Impact of the Political Turmoil on Madagascar s Garment Workers: Preliminary Study Based on the Firm Dataset

Transcription

1 Impac of he Poliical Turmoil on Madagascar s Garmen Workers: Preliminary Sudy Based on he Firm Daase Takahiro Fukunishi 1. Inroducion 1.1. Moivaion The garmen indusry in Madagascar is one of he few manufacuring indusries in sub-saharan Africa which have successfully peneraed he expor marke. Triggered by foreign direc invesmen from Mauriius, garmen expors, which were suppored by duy-free access o he EU and US markes, sared o grow in he early 1990s. Alhough he indusry experienced adverse shocks in he 2000s, namely, he poliical urmoil in 2002 and he liberalisaion of he apparel marke in 2005, he garmen expor s growh did no collapse. In paricular, Madagascar s garmen indusry overcame he rade liberalisaion ha eliminaed he quoa on main garmen exporers and resuled in he reducion of African and Lain American counries. Reaching he highes expor value in 2007, Madagascar s prospecive growh seemed promising. The garmen indusry provides a large number of jobs, paricularly for female and unskilled workers in developing counries. The employmen in his indusry is open o workers who have no compleed primary educaion, and he wage is subsanially higher han ha for informal jobs, which is an alernaive for female garmen workers (Acevedo and Roberson, 2011). The same rend is observed in Madagascar (Glick and Roubaud, 2006; Fukunishi and Ramiarion, 2011). In 2008, more han 100,000 workers were esimaed o be employed, majoriy of which (over 80% according o our survey) occupied posiions ha were also open o unskilled workers. Assuming he coninued improvemen of working condiions, he growh of he indusry was expeced o reduce povery (Glick and Roubaud, 2006). The growh rend was disruped by he financial crisis in 2008, and recovery was suffocaed by anoher poliical urmoil ha occurred in A collision beween he presiden and he former mayor of he capial ciy caused a series of rios, which ended wih he replacemen of he presiden afer he incursion of he army ino he presidenial palace. I no only caused he degradaion of public service, bu also invoked he cancellaion of he duy-free access by he US governmen [African Growh and Opporuniy Ac (AGOA)] in January In conras o Insiue of Developing Economies; fukunis@ide.go.jp 1

2 he recovery of he expor markes, garmen expors from Madagascar coninued o decrease in 2010, and expors o he US recorded a reducion by 74.0% (UN Comrade). The purpose of his paper is o deermine he changes in employmen and wages afer he poliical urmoil. Paricular focus is on differenial impac by skill given he indusry s conribuion o povery reducion. As a relaed issue, we also show he changes in expor value and price. Due o he lack of a firm survey or a labor survey afer 2009, i is no ye known how many jobs were los, how wages changed, and wha kind of workers were mos affeced. The auhor conduced firm surveys since 2009 in collaboraion wih he Universiy of Anananarivo. The surveys conain deailed employmen and wage informaion before and afer he urmoil, which allow he invesigaion of differenial impacs by posiion and gender. All firms surveyed were more or less affeced by he poliical urmoil. Hence, a rigorous idenificaion of he impac of his poliical urmoil on wages and employmen is difficul wih our daa. As a preliminary analysis, we have no explored he possibiliy of idenificaion ye. On he oher hand, we have a beer prospecive in idenifying he impac of he cancellaion of he duy-free access because, unlike he US marke, he EU marke has been mainaining duy-free access afer he urmoil Skill Inensiy and Skill Premium The lieraure on wage inequaliy in developing counries repored a significan increase in inequaliy afer rade liberalisaion, in conras o he Solper Samuelson heorem, and some argumens accouning for he resuls are suggesed (Goldberg and Pavcnik, 2007). Two facors seem relevan in he case of Madagascar, where expor price has changed significanly in The Solper Samuelson heorem predics ha a change in produc prices causes a change in inpu prices according o he inensiy of he inpus. Given he high inensiy in unskilled labor, a decrease (increase) in expor price reduces (raises) he wage of unskilled workers relaive o ha of skilled workers. The oher facor is produc qualiy. If a change in expor prices incorporaes a change in produc qualiy, skill inensiy may change and skill premium will follow. Aggravaion of he business environmen may have made buyers downgrade he qualiy of he producs sourced from Madagascar because producing high-qualiy producs is risky in a highly uncerain poliical sabiliy and in a degraded business environmen. The effec of cancelling he duy-free access on produc qualiy is less clear. Edwards and Lawrence (2010) argued ha he provision of he AGOA should increase he expor of producs whose fabric cos share is relaively high, assuming he lower labor produciviy of African manufacurers. Therefore, heir model suggess ha he cancellaion of he AGOA will increase he share of he producs wih low fabric inensiy, bu is effec on produc qualiy is no clear. Some empirical sudies showed ha firms producing high-qualiy producs employ skilled workers more 2

3 inensively (Scho, 2004), and such ransiion of labor demand increases skill premium (Zhu, 2005; Verhoogen, 2008). Therefore, if subsanial qualiy change occurred in Madagascar s garmen indusry, skill inensiy and skill premium migh have changed and workers migh have suffered from asymmeric adverse effecs by skill Mehodology and Daa We invesigaed he changes in expor value, price, employmen, skill inensiy, and skill premium afer he poliical urmoil and he cancellaion of AGOA. As discussed in he previous subsecion, expor price and skill inensiy affec skill premium. Since all garmen firms were affeced by he poliical urmoil, and because he changes before and afer he urmoil involve oher facors, a simple comparison does no clearly indicae he effecs of he chaos. Paricularly in 2009, he US and EU markes were seriously affeced by he financial crisis. Hence, he changes (in employmen, for example) beween 2008 and 2009 clearly overesimae he impac of he urmoil. For an analysis of he expor value and price, we uilised world expor daa in addiion o Madagascar s daa o separae he impac of he poliical urmoil from he oher facors ha affeced all exporers, such as he financial crisis. However, for an analysis of employmen and wage, we could no isolae he effec of Madagascar s urmoil from he financial crisis because he analysis relied on firm daa from Madagascar. Alhough i migh be possible o isolae he effec of he urmoil if he impac differed by firms or workers, we did no explore idenificaion in his paper. Insead, we mainly uilised he changes from 2008 o 2010 o minimise he adverse effecs of he financial crisis. However, because he financial crisis or any oher shocks migh have affeced he marke in 2010, we recognised ha i was a crude approximaion a bes. The effec of he AGOA cancellaion can be idenified somewha more effecively. Madagascar s garmens were graned duy-free access o US and EU markes unil 2009, and duy-free access o he EU marke has been mainained afer There are some differences in he condiions of duy-free applicaion; he mos visible difference is in he rule of origin. EU requires he producion of fabrics (weaving or kniing process) as well as sewing o be conduced in Africa, and US allows using impored fabrics. Since hese condiions did no change afer 2010, he difference-in-difference (DID) approach is applicable in isolaing he effec of he cancellaion of duy-free access by comparing changes in exporers o he EU marke and hose in exporers o he US marke. We used wo ypes of daa: rade saisics and original firm-level daa. Expor value and price were invesigaed using he impor daa of US and EU in he World Trade Alas daa, which conain EURO sa and US Inernaional Trade Commission daa. Employmen and wage changes are based on he firm-level daa from 2008 o The firm daa in 2008 conain 98 3

4 facories of expor-oriened garmen firms regardless of he EPZ saus and 19 non-exporing firms, which were randomly sampled from he firm lis. 1 The facories in 2008 were followed in 2009 and In his paper, we used only exporers as of 2008, which include he facories ha qui exporing afer 2009, so ha we can rack he changes in exporers as of 2008 afer he urmoil. The firm daa conain he number of workers and he monhly wages according o eigh posiions, gender, and hree experience caegories (less han 1 year, 1 5 years, and longer han 6 years) and various firm informaion. 2. Changes in Expor Value and Price 2.1. Expor Value and Price The expor value from Madagascar o he US and EU markes decreased by 18.0% in 2009, while mos of he major exporers, including China, recorded a negaive growh rae mainly due o he financial crisis (Table 1). The change in 2010 showed a sharp conras o he oher exporers, which experienced recovery from he financial crisis. The expor value from Madagascar dropped by 38.5%, which was mainly accouned for by he drasic decrease in expors o he US by 74.0% (Table 1 and Figure 1). Since almos all expors o he US marke uilised AGOA before 2009, suspension hardly hi garmen exporers in Madagascar. Neverheless, he reducion from 2009 o 2010 is no enirely aribued o he cancellaion of he AGOA because buyers may have cancelled Madagascar s producs regardless of he preferenial access. In fac, expors o he EU marke from Madagascar, which coninues o be benefied by duy-free access decreased by 10.4% in he same period, while he oal impors in he US and EU markes increased by 13.7% and 3.0%, respecively (UN Comrade), reflecing he recovery from he financial crisis. This implies ha he reducion in he US expor is no solely aribued o he cancellaion of AGOA. Reducion in expor value accompanied he closure of firms and he redirecion of he marke. Ou of 98 firms in our sample, 26 firms exied by mid The number of firms ha expor a leas some of heir producion o he US marke decreased from 36 in 2008 o 7 in 2010 (Tables A1 and A2). The uni price showed a relaively large change in The average uni price of Madagascar s garmen in he EU marke decreased by 29.0%, while ha in he US marke rose by 6.8% (Table 2). Madagascar s expor price in he US marked he highes price for he las seven years, while he average price of all expors o he US marke decreased and became he lowes price since I is no inuiively consisen as he cancellaion of he ariff exempion leads o a reducion in expor price when a buyer and a supplier share he ariff cos. 1 See he Appendix for he sampling procedure. 4

5 Price changes incorporae he upgrading or downgrading of a produc. I is inrinsically difficul o separae he conribuion of qualiy changes and ha of inflaion (or deflaion) in he absence of price deflaors defined a a deailed produc level. Among several frameworks, Harrigan and Barrows (2009) suggesed o regard he change in price wihin a narrowly defined produc caegory as a pure price change (inflaion) and he change in produc composiion as qualiy upgrading. Following his definiion, we esimaed wihin-produc price changes using he HS eigh-digi level informaion from he period (Table 3). Coefficiens on y2010 indicaed wihin-produc price changes in 2010, in which one ha is relaive o he average of is posiively significan in he US marke and one ha is relaive o 2009 is negaively significan in he EU marke. Price reducion in he EU marke is significan only in he shor erm, and he increase in he US marke is significan in he long erm. According o he Solper Samuelson heorem, such price changes imply ha skill premium increases in firms exporing o EU, whereas i decreases in exporers o he US Esimaion of he Impac of he Poliical Turmoil and AGOA Cancellaion Difference in changes of expor value beween hose from Madagascar and from all over he world indicae Madagascar s specific changes neing ou he worldwide rend of marke demand. If we could assume ha Madagascar s specific changes semmed from he poliical urmoil, he DID esimaor would represen he impac of he urmoil on expors. Given he absence of oher significan facors affecing expors, i is a good approximaion. If we consider only wo periods (2008 and 2010) for simpliciy, he DID esimae is Mada Mada World World ( lnv lnv ) ( lnv V ) DID = ln 2008, where Mada lnv 2010 is he log of he expor value from Madagascar in For he impac of he AGOA cancellaion, we can uilise he difference in he changes beween expor values from Madagascar o he US marke and from Madagascar o he EU marke, given he differenial applicaion of duy-free access. Addiionally conrolling he marke rend, he riple difference approach yields an esimaion of he AGOA cancellaion effec, TripleDID = Mada, US Mada, US Mada, EU Mada, EU [( lnv2010 lnv2009 ) ( lnv2010 lnv2009 )] US US World EU World EU [( V ) ( ] World, V World,,, ln ln lnv lnv ), where lnv World, US 2010 is he log of he expor value from all over he world o he US in The above wo esimaors can be obained in one regression model for he riple difference. We applied he following model: 5

6 lnv i, c, m, 0 ( US y2010 Mada) + β ( US y2010) + ( Mada y2010) = β + β β 1 m, c, + yeardummy + produc * counry * marke 2 i, c, m m, 3 c,, where US m is he marke dummy (=1 if an observaion represens an expor value o he US), y2010 is he year dummy (=1 if an observaion is an expor value in 2010), Mada is he dummy for he exporing counry (=1 if an observaion is an expor value from Madagascar), produc*counry*marke i,c,m is he produc-counry-marke fixed effec, and i, c, m, and denoe he produc, counry of expor, marke, and ime, respecively. The DID esimae for he impac of he poliical urmoil is β 3, and he riple DID esimae for he impac of he AGOA cancellaion is β 1. We esimaed he HS eigh-digi level impor values in he EU and US markes from Madagascar and he world. The worldwide value is he oal impor value in he US and EU markes raher han individual daa for each counry. Due o reliabiliy, we used impor daa raher han expor daa. The coverage is from 2005 o 2010 o avoid he influence of he Muli-fibre Arrangemen ha was in effecive unil The resuls are shown in Table 4 (lef panel). The coefficien for Mada*2010, he DID esimae, is posiive and weakly significan, while he riple DID esimae, US*2010*Mada, is boh negaive and significan. The resul indicaes ha Madagascar s expor o he US marke decreased by 80.0% (=exp[-1.612]-1), conrolling he changes in expor from Madagascar o he EU marke and he worldwide rend in expors. The effec of he AGOA cancellaion is quie large. Also, he resul indicaes ha conrolling he reducion in he expor o he US, Madagascar s expor increased in he eigh-digi produc level. Therefore, he effec of he poliical urmoil is no significan. We applied he same esimaion model for he expor price. 3 The DID esimae for he uni price is negaive bu insignifican, and he riple DID esimae is boh posiive and significan (Table 4, righ panel). Suspension of he AGOA increased he uni price of Madagascar s produc in he US marke by 26.7% (=exp[0.237]-1). As he produc s fixed effec is incorporaed, his change represens wihin-produc price changes. Based on he assumpion ha we made earlier, he resul indicaes inflaion of Madagascar s produc in he US marke. These exercises demonsrae ha he impac of he AGOA cancellaion was subsanial, and in conras, he impac of he poliical urmoil iself was no significan. Conrary o inuiion, he average price of Madagascar s garmen expored o he US marke rose. Because Madagascar s 2 Zero expor value is changed o 1 as long as a produc has a leas one posiive value beween 2005 and 2010 in order o avoid missing value in naural logarihm. 3 The uni price is he value per dozen for mos producs, while some producs repored in a differen uni, such as piece, are conrolled by he uni dummy. The resul does no differ when excluding he produc repored in unis oher han dozens. 6

7 producs enjoyed a duy-free saus in he US marke unil 2009, heir prices could be higher han hose from oher counries by up o ariff raes. Therefore, we expec ha he cancellaion of AGOA would have caused a drop in he prices so ha he prices of Madagascar s producs became equal o he prices of oher counries producs. A possible explanaion is qualiy upgrading wihin he eigh-digi produc level, which is conrary o our assumpion. This is lef for furher research. 3. Changes in Employmen and Skill Inensiy 3.1. Employmen The oal number of employmen in our sample is 57,250 in 2008, which includes managerial posiions. 4 The main labor force in he garmen indusry consiss of supervisors, machine operaors, and helpers who comprise an assembly line. These hree posiions accoun for 86.9% of workers, and operaors have he larges share (72.9%) (Table 5). Based on educaion, we deermined ha managers and officers are he mos skilled; engineers, qualiy conrollers, and supervisors are moderaely skilled; and operaors, helpers, and oher workers are low skilled. The composiion clearly shows ha he indusry is low-skilled labor inensive. A drasic reducion in he jobs is observed from 2008 o 2010, 46.5% of he jobs (26,611) in he exporing firms (Table 5). Engineers, operaors, helpers, and oher workers recorded a large reducion. This indicaes ha skill inensiy has been changed. The share of low-skilled workers slighly reduced from 89.9% o 87.2%. The gender difference is no clear, while he reducion rae is larger for male workers in oal bu no for all posiions (Table 6). The reducion in employmen is caused by a firm closure and downsizing of he coninuing firms; he former accouns for 53.2% of he oal job loss (14,159 jobs), and hence, he res (47.8%) is aribued o he reducion in employmen in he coninuing firms. The closure of firms exporing o he US marke represens 33.5% of he oal job loss (8,903 jobs) Skill Inensiy The measuremen of produc qualiy is pracically difficul for he available daa. Because skill premium changes occur hrough changes in skill inensiy in producion, i is more sraighforward o measure skill inensiy in order o undersand he mechanism of skill premium changes (Goldberg and Pavcnik, 2007). The change in skill inensiy wihin firms is explored based on he firms coninuing operaion 4 Firms ha was in operaion in 2010 bu missed in he 2011 survey, which collecing employmen in FY2010, are dropped in his aggregaion o ensure consisen esimaion of employmen changes. 7

8 from 2008 o Differences in employmen change (rae) beween 2008 and 2010 by posiion are esimaed using he following model: ln L i, p, = β + φ ( posiion * 2010) β ( US * 2010) + φ4 posiion + β3usi + φ5 year + firmi i, p, + φ 2 ( posiion * 2010 * US) i, p, + φ 3 ( posiion * year) p, + where lnl is he log of he number of jobs, posiion is he se of posiion dummy (helper is he base), y2010 and year is he year dummy (2008 is a base), US is he dummy for expor marke (=1 if an observaion belongs o a firm exporing o he US), firm is he firm fixed effecs, and i, p, and denoe he firm, posiion, and ime, respecively. 5 The coefficien of posiion dummy, φ 4, is he raio of employmen of posiion p relaive o he helper, a base caegory. The change in skill inensiy is represened in, which shows he change in he raio relaive o he helper. If i is posiive, he change in employmen in posiion p is greaer han ha in he helper; in his case, he reducion is milder. The changes in he skill inensiy of exporers o he US are capured by φ 2 φ 1, which is inerpreed as he effec of he AGOA cancellaion. The resul shows ha reducion in he employmen of officers, engineers, supervisors, and oher workers is significanly smaller han ha of helpers (Table 5). Given he higher skills required for officers, engineers, and supervisors, his resul indicaes ha he wihin-firm skill inensiy increased. The change in employmen by posiion does no differ beween US and non-us exporers excep for he engineer posiion. I means ha he change in he skill inensiy of he US exporers does no differ wih he non-us exporers, and hence, hey became more skill inensive. Because he increase in skill inensiy reduces he demand for low-skilled workers, hese resuls imply ha he skill premium increases in he exporers o boh EU and US. 4. Wage and Skill Premium Wage changes are esimaed using he wage panel daa colleced by posiion, gender, and experience (hree caegories: less han 1 year, 1 5 years, and longer han 6 years). Table 7 shows he esimaion resuls of he wihin-firm changes in real wage conrolling gender and experience. Ou of he eigh posiions, five in 2009 and six in 2010 have negaive changes. For mos posiions, reducion is larger and more saisically significan in I shows a subsanial decrease in real wages, which resuled from he drasic reducion in employmen. 5 The observaions wih zero values were dropped when convered o log form, hough he number of such observaions is small. 8

9 The change in skill premium wihin firms is esimaed by pooling all he wage observaions using he same esimaion model as he one used in skill inensiy esimaion, ln w i, p, = β + φ ( posiion*2010) β ( posiion* year) ( US *2010) + φ4 posiionp + β3usi + φ5 year + φ6 X p + φ7zi, + firmi i, p, + φ 2 ( posiion*2010* US) where lnw is he log of he real wage, X is experience and gender o which wage observaion corresponds, and Z is he vecor of ime-varian firm characerisics, such as operaion days and average uni price. The base posiion is he helper, and hence, he premium shows a difference in wages in each posiion wih he helper. ime, and φ 1 i, p, + φ 3 p, represens a base premium ha is consan across represens he change in premium in If i is posiive (negaive), he premium becomes larger (smaller). Addiional changes in he skill premium in exporers o he US are capured by φ 2 φ 4, which is inerpreed as he effec of he AGOA cancellaion. Table 8 shows he resuls for he concerned variables. The base caegory refers o he wages of female helpers wih less han one year experience in The gender dummy is no significan, while he experience dummy is posiively significan wih he larger effec for a longer experience. This is a plausible resul in he garmen indusry where he gender gap is generally small or negligible. The base skill premiums represened by he coefficiens of posiion dummies are posiively significan wih larger premiums for more skilled jobs. The coefficiens of he ineracion erms of he posiion dummy wih he 2010 dummy are negaive for all he posiions and significan for he posiions of managers, officers, qualiy conrollers, and operaors. This is evidence ha he wihin-firm skill premium beween helpers and more skilled jobs became smaller. Furhermore, he poin esimaes of reducion are larger for more skilled jobs; ha is, skill premiums are conraced proporionally o skill. The skill premium for US exporers did no differ significanly from ha for non-us exporers as he ripe ineracion erms are insignifican for all he posiions (Table 8). Hence, i means ha he cancellaion of AGOA did no affec he skill premium, and hus, he skill premium narrowed for he US exporers. For he exporers o he non-us marke, he increased skill premium couners he observed price and he skill inensiy changes. Because he prices in he EU marke have fallen and he skill inensiy has increased among non-us exporers, one expecs ha he wage of low-skilled workers decreases relaive o ha of high-skilled workers. On he oher hand, he change in he skill premium is undeermined for he exporers o US as he increased price is expeced o narrow he skill premium, while he increased skill inensiy would broaden he premium. If he laer effec has overcome he former, he observed reducion in skill premium is consisen. Alhough he impac of he AGOA cancellaion is no visible in he skill premium, i may be + 9

10 masked by he unexpeced reducion of skill premium in non-us exporers as our idenificaion ress on he difference in premium change beween he US and non-us exporers. The reason for he inconsisency in he relaionship beween skill premium and changes in price and skill inensiy among non-us exporer is unclear. However, he conradicion beween skill inensiy and skill premium suggess ha a possible explanaion is in he labor marke. Assuming an effecive and inegraed labor marke, he difference in he price elasiciy of he labor supply, ha is, low elasiciy for high-skilled workers, can explain i alhough i is unlikely in general. If low-skilled workers have alernaive employmen opporuniies in an informal secor and high-skilled workers have fewer opporuniies due o he recession caused by he poliical urmoil, such difference in elasiciy of he labor supply is possible. The efficiency wage may also explain he downward rigidiy of low-skilled workers. 5. Conclusion Afer he poliical urmoil in March 2009, garmen expors from Madagascar experienced a drasic reducion. The expor value and employmen halved from 2008 o 2010, and expor reducion was much greaer in he US marke. This was because wo effecs were a work: social confusion by poliical urmoil and cancellaion of he AGOA, which allowed duy-free access o he US marke. We explored he impac of he adverse shock on expor value, price, skill inensiy, and skill premium using he original firm daase. In erms of he expor value, he cancellaion of he AGOA had more significan impac han he poliical urmoil iself. The former accouned for he reducion in US expors by 80.0% on average over he producs defined by HS eigh-digi level, while he laer did no have a significan effec. The expor price made considerable changes; he average price decreased for expors in he EU marke, and i rose for hose o he US marke. Similar o he expor value, he AGOA cancellaion had a clear impac on price change, whereas he impac of he poliical urmoil was no significan. Surprisingly, i raised he expor price by 26.7% on average. In he drasic reducion in jobs, skill inensiy increased in garmen employmen as a whole as well as wihin individual firms. The change in skill inensiy does no significanly differ beween exporers o he US marke and exporers o he non-us marke. Hence, we found weak evidence ha he poliical urmoil increased he skill inensiy, bu we are more confiden ha he AGOA cancellaion did no affec i. Because increased skill inensiy should reduce he demand of low-skilled workers relaive o he high-skilled workers, i augmens skill inensiy. In conjuncion wih fall of expor price, he skill premium should have broadened in he firms ha expor o he non-us marke. On he oher hand, skill premium change in he exporers o he US is expeced o be ambiguous due o he increased expor price. 10

11 We found ha real wages dropped in mos of he posiions, whereas, conrary o he expecaion, wihin-firm skill premium conraced regardless of marke orienaion. The discrepancy beween skill inensiy and skill premium suggess ha he explanaion is in he labor marke, such as he difference in he elasiciy of labor demand by skill, efficiency wage, or insiuional rigidiy in low-skilled wages. Alhough he exporers o US did no have differenial changes wih he non-us exporers, i may possibly be because of he unexpeced reducion in skill premium in he non-us exporers. Overall, he poliical urmoil did no significanly affec he expor value and price, while suspension of he AGOA caused a drasic reducion in expors and an increase in expor price. A subsanial par of he employmen loss is, herefore, aribued o he AGOA suspension. Alhough causaion is no clear, low-skilled workers suffered more seriously from job desrucion han high-skilled workers. In conjuncion wih he decrease in expor prices in he EU marke, which is he larges marke for Madagascar s garmen firms, hese changes indicae a possible increase in skill premium and furher degradaion of he low-skilled workers welfare. However, i was no realised, and he premium was reduced. Two opposie changes in employmen and wage obscure he change in inequaliy. There are sill many issues o be explored. While we have shown he changes in expor price, skill inensiy, and skill premium, he mechanism linking hem has no been invesigaed ye. The observed relaionships beween skill inensiy and premium and beween expor price and skill inensiy are no consisen wih he heoreical implicaions or empirical evidences in he lieraure. A more rigorous idenificaion is necessary o undersand he impac of poliical urmoil on workers. Moreover, from he perspecive of povery reducion, scope should be broadened from wage inequaliy o income inequaliy of garmen workers, including hose who los a job. 11

12 Reference Acevedo, Gladys Lopez and Raymond Roberson (2011) Sewing Success? Employmen and Wages Following he End of he Muli-fibre Arrangemen, Washingon, D.C.: World Bank. Edwards, Lawrence and Rober Z. Lawrence (2010) AGOA Rules: The Inended and Uninended Consequences of Special Fabric Provision, NBER Working Paper Series, 16623, Cambridge: Naional Bureau of Economic Research. Fukunishi, Takahiro and Herinjaovo Aimé Ramiarion (2011) Expor-Oriened Garmen Indusry in Madagascar: Implicaion of FDI o he Local Economy, mimeo, Chiba: Insiue of Developing Economies. Glick, Peer and François Roubaud [2006] "Expor Processing Zone Expansion in Madagascar: Wha are he Labor Marke and Gender Impac?", Journal of African Economies, Vol.15, No.4, pp Goldberg, Pinelopi Koujianou and Nina Pavcnik (2007) Disribuional Effecs of Globalizaion in Developing Counries, Journal of Economic Lieraure, Vol. 45, No. 1, pp Harrigan, James and Geoffrey Barrows (2009) Tesing he Theory of Trade Policy: Evidence from he Abrup End of he Mulifiber Arrangemen, Review of Economics and Saisics, Vol. 91, No. 2, pp Scho, Peer (2004) Across-Produc versus wihin-produc Specializaion in Inernaional Trade, Quarerly Journal of Economics, Vol. 119, No.2, pp Zhu, Suzan Chun (2005) Can Produc Cycles Explain Skill Upgrading?, Journal of Inernaional Economics, Vo.60, No.1, pp Verhoogen, Eric A. (2008) Trade, Qualiy Upgrading and Wage Inequaliy in he Mexican Manufacuring Secor, Quarerly Journal of Economics, Vol. 123, No

13 Table 1. Growh rae of expor values in he US and EU markes (%) Madagascar World Bangladesh Cambodia Vienam China Pakisan India Source: UN Comrade (US and EU repors of impor daa) Table 2. Change in uni price (rae of change). US EU World World Madagascar average average Madagascar Noe: Uni price is he value per dozen (he producs repored in he uni oher han dozen are excluded). Source: World Trade Alas (US and EU repors of impor daa) Table 3. Wihin-produc price changes. EU US Dep. var Base year Base year ln (price) ** 0.179** y2010 (0.067) (0.071) (0.086) (0.097) N Noe: This able shows only esimaed coefficiens on y2010 alhough he regression models include a consan, and he models using 2009 as he base year include year dummies for The produc fixed effec is conrolled. Robus sandard errors are repored. ** indicaes significance a 5%. Source: Auhor s calculaion 13

14 Table 4. Difference in difference esimaion on expor value and price. Dependen var: ln(expor value) Dependen var: ln(uni price) USx2010xmada *** (0.418) USx2010xmada 0.237** (0.108) USx (0.193) USx (0.024) madax * (0.300) madax (0.064) y (0.083) unidummy (0.123) y (0.154) y *** (0.017) y (0.159) y *** (0.019) y (0.159) y *** (0.021) y (0.233) y *** (0.019) _cons *** (0.105) y *** (0.020) _cons *** (0.108) Fixed effec Produc*Counry*Marke Fixed effec Produc*Counry*Marke R R rho rho N 9096 N 7153 Noe: Robus sandard errors are repored. ***, **, and * indicae significance a 1%, 5%, and 10%, respecively. Source: Auhor s calculaion Table 5. Composiion and change in employmen by designaion (%). Composiion Rae of change (2008) Toal Male Female Managerial Officer Engineer Qualiy conroller Supervisor Operaor Helper Oher workers Toal Noe: Toal includes foreign workers who are no classified in he gender caegory. Source: Auhors calculaion based on he firm surveys 14

15 Table 6. Esimaion resuls of skill inensiy. Coefficiens Sandard error Descripion of variable US (0.233) y ** (0.177) y ** (0.224) USx (0.466) mngx (0.208) mng: manager offx *** (0.214) off: officer qcx (0.227) qc: qualiy conroller engx *** (0.316) eng: engineer spvx ** (0.232) spv: supervisor oprx (0.198) opr: operaor ohx ** (0.242) oh: oher worker mngx2010xus (0.282) offx2010xus (0.256) qcx2010xus (0.289) engx2010xus * (0.387) spvx2010xus (0.285) oprx2010xus (0.303) ohx2010xus (0.283) _cons (0.498) Fixed Effec firm Rho R N 1377 Noe: Some coefficiens are no repored. Robus sandard errors are repored. ***, **, and * indicae significance a 1%, 5%, and 10%, respecively. Source: Auhor s calculaion 15

16 Table 7. Real wage change by posiion (relaive o 2008) N Manager ** (0.072) (0.114) 262 Officer * (0.051) (0.061) 409 Engineer (0.063) (0.114) 164 Qualiy conroller * (0.050) (0.068) 315 Supervisor 0.111* (0.060) (0.060) 443 Operaor * (0.056) (0.055) 698 Helper (0.054) (0.078) 339 Oher (0.077) (0.081) 637 Noe: Esimaed coefficiens for year dummies based on he regression by posiion. The dependen variable is log of he wage by posiion, gender, and experience. Gender, experience, and firm fixed effecs are conrolled. N represens he number of wage observaions classified by posiion, gender, and experience. ***, **, and * indicae significance a 1%, 5%, and 10%, respecively. 16

17 Table 8. Esimaion resuls of skill premium. Model 1 Model 2 Descripion of variable Sex (0.011) (0.012) exp *** (0.019) 0.051** (0.021) exp *** (0.024) 0.154*** (0.027) price 0.001*** (0.000) y (0.060) (0.072) y (0.093) (0.106) US (0.083) (0.107) USx * (0.155) (0.155) USx (0.099) (0.093) mng 2.006*** (0.146) 2.011*** (0.161) mng: manager mngx ** (0.212) ** (0.233) off 1.309*** (0.089) 1.323*** (0.100) off: officer offx ** (0.114) (0.130) eng 1.020*** (0.107) 1.024*** (0.122) eng: engineer engx (0.182) (0.197) qc 0.635*** (0.077) 0.659*** (0.086) qc: qualiy conroller qcx * (0.104) * (0.116) spv 0.614*** (0.060) 0.606*** (0.058) spv: supervisor spvx (0.096) (0.100) opr 0.350*** (0.053) 0.344*** (0.056) opr: operaor oprx * (0.087) (0.098) oh 0.177*** (0.056) 0.110* (0.059) oh: oher worker ohx (0.093) (0.108) mngxusx (0.472) (0.495) offxusx (0.375) (0.395) engxusx (0.505) (0.507) qcxusx (0.234) (0.286) spvxusx (0.280) (0.299) oprxusx (0.148) (0.168) ohxusx (0.149) (0.139) _cons *** (0.163) *** (0.188) Fixed effec Firm Firm rho R N Noe: Some coefficiens are no repored. Robus sandard errors are repored. ***, **, and * indicae significance a 1%, 5%, and 10%, respecively. Source: Auhor s calculaion 17

18 Figure 1. Garmen expors from Madagascar. 700 Garmen Expors from Madagascar (mil.$) US EU Source: UN Comrade (US and EU repors of impor daa) 18

19 Appendix: Firm Daase The firm survey was conduced from 2009 o 2011 in collaboraion wih he Universiy of Anananarivo. Each survey colleced informaion from he previous fiscal year, namely, FY2008, FY2008, and FY2010. The firs survey colleced informaion on 118 facories: 86 facories of EPZ firms and 32 non-epz firms randomly seleced from EPZ and non-epz firm liss, respecively. The survey included exporing and non-exporing firms, and some non-epz firms were exporers. The survey covered 65.6% of EPZ firms, while he coverage of non-epz firms is unknown due o he incompleeness of he lis. The second survey followed he sample in he firs survey, and a small number of new samples were added in he hird survey. In his paper, we used facories exporing in 2008 regardless of heir expor saus afer The numbers of he samples are 98, 67, and 57, respecively, which reflec he closure of he firms (Table A1). Firm closure was confirmed by elephone and visi. Table A1. Transiion of operaion saus Mid-2010 Mid-2011 Operaion 98 (98) 75 (67) 64 (57) Closure Unknown 8 Toal Noe: The figures in parenheses are he number of facories successfully capured by he survey. Table A2. Marke orienaion US marke Non-US marke Unknown Toal Noe: The figures in he US marke row represen he number of facories supplying a leas some par of heir producs o he US marke. Therefore, he figures in he row of non-us marke are he number of facories supplying no producs o he US marke. 19