The Role of Sectoral Growth Patterns in Labor Market Development

Transcription

1 The Role of Secoral Growh Paerns in Labor Marke Developmen Absrac: This paper invesigaes he relaionship beween secoral growh paerns and employmen oucomes. A broad cross-counry analysis reveals ha in middle-income counries, employmen responds more o growh in less producive and more labor-inensive secors. Employmen in middle-income counries is suscepible o a resource curse, and grows rapidly in response o manufacuring and expor manufacuring growh. Wihin Brazil, Indonesia, and Mexico, he effecs of differen secoral growh paerns are conex dependen, bu differences in secoral growh effecs on employmen and wages are subsanially reduced in saes or provinces wih higher measured labor mobiliy. Consisen wih his, aggregae employmen and wage effecs of growh by secor are close o uniform when examined over longer ime horizons, afer labor has an opporuniy o adjus across secors. The resuls reinforce he imporance of growh in more labor-inensive secors, and sugges ha job mobiliy may be an imporan mechanism o diffuse he benefis of capial-inensive growh. Francisco Javier Arias-Vazquez Jean N. Lee David Newhouse* World Bank, Social Proecion and Labor, 1818 H Sree, Washingon DC, *Corresponding auhor: dnewhouse@worldbank.org JEL codes: O11, J20 Keywords: Economic Growh, Producion Srucure, Employmen, Unemploymen, Produciviy 1

2 1. Inroducion Developing counries are ypically characerized by large and persisen produciviy and earnings differences across secors, wih large numbers of workers oiling in low-produciviy jobs in agriculure or small-scale rading. The reallocaion of hese workers ino more producive employmen plays a criical role in increasing produciviy, promoing living sandards, and ulimaely lifing people ou of povery. This is consisen wih he longsanding view ha long-erm improvemen in incomes and reducions in income inequaliy require a srucural ransformaion ou of agriculure ino more producive secors, where wages are higher (Lewis, 1954; Chenery and Syrquin, 1975; Foser and Rosenzweig, 2007). Because he reallocaion of workers o more producive secors is necessary for developmen, boh researchers and policymakers are ofen keenly ineresed in he implicaions of differen paerns of secoral growh. Mos discussion, based on he Eas Asian experience, focuses on he imporance of expor-oriened manufacuring growh as an engine of employmen and produciviy growh (Balassa, 1978; Amii and Freund, 2010). This focus is a leas parially consisen wih a recen sudy, based on 55 counries, which finds ha growh in agriculure, consrucion, and noably manufacuring lead o he mos rapid povery reducion (Loayza and Raddaz, 2010). In his case, growh in secors where he poor can easily be employed appears o benefi hem mos. Oher sudies, however, have emphasized he implicaions of growh in oher economic secors, such as naural resources (Sachs and Warner, 1995; Sachs and Warner, 1999; Robinson e al, 2006). McMillan and Rodrik (2012), for example, consider 38 counries, 25 of which are low or middle-income, and find evidence for a naural resource curse refleced in employmen paerns in hese counries. Counries wih a higher share of expors in naural resources were more likely o experience growh-reducing changes in labor allocaion across secors, because he high produciviy resource secor was unable o absorb large amouns of labor. This is consisen wih he resuls from wo naural experimens, one following an increase in demand for coca pase in rural Columbia and he oher following he discovery of oil fields in Brazil, ha also sugges ha naural resource booms lead o disappoining employmen and growh oucomes. 1 Anoher sudy finds evidence for imporan increases in civil conflic in oil producing regions of Colombia as oil prices rise, and furher relaes he ramificaions of commodiy price shocks for poliical sabiliy o he labor inensiy and employmen effecs of increased producion, sudying price shocks o oil, coal, gold, coffee, bananas, sugar, palm and obacco (Dube and Vargas, forhcoming). Noneheless, a recen horough review highlighs he pauciy of evidence from panel-daa and quasi-experimenal sudies on he economic effecs of he resource curse. 2 Ye anoher srand of he lieraure has sressed he role of agriculural growh in promoing srucural change and reducing povery (Sadoule and de Janvry, 2010; Chrisiaenson e al, 2011). One sudy, for example, repors he resuls of model simulaions in which growh in agriculural produciviy hasens indusrializaion and is herefore cenral o developmen. 3 A subsequen sudy, however, suggess ha agriculural growh may be less imporan han non-agriculural growh in low-income seings, and ohers emphasize he imporance of he growh of nonfarm indusries for povery reducion even in rural areas. 4 Empirical evidence from China, India, and Indonesia indicaes ha agriculural growh has srong effecs on povery reducion in some conexs (Ravallion and Chen, 2007; Ravallion and Da, 1996; Suryahadi e al, 2010). 1 Angris and Kugler (2008), and Caselli and Michaels (2009). 2 Van Der Ploeg (2011), Haggblade e al (2002), Headey e al (2010), and Lanjouw and Lanjouw (2001). See also Ferreira e al (2010) for evidence on service secor growh and povery reducion in Brazil. 3 Gollin, e al (2002). 4 Alvarez Cuadrado and Poschke (2011). 2

3 In shor, he published evidence on secoral growh and povery generally agrees ha growh in labor-inensive secors such as manufacuring, consrucion, and agriculure may mos benefi he poor, while naural resource growh fails o spur produciviy increases for he majoriy of workers. However, his evidence varies considerably in differen conexs and is far from definiive, and he lieraure demonsraes no clear consensus on he role of secoral growh paerns in improving labor marke oucomes and hus reducing povery. The lack of clear evidence on he link beween growh and employmen paerns is surprising, given he srong inerlinkage beween employmen saus and povery. 5 Policymakers percepions regarding secoral growh paerns and employmen can have imporan implicaions for growh sraegies as well as policies owards rade, exchange raes, and indusrial developmen. This paper brings new evidence o bear on how changes in labor marke oucomes, such as employmen raes, unemploymen raes, wages, and wage and non-agriculural employmen, produciviy, vary wih secoral growh paerns. We focus on wheher growh in low produciviy secors or higher-produciviy secors generaes greaer growh in wages and employmen. While broad labor marke indicaors such as employmen are of ineres in heir own righ, more specific indicaors such as wages, produciviy, and ypes of employmen are closely relaed o povery and economic developmen, bu unforunaely less frequenly observed or analyzed. This analysis akes advanage of he wealh of labor marke daa available in he World Bank s Inernaional Income Disribuion Daabase (I2D2) and in several counry-specific labor marke surveys o include hese more specific indicaors in our empirical exploraion of he employmen consequences of secor-specific growh. 6 We begin wih a cross-counry analysis in which hese daa are combined wih UN counry-level daa on growh in seven secors o esimae he impac of growh in differen secors on employmen growh, wage growh and produciviy growh. The resuling daabase of growh paerns and labor marke oucomes includes 184 surveys in 81 counries. The cross-counry analysis is supplemened wih hree counry case sudies from Brazil, Indonesia and Mexico. These case sudies uilize annual and rerospecive daa from labor force and household surveys, and use saeand province-level variaion in secoral growh raes o idenify he relaionship beween secoral growh raes and labor marke oucomes. A unique feaure of hese hree counries is he availabiliy of panel daa on workers, which allows us o examine how he effec of differen secoral growh paerns is mediaed by workers mobiliy across jobs. When looking across counries a he shor-run effecs of secoral growh, growh in less producive secors, compared o more producive secors, increases employmen and reduces unemploymen. Growh in mining and uiliies, in conras, is associaed wih increased unemploymen and reduced employmen, providing some empirical suppor o he noion of a resource curse in employmen oucomes. Meanwhile, manufacuring growh is associaed wih a subsanial increase in employmen and decline in unemploymen among middleincome counries. Secoral growh has surprisingly few noiceable impacs on oher job characerisics ha are imporan indicaors of economic developmen, such as earnings, secoral produciviy, and he srucural ransformaion ou of agriculure and ino wage employmen. The analysis, however, lacks sufficien saisical power o rule ou poenially large effecs of secoral growh paerns on hese indicaors. The wihin-counry analyses of recen growh paerns in Brazil, Indonesia and Mexico do no conform o his or any oher consisen paern. In Brazil, employmen oucomes are comparable for high and low produciviy secor growh. In Indonesia, in conras, high produciviy secor growh reduces employmen effecs and increases 5 See Guierrez e al (2007). 6 For more informaion on an earlier version of he I2D2 daabase, see Monenegro and Hirn (2009). 3

4 wage growh. In Mexico, growh high produciviy secors are associaed wih sronger employmen growh, bu, if anyhing, slower growh in wages and labor produciviy. A he one digi level, employmen rises wih manufacuring growh in Brazil and Mexico, which is in line wih he cross-counry finding ha manufacuring growh increases employmen in middle-income counries. The Indonesian daa also show some evidence of a ype of resource curse, as mining growh is associaed wih saisically significanly slower employmen and wage growh. In general, however, growh paerns a he 1-digi secor level fail o show a consisen paern across he hree counries. Employmen is boosed more by agriculure growh in Indonesia and financial secor growh in Mexico. Meanwhile, wage increases are associaed wih growh in manufacuring in Indonesia, bu growh in he public and financial secors and in Mexico. In analyses of growh over longer periods using census daa from hese counries, we find lile evidence for differenial employmen or wage effecs of growh across secors over longer horizons, when echnologies and labor markes have an opporuniy o adjus, wih srikingly similar poin esimaes on agriculure, indusry and services growh. Consisen wih his differenial beween shor- and long-run effecs, we find ha higher measured labor mobiliy srongly aenuaes he differences beween he growh effecs of high and low produciviy secors in all hree counries in our shorer-run analyses. 7 Overall, he resuls confirm some commonly held preconcepions, while challenging ohers. In middle income counries, less producive secors are more effecive in increasing employmen, manufacuring growh increases employmen and reduces unemploymen, and naural resource growh does he opposie. However, here is a noable lack of evidence ha growh paerns affec wages, wage employmen, non-agriculural employmen, and produciviy, alhough mos of hese esimaed effecs are admiedly imprecise and canno rule ou large effecs. Agriculural growh also has surprisingly minor effecs on aggregae labor marke oucomes, even in low-income counries. Finally, he differenial effecs of secoral growh are likely o be more significan over he shorer run, when labor is relaively immobile. The resuls sugges ha i may be appropriae for longer run employmen generaion and povery reducion sraegies o maximize overall growh, raher han prioriizing growh in specific secors. In he nex secion we describe he daa and provide summary saisics and background informaion abou he Brazilian, Indonesian and Mexican economies and labor marke srucures. Secion 3 describes he empirical mehods. In secion 4, we examine labor marke oucomes and how hey respond o year-on-year shocks o secoral growh raes, in boh he cross-secion and he hree counries. In secion 4, we also discuss secors ha ofen receive special aenion in boh he lieraure and popular discussion, such as agriculure, expor manufacuring, and mining and naural resource exracion. In secion 5, we focus exclusively on he hree case sudies and examine he role of labor mobiliy and longer-erm esimaes of secoral growh impacs. Secion 6 concludes. 2. Daa and Background 7 Our resul suppors he noion ha as in Topalova (2008), differenial labor marke mobiliy may affec he disribuional implicaions of growh paerns. 4

5 To relae secoral growh paerns o labor marke oucomes, we firs obained informaion on GDP by secor a he naional level from he Unied Naions. The UN GDP daa are disaggregaed according o he following secors: agriculure, huning, foresry, and fishing (ISIC A-B); mining, manufacuring and uiliies (ISIC C-E), manufacuring (ISIC D); consrucion (ISIC F); wholesale, reail rade, resaurans and hoels (ISIC G-H); ranspor, sorage and communicaion (ISIC I); and oher aciviies (ISIC J-P). 8 We also use informaion on GDP by secor and sae or province from Mexico, Indonesia, and Brazil. These cover mos of he following secors in each counry: agriculure, mining, consrucion, elecriciy, manufacuring, reail rade, ranspor and communicaion, finance and business services, social services, oher services and he public secor. Daa on manufacuring expors come from he World Bank s World Developmen Indicaors daabase, and daa on labor regulaion from he World Bank s Doing Business indicaors daabase. For informaion on labor marke oucomes a he counry level, we urn o he sandardized survey daa included in he World Bank s Inernaional Income Disribuion Daabase (I2D2). For he wihin-counry analyses of annual growh and changes in employmen oucomes, we primarily use daa from Brazil s Monhly Employmen Survey (PME) from 2003 o 2009; he 1997, 2000, and 2007 waves of he Indonesia Family Life Survey (IFLS); and he Mexican Naional Survey of Occupaion and Employmen (ENOE) from 2005 o 2010 o measure changes in labor marke oucomes. Brazil s Monhly Employmen Survey (PME) is a roaing panel survey, bu i only covers major urban areas in six saes of Brazil, making i he mos limied in scope of he counry-level daa sources we use. We herefore supplemen he PME wih informaion from 2003 o 2009 from he Pesquisa Nacional por Amosra de Domicilios (PNAD), which is conduced annually and broader in geographical coverage, bu is no a panel survey. 9 Unlike in Brazil, he Mexican labor force survey (ENOE) is represenaive of all 32 saes, includes a quarerly roaing panel of survey respondens, and is a roaing panel. We use he firs and las inerview quarers for each individual o capure annual changes in employmen saus. Finally, he Indonesian esimaes are obained from he Indonesia Family Life Survey (IFLS), which sampled households from 13 provinces represening 83 percen of he populaion in is iniial round in Unlike he Brazillian and Mexican daa, he IFLS is no a roaing panel and more han 90 percen of he original households were successfully re-conaced in 1997, 2000, and The survey also racks movers, and obains a complee rerospecive labor marke hisory spanning from 1988 o For some oucomes, specifically o disinguish beween unemploymen and nonparicipaion in he labor force, we supplemen his wih informaion from he Naional Labour Force Survey for Indonesia (SAKERNAS). While hese household surveys are valuable, largely because hey are longiudinal, hey do no allow us o examine how secoral growh paerns relae o labor marke changes in he long run. We herefore supplemen he analysis using public use census microdaa samples for Brazil in 1960, 1970, 1980, 1990 and 2000; for Indonesia in 1971, 1980, 1990, 2000 and 2010; and for Mexico in 1990, 2000 and In hese longer-run analyses, we are consrained by he availabiliy of secoral GDP daa a he sub-naional level o analyze he hree broad secors of agriculure, services and indusry. Workers are mached o sae- or province-specific daa on secoral gross domesic produc (GDP) from naional saisical offices, based on heir sae or province of residence. Using each of hese oucome daases, we consruc five main measures of workers labor marke oucomes: Wheher hey are employed, unemployed, or ou of he labor force, heir log wage, and he secor s log produciviy (value added per worker) in he secor of employmen. For he cross-counry analysis, his is also supplemened wih an analysis of wage employmen and non-agriculural employmen. 8 For he purposes of our analysis, we separae mining and uiliies (ISIC C and E) from manufacuring (ISIC D). 9 We use aggregaes a he sae and secor level o generae esimaes of average wage and produciviy growh from he PNAD. 10 Noe ha wage daa are only available hrough

6 Based on average labor produciviy esimaes from hese daa, indusries are characerized as being high produciviy if hey fall in he following se: manufacuring, ranspor and communicaions, finance, elecriciy and uiliies, or mining. We group he following indusries ino he low produciviy se: oher services, agriculure, reail and wholesale rade, governmen and public adminisraion, and consrucion. This classificaion was based on he ranking of secors according o heir produciviy. The annual panel daa cover periods in Indonesia and Mexico including wo recen major recessions: he Indonesian financial crisis of , and he more recen recession in Mexico following he global financial crisis, beginning in During boh of hese recessions, indusrial growh was hi he hardes: indusrial growh declines disproporionaely in Indonesia in 1997 and never recovered o pre-crisis levels, and declines more han agriculural and service secor growh in Mexico during he more recen recession here Labor marke mobiliy, as evidenced by ransiions across secors and ransiions ou of unemploymen, is very high in Mexico, wih roughly one in six workers moving secors in any given year. Labor marke ransiions repored in he Indonesian daa are markedly less frequen, occurring for up o 9 percen of workers in any given secor, alhough changes in indusry may be underrepored in hese rerospecive daa. Labor marke ransiions are also very common in Brazil s PME, wih approximaely 1 in 5 workers moving secors wihin a year, alhough his number only reflecs he rae of ransiions for urban workers. Mexico and Brazil s higher labor marke mobiliy relaive o Indonesia, if no an arifac of he naure of he daa, may perhaps be relaed o heir much more diverse economic srucures. Indonesia s economy in 2007 was more heavily agriculural han ha of he oher wo counries, wih 33 percen of workers employed in agriculure, one of he secors wih he lowes labor produciviy, relaive o Mexico s 13.5 percen share. 3. Empirical Mehods In our cross-counry analysis, counries are observed in differen years and a disparae inervals, and we herefore follow Loayza and Raddaz (2010) in calculaing annualized growh raes in employmen oucomes and in secorspecific GDP over hese inervals. We also weigh secoral growh by GDP share. Besides adjusing for he size of differen secors in he economy, his allows for ess of he equaliy of coefficiens across secors, as in Loayza and Raddaz (2010) and Ravallion and Chen (2007). As noed in hose papers, he case in which he coefficiens on differen share-weighed secors are saisically indisinguishable corresponds o he case ha only overall growh maers and he secoral composiion of growh does no. We regress hese annualized changes in employmen oucomes on share weighed GDP by secor: eˆ 7 ˆ c, 0 ksc, k, yc, k, k 1 (1) c, where e ˆc, is he annualized change of an employmen oucome in counry c a period, s c, k, is he GDP share of secor k in counry c a period, and ˆ is he growh rae of oupu in secor k in counry c a period. We es y c, k, for significan effecs of growh composiion on employmen oucomes by esing equaliy of 1 hrough 7. Unlike Loayza and Raddaz (2010), we are iniially ineresed in he effecs of shorer-run growh on employmen oucomes, o keep esimaes comparable o hose using annual growh and individual-level panel daa for Brazil, 6

7 Indonesia and Mexico. Therefore, he analysis uses every inerval in which growh and changes in employmen oucomes are observed o esimae he relaionship beween secoral growh paerns and employmen oucomes. Because here are muliple observaions per counry, we cluser sandard errors a he counry level. We hen resric he coefficiens o be equal among high and among low produciviy secors, and esimae he following varian of equaion (1): s ˆ ˆ c, k, yc, k, Low sc, k, yc, k, c. (2) eˆ c, 0 High High Low, To es for differences in effecs of growh of high and low produciviy secors, we es for equaliy of. and Expor-led manufacuring growh may have disincive effecs on labor markes. To es for his, in he crosscounry daa, we simply regress growh in employmen oucomes on oal GDP growh and growh in expor manufacuring, denoed xˆ : e ˆc, 0 yˆ c, xˆ c, (3) The specificaions for he counry-specific analysis using panel daa for Brazil, Indonesia, and Mexico are similar in spiri o he cross-counry analysis. Where possible, he specificaions include individual-level fixed effecs o empirically assess he relaionship beween secoral growh paerns and changes in employmen saus. Individual fixed effecs, raher han sae or province-level dummies, are included o improve efficiency, as hey affec consisency only in he Indonesian case. 11 We esimae he following specificaion: e i, r, 10 s y (4) 0 k 1 k r, k, r, k, i i, r, Where e i r,, is employmen oucome for person i in region (sae or province) r a ime, r, k is he GDP share of secor k in region r a ime, and y r k, s,, is oupu in secor k in region r a ime, and i is an individual fixed effec. In pracice, as we have an iniial and final observaion for each individual in he panel daa, we esimae his equaion in firs differences: e i, r, 10 k 1 ksr, k, yr, k, i, r, ~ ~ (5) Then, as wih he cross-counry daa, we resric coefficiens o be equal among high produciviy secors and among low produciviy secors, and es for differences in effecs of high and low produciviy secor growh 12 : 11 This is because only he Indonesian daa racks inerprovincial migrans. 12 For he specificaions described above, we also es for robusness o he inclusion of counry-, sae- or province-specific rends. 7

8 sr k yr k Low sr k yr k d ~,,,,,, ~,, i, r (6) e i, r, High High Low, The impac of secoral growh paerns may depend on worker mobiliy beween jobs. To assess his, we consruc sae- or province-level measures of labor mobiliy in Brazil, Indonesia, and Mexico, using he individual-level panel daa on indusry of employmen. Our primary measure is he average share of individuals in a sae who change indusries wihin one year. We also use he occupaional informaion o consruc an alernaive measure of labor mobiliy: he share of individuals in a sae who change occupaions wihin a year. The resuls are largely robus o use of his alernaive measure. e i, r, High HighLM s High r, k, yr, k, Low s Low r, k, yr, k, ~ sr, k, yr, k, * LM LowLM sr, k, yr, k, * LM ~ r i, r, High Low (7) For each counry, o es wheher he differenial beween he effecs of high and low produciviy growh changes wih flexibiliy in he labor marke, we regress changes in labor marke oucomes on he share weighed growh raes of high and low produciviy secors, ineraced wih hese labor mobiliy measures: where is he measure of average labor mobiliy a he sae or province level. We aemp o conduc an analogous es in he cross-counry daa. Because empirical informaion on labor marke ransiions are no available for a broad se of counries, we use measures of labor regulaion from he Doing Business daabase, despie he inconclusive evidence on he link beween labor regulaions and job eˆ c, mobiliy. 13 s yˆ s yˆ 0 HighLR High c, k, sc k yˆ,, c, k, * LRc LowLR sc k yˆ,, c, k, * LRc c, High High c, k, Low Low Low c, k, c, k, (8) Finally, we urn o census daa for Brazil, Indonesia and Mexico o assess he longer run effecs of secoral growh paerns on employmen oucomes. We esimae: 13 For example, Kugler (2004) and Saavedra and Torrero (2004) found ha loosening employmen proecion legislaion led o a moderae increase in job separaions and hiring in Colombia and Peru, bu Paes de Barros and Corseuil (2004) and Perin and Sivadasan (2006) find lile evidence ha increased regulaions decreased mobiliy in Brazil or Chile. 8

9 e i, r, 3 s y (9) 0 k 1 k r, k, r, k, r i, r, where are sae or province fixed effecs and are year effecs and secors include agriculure, services and indusry. 4. Secoral growh impacs on employmen 4.1 High and low produciviy secor growh Is growh in high or low-produciviy secors more srongly associaed wih employmen creaion? Before addressing his quesion, we firs pu changes in employmen and unemploymen ino conex by comparing how hey evolve wih per capia GDP. Figure 1a shows a smoohed average of counries employmen o populaion raio a differen levels of per capia GDP. For low-income counries, employmen falls rapidly as per capia GDP increases, as a larger share of he populaion can afford o exi he labor force. This fall is dramaic, from abou wo hirds of he populaion for he poores counries, o a low of abou half for lower middle-income counries, a abou $2000 per capia. Employmen hen gradually rises, o abou 60 percen for he riches counries. Figure 1b ells a similar sory in reverse. Average unemploymen raes sar a abou 12 percen and rise o jus over 15 percen, as more workers can afford o search, before falling again o 15 percen. These figures illusrae wo key poins. Firs, for low-income counries, employmen declines and unemploymen increases may arguably be a welcome sign of economic growh and developmen. Secondly, even a relaively small employmen increase or unemploymen decline, on he order of one half o one percenage poin, consiues an imporan employmen improvemen in middle and high-income counries. Alhough he inerpreaion of employmen changes depends heavily on he counry s level of developmen, we begin firs by presening aggregae resuls in Table 2. These are aken from Equaion (2), which esimaes how annualized growh raes of employmen relae o share-weighed secoral growh raes. They show moderae differences in he effecs of high and low produciviy secor growh on employmen and unemploymen. In a sample of 81 counries wih mached growh and employmen oucome daa, low produciviy growh appears o be associaed wih significanly faser growh in employmen and reducions in unemploymen relaive o high produciviy growh. However, if anyhing, high produciviy growh may be associaed wih faser wage and labor produciviy growh, alhough he differences are no saisically significan. In erms of magniudes, a one percenage poin increase in overall GDP due o high-produciviy secors raises employmen, as a share of he populaion, by 0.2 percenage poins (significan a he 5 percen level), while low produciviy growh is associaed wih an increase of 0.08 percenage poins (no saisically differen from zero), and he difference ( is marginally saisically significan a he 10 percen level (es p-value = 0.075). This suggess ha on average, low produciviy growh leads o faser employmen growh han does high produciviy growh. Similarly, low produciviy secor growh appears o be associaed wih faser declines in unemploymen. High produciviy growh is associaed wih a coefficien of 0.067, meaning ha a one percenage 9

10 poin increase in GDP growh due o he producive secors is associaed wih a roughly 0.07 percenage poin increase in he unemploymen rae. Meanwhile, a comparable increase in he growh of low-produciviy secors is associaed wih a reducion of 0.16 percenage poins (significan a he 1 percen level). The difference ( is significan a he 5 percen level (es p-value = 0.021). Wih respec o inaciviy, growh in more producive secors may marginally raise he share of populaion ha is ou of he labor force, wih a poin esimae of 0.149, bu he difference in coefficiens for high and low produciviy growh is no significan (. In conras o hese resuls on employmen growh, i appears ha if anyhing, high produciviy secor growh may lead o faser wage and labor produciviy growh relaive o low produciviy secor growh, alhough he coefficiens are imprecisely esimaed and he differences beween coefficiens are no saisically significan. 14 As noed above, he inerpreaion of employmen and unemploymen changes depend on counries sage of developmen, as declines in employmen and increases in unemploymen ofen reflec economic growh in lowincome counries. Figures 2, 3 and 4 show how he esimaed effecs of growh in high and low produciviy secors vary according o counries per capia GDP. For low-income counries, high-produciviy secors are associaed wih a slighly greaer reducion in employmen, bu differences wih less producive secors is no saisically significan, as he op dashed blue line in figure 2 exceeds he lower dashed red line. The bigger sory, however, is he posiive effec of low-produciviy growh on employmen for middle-income counries, paricularly hose in he $2000 o $10,000 range. Here, a 1 percen increase in aggregae growh due o more producive secors is associaed wih an increase in employmen of half a percenage poin, while growh in more producive secors is associaed wih a sligh decline. The unemploymen resuls also show a beneficial effec of less producive secors in middle-income counries. For low income counries, growh in more producive secors is associaed wih a slighly greaer increase in unemploymen, bu he differences wih less producive secors are relaively minor. Esimaed impacs on unemploymen sar o diverge a jus under $2000 per capia, as growh in less producive secors begins o have a subsanial negaive and saisically significan effec; a is peak, a one percenage poin increase in growh due o less producive secors lowers unemploymen raes by half a percenage poin. To sum up, employmen and unemploymen in middle-income counries responds favorably o growh in less producive, more labor-inensive, secors. Turning o our case sudy counries of Brazil, Indonesia and Mexico, we find ha he effecs of high and low produciviy secor growh are no uniform across conexs, and may poenially differ based on insiuional facors or srucural feaures of he labor marke. Wihin Brazil, unlike he cross-counry analysis, he effecs of high and low produciviy secor growh on employmen and unemploymen are broadly similar. High produciviy secor growh is associaed wih weakly posiive effecs on employmen growh, while low produciviy secor growh is associaed wih weakly negaive effecs on employmen growh. The difference beween coefficiens, however, is no saisically significan (es p-value = 0.384). A one percenage poin increase in aggregae GDP due o high-produciviy secors is associaed wih a iny 0.03 percenage poin reducion in unemploymen, hough he esimae is precise (significan a he 1 percen level). Low produciviy growh is associaed wih an even weaker 0.02 percenage poin reducion in he unemploymen rae, and he difference beween high produciviy and low produciviy coefficiens is no saisically disinguishable from zero (es p-value = 0.274). High produciviy growh and low produciviy growh are also indisinguishable in heir effecs on growh or reducions in he share of he populaion ha is ou of he labor force (es p-value = 0.526). 14 In specificaions allowing for counry-specific growh raes, we similarly find poin esimaes ha sugges ha growh in lower produciviy secors differenially reduces unemploymen. These esimaes, however, sugges ha his reducion in unemploymen may largely come from increases in dropou from he labor force, and ha growh in higher produciviy secors significanly increases non-agriculural employmen relaive o growh in lower produciviy secors. 10

11 One cavea is ha for hese hree oucomes in Brazil, we use daa from he PME, which is only represenaive of urban workers in six saes. We herefore supplemen he PME wih daa from he PNAD, which is a naionally represenaive household survey, o esimae effecs on he remaining wo oucomes. In fac, using he PNAD makes lile difference, as he resuls on high produciviy and low produciviy secor growh on log wages and log produciviy are comparable o hose from he PME. Boh high produciviy and low produciviy secor growh are associaed wih modes increases in wage growh, alhough boh poin esimaes are saisically indisinguishable from zero and from each oher (es p-value = 0.256). The esimaes of he effecs of high produciviy and low produciviy secor growh on produciviy growh are also srikingly similar in Brazil, wih high produciviy growh associaed wih a poin esimae of (significan a he 1-percen level) and low produciviy growh associaed wih a poin esimae of (also significan a he 1-percen level). Despie he increased precision due o he use of daa from many more provinces, he poin esimaes are no saisically differen from each oher (es p-value = 0.250). Indonesia is more consisen wih he resuls from he cross-counry analysis, as high-produciviy growh appears o be more effecive a reducing employmen han growh in less producive secors. The coefficien on high produciviy growh is larger in magniude (-0.235) han he coefficien and low produciviy growh. Thus, in Indonesia, a one percenage poin increase in GDP growh due o more producive secors is associaed wih a 0.2 percenage poin decline in he employmen o populaion raio. The difference beween coefficiens is marginally significan (es p-value = 0.061). Because Indonesia s per capia GDP was roughly $800 in 2000, i is possible ha growh in more producive secors proved more beneficial o workers by allowing more o volunarily exi he labor force. The wage resuls also appear o mach boh he cross-counry analysis and are consisen wih beneficial labor marke effecs from more producive secors, wih he cavea ha mos of he wage daa is rerospecive. Specifically, high produciviy growh appears o raise wages more han low produciviy growh in Indonesia. The coefficien on high produciviy growh (1.790, significan a he 1-percen level) suggess ha a one percen increase in aggregae GDP increases wages by 1.8 percen, and is much larger han and saisically disinguishable from he coefficien on low produciviy growh (0.022, no saisically disinguishable from zero), wih a es p- value of However, he Indonesian resuls are heavily influenced by he evens of he 1998 financial crisis, which led simulaneously o a seep decline in higher-produciviy secors and exraordinary real wage declines on he order of 40 percen. Looking a produciviy growh, which is available for a wider se of years, ells a differen sory. Low produciviy growh appears o raise labor produciviy significanly more han does high produciviy growh, wih he difference beween coefficiens significan a he 5 percen level. Turning o Mexico, esimaes of he impacs of high and low produciviy growh also sugges much larger effecs of high produciviy secor growh in ha conex. A 1 percen increase in growh due o high produciviy secors is associaed wih employmen growh ha is faser by 0.25 percen (coefficien significan a he 1 percen level), while a 1 percen increase in growh due o low produciviy secors is associaed wih employmen growh ha is faser by only 0.07 percen (no saisically differen from zero). The difference beween coefficiens, however, is no saisically differen from zero (es p-value = 0.135). Unemploymen raes also appear o drop more wih high produciviy secor growh (, significan a he 1 percen level) han wih low produciviy secor growh (, no saisically disinguishable from zero). Resuls on he likelihood of being ou of he labor force look very similar o hose for unemploymen, wih growh in high produciviy secors leading o much larger declines in he likelihood of being ou of he labor force (, significan a he 1 percen level) han growh in low produciviy secors (, no saisically disinguishable from zero), bu again, he coefficiens are no saisically differen from each oher (es p-value = 0.267). 11

12 In Mexico, he impacs of secoral growh paerns on wage growh appear small, bu as in Indonesia, low produciviy secor growh appears o lead o more rapid increases in average labor produciviy han does high produciviy secor growh. Wage impacs of boh high and low produciviy secor growh are small and saisically insignifican, as well as indisinguishable from each oher (es p-value = 0.733). The impac of low produciviy secor growh on labor produciviy appears large and posiive (, significan a he 1 percen level). The impac of high produciviy secor growh on labor produciviy also appears posiive, bu smaller (, marginally significan a he 10 percen level). The difference in coefficiens for produciviy appears o be significan (es p-value = 0.021). In sum, he cross-counry evidence suggess ha growh in low-produciviy secors do more o increase employmen and reduce unemploymen, especially in middle-income counries. The counry case sudies are less conclusive, however, and sugges cauion before generalizing cross-counry paerns o paricular conexs. In none of he hree cases did low-produciviy secors clearly have sronger employmen or unemploymen effecs han more producive secors. 4.2 Growh disaggregaed a he 1-digi secor level: agriculure, expor-led manufacuring, and naural resources The differences in he resuls beween low and high-secor growh, boh across and wihin counries, sugges ha growh in individual secors may have disincive effecs. We herefore urn o esimaing he effecs of growh in more finely disaggregaed secors on employmen oucomes. In he cross-counry daa, we use daa from he Unied Naions disaggregaed primarily a he 1-digi secor level o creae seven oupu caegories (Table 4). In conras o he lieraure emphasizing he special imporance of agriculural growh or manufacuring, we find only evidence for srongly differenially negaive effecs of growh in mining and uiliies on employmen growh. 15 Similarly, we find large and significan increases in he unemploymen rae associaed wih mining and uiliies growh in he cross-counry analysis. 16 Besides mining and uiliies, here is lile evidence of oher secors differenially affecing employmen oucomes, excep for he anomalous resul ha growh in oher indusries (largely services) has more posiive effecs on wage growh. The previous secion illusraed he imporance of disaggregaing he average effecs presened in he crosscounry regressions, o consider how impacs on employmen and unemploymen depend on counries level developmen. In low-income conexs, employmen ends o declines wih growh, and open unemploymen is ypically a poor indicaor of labor marke healh. For his reason, we esimae impacs separaely for each period using locally weighed regressions, based on counries log GDP per capia in 2000, shown in Figures Each do represens a differen period for which boh changes in GDP by secor and labor marke oucomes are observed. 15 This resul is somewha sensiive o specificaion choice. We separaely esimae effecs of mining and uiliies growh and manufacuring growh by subracing manufacuring oupu from he groupings of ISIC C-E (mining, manufacuring and uiliies) provided in he raw daa from he Unied Naions. When using he raw daa as provided, we find a large and srongly posiive coefficien of manufacuring growh on employmen growh, alhough he ne effec, adding he coefficiens on ISIC C-E and on ISIC-D, manufacuring alone, is much more weakly posiive. 16 This resul is robus o he inclusion of counry or province-specific rends in boh he cross-counry analysis and when looking a Indonesia, alhough in he laer case, in specificaions including province-specific rends, he negaive effecs of mining growh show up on wages raher han employmen. 17 The bandwidh for log per capia GDP in he non-parameric regressions is 1 and he kernel is Gaussian. 12

13 Figures 5a-5c indicae ha manufacuring is srongly, significanly and posiively associaed wih employmen in he range of abou $3000 or $3500 o jus under $20,000, wih effecs larges for upper middle-income counries as defined by he World Bank. For hese counries, a 1 percen increase in aggregae growh due o manufacuring is associaed wih up o a 1 percen increase in he employmen o populaion raio and a comparable decline in he unemploymen rae. In conras, he correlaion beween naural resource growh and employmen is far more evenly spread among he disribuion (Figure 6a). Figure 6b, however, is suggesive of an unemploymen effec even a low levels of GDP, which grows for upper middle-income counries. There is also suggesive evidence ha mining growh may be associaed wih increased wages. In conras o manufacuring and mining, here is srikingly lile evidence of srong effecs of agriculural growh on employmen growh or decline a low levels of income, where we would a priori expec such effecs o be mos significan (Figure 7a). Agriculural growh does appear o be associaed wih reduced unemploymen among middle-income counries and surprisingly increased employmen in high income counries. Esimaed wage effecs a low levels of income are also surprisingly small in magniude, hough in mos cases wages are no defined for many in agriculure, so his may also reflec changes in he composiion of he work force. Regardless, here is remarkably lile evidence ha agriculural growh is associaed wih improved labor marke oucomes for low and lower-middle income counries. To assess he possibiliy of a differenial impac of expor manufacuring on employmen and wages, we combine daa on expor manufacuring from he World Bank s World Developmen Indicaors daabase, and regress growh in employmen oucomes on overall GDP growh and growh in expor manufacuring (Table 9). Alhough expor manufacuring does have an independen effec on employmen oucomes, he poin esimae is small. Susained growh over five years a he 75 h percenile of growh in expor manufacuring would correspond o a roughly one hundred percen change in expor manufacuring oupu, and a 2.8 percenage poin increase in employmen raes. Growh in expor manufacuring is also associaed wih a reducion in unemploymen, bu is weakly associaed wih wage growh. I is possible ha episodes of growh in expor manufacuring coincide wih increases in overall openness o rade, which may inroduce compeiive pressures ha preven wages from growing. The nonparameric resuls (Figures 8a-8c) generally confirm hese resuls, as effecs on employmen are generally posiive bu small, and here is no paricularly srong effec for upper-middle income counries. Manufacuring expors are associaed wih reduced unemploymen in middle income counries, bu he effecs are mild, and here is lile effec of wages. Overall, he resuls sugges ha he srong effecs of manufacuring growh on employmen and unemploymen in middle income counries are mainly capured in overall growh, and here is lile addiional benefi from increasing manufacuring expors. The evidence from Brazil is also consisen wih a mild impac of manufacuring (Table 6). Alhough manufacuring is he only secor significanly associaed wih employmen growh, a 1 percen increase in growh from manufacuring growh would lead o only a 0.05 percen increase in employmen growh (poin esimae significan a he 5 percen level). For unemploymen, governmen, ranspor and communicaions, and mining growh are all associaed wih reducions in unemploymen (a he 1 percen, 5 percen, and 10 percen level, respecively), which he larges effecs for growh in governmen and ransporaion and communicaions. Looking a he poin esimaes on growh in he share of he populaion ha is inacive, however, suggess ha hese effecs on unemploymen are largely driven by increased dropou from he labor force. We find no effecs of individual secor growh on wage growh ha are saisically significanly differen from zero, and find roughly similar effecs of growh in all secors on growh in average labor produciviy, alhough no all coefficiens are sufficienly precisely esimaed o be saisically disinguishable from zero. In sum, alhough he hypohesis ha he coefficiens are equal is rejeced, he Brazilian evidence does no sugges large differenial effecs beween he seven secors. 13

14 Wihin Indonesia, growh paerns sugges ha agriculure growh has a srongly posiive effec on employmen growh, while growh in rade and commerce and growh in mining have significanly negaive impacs on employmen growh (Table 7). 1 percenage poin of addiional growh in agriculure is associaed wih 0.35 percenage poins of addiional employmen growh (poin esimae significan a he 5 percen level), while 1 percenage poin of growh from addiional growh in services is associaed wih a 0.38 percenage poin reducion in employmen growh (poin esimae marginally significan a he 10 percen level). Consisen wih he crosscounry resuls on a ype of resource curse, 1 addiional percenage poin of growh from mining is associaed wih boh a 0.05 percenage poin reducion in employmen growh (significan a he 5 percen level) and a 0.12 percenage poin reducion in log wage growh (marginally significan a he 10 percen level). In conras, manufacuring growh is associaed wih large increases in wage growh, as he implied effec of a 1 percenage poin increase in growh due o manufacuring is a 2.5 percen poin increase in wages (significan a he 1 percen level). We can rejec equaliy of effecs on employmen growh (es p-value = 0.004) and of effecs on wage growh (es p-value = 0.011). Effecs of 1-digi secor growh on labor produciviy growh also vary, wih he larges and mos significan effecs coming from growh in agriculure, rade and commerce, and consrucion. 18 Wihin Mexico, growh in manufacuring and in finance is associaed wih significanly faser employmen growh (esimaes saisically significan a he 5 percen and 1 percen levels, respecively, Table 8). The magniudes of he effecs are fairly large, wih a 1 percenage poin increase in growh from manufacuring associaed wih a 0.24 percenage poin increase in employmen growh, and a 1 percenage poin increase in growh from he financial secor associaed wih a 0.39 percenage poin increase in employmen growh. Only mining growh and growh in finance are significanly associaed wih a reducion in he growh of he share of he populaion ou of he labor force, wih a poin esimae of (significan a he 5 percen level) for mining and a poin esimae of (marginally significan a he 10 percen level) for finance. Faser growh due o growh in governmen and public adminisraion and from growh in he financial secor is associaed wih much faser wage growh in Mexico. Faser produciviy growh appears o follow from growh in agriculure, consrucion, and manufacuring in Mexico. Differences among coefficiens are significan for only unemploymen, wage growh, and produciviy growh. Overall we find lile evidence ha agriculure is a leading candidae o improve employmen oucomes and raise wage growh, alhough in our analysis here we are neing ou iniial employmen shares in order o es for composiional effecs of growh. We do find evidence for effecs of manufacuring growh on wage growh and employmen growh in he cross-counry analysis, and in each of he wihin-counry analyses. We also find evidence for a resource curse refleced in employmen oucomes in boh he cross-counry daa and in Indonesia, alhough we find no evidence for such in Brazil or Mexico. 5. Labor marke mobiliy and long run growh effecs The previous secions have described noable differences in secoral growh paerns and employmen oucomes, paricularly wih respec o differences beween high and low produciviy secors, as well as growh in manufacuring and naural resources. Moivaed by hese findings, we nex examine wheher he effecs of he 18 In specificaions including province-specific rends, growh in all secors is similarly unemploymen-reducing and has few oher saisically significan effecs on hese oucomes, alhough as noed earlier, mining growh is associaed wih significanly slower wage growh. 14

15 secoral composiion of growh on employmen oucomes vary wih measured labor mobiliy, or appear o be differen over a much longer ime horizon. Table 10 shows he resuls of regressions where high and low produciviy growh, again weighed by shares, are ineraced wih sae or province level measures of average annual labor mobiliy. The measures of labor mobiliy are rough, and reflec he share of workers who repor changing indusries wihin one year. In Brazil, Indonesia and Mexico, differences beween he effecs of high and low produciviy secor growh on changes in employmen oucomes appear o be larges in low-mobiliy saes and provinces. These differences are grealy miigaed by increases in labor mobiliy. In paricular, he poin esimaes on he ineracions beween secoral growh and labor mobiliy imply ha in all hree counries, he effecs of high and low produciviy growh on changes in employmen oucomes are roughly equal ha in he mos mobile saes or provinces. These effecs are mos precisely esimaed for employmen growh and in Mexico. In Brazil, despie he fac ha we are consrained by he coverage of he survey conaining individual-level panel daa on employmen o only six saes, we see similar qualiaive paerns. Alhough pas sudies come o conflicing conclusions on he role of regulaions in resricing job mobiliy, a consensus has emerged ha employmen proecion regulaions and relaed laws reduce gross labor mobiliy (Freeman, 2010). We herefore examine cross-counry daa on labor regulaion o see if policy-induced variaion in labor mobiliy also miigaes he differenial effecs of secoral growh. The resuls, presened in Table 11, give no indicaion ha improvemen in a counry s Doing Business rank aenuaes he effec of differen secoral growh paerns. We aribue his o he limied abiliy of he Doing Business measures o capure much of he underlying empirical variaion in labor marke mobiliy and flexibiliy. Finally, we urn o he census daa o esimae he long-run effecs of growh composiion on employmen oucomes, afer labor has an opporuniy o adjus across secors. The daa from Brazil and Indonesia span four decades, from 1960 o 2000 and 1971 o 2010 respecively, while he Mexican daa covers he 20 years from 1990 and 2010, wih employmen and oupu daa a approximaely 10 year inervals. The magniudes of he poin esimaes on growh from agriculure, services, and indusry are srikingly similar over longer horizons in Brazil, Indonesia and Mexico (Table 12). In he long run, i appears o make lile difference wheher sae or provincial growh was led by agriculure, indusry, or services. We herefore conclude ha differences in he effecs of growh in differen secors are largely miigaed by labor mobiliy. 6. Conclusion Broad discussions on how policy can creae more and beer jobs in developing counries ofen spark debae on wheher policy should aim o prioriize growh in paricular economic secors. Indeed, shor-run paerns of secoral growh appear o have disparae impacs on employmen, wage and produciviy growh around he developing world. In cross-counry regressions, growh in less producive secors, such as agriculure, consrucion, and reail and wholesale rade and oher services, end o lead o faser employmen growh han growh in higher produciviy secors, such as manufacuring, finance and business services, and mining and uiliies, wih paricularly noiceable effecs for middle-income counries. A he one digi level, he sronges effecs are apparen in naural resources and manufacuring. Naural resource growh in he cross-counry analysis, while conribuing o reducions in employmen for low-income counries, 15

16 appears o harm labor markes in middle-income counries by reducing employmen and increasing unemploymen. Manufacuring growh appears o reduce employmen for low-income counries, suggesing ha hese more producive jobs allow a greaer share of workers o volunarily exi he labor force. Manufacuring growh in middle-income counries, unlike naural resource growh, leads o large increases in employmen and reducions in unemploymen, boh when comparing across counries and across Mexican and Brazilian saes. This suggess ha as counries develop, he sakes in he global compeiion for growh in labor-inensive secors, paricularly in manufacuring, rise. In conras o previous lieraure, we find lile oher evidence for special effecs of growh in paricular secors, such as agriculure, on employmen and wage growh. While growh in he financial secor is srongly associaed wih increased employmen and wage growh in Mexico, here is lile evidence for a special role for financial services growh in oher conexs. While growh in low produciviy secors may lead o faser employmen growh, he evidence boh wihin and across counries suggess ha lower employmen growh may be coupled wih faser wage growh. This suggess ha labor marke rigidiies may play a role in deermining he disribuional impacs of growh, boh overall and in paricular secors. The specific impacs will depend on srucural and insiuional feaures of he indusry as well as local labor markes, which may no be capured by sandard measures of regulaion. The varying paerns of secoral growh on employmen oucomes in he hree counry case sudies of Brazil, Indonesia and Mexico are also consisen wih a rade-off beween wages and employmen. To he exen ha expansions in employmen will differenially benefi he poor relaive o increases in wages, holding employmen fixed, we consider our main cross-counry resuls consisen wih he lieraure emphasizing he imporance of labor-inensive (he inverse of producive) growh for povery reducion around he world. 19 We also find ha measured labor mobiliy significanly aenuaes differences in he impacs of growh composiion on employmen oucomes, as does examining he longer run. Taken ogeher, our esimaes sugges ha in he shor-run, and o he exen ha wealh is accumulaed in savings and asses over any ransiion pah over he longer-run, differences in he secoral composiion of growh do have disribuional implicaions, as hey differenially bring he unemployed ino he workforce, or raise he wages of hose already employed. However, hese differences may be dissipaed by higher labor mobiliy and over longer ime horizons. Overall, our resuls sugges ha in middle-income counries, he composiion of growh as well as overall growh has a leas a shor-run effec on he growh of employmen and wages. Composiion effecs are much smaller in low income conexs, and appear o dwindle over longer ime horizons. This is consisen wih argumens ha focusing on overall growh raher han paricular secors may be an appropriae long-erm policy objecive (Kraay, 2006). Labor marke mobiliy ends o aenuae hese composiional effecs, and promoing mobiliy beween jobs may herefore be an imporan mechanism for more broadly disribuing he benefis of growh originaing in higher produciviy and more capial-inensive secors. 19 Relaedly, Guierrez e al (2007) find ha he povery impacs of employmen inensiy vary by secor, wih produciviy improvemens necessary for povery reducion in agriculure. 16

17 Acknowledgmens This is a background paper for he 2013 World Developmen Repor on Jobs. The auhors would like o hank Arup Banerji, Kahleen Beegle, Samuel Freije, David Robalino and seminar paricipans in Washingon DC and he Eas Asia and Pacific region for helpful commens and suggesions, as well as Claudio Monenegro and his eam for compiling and providing he daa. The views expressed in his paper are hose of he auhors and do no necessarily reflec hose of he World Bank, is Board of Direcors, or he counries hey represen. References Amii, Mary and Caroline Freund (2010). The Anaomy of China s Expor Growh. In China s Growing Role in World Trade. Eds. Rober Feensra and Shang Jin Wei, Naional Bureau of Economic Research, Cambridge, MA. Alvarez-Cuadrado, Francisco, and Markus Poschke (2011). Srucural Change Ou of Agriculure: Labor Push versus Labor Pull." American Economic Journal: Macroeconomics, 3(3): Angris, Joshua D., and Adriana D. Kugler (2008). Rural Windfall or a New Resource Curse? Coca, Income, and Civil Conic in Colombia." The Review of Economics and Saisics, 90(2): 191{215. Balassa, Bela (1978). Expors and Economic Growh. Journal of Developmen Economics, 5: Berg, Janine, Chrisopher Erns and Peer Auer (2006). Meeing he Employmen Challenge: Argenina, Brazil and Mexico in he Global Economy. Inernaional Labour Organizaion. Caselli, Francesco and Guy Michaels (2009). Do Oil Windfalls Improve Living Sandards? Evidence from Brazil." NBER Working Papers Chenery, Hollis and Moises Syrquin (1975). Paerns of developmen, , Oxford Universiy Press, London. Chrisiaensen, Luc, Lionel Demery, and Jesper Kuhl (2011). The (Evolving) Role of Agriculure in Povery Reducion An Empirical Perspecive. Journal of Developmen Economics, 96(2): Da, Gaurav and Marin Ravallion (2011). Has India s Economic Growh Become More Pro-Poor in he Wake of Economic Reforms? World Bank Economic Review, 25(2): De Janvry, Alain and Elizabeh Sadoule (2009). Agriculural Growh and Povery Reducion: Addiional Evidence. World Bank Research Observer, 25(1): Dube, Oeindrila and Juan Vargas (2012). Commodiy Prices and Civil Conflic: Evidence from Colombia. Forhcoming, Review of Economic Sudies. Ferreira, Francisco, Phillippe Leie, and Marin Ravallion (2010). Povery Reducion Wihou Economic Growh? Explaining Brazil s Povery Dynamics, Journal of Developmen Economics, 93(1): Foser, Andrew and Mark Rosenzweig (2007). Economic Developmen and he Decline of Agriculural Employmen. In Handbook of Developmen Economics, Volume 4. Eds. T Paul Schulz and John Srauss, Elsevier. 17

18 Freeman, Richard B. (2010). Labor Regulaions, Unions, and Social Proecion in Developing Counries: Marke Disorions or Efficien Insiuions? In Dani Rodrik and Mark Rosenzweig (Eds.), Handbook of Developmen Economics (Vol. Volume 5, pp ), Elsevier: Amserdam. Freund, Caroline and Bob Rijkers (2012). Employmen Miracles. Mimeo, World Bank. Gollin, Douglas, Sephen L. Parene, and Richard Rogerson (2002). The Role of Agriculure in Developmen." American Economic Review, 92(2): Guierrez, Caalina, Carlo Orecchia, Pierella Paci, and Pieer Serneels (2007). Does Employmen Generaion Really Maer for Povery Reducion? World Bank Policy Research Working Paper No Haggblade, S., Hazell, P., Reardon, T. (2002). Sraegies for Simulaing Povery-Alleviaing Growh in he Rural Non-farm Economy in Developing Counries. Mimeo, Inernaional Food Policy Research Insiue, Washingon, DC. Headey, Derek, Dirk Bezemer and Peer B. Hazell (2010). Agriculural Employmen Trends in Asia and Africa: Too Fas or Too Slow? World Bank Research Observer, 25(1): Kraay, Aar (2006). When Is Growh Pro-poor? Evidence from a Panel of Counries. Journal of Developmen Economics, 80(1): Lanjouw, Jenny and Peer Lanjouw (2001). The Rural Non-farm Secor: Issues and Evidence from Developing Counries. Agriculural Economics, 26 (1): Loayza, Norman and Claudio Raddaz (2010). The Composiion of Growh Maers for Povery Alleviaion. Journal of Developmen Economics, 93: McMillan, Margare S. and Dani Rodrik (2012). Globalizaion, Srucural Change and Produciviy Growh. NBER Working Paper Monenegro, Claudio, and Maximillian Hirn (2009). A New Disaggregaed Se of Labor Marke Indicaors Using Sandardized Household Surveys from Around he World, World Developmen Repor Background Paper, available a: hp://sieresources.worldbank.org/intwdr2009/resources/ /monenegro- Hirn.pdf Ravallion, Marin and Shaohua Chen (2007). China s (Uneven) Progress Agains Povery. Journal of Developmen Economics, 82(1): Ravallion, Marin and Gaurav Da (1996). How Imporan o India s Poor is he Secoral Composiion of Economic Growh? World Bank Economic Review, 10(1): Robinson, James, Ragnar Torvik, and Thierry Verdier (2006). Poliical Foundaions of he Resource Curse. Journal of Developmen Economics, 79(2): Sachs, Jeffrey D. and Andrew Warner (1995). Naural Resource Abundance and Economic Growh. Harvard Insiue for Inernaional Developmen, Developmen Discussion Paper No Sachs, Jeffrey D. and Andrew Warner (1999). The Big Push, Naural Resource Booms and Growh. Journal of Developmen Economics 59,

19 Suryahadi, Asep, Daniel Suryadama, and Sudarno Sumaro (2009). The Effecs of Locaion and Secoral Componens of Economic Growh on Povery: Evidence from Indonesia. Journal of Developmen Economics, 89(1): Thurlow, James and Peer Wobs (2006). No All Growh is Equally Good for he Poor: The Case of Zambia. Journal of African Economies, 15(4): Topalova, Peia (2008). India: Is he Rising Tide Lifing All Boas? Issues IMF Working Paper. Inernaional Moneary Fund, Washingon, D.C. Van der Ploeg, Frederick (2011). Naural Resources: Curse or Blessing, Journal of Economic Lieraure, 49:2, p World Bank (2012). World Developmen Indicaors. World Bank, Washingon D.C. 19

20 Figure 1a: Smoohed Employmen o Populaion raio by per capia GDP, Unemploymen rae GDP in 2000 Noes: Each do represens he average employmen o populaion raio from he mean of employmen o populaion, locally weighed according o per capia GDP. The daa consis of 62 counries for which he laes observed year for each counry for surveys beween 2000 and 2010.The bandwidh is 1, wih an Epanchnikov kernel. Figure 1b: Unemploymen rae by per capia GDP, GDP in 2000 Noe: See noes o figure 1a 20

21 Figure 2: Esimaed impac of high and low produciviy secor growh on employmen o populaion raio, by counry GDP in 2000 Noes: Each do represens a coefficien esimae on from equaion (2), locally weighed according o log per capia GDP in 2000 using a bandwidh of one and an Epanechnikov kernel. Dashed lines represen 95% confidence inervals. Secoral growh is weighed by is share of GDP. The coefficiens represen he percenage poin change in he employmen o populaion raio associaed wih he amoun of addiional growh in high or low producive secors necessary o raise aggregae growh one percenage poin. High-produciviy secors are manufacuring, ranspor and communicaions, finance, elecriciy and uiliies, or mining. Low-produciviy secors are oher services, agriculure, reail and wholesale rade, governmen and public adminisraion, and consrucion. Sandard errors are clusered on counry. 21

22 Figure 3: Esimaed impac of high and low produciviy secor growh on unemploymen growh, by counry GDP in 2000 See noes o Figure 2. 22

23 Figure 4: Esimaed impac of high and low produciviy secor growh on wage growh, by counry GDP in 2000 See noes o Figure 2. 23

24 Figure 5a: Effecs of manufacuring growh on employmen growh, by counry GDP in 2000 See noes o Figure 2. Figure 5b: Effecs of manufacuring growh on unemploymen growh, by counry GDP in 2000 See noes o Figure 2. 24

25 Figure 5c: Effecs of manufacuring growh on wage growh, by counry GDP in 2000 See noes o Figure 2. 25

26 Figure 6a: Effecs of mining growh on employmen growh, by counry GDP in 2000 See noes o Figure 2. Figure 6b: Effecs of mining growh on unemploymen growh, by counry GDP in 2000 See noes o Figure 2. 26

27 Figure 6c: Effecs of mining growh on wages growh, by counry GDP in 2000 See noes o Figure 2. 27

28 Figure 7a: Effecs of agriculure growh on employmen growh, by counry GDP in 2000 See noes o Figure 2. Figure 7b: Effecs of agriculure growh on unemploymen growh, by counry GDP in 2000 See noes o Figure 2. 28

29 Figure 7c: Effecs of agriculure growh on wage growh, by counry GDP in 2000 See noes o Figure 2. 29