Deposited on: 27 March 2009

Size: px
Start display at page:

Download "Deposited on: 27 March 2009"

Transcription

1 Harrs, R. and L, Q.C. (2008) Exportng, R&D, and absorptve capacty n UK establshments. Oxford Economc Papers, 61 (1). pp ISSN Deposted on: 27 March 2009 Enlghten Research publcatons by members of the Unversty of Glasgow

2 Exportng, R&D and Absorptve Capacty n UK Establshments * by Rchard Harrs (Unversty of Glasgow) and Qan Cher L (Unversty of Strathclyde) * We wsh to express our thanks to comments receved from the edtor and two anonymous referees, whch much mproved the paper. However, the authors reman solely responsble for any remanng errors or defcences. We would also lke to thank UKTI for ther nvolvement n ths work. Fnally, ths work contans statstcal data from ONS whch s Crown copyrght and reproduced wth the permsson of the controller of HMSO and Queen's Prnter for Scotland. The use of the ONS statstcal data n ths work does not mply the endorsement of the ONS n relaton to the nterpretaton or analyss of the statstcal data. Ths work uses research datasets whch may not exactly reproduce Natonal Statstcs aggregates. Correspondng author. Address: Department of Economcs, Unversty of Strathclyde, Sr Wllam Duncan Buldng, 130 Rottenrow, Glasgow, G4 0GE,UK; Tel: +44 (0) ; Fax: +44 (0) ; Emal: Cher.L@strath.ac.uk 1

3 Exportng, R&D and Absorptve Capacty n UK Establshments Abstract Ths paper models the determnants of exportng (both n terms of export propensty and export ntensty), wth a partcular emphass on the mportance of absorptve capacty and the endogenous lnk between exportng and undertakng R&D. Based on a merged dataset of the 2001 Communty Innovaton Survey and the 2000 Annual Respondents Database for the UK, our results suggest that establshment sze plays a fundamental role n explanng exportng. Meanwhle, alongsde other factors, undertakng R&D actvtes and havng greater absorptve capacty (for scentfc knowledge, nternatonal co-operaton, and organsatonal structure) sgnfcantly reduce entry barrers nto export markets, havng controlled for self-selectvty nto exportng. Nevertheless, condtonal on entry nto nternatonal markets, only greater absorptve capacty (assocated wth scentfc knowledge) seems to further boost export performance n such markets, whereas spendng on R&D no longer has an mpact on exportng behavour once we have taken nto account ts endogenous nature. JEL codes: L25; O24; O32; R11 Keywords: exports; R&D; absorptve capacty; sample selecton I. Introducton Exportng s mportant to the UK economy; t accounted for 28.4% of UK GDP n 2006, whle the UK s the ffth largest exporter of manufactures n the world. Moreover, data from the UK Communty Innovaton Survey of 2001 and 2005 show that around 26% of all frms producng marketable output were nvolved n exportng (the fgure was around 45% for manufacturng frms). Thus, a better understandng of what determnes exportng actvtes s mportant for the UK economy. Recent lterature has tended to concentrate on the mcroeconomc approach to trade, renforcng the mportance of exportng for (natonal) economc growth (e.g. the varous studes by Bernard and assocates 1, and Meltz, 2003). Exportng tends to be concentrated among a (very) small number of frms who nevertheless are large and account for the preponderance of trade undertaken. Such frms have a greater probablty of survval; hgher growth rate; greater productvty; hgher captal-ntensty; they pay hgher wages; and employ better technology and more sklled workers (after controllng for other relevant covarates). To put thngs n context, Bernard and Jensen (2004a) show that ncreased export opportuntes are assocated wth both ntra- and nter- ndustry reallocatons whch account for 40% of TFP growth n the manufacturng sector; whle for the UK Harrs and L (forthcomng) confrm that exportng 1 For example, Bernard and Jensen (2004a), Bernard et. al. (2003) and Bernard et. al. (2007). 1

4 frms experence faster productvty growth than non-exporters and therefore contrbute more to overall productvty growth. In partcular, wthn productvty gans for exporters are relatvely large and new export entrants (through beng ether taken-over or merged, or as new start-ups) also contrbute about 38 per cent each of the overall growth n UK aggregate productvty durng Thus, hgher productvty levels as well as faster growth rates are assocated wth exporters, provdng an mportant reallocatve channel for explanng aggregate productvty growth. In ths paper we attempt to obtan a better understandng of the frm s behavour when facng ntense nternatonal competton, so as to shed lght on ths mportant export-productvty nexus 2. Hence ths study concentrates on what determnes who exports (and thus barrers to exportng) and how much s exported, and whch factors are most mportant n drvng such exportng actvtes. In partcular, we are nterested n the lnkage between exportng and R&D, and how any (causal) relatonshp between these varables s affected by ntroducng other varables (partcularly absorptve capacty ). Despte the mportance of ths area there are stll only a lmted number of mcro-based studes n the lterature, especally wth regard to UKbased emprcal analyses. Thus, the next secton summarses some recent lterature on the lnks between exportng, absorptve capacty and nnovaton actvtes (such as R&D spendng), whle also recognsng other factors that determne exportng. Secton III dscusses the data used, whch comprses establshment data from the 2001 Communty Innovaton Survey along wth the 2000 Annual Respondents Database for the UK. Ths s followed by estmatng a Heckman-type sampleselecton model of exportng n Secton IV. Fnally, the paper concludes wth a summary, a dscusson of the polcy mplcatons and some caveats of ths study. II. Lterature Revew Absorptve Capacty In ths sub-secton we defne what s meant by absorptve capacty, and ts role n determnng both exportng and R&D (how absorptve capacty s measured s taken up n the next secton). As wll become evdent, we argue that absorptve capacty s expected to help frms break down barrers to both operatng n nternatonal markets as well as undertakng R&D (t s also lkely to have an mpact on how much a frm exports and/or spends on R&D, subject to havng overcome entry barrers to dong ether or both actvtes). We also wll argue that absorptve 2 For recent surveys of the lterature on ths export-productvty lnkage, see López, 2005; Greenaway and Kneller, 2005, 2007; Wagner,

5 capacty can be treated as pre-determned, gven that t s based on pror accumulated abltes and competences emboded n the frm, and thus s not endogenous to current R&D and exportng. Absorptve capacty s defned here as the ablty to explot knowledge (obtaned both nternally and especally externally) that s emboded n ntangble assets, wth the latter beng recognsed as a key drver of enterprse performance. Knowledge and learnng can be expected to have a fundamental mpact on growth n that frms must apprehend, share, and assmlate new knowledge n order to compete and grow n markets n whch they have lttle or no prevous experence (Auto et. al., 2000). In a semnal paper, Cohen and Levnthal (1990) demonstrate that the ablty to explot external knowledge s a crtcal component of a frm s capabltes. They argue that:...the ablty to evaluate and utlse outsde knowledge s largely a functon of pror related knowledge. At the most elemental level, ths pror knowledge ncludes basc sklls or even a shared language but may also nclude knowledge of the most recent scentfc or technologcal developments n a gven feld. Thus, pror related knowledge confers an ablty to recognze the value of new nformaton, assmlate t, and apply t to commercal ends. These abltes collectvely consttute what we call a frm s absorptve capacty (p. 128). Gven these arguments, t s possble to conclude that when a frm nternatonalses, t must have suffcent resources and capabltes through absorbng new knowledge to overcome the ntal (sunk) costs of competng n nternatonal markets n order to organse for foregn competton, thus facng the dual challenge of overcomng rgdtes and takng on novel knowledge (Erksson and Johanson, 1997). Smlarly, when a frm engages n R&D, t has to apprehend, share, and assmlate new knowledge to overcome barrers to nnovaton (Aw et. al., 2007). Furthermore, Cohen and Levnthal argue that the development of absorptve capacty s hstoryor path- dependent (see also Davd, 1985, and Arthur, 1989). Ths results from the effectve assmlaton of new knowledge beng dependent on accumulated pror knowledge. For example, the possesson of related expertse permts a frm to assess more accurately the nature and commercal potental of technologcal advances and/or operaton n new markets. Ths n turn wll affect the ncentve to make further nvestments n developng capabltes n that doman. Further, where a frm has not nvested n a doman of expertse early on, t s lable to fnd t less attractve to nvest n subsequently even where t s a promsng feld because of the mpact on current output. The result s that frms may become locked nto nferor procedures, locked out of technologcal opportuntes and/or new markets, and exhbt hgh degrees of nerta wth respect to changes n ther external envronment. Ths noton that absorptve capacty s path- 3

6 dependent therefore leads to the assumpton that t can be treated as a pre-determned varable and t s not endogenous to current R&D spendng and exportng. The Export-R&D Nexus The lnkage between nnovaton actvtes (such as R&D) and exports has been charactersed by ncreasng nterdependence n the process of globalsaton, and s often regarded to be of paramount mportance to an economy: nnovaton s commonly taken as a proxy for productvty and growth, and exportng for compettveness of an ndustry/country. Export orentaton at the frm level has been extensvely nvestgated n the lterature, and varous emprcal studes have emphassed the role of technology and R&D as one of the major factors facltatng entry nto global markets and thereafter mantanng compettveness and boostng export performance (see Harrs and L, 2005, for further detals). For nstance, recent studes nclude Bleaney and Wakeln (2002), and Gourlay and Seaton (2004), for the UK; Baldwn and Gu (2004), for Canada; Basle (2001), for Italy; and fnally n comparatve studes, Roper and Love (2002), for both UK and German manufacturng frms. Stll evdence at ths mcro level does not seem to be conclusve, as nconsstent results have been found by Lefebvre et. al. (1998) and Sterlacchn (2001). Wth respect to the causalty ssue assocated wth the lnkage between R&D and exports, the early consensus n the lterature was that causalty runs from undertakng nnovaton actvtes to nternatonalsaton. Ths can be easly understood from the perspectve of product dfferentaton or nnovaton-led exports, n lne wth the predctons of both the more conventonal product-cycle models as well as the recently developed neo-technology models that have tended to domnate the well-establshed trade-nnovaton theores n the macroeconomcs lterature (e.g. Posner, 1961; Dollar, 1986; and Krugman, 1979). In short, product dfferentaton/nnovaton translates nto compettve advantages that allow the frm to compete n nternatonal markets. It can also be argued that causalty may go from exportng to nnovatveness,.e. there exsts a learnng-by-exportng effect. Ths reverse drecton of causaton s n accordance wth the theoretcal predctons of global economy models of endogenous nnovaton and growth, such as those n Romer (1990), Grossman and Helpman (1991), and Aghon and Howtt (1998). From a resource-based perspectve, beng exposed to a rcher source of knowledge/technology that s often not avalable n the home market, exportng frms could well take advantage of these dverse knowledge nputs and enhance ther competency base, and hence n ths sense, such learnng from global markets can foster ncreased R&D and nnovaton wthn frms. Ths learnng effect nduced by partcpaton n nternatonal markets s often not drectly measured 4

7 but consdered ndrectly through the lnk between nnovaton and productvty growth. The process of gong nternatonal s perceved as a sequence of stages n the frm s growth trajectory, whch nvolves substantal learnng (and nnovatng) through both nternal and external channels, so as to enhance ts competence base and mprove ts performance. Gven that causalty can run n both drectons, ths two-way lnkage between a frm s exportng and nnovatng actvtes (such as undertakng R&D) has also been tested and confrmed emprcally, partcularly n studes of frms operatng n emergng economes (e.g. Alvarez, 2001, for Chle), where the learnng effect s lkely to be more pronounced, from the perspectve of technology catchng-up or economc convergence (Ben-Davd and Loewy, 1998). Overall, the paucty of evdence on ths feedback relatonshp may be partly explaned by the lmtatons of the data as well as the econometrc methods avalable to explore ths causalty ssue. In general, we would a pror expect that undertakng R&D and exportng are endogenous; hence, we both test and allow for ths n our emprcal work n Secton IV. Other Determnants of Exportng Actvtes A number of other factors have been suggested n the lterature to exert an mpact on a frm s exportng behavour, and therefore moderate the way exportng and R&D actvtes affect (and nteract wth) each other. To begn wth, there s well-documented evdence on how the sze of frms affects the probablty of enterng foregn markets, as larger frms are expected to have more (technologcal) resources avalable to ntate an nternatonal expanson (e.g. Roberts and Tybout, 1997; Gourlay and Seaton, 2004; and Kneller and Psu, 2007). Nevertheless, condtonal on havng overcome entry barrers, the sze effect on export performance could become negatve - as frms grow larger (and presumably more productve), they mght have an ncentve to expand ther foregn-market penetraton through FDI (rather than exports), whch often consttutes an alternatve (and more attractve) strategy for nternatonal expanson (c.f. Head and Res, 2004; and Helpman et. al., 2004). Ths possbly explans why a non-lnear relatonshp between sze and export actvtes s frequently captured n emprcal studes where export propensty and ntensty are not estmated separately (e.g. Wagner, 1995; Bernard and Jensen, 1999; and Bleaney and Wakeln, 2002). In addton to ths sze effect, the sectoral context n whch a frm operates s also lkely to be mportant snce belongng to a specfc ndustry may condton the frm s strategy as well as performance to some degree (both n terms of nnovaton and nternatonalsaton actvtes). As ndustres are nether homogeneous n ther technologcal capacty nor exportng patterns, the sectoral effect (reflectng technologcal opportuntes and product cycle dfferences) s usually expected to be sgnfcant n condtonng the frm s export-nnovaton relatonshp (for nstance, 5

8 Hrsch and Bjaou, 1985; Hughes, 1986; and Gourlay and Seaton, 2004). Moreover, the role of some ndustral/spatal factors could also be expected to be mportant. Frstly, the mportance of geographc factors s captured n Overman et. al. s (2003) survey of the lterature on the economc geography of trade flows and the locaton of producton. If nformaton on foregn market opportuntes and costs s asymmetrc, then t s reasonable to expect frms to cluster wthn the same ndustry/regon so as to acheve nformaton sharng and therefore mnmse entry costs. Co-locaton may help mprove nformaton about foregn markets and tastes so as to provde better channels through whch frms dstrbute ther goods (Atken et. al., 1997). There are usually two dmensons to these agglomeraton effects a regonal effect and an ndustral effect. The former comprses the spatal concentraton of exporters (from varous ndustres) whereas the latter effect s where exportng frms from the same ndustry co-locate. Greenaway and Kneller (2004) provde emprcal evdence that shows the ndustral dmenson of agglomeraton would appear to be more mportant for the UK whle Bernard and Jensen (2004b) fnd t to be nsgnfcant n explanng the probablty of exportng n the US. The benefts brought about by the co-locaton of frms on the decson to export have also been documented n for nstance, Atken et. al. (1997) for Mexco. Lastly, n a smlar fashon, market concentraton s also expected to postvely mpact on exportng actvtes. A hgh level of concentraton of exporters wthn an ndustry may mprove the underlyng nfrastructure that s necessary to facltate access to nternatonal markets or to access nformaton on the demand characterstcs of foregn consumers. Therefore, we mght expect a hgher propensty for non-partcpants to go nternatonal n a market wth a hgher degree of concentraton of export actvty (see Greenaway and Kneller, forthcomng, for evdence from UK manufacturng). In summary, we expect absorptve capacty and R&D spendng, plus a number of other factors (such as frm sze and ndustral sector) to mpact on whether exportng takes place, and condtonal on ths, how much s sold abroad. We also expect that R&D spendng s lkely to be endogenous n any model estmated. III. The Data The ablty to undertake a mcro-level analyss of the determnants of exportng, wth partcular focus on ts relatonshp wth nnovatve actvtes, depends on the data avalable. There are 2 major mcro-based sources of data that are approprate, both of whch nclude establshment- 6

9 level data for the UK: () the Communty Innovaton Survey 2001 () 3 ; and () the Annual Respondents Database (ARD) 4. The dataset s a cross-sectonal survey of nnovaton coverng the perod, ncludng the characterstcs of the reportng unt surveyed (e.g. turnover, employment and, most mportantly, exports). The survey only acheved a 43% response rate and over-represents large establshments (wth only frms employng 10 or more ncluded), but the weghts avalable n can be used to ensure the sample obtaned s representatve of the populaton of all UK establshments. The dataset covers all sectors of the economy and can be lnked nto the ARD, snce the Inter Departmental Busness Regster (IDBR) reference numbers are common to both datasets. Thus ancllary nformaton (partcularly on ownershp and spatal characterstcs) avalable n the 2000 ARD has been added to the data for use n our subsequent analyss of what determnes exportng 5. Of the 8172 establshments covered n, t was possble to locate 7709 of these n the ARD at the reportng unt level; where necessary, plant level ARD nformaton (e.g. on captal stocks n manufacturng) was aggregated to reportng unt level to ensure comparablty wth 6. Table 1 sets out the lst of varables we use n ths study, along wth the data sources used. R&D spendng s defned here as ntramural R&D, acqured external R&D or acqured other external knowledge (such as lcences to use ntellectual property). Of partcular mportance s the absorptve capacty of the establshment. No drect nformaton on ths varable s avalable, but does contan nformaton on key elements of organsatonal, learnng and networkng processes that can be related to absorptve capacty,.e. external sources of knowledge or nformaton used n technologcal nnovaton actvtes 7 ; partnershps wth external bodes on nnovaton co-operaton 8 ; and the ntroducton of changes n organsatonal structure and human resource management (henceforth HRM) practces whch wll be related to nternal aspects of absorptve capacty 9. 3 The more up-to-date Communty Innovaton Survey 2005 (CIS4) s also avalable but does not contan nformaton on how much was sold abroad (only whether the establshment engaged n exportng). 4 For a detaled descrpton of the ARD, see Oulton (1997), Grffth (1999), and Harrs (2002, 2005). 5 The 2000 ARD data s used as the sample was drawn from the 2000 verson of the IDBR, and thus matches ARD data on establshments operatng n that year. 6 Non-matched observatons were mostly n those sectors not covered n the ARD (.e. fnancal servces). 7 See queston 12.1 n the questonnare (avalable at Table A1 lsts the 16 varables ncluded n CIS, and respondents were asked to rank how mportant each factor s (from 0 not used, to 4 hgh mportance). 8 See queston 13.2 n the questonnare and Table A1 for the 8 varables ncluded n CIS. Snce respondents were asked to ndcate whether cooperaton was wth organsatons that were local, natonal, European, US or n other countres, we could separately dentfy cooperaton at the natonal and nternatonal level. 9 See queston 17.1 n the questonnare and Table A1 for the 4 varables ncluded n CIS. 7

10 Table 1: Varable defntons used n CIS-ARD merged dataset for 2000 Varable Defntons Source Export Whether the establshment sold goods and servces outsde the UK (coded 1) or not n 2000 Export ntensty Establshment export sales dvded by total turnover n 2000 R&D Whether the establshment undertook any R&D as defned n the text (coded 1) or not n 2000 R&D contnuous Whether the establshment undertook R&D contnuously (coded 1) or not durng Sze Number of employees n the establshment, broken down nto 5 sze-bands,.e. 0-9, 10-19, 20-49, and 200+ Enterprse sze Number of employees n the enterprse ARD Age Age of establshment n years (manufacturng only) ARD Employment Current employment for establshment n 2000 ARD Captal Plant & machnery captal stock for establshment n 2000 (source: Harrs and Drnkwater, 2000, updated) ( m 1980 prces) ARD Labour productvty Establshment turnover per employee n 2000 Mult-plant Dummy coded 1 when establshment belongs to a multple-plant enterprse ARD >1 SIC multplant Dummy varable =1 f establshment belongs to a multple-plant enterprse operatng n more than 1 (5-dgt) ndustry ARD >1 regon multplant Dummy varable =1 f establshment belongs to multple-plant enterprse operatng n more than 1 UK regon ARD US-owned Dummy coded 1 f establshment s US-owned ARD Other foregn-owned Dummy coded 1 f establshment s other-owned ARD AC for external knowledge Absorptve capacty AC for natonal co-operaton (5 factors, see text AC for organsatonal structure & human resource management (HRM) for detals) AC for nternatonal co-operaton AC for scentfc knowledge Excessve perceved economc rsks Hgh costs of nnovaton Cost of fnance Barrers to Avalablty of fnance nnovaton a Organsatonal rgdtes wthn the enterprse (10 factors dentfed Lack of qualfed personnel n CIS) Lack of nformaton on technology Lack of nformaton on markets Impact of regulatons/standards Lack of customer responsveness Industry % of ndustry output (at 5-dgt SIC level) located n local authorty dstrct agglomeraton n whch establshment s located ARD Dversfcaton % of 5-dgt ndustres (from over 650) located n local authorty dstrct n whch establshment s located ARD Herfndahl Herfndahl ndex of ndustry concentraton (5-dgt level) ARD Densty Populaton densty n 2001 n local authorty dstrct n whch establshment s located CoP, 2001 Industry Dummy varable =1 f establshment located n partcular ndustry SIC (2- dgt) GO regons Dummy varable =1 f establshment located n partcular regon Greater South East Dummy varable =1 f establshment belongs to enterprse operatng n Greater South East regon ARD a Each dummy varable s coded 1 f the barrer s of medum-to-hgh mportance to the establshment. In order to extract core nformaton, a factor analyss (prncpal component) was undertaken usng the 36 relevant varables coverng the above dmensons of absorptve capacty (for detals see Table A1). Based on the Kaser crteron (Kaser, 1960), fve prncpal components were 8

11 retaned (wth egenvalues greater than 1), accountng for some 62% of the combned varance of these nput varables. In order to obtan a clearer pcture of the correlaton between those varables related to absorptve capacty and the fve factors extracted, the factor loadngs matrx was transformed usng the technque of varance-maxmsng orthogonal rotaton (whch maxmses the varablty of the "new" factor, whle mnmsng the varance around the new varable). As can be seen n Table A1, all 36 nput varables used to measure absorptve capacty are supported by the Kaser-Meyer-Olkn (hereafter KMO) measure of samplng adequacy most of the KMO values are above 90% and an overall KMO value of nearly 95% suggests a marvellous 10 contrbuton of the raw varables. Based on the correlatons between these 36 underlyng varables and the fve varmax-rotated common factors n Table A1 (each wth a mean of zero and standard devaton of 1), we were able to nterpret these factors as capturng the establshment s capabltes of explotng external sources of knowledge; networkng wth external bodes at the natonal level; mplementng new organsatonal structures and HRM strateges; buldng up partnershps wth other enterprses or nsttutons at the nternatonal level; and acqurng and absorbng codfed scentfc knowledge from research partners respectvely 11. Comparng these 5 dstnct dmensons of absorptve capacty, we could expect the absorptve capacty for scentfc knowledge to be partcularly mportant n ndcatng the technologcal opportuntes an establshment possesses, as ths noton of technologcal opportuntes was orgnally put forward to reflect the rchness of the scentfc knowledge base (Scherer, 1992). Moreover, as research grows ncreasngly expensve and rsky nowadays, ndustry has sought for specalst technology n academa or other government research nsttutes to complement or substtute ther n-house R&D efforts drawn on ts own resources. Varous hypotheses on the components of absorptve capacty have been put forward n the lterature (partcularly, n management studes), such as human captal, external network of knowledge and HRM practces as n Vndng (2006), and potental and realsed absorptve capacty as re-conceptualsed by Zahra and George (2002). Nevertheless, there seems to be an mbalance between the relatve abundance of varous defntons of absorptve capacty and a defcency of emprcal estmates of ths concept, wth R&D-related varables most commonly used as proxes (e.g. Cohen and Levnthal, 1990; Arora and Gambardella, 1990; Veugeler, 1997; Cassman and Veugelers, 2002; Belderbos et. al., 2004). However, gven the path-dependent nature of absorptve capacty, R&D fals to capture the realsaton and accumulaton of absorptve capacty, not to menton ts dstnct elements (Schmdt, 2005). Notably, whlst 10 Hstorcally, the followng labels are gven to dfferent ranges of KMO values: Marvellous, Mertorous, Mddlng, Medocre, Mserable, Unacceptable. 11 The correlatons wth the hghest values for each factor have been hghlghted (usng bold, talcsed values) n Table A1 to show why a partcular factor s nterpreted as representng a specfc aspect of absorptve capacty. 9

12 allowng R&D to be potentally endogenous, we treat the path-dependent absorptve capacty as predetermned n our emprcal models,.e. such capacty takes a (relatvely) long tme to buld. To our knowledge the approxmaton of absorptve capacty used n ths study provdes the most drect, and comprehensve set of emprcal measures avalable for the UK. Others have taken a dfferent approach wth regard to how the above varables used to measure external absorptve capacty should be classfed. For example, Dachs et. al. (2004) use the nformaton on sources of knowledge from supplers and customers to compute a varable that attempts to capture vertcal spllovers (of knowledge). We have chosen not to take a smlar approach. The pragmatc reason s that n our statstcal analyss (Secton IV) we fnd these spllover measures are nsgnfcant n the models determnng exportng and R&D, whereas our measures of absorptve capacty are found to be mportant determnants. In addton, the proporton of establshments that stated that such sources of knowledge had hgh mportance s relatve small (15.1% for vertcal spllovers; 3.5% for horzontal spllovers; 1.3% for nsttutonal spllovers; and 4.5% for publc spllovers). Taken together, over 90% of establshments have a zero value for spllovers; whereas the absorptve capacty measures are based on much more nformaton and span a greater range. Lastly, there s a hgh correlaton between these types of spllover measures and our measures of absorptve capacty; therefore t s clear that knowledge spllover effects wll be captured wthn the absorptve capacty measures we use n ths study. Indeed, by defnton absorptve capacty captures the ablty of frms to nternalse external knowledge spllovers. Most other varables ncluded n Table 1 are self-explanatory, although a clear dstncton needs to be made between the sze of the establshment (usually comprsng just one plant) and the sze of the enterprse (larger than the sze of the establshment for mult-plant frms). Moreover, ndustral agglomeraton s ncluded to take account of any Marshall-Romer external (ds)economes of scale (Davd and Rosenbloom, 1990; Henderson, 1999). The greater the clusterng of an ndustry wthn the local authorty, the greater the potental benefts from spllover mpacts. Conversely, greater agglomeraton may lead to congeston, and therefore may lower productvty. The dversfcaton ndex s also ncluded to pck up urbansaton economes assocated wth operatng n an area wth a large number of dfferent ndustres. Hgher dversfcaton s usually assumed to have benefts to producers through spllover effects. The Herfndahl ndex of ndustral concentraton s measured at the 5-dgt 1992 SIC level to take account of any market power effects (whch are expected to be assocated wth the propensty to undertake both exportng and R&D). The varable that measures f the establshment belongs to an enterprse operatng n more than one (5-dgt) ndustry (.e. >1 SIC multplant) s ncluded to proxy for any economes of scope. 10

13 Table 2: Dstrbuton of establshments, 2000, by whether exported and/or undertook R&D Do not export Export All Manufacturng No R&D Undertake R&D Total Non-manufacturng No R&D Undertake R&D Total Source: authors own calculatons usng weghted data from (populaton weghts avalable n ) Lastly, we present some basc comparsons between exporters, those undertakng R&D and some establshment characterstcs before dscussng multvarate modellng results n Secton IV. Frstly, Table 2 shows that n manufacturng some 44% of establshments were nvolved n exportng, whle only 18.6% ncurred spendng on R&D n The table also shows that some 30.5% of exporters also engaged n R&D actvtes (or alternatvely, nearly 73% of those manufacturng establshments undertakng R&D also exported). Ths suggests a strong relatonshp between the two actvtes, although there were a substantal number of establshments that exported but wthout fndng t necessary to also engage n R&D. Table 3: Exportng (and export ntensty) n UK establshments, 2000, by sze (percentage fgures) Employment sze Manufacturng Non-manufacturng Total % export exports/sales % export exports/sales % export exports/sales Total Source: authors own calculatons usng weghted data from (populaton weghts avalable n ) There was a wde varaton across ndustres n the propensty for frms to export (e.g. nearly 74% of establshments n the Chemcals sector were engaged n exportng, wth some 26% of goods sold abroad; n comparson, most non-manufacturng sectors had low levels of 11

14 exportng 12 ); there was a much smaller, although sgnfcant, varaton across regons n exportng (e.g. over 61% of manufacturng establshments n Northern Ireland exported, whle only 35% n London dd so). Table 3 also shows establshment sze seemed to determne whether goods and servces were sold abroad - exportng ncreased wth establshment sze. Further detals are avalable n Table 3.4 n Harrs and L (2005), coverng the characterstcs of those establshments that exported separately from those that dd not (for both manufacturng and non-manufacturng sectors). In summary, ths shows that all of the followng were hgher for exporters: the lkelhood of engagng n (contnuous) R&D and to be nnovatve (as measured by whether they produced new product and/or process nnovatons; whether novel or otherwse); level of co-operaton wth (nternatonal) partners outsde the enterprse; captal ntensty; age of the establshment; the level of ndustral concentraton; the mportance of agglomeraton economes (but not dversfcaton); the propensty to have producton capacty n the Greater South East regon; and the probablty of belongng to a mult-regon, mult-plant frm, operatng n more than one ndustry, and/or beng foregn-owned. IV. Estmatng the Determnants of Exportng In modellng the determnants of exportng usng the CIS-ARD merged dataset for 2000, separate models have been estmated for manufacturng and servces (gven the dfferent export ntenstes between these two sectors). We only report the results for manufacturng n ths study (although those for servces are smlar), gven space constrants and the fact that a much larger proporton of establshments engaged n (hgher levels of) exportng n ths sector 13. Wth respect to the econometrc modellng of exportng behavour (wth R&D actvtes as explanatory varables), we use a Heckman (1979) approach, whch recognses that those that export are not a random sub-set of all establshments; rather, modellng export ntensty (exports per unt of sales) needs to take nto account that those wth non-zero exportng levels have certan characterstcs that are also lnked to how much s exported. Falure to take nto account ths self-selecton element when modellng exportng ntensty would lead to results that suffer from selecton bas. Note, maxmum lkelhood estmators have to be employed to obtan both effcent and consstent coeffcents (see, for nstance, Barros et. al., 2003), and both equatons 12 Indeed, the data show that sgnfcant proportons of frms export n only the wholesale trade, computng and R&D sectors of non-manufacturng. 13 The data shows that 64% of the value of all exports n 2000 orgnated from the manufacturng sector (even though ths sector accounted for some 26.7% of total turnover); manufacturng also accounted for some 74% of total R&D spendng (when omttng the R&D sector). Manufacturng establshments accounted for nearly 61% of all those engaged n exportng (and 51% of those engaged n R&D), despte only accountng for 36.5% of UK establshments. 12

15 must be estmated smultaneously (usng for example the FIML estmator) 14. A frst ssue that needs to be tackled s that of dentfcaton n the Heckman model (.e. whch varables appear n the probt estmaton but not n the sample selecton equaton). Our approach s to nclude varables assocated wth the fxed costs of exportng (such as the captal/employment rato, ndustry agglomeraton, the Herfndahl ndex and the mpact of regulatons/standards) only n the equaton determnng whether exportng takes place or not 15. These varables are a pror lkely to be assocated wth breakng down barrers to enterng export markets, rather than how much s exported (condtonal on enterng such markets). In addton, a method of smultaneous estmaton has also been proposed to take nto account the endogenety of exportng and R&D decsons n modellng exportng behavour. Ths nvolves the estmaton of smultaneous probt models that treat exports and R&D as jontly endogenous varables. For nstance, usng a technque frst devsed by Maddala (1983), t s possble to regress the endogenous varables on the entre set of assumed exogenous varables and construct the predcted varables as nstruments. In the second stage, export and nnovaton varables need to be replaced wth these nstruments to yeld unbased estmates of the mpact of nnovaton on exports (and vce versa). Smlar smultaneous approaches have been employed n several emprcal studes treatng nnovaton and exports as nextrcably nterdependent (Hughes, 1986; Zhao and L, 1997; Smth et. al., 2002; and Lachenmaer and Woessmann, 2006). We have estmated two versons of the Heckman model: the frst (denoted Model 1) takes no account of the lkely endogenety between exportng and R&D (.e. the latter s assumed to be predetermned). In Model 2 we allow R&D to be endogenous, and replace t wth ts predcted value obtaned from the reduced-form model determnng R&D (see Table A2 n the appendx). The results for the manufacturng sector, as to whether establshments export or not, are provded n Table 4(a), wth margnal effects reported 16. The dagnostc tests provded n the lower part of the table show that the Heckman selecton procedure s clearly justfed 17 : the correlaton between the error terms of the two equatons n the model s large ( ρ = ) and statstcally sgnfcant from zero (as suggested by the Wald test of ndependent equatons wth a χ 2 (1) = value that rejects the null hypothess that ρ = 0 at better than the 1% sgnfcance level). We have also undertaken a Smth-Blundell test for exogenety based on Model 2, whch 14 Note, the use of the Heckman sample selectvty approach s not about separatng out the exportng decson nto two stages. The latter has been crtcsed by, for nstance, Wagner (2001), who argues that (based on the ex post nature of sunk costs) there s no such thng as a two-step decson nvolvng () the decson to export and () how much to export. These are not mutually exclusve, as costs are carefully consdered when frms decde (by producng the proft-maxmsng quantty at the gven prce) whether to partcpate n such export markets or not. 15 Note, n any event these varables were not sgnfcant when ncluded n the equaton determnng export ntensty. 16 The z-values for Model 2 have not been corrected for bas that may result from usng a generated varable (predcted R&D) based on the model estmated n Table A2; nevertheless, ths bas s unlkely to be very large. 17 An outlne of the Heckman model, and thus defntons of the parameters ρ, σ, and λ, s provded n the appendx. 13

16 ncludes all the (sgnfcant) varables n the model as determnants of the probablty of exportng and wth R&D nstrumented by those 8 varables hghlghted n Table A2 (e.g. hgh cost of nnovaton), whose parameter estmates are also hghlghted n Column 1. These nstruments were chosen on the bass of whether they were sgnfcant determnants of R&D (see Table A2) but not sgnfcant n determnng whether the establshment exported (.e. Model 2). The test obtaned a χ 2 (1) value of 22.6, whch rejects the null of exogenety at better than the 1% sgnfcance level 18. An establshment undertakng R&D was assocated wth a sgnfcantly hgher lkelhood of non-zero exports,.e. a (cet. par.) 17.5% hgher probablty of sellng nternatonally when R&D s treated as exogenous. However, when we allow for R&D to be endogenous (by replacng R&D wth ts predcted value), the margnal effect for ths varable falls from to The fnal column n Table 4(a) shows that only some 18.6% of UK manufacturng establshments undertook R&D n 2000; thus, ths had an mportant mpact on the propensty to export. The parameter estmates for the remanng varables, whch enter as determnants of whether exportng s undertaken or not, are mostly very smlar for Models 1 and 2. Thus, we shall refer only to those reported for Model 2, where R&D enters as an endogenous varable (.e. the preferred model). 18 Note, ths test s ndcatve, as the endogenous varable we nstrument s dchotomous (the test would normally requre R&D to be a contnuous varable). 14

17 Table 4(a): Determnants of exportng n UK Manufacturng, 2000 Dependent varable: exportng undertaken or not Model 1 Model 2 Means pˆ / x pˆ / x z-value z-value ( x ) R&D Establshment sze employees employees employees ln enterprse sze x Mult-plant Other factors Absorptve capacty (external knowledge) Absorptve capacty (natonal co-op) Absorptve capacty (org structure & HRM) Absorptve capacty (nternatonal co-op) Absorptve capacty (scentfc knowledge) ln Captal/employment rato ( m per worker ARD data) ln Labour productvty ( '000 per worker) Industry agglomeraton ln Herfndahl ndex Impact of regulatons/standards Industry sector (2-dgt 1992 SIC) Food & drnk Textles Clothng & leather Wood products Paper Publshng & prntng Chemcals Rubber & plastcs Non-metallc mnerals Basc metals Fabrcated metals Machnery & equpment nes Electrcal machnery Medcal etc nstruments Motor & transport Furnture & manufacturng nes Regon Eastern England Northern Ireland ρ σ λ (unweghted) N N (export > 0) Log pseudo-lkelhood Wald test of ndependent equatons: χ 2 (1) Smth-Blundell test of exogenety of R&D: χ 2 (1) Notes: Weghted regresson s used (wth populaton weghts avalable n ). Model 1 s the baselne model, whle Model 2 controls for endogenety of R&D (hence the predcted value s used based on the reduced-form model n Table A2). The reported parameter estmates are all statstcally sgnfcant at the 10% level or better. For varable defntons, see Table 1. 15

18 The sze of the establshment had a major mpact on whether any exportng took place; vs-à-vs the baselne group (establshments employng less than 10), movng to employees ncreased the probablty of exports > 0 by 17.5%, an ncrease n the probablty by 28.4% n the group and up to an ncrease of almost 36% for establshments wth 200+ employees. Ths confrms the results presented n Table 3 that sze and the propensty to export are postvely related 19. Gven that the last column n Table 4(a) shows the dstrbuton of establshments by sze, t can be seen that the UK has relatvely fewer establshments n the largest sze band lsted, thus to some extent lmtng the number of establshments that export. We have also ncluded the sze of the enterprse as well as establshment sze, but only for those establshments that belong to mult-plant enterprses (employment sze for sngle-plant enterprses s already accounted for usng the dummy varables for establshment sze-band). Our results show that ncreasng the sze of the enterprse s negatvely related to the probablty of sellng overseas f the establshment belonged to an enterprse that had multple plants n 2000, suggestng that (havng controlled for the large postve relatonshp between establshment sze and exportng) large mult-plant enterprses have a slghtly hgher propensty to supply UK markets vs-à-vs sngle-plant enterprses. To confrm ths, we also tred enterng the mult-plant dummy varable n addton to the composte varable comprsng ln enterprse sze mult-plant. If both varables nvolvng mult-plant status are entered, they are both nsgnfcant (due to collnearty problems); f just the mult-plant varable s entered a sgnfcant (and large) negatve value s obtaned for the parameter estmate confrmng that enterprses wth more than one plant have a (slghtly) lower lkelhood of sellng overseas. Overall absorptve capacty was mportant n determnng whether an establshment had nonzero exports n the manufacturng sector, but the varables representng the acquston of external knowledge and natonal co-operaton for nnovaton purposes become nsgnfcant when R&D s treated as endogenous. Ths suggests that these aspects of absorptve capacty (whch by constructon are drectly based on nnovaton actvtes) are mportant drvers of whether any R&D s undertaken, and then ndrectly mpact on whether the establshment exports through the ncluson of (endogenous) R&D n the exportng equaton 20. Establshments 19 Instead of usng sze-bands (to pck up potental non-lnear effects), actual ln employment for the establshment (and ln employment-squared) can be entered. Usng ths form of Model 2, we obtan margnal effects of (4.04) and (-2.26) for ln employment and ln employment-squared respectvely (z-values n parentheses). Other parameter estmates n the model change very lttle (n terms of ther value or sgnfcance). Ths (alternatve specfcaton) confrms that sze and the propensty to export are postvely related, and that at (very) large levels of employment there s a flattenng of the relatonshp. However, usng ln employment and ln employment-squared does not pck up any postve non-lnear effect as shown n Table 4(a), where movng from lower to hgher szebands results n an ncreasng postve relatonshp between sze and the lkelhood of exportng. Moreover, the ncluson of a cubc-term for ln employment was not successful (the parameter estmate on ln employment becomes nsgnfcant). 20 Ths can also be seen by comparng the results for the structural equaton (Model 2) n Table 4(a), and for the reduced-form model n Table A2. 16

19 that had hgher nternal absorptve capacty (based on ther organsatonal and HRM characterstcs) were margnally more lkely to overcome barrers nto export markets; ncreasng ths aspect of absorptve capacty by one standard devaton from ts mean value ncreased the probablty of exportng by over 2%. The ablty to nternalse external knowledge ganed from nternatonal co-operaton ncreased the lkelhood of exportng by 4.4% (agan based on one standard devaton ncrease), whle absorbng scentfc knowledge (from research organsatons) resulted n an ncrease n the lkelhood of sellng overseas by around 6%. Here the relatve magntude of dfferent dmensons of absorptve capacty s perhaps not surprsng. From the perspectve of technologcal opportuntes, scence-based technologcal opportuntes generally requre a hgher level of absorptve capacty than those generated by other sources of knowledge, such as supplers and customers (Becker and Peters, 2000). Gven that the largest absorptve capacty s lkely to be called for to assmlate scentfc knowledge stemmng from research nsttutes (Leponen, 2001), we could therefore expect the absorptve capacty for ths type of knowledge to have the largest mpact on establshment s nternal capabltes (wth respect to exportng n ths context). Establshments wth hgher labour productvty were also more lkely to enter export markets; a doublng of ths varable (from ts mean value of just under 60k turnover per worker to just over 119k) ncreased the probablty of exportng by some 7.2%. In all, these results confrm those often gven n the lterature that better establshments (n terms of ther ablty to nternalse external knowledge, and productvty) were more lkely to export. Turnng to the varables assocated specfcally wth the fxed costs of entry nto export markets (and whch dentfy the Heckman model), more captal-ntensve establshments were also more lkely to export; doublng the captal-to-labour rato (from a mean of just over 3.5k per worker n 1980 prces) ncreased the probablty of exportng by about 1.4%. Industry/market concentraton was also lnked to a greater probablty of exportng; ncreasng the Herfndalh ndex of market concentraton, from ts mean value of 0.06 to 0.16 (the latter beng the average value for the 90 th decle group n manufacturng), rased the probablty of exportng by 7.3%. The mpact of regulatons/standards as a barrer to nnovaton also reduced the lkelhood of the establshment exportng (by some 7.7%). Lastly, sector also mattered, wth all those ndustres lsted havng hgher probabltes of exportng (by between 19.9 to 47.7%) vs-à-vs mnng & quarryng (the baselne group). The sectors wth the hghest propenstes to export were textles, chemcals, rubber & plastcs, basc metals, machnery & equpment, electrcal machnery, and medcal & precson nstruments. Establshments n Northern Ireland were more lkely to engage n sellng overseas, wth a 23.6% 17

20 hgher probablty of exportng. There were no other sgnfcant regonal effects for the manufacturng sector. None of the other varables entered (see Table 1) proved to be sgnfcant n determnng entry nto export markets (e.g. age of the establshment, foregn ownershp, ndustry dversfcaton, whether the establshment belonged to an enterprse operatng n more than one ndustry, more than one regon, or n the Greater South East). In modellng how much of turnover s exported, the results for manufacturng are reported n Table 4(b), coverng just those wth postve export sales (gven the two-stage Heckman approach used, these results are condtonal on the model determnng whether exportng takes place at all). The models presented concde wth the treatment of contnuous R&D as beng ether exogenous or endogenous (n a comparable way to how R&D s treated n Table 4(a)). Agan we have undertaken a Smth-Blundell test for exogenety based on Model 2, whch ncludes all the (sgnfcant) varables n the model as determnants of the probablty of exportng and wth contnuous R&D nstrumented by those 17 varables whose estmated parameters are hghlghted n Column 5 of Table A2 (e.g. US-owned). These nstruments were chosen on the bass of whether they were sgnfcant determnants of contnuous R&D (see Table A2) but not sgnfcant n determnng exportng ntensty (.e. Model 2). The test obtaned an F-statstc of (whch rejects the null of exogenety at better than the 1% sgnfcance level). 18

21 Table 4(b): Determnants of exportng ntensty n UK Manufacturng, 2000 (cont.) Dependent varable: Model 1 Model 2 Means ( x ) ln exportng ntensty z-value z-value βˆ βˆ R&D actvtes R&D contnuous Establshment sze employees employees employees employees Other factors Absorptve capacty (natonal co-op) Absorptve capacty (scentfc knowledge) Industry sector (2-dgt 1992 SIC) Food & drnk Paper Non-metallc mnerals Machnery & equpment nes Electrcal machnery Medcal etc nstruments Motor & transport Regon London Northern Ireland South West Scotland Wales Smth-Blundell test of exogenety of R&D contnuous: F (1, 3298) Notes: Weghted regresson s used (wth populaton weghts avalable n ).. Model 1 s the baselne model, whle Model 2 controls for endogenety of contnuous R&D (hence the predcted value s used). All parameter estmates are statstcally sgnfcant at least at the 10% level. Values of dagnostc tests are the same as n Table 4(a). For varable defntons, see Table 1. In Model 1, undertakng contnuous R&D was assocated wth an over 52% hgher level of export ntensty 21, but when contnuous R&D s nstrumented t s no longer statstcally sgnfcant (rather, as dscussed below, the mportance of the sze of the establshment on ntensty ncreases sgnfcantly when the contnuous R&D varable s omtted, suggestng a postve relatonshp between the undertakng of contnuous R&D and the sze of the establshment condtonal on havng controlled for entry nto export markets). Whle Table 4(a) shows that the sze of the establshment had a major mpact on whether any exportng took place (.e. the larger the establshment, the greater the probablty of exportng, 21 Snce the dependent varable n the model s the natural log of export ntensty, the elastcty wth respect to a dchotomous varable s gven byexp( ˆ) β 1. 19