THE CONTRIBUTION OF ICT TO ECONOMIC ACTIVITY: A GROWTH ACCOUNTING EXERCISE WITH SPANISH FIRM-LEVEL DATA

Transcription

1 THE CONTRIBUTION OF ICT TO ECONOMIC ACTIVITY: A GROWTH ACCOUNTING EXERCISE WITH SPANISH FIRM-LEVEL DATA Ignaco Hernando and Soledad Núñez Banco de España Banco de España Servco de Estudos Documento de Trabajo n.º 23

2 THE CONTRIBUTION OF ICT TO ECONOMIC ACTIVITY: A GROWTH ACCOUNTING EXERCISE WITH SPANISH FIRM- LEVEL DATA (*) January 22 Ignaco Hernando and Soledad Nuñez Banco de España. Research Department (*) We would lke to thank J. J. Camo, J. M. Campa, J. R. Cano, A. Esteban, A. Estrada, E. Ortega, M. Ortega, J. I. González Pascual, G. Pérez Qurós, F. Restoy, P. Schreyer, J. Vallés and semnar audences at Banco de España, 23 rd SUERF Colloquum, IVIE and Unversdad Públca de Navarra for helpful comments and suggestons. We also thank the Central Balance Sheet Offce of the Banco de España and A. Estrada and D. López-Saldo for provdng the data.

3 Abstract Ths paper uses a well-establshed growth accountng framework to measure the contrbuton of ICT goods (consdered as captal nputs) to output and labour productvty growth n the Spansh economy. We apply ths framework to a sample of around 13 Spansh frms per year over the perod The use of mcro-level data s especally useful for the purpose n hand. Frstly, our database provdes detaled breakdowns of captal. Ths helps mtgate the usual msmeasurement problems n obtanng captal stocks. Secondly, by avodng the usual avalablty lags assocated wth the use of aggregate data, we can focus on a more recent perod. The man fndngs may be summarsed as follows. 1) The use of ICT as a captal nput has made a postve and, relatve to ts cost share, sgnfcant contrbuton to output and productvty growth. 2) Ths contrbuton was hgher n the second half of the 199s. For ths perod, we estmate that the use of ICT nputs accounted for nearly one-fourth of labour productvty growth. 3) At a sectoral level, we fnd that there s a general rse n the share of ICT n total captal and a general reducton n ICT cost shares, drven by the sharp downward trend n the prces of ICT products. However, the contrbuton of ICT nputs dsplays a degree of heterogenety across sectors, owng to the dsparty of sectoral accumulaton rates of ICT nputs. Fnally, results at the frm level exhbt a notable heterogenety, although a majorty of frms have experenced an ncrease n the ICT captal growth rates and n the ICT contrbuton to growth. JEL categores: O33, D24, L63 Keywords: Informaton and Communcaton Technologes, Growth Accountng, Technologcal Change. 2 BANCO DE ESPAÑA / DOCUMENTO DE TRABAJO N. 23

4 1. Introducton Over the last decade, there has been enormous techncal progress n the nformaton and communcaton technologes (ICT) ndustres. These effcency gans have drven down the relatve prces of computers, software and communcatons equpment and have sgnfcantly stmulated the demand for ths type of good. As a consequence, the ICT- producng ndustres have experenced, at least n some economes, unprecedented growth rates and have contrbuted to the acceleraton of total factor productvty growth. Furthermore, the mpact of techncal advances n ICT on economc actvty goes beyond the drect mpact on ICT-producng ndustres. A potentally strong mpact stems from the adopton and use of new technologes n most sectors of the economy. In ths respect, the reducton n the prces of ICT captal goods encourages the accumulaton of ths type of nput. Consequently, the dffuson of ICT as a captal nput mght have contrbuted sgnfcantly to output and labour productvty growth. Fnally, an addtonal contrbuton of ICT to growth may arse from an acceleraton of TFP growth due to effcency enhancng effects arsng from the producton and adopton of ICT. However, the emprcal assessment of the role of ICT n economc actvty poses consderable statstcal problems. Frstly, the relevant nformaton s not avalable on a tmely bass (n the Spansh case, sectoral nformaton s avalable only wth a four-year lag). Secondly, detaled breakdowns of captal and nvestment are not usually accessble. Thrdly, sgnfcant measurement problems arse from the dffculty of constructng adequate prce ndces and of calculatng economc deprecaton for ICT captal goods. Gven these data lmtatons, the use of frm-level data, though they do not solve all measurement problems, represents a promsng avenue, and t s that whch we explore n ths paper. Our objectve s to examne the relatonshp between the use of ICT as a captal nput and the recent performance of productvty growth n Span. Therefore, rather than analysng the contrbuton of ICT-producng ndustres to economc development we adopt an nput-orented approach that focuses on the role of ICT as a captal nput n all sectors of the economy. For ths purpose, we estmate, usng a standard growth accountng framework, the contrbutons to output and productvty growth from the use of the dfferent nputs (ICT among them). Ths analytcal framework has already been used to estmate, wth aggregate nformaton, the growth contrbuton from the use of ICT captal n the U.S. and other ndustralsed economes -Olner and Schel (2), Schreyer (2), Coleccha and Schreyer (21) and Daver (21)-. 1 In ths paper, we conduct the analyss usng frm-level data. More precsely, we make use of a sample of Spansh frms over the perod , obtaned from the Central Balance Sheet Offce of the Bank of Span. The fnal sample ncludes about 13 frms per year and t provdes nformaton for suffcently detaled breakdowns of captal. By conductng the analyss at the frm level, we reduce dffcultes arsng from msmeasurement of captal stocks. As potental addtonal advantages, the use of ndvdual data 1 Most of the estmates of the contrbuton of ICT nputs to output growth concern the US economy. Addtonal references are Jorgenson and Stroh (2) and Whelan (1999). BANCO DE ESPAÑA / DOCUMENTO DE TRABAJO N. 23 3

5 would allow the dstnctve features of the fnancal structure of technology-ntensve frms to be dentfed and offers some ponters to the factors nfluencng the effect of ICT captal on productvty growth. Nevertheless, the ndvdual data also ental some problems. Thus, captal stock fgures have to be converted from book value to market value. Next, sample coverage, n terms of value added and employment s, for a few sectors, low. Further, the sample we use s defntely based towards large frms. Fnally, our sample does not enable us to consder labour qualty as a contrbutng factor to growth. Gven these drawbacks, our results should be vewed wth some cauton. Although we use ndvdual frm data, our man objectve s to derve some general conclusons about the ICT contrbuton to growth for the whole non-fnancal market economy. For ths purpose, we frst obtan sectoral fgures (we consder 17 sectors) by averagng frms behavour by sector. Thus, we mplctly assume that the average performance of the frms n the sample s representatve of the sector they belong to. We then obtan aggregate fgures by averagng sectoral results, weghtng them by ther share n the whole market economy. We check that our concluson about the ICT contrbuton to growth for the aggregate economy s robust to alternatve procedures of aggregaton of the ndvdual ICT contrbutons. Our results suggest that the ICT contrbuton to growth and ICT captal accumulaton rates have been relatvely sgnfcant although quanttatvely smaller than n the US. Thus, accordng to Olner and Schel (2), over the perod , ICT captal deepenng explaned almost one-quarter of each percentage pont of US output growth, whle ths fgure was around.1 p.p. for the Spansh case n the second half of the nnetes (1996-2). Over the perod , these authors report US annual growth rates for hardware and software of about 21% and 13%, respectvely. The correspondng growth rates for the Spansh economy were 12% and 9%, respectvely. These growth rates can be consdered as low f we take nto account that there s a most szeable gap between US and Span n terms of ICT captal deepenng. To our knowledge, for the Spansh case, only Daver (21) has analysed the growth mpact of ICT accumulaton wthn a growth accountng framework. Compared wth our fndngs, Daver reports a smlar ICT contrbuton for the whole perod consdered but a lower one for the second half of the nnetes. 2 Nevertheless, there are some substantal dfferences between Daver s study and the one presented here. Frst, Daver assumes perfect competton, computng the contrbuton to growth of factor nputs n terms of ncome shares, whle we relax ths assumpton and, therefore, compute these contrbutons n terms of cost shares. Second, he uses aggregate data from a very dfferent data set (ICT expendture taken from WITSA/IDC and Natonal Accounts OECD seres). 2 Thus, for the perod , Daver reports an annual ICT contrbuton to output growth of.36 p.p.. In ths paper the correspondng fgure s.35 p.p.. However, for the perod Daver fnds an ICT contrbuton of.34 p.p., meanwhle we fnd a correspondng fgure of.42 p.p.. 4 BANCO DE ESPAÑA / DOCUMENTO DE TRABAJO N. 23

6 The rest of the paper s organsed as follows. Secton 2 descrbes the analytcal framework. Secton 3 ntroduces our database payng specal attenton to the descrpton of the prce ndces for captal nputs and to the constructon of the captal stocks and the user costs of the captal nputs. A more detaled descrpton of the sample and the defnton of varables are relegated to Annex 1. Secton 4 presents the results for the whole economy, as well as for the 17 sectors consdered. Fnally, secton 5 concludes. 2. The analytcal framework: Neoclasscal growth accountng In ths paper we apply the neo-classcal growth accountng framework developed orgnally by Solow (1957). Ths framework has been extensvely appled n other studes on the ICT contrbuton to growth, such as Olner and Schel (2), Jorgenson and Stroh (2), Schreyer (2), Daver (2), among others. Our man departures from these authors are twofold. Frst, we use ndvdual frm data, and, second, followng Hall (199), we do not mpose perfect competton. 3 We start from a Cobb-Douglas producton functon (F) that relates frm value-added (Q) to seven nputs: labour (L), software (K sw ), hardware (K hw ), non-resdental buldngs (K bld ), ndustral equpment (K eq ), other equpment and furnture (K oeq ) and transportaton equpment (K trp ). 4 Thus: Q = θ F(L,K sw, K hw, K bld, K eq, K oeq, K trp ) where we assume that F dsplays constant returns to scale n factor nputs. Computng growth rates: q = L F K F θ + l + k (1) Q L Q K = sw, hw, bld, eq, oeq, trp where lower case letters correspond to the logarthms of the correspondng upper-case varables. The term θ captures output growth not accounted for by changes n factor nputs, and approxmates total multfactor productvty (TFP). Frst order condtons for cost mnmzaton are: 3 Gven that we use ndvdual data on a yearly bass, we consder ths strategy more approprate. 4 These are the breakdowns of captal that are avalable n our database. McGuckn and Stroh (22) hghlght the mportance of usng a detaled breakdown of captal n order to avod the mportant bases n the measurement of the productve mpact of the dfferent nputs that arse when the elastcty of all types of non-computer captal s ncorrectly restrcted to be equal. BANCO DE ESPAÑA / DOCUMENTO DE TRABAJO N. 23 5

7 F K r = mc F w and = (2) L mc where r s the rental prce of captal and w s the labour market wage. Gven that, wth constant returns, margnal cost (mc) s equal to average cost at the cost mnmzaton value of nputs, we can wrte: F L w Q = wl + r k (3) and F K r Q = w L + r k (4) Some smple algebra leads to: F L w L = w L + r k Q L = α L Q L (5) F K r K = w L + r k Q K = α K Q K (6) where α s the cost share of nput. Substtutng (5) and (6) n (1): q = α k + α k L l + TFP (7) Thus, n equaton (7), each nput s contrbuton s obtaned by multplyng ts rate of change by each factor s share n total cost (α ). 5 Addtonally, n the computaton of the cost shares, we ntroduce the assumpton that all types of captal earn the same compettve rate of return at the margn, net of deprecaton and captal gans or losses mpled by the changes n the prces of captal goods. Thus, we are assumng that frms allocate resources effcently. To mpose the same rate of return for all captal assets mples a very hgh gross rate of return for ICT to offset the rapd deprecaton and the captal losses arsng from the declne n ICT prces. Groupng terms n equaton (7) yelds: 5 An alternatve approach would be to estmate the parameters of the producton functon. Although such an approach does not requre the ntroducton of the neo-classcal assumptons, t requres assumng the homogenety of the parameters of the producton functon, at least, across sectors. 6 BANCO DE ESPAÑA / DOCUMENTO DE TRABAJO N. 23

8 q = c l + c ITC + c OTHERK + TFP (8) where c l s the contrbuton of the labour nput to value added growth, c ICT s the contrbuton of ICT captal and c OTHER K s the contrbuton of non-ict captal, beng: c l =α l l (9) c ITC = α k + α sw sw hw k hw (1) c OTHERK = α k + α k + α k + α bld bld eq eq oeq oeq trp k trp (11) Alternatvely, by rearrangng equaton (7) we can obtan a smlar decomposton for labour productvty growth: q l = cl ITC + cl OTHERK + TFP (12) where cl ITC = α ( k l) + α ( k l) sw sw hw hw (13) cl OTHERK = α ( k l) + α ( k l ) + α ( k l) + α ( k l) bld bld eq eq oeq oeq trp trp (14) Accordng to ths expresson, growth n labour productvty s explaned by the ntensty of the process of captal deepenng (ncrease n the amount of captal per unt of labour) and by the growth rate of TFP. The neoclasscal growth accountng framework provdes a smple analyss of the proxmate sources of economc growth. It decomposes the growth rate of output nto the sum of two factors: the rate of ncrease of nputs and the multfactor productvty growth. Thus, ths framework represents a lmted approach to understandng the process of economc growth. It does not adequately explan whch are the underlyng factors drvng the substtuton processes between factors or whch are the causes behnd the growth of TFP. In our case, we calculate each component of equatons (8) and (12) (that s, the value added growth rate, factor nput contrbutons and the TFP growth rate) for each frm n the sample. To obtan the components of equatons (8) and (12) for the total non-fnancal market economy from the components computed at the frm level we take two addtonal steps. Frst, we average these components by sector. For the sake of robustness, n ths step, we consder two alternatve sectoral breakdowns (the Natonal Accounts sectoral breakdown BANCO DE ESPAÑA / DOCUMENTO DE TRABAJO N. 23 7

9 nto 71 ndustres and the breakdown nto 17 sectors used n Estrada and López-Saldo(21) 6 ) and we use two methods of aggregaton: a) we compute the sectoral contrbutons as smple averages of the ndvdual ones, and b) we add up the ndvdual data for value added and productve nputs and then obtan the sectoral contrbutons. 7 Second, we obtan the fgures for the total non-fnancal market economy by takng the average of the sectors for each component, the sectors beng weghted by ther share n total value added. Sectoral weghts are calculated usng data taken from Estrada and López-Saldo (21) n the case of the 17-sector breakdown and are drectly taken from the Natonal Accounts n the case of the 71-sector breakdown The data From the prevous secton, the contrbuton of each type of captal to output growth depends on ts cost share and on ts accumulaton rate. Therefore, the valdty of ths exercse depends on the accurate measurement of two elements: the captal stocks and ther user costs. Before descrbng the method of constructon of these two elements, we dscuss our choce of prce ndces for captal nputs as these are an essental component n the computaton of both the captal stocks and the user costs. Prce ndces for captal nputs. The choce of an approprate deflator for captal nputs s crucal both for the measurement of the captal stocks and for the computaton of the user costs. Ths task s partcularly delcate n the case of ICT captal goods. Most of these ICT captal goods have undergone sgnfcant qualty changes that f not properly taken nto account wll lead to an overestmaton of the prce change n ICT captal goods and to an underestmaton of the correspondng captal stocks. Therefore, the use of prce ndces for ICT captal goods based on the applcaton of hedonc technques seems to be an essental tool to decompose the change n the nomnal captal stocks nto ther prce and quantty components. Gven that, for the Spansh economy, there s no prce ndex for ICT goods n constant qualty terms 9, we apply an ndrect approach based on Schreyer (2)- to obtan an adequate ICT deflator. Schreyer constructs the ICT prce deflator for a gven country n such a way that the dfference between the ICT prce change and the prce change n all other nvestment goods for that country s equal to the dfference between the same prce changes for the US economy. 6 Estrada and López-Saldo (21) construct a database on a yearly bass, usng Natonal Accounts, wth nformaton on several economc varables for 17 sectors, excludng the non-market economy and fnancal sector, for the perod The use of ths breakdown of the market economy nto 17 sectors was determned by the avalablty of ths database. 7 In Annex 2 we dscuss the choce of the aggregaton method and we present the results for the alternatve procedures of aggregaton of the nformaton computed at the ndvdual level. 8 In the case of the 71-sector breakdown, snce sectoral data from Natonal Accounts (ESA 95) s only avalable for the perod , we use 1995 weghts for perod and 1997 weghts for the perod In order to obtan annual varaton for these weghts, we have corrected them wth the weghts for the 17 sectors consdered n Estrada and López-Saldo (21). 9 Izquerdo and Matea (21) provde a seres of hedonc prces for personal computers n Span. We have not used ths seres because personal computers are just one product among those ncluded n our hardware category. 8 BANCO DE ESPAÑA / DOCUMENTO DE TRABAJO N. 23

10 We closely follow Schreyer s methodology and compute the prce deflator for captal nput (P ) by assumng that the rato of the deflator of captal nput to the GDP deflator n Span s the same as the correspondng rato n the US. We have appled ths procedure for deflaton to ICT captal nputs: hardware and software To test the senstvty of the results to the choce of deflator for the ICT captal nputs we have alternatvely used a set of prce ndces for these nputs taken drectly from Spansh offcal statstcal sources. Fgure 1 dsplays the tme profle of the deflators for the captal nputs and, n partcular, t provdes a comparson between the two sets of deflators for the ICT nputs. 1 In the case of hardware, the ndex computed usng US deflators shows a sgnfcantly more pronounced declne. Ths result clearly shows to what extent not takng nto account the qualty changes n these ICT goods ntroduces a serous bas n the estmaton of ther prce changes. Usng Spansh statstcal sources, we are unable to obtan a deflator for software. As ths fgure makes clear, usng a common deflator for hardware and software s hghly msleadng. Captal stocks Our database provdes accountng data correspondng to the sx types of captal assets already mentoned: software (K sw ), hardware (K hw ), non-resdental buldngs (K bld ), ndustral equpment (K eq ), other equpment and furnture (K oeq ) and transportaton equpment (K trp ). It should be mentoned that n our sample, software captal comprses successful R+D nvestment, and hardware captal ncludes communcatons equpment. In ths paper we construct measures of the captal stocks usng ths accountng nformaton. More precsely, we have nformaton on the net book value (at hstorc prces) of the sx types of captal and we can construct the average age of each captal tem (as the two-year average of the rato of total accumulated deprecaton to current deprecaton). Usng the prce ndces for nvestment goods already descrbed, the book values of the captal stocks and ther average ages we can obtan the value of the captal stocks at constant and current prces. 11 We apply ths procedure to all the observatons for each frm. An alternatve approach to ths procedure would be the perpetual nventory method, whch combnes the nformaton on the captal stock at constant prces n an ntal year wth nformaton on nvestment volumes for the subsequent years. 12 As already mentoned, the avalablty of mcro-level nformaton has undenable advantages for the purpose of ths paper. However, the use of accountng data to obtan measures of captal stocks also has some lmtatons. In partcular, gven that the avalable nformaton s on the book value (net of economc deprecaton) of fxed captal we are constraned to construct wealth measures of the captal stocks,.e, measures of the market value of the assets of the frm. However, as s thoroughly dscussed n Olner and Schel (2) and Schreyer (2), the relevant measure of captal nputs for a growth accountng exercse s 1 See Annex 1 for the detaled defnton of both sets of prce deflators. 11 We essentally apply the same methodology as n Hall (199). Further detals on the computaton of the captal stocks are gven n Annex Unfortunately, the lack of suffcently detaled breakdowns of nvestment prevents us from adoptng ths standard approach n the constructon of captal stocks. BANCO DE ESPAÑA / DOCUMENTO DE TRABAJO N. 23 9

11 that provded by the productve stocks of the nputs, that s, the productve capacty of the stock. In other words, the productve stocks take nto account the physcal decay of the assets whereas the wealth stocks reflect the economc deprecaton. For most of the captal assets, these concepts are related but ths s not the case for computers. Computers experence very lttle physcal decay but they suffer a very hgh economc deprecaton (as they have a very short lfecycle). As we are constraned to use a wealth measure for the captal stocks, our estmates of the growth contrbutons of ICT captal assets (for whch the dfference between the productve and the wealth stock s relevant) wll be based downwards. Fgure 2 shows the growth rates of the ICT (types K sw and K hw ) captal stocks at constant prces and the changes n the rato of ICT captal to total captal. Both these ICT captal goods have experenced outstandng growth rates (much hgher than those for non-ict captal). As a consequence, the share of ICT captal goods n the total captal stock has steadly ncreased over the perod consdered and ths accumulaton process has substantally accelerated n the second half of the decade. Thus, the weght of ICT captal n the total captal stock was almost 11% n 2, twce the correspondng fgure for 1992 (5.2%). Ths process has been smlarly ntense for software (ts weght n the total captal stock has rsen from 1.5% to 3.1%) and for hardware (from 3.7% to 7.8%). These fgures suggest that a strong process of substtuton of ICT captal for other types of captal nput has taken place, manly drven by the sharp downward trend n the prces of ICT nputs. Cost shares Each factor s cost share s defned as the rato of the cost of the nput to total cost of output whch, under the neoclasscal assumptons, s equal to total costs. In the case of labour, ts cost can be drectly obtaned from the accountng data. In the case of the captal nputs, ts computaton gven by the product of the captal stock and ts rental prce or user cost- s not so straghtforward. The defnton of the user cost of the captal nput K s gven by the product of three terms: the acquston prce (P ), the gross rate of return (R ) and a fscal correcton factor (f). UC = P R f In what follows, we focus on the computaton of the gross rate of return. The acquston prce has been prevously dscussed and the fscal correcton factor, whch s constructed at a sectoral level, s descrbed n more detal n Annex 1. Ths fscal correcton factor, whch s assumed to be common to all types of captal, reflects taxes and fscal ncentves. The gross rate of return for captal nput K s gven by the followng expresson: R = r +δ π 1 BANCO DE ESPAÑA / DOCUMENTO DE TRABAJO N. 23

12 where r s the net rate of return common to all types of captal (representng the opportunty cost of the nvestment) 13, δ s the deprecaton rate (whch proxes the loss n market value due to ageng) and π s the captal prce nflaton (reflectng captal gans or losses). Two factors determne the evoluton of the cost share of each captal nput: ts user cost and ts weght n total captal. Fgure 3 dsplays the path, n real terms 14, of the frst of these factors, the rental prce or user cost, for all the types of captal nputs consdered. Gven that the deprecaton rates and the fscal correcton factor have remaned qute stable over the sample perod, the tme profle s mostly explaned by the captal prce nflaton and by the opportunty cost of nvestment. Provded that ths last factor s assumed to be common to all types of captal, the prce changes n captal goods s left as the man cause explanng dfferences n the changes n the cost shares across types of captal. Especally remarkable s the fall n the user costs of ICT captal, partcularly hardware, relatve to the user costs of other types of captal nput. Ths relatve behavour of user costs s decsve n explanng the exstence of strong substtuton effects between dfferent types of captal. As Fgure 4 shows, there has been a markedly dfferent tme profle for the cost shares of ICT and non-ict captal nputs. On the one hand, n the case of non-ict captal nputs, ther cost share has shown a sgnfcant downward trend throughout the sample perod, manly drven by the declnng weght of non-ict captal n total fxed captal. On the other hand, the cost share of ICT captal goods has exhbted a slght downward trend whch s the result of two effects of large magntude but opposte sgn: the ncreasng weght of ICT captal nputs and the sharp declne n ther rental prce. Ths declnng trend n the cost share of total ICT captal s mostly explaned by the behavour of the cost share of hardware. However, n the case of software, the declne n ts rental prce has not been so sharp as to cancel out the ncrease n ts weght n total fxed captal. Thus, we observe a slghtly growng cost share for software. 4. Growth contrbuton from the use of ICT as a captal nput Once the nformaton on the rates of growth of dfferent nputs and ther costs shares s avalable, usng equatons (8) and (12) we can straghtforwardly approxmate the decomposton of output and labour productvty growth. Gven that our analyss s performed at the frm level, we report three types of results. Frst, we provde a decomposton of output growth for the whole market economy. As already mentoned, we compute ths decomposton n three steps. In the frst step, we compute each element of equatons (8) and (12) at the ndvdual level. Next, by takng sectoral averages for each of these components, we obtan a decomposton of output 13 In the constructon of the net rate of return r, whch has been assumed to be common for all frms wthn the same sector, we have not taken nto account the composton of fnancng. Pt Kt r + δ π, where P tk t s captal nput, n nomnal terms, and P 95K t s captal nput at constant prces. P 95Kt 14 That s ( ) BANCO DE ESPAÑA / DOCUMENTO DE TRABAJO N

13 growth at the sectoral level. 15 Then, usng value-added weghts, we aggregate these sectoral results. Second, we provde a dscusson of the results at the sectoral level. Fnally, we report the dstrbuton of ndvdual ICT captal contrbutons to growth AGGREGATED MARKET ECONOMY RESULTS Table 1 presents the decomposton of output growth. The frst column reports the results for the overall perod, Durng these years, value-added for the non-fnancal market economy rose at an annual average rate of 2.9%. The contrbuton of ICT captal (lne 5) represented.38 percentage ponts, of whch.16 was explaned by computer software and.22 by hardware. Therefore, the ICT contrbuton to growth was small durng the 9s, beng around half of the contrbuton accounted for by other fxed captal. Nevertheless several comments should be made. Frst, over the perod under study, the ICT captal stock ncreased at an annual average rate of 7.5%, whle non-ict captal rose at a rate of.9%. Consequently, the contrbuton of ICT captal to growth was moderate because ICT captal stll represents a very modest fracton of total captal stock (7.6%) and so, ts share n total cost s rather small (2.%). In other words, relatve to ts share n ether total cost or n the total fxed captal stock, the contrbuton of ICT captal to growth has been consderable (see lnes 26 and 3 of Table 1). Second, throughout the analysed perod, the ICT contrbuton has been ncreasng (see columns 2 and 3 of Table 1). Thus, for the perod , the ICT contrbuton to annual value-added growth reached.45 percentage ponts, 55% hgher than the average contrbuton for the perod Conversely, non-ict captal contrbuton has sgnfcantly decreased. Thrd, the rse n the ICT contrbuton to growth durng the second half of the 9s s explaned by an acceleraton n ICT accumulaton rates, snce the ICT cost share declned slghtly between these two perods. Thus, annual growth rates for new technology equpment rose from 4.4% durng to 1.1% durng Gven these growth rates, whch were also consderable n nomnal terms, the reducton n the ICT cost share s explaned, as already mentoned, by the exceptonal declne n ICT captal goods prces. Fnally, t should be mentoned that there are some dfferences between the contrbutons of computer software and hardware to growth. Thus, whle the contrbuton of computer software to growth rose sgnfcantly between and , the contrbuton of computer hardware ncreased moderately. Nevertheless, the growth rate of computer hardware accelerated consderably more than that of computer software. In spte of 15 As already mentoned, n ths step we have consdered two alternatve sectoral breakdowns and we have computed both smple and weghted averages. In what follows, we present the results correspondng to the case of the 17-sectors breakdown and where the sectoral fgures are computed as smple averages of the ndvdual ones. Ths aggregaton method s the one that best approxmates the growth rates observed for several economc varables wth Natonal Accounts data. Nevertheless, the results are robust to the alternatve procedures of aggregaton (see Annex 2 for detaled results). 12 BANCO DE ESPAÑA / DOCUMENTO DE TRABAJO N. 23

14 that, the sharp declne n user costs of computer hardware explans the moderate ncrease of ts contrbuton to output growth. Table 2 presents a decomposton of labour productvty growth. For the overall perod, labour productvty grew at an average annual rate of 2.22%. The process of ICT captal deepenng showed an average contrbuton to productvty growth of.35 percentage ponts. Furthermore, ths contrbuton surged n the second half of the 9 s, when the average annual ICT contrbuton reached.38 percentage ponts, up from.31 n the perod In relatve terms, ths ncrease n the ICT contrbuton s even sharper. Thus, whle n the perod the ICT contrbuton explaned, on average,.11 of each percentage pont of labour productvty growth, ths fgure was.23 n the perod By contrast, the contrbuton of non-ict captal to labour productvty growth declned over the perod consdered. These results suggest then that the slowdown n labour productvty growth durng the second half of the 9s was mostly explaned by a reducton n non-ict captal deepenng growth rate, snce the slowdown n TFP growth was not as sharp as that n labour productvty growth SECTORAL RESULTS Table 3 shows the sectoral breakdown we have consdered, together wth each sector s share n value-added, the number of observatons n the sample, ts coverage (n terms of value added and employment) and the number of frms n the sample for whch the ICT captal stock s zero. For most sectors, sample coverage can be consdered as relatvely hgh, although for sx of them t s lower than 15% (n terms of value added), and, therefore, results for these sectors should be vewed wth more cauton. Fgure 5 presents sectoral ICT cost shares, ICT shares n total fxed captal, ICT captal growth rates and the contrbuton of ICT captal to growth. These sectoral varables have been computed by averagng frm values wthn the sector. Therefore, the results presented here correspond to the average frm behavour n the sample, and we take them to be representatve of the correspondng sector. The results at the ndustry level can be summarsed as follows: Frst, all sectors, except Other Market Servces 16, dsplay a small ICT captal share n total fxed captal (see panel 1 of Fgure 5), rangng from 2.2% (for Rubber and Plastc products) to 8.1% (Communcaton Servces). Consequently, cost shares are also small. However, throughout the perod consdered, all sectors experenced a sgnfcant ICT captal growth rate, contrastng wth that for non-ict captal (see panel 3). Gven the modest fracton of new technology captal, the average ICT contrbuton to output growth was small, rangng from.15 percentage ponts for Buldng and Constructon to 1.4 p.p. for Communcaton Servces. However, for all sectors ths contrbuton was been, relatve to cost shares, much hgher than that of the non-ict captal stock (see panel 4). 16 Other Market Servces, whch nclude real estate, busness servces and computer servces actvtes, presents an average ICT captal share of 15.5%, far above that n the other sectors. BANCO DE ESPAÑA / DOCUMENTO DE TRABAJO N

15 Second, all sectors have experenced a rse n the share of ICT captal n total fxed captal throughout the perod consdered. In most sectors, ths substtuton of ICT captal for non- ICT captal, explaned by relatve prce developments, have accelerated n the second half of the perod consdered. Thus, for 15 sectors, annual ICT growth rates were hgher durng than durng (see panel 7 of Fgure 5), ths acceleraton beng especally remarkable for Communcaton Servces and Other Market Servces. In spte of these accumulaton rates, cost shares have been lower n the second half of the analysed perod for most sectors (13 of them), reflectng the sgnfcant declne experenced n ICT captal good prces (see panel 6). Thrd, for most sectors (13 of them), the ICT growth rate acceleraton outweghted the declne n the cost share, and, consequently, the ICT contrbuton to output growth ncreased n relatve to (see panel 8). However, n terms of labour productvty growth, only 9 sectors experenced a hgher ICT contrbuton n the second half of the 9s (see panel 9) 17, despte the general rse n ICT captal deepenng 18. In sum, although the ICT contrbuton to growth across sectors dsplays a degree of heterogenety, most sectors show smlar man results to those for the whole market economy. That s, the ICT contrbuton to growth s small n absolute terms, but t was ncreasng over the perod covered. Ths ncrease s mostly explaned by an acceleraton n ICT captal accumulaton. ICT contrbuton to growth n ICT-producng sectors It s often argued that the dramatc prce declne n ICT captal goods over recent decades can be explaned by the effcency gans n the ICT-producng sectors. Therefore, t seems worth analysng growth developments, n terms of output and productvty, n those sectors producng goods and servces related to ICT 19. We have consdered three ICT producng sectors: ICT manufacturng, whch comprses the producton of ICT goods, ICT Communcatons and ICT Computer Servces 2. It should be ponted out that we do not have value-added deflators wth the requred level of dsaggregaton. Therefore value-added, labour productvty and TFP are, probably, mperfectly measured. Results for the analyss performed are presented n Fgure 6. ICT-producng sectors have experenced hgher ICT captal growth rates and hgher ICT contrbutons to value-added growth than other economc sectors. Ths s especally the case for the second half of the sample perod. Value-added growth rates n these sectors have been sgnfcantly hgher than those for the rest of the economy. Therefore, ICT-producng sectors have contrbuted postvely to output 17 Nevertheless, relatve to labour productvty growth, the ICT contrbuton ncreased n 11 of the 17 sectors. 18 The ICT captal labour rato rose n 16 of the 17 sectors. 19 For a more detaled study of ICT-producng sectors see Nuñez (21). 2 More specfcally, usng NACE/93, ICT manufacturng ncludes dvsons 3 and 31 and groups 313, 332 and 333, ICT Communcatons group 641 and ICT computer servces corresponds to dvson BANCO DE ESPAÑA / DOCUMENTO DE TRABAJO N. 23

16 growth, although gven the modest share of ICT value-added n total economy (around 5.2%), ths contrbuton has been small. In terms of total factor productvty, ICT manufacturng and ICT Communcatons have experenced much hgher growth rates than other sectors. Besdes, these growth rates have accelerated n the second half of the sample perod, n contrast to the slowdown n TFP growth n the whole market economy. Table 4 reports the contrbuton of ICT ndustres to total market economy TFP growth 21. For the perod , ths contrbuton was, n annual average terms,.17 p.p., rsng to.19 p.p. n the perod Relatve to total economy TFP growth, these contrbutons were 14% and 19%, respectvely, whch can be consdered hgh f we take nto account that ICT producng sectors account for only 5% of total value added. More mportantly, these hgh relatve contrbutons mply that the other branches of actvty, wth a much hgher weght, have recorded a very low and declnng rate of TFP growth. These results mght suggest therefore that the use of ICT has not, as yet, gven rse to postve spllover effects that have translated nto ncreases n productve effcency for the whole economy, or, f there have been any, they have not been able to offset the negatve effect of other determnants of total productvty. 4.3 CROSS-SECTIONAL DISTRIBUTION OF ICT CONTRIBUTIONS TO VALUE ADDED GROWTH In ths secton we try to provde an overvew of the results obtaned at the ndvdual level. For ths purpose, Fgure 7 dsplays the cross-sectonal dstrbutons of ICT contrbutons to value-added growth and of ther two man determnants, the ICT cost shares and the ICT accumulaton rates. These dstrbutons are presented both for the and perods. As was already clear from the sectoral results, the average decomposton of output growth hdes very heterogeneous ndvdual behavours. Panel 1 of Fgure 7 shows that the dstrbuton of ICT contrbutons to output growth s hghly skewed to the left. Whereas the average ICT contrbuton to output growth was nearly.4 percentage ponts (see Table 1), around 75% of the frms exhbt an ICT contrbuton below ths average value. And ths contrbuton s even negatve for a sgnfcant fracton of frms. As can be seen from Panel 2, these negatve contrbutons are drven by the presence of negatve accumulaton rates. In most cases, these negatve accumulaton rates arse from the fact that the gross ICT nvestment s not enough to offset the hgh deprecaton rates of the nstalled ICT captal. Only n a small number of cases are sales of ICT captal goods observed. The cross-sectonal dstrbuton of ICT cost shares s agan extremely skewed to the left (see Panel 3). Almost 8% of the frms have ICT cost shares below the average ICT cost share (2.% for the whole perod). It s also noteworthy that the fracton of frms wth a zero ICT cost share s very low (around 5% of the sample). Fnally, t s nterestng to analyse how these dstrbutons have evolved over the sample perod. Comparng the perods and , the rse n the average ICT contrbuton from.29 to.45 percentage ponts (see Table 1) s also reflected n a slght shft to the rght n the dstrbuton of ICT contrbutons. Ths shft s especally vsble n the lower tal of 21 Ths contrbuton s computed as the product of TFP growth n ICT ndustres and the value added weght of ICT ndustres (see Schreyer, 21). BANCO DE ESPAÑA / DOCUMENTO DE TRABAJO N

17 the dstrbuton wth the reducton n the percentage of frms wth negatve contrbutons. Thus, the percentage of frms wth a negatve ICT contrbuton decreases from 33% n the early nnetes to 24% n the second half of the sample perod. Nevertheless, the changes, of opposte sgn, n accumulaton rates and prces of ICT goods explan the notable stablty of ths dstrbuton. Thus, the small change n the dstrbuton of ICT contrbutons s mostly drven by the sgnfcant shft to the rght n the dstrbuton of ICT accumulaton rates that offset the shft to the left n the dstrbuton of ICT cost shares. Agan, ths shft s more perceptble n the lower tal of the dstrbuton. For example, the percentage of frms wth an ICT cost share below 1% rses from 51% n the frst half of the decade to 6% n the late nnetes. 5. Conclusons Ths paper examnes the role played by ICT captal as an nput factor and, more specfcally, as a factor contrbutng to output growth n the Spansh Economy n the perod For ths purpose, we use a standard growth accountng framework and a frm-level database. In order to obtan a general concluson regardng the ICT contrbuton to growth for the whole non-fnancal market economy, we aggregate the ndvdual results n two steps. Frst, we obtan sectoral fgures by averagng frms results by sector. Thus, we mplctly assume that the average performance of the frms n the sample s representatve of the sector they belong to. We then obtan aggregate fgures by averagng sectoral results, weghtng them by ther share n the whole market economy. The use of frm-level data s helpful to overcome some dffcultes assocated wth the use of aggregate data, manly the avalablty lags and the msmeasurement of captal stocks. However, ndvdual data also pose some problems for the purpose n hand. In partcular, the uneven coverage of the sample, both by sector and by sze of frm, and the need to transform accountng data nto nformaton that s meanngful n economc terms, represent mportant lmtatons. Bearng n mnd these drawbacks, our results should be vewed wth some cauton. Our man fndngs may be summarsed as follows. Frst, the use of ICT as a captal nput has made a postve and, relatve to ts cost share, mportant contrbuton to output and productvty growth. Over the whole sample perod consdered, the contrbuton of ICT equpment amounts to about one-thrd of the entre contrbuton of fxed captal to both output and labour productvty growth. Ths s especally noteworthy f we take nto account that the cost share for ICT captal nputs represents around one-tenth of the cost share for the total fxed captal. Second, ths contrbuton has been hgher n the second half of the 199s, n spte of the slght decrease n the cost share of ICT captal goods. For ths perod we estmate that the use of ICT nputs accounted for nearly one-fourth of the labour productvty growth, representng around 55% of the entre contrbuton of fxed captal. Thrd, at a sectoral level, we fnd that there s a general rse n the weght of ICT n total fxed captal and a general reducton n ICT cost shares drven by the sharp downward trend n the prces of ICT products. However, the contrbuton of ICT nputs dsplays a certan sectoral heterogenety explaned by the dsparty of accumulaton rates of ICT nputs across sectors, although most sectors have experenced a 16 BANCO DE ESPAÑA / DOCUMENTO DE TRABAJO N. 23

18 hgher contrbuton to growth n the second half of the 199s. Fnally, at the ndvdual level frms exhbt notable heterogenety, although a majorty recorded hgher ICT captal growth rates n the second half of the 9s. Although ICT captal growth rates have been notable, they are stll well below those observed n the US economy. Consequently, they are not suffcently hgh to narrow the gap n new technology captal observed between the Spansh and US economes. A fnal remark concerns TFP growth. The results presented here show a slghtly lower TFP growth rate for the second half of the 9s. However, our approach does not allow us to draw any concluson on the lnk between ICT growth and TFP growth rates. In other words, the growth accountng framework provdes a valuable analyss of the proxmate sources of economc growth, but t does not adequately explan whch are the underlyng factors drvng the processes of substtuton between factors or whch are the causes that le behnd TFP growth. These are the types of ssue we plan to address n future research. We thnk that frm-level data s especally useful to deal wth them, snce they allow the dstnctve features of the technology-ntensve frms and of the frms dsplayng a hgh productvty growth to be dentfed. BANCO DE ESPAÑA / DOCUMENTO DE TRABAJO N

19 Table 1 ICT CONTRIBUTION TO VA GROWTH Results for the whole non-fnancal market economy (1) Total perod VA growth rate Contrbuton from: 2. Labour (n hours) Software Hardware ICT (3+4) Rest of captal Total factor productvty Cost shares (%) (2) 8. Software Hardware ICT (8+9) Rest of captal Growth rate of captal stocks (%) (2) 12. Software Hardware ICT Rest of captal Rato of ICT captal to total fxed captal (%) 2. Software Hardware ICT Rest of captal Contrbuton to VA growth relatve to cost shares 24. Software (3/8) Hardware (4/9) ICT (5/1) Rest of captal (6/11) Contrbuton to VA growth relatve to total fxed captal share 28. Software (3/2) Hardware (4/21) ICT (5/22) Rest of captal (6/23) (1) Computed by averagng sectoral results, weghted by ther share n total value-added. Sectoral results correspond to the average for ndvdual frms n the correspondng sector. (2) Note that the product of average cost share by average captal growth rates s not the same as the average contrbuton to growth. 18 BANCO DE ESPAÑA / DOCUMENTO DE TRABAJO N. 23

20 Table 2 ICT CONTRIBUTION TO LABOUR PRODUCTIVITY GROWTH Results for the whole non-fnancal market economy(1) Total perod Labour productvty growth (2) contrbuton from: 1. Software Hardware ICT (1+2) Rest of captal Total factor productvty Memorandum tems: Growth rate of labour Captal-labour rato (3) Software Hardware ICT Rest of captal (1) Computed by averagng sectoral results, weghted by ther share n total value-added. Sectoral results correspond to the average for ndvdual frms n the correspondng sector. (2) In hours (3) Captal stock (n mllons of pesetas) per 1 hours of labour BANCO DE ESPAÑA / DOCUMENTO DE TRABAJO N

21 Sector Table 3 Sectoral classfcaton and fnal sample coverage Correspondence NACE/93 (2 dgt classf.) Value added sample coverage Sample descrpton Employment Value sample added per coverage frm (1995 ESP m) Employees per frm % of frms wth ICT= 1.Agr., Forestry &Fshery 1, 2, Fuel and Power Products 1, 14, 23, Ferrc & Non-Ferrc & Metals Non Metal. Mnerals & Mneral Prod Chemcal Products Machnery Transport Equpment Food, Beverages & Tobacco Textles,Cloth., Leather & Footw Other Manufacturng Products 2, Paper and Prntng Products Rubber and Plastc Products Buldng and Constructon Repar, Wholesale,Retal & hostn Transport Communcaton Servces Other Market Servces (1)Taken from Estrada and López-Saldo (21) Share n total value added (%) (1) Total num obs Annual averages 2 BANCO DE ESPAÑA / DOCUMENTO DE TRABAJO N. 23

22 TABLE 4 CONTRIBUTION OF ICT PRODUCING SECTORS TO DTFP D TFP Weght n VA Contrbuton to DTFP total eco. (a) % % p.p. % (b) % % p.p. % (b) ICT manufacurng ICT comuncatons Computer servces Total ICT (c ) Memorandum tem: Total market economy (a) Computed as the product of TFP and weght n VA (b) Relatve to TFP of total market economy (c) Computed by addng the contrbutons of the 3 sectors nvolved D TFP Weght n VA Contrbuton to DTFP total eco. (a) BANCO DE ESPAÑA / DOCUMENTO DE TRABAJO N

23 FIGURE 1 PRICE DEFLATORS FOR INVESTMENT GOODS 12 Software 11 Hardware Non-resdental constructon 15 Industral equpment Other equpment and furnture 15 Transportaton equpment Usng US deflators Usng Spansh deflators 22 BANCO DE ESPAÑA / DOCUMENTO DE TRABAJO N. 23

24 FIGURE 2 1. Growth rates of ICT captal nputs (whole economy) Software Hardware Total ICT 2. Composton of captal stock (whole economy) ICT captal / total captal rato Software Hardware Total ICT Fgures dsplayed correspond to the average of sectoral ones, weghted by ther shares n total value added. BANCO DE ESPAÑA / DOCUMENTO DE TRABAJO N

25 FIGURE 3 USER COSTS OF CAPITAL (1992=1) Software Non-resdental buldngs Industral equpment Other equpment Transportaton equpment Hardware FIGURE 4 Cost shares of ICT captal nputs (whole economy) (%) Non-ICT Total ICT 24 BANCO DE ESPAÑA / DOCUMENTO DE TRABAJO N. 23