Firm Productivity in Jamaica: An Investigation of Total Factor Productivity with Micro-Level Firm Data.

Size: px
Start display at page:

Download "Firm Productivity in Jamaica: An Investigation of Total Factor Productivity with Micro-Level Firm Data."

Transcription

1 Aprl 2011 (DRAFT) Frm Productvty n Jamaca: An Investgaton of Total Factor Productvty wth Mcro-Level Frm Data. Taff Bryson 1 Research Servces Department Research and Economc Programmng Dvson Bank of Jamaca Abstract Ths paper examnes productvty at the frm level usng panel data of publcly lsted companes n Jamaca between 1998 and The paper uses a Taylor seres expanson to approxmate the two nput CES producton technology from whch estmates of TFP are derved. The estmates of TFP were then used n a dynamc panel estmaton model wth factors thought to nfluence productvty changes. The analyss fnds that low productvty among frms n Jamaca s due to the low skll level, whch lmts the postve mpact on productvty growth of technology and economes of scale. The paper posts that mprovng percentage of sklled workers level by 1 per cent can ncrease productvty growth by 0.03 per cent. It also concludes that ncreased capacty utlzaton has leads to producton growth. The development of human captal to boost labour productvty remans a key area to be addressed. JEL classfcaton: C23, O14, D24, J24 Keywords: Panel Data, Total Factor Productvty 1 The author would lke to thank all the staff of the Research Servces Department of the Bank of Jamaca for ther nvaluable comments and contrbutons to the development of ths paper. All errors are, however, my own. The vews expressed n ths paper are those of the author and do not necessarly reflect those of the Bank of Jamaca.

2 Table of Contents 1. Introducton Revew of Lterature Theoretcal Framework The Emprcal Analyss Results Summary and Concludng Remarks References... 17

3 Introducton The rate of economc growth depends on the effcent use of a country s resources as well as ts ablty to acqure and apply modern technology (Ayanwale and Bamre, 2004). In both developed and developng countres, productvty generated by these nnovatve actvtes s seen as the lnk between nvestment and economc growth. As such, an examnaton of the determnants of productvty at the macro and frm levels s essental to the desgn of a country s growth and development strategy. Ths analyss s mportant for Jamaca as despte sgnfcant nvestment averagng 25.1 per cent of GDP over the last 20 years, economc growth has averaged a mere 0.7 per cent. Ths s n contrast to countres such as Japan, New Zealand and Trndad & Tobago that had smlar levels of nvestment. Usng data for 23 publcly lsted companes on the Jamaca Stock Exchange over the perod 1998 to 2009, the paper calculates productvty at the frm level n an attempt to derve prelmnary nsght nto the determnants of frm level effcency n the Jamacan economy. The study examnes frm-level productvty as opposed to aggregate productvty, whch has been found to mask substantal dfferences at the ndustry level (Bernard and Jones, 1996). Admttedly, the number of frms on the Jamaca Stock Exchange s lmted; however, these frms are large players n ther respectve ndustres and behave n a typcal manner. Consequently, the paper employs the two-step GMM estmator gven that the avalable data are yearly and capture a small perod from Addtonally, Wndmejer s bas-corrected estmators are derved, to mprove the estmatons. In ths context, the paper s nterested n answerng a few questons. Frst, how does productvty levels vary durng the revew perod and what accounts for these varatons. Secondly, what varables have mpacted frms productvty? And fnally does the productvty of the publcly lsted frms reflect the aggregate productvty level? The man results of ths paper are summarzed as follows: (1) Captal plays a more sgnfcant role n the producton process than labour. (2) The low skll level of labour s a major determnant of low total factor productvty (TFP) growth. (3) The use of technology and economes of scale postvely nfluence productvty growth. 1

4 The rest of the paper s organzed as follows: Secton II outlnes smlar emprcal studes. Secton III outlnes the theoretcal framework for the emprcal analyss. Secton IV contans a descrpton of the data. Fnally, Secton V presents the results and some concludng remarks and possble extensons to the current work. 1.0 Revew of Lterature Estmatng growth n TFP s essental n assessng economc performance, although the process of estmaton s at tmes dffcult. The dffculty n estmatng and nterpretng TFP les n the fact that dfferences n the assumptons about factor nputs or the functonal form of TFP can yeld dfferent estmates of TFP growth. The startng pont for estmatng TFP s a producton functon that represents how nputs, physcal captal and human captal are combned to produce output. Ghosh and Kraay (2000) note that the estmated productvty growth, beng not drectly observable, depends on assumptons of constant returns to scale as well as constant elastcty of substtuton between nputs over tme. Consequently, estmates of the varaton n TFP growth overtme can be qute senstve to the perod for whch they are calculated, and the results should be thoroughly checked for robustness. The authors also note that whle nterpretng estmates of TFP growth at the frm level s straghtforward, the estmate at aggregate level s less clear 6. Accordngly, t s mportant to dstngush between the dfferent sources of TFP growth as there are lkely lmts to the gans from structural change once resources have been reallocated to the most effcent sectors. Gven the aforementoned apprehensons, Ghosh and Kraay (2000) propose two approaches to estmate TFP growth. Frstly, estmate the producton functon drectly, thereby mposng the assumptons of constant returns to scale and prefect competton. Ths approach, however, s noted by the authors as beng lmted by the paucty of relable data on wages along wth the fact that the underlne assumptons seem mplausble of many developng countres. The second approach, whch s employed n 6 The authors purport that at the aggregate level, f resources shft from less productve to more productve frms faster growth n aggregate TFP may be observed even n the absence of techncal progress at the frm level. 2

5 ths paper, nvolves estmatng the producton functon econometrcally, n whch case the resduals n the regresson are nterpreted as the TFP growth. In examnng the lterature on productvty there are two man areas of nterest. Frstly, t s mportant to note the functonal form of the producton functon as well as the assumptons used to derve estmates of TFP. Secondly, attenton s gven to the analyss of factors that are thought to possbly nfluence changes n the estmated productvty. Wthn producton theory, the Constant Elastcty of Substtuton (CES) producton functon, the Cobb-Douglas (CD) and the translog functon are wdely appled n the lterature. The translog form and the logarthm of the CD, beng lnear, are easy to estmate, whle the CES form s non-lnear and ts estmaton ncludes more complcated technques. Algebracally, the CD producton functon s merely the CES producton functon under the assumpton of constant substtuton of producton nputs. In the case of the CD, the estmaton of productvty by ordnary least squares (OLS) wll yeld based parameter estmates because t does not account for the unobserved productvty shocks (Yasar, et al., 2008). The smultanety bas arses, as the dsturbance n the producton functon s observed by the frm but not by the econometrcan. Blundell and Bond (1988) propose usng the Arellano and Bond (1991) estmator to correct for the bas. The Arellano and Bond (1991) estmator s partcularly useful for panel estmaton. In the case where the regressors may be correlated wth the error term, there exst fxed effects that may be correlated wth the explanatory varables and the panel dataset has a short tme dmenson and a larger ndvdual (frm) dmenson. Further, n an evaluaton of several dfferent technques for estmatng dynamc panel models, Judson and Owen (1996) conclude nter ala that the restrcted generalzed method of moments (GMM) estmator such as that dscussed n Arellano and Bond (1991) ncreases computatonal effcency wthout sgnfcantly detractng from ts effectveness. Alternately, Kmenta (1967) presents a lnearzaton of the two-nput CES functon, by employng a Taylor approxmaton, the result of whch s a restrcted form of the general 3

6 translog functon. Ths study follows the work of Kmenta (1967) by assumng that the general functonal form of producton s emboded by the CES technology and apples the Taylor approxmaton of the CES n dervng estmates of TFP wthout the mposton of constant returns to scale as n the case of alternate functonal forms. Wth regard to the determnants of productvty, most of the recent studes nvestgate the mpact of polcy change on productvty changes and consequently economc growth. In partcular, there s a large body of lterature on productvty growth and ts determnants for the manufacturng sector. These studes focus on the relatonshp between economc reforms and productvty n the manufacturng sector. Other studes n the lterature explore the factors underlyng productvty such as foregn drect nvestment (FDI) and research and development (R&D) amongst others. Intutvely, greater trade lberalzaton, FDI and/or R&D ncreases productvty n domestc ndustres and result n more effcent allocaton of resources and overall greater output (Topalova, 2004). Unel (2003) usng a smple producton functon wth the assumptons of perfect competton and constant returns to scale found that labour and total factor productvty growth n Indan manufacture ndustry were hgher after trade reforms n the late 1970s. Further, the author found that whle aggregate productvty ncreased, sectoral productvty grew only n sectors whch were opened to trade and FDI whle those that were not opened dd not perform as well. However, a smlar analyss on Indan manufacture ndustry by Kaur and Kran (2008) analyzed ndustry productvty at the aggregate and dsaggregated level for the perod 1980 to 2003 as well as for the sub-perods, pre and post the 1991 lberalzaton reforms. The paper estmated TFP growth by employng the trans-log producton functon and found that TFP ncreases were hgher durng the pre-lberalzaton perod. Further, the analyss of the partal productvty and total factor productvty for Indan manufacturng sector were found to have decelerated n the new polcy regme wth approxmately 72 per cent of the ndustral groups showng hgher total factor productvty n the pre lberalzaton phase. However Glodar and Kumar (2003), n a smlar study, found that 4

7 the reducton n effectve protecton to ndustres appears to have had a favourable effect on productvty growth n Indan ndustres. Ther study, whch used a trans log ndex of TFP at the ndustry level wth an explct varable representng trade lberalzaton, hghlghted that the deceleraton n TFP growth should not be attrbuted to mport lberalzaton but rather to gestaton lags n nvestment projects and a slowdown n agrculture growth n the 1990s. Other studes n the lterature focus on possble stmul to productvty from FDI and/or R&D and the lkely postve spllover effects. Ayanwale and Bamre (2004) nvestgated the mpact of FDI on productvty at the frm-level n agro/agro-alled sector of the Ngeran economy. FDI was seen as a productvty enhancng package. The results obtaned from the use of a neoclasscal two factor producton functon demonstrated that foregn frms productvty were hgher than that of ther domestc counterpart. Accordngly, the authors alluded to the exstence of postve and sgnfcant spllover effects of FDI at the frm level, but noted, however, that the spllover may not extend to the sectoral level. Kyota (2006) usng a frm-level longtudnal data from 1995 to 2002 examned the mpact of R&D on the productvty of Japanese manufacture frms. The paper used a productvty catch up model proposed by Bernard and Jones (1996) and found postve effects of both local and nternatonal R&D spllovers. Smlar results were obtaned by Yeon Lee and Km (2006) who analyzed the TFP for 14 OECD countres usng R&D among the nput proxes, to derve estmates of Malmqust TFP ndexes 7. The results ndcate that R&D s an mportant determnant of productvty growth n manufacturng ndustres as well as ts assocated spllover effects. Wagner (2005) use mcro data to emprcally nvestgate the relatonshp between export actvtes and productvty. More specfcally, the author examned the extent and causes 7 The Malmqust ndex s an ndex that s used to compare the producton technology of two economes. A functon of maxmum possble producton, wth respect to a set of nputs the Malmqust Index of economy A wth respect to economy B s calculated by substtutng the labour and captal nputs of economy A nto the producton functon of B, and vce versa. 5

8 of productvty dfferentals between frms that are export-orented and ther counterparts, whch produce for the domestc market. By lookng at dfferences n average labour productvty or average total factor productvty between exporters and non-exporters, the author found evdence to support the hypothess that exporters are more productve than non-exporters, and that whle the more productve frms self-select nto export markets, exportng does not necessarly mprove productvty. Wth respect to productvty studes on Jamaca, Downes (2003) revewed the factors that affect productvty and compettveness n Jamaca at the mcro and macro level durng the 1980s and 1990s. The author utlzed feld ntervews wth key nformants, case studes of companes and survey of enterprses. The study found that the process of enhancng compettveness can be acheved by boostng the productvty of the factors of producton (captal, labour, manageral ablty), mantanng a stable macroeconomc envronment and lowerng the overall costs of producton as well as the cost of dong busness n Jamaca. Drawng on the Kmenta (1967) and Hoff (2004), ths study frst estmates the CES functonal form of the producton functon by OLS then test the sgnfcance of frm level ndcators by usng the GMM procedure. Ths, to the best of the author s knowledge, represents the frst attempt to statstcally test the possble determnants of productvty n Jamaca at the frm s level. 1. Theoretcal Framework The theoretcal framework begns wth the assumpton of a two-nput CES producton functon for each frm as follows: Y 1 p p p K 1 L. (1) 6

9 Where Y, K, L represent output, captal and labour for the th frm, respectvely, and, and p are symbolc of factor productvty, the share/dstrbuton parameter and the elastcty of substtuton. For the above functon, constant (Hcks, 1932) elastcty of substtuton between nput factors s gven by 1 1 p Expressng equaton 1 n logarthms gves:,when 1 (.e. p 0 ) 8. lny p p K (1 ) L u 1 ln ln (2) p Where u s the stochastc error term assumed to be ndependently and normally dstrbuted wth zero mean and constant varance. Followng Kmenta (1967) and Hoff (2004), the CES producton functon s approxmated by a lnear Taylor seres expanson, when p s n the neghbourhood of zero (.e. when the elastcty of substtuton s n the neghbourhood of unty). Usng ths form, the parameters of equaton 2 can therefore be estmated by least squares: lny ln( ) v ln( K) v (1 ) ln( L) pv (1 )ln( L) 2 2 pv (1 )ln( K) 2 2 pv (1 )ln( K) ln( L) (3) By algebrac manpulaton, equaton 3 s equvalent to the followng translog functon: 2 2 ln( Y ) 0 1 ln( L) 2 ln( K) 11 ln( L) 22 ln( K) 12 ln( k) ln( L). (4) for whch the parameters fulfll the condtons: v ; p v (5) 1 2 (1 ) As stated above the two nput CES functon reduces to the Cobb-Douglas functon n the lmt 1, (.e. p=0). 7

10 Intutvely, Hoff (2004) argue that f a two nput technology s beleved to have constant elastcty of substtuton n the neghbourhood of unty, then the nput and output values observed for the technology maybe ftted to the above translog form. The resduals from the estmaton of equaton 4 are then used as a measure of TFP. The second component of the framework looks at estmatng the relatonshp between total factor productvty and a set of potental explanatory varables for each frm n the panel. Recognzng the lmtatons of workng wth panel data, ths study augments the theoretcal framework by removng frm fxed effects, whch maybe correlated wth the explanatory varables and contaned n the error termu. Accordngly, the regressors n equaton 4 were transformed by frst dfferencng. 9 Specfcally, we consder the explanatory power of skll level, captal ntensty, the level of technology and the economes of scale of the frms. The functonal form that we estmate s: (6), t 0, t 1 1, t 2, t 3, t 4, t where,,, and represent productvty, level of skll n the labour force, captal ntensty, level of technology and the economes of scale of the frms, respectvely. It s expected that the level of skll, whch s measured as the sklled porton of the labour force, s postvely related to productvty. Smlarly, postve relatonshps are expected between productvty and the level of technology, captal ntensty and economes of scale, albet subdued due to the hgh percentage of unsklled labour n the Jamacan labour force. Of note, a more granular analyss by further dchotomzng frms nto goods or servce producers or exporters and non-exporters was not done due to data lmtatons gven the small sample. Gven the panel structure of the data as well as the characterstc of a short tme dmenson and a larger frm dmenson, the Arellano- Bond (1991) GMM estmator s employed where lagged frst-dfferences are used as nstruments n addton to the usual lagged levels. Ths removes the serally correlated shocks to the frm-specfc producton technology (Judson and Owen, 1996). 9 Transformaton by frst dfferencng removes the fxed frm specfc effect because t does not vary over tme. 8

11 2. The Emprcal Analyss The analyss uses data from the fnancal reports of 23 publcly lsted frms on the Jamacan Stock Exchange from 1998 to The data draws on 12 years of nformaton from the audted fnancal statements of each company/group to create an unbalanced panel dataset. The use of an unbalanced panel s deal for storng as much nformaton as possble from the dataset, whch should lead to more effcent estmates (Buzten et. al, 2001). The paper recognzes the fact that the number of frms on the Jamaca Stock Exchange s lmted; however, these frms are large players n ther respectve ndustres and tend to behave n a typcal manner. In ths regard, t s beleved that the dataset s a reasonable representaton of the corporate sector n Jamaca and as such the results are useful. From the frm-level data, the varables examned n ths nvestgaton are labour, captal, output, captal ntensty and economes of scale. Wth regard to the elements of the producton functon, captal s measured by the value of plant, equpment and machnery, labour s captured by the number of employees 10 and output s measured by total revenue as dentfed n the company s annual fnancal reports. The qualty of the labour force and captal ntensty s measured by the percentage of the labour force classfed as havng no tranng and the rato of the real value of fxed assets to labour, respectvely 11. Economes of scale s captured by the value share of captal adjusted for nflaton and was ncluded to test the hypothess that larger frms are nherently more productve. Wth regard to the varables examned usng aggregate data, nvestment n technology s measured by nternet users per 100 people seeks to capture the mpact of technology savvy ndvduals on productvty levels 12. Further, the level of skll s measured as the recprocal of the percentage of the Jamacan labour force that s classfed as havng no skll The number of workers was grown by rato of average change n sectoral wages to average change n frm wages where there was mssng data. 11 The qualty of the labour force measured as the percentage of the labour force was estmated wth varyng levels of tranng n prelmnary estmatons, No tranng was used as other measures were correlated wth other determnants n the regresson, such as technology. 12 Technology, measured as moble and fxed-lne telephone subscrbers per 100 people, was also estmated but was found not to be statstcally sgnfcant. 13 An average of 75.9 persons n the labour force s wthout skll and receves low wages. 9

12 A summary of the data statstcs s provded n Table 1. It shows that durng the sample perod, the dstrbuton of the data s postvely skewed and leptokurtc for all varables, ndcatve of the domnance of larger companes/group, and substantal number of observaton close to the mean, respectvely. Table 1: Summary Statstcs Statstcs Labour Captal Output Captal Intensty Economes of Scale mean se(mean) p sd skewness kurtoss N To assst n shapng the apror expectatons, a scatter plot of labour aganst output, captal wth output and captal ntensty wth output were created (see Fgures 1-3 n the Appendx). The plots ndcated that hgher levels of captal are assocated wth hgher output whle hgher labour nput does not necessarly translate nto greater output. From the plot of the relatonshp between the captal ntensty and output t was deduced that hgher captal ntensty s not necessarly assocated wth hgher output. Ths could possbly ndcate the lack of effcent use of captal employed. The man suppostons are therefore that all the aforementoned varables postvely mpact productvty. Wth regard to qualty of labour force, lower sklled workers are expected to contrbute very lttle to changes n productvty. 3. Results Ths secton outlnes the results from the estmaton of equaton 4 and 6. The assessment starts wth an examnaton of the propertes of the data. The exstence of a panel unt root s formally tested usng the Levn, Ln & Chu panel unt root test, whch assumes a 10

13 common unt root process. The null hypothess of a unt root s rejected at the 5 per cent level for the log of captal, labour and output (See Table 4 n the Appendx). However, the Im, Pesaran and Shn Unt root test, whch assumes ndvdual unt root process, fals to reject the hypothess of a unt root for the labour, output and captal varables. The major advantage of the Im, Pesaran and Shn Unt root test s that fewer tme observatons are requred for the test to have power as t combnes nformaton from the tme seres dmenson wth that from the cross secton dmenson. As such the test allows for unbalanced panels. Notwthstandng the results of the Im, Pesaran and Shn Unt root test, the Pedron Resdual Contegraton Test rejected the null hypothess of no co-ntergraton and equaton 4 was subsequently estmated by ordnary least squares method wth the mposton of the restrctons stated n equaton 5. The results for the OLS estmaton of ths producton functon are presented n Table 2. The percentage contrbuton of each factor to total frm output can be deduced from the magntude of the coeffcent of labour. The labour share, 1, s ndcatve of the low contrbuton of labour towards output, possble capturng the concentraton of unsklled workers n frms. The estmated labour share of 0.24 dffers from aggregate computaton of 0.59 n Jamaca. However, the estmated labour share corresponds wth mean estmates of labour share of developng countres of 0.34 by Decreuse and Maarek (2008) but s below average labour share of 0.39 for developed countres. The results confrm that the captal and labour varables are statstcally sgnfcant at the 5 per cent level (see Table 2). 11

14 Table 2. OLS Estmaton of CES Producton Functon Dependent Varable s Coef. t P>t Output 3.79* (0.34) * (0.03) * (0.00) Number of obs 208 R-squared Adj R-squared Root MSE Res. dev * Sgnfcant at 5% level of sgnfcance. Bootstrapped standard errors n the parenthess. Followng the work of Hoff (2004), the results were examned to determne whether the translog functon can be reduced to a Cobb-Douglas functon. The Wald test was used to evaluate the null hypothess that the trans log functon can be reduced to a Cobb-Douglas functon by testng that none of the cross parameters, j, are sgnfcantly dfferent from zero. The test rejects the null hypothess that the cross parameters are not sgnfcantly dfferent from zero at the 5% level, and therefore the hypothess that the translog functon can be reduced to a Cobb-Douglas functon rejected (see Table 6). The results are evaluated to determne whether the translog functon s an approxmaton of the CES functon by examnng the followng condton: j j n k 1 k j 2 2 n k 1 k 2 2 j n j k 1 k jj Intutvely, f the hypothess of the CES structure s beleved to be true, the substtuton factor can be evaluated as the average of the factors and ndcatvely the elastcty of 12

15 substtuton. The average of substtuton factor was found to be mplyng elastcty of substtuton of Ths supports the pont that the estmated translog functon s an approxmaton of the CES functon as the elastcty of substaton s not equal to unty and the substtuton parameter, p, not equal to zero. Havng derved estmates of the change n TFP, the regresson of possble determnants of changes n TFP s examned. Equaton 6 s estmated wth frst dfference cross-secton effects to correct for possble autocorrelaton of the error term. Addtonally, the Sargan test of over-dentfyng restrctons faled to reject the null hypothess that the overdentfyng restrctons are vald. Further, usng Wooldrdge s test for seral correlaton, n lnear panel-data models, reveals that the model s robust to seral correlaton as well as condtonal heteroskedastcty (see Table 5 n the Appendx) 14. Further, the Arrellano and Bond test ndcates a falure to reject the null hypothess of no seral correlaton. Wth the excepton of changes n captal ntensty, the results show that all varables are statstcally sgnfcant and are n lne wth a pror expectatons (See Table 3). Intutvely, the results demonstrate that hgher qualty of the labour force wll have a postve mpact on productvty albet at the 10 per cent level of sgnfcance. Smlarly, the level of technology proxed by the level of nternet access also mproves productvty. The results show that captal ntensty s negatvely related to greater effcency n the producton process. Intutvely, ths result mples that ncreasng the real value of fxed captal relatve to the number of workers reduces output as was nferred from the scatter plot. One plausble explanaton could be that the type of captal employed s not accompaned by the requste labour sklls to ncrease output. Fnally, economes of scale, although beng statstcally sgnfcant was found to have lttle or no mpact on productvty. The low coeffcent on the economes of scale varable mples that larger frms are not nherently more productve that ther smaller counterparts. The results of the estmaton s summarzed n Table Wooldrdge posts that there s no seral correlaton f the resduals from the regresson are regressed wth the frst-dfferenced varables on ther lags and the coeffcent on the lagged resduals s equal to 0.5. In ths study the coeffcent was whch was the Wald test ndcated was not sgnfcantly dfferent from

16 Table 3: Dynamc Panel Data Estmaton of Total Factor Productvty Dependent Varable s TFP Coef. z P> z Lag of TFP * (0.06) Qualty of Labour ** (0.02) Technology * (0.00) Captal Intensty * (0.02) Economes of Scale 0.000* (0.00) Arellano Bond test 1 st Oder z value nd Order z value 1.68 Sargan test ch2(64) Prob > ch No. of obs 141 No. of groups 21 Wald ch2(5) Prob > ch * Sgnfcant at 5% level of sgnfcance and **sgnfcant at 10% level of sgnfcance. (fgures n parenthess are the Wndmejer s bas-corrected standard errors) 4. Summary and Concludng Remarks Ths paper estmates a producton functon and analyses the factors whch nfluence productvty of frms n Jamaca. Usng data from publcly lsted frms between 1998 and 2009, the study estmates a CES producton technology functon, whch s approxmated as a Taylor seres expanson for the average frm. Wth estmates of total factor productvty, the study further examnes possble determnants of changes n the estmated frm s productvty. The paper fnds that for the typcal frm captal plays a major role n the producton of output. The effcency by whch ths s done s negatvely nfluenced by the hgh levels of unsklled labour. However, productvty has been 14

17 postvely nfluenced prmarly by the utlzaton of technology and the economes of scale. Notwthstandng, the paper hghlghts that low labour productvty as a major hurdle to achevng ncreased nternatonal compettveness and subsequently economc growth. Polcy acton should be amed at enhancng human resource development as well as fosterng mproved management employee relatons and approprate ncentve systems to foster a more capable and motvated workforce. Incentvzng the tranng of workers as well as research and development are possble areas to explore. The results above concur wth studes by Shrley (1991), Krton (1992) and Crck (2001) who found human resource underdevelopment as well as the absence of techncal and problem solvng sklls as the man reasons for low productvty n Jamaca. Addtonally, Shrley (1991) hghlghted that the dchotomy between lne and staff workers foster dstrust n workplace relatons, whch produces worker demotvaton and low productvty. Ths low productvty scenaro partally explans the hgh nvestment and low growth paradox whch exsts n Jamaca. As t relates to manufacturng, Krton (1992) suggested that labour productvty maybe enhanced by engagng n, amongst other thngs, plant reorganzaton and lay-out, proper producton schedulng of mantenance, mproved mantenance of equpment, tranng n qualty management and an ncentve programme to reduce absenteesm. Moreover, Crck (2001) n examnng sx companes n Jamaca found that even wth superor wages and salares, productvty can be low. Companes should therefore focus on human resource management factors such as tranng, the qualty of the work force and performance based ncentves so as to enhance labour productvty and ultmately mprove nternatonal compettveness for frms n general. Gven the lmtatons of data, there exsts possble room for mprovement n subsequent studes. Ths could possbly nclude ncreased data to analyze the dfferences n productvty of goods and servce producers or exporters and non-exporters as well as the determnants whch characterze TFP growth n the respectve market segments. Further, more granular categorzaton of wage data could further nform the mpact of labourers, mddle management and senor management. Smlarly, results could be further 15

18 strengthened by more detaled data on ndvdual frm expendture on technology and research and development. 16

19 5. References Arellano, M. and Bond, S., Some tests of specfcaton for panel data: Monte Carlo evdence and an applcaton to employment equatons, The Revew of Economc Studes, 58. pp Ayanwale,A.B. and Bamre, S Drect Foregn Investment and frm-level Productvty n the Ngeran Agro/agro-alled Sector. Journal f Socal Scence, 9 (1) pp Balakrshnan, P., K. Pushpangadan and M.S. Babu Trade Lberalzaton and Productvty Growth n Manufacturng: Evdence from Frm-Level Panel Data. Economc and Poltcal Weekly, October. Bartelsman, E. and Doms,M., Understandng Productvty: Lesson from longtudnal Mcrodata, Journal of Economc Lterature, 38 vol. 3 pages Baum,C., Schaffer, M. and Stllman, S Instrumental varables and GMM: Estmaton and testng. The Stata Journal, Vol. 3, No. 1, pp Bernard, Andrew B. and Jones, Charles I., Comparng Apples to Oranges: Productvty Convergence and Measurement Across Industres and Countres. Amercan Economc Revew, Vol. 86, No. 5, December Avalable at SSRN: Blundell, R., Bond, S.,1998. Intal condtons and moment restrctons n dynamc panel data models, Journal of Econometrcs 87, Butzen, P.l, Fuss, C., Vermeulen, P., Sevestre, P. and Worms, A. (2001).The Interest Rate and Credt Channels n Belgum: An Investgaton wth Mcro-level Frm Data. ECB Workng Paper No Crck, A. P Celebratng Ourselves: Best Practces n Jamacan Organsatons (paper presented at the JATAD Conference/Dept of Management Studes, UWI, Mona). Bruno Decreuse & Paul Maarek, "FDI and the labor share n developng countres: A theory andsome evdence," Workng Papers halshs , HAL. Drukker,D Testng for seral correlaton n lnear panel data models, The Stata Journal, vol. 3, No. 2, pp Downes A. S., Productvty and Compettveness n the Jamacan Economy. Economc and Sector Study Seres, RES03002, InterAmercan Development Bank, Washngton, DC, July. 17

20 Fernandes, Ana M., "Frm Productvty n Bangladesh Manufacturng Industres," World Development, Elsever, vol. 36(10), pages , October. Glodar, B., and Kumar, A., Import Lberalzaton and Productvty Growth n Indan Manufacturng Industres n the 1990s, The Developng Economes, Vol. XLI, No. 4, pp Gomme, P. and Rupert, P., Measurng Labor s Share of Income. Federal Reserve Bank of Cleveland. Polcy Dscusson Papers. Ghosh, S.R. and Kraay, A. (2000). Measurng Growth n Total Factor Productvty, PREMnotes Economc Polcy, No.42, The World Bank, 2000 Gupta, A.,2009. Lookng Beyond the Methods: Productvty Estmates and Growth Trends n Indan Manufacturng. Avalable at SSRN: Hcks, J.R., 1932, The theory of wages, (McMllan: London). Hodges, D. 1969, A Note On Estmaton Of Cobb-Douglas And CES Producton Functon Models, Econometrca,vol. 37.no.4. Hoff, A (2004). The Lnear Approxmaton of the CES Functon wth n Input Varables. Marne Resource Economcs, 19, Judson, R., and A. Owen, "Estmatng Dynamc Panel Data Models: A Practcal Gude for Macroeconomsts," Federal Reserve Board. Kathura, Vnsh, Seethamma Natarajan, Rajesh Raj and Sen, Kunal, Organzed versus Unorganzed Manufacturng Performance n Inda n the Post-Reform Perod. Unpublshed. Kaur, M. and Kran, R Indan Manufacturng Sector: Growth of Productvty under The New Polcy Regme, Internatonal Journal of Busness Research Papers, Vol. 4 No.2 March 2008, pp Kyota, K "Reconsderng the Effects of Intranatonal and nternatonal R&D Spllovers on Productvty Growth: Frm-level Evdence from Japan," Dscusson papers 06001, Research Insttute of Economy, Trade and Industry (RIETI). Krton, G.P Organzng for Greater Productvty, Carbbean Labour Journal, vol 2, no 1, Sprng, pp Kmenta, J "On Estmaton of the CES Producton Functon", Internatonal Economc Revew, vol. 8, p

21 Levnsohn, J. & Petrn, A., "Estmatng Producton Functons Usng Inputs to Control for Unobservables", Revew of Economc Studes, Blackwell Publshng, vol. 70(2), pages , 04. Mshra, SK A Note on Numercal Estmaton of R Sato s Two-Level CES Producton Functon, MPRA Paper No Nshmura, K., Nakajma,T and Kyota, K., Innovaton versus Dffuson: Determnants of productvty Growth among Japanese Frms. CIRJE Dscusson papers. Sergo DeSouza, "Levnsohn and Petrn's (2003) Methodology Works under Monopolstc Competton," Economcs Bulletn, AccessEcon, vol. 12(6), pages Shrley, G. (1991): Technology, Productvty and Global Compettveness, mmeo, Dept of Management Studes, UWI, Mona, September. Shrley, G Jamaca and Free Trade n the Western Hemsphere: A Comparatve Analyss of Selected Manufacturng and Servce Industres n A Weston and V Vswanathan (eds): Jamaca After NAFTA: Trade Optons and Sectoral Strateges (Kngston, Jamaca, Ian Randle Publshers), pp Topalova, P., "Trade Lberalzaton and Frm Productvty: The Case of Inda," IMF Workng Papers 04/28, Internatonal Monetary Fund. Unel, B., Productvty Trends n Inda s Manufacturng Sectors n the Last Two Decades. IMF Workng Paper, No WP/03/02. Wagner, J., "Exports and Productvty: A Survey of the Evdence from Frm Level Data," Dscusson Paper Seres 26308, Hamburg Insttute of Internatonal Economcs. Yasar, Racborsk and Po, "Producton functon estmaton n Stata usng the Olley and Pakes method," Stata Journal, StataCorp LP, vol. 8(2), pages , June. Yeon Lee, J. and Km, J "Total Factor Productvty and R&D Captal n Manufacturng Industres," Economcs Study Area Workng Papers 89, East-West Center, Economcs Study Area. 19

22 Appendx Fgure 1. Scatter Plot of Captal and Output ly lk Fgure 2. Scatter Plot of Labour and Output ly ll Fgure 3. Scatter Plot of Captal Intensty and Output ly c 20

23 Table 4: Panel unt root test: Summary Sample: Exogenous varables: Indvdual effects Automatc selecton of maxmum lags Automatc selecton of lags based on SIC: 0 to 2 Newey-West bandwdth selecton usng Bartlett kernel Seres: LK Seres: LL Seres: LY Seres: LTRAIN Seres: DNET Seres: LSIZE Method Statstc Prob.** Statstc Prob.** Statstc Prob.** Statstc Prob.** Statstc Prob.** Statstc Prob.** Null: Unt root (assumes common unt root process) Levn, Ln & Chu t* Null: Unt root (assumes ndvdual unt root process) Im, Pesaran and Shn W-stat ADF - Fsher Ch-square PP - Fsher Ch-square

24 Table 5. OLS Estmaton of Wooldrdge Test for Autocorrelaton n Panel Data Dependent Varable Is The Resduals Of Producton Functon Coeffcent Std. Error t Statstc Prob * R squared Adjusted R squared Durbn Watson stat No. of Observatons 105 * Sgnfcant at 5% level of sgnfcance. Table 6. Wald Test of the Cross Parameters Null Hypothess 11 = 0 Test Statstc Value df Probablty F statstc (1, 205) Ch square Table 7. Arellano-Bond test for zero autocorrelaton n frst-dfferenced errors Null hypothess: no autocorrelaton Order z Table 7. Sargan Test of Over-dentfyng Restrctons Null hypothess: Over-dentfyng restrctons are vald ch2(64) Prob > ch