Pertanika J. Soc. Sci. & Hum. 13(1): 95-102 (2005) ISSN: 0128-7702© Universiti Putra Malaysia Press

Relationships Among Output, Wages, Productivity and Employmentin the Malaysian Electronic and Electrical Sub-sector

*ZULKORNAIN YUSOP, **LAW SIONG HOOK & **NORASHIDAH MOHD NORDepartment of Economics

Faculty of Economics and ManagementUniversiti Putra Malaysia

43400 UPM Serdang, Selangory Malaysia

Keywords: Output, wages, productivity, electronic and electrical industries, employment

ABSTRAKIndustri elektrik dan elektronik Malaysia telah menjadi komponen paling pentrhg dalam sektorpembuatan negara. Seperti yang diketahui umum, sektor tersebut memainkan peranan pentingdalam eksport negara, guna tenaga sepertimana pertumbuhan ekonomi. Memikirkan kepentingansubsektor tersebut, adalah mustahak supaya kita memastikan pertumbuhan subsektor itu berterusandan persaingan dikekalkan atau dipertingkatkan. Kajian ini secara khususnya bertujuan untukmenyelidik pertalian antara output, produktiviti, gaji dan buruh. Seperti yang diketahui umum,gaji boleh meningkat selagi ia setara dengan peningkatan tinggi dalam produktiviti. Walaubagaimanapun, perubahan gaji sebenarnya boleh mempengaruhi guna tenaga. Ujian punca unitmenunjukkan bahawa semua pemboleh ubah gaji tersebut adalah I (1). Prosedur Johansendikendalikan untuk melihat hubungan jangka panjang dan jangka pendek dengan pembolehubah tersebut. Keputusan ujian kointegrasi Johansen menampakkan bahawa hubungan jangkapanjang keseimbangan wujud di kalangan pemboleh ubah tersebut. Daripada analisis dinamikjangka pendek, kami dapati bahawa kecuali gaji sebenar, produktiviti buruh dan guna tenagasecara statistiknya signifikan dalam mempengaruhi output.

ABSTRACTThe Malaysian electronic and electrical sub-sector has been the most important component of thenation's manufacturing sector. As is widely known, the sector has played a vital role in the nation'sexport, employment as well as overall economic growth. Considering the importance of the sub-sector, it is imperative that we ensure sustainable growth of the sub-sector and that competitivenessis maintained or even improved. This study is particularly aimed at investigating the linkagesbetween output, productivity, wage and labour. As is widely acknowledged, wages may increase aslong as it is commensurate with a higher increase in productivity. However, changes in wages canactually affect employment. Unit root tests indicate that all of the above variables are 1(1).Johansen's procedure was conducted to see the long run and short run relationships between thevariables. The Johansen cointegration test results revealed that a long-run equilibrium relationshipexists among the variables. From the short run dynamic analysis, we found that except for realwages, labour productivity and employment are statistically significant in influencing output.

INTRODUCTION in the sector (Table 1). The manufacturing

Since the early 1980s, the Malaysian economy s e c t o r i s currently the major contributor to thehas relied heavily on the manufacturing sector. nation's GDP (about 30% of the total) and itAbout 27% of the working population is engaged accounts for about more than 80% of the nation's

total merchandise export.

* Associate Professor** Lecturers at the Department of Economics, Faculty of Economics and Management, Universiti Putra Malaysia.

Zulkornain Yusop, I aw Siong Hook & Norashidah Mohd Nor

TABLE 1Malaysian manufacturing sector: annual growth of output, exports, value added,

percentage share of GDP and employment

Year Output Growth Exports GrowthPercentage

Share of GDPPercentage Shareof Employment

1988198919901991199219931994199519961997199819992000200120022003

17.020.315.314.07.014.611.411.418.210.1-13.413.517.012.04.06.5

31.9736.2128.0730.9116,5325,5234.1122.417.6612.8732.8014.3420.6124.929.326.6

21.123.324.625.625.126.226.727.129.129.927.930.33.430.230.130.6

15.2915.9419.9421.3323.1023.5524.8925.9226.6827.5026.5027.1027.626.721.721.7

Source: Malaysia Economic Report

Within the manufacturing sector, theelectronic and electrical (E&E) sub-sector hasplayed a leading role in the nation's growth,exports and employment. In 2003, the sub-sectorcontributed 61% of the total manufacturingoutput, about 35% of the total manufacturingvalue added, and accounted for 34% of totalmanufacturing employment. The E&E sub-sectoralso contributed 68% (worth of RM223 billion)of the total manufactured exports in year 2003.Major export products include electricalmachinery, apparatus, appliances, electrical parts,office machines and automatic data processingequipment. In terms of value added, much of ithas been contributed by semiconductors andother electronic components and communicationequipment and apparatus, radio and TV sets,sound reproducing, and reproducing andrecording equipment.

Obviously, the country's exports are heavilydependent on manufactured products whilemanufactured exports are narrowly based onE&E products. Thus, our exports would be verysensitive and vulnerable to changes in worldsupply and demand for electrical products.Should there be any severe drop in the demandfor electronic and electrical products in theworld market, the country's economy would beadversely affected in terms of growth andemployment. The world electronic industry did

experience severe doldrums in the middle ofthe 1980s in the aftermath of a massive supplyexpansion in anticipation of increased demandwhich did not materialise. The effects of thedoldrums were severely felt by many producers,exporters and workers in the electronic sector.As such, it is important to understand therelationship or interdependence between certainimportant variables in the sub-sector i.e. output,productivity, employment and wage.

A very first question here is whether wage,labour productivity and employment significandyaffect the real output of the Malaysian electronicsub-sector? It is also of interest to examine themulti-directional linkages among the real output,wages, productivity growth and employment inthe electronic sub-sector. The establishment ofcausal dynamic linkages among these variableshas important implication for Malaysia. If thewage rate leads to a higher productivity, thenincreasing it can increase the manufacturingoutput and competitiveness in the internationalmarket. However, if the increase in wage rateleads to a significant reduction in employment,then the objective of reducing unemploymentmay contradict that of reducing poverty andimproving the living standards of the workers.

Numerous studies have been conducted toexamine the relationship among wages,productivity, employment and output. Huh and

96 PertanikaJ. Soc, Sci. & Hum. Vol. 13 No. 1 2005

Relationships Among Output, Wages, Productivity and Employment

Trehan (1995) found that Granger causality runsfrom prices and productivity to wages but notthe other way round in the United States byusing a simple dynamic labour demand model.Parker (1995) had shown that increase in theUnited States wage rate increased the labourproductivity, but it reduced the level ofemployment by about 2%. Hostland (1996)supported the neoclassical view of a long runequilibrium relationship between the real wagerate and labour productivity in the case ofCanada. Deviations between the real producerwage rate and average labour productivity arelarge and persistent but are not found to bepermanent. Growth in the real producer wagegrowth in excess of average labour productivityhas generally had relatively minor implicationson the employment rate.

The effect of productivity growth on wageshas been carried out by Carneiro (1998) inBrazil using time series data for 22 manufacturingsub-sectors for 1985-1993. He found changes insectoral productivity to be a relevant explanationfor the changes in sectoral nominal wages. Pausand Robinson (1997) demonstrated thateconomic growth, investment share growth andproductivity growth are the determinants of realwage growth. They also concluded thatgovernments who want to promote growth andliving standards of their workers have to focusexplicitly and primarily on increasing investmentand productivity growth. On the other hand,Pehkonen (1995) found that there areconsiderable differences across the differentsectors of the economy and inter-countrydifferences of productivity growth.

This paper is divided into several parts. Thenext section explains the sources of data and theeconometric methodology employed followedby discussion on the estimated results. The lastsection provides the conclusion and policyimplication.

THE DATA AND METHODOLOGYThe Data

This study involves annual data spanning from1973 to 1997 of the E & E sub-sector real output(RO), employment (EMP), labour productivity(LP) and real wages (W). They were collectedfrom various issues of the Malaysia Year Book ofStatistics, published by Department of Statisticsand Annual Productivity Report by NationalProductivity Corporation (NPC).

Value-added is used as a proxy for output ofE&E sub-sector. The value added output is grossex-establishment values minus cost of inputs andis deflated by the producer price index (PP1) toachieve real terms at the base year 1990. Value-added reflects the true economic activity of theindustry and it also tended to yield results thatare more closely associated with business changes.The nominal wages is derived by dividing thelabour costs (salaries and wages paid, includingbonuses, cash allowances, etc) by the number ofpaid employees. The real wage is then obtainedby deflating the nominal wage with consumerprice index (CPI). The number of paidemployees comprises both full time and part-time workers, where two part-time workers aremade equivalent to one full-time worker. Labourproductivity (LP) is defined as value added peremployee in nominal terms.

Vector Autoregressive (VAR) Model

For the purpose of this study, a vectorautoregressive (VAR) model was set up toinvestigate the relationship among output, labourproductivity, employment and real wages in theE&E sub-sector.

RO,

EM.

RO,LP,EM,W

a.

« 3

A. PJD. A(1)

where RO denotes the real output; LP is labourproductivity; EM is employment and W is realwages.

The long-run relationships amongst thevariables are investigated by the Johansenjuselius(1990) multivariate cointegration test. The short-run relationships, on the other hand, areanalyzed by the Granger-causality analysis withthe vector error-correction model (VECM) (1988).

Multivariate Cointegration TestBefore conducting the multivariate cointegrationtest, it is necessary to establish whether therelevant variables are stationary and to determinethe order of integration of the variables. Thiscan be achieved by employing unit root tests,namely the Augmented Dickey-Fuller (ADF)(1979) and unit root tests in the levels and first

PertanikaJ. Soc. Sci. & Hum. Vol. 13 No. 1 2005 97

Zulkornain Yusop, Law Siong Hook & Norashidah Mohd Nor

differences of the variables. After having theorder of integration of each series or stationaryproperties of each individual series, maximumlikelihood multivariate cointegration test is thenutilized to determine the number of linearlyindependent cointegrating vectors in the system.In the case of non-stationary data, a cointegrationanalysis will then be conducted in a vectorautoregression (VAR) model:1

(2)

where

for I * 1, ..., ft- 1.

Xt is a vector of p variables (or p = 4 for thisstudy), /i are the intercepts, / are deterministictrends and £{ is a vector of Gaussian randomvariables. The coefficient matrix II, also referredto as the long-run impact matrix, containsinformation about the stationarity of the fourvariables and the long-run relationship amongstthem. The rank (r) of the matrix determines thenumber of cointegrating vectors in the system.In the absence of cointegration, II is a singularmatrix (its rank, r- 0). Hence, in a cointegratedcase, the rank of FI could be anywhere betweenzero and four. If = 1, there is a singlecointegrating vector, whereas for 1 < r< 4, thereare multiple cointegrating vectors. This is anindication that the variables in the system arecointegrated in the long run with r cointegratingvectors. In other words, these variables possess along-run equilibrium relationship, and aremoving together in the long run. The II matrixcan be factored as, IT • ctjS7, where the a matrixcontains the adjustment coefficients and thematrix contains the cointegrating vectors.(Johansen andjuselius 1990) approach uses twolikelihood ratio statistics, the trace and themaximum eigenvalue statistics, to test for thepossible number of cointegrating vectors in thesystem. Critical values for these statistics are

tabulated in Osterwald-Lenum (1992). Theoptimal lag structure of the system is determinedby using the Likelihood ratio test.

Vector Error-Cmrection Model (VECM)If cointegration is detected amongst the variables,then the short-run Granger-causality analysis onthese variables must be conducted in a vectorerror-correction model (VECM) to avoid problemof misspecification (see Granger 1988) .2

Otherwise, the analysis may be conducted as astandard vector autoregressive (VAR) model.3

The direction of Granger-causal effect runningfrom one variable to another can be detectedusing the vector error-correction model (VECM)derived from the long-run cointegrating vectors.The VECM model employed for the testing ofGranger-causality across various variables in thesystem can be represented by:

AROt

ALP,

AEMt

a, pu(L) j812(L) •../3n(L) p22(L) ---

[pAl(L) j942(L) -

AROt

ALP,

AEMt

00

(3)

where Xt is an ( 4 x 1 ) vector of the variables inthe system, a's represent a vector of constantterms, j3*s are estimable parameters, A is adifference operator, L is a lag operator, j3(L)and O(L) are finite polynomials in the lagoperator, z ,'s are error-correction terms, and6f's are disturbances.

The Granger causality test is applied bycalculating the F-statistic based on the null

A variable that is found to be stationary at level, or is 1(0), is treated as an exogenous variable in the system.If the variables in a system are cointegrated, then the short-run analysis of the system should incorporate the error-correction term (ECT) to model the adjustment for the deviation from its long-run equilibrium.When an ECT is added to the vector autoregressive model (VAR), the modified model is referred to as the vectorerror-correction model (VECM). VECM is thus a special case of VAR.

98 PertanikaJ. Soc. Sci. & Hum. Vol. 13 No. 1 2005

Relationships Among Output, Wages, Productivity and Employment

hypothesis that the set of coefficient for thelagged values of independent variables are notstatistically different from zero. If the nullhypothesis is not rejected, then it can beconcluded that the independent variable doesnot cause the dependent variable. For instance,if the F-statistic of the real wages (W as anindependent variable in the equation) issignificant at a 5% level (i.e. Ho: j3.(L) = 0, for irefers to W, is rejected at a 5% significancelevel), and the employment (EM) is thedependent variable of the equation, then we cansay that there is a short-run causal effect runningfrom W to employment. Besides the detection ofthe short-run causal effects, the VECM also allowsus to examine the effective adjustment towardsequilibrium in the long run through thesignificance or otherwise of the Mest of thelagged error-correction terms (ECT) of theequation.

ESTIMATED RESULTS ANDINTEPRETATIONS

Unit Root and Cointegration Tests

Table 2 presents the result of the ADF andPhillip-Perron (P-P) unit root tests of real output,labour productivity, employment and real wages.The results support the presence of a unit rootat the level of all variables and the absence ofany unit root after first differencing; in otherwords, all variables are 1(1). This reveals that allvariables are nonstationary in the levels, butstationary in the first differences.

TABLE 2Results of the unit root tests

The Johansen cointegration test is performedto a system of four 1(1) variables for the E&Esub-sector and the estimated results are reportedin Table 3. The results reveal that there is onecointegrating vector in the system.

This indicates that there a long-runequilibrium relationship exists in the fourvariables, namely real output, labour productivity,real wages and employment.

It is assumed that there is no deterministictrend in data, no intercept and trend in thecointegrating equation. Outputs of theJohansen's test suggest that one cointegratingvector exists based upon the A,max and tracestatistics at 1 % level (Panel I). Both of the testssuggest rejection of zero cointegrating vector infavour of one. The cointegrating equation wasestimated with a provision for three lags and noserious serial correlation or normality problemwas found with the inclusion of this number oflags (Panel III). The estimated cointegratingvector has theoretically plausible coefficients.The long run relationship may be written as:

LROt = 3.11*LLP + 1.85*LEMt - 1.717*LWAGE

The equation indicates that higher labourproductivity and employment yield positiveinfluence on the real output of the industry inthe long run with estimated elasticities of 3.11and 1.85 respectively. On the contrary, real wageincrease seems to cause a decrease in the E&Eindustry real output with estimated elasticity of

ROLPEMW

ROLPEMW

Augmented

Constant withTrend

-2-7138(1)-1.0618(0)-1.4506(0)-3.3737(0)

-5.9108(1)**-5.1127(0)**-3.0697(0)-6.1585(0)**

Dickey Fuller Test

Constant withoutTrend

Level-0.8549(1)0.8572(1)-0.8852(0)-1.2608(0)

First Difference-6.0524(1)**-4.3324(1)**-3.0725(0)*-6.3436(0)**

Constant withTrend

-2.5292(2)-0.6678(2)-2.0031(2)-3.4764(1)

-3.7785(2)*-6.0199(2)**-3.2019(2)-6.3505(2)**

P-P Test

Constant withoutTrend

-1.2715(2)1.6605(2)

-0.8645(2)-1.3725(1)

-4.0262(2)**-4.9233(2)**-3.1561(2)*-6.5678(2)**

Notes: RO = real output; LP = labor productivity; EM • employment; W • Real wages. The asterisk * and ** indicatesthe level of significance at a 5 % and 1 % level respectively. The number in each parenthesis indicates the optimal laglength used in the regression, which is determined by the Akaike information criterion (AIC), to ensure the whitenessof the residual.

PertanikaJ. Soc. Sci. & Hum. Vol. 13 No. 1 2005 99

Zulkornain Yusop, Law Siong Hook & Norashidah Mohd Nor

TABLE 3Results of Johansen and Juselius multivariate procedure, VAR with 3 lags

sample period: 1973-1997 (25 observations)

I. Eigenvalue0.942627 0.587569 0.281412 0.0019842

Hypothesis Maximum Eigenvalue TraceHo:rank=p -Tlog(l-Vnu) using T-nm 95% -T\Sum log(.) using T-nm

P = = 0 62.88**19.497.270.0437

40.01**12.44.6270.02781

27.121.0.14.1

3.8

89.68**26.8

7.3140.0437

57.07**17.054.6540.02781

47.229.715.43.8

II.

HI.

i)

ii)

i i i )

iii)

Estimated Cointegrating VectorLRO

1.0000

Test for Appropriate Lag LengthLRO

Serial CorrelationX2(l) 2.0147 [0.1558]F(l, 12) 1.2097 [0.2930]Normality: %z{2)5.5816 [0.0614]Vector AR 1-1 F(16, 18)Vector normality %2 (8)

LLP-3.1105

(3)LLP

0.25286 [0.6151]0.13953 [0.7153]

0.96614 [0.6169]= 2.0802= 11.826

LEM-1.8496

LEM

8.656 [0.0342]*0.1976 [0.6645]

5.2299 [0.0732][0.1870][0.1591]

LW1.7167

LW

0.0827 [0.7736]0.0453 [0.8350]

0.5446 [0.7617]

Note:** indicates significance at 1% level.* indicates significance at 5% level.Figures in square parentheses [] refer to marginal significance level.

TABLE 4Causality results based on vector error correction model for electrical and electronic sub-sector (lag 3)

Estimated F-Statistic (joint Wald test)D(LRO) D(LLP) D(LEM) D(W)

ECTCoefficient / (t-Statistic)

D(LRO)D(LLP)D(LEM)D(LW)

2.030*2.880**2.014*

2.689*** 0.9398 4.348*** -0.8302 / (-2.15)**2.593** 2.194** -0.894 / (-14.34)***

2.456** - 1.1109 -0.0306 / (-1.1867)0.882 1.6238 - -0.545 / (-3.101)**

Note: * , ** and *** denote statistically significant at 10%, 5% and 1% respectively.

-1.717.

The Short Run Dynamic RelationshipThe short run interaction among the fourvariables are estimated using the vector errorcorrection model. The results of the VECM arereported in Table 4. Following the Johansenmultivariate cointegration test, one error-correction term is incorporated in the VECM.The error correction coefficients are significantin three of the equations i.e. that of real output,labour productivity and real wage suggesting

that real output, labour productivity and realwage are adjusted to divergence from long-runequilibrium steady state.

The results in Table 4 imply that RO, LPand W are endogenous while EM is weaklyexogenous. The estimated results also revealthat there is a unidirectional causal effect runningfrom labour productivity and employment toreal output. Both labour productivity and realwage significantly cause real output at 1% levelof significance. We could also see a unidirectionalcausal effect from real wage to labour productivity

100 PertanikaJ. Soc. Sci. & Hum. Vol. 13 No. 1 2005

Relationships Among Output, Wages, Productivity and Employment

and from real output to employment. Bi-directional causal effect seems to prevail betweenemployment and labour productivity.

The summary of the short run dynamicrelationship between the variables may beillustrated using the following diagram:

Results of diagnostic tests (Table 5) indicatethat the estimated short run dynamic modelsare quite robust as they generally pass all thetests of LM for auto-correlation, J-B for normality,and the Ramsey general specification test forspecification (except for the LLP which issignificant barely at 5%).

• LP

CONCLUSION AND IMPLICATIONS

Global development and the domesticdevelopments in the Malaysian manufacturingscenario which is basically highly dependent onE&E sub-sector provides justification for athorough understanding of the short run andlong relationship between real output, labour

TABLE 5Diagnostic checking for the VECM

Test LM (4) ARCH (4) J-B RESET (1)

LROLLPLEMLW

0.5280.3320.8320.750

0.0820.2390.1690.155

1.770.5710.8153.304

2.624.897*0.081.35

Note: *and ** denote statistically significant at 5% and 1%respectively.

productivity, employment and wage in the sub-sector. As is widely known, manufacturing as awhole sector is positioned as the main engine ofgrowth for the nation's development and hasbeen experiencing fundamental changes intechnology and liberalization of competition.Among all industries, the E & E sub-sector hasabsorbed the highest number of employment.Therefore, changes in wage rate in order to

influence productivity growth may amplifyfluctuations in employment rate. This studyattempted to investigate the linkages betweenthese variables over the 1973 - 1997 period inorder to provide a historical perspective on thisissue using the VAR model.

The estimated results indicate that there isa long run equilibrium relationship among realoutput, labour productivity, employment andreal wages of the E&E sub-sector. The short-rundynamic interaction shows that productivity andreal wage are quite significant factors affectingthe output. In the meanwhile, changes inproductivity are associated with changes in realwage, employment and real output. Real outputand labour productivity are also responsible forchanges in employment and finally, real outputdoes indeed affect real wage.

An important implication from this study isthat productivity can indeed play an importantrole in the real output of the E&E sub-sector inthe long run as well as short run. Our long runmodel reveals that availability of more labour isassociated with an increase in output while areal wage increase can adversely affect the realoutput. Thus, it is imperative that policy makersintensify efforts to improve the level ofproductivity in the sub-sector, and at the sametime ensure that the increase in level of realwage reflects the productivity of labour.

REFERENCES

CARNEIRO, F.G. 1998. Productivity effects in Brazilianwage determination. World Development 26: 139-153.

DICKEY, D.A. and W.A. FULLER. 1979. Distribution of

the estimators for autoregressive time-serieswith a unit root. Journal of the American StatisticalAssociation 74: 427-431.

ENGIZ, R. F. and C.W.J. GRANGER. 1987. Cointegrationand error correction: representation,estimation and testing. Econometrica 55: 251-276.

GRANGER, C.W.J. 1988. Some recent development ina concept of causality. Journal of Econometrics39: 199-211.

HOSTIAND, D. 1996. Real wages, labor productivityand employment in Canada: a historicalperspective. Research Paper for AppliedResearch Branch, Strategic Policy, HumanResources Development Canada.

PertanikaJ. Soc. Sci. & Hum. Vol. 13 No. 1 2005 101

Zulkornain Yusop, Law Siong Hook & Norashidah Mohd Nor

HUH, C.G and B. TREHAN. 1995. Modeling the time-series behaviour of the aggregate wage rate.Federal Reserve Bank of San Francisco EconomicReview 1: 3-13.

JOHANSEN, S. 1988. Statistical analysis of cointegrationvectors. Journal of Economic Dynamics and Control12: 231-254.

JOHANSEN, S. and K. JUSELIUS. 1990. Maximum

likelihood estimation and inference oncointegration with applications to the demandfor money. Oxford Bulletin of Economics andStatistics 52: 169-211.

MACKINNON, J. 1991. Critical value for cointegrationtests. In Long-run Economic Relationships: Readingin Cointegration, ed. R.F. Engle and C.WJ.Granger. New York: Oxford University Press.

OSTERWALD-LENUM, M. 1992. A note with quantitiesof the asymptotic distribution of the maximumlikelihood cointegration rank test statistics.Oxford Bulletin of Economics and Statistics 54:461-472.

PARKER, S. 1995. The real wage-employmentrelationship in the United States. Journal ofLabour Economics 8: 1-15.

PAUS, E.A. and M.D. ROBINSON. 1997. The

implications of increasing economic opennessfor real wages in developing countries, 1973 -1990. World Development 25: 537-547.

PEHKONEN, J. 1995. Wages and productivity growthin the nordic countries. European EconomicReview 39: 1181-1196.

(Received: 24 April 2002)

102 PertanikaJ. Soc. Sci. & Hum. Vol. 13 No. 1 2005