i u ions. In: Bowless S, Du lau S, Ho K (eds) Po e y aps. P ince on Uni e si y P ess, P ince on E man L, e Kaa DM (2019) Inequali y and g ow h: indus y-le el e idence. J Econ G ow h 24:283–308 Fe ei a FHG, Lei e PG, Ra allion M (2010) Po e y educ ion wi hou economic g ow h? Explaining B azil’s po e y dynamics, 1985–2004. J De Econ 93(1):20–36 Fe ei a FHG, Chen S, Dabalen A e al (2016) A global coun o he ex eme poo in 2012: da a issues, me hodology and ini ial esul s. J Econ Inequal 14(2):141–172 Fo bes K (2000) A eassessmen o he ela ionship be ween inequali y and g ow h. Am Econ Re 90:869–887 Fosu AK (2017) G ow h, inequali y, and po e y educ ion in de eloping coun ies: ecen global e idence. Res Econ 71(2):306–336 Fu ce i D, Loungani P, Os y JD, Pizzu o P (2020) Will Co id-19 a ec inequali y? E idence om pas pandemics. Co id Econ 12:138–157 Galo O (2009) Inequali y and economic de elopmen : an o e iew. WP 2009-3, B own Uni e si y, Depa - men o Economics Galo O, Moa O (2004) F om physical o human capi al accumula ion: inequali y and he p ocess o de elopmen . Re Econ S ud 71:1001–1026 Galo O, Zei a J (1993) Income dis ibu ion and mac oeconomics. Re Econ S ud 60(1):35–52 Gib a R (1931) Les Inégali és Économiques. Si ey, Pa is GoldinC,Ka z LF(2008) The acebe ween educa ionand echnology. Ha a d Uni e si yP ess,Camb idge Haide LJ, Boons a WJ, Pe e son GD, Schlü e M (2018) T aps and sus ainable de elopmen in u al a eas: a e iew. Wo ld De 101:311–321 Hal e D, Oechslin M, Zweimulle J (2014) Inequali y and g ow h: he neglec ed ime dimension. J Econ G ow h 19(1):81–104 Hanushek EA (2017) Fo long e m economic de elopmen , only skills ma e . IZA Wo ld o Labo 343 IMF (2020) Wo ld economic ou look, Oc obe 2000 Kalwij A, Ve schoo A (2007) No by g ow h alone: he ole o he dis ibu ion o income in egional di e si y in po e y educ ion. Eu Econ Re 51(4):805–829 Kleibe gen F, Paap R (2006) Gene alized educed ank es s using he singula alue decomposi ion. J Econom 133(1):97–126 Knowles S (2005) Inequali y and economic g ow h: he empi ical ela ionship econside ed in he ligh o compa able da a. J De S ud 41:135–159 K aay A (2006) When is g ow h p o-poo ? E idence om a panel o coun ies. J De Econ 80(1):198–227 K aay A (2015) Weak ins umen s in g ow h eg essions: implica ions o ecen c oss-coun y e idence on inequali y and g ow h. Policy Resea ch Wo king Pape No. 7494. The Wo ld Bank K aayA,McKenzieD(2014)Dopo e y apsexis ?Assessing hee idence.JEconPe spec 28(3):127–148 K usell P, Ohanian LE, Ríos-Rull JV, Violan e GL (2000) Capi al-skill complemen a i y and inequali y: a mac oeconomic analysis. Econome ica 68(5):1029–1054 Kuzne s S (1955) Economic g ow h and income inequali y. Am Econ Re 45(1):1–28 123 790 G. A. Ma e o, L. Se én La Fe a a E (2019) P esiden ial add ess: aspi a ions, social no ms, and de elopmen . J Eu Econ Assoc 17(6):1687–1722 Lakne C, Ge szon Mahle D, Neg e M, Bee P ydz E (2020) How much does educing inequali y ma e o global po e y? Global Po e y Moni o ing Technical No e 13, Wo ld Bank, Washing on, DC Lakne C, Yonzan N, Mahle D, Cas aneda R, Wu H (2021) Upda ed es ima es o he impac o COVID-19 on global po e y: looking back a 2020 and he ou look o 2021. h ps://blogs.wo ldbank.o g/ Li H, Zou H (1998) Income inequali y is no ha m ul o g ow h: heo y and e idence. Re De Econ 2(3):318–334 Liu Z, S engos T (1999) Non-linea i ies in c oss-coun y g ow h eg essions: a semipa ame ic app oach. J Appl Econom 14:527–538 Loayza NV, Radda z C (2010) The composi ion o g ow h ma e s o po e y alle ia ion. J De Econ 93(1):137–151 Loayza N, Fajnzylbe P, Calde ón C (2005) Economic g ow h in La in Ame ica and he ca ibbean: s ylized ac s, explana ions and o ecas s. The Wo ld Bank, Washing on López H, Se én L (2006) A no mal ela ionship? Po e y, g ow h and inequali y. Policy Resea ch WP S3814, Wo ld Bank LópezH, Se énL (2015) Too poo o g ow. Chap e 13 in economic policiesin eme ging-ma ke economies es sch i in Hono o Vi o io Co bo. In: Caballe o RJ, Schmid -Hebbel K (eds) 1s edn, ol 1, no 21. Cen al Bank o Chile, pp 309–50 Ma e o G, Rod íguez JG (2013) Inequali y o oppo uni y and g ow h. J De Econ 104:107–122 Ma e o GA, Se én L (2018) Po e y, inequali y and g ow h: a obus ela ionship? Policy Resea ch WP Se ies 8578, Wo ld Bank Mookhe jee D, Ray D (2002) Con ac ual s uc u e and weal h accumula ion. Am Econ Re 92:818–849 Nelson C, S a z R (1990) The dis ibu ion o he ins umen al a iables es ima o and i s - a io when he ins umen is a poo one. J Bus 63(1):S125–S140 Newson R (2000) sg151: B-splines and splines pa ame e ized by hei alues a e e ence poin s on he x- axis. S a a Tech Bull 57:20–27. Rep in ed in S a a Tech Bull Rep in s 10. S a a P ess, College S a ion, TX, pp 221–230 Owen D, Knowles S, Lo gelly PK (2002) A e educa ional gende gaps a b ake on economic de elopmen ? Some c oss-coun y empi ical e idence. Ox Econ Pap 54(1):118–149 Panizza U (2002) Income inequali y and economic g ow h: e idence om Ame ican da a. J Econ G ow h 7:25–41 Pa idge MD (1997) Is Inequali y ha m ul o g ow h? Commen . Am Econ Re 87(5):1019–1032 Pe o i R (1996) G ow h, income dis ibu ion and democ acy. J Econ G ow h 1:149–187 Pesa an MH (2021) Gene al diagnos ic es s o c oss-sec ional dependence in panels. Empi Econ 60:13–50 Pinko skiy M, Sala-i-Ma in X (2013) Wo ld wel a e is ising: es ima ion using nonpa ame ic bounds on wel a e measu es. J Public Econ 97(C):176–195 Quah D (1993) Empi ical c oss-sec ion dynamics in economic g ow h. Eu Econ Re 37:426–434 Ra allion M (2004) P o-poo g ow h: a p ime . Policy Resea ch Wo king Pape 3242, WB Ra allion M (2005) A po e y-inequali y ade o ? J Econ Inequal 3:169–181 Ra allion M (2012) Why don’ we see po e y con e gence? Am Econ Re 102(1):504–523 Roodman D (2009) A no e on he heme o oo many ins umen s. Ox Bull Econ S a 71(1):135–158 Sala-i-Ma in X (2006) The wo ld dis ibu ion o income: alling po e y and… con e gence, pe iod. Q J Econ 121(2):351–397 Sande son E, Windmeije F (2016) A weak ins umen F- es in linea IV models wi h mul iple endogenous a iables. J Econom 190:212–221 Seh awa M, Gi i AK (2018) The impac o inancial de elopmen , economic g ow h, and income inequali y on po e y: e idence om India. Empi Econom 55(3):1585–1602 Shah A, Mullaina han S, Sha i E (2012) Some consequences o ha ing oo li le. Science 338(6107):682–685 Sianesi B, Reenen JV (2003) The e u ns o educa ion: mac oeconomics. J Econ Su 17(2):157–200 S igli zJ(1969)Thee ec so income,weal handcapi algains axa ionon isk- aking. Q JEcon83:263–283 S igli z J (2012) The p ice o inequali y: how oday’s di ided socie y endange s ou u u e. W.W. No on and Company, New Yo k S ock J, Yogo M (2005) Tes ing o weak ins umen s in linea IV eg ession. In: Iden i ica ion and in e ence o econome ic models: essays in Hono o Thomas Ro henbe g, ol 5, pp 80–108 UNU-WIDER (2008) Wo ld income inequali y da abase, Ve sion 2.0c, May 2008 123 G ow h, inequali y and po e y: a obus ela ionship? 791 Van de Weide R, Milano ic B (2018) Inequali y is bad o g ow h o he poo (bu NOT FOR THAT OF THE RICh). Wo ld Bank Econ Re 32(3):507–530 Voi cho sky S (2005) Does he p o ile o income inequali y ma e o economic g ow h? Dis inguish- ing be ween he e ec s o inequali y in di e en pa s o he income dis ibu ion. J Econ G ow h 10:273–296 Voi cho sky S (2011) Inequali y and g ow h. In: Nolan B, Sal e da W, Smeeding T (eds) The Ox o d handbook o economic inequali y. Ox o d Uni e si y P ess, Ox o d Windmeije F(2005)A ini esampleco ec ion o he a ianceo linea e icien wo-s epGMMes ima o s. J Econom 126(1):25–51 Publishe ’s No e Sp inge Na u e emains neu al wi h ega d o ju isdic ional claims in published maps and ins i u ional a ilia ions. 123