Preview only show first 10 pages with watermark. For full document please download

Supply Restricted Telecommunications Markets: The Effect Of Technical Efficiency On Waiting Times

The paper is concerned with supply constraints in the provision of telecommunications services. As a measure of supply constraint we use the average waiting time for telephone connections. Duration models are employed to analyze a panel data set for




  Journal of Productivity Analysis, 18, 161–169, 2002  C  2002 Kluwer Academic Publishers. Manufactured in The Netherlands. Supply Restricted Telecommunications Markets:The Effect of Technical Efficiency on Waiting Times ∗ ROBERT BARTELS † [email protected] Faculty of Economics and Business, University of Sydney, Australia TOWHIDUL ISLAM [email protected]  Business Program, UNBC, 3333 University Way, Prince George, B.C., Canada V2N 429  Abstract  The paper is concerned with supply constraints in the provision of telecommunicationsservices. As a measure of supply constraint we use the average waiting time for telephoneconnections. Duration models are employed to analyze a panel data set for 28 countries. Inadditiontoeconomicvariables,weconsidertheroleoftechnicalefficiencyincausingsupplyconstraints. Stochastic frontiers are used to determine the technical efficiency with whichcountries use labor and capital inputs to connect customers. When technical efficiency isincluded in duration models for waiting times until connection, we find that it is the majordeterminant. JEL classification:  L96, C23, C41, D24 Keywords:  telephone connections, duration models, panel data, stochastic frontiers, technical efficiency In less developed countries the provision of telecommunications services is frequentlysupply constrained, resulting in unmet demand and unreasonably long waiting times forcustomers to be connected. According to ITU (1999), about 50 million potential customers,in over 100 countries, were waiting for connections, with an average waiting time of twoyears.Duetotheflow-oneffectsoftelecommunicationsintoalmosteveryareaofeconomicactivity, the failure to provide phone connections is quickly felt throughout a country’sproductive system and affects its ability to compete in the global economy.Studiesoftelecommunicationsdemandoftenomitdatafrommarketsthataresupplycon-strained,ortheyincludethedataandignoretheproblem.Whensupplyconstraintsareincor-poratedintheanalysis,itistypicallyasanexplanatoryvariableinexplainingthepenetrationor diffusion of telephone services. 1 While Bowles (1995) and Maddock (1997) have em-phasizedtheimportanceofabetterunderstandingofthedeterminantsofsupplyconstraints, ∗ An earlier draft of the paper was presented at the  19th International Symposium on Forecasting,  June 27–30,1999, Washington DC. Towhidul Islam gratefully acknowledges the support of a U2000 Postdoctoral Fellowshipfrom The University of Sydney. † Address correspondence to: Head, School of Business, Faculty of Economics and Business, The Universityof Sydney, NSW 2006, Australia.  162  BARTELS AND ISLAM we are not aware of any studies that investigate this question empirically. The present paperaddresses this gap in the literature. More speci fi cally, we investigate the determinants of supplyconstraintsinthedeliveryoftelecommunicationsservices,usingtheaveragewaitingtime for telephone customers to be connected as a measure of supply constraint. 2 Technical ef  fi ciency is the main explanatory factor of interest in this study. More specif-ically, we consider the ef  fi ciency with which labor and capital inputs are used to connectcustomers.Ef  fi ciencymeasuresarerarelyconsideredinempiricalstudiesofthetelecommu-nications industry. This is somewhat surprising given the importance ascribed to ef  fi ciencyconsiderations in theoretical studies of industry structure. For example, Laffont and Tirole(1993) argue that regulatory and structural changes are effective means of achieving greaterallocative and productive ef  fi ciency. Referring speci fi cally to the telecommunications sec-tor, Crandall and Flamm (1989), Cowan (1990) and Duch (1991) suggest that, in a numberof countries, problems of industry structure are the main cause of low ef  fi ciency. And,according to Maddock (1997), low ef  fi ciency, regulatory barriers, and a lack of standards,all inhibit successful diffusion of telecommunications technology. Taking heed of thesearguments, in this paper we explicitly incorporate a measure of technical ef  fi ciency as anexplanatory variable in modeling supply restrictions.In Section 1 we describe the data used in the study. In the following section we derivemeasures of the technical ef  fi ciency of the telecommunications industries in different coun-tries. Since technical ef  fi ciency can not be measured directly, we estimate a stochasticfrontierproductionfunctiontoobtainestimatesoftechnicalef  fi ciency.InSection3wethenestimate a model that relates the supply constraints in a country to its technical ef  fi ciencyandanumberofotherexplanatoryvariables.Policyimplicationsandsuggestionsforfurtherresearch are presented in Section 4. 1. Data Description Thedataavailableforthisstudycovertheperiod1982 – 1992.Sincethewaveofprivatizationthat has hit the telecommunications industry started only towards the end of this period, wehave not analyzed the impact of privatization in this study. Waiting time for connection isused as an indicator of supply restrictions. In previous studies, the major reason for supplyrestrictionsintelecommunicationsisusuallyconsideredtobeeconomicunderdevelopment.Here we focus instead on the role of technical ef  fi ciency (or rather, inef  fi ciency) as a causeof supply restrictions. Derivation of measures of technical ef  fi ciency is discussed in thenext section. Other explanatory variables in the model are two indicators of the level of capital investment, namely, per capita national investment, and per capita investment intelecommunications. Finally, we include telecommunications staff per 100 telephone linesas a measure of labor use.Allmonetaryvariablesaremeasuredinconstant1985USdollarsadjustedforinternationalpurchasing power parity (PPP). Data on PPP and national investment were taken fromthe Penn Table (Summers and Heston, 1991) and from the CHASS website at TorontoUniversity.DataonthetelecommunicationssectorweretakenfromITU(1997)andvariousissues of the statistical yearbook of the International Telecommunication Union. Table 1gives the de fi nitions of the variables used in the study. Sample means across the period1982 – 1992 for each the 28 countries used in the panel data set are listed in Table 2.  SUPPLY RESTRICTED TELECOMMUNICATIONS MARKETS  163 Table 1.  Variable de fi nitions.Variables Type Description Sources (a) Variables used in stochastic frontier analysis (see Section 2) CONNECT Output No. of main telephone connections ITU (1997)INV TEL Input Total telecom investment (1985 $US) ITU (1997)INVEST Input Total investment in the economy (1985 $US) Penn tableLABOR Input Total telecom staff ITU (1997) (b) Variables used in modeling waiting times (see Section 3) WAIT Dept Expected waiting times, calculated ITU (1997)from waiting listsINV TEL/CAP Indept Per capita telecom investment (1985 $US) ITU (1997) and Penn TableINVEST/CAP Indept Per capita national investment in the Penn tableeconomy (1985 $US)LABOR/LINES Indept Staff per 100 main lines ITU (1997) Table 2.  Sample averages for the period 1982 – 1992. a WAIT INV TEL/CAP INVEST/CAP LABOR/LINESCountry (years) (1985 $US) (1985 $US) (per 100 Lines)Korea 0.450 67.8 1742.6 0.61Cyprus 0.641 60.8 1721.5 1.34Portugal 0.788 54.7 1112.8 1.38Malaysia 1.058 52.7 1335.4 2.75Ireland 1.112 64.48 1768.4 2.01S Africa 1.192 27.14 495.3 3.33Panama 1.714 13.76 479.9 2.01Chile 2.492 24.0 913.0 1.64Turkey 2.506 21.9 687.4 2.47Mexico 2.687 22.6 837.5 0.98Costa Rica 2.834 17.7 562.3 1.28India 3.391 3.92 158.6 8.48Indonesia 3.827 6.09 457.7 5.08Morocco 4.024 13.6 193.6 2.41Tunisia 4.511 12.7 321.8 2.37Fiji 4.873 25.5 453.1 3.77Greece 5.248 38.7 1221.6 0.88Venezuela 5.618 20.4 948.5 1.41Thailand 5.663 10.6 633.8 2.14Kenya 6.274 7.8 96.3 7.92Malawi 6.892 4.4 37.3 6.02El Salvador 7.353 7.6 133.8 5.79Peru 7.582 8.7 439.4 3.03Hungary 7.868 33.6 1270.5 6.65Sri Lanka 7.987 7.4 248.5 9.40Algeria 9.402 8.74 686.2 2.79Zambia 9.964 8.04 82.5 5.57Zimbabwe 21.008 5.96 152.8 4.64 a See Table 1 for variable de fi nitions.  164  BARTELS AND ISLAM 2. Measurement of Technical Ef  fi ciency (TE) Technical ef  fi ciencies can ’ t be measured directly and are usually derived from an estimatedfrontier production function. A decision-making unit ’ s (DMU) technical ef  fi ciency is mea-sured by taking the DMU ’ s actual output as a percentage of the potential output for thatDMU if it were operating at the estimated frontier. In the literature there are two commonlyused approaches to  fi tting frontier production functions, namely, stochastic frontiers, anddata development analysis (DEA). We have used both approaches in our analysis, but sincethe outcomes are qualitatively similar, we only describe the results based on the stochasticfrontier production function.Stochastic frontier production functions were  fi rst introduced, independently, by Aigneretal.(1977)andMeeusenandvandenBroeck(1977).Hereweusethepaneldataextensionof the srcinal model due to Pitt and Lee (1981):ln (  y it  ) = β  ln (  x  it  ) + ν it  − u it  , ( i  = 1 , 2 ,...,  N  ;  t   = 1 , 2 ,...,  T  )  (1)where  y it   is the output of the  i th  fi rm in time period  t  , and  x  it   is a  k  -vector of inputquantitiesusedbythe i th fi rminperiod t  .Therearetworandomterms; u it   isanon-negativerandom variable that represents the technical inef  fi ciency of the  fi rm, and  ν it   is the usualdisturbance term, which accounts for the effect on the output variable of measurement errorand other unobserved random factors. In line with common practice, the  ν ’ s are assumed tobe independent, identically distributed  N  ( 0 ,σ  2 ν  )  variables, and the  u ’ s non-negative  half  -normal  N  ( 0 ,σ  2 u )  variables. Maximum likelihood estimates of the model were obtainedusing the software package FRONT41 (Coelli, 1996). For computational convenience, thispackage uses an alternative parameterization to that shown in (1) suggested by Battese andCorra(1977),namely σ  2 = σ  2 u  + σ  2 ν  and λ = σ  2 u /σ  2 .Theadvantageofthisparameterizationis that  λ  always lies between zero and one.Given input vector,  x  it  , and output,  y it  , the technical ef  fi ciency of the  i th  fi rm at time  t  is de fi ned as the ratio of the observed output of the  fi rm to its potential output as obtainedfrom the frontier production function: TE  it   =  y it  exp[ β  ln (  x  it  + ν it  ) ] = exp[ β  ln (  x  it  ) + ν it  − u it  ]exp[ β  ln (  x  it  ) + ν it  ] = exp ( − u it  )  (2)Note that since  u it   is non-negative, the technical ef  fi ciency measure,  TE  it  , is bounded by 0and 1.Parameterestimatesforthestochasticproductionfrontierforourpaneldatasetof28coun-tries are shown in Table 3. The output variable used was CONNECT and the input variableswere INVEST = total investment in the economy; INVEST TEL = total investment in thetelecommunications sector; and LABOR = telecommunications staff (see Table 1).Wecantestthenullhypothesisthattherearenotechnicalinef  fi ciencyeffectsinthedatabytesting  H  0 : λ = 0againstthealternative  H  1 : λ >  0.Aone-sidedgeneralizedlikelihoodratiotestfortestingthesehypotheseshasbeenrecommendedbyCoelli(1995).Thecriticalvaluefor the test statistic, with size  α = 0 . 05, is 2.71. For our estimated model the appropriatestatistic has a value 192.6, and hence the hypothesis that there are no technical inef  fi ciencyeffects in the data is  fi rmly rejected. The estimated value of   λ  is 0.895, indicating that 90%  SUPPLY RESTRICTED TELECOMMUNICATIONS MARKETS  165 Table 3.  Maximum likelihood estimates of stochastic frontier. a Parameter Coef  fi cient Standard Error  t  -StatisticCONST 6.760 0.697 9.70INV TEL 0.431 0.067 6.41INVEST 0.499 0.122 4.10LABOR  − 0.150 0.131  − 1.15 σ  2 3.194 0.889 3.60 λ  0.895 0.031 29.0 a The output variable is CONNECT (see Table 1 for variable de fi nitions). Estimateswere obtained using the software package FRONT41 (Coelli, 1996). Table 4.  Estimated technical ef  fi ciencies (TE). a WAIT (Average for TE (Average forCountry Period 1982 – 1992) Period 1982 – 1992)Korea 0.450 0.786Cyprus 0.641 0.920Portugal 0.788 0.533Malaysia 1.058 0.290Ireland 1.112 0.323S Africa 1.192 0.405Panama 1.714 0.258Chile 2.492 0.253Turkey 2.506 0.859Mexico 2.687 0.281Costa Rica 2.834 0.245India 3.391 0.264Indonesia 3.827 0.092Morocco 4.024 0.178Tunisia 4.511 0.282Fiji 4.873 0.111Greece 5.248 0.718Venezuela 5.618 0.227Thailand 5.663 0.229Kenya 6.274 0.131Malawi 6.892 0.063El Salvador 7.353 0.194Peru 7.582 0.190Hungary 7.868 0.202Sri Lanka 7.987 0.054Algeria 9.402 0.230Zambia 9.964 0.089Zimbabwe 21.008 0.050 a TE ’ s are derived from the estimated stochastic frontier in Table 3. of the residual variance is due to the inef  fi ciency effect and only 10% to the random erroreffect.Using the estimated stochastic frontier production function, technical ef  fi ciencies can becalculated for each country and year. Table 4 shows, for each country, the mean of theseef  fi ciencies over the period 1982 – 1992.