CEP Discussion Paper No 629 April A Statistical Framework for the Analysis of Productivity and Sustainable Development.

Transcription

1 CEP Dscusson Paper No 629 Aprl 2004 A Statstcal Framework for the Analyss of Productvty and Sustanable Development Ncholas Oulton

2 Abstract To analyse the consequences of the changng economc structure of the UK, we need a set of statstcs broken down by ndustry that are consstent wth the whole economy measures avalable from the natonal accounts. The theory of growth accountng then provdes a framework n whch the contrbuton of each ndustry to the natonal economy can be measured and assessed. Ths paper dentfes the obstacles currently facng a researcher tryng to mplement ths approach. It makes a number of recommendatons for the mprovement of offcal statstcs. Keywords: Natonal accounts, growth accountng, productvty JEL: C82, O47, Q01 Ths paper was produced as part of the Centre s Productvty and Innovaton Programme. The Centre for Economc Performance s fnanced by the Economc and Socal Research Councl. Acknowledgements Ths paper has been prepared for the Allsopp Revew of Statstcs for Economc Polcy Makng. I am grateful for fnancal support from the Allsopp Revew, HM Treasury, towards the preparaton of ths paper and for comments from Chrs Allsopp, Mke Wllams and Geoff Tly. But the vews expressed are my own and are not necessarly those of the Revew team and of course I am responsble for any errors. Ncholas Oulton s a Senor Vstng Research Fellow at the Centre s Productvty and Innovaton Programme. Publshed by Centre for Economc Performance London School of Economcs and Poltcal Scence Houghton Street London WC2A 2AE All rghts reserved. No part of ths publcaton may be reproduced, stored n a retreval system or transmtted n any form or by any means wthout the pror permsson n wrtng of the publsher nor be ssued to the publc or crculated n any form other than that n whch t s publshed. Requests for permsson to reproduce any artcle or part of the Workng Paper should be sent to the edtor at the above address. Ncholas Oulton, submtted 2004 ISBN

3 1. Introducton To analyse the consequences of the changng economc structure of the UK, we need a set of statstcs broken down by ndustry that are consstent wth the whole economy measures avalable from the natonal accounts. For each ndustry, we need at a mnmum measures of output and of nputs. The theory of growth accountng then provdes a framework n whch the contrbuton of each ndustry to the natonal economy can be measured and assessed. Growth accountng also enables us to see how these contrbutons are evolvng over tme. The poneerng growth accountng studes were at the whole economy level. Whole economy studes can be llumnatng, but n practce polcy makers frequently want to know what s gong on at the sectoral and ndustry level. Through the work of Domar (1961), Hulten (1978) and Jorgenson et al (1987), the growth framework has been extended to the ndustry level. Best practce n ths area s the Jorgensonan growth accountng framework, now enshrned n two OECD manuals (OECD (2001a) and (2001b)). Growth accountng analyses have been very nfluental n the ongong debate about the US productvty acceleraton of the 1990s (Olner and Schel (2000); Jorgenson and Stroh (2000a) and (2000b)). They have also been employed to nform the UK polcy debate (O Mahony (1999)) and to analyse the falure of Europe or the UK to experence a smlar productvty acceleraton (Coleccha and Schreyer (2002); van Ark et al (2002); O Mahony and van Ark (2003); Basu et al (2003); European Commsson (2003)). To some, growth accountng, based as t s on the assumpton of perfect competton, may seem smplstc. But f one lacks the statstcal tools to do growth accountng then t s unlkely that one wll be able to carry out supposedly more sophstcated analyses ether. Also, the statstcs requred for growth accountng can be used as a bass for testng more complex hypotheses gong beyond those requred for growth accountng tself (as eg s done by Basu et al (2003)). Another advantage of the growth accountng framework s that mplementng t mposes an mportant dscplne: more tests are mposed on the consstency and coherence of the natonal accounts than s the case at present. Outlne of the paper To mplement the growth accountng framework at the ndustry level, we need for each ndustry measures of both gross output and value added on the one hand and of all the nputs captal, labour and ntermedate on the other. It turns out that at the moment ths s

4 qute a dffcult task usng offcal UK data. The frst dffculty arses from the way n whch the growth of output s measured by the ONS at the ndustry level, whch s nconsstent conceptually wth the way that the growth of GDP s measured. Ths dffculty, whch s mportant even f we do not seek to do a full growth accountng analyss, s dscussed n Secton 2. Secton 3 sets out the growth accountng framework n an nformal manner, showng t to be an extenson of, not a substtute for, the natonal accounts as they currently exst; Annex A provdes a more formal treatment. Secton 4 s a bref dscusson of the concept of sustanablty. It shows how one partcular measure of sustanablty, whch I call Wetzman s Net Domestc Product, can be calculated wthn the same framework. Secton 5 sets out the barrers whch currently confront a researcher seekng to mplement the growth accountng framework. These can be dvded nto mundane dffcultes whch could be fxed relatvely easly, and deeper conceptual and measurement ssues. The latter centre round (1) the measurement of output and prces n the prvate servces ndustres (consdered n more detal n Annex B); (2) the measurement of captal servces; and (3) ssues surroundng nformaton and communcatons technology (ICT). Lest the reader feel overwhelmed by the dffcultes, Secton 6 consders how the US statstcal agences tackle parallel problems and shows how they have solved some (though not all) of them. Secton 7 draws together the conclusons. 2. GDP and Industry Output: Ensurng Consstency It s well known that GDP at current prces can be measured n three ways: from ncome, output or expendture. In prncple, the three estmates must yeld the same answer, n the absence of errors and omssons. It s not so often apprecated that the growth of GDP n real terms can be measured n several ways too and that the dfferent ways must also be consstent n prncple. For example, the growth of aggregate output (GDP) can be thought of as an average of the growth rates of the ndustres that make up the natonal economy. Equally, t can be thought of as an average of the growth rates of the dfferent forms of expendture on fnal output. The growth of the natonal economy can also be thought of as an average of the growth rates of the regons, and the latter n turn can be thought of as averages of the growth of regonal output or of regonal expendture. 2

5 In nomnal terms GDP s ether a sum of value added n each ndustry or t s a sum of fnal expendtures net of mports (fnal outputs) and these two sums are equal n prncple. We can show that ths equalty must hold by defnton of value added. 1 But n addton to the level of GDP n current prces, we are also nterested n the growth rate of the volume of GDP (real GDP). We obvously want the growth rate of real GDP to be the same, whether we use the output or the expendture approach. It would be very strange f expendture-sde GDP n current prces was always equal to output-sde GDP, but the growth rate of the two measures dffered. If ths was the case, GDP would not be a very useful concept. It can readly be shown that (n the absence of errors, omssons or nconsstences n the underlyng statstcs) the two measures of growth wll yeld the same answer. But for ths to be the case, we must measure real value added n each ndustry by double deflaton. 2 Double deflaton means that an ndustry s gross output s to be deflated by the prce of ts output, whle each nput, whether domestc or mported, s to be deflated by ts own prce ndex. For example, f n addton to labour and captal, flour s requred to make bread, then to measure real value added n bread-makng, we must deflate the value of output by the prce of bread and the value of the purchased nput (flour) by the prce of flour. For any ndustry, the defnton of value added n current basc prces s: Value added = Gross output mnus Intermedate nput (domestc and mported) = Gross operatng surplus (profts) + Wage bll + Taxes on producton The sum of value added across ndustres s GDP at current basc prces. There are two ways n whch real GDP may be measured, from the output or from the expendture sdes. From the output sde, an ndex of GDP growth (GDP(O)) s: n n Growth of GDP(O) = wvˆ, w = 1 = 1 = 1 1 For the moment we are neglectng the dfference between basc prces, whch are normally used to measure GDP(O), and market prces, normally used to measure GDP(E). 2 Ths fact s well known to natonal ncome statstcans: see Commsson of the European Communtes Eurostat et al. (1993). 3

6 Here w s the share of nomnal value added of ndustry n current prce GDP, V s real value added n the th ndustry, a hat denotes a growth rate, and there are n ndustres. Second, from the expendture sde (GDP(E)): n n Growth of GDP(E) = seˆ, s = 1 = 1 = 1 where E s fnal expendture on the products of ndustry and s s the share of fnal expendture on n current prce GDP. We can readly show that these two measures of GDP growth are equal, n the absence of errors or omssons n the statstcs. 3 But note that equalty s only guaranteed n prncple f the growth of real value added s measured by double deflaton: see Annex A. In practce, of course the two estmates of growth wll dffer. The reconclaton between them s a complcated process. Frst of all, there s the reconclaton between levels n current prces. In addton to the quarterly balancng process, there s a qute separate annual balancng process to reconcle the output and expendture approaches, usng the nput-output supply and use tables. 4 Next, successve estmates of levels have to be converted nto estmates of growth rates. The ONS takes the vew that for measurng growth on an annual bass the expendture sde estmate s the most relable. For quarterly data, the ONS consders that movements wthn the year are best measured by GDP(O), but growth over the whole year s constraned to match the expendture measure. Concepts, Sources and Methods states: In the UK economc accounts, the expendture approach s used to provde current prce and volume measures of GDP. [paragraph ]. The 1998 Blue Book (Offce for Natonal Statstcs (1998b), pages ) spells ths out n more detal: Constant prce gross value added provdes the lead ndcator of economc change n the short term. However n the long term, constant prce gross value added s requred to follow reasonably closely the path ndcated by the constant prce expendture measure of GDP. To acheve ths, specal addtonal balancng adjustments (or coherence factors) are sometmes requred to be ncluded wthn gross value added at constant basc prces, as explaned below. These nstances can occur partcularly n perods of rapd change n economc actvty.... An examnaton of the constant prce gross value added and 3 Ths gnores the dfference between market prces at whch expendture s usually measured and basc prces at whch output s usually measured. But the argument can easly be extended to encompass ths pont. 4 At the moment ths s done only n current prces, though eventually t may be done n constant prces too (Ahmad (1999); Tuke and Aldn (2004)). 4

7 expendture measures of GDP suggested that the growth n the measure of gross value added was over-stated by around half a per cent between 1988 and Ths makes clear that, eventually, the annual constant prce output sde estmates are constraned to follow the expendture ones. In the publshed fgures, there s no dscrepancy between the two annual estmates (unlke n the US Natonal Income and Product Accounts: see below), but ths s because the output sde estmate s adjusted to conform to the expendture sde one. The reason for preferrng the expendture sde estmate s threefold: 1. The output of some ndustres s less well measured than others, ether because of conceptual dffcultes or because less statstcal effort s devoted to them. There s as yet no counterpart n the prvate servce ndustres to the Producer Prce Index (PPI) programme that covers the producton sector. Much of the hard-to-measure part of the economy s engaged n producng ntermedate products (eg busness servces or wholesale bankng) and these actvtes largely drop out of GDP measured from the expendture sde. 5 See Annex B for more on ths. 2. More detal s avalable on the expendture sde and the prce ndces are more relable. The Retal Prces Index programme s far more extensve than the PPI programme. 3. In practce the ONS does not use double deflaton to estmate real value added (except n electrcty supply and agrculture); nstead t uses real gross output as a proxy for real value added: see Offce for Natonal Statstcs (1998), chapters 11 and 13, and Sharp (1998). 6 So even n the absence of errors and omssons, the output and expendture sde estmates would not be equal and the output sde estmate would be wrong, unless n each ndustry real gross output and real ntermedate nput were growng at the same rate, whch s clearly unlkely. The dscrepancy between the output and expendture sde estmates s removed by adjustng output growth n the prvate servce ndustres; output n the producton sector (about a thrd of the economy) and n the government sector (about a ffth) s not adjusted. 5 6 Not completely, snce some enter nto nternatonal trade. In most cases, real gross output s measured by sales deflated by an approprate prce ndex. 5

8 Even though the growth of the producton sector s ncorrect as a measure of ts contrbuton to GDP growth (snce the Index of Producton uses sngle deflaton), t s left unadjusted. 7 So the whole burden of errors and omssons, together wth the error caused by the use of sngle rather than double deflaton, s thrown onto the estmates of growth n the prvate servces ndustres. The sze of the adjustment needed to make GDP(O) conform to GDP(E) s not publshed. It s clearly very unfortunate that, at a tme when the prvate servce ndustres have been growng n mportance, our understandng of ther performance should be dstorted n ths way. In fact, the ONS has the basc means avalable to mplement double deflaton, at least for recent years, usng the Input-Output Supply and Use Tables, whch are avalable on a fully consstent bass from 1992 onwards. These are n current prce terms and play an essental role n the current prce balancng process. The ONS has also had a programme, called the Constant Prce Input Output (KPIO) programme, whch has been nvestgatng balancng n real terms, n effect double deflaton (Ahmad (1999); Powell and Swatch (2002)). But so far double deflaton has not been ncorporated nto the natonal accounts process. The Frst Report of ths Revew (Allsopp (2003), paragraph 6.10) recommended use of the output approach for estmatng regonal GDP. Adopton of double deflaton would be consstent wth ths. 3. The Growth Accountng Framework 8 The growth accountng framework s now more than ffty years old. Though t was frst formalsed by Solow (1957), the earlest emprcal applcatons preceded hs semnal paper (Hulten (2000)). Subsequently, the framework has been deepened by Jorgenson and hs varous collaborators (eg Domar (1961), Jorgenson and Grlches (1967), Hulten (1978), Jorgenson, Gollop and Fraumen (1987), and Jorgenson (1989). Ths approach has now been codfed n two OECD manuals (OECD (2001a) and (2001b)). Annex A sets out the approach more formally. Here we gve an nformal treatment. 7 The Revew of Short Term Output Indcators suggested that the IoP too should bear a share of the adjustment (ONS (2000)). 8 See Annex A for a more formal dervaton. 6

9 3.1 The aggregate framework The startng pont s the famlar aggregate growth accountng equaton: Growth of GDP = [Captal s share tmes growth of captal nput] plus [Labour s share tmes growth of labour nput] plus growth of TFP (1) where GDP, captal and labour are all chan ndces of ther respectve components. Thus labour nput s a weghted average of the growth rates of the dfferent types of labour, where the weghts are the shares of each type n the aggregate wage bll; the shares are equal to the elastctes of output wth respect to the nputs under the assumpton of perfect competton. Captal nput s defned as aggregate captal servces, not captal stock. The treatment of captal s therefore exactly analogous to the treatment of labour, where each type of labour nput s measured by the flow of labour servces, e hours worked. 9 Labour nput grows ether f hours worked ncrease or f the qualty of labour ncreases; the latter occurs f the composton of the labour force shfts towards better pad workers (Burrel-Llombart and Jones (2004)). TFP growth s calculated as the resdual. Ths equaton can be rearranged n per hour worked terms as where Growth of GDP per hour worked = Captal deepenng plus Labour qualty contrbuton plus TFP growth (2) Captal deepenng = [Captal s share tmes growth of captal nput per hour worked] Labour qualty contrbuton = [labour s share tmes growth of labour nput per hour worked] 9 The mportant dstncton between captal servces and captal stock was ntroduced n Jorgenson and Grlches (1967) and set out theoretcally n Jorgenson (1989). For further dscusson, see Dewert and Lawrence (2000) and Oulton and Srnvasan (2003a). 7

10 Buldng up from the ndustry level The next step s to buld up the aggregate relatonshp of equatons (1) or (2) from correspondng relatonshps at the ndustry level. Suppose we have data on the prce and quantty of gross output and the prces and quanttes of nputs and outputs for a set of ndustres coverng the whole economy. These data can be used to measure the growth of TFP n each ndustry. For the th ndustry: Growth of TFP n ndustry = Growth of gross output of mnus cost-share-weghted growth of captal servces, labour and ntermedate nput (3) We can also calculate TFP growth usng prces rather than quanttes, the so-called dual approach: Growth of TFP n ndustry = mnus [Growth of prce of gross output of mnus cost-share-weghted growth of prces of captal servces, labour and ntermedate nput] (4) Provded that the accountng system s consstent, the dual approach of (4) must yeld exactly the same answer as the prmal one usng quanttes, equaton (3). 10 The crucal lnk between the ndustry TFP rates and the aggregate TFP growth rate s provded by the concept of Domar aggregaton (Domar (1961)): Aggregate TFP growth rate = Domar-weghted sum of ndustry TFP growth rates (5) where the Domar weght for ndustry s Nomnal gross output of ndustry Nomnal GDP 10 At the aggregate level there s a dual analogue to equaton (1) as well. 8

11 Note that the Domar weghts sum to more than 1; ths reflects the fact that each ndustry makes a double contrbuton to aggregate TFP, once n ts own rght and once through reducng the costs of ndustres that buy from t. Equaton (5) can be shown to hold exactly f any gven nput (eg unversty-educated female workers aged 30-34) s pad the same wage n all ndustres. If ths s not the case, then the equaton for aggregate TFP growth also contans terms reflectng the reallocaton of captal and labour towards or away from hgher value uses (Jorgenson et al (1987)). But the Domar-weghted sum of equaton (5) can stll be regarded as the best measure of underlyng productvty growth at the aggregate level. Equaton (5) enables us to trace the sources of aggregate TFP growth rate to ts ndustres of orgn. Changes n aggregate TFP can be assgned ether to changes n the underlyng ndustry rates or to structural change (changes n the Domar weghts). 4. Sustanablty Sustanablty s a contentous concept that can no doubt be defned n more than one way. But the central noton behnd the concept s the level of consumpton that can be mantaned ndefntely, wthout runnng down or usng up stocks of resources, whether natural or manmade. Ths noton has a long hstory n economcs. It s related to the concept of permanent ncome famlar to macroeconomsts and to the concept of real ncome suggested by Hcks (1940). The modern formulaton s due to Wetzman (1976), elaborated n Wetzman (1997) and (2003). It turns out that the statstcal framework requred for the analyss of productvty, namely growth accountng, can also be an mportant buldng block for the analyss of sustanablty. Wetzman s net domestc product, denoted by WNDP, s defned as net domestc product (NDP), 11 measured n consumpton unts. Alternatvely, t s current prce NDP deflated by the prce ndex for consumpton. In symbols, WNDP = [ PC + P K ]/ P t Ct t = 1 t t Ct = C + p K n t = 1 t t n (6) 11 Net natonal product s Wetzman s term, though net natonal ncome mght be more approprate. I use the term net domestc product to retan the lnk wth the natonal accounts and because I am neglectng net ncome from abroad. See Sefton and Weale (1996) for estmates of WNDP for the Unted Kngdom. 9

12 where C t s real consumpton; equvalently real net nvestment n the th stock; goods; P t s the nomnal prce of the th type of captal; and type of captal n terms of consumpton goods [ = P / P ]. Kt s the growth of the th captal stock ( = 1,,n) or t P Ct s the nomnal prce of consumpton Ct p t s the relatve prce of the th Wetzman shows that WNDP can be consdered a monetary measure of socal welfare. The ntuton behnd WNDP s that only consumpton matters for welfare. So the current level of consumpton must obvously be part of the measure. In addton, net nvestment ncreases future consumpton. The second term on the rght hand sde n equaton (6) above can be nterpreted as the present value of the future stream of consumpton that s generated by addng to captal stocks. Alternatvely, t can be nterpreted as the addtonal amount of consumpton (over and above actual current consumpton) whch could be enjoyed on a permanent bass, f captal stocks were just mantaned rather than augmented. Wetzman (1976) gave a more formal justfcaton. He showed that WNDP s proportonal to the yeld on wealth (assumng perfect competton and no externaltes). That s, t s equvalent to permanent ncome. Consder an economy whch behaves as f t s governed by a socal planner who seeks to maxmse wealth, defned as the present value of the stream of real consumpton, subject to technologcal constrants and to gven ntal captal stocks. Along the optmal path, Wetzman showed that WNDP s the yeld on wealth, e t s equvalent to permanent ncome. WNDP can therefore be consdered a cardnal measure of welfare. Ths s mportant because WNDP s drectly observable: only current prces and quanttes enter nto WNDP, whle measurng wealth drectly would requre forecastng an nfnte stream of future consumpton. It mght seem at frst sght that measurng WNDP s easer than measurng GDP, snce the only prce we need to know s the prce of consumpton. But ths would be ncorrect. WNDP requres us to measure captal stocks and to do ths properly, we need to know the prces of nvestment goods (correctly adjusted for qualty) and also deprecaton rates. We could estmate WNDP at the aggregate level only. But just as n practce polcy-makers and analysts are nterested n tracng the sources of GDP growth, so t s lkely that they would want to do the same for WNDP, f t became customary to calculate t. It can be shown that a smlar knd of growth accountng decomposton, requrng the same statstcal materals as for GDP, s possble for WNDP too (Oulton (2002b)). The extent to whch we can regard WNDP n practce as a satsfactory welfare measure depends on (a) the extent to whch actual market prces correspond to true socal values and 10

13 (b) whether or not all relevant captal stocks have been ncluded. Under (a), we may note that the measurement of GDP also reles on market prces reflectng socal values. Under (b), apart from land envronmental assets are not currently ncluded n the SNA. 12 Nevertheless, Wetzman s analyss shows us the relatonshp between GDP, a measure of output derved from the natonal accounts as currently consttuted, and WNDP, a measure of sustanablty, and ndcates what would be requred statstcally n movng from one measure to the other. 13 Despte ts theoretcal mportance, Wetzman s NDP s not generally calculated by statstcal agences snce t s not formally part of the 1993 System of Natonal Accounts (SNA93), nor of the latter s European formulaton (ESA95). In the Unted Kngdom for example, t s possble to derve net domestc product at 2000 basc prces from publshed seres. 14 And the ONS publshes net natonal dsposable ncome at 2000 market prces. Ths concept equals GDP at 2000 market prces adjusted for fxed captal consumpton, net ncome from abroad, and for gans or losses on the terms of trade. But n both these cases, nvestment, mports and government expendture are valued n ther own base year prces, not n consumpton unts. However, whether or not the ONS tself chose to publsh estmates of WNDP, mplementaton of the growth accountng framework would enable anyone who wanted to to do so. 5. Barrers to Implementng the Growth Accountng Framework Any researcher currently seekng to mplement the growth accountng framework usng offcal data runs nto a number of dffcultes, rangng from the mundane to deeper conceptual and measurement ssues. These can be llustrated by reference to the Bank of England Industry Dataset (BEID) that dd attempt to mplement the framework: see Oulton and Srnvasan (2003b). The frst ssue s the number of ndustres that t s possble to dstngush, 34 n the case of the BEID. Of these, 18 fall n the producton sector (ncludng 11 n manufacturng), 3 n publc servces, but only 13 n prvate servces whch now accounts for half of GDP. It would have been possble to dstngush more ndustres n 12 ESA95 recognsed mneral ol exploraton as a type of gross nvestment but depleton of ol felds s not counted as a form of deprecaton. 13 See Crafts (2002) for a dscusson of some of the wder ssues, ncludng changes n lfe expectancy, nvolved n measurng real natonal ncome. 14 Ths s gross value added at 2000 basc prces [ABMM] mnus fxed captal consumpton at 2000 prces [YBFX]. 11

14 manufacturng, whch accounts for less than a ffth of GDP, but not n prvate servces. Ths statstcal mbalance between the producton and servces sectors was noted n the Frst Report of ths Revew (Allsopp (2003), chapter 9). 5.1 Mundane dffcultes 1. For analytcal purposes we need long runs of data, eg years. But these are hard to put together n practce because The SIC s changed at roughly 10 year ntervals and earler seres are not always revsed to the new bass. For example, for the Bank of England Industry Dataset (BEID), whch covered the perod , t was necessary to brdge across the 1968, 1980 and 1992 SICs. Some basc offcal seres do not go back very far. For example, the offcal seres for current prce value added n communcatons (ONS code: QTPP) goes back only to 1985, whle that for constructon (ONS code: EWSX) goes back only to The offcal seres for whole economy total weekly hours worked (ONS code: YBUS) goes back only to 1992Q2. 2. Some basc economc seres are not produced by the ONS and have to be constructed out of a varety of sources. For example, hours worked s not publshed at the ndustry level. So for the BEID t was put together from unpublshed data on employees by ndustry (from the AES/ABI surveys), grossed up for self-employment (from the Labour Force Survey), and multpled by weekly hours worked per worker (from the New Earnngs Survey). 3. There are nconsstences between the levels of aggregaton at whch dfferent seres are publshed. For example, nomnal data on sales and purchases are publshed n the nputoutput tables (also n the supply and use tables) for nowadays 123 ndustres. But the correspondng prce ndces are not publshed at ths level, though prce ndces are n fact avalable at a much more detaled level. 4. The asset breakdown for the ONS s seres for nvestment n fxed assets s poor. Only four types of asset are dstngushed: buldngs, other machnery and equpment, vehcles, and ntangbles. Software nvestment s ncluded but, bafflngly, s dvded half and half between ntangbles and other machnery and equpment. A much more detaled breakdown s avalable from the supply and use tables, but on a consstent bass only from 12

15 1992 and only n current prces. Lack of asset detal would not matter much f asset prces all rose at about the same rate, but we know ths s far from beng the case (see below on ICT). 5.2 Deeper ssues Output and prces n the servce sector It s wdely beleved that measurng output and prces n servces s harder than n goods. Ths ssue s dscussed n detal n Annex B. The man problem s not the nherent dffculty of measurng servce output (though for some ndustres there are genune conceptual problems) but the absence of true prce ndces for an mportant part of servce output. There s no equvalent for the servce sector of the PPI programme for the producton sector, though Corporate Servces Prce Indces are under development. When true prce ndces are absent, real output has to be measured ether by deflatng nomnal gross output by an napproprate prce ndex such as RPIX (called rough deflaton n Annex B) or by usng an nput ndcator such as employment as a proxy for output (usually adjusted by a guesstmate of productvty growth derved from another ndustry group n whch productvty growth s easer to measure). Use of RPIX to deflate an ndustry s output mples that labour productvty s assumed to be growng at the same rate as n the sectors to whch RPIX properly apples (consumer goods and servces). Whether nput ndcators or rough deflaton s employed, ether way the productvty growth rate s beng assumed, not measured. How large then s the proporton of GDP where productvty growth rate s assumed, not measured? For the publc sector, ths amounted to about 18% of GDP pror to the changes ntroduced n the 1998 Blue Book (the partal adopton of ESA95). For 1986 onwards, ths fgure s now reduced to about 9% of GDP, the remander beng covered by output ndcators. (Output measurement n the publc sector s to be covered by a separate nqury (the Atknson Revew)). For prvate servces, the proporton where productvty growth s assumed not measured comes almost entrely from sectons J (Fnancal ntermedaton) and K (Real estate, rentng and busness actvtes) of the 1992 SIC and amounts to about 13% of GDP. The contrbuton of an ndustry to GDP growth s the growth of ndustry output multpled by the share of that ndustry n nomnal GDP. For fnance and nsurance, even these shares are problematc. The contrbuton of the fnancal sector to GDP growth s about half what t otherwse would be, because of a negatve tem called the Adjustment for fnancal servces. Ths arses because profts n the rest of the economy are measured gross of 13

16 nterest payments to fnancal frms, whle the profts of fnancal frms nclude ther nterest recepts. To avod double countng, the Adjustment for fnancal servces s requred. Under ESA95, profts n the non-fnancal sector are reduced by the amount that frms pay for the servces of fnancal ntermedaton, so the non-fnancal sector s weght n the economy wll be somewhat lower and the fnancal sector s weght wll be consderably hgher (Begg et al (1996)). However, ths part of ESA95 has not as yet been mplemented n the UK. In nsurance, the weght s measured by premums net of clams as a proporton of GDP and ths can be negatve. It was n fact negatve for non-lfe nsurance n 1990 and ths meant that whle 1990 weghts were beng used, any growth n the non-lfe nsurance ndustry had the effect of reducng GDP. Captal In most countres, ncludng the UK, captal stocks are estmated by the Perpetual Inventory Model (PIM). That s to say, for each type of asset the stock s estmated by cumulatng flows of gross nvestment, then subtractng retrements at the end of what s beleved to be the servce lfe, and fnally (n the case of so-called net stock estmates) allowng for deprecaton on older but stll survvng assets. Emprcally, estmates of captal stocks are bedevlled by two major areas of uncertanty. The frst relates to the servce lves of assets. Second, there s the choce of the approprate pattern of deprecaton: should we use for example geometrc, hyperbolc, lght bulb (also called one-hoss shay ), or straght-lne (the ONS uses the last of these)? 15 Lttle s known about the servce lfe of assets n the Unted Kngdom. Tll 1983, the offcal estmates of the captal stock were based on the work of Redfern (1955) and Dean (1964); see also Grffn (1976). Ther estmates of servces lves were n turn based on the lfe lengths used by the Inland Revenue for tax purposes, from a perod before the tax system encouraged busness frms to deprecate assets more rapdly n ther accounts than would be justfed by true economc lves (Inland Revenue (1953)). In 1983, the Central Statstcal Offce (the predecessor of the ONS) revsed the servce lves downwards, ctng (unpublshed) dscussons wth manufacturers as ts authorty (Central Statstcal Offce (1985), page 201). 15 Estmates of captal stocks are needed for balance sheet analyss but for measurng productvty we need estmates of captal servces: the dstncton between these two concepts of captal s extensvely dscussed n OECD (2001b) and n Oulton and Srnvasan (2003a). At the moment, the ONS publshes only stock estmates as offcal statstcs. But the emprcal requrements for estmatng stocks and servces are very smlar, so I do not labour the dstncton here. 14

17 Followng a report commssoned from the Natonal Insttute of Economc and Socal Research (Mayes and Young (1994)), ths reducton was reversed. But at the same tme two other changes were ntroduced. Frst, a new category of asset, numercally-controlled machnery, was ntroduced nto the ONS s Perpetual Inventory Model (PIM) of the captal stock. Ths type of asset s assumed to have only 40% of the servce lfe of other types of plant and machnery (Vaze et al (2003)). Second, some plant and machnery s assumed to be scrapped prematurely; the rate of scrappng s assumed to be related to the corporate nsolvency rate, whch has been on a rsng trend snce the 1970s. Clearly then the emprcal evdence for servce lves n the Unted Kngdom s weak. 16 Ths judgment s confrmed by the OECD. In ts captal stock manual (OECD (2001b), Appendx 3) t lsts four countres (not ncludng the Unted Kngdom) for whch servce lves appear to be based on nformaton that s generally more relable than s usually avalable for other countres : the Unted States, Canada, the Czech Republc, and the Netherlands. It s noteworthy that n each of these countres servce lves are lower than assumed n the Unted Kngdom for both buldngs and plant and machnery (at least before the effects of premature scrappng are consdered). The emprcal evdence for deprecaton rates n the UK s even weaker that that for servce lves. The most extensve research on deprecaton has been done for the Unted States, though even here the evdence s patchy and much of t s out of date. 17 Consequently, most ndependent researchers apply the US rates to the UK, possbly adjustng for assumed dfferences n servce lves between the two countres. No nternatonal consensus has yet been reached on the approprate assumpton to make about deprecaton (OECD (2001b)). In practce, a varety of approaches has been used. In the Unted States, the Bureau of Labor Statstcs (BLS) produces estmates of the productve captal stock, e aggregate captal servces, that assume a hyperbolc pattern of decay rates, argung that these are more realstc than geometrc decay. The Australan Bureau of Statstcs follows a smlar approach (Australan Bureau of Statstcs (2001)). But ths pattern s not based on any strong emprcal evdence. Statstcs Canada on the other hand uses geometrc decay. The BEA does not estmate captal servces but does produce wealth measures of the captal stock usng geometrc deprecaton (Fraumen (1997); Herman (2000)). Academc researchers have tended to assume geometrc deprecaton, eg Jorgenson 16 Knowledge may be mproved f the results of the ONS s captal stock survey are publshed (West and Clfton-Fearnsde (1999)). However, the survey was manly confned to manufacturng and has now been dscontnued. 17 The U.S. evdence s revewed n Fraumen (1997); see also Oulton and Srnvasan (2003a). 15

18 and hs varous collaborators n numerous studes (eg Jorgenson et al (1987); Jorgenson and Stroh (2000a) and (2000b)). By contrast the ONS n common wth many other natonal statstcal agences employs straght-lne deprecaton n ther net stock estmates. Recently, the ONS has revsed the PIM so that the underlyng data are now consstent wth the natonal accounts and there are also some mprovements n the model (Vaze et al (2003)). An expermental ndex of captal servces has also been publshed (Vaze (2003)). But sgnfcant weaknesses reman. Frst, as just noted, the estmates are not based on frm emprcal evdence about servce lves or deprecaton rates. Second, the underlyng nvestment data are only dsaggregated nto a small number of asset types, currently nne: 1. Industral and commercal buldngs 2. Dwellngs 3. Plant and machnery (excludng tems 4 and 5) 4. Computer- and numercally-controlled plant 5. Computers 6. Vehcles 7. Artstc orgnals 8. Mneral exploraton 9. Software By contrast, the US captal stock estmates are based on more than 50 asset types (see below). Lack of dsaggregaton has a large mpact on the estmates when the prces of assets are changng rapdly relatve to the average (Oulton and Srnvasan (2003a)). Ths s the case of course wth computers, so separatng them out s a step n the rght drecton. But the effect of dong so s greatly reduced n the revsed UK PIM snce the same deflator s employed for computers as for plant and machnery as a whole. Computers are not the only example of assets wth rapdly changng relatve prces. Others are telecommuncatons equpment and numerous other h-tech products where semconductors are mportant components. ICT Whatever one thnks about the New Economy, the measurement of ICT outputs and nputs s lkely to reman a crucal ssue for many years. It seems lkely that there are dfferences 16

19 between Europe and the US n the way that the same methodology, the 1993 SNA, s beng appled and even dfferences between European countres (Lequller (2001)). However, there s no doubt that at the moment the US statstcal agences are consderably n advance of the ONS n the measurement of ICT, though to be far the rest of the EU generally lags as far behnd. It has been known for some whle that the evoluton of offcal prce ndces for computers dffers wdely across countres. Most observers have judged the dfferences to be much too great to be plausble reflectons of genune dfferences between countres, gven that computers are wdely traded nternatonally (Schreyer (2002)). Outsde analysts nvarably use the US prce (adjusted for exchange rate changes), whchever country they are studyng. Other countres (ncludng the UK s) domestc PPIs for computers have up to now not carred convcton. The correct measurement of ICT s of crtcal mportance to the analyss of productvty growth, snce applyng US methods and prce ndces to ICT rases the growth rate of the UK s GDP and also rases substantally the contrbuton of captal to economc growth, whle reducng that of TFP; (Oulton (2001b) and (2002a); Oulton and Srnvasan (2003b); see also O Mahony and van Ark (2003)). One reason that the US s statstcally n advance of the UK s smply that t started earler. The offcally-sponsored study whch led to the ncorporaton of a hedonc prce ndex for computers nto the US NIPAs was Cole et al (1986). Ironcally, ths was preceded by a poneerng study by Stoneman (1976) for the UK. Stoneman s analyss suggested, agan on the bass of hedonc analyss, that the UK s offcal PPI for computers was overstatng prce change by some 10 percentage ponts per annum. But ths work was gnored and only recently has the ONS adopted hedonc methods for the PPI for computers (Ball and Allen (2003)). Computer prces are mportant but they are far from the only ssue. Concentraton on computer prces may have masked more pressng problems n software. In the US software nvestment s much larger than nvestment n all types of computer. So t seems we should be at least as concerned wth measurng the level and growth of software stocks as wth those of computer stocks. But here we have a puzzle. In the UK the offcal seres for current prce software nvestment shows t to be much smaller than computer nvestment. In earler work done at the Bank of England (Oulton (2001b)), I questoned ths, argung that the 1991 survey to whch up to now the offcal seres has been pegged yelds qute a dfferent concluson. The UK also appears to be out of lne wth other European countres. Lequller (2001) has compared France wth the US. He fnds that the rato of software nvestment to IT equpment 17

20 nvestment was about the same n the two countres n 1998 (hs page 25 and chart 5). He also fnds that the rato of software nvestment to ntermedate consumpton of IT servces s substantally lower n France than n the US (page 26-27). Ths rato s exceptonally hgh n the US, but equally hs chart 6 shows that t s exceptonally low n the UK. In fact, the reported UK rato s substantally lower than n France, the Netherlands, Italy and Germany. These consderatons led me to conclude that the level of software nvestment n the UK should be multpled by a factor of at least three, addng about 1% to the level of GDP at the end of the century. There s now an opportunty to settle ths ssue snce the results from the recent SERVCOM survey are now avalable, though not as yet fully publshed (Prestwood (2001)). An even bgger gap n our statstcal knowledge relates to software prces. There s no offcal prce ndex for software n the UK and so far as I am aware no programme to create one. (The US has a prce ndex for pre-packaged software, one thrd of the total, but as yet no prce ndces for the other two thrds, custom and own account software: see Parker and Grmm (2000)). In the UK natonal accounts, software nvestment s deflated by a prce ndex based on computer programmers wages, wth a guesstmated allowance for programmers productvty. We also have no data for the UK on the average servce lves of computers and software. A survey focused on ICT servce lves, perhaps concentrated on the ndustres whch are the heavy users (wholesale and retal trade; fnance and busness servces), would seem to be a relatvely straghtforward way of makng progress here. Other measurement ssues 1. Computer and software prces are not the only areas of concern even n ICT. It seems lkely that measurng prces for many types of h-tech equpment, such as telecoms and medcal equpment, requres a smlar approach. The matched models method commonly used by natonal statstcal agences, ncludng the ONS, may yeld systematcally based results, though the drecton of bas s not necessarly upwards. In the US, t s expected that hedonc methods wll be appled more and more wdely (Moulton (2001)). 2. Techncal progress n semconductors appears to le at the heart of techncal progress n computers, telecommuncatons equpment, and many other types of machnery. The qualtyadjusted prces of semconductors have been fallng extraordnarly rapdly; from , 18

21 the prce of ntegrated crcuts (mcroprocessors and memory chps) fell at between 36 and 56% per annum n the US; mcroprocessor prces fell at over 50% per annum (Azcorbe et al (2000; Azcorbe (2002)). Semconductors are an ntermedate good, so on the expendture sde they largely drop out, ther effect showng up n the prces of the goods n whch they are ncorporated. But they stll have some effect on GDP(E) unless the semconductors trade balance s zero. Even a small trade balance can have a sgnfcant effect on GDP because of the extreme sze of the prce change. On the output sde there s the potental for a much larger effect snce the UK semconductor ndustry s not neglgble. So gettng the semconductor prce rght s mportant. 3. The 1993 SNA, from whch ESA95 derves, treats expendture on software as a form of nvestment for the frst tme. But strangely, t contnues to regard R&D expendture as a current expense (ntermedate consumpton). Ths makes lttle sense. It would be hghly desrable for the ONS to produce estmates of R&D expendture for each ndustry on the same bass as other types of nvestment; at the moment publshed R&D spendng s only broken down nto 11 ndustry groups (Morgan (2003), Table 8). Then researchers could estmate R&D stocks and f desred ncorporate these nto the growth accountng framework. There s no need to wat untl some future revson of the SNA mandates that R&D be treated as an nvestment. 6. Economc Statstcs n the Unted States The feasblty of employng the growth accountng framework can be assessed n part by lookng at what other statstcal agences do. Here I concentrate on practce n the Unted States. The U.S. Bureau of Economc Analyss (BEA) publshes estmates of value added n current prces ( Gross Product Orgnatng or GDP by ndustry ) for 72 ndustres coverng the whole economy; the sum of value added across ndustres, plus a statstcal dscrepancy, equals GDP. The BEA also estmates chaned-volume ndces of the growth of each ndustry s real value added; these are calculated by double deflaton (Yuskavage (1996)). Lke the ONS, the BEA beleves the expendture estmate of GDP, both n current prces and 19

22 n chaned volume terms, to be more relable than the output one. 18 Yuskavage (1996) as follows: Its reasons were stated by The amount of detaled expendtures data that are avalable for weghtng the prce ndexes used n calculatng GDP [e, GDP(E)] s greater than that for gross outputs and ntermedate nputs used n calculatng GPO [e, GDP(O)], and lttle nformaton s collected annually on the composton of nputs or of non-manufacturng outputs. To make the two estmates of GDP agree, the BEA adds a statstcal dscrepancy to the output estmate. From ths has ranged between plus 0.4% and mnus 1.4% of current prce GDP(E), averagng about mnus 0.3%. In growth rate terms, there s also a statstcal dscrepancy whch averaged vrtually zero over the perod , though rangng from 0.9 to +1.2 percentage ponts per annum n ndvdual years (Yuskavage (1996), Table 2). The dscrepancy s beleved to be located n the prvate sector, not the government. Unlke the ONS, the BEA does not constran the prvate sector growth rates of output so that the aggregate of prvate sector and government growth equals that of GDP(E), but nstead s content to reconcle the two estmates of GDP growth by addng a statstcal dscrepancy. The BEA also publshes estmates of the captal stock held by broad sectors (U.S. Department of Commerce (1999); Herman (2000)). These are calculated by the perpetual nventory method. In the prvate sector some 28 types of equpment are separately dstngushed, plus 21 types of non-resdental, and seven types of resdental structures. For nearly all asset types deprecaton s geometrc wth the rates deduced from studes of second hand asset prces (Fraumen (1997); Oulton and Srnvasan (2003a)). The U.S. Bureau of Labor Statstcs (BLS) frst publshed estmates of TFP growth n 1983, coverng the perod (The BLS uses the term multfactor productvty (MFP), but the concept s the same as TFP). Currently t publshes TFP growth estmates for broad aggregates lke prvate busness and manufacturng and also at a more detaled level, eg for 20 2-dgt manufacturng ndustres and for dgt manufacturng ndustres. In addton, t publshes comparsons between US, German and French TPF growth. The more detaled estmates employ the so-called KLEMS approach: the growth of real gross output (not value added) s accounted for by the growth of captal (K), labour (L), energy (E), materals (M) and busness servces (S). Captal nput s measured by aggregate captal servces, not by the captal stock. The captal servces estmate s bult up from the same underlyng data as the BEA s stock estmates, though the BLS assumes that the servces of captal assets decay 18 Confusngly, t calls the output sde estmate the ncome estmate. 20

23 hyperbolcally, not geometrcally. Labour nput s measured by hours worked, adjusted for qualfcatons, age and sex. Inputs of energy, materals and servces are measured by deflated purchases, at a detaled level. 7. Conclusons: the Need for Joned-Up Statstcs Growth accountng provdes, I would argue, a very frutful way of lookng at the changng structure of the economy, and assessng the contrbuton of each sector or ndustry to the whole. It also provdes a very useful framework for the collecton of economc statstcs. Ths approach mght be called joned-up statstcs. The advantages of ths approach have long been recognsed n the US, where the BEA and the BLS routnely supply the knd of productvty statstcs for whch I am argung here. At the moment, the ONS does not provde ths servce to users, nor does t supply the raw materals that would enable outsders readly to create ther own statstcs. What of the future under current plans? The ONS does have a productvty programme, but so far t has concentrated on labour productvty and progress has been slow. The Corporate Servces Prces programme promses one day to mprove the measurement of prces and real output n servces but agan progress has been slow and the hardest sectors have yet to be covered. The ONS does have a programme for measurng real value added by double deflaton, the Constant Prce Input Output (KPIO) programme, but ths has as yet not been ncorporated nto the natonal accounts process. The Captal Stocks Survey mght have ncreased our knowledge of asset lves, but ths has been dscontnued wth only part of manufacturng covered. In my vew, the ONS ought to adopt the growth accountng approach as the framework for ther economc statstcs. Ths s just an extenson of the SNA whch les behnd the natonal accounts and s now well sanctoned by nternatonal organsatons, eg the OECD and the European Commsson. As a part of ths, the ONS should expand ts nput-output programme and ntegrate t wth ther ongong PPI and CSPI programmes, e reactvate ther KPIO programme. Ths would enable them n due course to produce estmates of double deflated value added. The latter would n turn ncrease the coherence of the natonal accounts and greatly assst n the analyss of productvty. 21