Impact of trade liberalization on firm s labour demand by skill: The case of Tunisian manufacturing

Size: px
Start display at page:

Download "Impact of trade liberalization on firm s labour demand by skill: The case of Tunisian manufacturing"

Transcription

1 Impact of trade lberalzaton on frm s labour demand by skll: The case of Tunsan manufacturng Rm Ben Ayed Mouelh October 2003 (Frst draft) (LEA, ISCAE, Unversty of la Manouba, Campus unverstare la Manouba) ou Abstract Ths paper nvestgates the mpact of trade lberalzaton process n Tunsa on employment by dstngushng dfferent sklls and dfferent types of frms usng mcro level data coverng the perod There s consderable dsagreement among analysts on the mpact of recent trade reforms on labour. Our contrbuton to these debates n ths paper s essentally an emprcal ssue. The analyss of a Tunsan frms data may be vewed as an attempt to apprehend how employment n Tunsa, a developng country, adjusted to the trade reforms. Usng a mcro-level detal on ndvdual frms, we are able to trace the relatonshp between changes n trade polces and manufacturng employment at the frm level and by skll. Although trade reforms are generally mplemented at the sector level, ther effects may vary sgnfcantly across frm characterstcs such as ownershp (publc vs. prvate) and degree of export orentaton. We can then measure the effects of trade polcy on employment for dfferent types of frms. We also try to assocate changes n employment drectly wth measure of change n trade protecton, rather than lnk them to changes n mports and exports whch s more common.

2 1. Introducton Tunsa was one of the countres that abandon mport substtuton ndustralzaton as a development strategy. Import substtuton was desgned to protect domestc nfant manufacturng ndustres from mport through the use of protectve nstruments, such as tarffs and quotas controls. Snce 1986, numerous measures have been taken to further lberalse the trade. Ths paper nvestgates the mpact of trade lberalzaton process n Tunsa on employment by dstngushng dfferent sklls and dfferent types of frms usng mcro level data coverng the perod The basc nsght of the H-O model that trade should beneft a country s abundant factor s qute compellng. More recent epsodes of trade lberalzaton appear not to have been assocated wth large mprovements n prospects for the typcal worker ( Hasan 2001). The emprcal evdence concernng the relatonshp between employment and trade reform n developng countres s not defntve. Our goal s to hghlght some contnung puzzles n our understandng of the effects of trade reform. A possble explanaton of the apparent dvergence between the expectatons of lberalzaton advocates and the recent evdence s that over the past couples of decades world technology changed n a way that rased the relatve demand for sklled labor. The net mpact on employment depends on the sze of the two effects, H-O argument and technologcal one. Several studes that measure the mpact of trade reform on employment fnd almost no mpact or small effects on employment. A possble explanaton of the absence of the mpact of trade polces on employment s aggregaton of employment. Ths aggregaton can mask the effects. In ths paper we dsaggregate labor nto sklled and unsklled categores to analyse the effects of trade polces on dfferent sklls. Another possble reason for the lack of an employment response s that labor regulatons, partcularly n developng countres, nhbt the reallocaton of labor (Curre and Harrson 1997). In ths paper, we analyse and we ntroduce ths phenomena. In sum, there s consderable dsagreement among analysts on the mpact of recent trade reforms on labour. Our contrbuton to these debates n ths paper s essentally an emprcal ssue. The analyss of a Tunsan frms data may be vewed as an attempt to apprehend how employment n Tunsa, a developng country, adjusted to the trade reforms. Usng a mcrolevel detal on ndvdual frms, we are able to trace the relatonshp between changes n trade polces and manufacturng employment at the frm level and by skll. Although trade reforms are generally mplemented at the sector level, ther effects may vary sgnfcantly across frm characterstcs such as ownershp (publc vs. prvate) and degree of export orentaton. We can then measure the effects of trade polcy on employment for dfferent types of frms. We also try to assocate changes n employment drectly wth measure of change n trade protecton, rather than lnk them to changes n mports and exports whch s more common. By usng frm-level data we are also able to drectly apply a model derved from the frm s labor demand decson and to control for unobserved, constant frm-level determnants of labor demand usng frm specfc effects models. Introducton of technologcal change, labor adjustment and export orentaton phenomena n the model plays an mportant role and have an mpact on the response of employment to trade reform.

3 The rest of the paper s organzed as follows. Secton 2 presents a revew of trade reform n Tunsa. Secton 3 presents a dscusson of necessary background about the mpact of trade on employment. Secton 4 lays down the man model to be used as framework for the emprcal analyss. Secton 5 presents the data and some basc descrptve statstcs wth some ndcatons about the estmaton method. Secton 6 shows and dscusses the man econometrc results. Fnally, secton 7 concludes. 2. Trade lberalzaton n Tunsa Snce 1986, numerous measures have been taken to further lberalse the trade: the structural adjustment plan (1986), adherence to the GATT (1989), adherence to the OMC (1994) and the sgnature of a free-trade agreement wth the European Unon (1995). The Tunsan trade lberalzaton program can be dvded nto three perods (Cherkaou and Nan 2002): The prestructural adjustment program (SAP) adopted n 1986, durng the SAP perod ( ) and after Before 1986, over 94% of mports were subject to lcensng, and tarff rates averaged 41 percent. In 1987, mport lcensng was gradually decreased and by 1991 only 30 percent of mports was subject to a lcense. The average mport dutes declned from 41 percent to 33 percent n 1987 and to 29 percent n The hghest duty rate was reduced from 200 percent to 43 percent. There was a contnuaton of the removal of lcensng requrements n 1992 and Snce 1995 trade polcy has been domnated by the free trade assocaton agreement wth the European Unon. The agreement mples a removal of tarffs on ndustral mports from Europe over a twelve-year perod. Ths removal started n The scope and speed of ths trade lberalzaton epsode s apparent from table1. The mean effectve rate of protecton 1 fell from 555 n 1985 to 80 n 1991 for the IAA sector, from 203 n 1985 to 58 n 1991 n the ITHC sector and from 203 n 1985 to 49 n 1991 n the ICH sector. Dsaggregated by ndustres, the percentage declnes n effectve rates of protecton, partcularly between 1986 and 1990, are mpressve n all ndustres. <Insert table 1 here> In Summary, durng the ntal perod of trade lberalzaton ( ), the degree of protecton was greatly lowered. However, durng the second perod ( ) the nomnal and the effectve rates of protecton ncreased except for some products. Ths ncrease s explaned by the consequences of the Uruguay Round that transformed non-tarff protecton nto ther tarff equvalent. We have to note that lberalzaton program remans relatvely tmd over ths perod and concerns partcularly equpment and nputs. The government has adopted a more actve lberalzaton polcy after 1995, but there s no plant-level data coverng ths perod n Tunsa. However, our sample perod covers an mportant phase of Tunsan trade reform ( ), relatvely to the protectonst perod before The cross secton dmenson of our data covers frms belongng to dfferent actvtes whch are dfferently protected. Ths suggest that f trade has a sgnfcant effect on employment, t should be apparent n ths data characterzed by the two dmensons: temporal and ndvdual. 1 The effectve rate of protecton s defned as the proportonal ncrease n value added resultng from the mposton of protectve measures It measures the percentage by whch value added can ncrease over the freetrade level as a consequence of a tarff structure. The effectve rate of protecton captures protecton of ntermedate and fnal goods. It also captures tarff or non-tarff protectve measures. A negatve rate mples that nput ndustres are partcularly favoured. Ths negatve rates ndcate hgher tarffs on nput mports than on fnal goods

4 The changes n the level of manufacturng employment by sector n Tunsa over the perod are shown n graph1. We can dentfy a phase of rapd growth after 1987 whch s accelerated after 1992 especally for the textle and ndustres dverses. We can also dentfy the domnant role of the textle sector n terms of job creaton. <Insert graph 1 here> 3. Trade lberalzaton and labour demand by sklls: Some background Let us consder the predctons of the H-O model and Stolper-Samuelson theorem n a smplfed world that has two sectors, two factors of producton (sklled and unsklled labor), and conssts of a developed and a developng country. The developng country s assumed to be relatvely unsklled labor abundant. When the developng country reduces trade barrers on the mported product, the Stolper-Samuelson theorem predcts that the declne n the prce n the mport-competng sector wll hurt the factor of producton used relatvely ntensvely n the producton of the mported good (sklled labor n developng countres) and beneft the factor of producton used ntensvely n the export sector (unsklled labor n developng countres). Edwards (1988), used a three-goods model (mportables, exportables, and nontradables) to analyze labor-market adjustment to changes n the terms of trade and mport tarffs for a small, open economy. The analyss looked at both long-run and short-run and assumed alternatve assumptons regardng wage flexblty. The effects of the fall n the relatve prce of mportables followng lberalzaton n ths type of model are n lne wth those predcted by the Stopler-Samuelson theorem; where exportables are relatvely unsklled labourntensve tarff reducton ncreases demand for the economy s abundant factor. The wthntradeables shft n producton and employment s towards exportables and away from mportables, gven the rse n the relatve prce of exportables. These changes ental an ncrease n the demand for unsklled labor n the unsklled labor abundant country. In developng countres, where unsklled labor s abundant and sklled labor s scarce, trade tends to rase both the demand for unsklled labor and the unsklled wages relatve to the sklled wages. However, ths basc nsght of the H-O model that trade should beneft a country s abundant factor s qute compellng. More recent epsodes of trade lberalzaton appear not to have been assocated wth large mprovements n prospects for the typcal worker ( Hasan 2001), (Pavnck and al 2003). It s hard to avod the concluson that there s a genune conflct of evdence. There are varous factors that may explan the apparent dvergence between the expectatons of lberalzaton advocates and the recent evdence (Wood 1997).. A possble explanaton s that over the past couples of decades world technology changed n a way that rased the relatve demand for sklled labor. Technologcal change has been skll based. Increased openness n a developng country affects the skll structure of labor demand by changng the producton technology avalable through ncreased mports of advanced captal goods or through opportuntes for exporters to learn from foregn buyers and be exposed to foregn markets. The net mpact depends on the sze of the two effects, H-O argument and technologcal one, and on the dfference between domestc and world technology.

5 Another possble explanaton s that realzng the gans from trade as expected by H-O model requres that factors are moble and reallocate from mport competng sectors to exportng sectors. Then labor market regulatons can explan the sluggsh labor market response to trade reforms n some countres, organzed labor s usually vewed as an obstacle to labor market adjustment. In fact, hrng and frng laws may affect labor moblty. Legslatons whch make t dffcult for frms to lay off workers are lkely to mpnge on frm s ablty to reallocate resources to new lnes of producton (Hasan 2001). Regulatons and barrers to labor market moblty are lkely to mpede adjustment and dlute the benefts of trade reform for workers as a whole. Then labor markets were unaffected and frms adjusted to greater competton through other changes as reductons n proft margns (Curre, Harrson, 1997). 4. Model Ths secton outlnes a smple model of employment determnaton whch ncorporates the effects of trade takng nto account the effects of technologcal changes and the delay of adjustment of labor. We begn by assumng that a frm-specfc producton functon can be descrbed by a Cobb- Douglas form as γ α β y t= At Kt Lt (1) where y ndcates the output, K and L are captal and labor nputs, respectvely. Captal stock s supposed to be fxed. α and β are parameters to be estmated representng factor share coeffcents. γ allows for factors affectng and changng the effcency of the producton process (Mlner and Wrght 1998). The factors consdered here are related to trade lberalzaton. These factors vary over tme and across frms (of the same sectors) n the followng manner: At= exp( t γ tdt) tpe δ1 t where tpe s the effectve protecton rate. D t s a dummy varable havng a value of one for the t th tme perod and zero otherwse and where γ t are parameters to be estmated. The dummy varable D t s ntroduced to model pure technology change accordng to the general ndex (GI) approach (Baltag and Grffn 1988). Ths has the advantage of not mposng any structure on the behavour of techncal change. Ths tme dummy model allows for the tme effects to swtch from postve to negatve and back to postve effects. The change n γ t between perods s a measure of the rate of techncal change. Ths can be wrtten as: (3) TC t, t+ 1= γ t+ 1 γ t The hypothess of no techncal change mples that t γ =constant t n the model (2). A frm s derved demand for labour would therefore depend on output, stock of captal and the wage rate ( w ), together wth the effectve protecton rate and tme dummes. (2)

6 However, the assumpton of homogeneous workers n a frm can be consdered as a strong hypothess snce t s lkely that frms employ workers of dfferent sklls. In partcular, frms most often employ both sklled and unsklled workers. As long as data about dfferent categores of workers are avalable, one can estmate dsaggregated labor demand models (Bresson, Kramarz et Sevestre 1992). Ths wll permts us to analyse the effect of trade varables on demand for dfferent categores of labour. Consder that the frms producton technology can be wrtten as 1 2 y t = f( Kt, Lt, Lt, At) (4) 1 Lt the level of employment of sklled workers, workers. 2 L t the level of employment of unsklled We assume that, at frst, frms determne an expected producton and then mnmze costs under the constrant of ther expected level of producton. Then the expresson of the desred * levels of employment for category j L j t j = 1, 2 n terms of ther determnants (captal, producton level.) can be derved from the soluton of the frm optmsaton program. In our context they depend on the producton level, on the captal stock K, on the relatve wages 1 w t ( 2 ) for the two workers categores consdered together wth the trade varables and the wt tme trend (or dummes) whch account for the techncal progress. Snce the wages by skll are not avalable n our data, the relatve wage wll be consdered as fxed over tme and vary only over frms, so that n estmatons ts varaton wll be captured by specfc effects α. We rely on the frm effects and on the tme effects (tme dummes) to absorb the effect of relatve wages over the perod and the frms. The labor demand functon for the category j can be wrtten as L j t = g( yt, Kt, tpet, Dt, α) (5) However, to take nto account the fact that when frms face a change n ther envronment, partcularly trade reforms, they do not necessarly adjust mmedately ther level of employment to the new busness condtons due to the exstence of adjustment costs, namely hrng and frng costs, we use a dynamc adjustment process whch can be represented as L j t j t 1 j* j ( t t 1 j L = λ L L ) (6) Ths would allow us to examne whether a frm s response to trade shocks s related to the speed wth whch t adjusts to changes n desred employment levels. We suppose that the speed of adjustment depends on the skll level j, the dynamcs of adjustment among labor nputs are dfferent. One would expect hrng costs to be larger the hgher the skll of workers, snce tranng costs are expected to be lower for unsklled labor. Furthermore, snce severance pay depends on the worker s earnngs and they depend on hs skll, frng costs wll ncrease wth worker s skll (Borrego, 1998).

7 For analytcal convenence and computatonal easness, adjustment costs were assumed to be symmetrc and quadratc to derve explct partal adjustment models of labor demand (Bresson and all, 1992). Then, the labor demand functon for categore j can be rewrtten as a reduced form equaton for estmaton: j t j j j j j λ j)ln Lt 1 + λ jθ 1 ln yt + λ jθ 2 ln K t + λ jθ 3 tpe t + λ jθ 4 ln L = (1 t+ α + v j=1,2 (7) Coeffcents are specfc to the employment category j. α s the frm s specfc effect. V t s the classcal error term. 5. Data and estmaton method 5.1. Data The data used n ths study are taken from the natonal annual survey report on frms (NASRF) carred out by the Tunsan Natonal Insttute of Statstcs (TNIS). The data covers nearly all frms for dfferent sectors (ntally 5000) over the perod Although the data s collected by ntervews, the Tunsan NASRF stll suffers from a non-response mean rate of 57.5% for the perod Unfortunately, for the perod , the TNIS does not report any nformaton concernng both the non-response rate of frms and the reasons of non-response. In the frst stage, the data set has been cleaned from observatons whch could be seen as erroneous or whch were clearly outlers. We have taken out frms observed less than 5 perods, ths s an unavodable consequence of the dynamc nature of the model and panel data technques used. In the estmaton of a dynamc functon, we also requred that all sample frms be observed consecutvely. Thereby, our emprcal analyss s based on an unbalanced panel consstng of a sample of 660 frms from the IAA, ICH and ITHC sectors (see table 3) wth between 5 and 12 annual and contnuous observatons over the perod (see table 2) 2. Ths sample s the only plant-level data avalable n Tunsa. <Insert table 2 here> <Insert table 3 here> j j t The data set ncludes: value added (y), captal stock (K) evaluated at hstorcal values, labour (number of employees L), a decomposton of labour nto sklled (L 1 ) and unsklled labour ( L 2 ), and the total expendture on personnel all categores s ncluded. The actvtes of unsklled workers nclude machne operaton, producton supervson, repar, mantenance and cleanng; those of sklled workers nclude management, admnstraton, and general offce tasks. The number of employees s adjusted for whether t s part or fulltme equvalent employment. Labour costs for sklled and unsklled employees are combned, makng t dffcult to dstngush between the wages of the two types of workers. An average wage was calculated by dvdng the total labour bll by the total number of employees. Ths wages 2 Two years are lost n constructng lags and takng frst dfferences, so that the estmaton covers the perod

8 measure reflects both changes n worker remuneraton and changes n the composton of the labour force. The frm s actvty s descrbed by a two-dgt Tunsan nomenclature of economc actvtes. An aggregaton to one-dgt codes leads to the 6 ndustral sectors. As shown n table 4, our sample represents mportant percentages relatvely to ndustral employment. <Insert table 4 here> Ths table shows that our sample covers 27.5% of ndustral employment and s very representatve of the IME and ICH actvtes. The last two columns show that the dstrbuton of employment across sectors n our sample and n the ndustral data are comparable. Snce we are dealng wth a pooled sample of ndvdual frm data, the ssue of frm heterogenety s an mportant one. In our sample heterogenety can arse from the fact that frms n dfferent sectors of the ndustry can be expected to operate under dfferent technologes whch leads to dfferences n labour demand functons. Ths wll be assumed to be a source of ndustry-specfc heterogenety, although n practce ndustres cannot be defned so fnely that frm-specfc dfferences n technology wthn a partcular branch are elmnated. <Insert table 5 here> Table 5 provdes an overvew of the frms n our data set. The frst column shows the mean characterstcs of all frms n the sample, whle columns two to sx llustrate the dfference between varous groups of frms. Ths table shows that prvate frms are much larger than publcly owned frms n our sample and have a hgher share of sklled workers. In addton, prvate ndustral frms pad hgher wages. Ths confrms the fact that publc sector employees are underpad. The last three columns break out frms n the three ndustry sectors consdered n ths study. Export-orented frms belongng to the ITHC tend to be smaller, to have a smaller share of sklled workers and to have lower wages than other frms belongng to IAA and ICH ndustres. Ths s n conformty wth the realty of the ITHC sector whch s laborntensve, employng an unsklled workng populaton. Small frms and famly owned frms domnate the Tunsan textle ndustry. ITHC s also domnated by frms that are totally orented towards exportaton, whch specalze n subcontractng for foregn brands and platforms. Frms n ICH ndustres tend to be the larger ones, to have the hghest share of sklled workers and to have the hghest wages. <Insert table 6 here> Table 6 shows the rato of sklled to unsklled employment for all frms. Between 1983 and 1993 the rato of sklled to unsklled employment ncreased from 0.14 to Ths rato has doubled n the IAA and ICH sectors between 1983 and <Insert table 7 here>

9 Table 7 shows relatve employment by one dgt ndustry for all frms n 1983 and Two ndustres experenced a sgnfcatve ncrease n the relatve employment of sklled labor (IAA, ICH), whle one ndustry (ITHC) experenced a declne. The change n composton towards less use of unsklled workers relatvely to sklled ones s farly vsble for the IAA and ICH sectors. One possble explanaton for the rse n the relatve demand for sklled labour s the skll-based techncal change whch s accelerated by trade lberalzaton. In fact, trade s the vehcle through whch new technologes enter most developng countres. Ths s because most nnovatons occur abroad and are mported n the form of machnery and equpment. The relaxaton of trade barrers reduces the prce of mportng technology from abroad and could lead to surges n foregn drect nvestment and exports. If foregn frms and exporters demand more sklled workers, due to more advanced technology or dfferng product mxes, then ths could explan a shft towards sklled employees (Harrson 1995). Trade lberalzaton permts the acceleraton of the mported physcal captal stock to GDP; the attendant captal-skll complementartes and bundled technology would then rase the relatve demand for sklled workers. Ths explanaton s also compatble wth a greater degree of ndrect skll nputs for exports requred for marketng and dstrbuton (OCDE 1996). Our plant-level data were combned wth measures on the effectve rate of protecton (tpe) calculated by the Insttut d Econome Quanttatve (IEQ) at the 2-dgt sector level and also at one dgt level. Formally the effectve rate of protecton s gven by the proportonal ncrease n value added resultng from the mposton of protectve measures. Thus the effectve rate of protecton s ncreased f the value added n the consdered actvty can be rased as a consequence of the mposton of any tarff or non-tarff protectve measures. In prncple, effectve protecton fgures, whch s a drect measure of trade polcy, should summarze all of the product market forces that drve a wedge between domestc and world prces (Tybout and Westbrook 1995). The tpe measure combnes nformaton on tarffs wth non-tarff barrers. However, It seems worthwhle to also consder alternatve measures of exposure to nternatonal competton: mport lcence coverage, offcal tarff rates, mport penetraton, but these measures are not avalable. We have already dscuss the large decrease n tpe durng the frst observed perod (84-90) and the somewhat stablsaton between Our sample perod covers an mportant phase of Tunsan trade reform ( ), relatvely to the protectonst perod before The cross secton dmenson of our data covers frms belongng to dfferent actvtes whch are dfferently protected. Ths suggest that f trade has a sgnfcant effect on employment, t should be apparent n ths data characterzed by the two dmensons: temporal and ndvdual. However, we have to note that lberalzaton program remans relatvely tmd over ths perod and concerns partcularly equpment and nputs and not fnal goods. Ths s manly due to the government s preoccupaton wth mantanng socal stablty and preparng companes for competton. The government has adopted a more actve lberalzaton polcy after 1995, but our plant-level data don t cover ths perod Estmaton method We wll estmate the dynamc model specfed n equatons (7) by GMM as suggested by Blundell and Bond (1998) 3, wthout assumng any dstrbuton for the error terms, takng nto 3 See appendce for more detals on ths estmaton method

10 consderaton the dynamc form and the presence of varables that are nvarants over tme. Estmaton of the dynamc error component model s consdered usng an alternatve to the standard frst dfferenced GMM estmator of Arellano and Bond (1991). It s a system GMM estmator deduced from a system of equatons n frst dfferences and n levels. Ths estmator s defned under extra moment restrctons that are avalable under qute reasonable condtons relatng to the propertes of the ntal condton process. Explotng these extra moment restrctons offers effcency gans and permts the dentfcaton of the effects of tme nvarant varables. 6. Results The emprcal results are based on the estmaton of equatons (7) descrbed above. The dependent varable n equaton (7) s annual sklled, unsklled or total labour. The ndependent varables nclude the output, the captal stock and the protecton varable: the effectve rate of protecton. All varables are expressed n logarthms. We also nclude year effects to control for common aggregate shocks that are not otherwse captured by our specfcaton. In addton, to allow for the possblty of slow employment adjustment, we consder specfcatons that nclude lagged employment among the rght-hand sde varables. Introducng output among the lst of explanatory varables n the employment equatons, as s done n the regresson equatons 7, gves us a way of assessng the mpact of openness after controllng for state of economc actvty. All regressons nclude year ndcators. Note that t s crucal to control for unobserved yearspecfc varables that could nfluence employment concurrently wth tarffs. Wthout controllng for year effects (macroeconomc shocks as exchange rate devaluaton), one would falsely conclude that there s (or there s not) a relaton between tarffs and employment. Table 8 reports the estmaton results of the dynamc employment functon defned by equatons (7) and estmated by the system GMM method. The frst column reports the results for the total labour, the second reports the results for the unsklled labour demand and the 4 thrd one reports the results for the sklled labour demand. The test of the hypothess λ = 1 that there s no dfference between target and actual labour use reject t at the 5 percent level of sgnfcance for the dfferent categores of labour. Ths hypothess s not approprate for the data at hand. <Insert table 8 here> Output, captal and labour are consdered as predetermned and correlated wth frm-specfc effects error. Consequently, the nstruments used for equatons n frst dfferences are observatons of captal and labour dated (t-2) and earler n addton to output dated (t-2) and earler. For the system GMM estmator, we add the observatons on ( k, l, y) dated (t-1) and tme dummes as nstruments for the equatons n levels. The valdty of the nstrument set s checked usng a sargan test. Ths s asymptotcally dstrbuted as ch-squared under the null. The nstruments used n the frst dfferenced GMM or n the system GMM are not rejected by the Sargan test of over-dentfyng. Tests of no 4 For testng ths hypothess, a Student test of sgnfcatvty of the coeffcent assocated to the lagged varable, that s(1- λ ), s suffcent.

11 seral correlaton n the v t (M 1 and M 2 ) 5 provde evdence to suggest that ths assumpton of serally uncorrelated errors s approprate n the dynamc model as s shown n the dfferent colulmns 6. We note that the dynamc labour equatons perform well n conventonal statstcal terms wth no second order seral correlaton and wth a sargan test for nstrumental valdty ndcatng that the nstrument set and the resduals are not correlated. However, the test of no seral correlaton of resduals s rejected n the statc model. Ths ndcates the presence of msspecfcaton may be due to omsson of the lagged dependent varable. The dfference between the results for the statc model and the dynamc are not neglgble. The coeffcents values estmates are dfferent wth more precson when costly adjustment of labour s accounted for n the model. The estmaton of the system has made the dentfcaton of tme-nvarant varables effects possble n contrast to the frst dfferenced estmaton. The coeffcent estmates assocated to the total labour shows that the coeffcent on the lagged dependent varable s of 0.88 and strongly sgnfcant. The speed of adjustment s of 0.12, that s frms adjust only 12 percent of ther devatons off the optmalty n one year. Ths confrms the fact that labour takes tme to reach ts optmal level whch s consstent wth the exstence of substantal employment protecton. Ths s more pronounced for the sklled labour confrmng the fact that adjustment costs are dfferent for dfferent sklls. Ths also confrms the fact that to dentfy the mpact of trade reform on labour we must take nto account the adjustment delays of the labour. It suggests that the dynamcs of the equatons are mportant. As regards the employment equatons, we obtan farly reasonable estmates. The coeffcent on the output varable s postve and sgnfcant for the dfferent categores of labour. Introducng the output varable among the explanatory varables n the employment equatons, as s done n our regresson equatons, gves us a way of assessng the mpact of openness on employment after controllng for state of economc actvty. The coeffcent on captal s not sgnfcant for the unsklled labour but sgnfcant and postve for the sklled labour demand. Ths result ndcates that sklled labour and captal are complements. Wth respect to the protecton varable, our estmates are reasonable. We obtan a negatve and sgnfcant coeffcent on the varable that captures protecton (tpe)for the dfferent categores of labour. Ths suggests that a reducton n the protecton rate s assocated wth an ncrease n labour demand. The coeffcent on protecton, -0.02, mples that a reducton n protecton of 10 percentage ponts would lead to an ncrease n employment of 0.2 percent. The results suggest that unsklled employment s more responsve to changes n protecton levels than that of sklled employment. The estmated elastcty of unsklled worker employment wth respect to a change n protecton levels s 0.05, whereas that for sklled employment s It appears that frms adjusted to a contracton of the domestc market usng other means of adjustment than cuttng employment. Ths could reflect an mprovement n productvty, effcency and compettvty resultng from more exposure and competton from abroad. Another explanaton s the release of the nput and captal constrant that allows the frm to use more (mported) captal equpment and other ntermedate nputs at lower prces after lberalzaton. It s worth mentonng that the Tunsan ndustres, partcularly the clothng ndustry, rely heavly on the mportaton of raw materals n order to meet producton needs. 5 See the appendx for precson about these two tests M 1 and M 2 6 These tests statstcs, dstrbuted normally under the null of no seral correlaton, are calculated and presented n the table 3.

12 In partcular, ncreased exposure to trade could have ncreased the demand for sklled labor va skll-based technologcal change. Wood (1995) argue that frms mght adapt skll-based technology n response to ntensfed competton from abroad. Lower prces of foregn machnery and technology provde an addtonal ncentve for the frms to adapt new technology and then to ncrease ther demand for sklled labor. Another explanaton for the postve responses of employment s that the supply of labour ncreased dramatcally n Tunsa as women and more educated people entered the labour market n the consdered perod. Ths allowed employment n the dfferent sectors and for the dfferent sklls to be ncreased. However, the results change somewhat when we consder dfferent types of frms. In tables 9 and 10, we report estmatons for dfferent types of frms. The results suggest that frm orentaton and frm characterstcs sgnfcantly affect both the magntude and the sgn of employment responses to trade polcy. Columns 5 to 7 n table 9 show estmates for exportng frms (wth mean export shares greater than 80%), the results suggest that decreases n protecton were assocated wth sgnfcant ncreases n employment for the dfferent categores. Columns 2 to 4 n table 9 show estmates for domestcally orented frms. The coeffcent on protecton s postve and sgnfcant suggestng that decreases n protecton were assocated wth sgnfcant decreases n employment. It appears that domestcally orented frms adjusted to a contracton of the domestc market by cuttng employment, whle exporters used other means of adjustment. Trade reforms may have ncreased exports by reducng tarffs on mported nputs. Ths suggest that export-orented frms may have been net benefcares of the whole package of new polces. Ths can be also assocated to the nsttutonal reforms whch made t attractve to produce for the export market. <Insert table 9 here> In summary, the effects of openness on exportng manufacturng frms are postve rsng employment as standard models of trade would predct for these unsklled labour-ntensve frms n a developng country. Ths effects are negatve for the domestcally orented frms leadng to some reducton n employment. It s nterestng to note that, sometmes, trade varable has no effects on aggregate employment but has effects on sklled and unsklled labour separately. Hence, the effects of trade lberalzaton may not show up n global employment, ths justfes the desegregaton of employment nto sklled and unsklled one. The response of labour to output s more pronounced n the domestcally- orented frms than n the exportng ones. The results obtaned by dstngushng prvate from publc frms (table 10) suggest that employment n the prvate and publc frms react n a comparable manner to the measures of trade polcy. For the majorty of plants, the employment of sklled workers s not very responsve to changes n protecton levels. The coeffcents on effectve protecton rate are statstcally nsgnfcant, wth pont estmates close to zero. However, the effects of trade

13 reform are postve, rsng the demand for unsklled labor n the prvate and n the publc plants. The ncrease n employment due to a reducton n tarff protecton s hgher for prvate frms. Reductons n quota coverage and tarffs on mported nputs are assocated wth ncreases n unsklled labor demand perhaps reflectng ncreases n actvty levels resultng from greater access to mported captal or other nputs n the prvate sector. However, the sgnfcant ncrease n the unsklled labor demand assocated to decreases n protecton level n the publc sector s consstent wth the dea that publc employment acts as socal safety net, absorbng employees n responses to an ncrease n labour supply. <Insert table 10 here> The coeffcents on captal are postve and statstcally sgnfcant for the sklled labor demand whch s consstent wth the hypothess of complementarty between captal and sklled workers. Total labor adjustment appear to be more sluggsh n the publc frms than n the prvate one. It s possble that publc sector are more constraned n ther ablty to fre workers than prvate frms, whch s consstent wth the exstence of substantal employment protecton n the publc sector. Publc sector act usually as a socal safety. In order to test whether labor market adjustment n Tunsa s consstent wth mplcatons of Hecksher-Ohln adjustments to trade reform, we also dstngush between ndustres. Industres are classfed as mportables or exportables on the bass of nformaton about market orentaton, the sze of the mport and export share and the polcy regme. Textles, clothng and leather (ITHC) are among the exportable sectors, whle food (IAA) and chemcal (ICH) ndustres are among mportables sectors. The stolper-samuelson theorem predcts that the declne n the prce n the mport-competng sector wll hurt the factor of producton used relatvely ntensvely n the producton of the mported good (sklled labor) and beneft the factor of producton used ntensvely n the export sector (unsklled labor). So let us understand how adjustment occurred at the sector level to analyze whether certan sectors benefted more from the lberalzaton. In fact, much of the changes n employment followng reforms could occur through wthn ndustres. Hence, the effects of trade lberalzaton may not show up n net ndustry employment. <Insert table 11 here> The effects of a fall n the relatve prce of mportable followng lberalzaton are n lne wth those predcted by the model of Edwards (1988). Ths model suggests that the mpact of tarff reductons would serve to reduce the prce of mportables relatve to that of exportables, and lead to a swtch of producton and employment n favour of exportables. Lookng at the estmated responses for exportables (ITHC) (table 11), t may be seen that these correspond to those predcted theoretcally. When exportables (ITHC here) are relatvely unsklled-labour ntensve, tarff reducton ncreases demand for the economy s abundant factor and the wthntradeables shft n employment s towards exportables and away from mportables (IAA and ICH), gven the rse n the relatve prce of exportable. Trade reform should restore the comparatve advantage to unsklled labour ntensve goods (ITHC) and mples a reallocaton of output toward unsklled labour ntensve frms.

14 Fnally, to nvestgate whether trade reforms were assocated wth the ncrease n the share of sklled workers observed n the IAA and ICH sectors (see tables 6 and 7), we regress the share of sklled workers aganst ndustry effectve rate of protecton and tme ndcators. The results ndcate that the share of sklled workers s not drectly related (n the dfferent sectors: ITHC, ICH and IAA) to protecton. These results are n conformty wth the Stolper- Samuelson theorem and suggest that the ncrease n the share of sklled workers partcularly observed n IAA and n ICH (between 1983 and 1993) cannot be attrbuted drectly to lberalzaton. However, ths s consstent wth skll based technologcal change. Sklled based technologcal change could have been partcularly nduced by changes n foregn competton, so that trade lberalsaton may have an ndrect and not a drect effect on the rse of skll n IAA and ICH sectors. 7. Concluson Ths artcle presented mcro-level evdence regardng the connecton between trade polcy and ndustral sector employment. Ths artcle s one of only a few studes that has examned ths ssue at the plant level and that are based on panel data and hence able to follow ndvdual plants through tme. It s the only study to address the effect of trade reform on employment by sklls and usng frms data n Tunsa. However, lmted access to qualty trade polcy measures prevented us from fully explotng the panel nature of the data. The results suggest, souvent, that reductons n tarffs and non tarffs levels conducted n ths frst phase of lberalzaton n Tunsa are assocated wth moderate ncrease n frm labor demand. A 10 pont reducton n TPE s assocated wth a 2 to 3% ncrease labour demand. When equatons are estmated separately for unsklled and sklled labor, the paper fnds that employment of the former s sgnfcantly more responsve to changes n protecton levels than those of the latter. The results also suggest that the mpact of trade lberalzaton on labour demand depends on the characterstcs of frms. In partcular, the estmates obtaned suggest that trade lberalzaton have benefcal effects on employment for exportng-frms. Conversely, trade lberalzaton have negatve effects on employment for domestcally orented frms. In sum, the trade lberalzaton have effects on labor market adjustment n Tunsa. It seems to have more effects on unsklled labour than on sklled labor, ths justfy the consderaton of two sklls. The exportable sectors react by ncreasng unsklled labor demand and the mportable react by decreasng unsklled labor demand. Ths s some what consstent wth the predcton of H-O model. The ncrease n demand of sklled labor n reacton to trade lberalzaton sometmes observed can be assocated to skll based technologcal change that was nduced or accelerated by trade reforms. Sklled-based technologcal change could have been partally nduced by changes n foregn competton.

15 Table 1: Effectve rates of protecton n Tunsa n percentage terms IAA IMCCV IME ICH ITHC ID Industres manufacturéres Graph 1:employment trend for sx ndustres $ %$ $ %$ $ %$ $ %$ $ %$ Ind. Manuf. Dverses Textles Habllement & Cur Ind. Chmques Ind. Mecanques & Electrques Materaux de Constructon C.& V. Ind.Agro -Almenta res $ %$ 0 %$ Table 2 : Number of frms by perods Perods Frms Table 3: Number of frms observed by ndustry

16 ITHC IAA ICH Table 4:Representatvty of the sample IAA ICH ITHC total Percentage of sectoral employment 31% 62% 18.5% 27.5% Percentage of sample employment 12.5% 10% 33.7%.. Percentage ndustral employment 11% 4.3% 50%.. of Table 5: Varables means Varables All frms Prvate frms Employment 98.4 (203) 106 (214) Sklled 14.7 (62) 16.1 (66) Unsklled 83.7 (152) 90 (159) Rato Publc frms 39.3 (66) 4 (5.8) 36 (62) 0.11 ITHC IAA ICH 110 (157) 11.7 (34) 98 (129) (166) 12.8 (5.1) 57 (129) (366) 29 (130) 85.6 (244) Added value Captal Stock ( ) ( ) ( ) ( ) (323527) ( ) ( ) ( ) ( ) ( ) ( ) ( ) Wage 2731 (1708) 2804 (1775) 2189 (909) 2163 (1091) 3307 (1960) 3618 (2171) Exportng rato 0.29 (0.39) 0.31 (0.39) 0.15 (0.31) 0.46 (0.41) (0.22) 0.15 (0.29)

17 Table 6: Rato of sklled workers to unsklled workers (n means) year 3 ndustres ITHC ICH IAA Table 7: Rato of sklled to total workers year 3 ndustres ITHC ICH IAA Table 8: Estmaton results for all frms (660 frms) Coeffcents Total labour Unsklled labour Sklled labour (L) (LU) (LS) Ln L(-1) 0.88 (0.023) t= (0.019) t= (0.02) t=27.6 Ln y (0.026) (0.027) t= (0.027) t=10.3 Ln K (0.019) t= (0.026) t=-0.42 (0.024) t=6.44 Ln tpe (0.006) t= (0.008) t= (0.0079) t= Year effects yes Yes yes M1:1 st order seral correlaton M2: 2 nd order seral correlaton Sargan nstrumental valdty test (df) (117) (117) (172)

18 Tableau 9: estmaton results for exportng and domestcally orented frms Coeffcents Domestcally orented Exportng frms L total L unsklled L sklled L total L unsklled Ln L(-1) (0.018) (0.013) (0.029) (0.021) t= (0.024) Ln y 0.21 (0.022) 9.79 Ln k (0.017) Ln tpe (0.010) (0.019) (0.016) (0.009) (0.05) (0.031) (0.015) (0.020) t= (0.023) (0.0087) (0.032) (0.012) L sklled 0.58 (0.022) t= (0.040) (0.041) (0.016) Year effects Yes Yes Yes yes yes yes M1:1 st order seral correlaton M2: 2 nd order seral correlaton Sargan nstrumental valdty test (df) (62) (62) (62) (117) (117) (117) Tableau 10: Estmaton results for prvate and publc frms Coeffcents Prvate frms Publc frms L total L unsklled L sklled L total L unsklled Ln L(-1) (0.022) (0.021) (0.028) (0.038) (0.031) Ln y (0.021) 4.86 Ln k (0.020) Ln tpe (0.0064) (0.03) (0.034) (0.013) (0.055) (0.044) (0.012) (0.037) (0.015) (0.006) (0.037) (0.027) (0.008) L sklled 0.61 (0.012) (0.047) (0.042) (0.027) 0.56 Year effects Yes yes Yes yes yes yes M1:1 st order seral correlaton M2: 2 nd order seral correlaton Sargan nstrumental valdty test (df) (117) (117) (117) (62) (62) (62)

19 Table 11: Estmaton results for the dfferent sectors Coeffcents ITHC IAA ICH L total Ln L(-1) 0.83 (0.043) Ln y (0.045) 3.25 Ln k (0.032) Ln tpe (0.0098) L unskll ed 0.62 (0.049) (0.061) (0.056) (0.014) L sklled 0.52 (0.037) (0.10) (0.081) (0.023) 1.32 L total 0.78 (0.029) (0.038) (0.042) (0.0062) L unsklled 0.44 (0.016) (0.01) (0.02) (0.003) L sklled 0.57 (0.025) (0.021) (0.005) L total 0.77 (0.018) (0.011) (0.012) (0.0038) L unsklled 0.23 (0.020) (0.012) L sklled 0.58 (0.019) (0.034) (0.026) (0.016) Year effects yes yes yes yes yes yes yes yes yes M M : Sargan nstrumental valdty test (df) (62) (62) (62) (62) (62) (62) (51) (51) (51)

20 References Arellano, M. and S.Bond Some tests of specfcaton for panel data: Monte Carlo evdence and an applcaton to employment equatons, Revew of Economc studes, vol.58, pp Blundell, R and S.Bond Intal condtons and moment restrctons n dynamc panel data models, Journal of Econometrcs, vol 87, pp Bresson, G., F.Kramarz and P.Sevestre Dynamc labor demand models, n Matyas.L and P.Sevestre, The Econometrcs of Panel Data, Kluwer Academc Publshers, pp Baltag, B.H and J.M. Grffn A general ndex of techncal change, Journal of poltcal Economy, vol 96, pp Borrego (1998) Cherkaou, M and A.Nan Trade lberalzaton, real exchange rate and export dversfcaton n selected North Afrcan Economes, Paper presented n the Eghth annual conference of ERF. Curre, J. and A.Harrson Sharng the costs: the mpact of trade reform on captal and labor n Morocco, Journal of Labor Economcs, vol 15, pp Edwards, S Terms of trade, Tarffs, and labor market adjustment n developng countres, the World Bank. Hasan, R The mpact of trade labor market regulatons on employment and wages: Evdence from developng countres, East-west center workng paper. Levnsohn, J Employment responses to nternatonal lberalzaton n chle, Journal of Internatonal Economcs, vol 47, pp OCDE Impact of trade upon wages n developng countres, Document technque. Pavnck, N., A.Blom, P.Goldberg. and N.Schady Trade lberalzaton and labor market n Brazl, Polcy research workng paper. Mlner, C. and P.Wrght Modelng labour market adjustment to trade lberalzaton n an ndustralsng economy, The Economc Journal, vol 108, pp Wood, A Openess and wage nequalty n developng countres: The Latn Amercan challenge to East Asan conventonal wsdom, The World Bank Economc Revew, vol 11, number 1. Tybout, J. and D.Westbrook Trade lberalzaton and the dmensons of effcency change n Mexcan manufacturng ndustres, Journal of Internatonal Economcs, vol 39, pp

21 Appendce: Estmaton methods The labour demand functon specfed n equatons (7) s charactersed by the presence of the lagged dependent varable n addton to tme-varant and tme-nvarant varables (actvty) among the regressors. Ths s presented as follows y = α + λy ρ + t t 1 + xt ' β + F ' ut =1, N et t=2,..t (1) where ' x t s 1xk vector of explanatory varables, F s a vector of tme-nvarant regressors, ( β, ρ) are the vectors of coeffcents, α and λ are scalars. T s the number of ways the frm belongs. The error term s specfed as follows u + t = η vt where η = µ α ( α a µ ) = 0 v t represents the effects, whch can not be controlled by the frms and are assumed to be ndependently and dentcally dstrbuted, v t d(0, σ v ), η s a specfc term assumed to be 2 d (0, ση ). The model (1) can be rewrtten as a standard error component model y = w ' θ + η + v Where w = (1, y, x ', F ') (2) t t t t ' t 1 t We suppose the famlar error components structure as t appears n Ahn and Schmdt (1995) where ) For all, v t s uncorrelated wth η for all t. E( v t η ) = 0 for =1,.,N and t=1,,t. ) For all, the vt are mutually uncorrelated. E( v t v s ) = 0 for =1,,N and t s. In addton, there s the assumpton concernng the ntal condtons y 0 (Ahn and Schmdt 1995,1997) ) For all, vt s uncorrelated wth y 0 for all t. Snce y t s a functon of η, y t 1 s also a functon of η. Therefore, y t 1, a rght-hand regressor n (1), s correlated wth the error term. Ths renders the classcal estmator based and nconsstent (Baltag 1995). The wdely used estmator n ths context s obtaned by generalsed method of moments (GMM) after frst dfferencng to elmnate the correlated ndvdual specfc effects. Lagged levels of y t are used as nstruments for equatons n frst dfferences (Arellano and Bond 1991). The frst dfference transformaton wpes out the tmenvarant varables, consequently the outcomng effcency measure wll be contamnated by 2