On the relationship between innovation and export: the case of Australian SMEs

Transcription

1 On he relaionship beween innovaion and expor: he case of Ausralian SMEs Alfons Palangkaraya Melbourne Insiue of Applied Economic and Social Research & Inellecual Propery Research Insiue of Ausralia The Universiy of Melbourne Inellecual Propery Research Insiue of Ausralia Working Paper No. 3/13 ISSN February 2013 Inellecual Propery Research Insiue of Ausralia Level 5, Faculy of Business and Economics Building The Universiy of Melbourne Vicoria 3010 Ausralia Telephone: 61 (0)

2 On he relaionship beween innovaion and expor: he case of Ausralian SMEs ALFONS PALANGKARAYA Melbourne Insiue of Applied Economic and Social Research & Inellecual Propery Research Insiue of Ausralia The Universiy of Melbourne This paper invesigaes he link beween innovaion and expor marke paricipaion using Ausralian small and medium enerprises (SMEs) daa. The resuls show ha expor and innovaion are posiively linked. Depending on he indusry and he ype of innovaion (process or produc), innovaion may lead o expor and, o a lesser exen, expor may lead o innovaion. Firms in he primary secor (agriculure and mining) show he sronges evidence ha innovaion leads o expor. From firms in he services secor, here is indicaion ha only process innovaion leads o expor. Also, only in his secor, here is evidence ha expor may lead o ( process) innovaion. Keywords: innovaion; expor; small and medium enerprises; propensiy score maching JEL Classificaion: F14, O12, O14, O31 The auhor acknowledges and hanks he Economic Research Insiue of ASEAN and Eas Asia (ERIA) under he ERIA Research Projec on Globalizaion and Innovaion in Eas Asia for financing he sudy. The auhor also hanks he Ausralian Bureau of Saisics for providing online daa access via RADL o he business level panel daa 2004/ /07 of he Business Longiudinal Daabase. Finally, he auhor is graeful o receive valuable commens and suggesions from Ari Kuncoro, Dionisius Narjoko, John Haisken-DeNew, and paricipans of he 41 s Ausralian Conference of Economis in Melbourne in 2012, and paricipans of ERIA Working Group Workshop on Globalizaion and Innovaion in Eas Asia in Jakara, Denpasar and Bangkok in 2010 and 2011 in which earlier versions of he paper was presened. The auhor claims responsibiliy for all he views expressed in his paper. 1

3 I Inroducion This paper empirically invesigaes he direcion of causaliy beween innovaion and expor marke paricipaion using firm level daa of Ausralian small and medium enerprises (SMEs) employing 200 or fewer workers. A beer undersanding of he effecs of globalisaion on economic performance, paricularly on he performance of firms, is imporan o ensure ha relevan public policy aains is opimum benefi. One poenial benefi of globalisaion ha could serve for he raionale of such policy comes in he form of produciviy improving mechanism resuling from he paricipaion in he inernaional marke such as he expor marke. Exising empirical evidence and heoreical predicions indicae ha, insead of being a by-produc of he hypohesised learning-by-exporing effecs, he produciviy advanage of exporing firms relaive o non-exporing firms comes from heir preexpor differences in performance (Aw e al, 1997; Bernard and Jensen, 1999 and 2004; Meliz, 2003). This pus he asserion of he produciviy improving benefis of expor marke paricipaion ino quesion. However, recen sudies have analysed he inermediae role of innovaion and invesmen in Research and Developmen (R&D) in he expor and performance relaionship more closely. They have found evidence of possible learning effecs from paricipaing in he expor marke via ha inermediae channel (Crespi e al. 2008; Damijan e al., 2010; Girma e al., 2003 and 2008; MacGarvie, 2006; Fernandes and Paunov, 2010; Busos, 2011). Thus, if he role of learning from exporing is really imporan as he above sudies seem o sugges, hen a failure in recognising i could lead o subopimal policy formulaion for aking advanage of globalisaion. However, due o firm and counrylevel heerogeneiy, we sill need furher evidence on he imporance of innovaion in deermining he direcion of causaliy beween globalisaion and economic performance. In oher words, we need addiional sudies which focus on he role of he inermediae sep, namely innovaion. This paper conribues o he exising lieraure by evaluaing he link beween expors and innovaion in a similar fashion o recen sudies such as Damijan e al. (2010) and Crespi e al. (2008). However, unlike hese and many of he exising sudies which concenrae on medium and large manufacuring firms, we uilise daa from a sample of small and medium firms from all indusrial secors: primary (agriculure, 2

4 fishery and foresry, and mining), secondary (manufacuring), and eriary (services). Our more specific conribuion for he relevan lieraure in he case of Ausralian economy is ha his paper provides furher insighs o hose esablished by exising sudies such as Palangkaraya and Yong (2007; 2011) on he relaionship beween inernaional rade and produciviy by looking a innovaion as he likely inermediae sep. More generally, his paper conribues o debae on he benefis of proglobalisaion policy o small and medium firms from across differen secors. A confirmaion of he learning-by-exporing hypohesis, for example, indicaes ha expor marke paricipaion improves small and medium firms performance hrough he simulaion of innovaive aciviies. Thus, he poenial benefis from policies designed o improve global marke aciviies (paricularly in he expor marke) would be higher han in he case when here is no learning effecs and ha such policies may be more effecive if coordinaed wih innovaion policy. Furhermore, he findings of his paper could also demonsrae how he learning effecs are generaed given exising differences in he ypes of innovaion involved. Wih a beer undersanding of hese issues, he governmen would be in a beer posiion in designing policies ha can address any marke failure which may lead subopimal resource allocaions on differen ypes of innovaive and expor marke aciviies. To achieve our objecive, in his paper we apply he propensiy maching score (PSM) mehodology on firm level panel daa obained from he Ausralian Bureau of Saisics Business Longiudinal Daabase 2004/05 o 2006/07. The panel daa cover approximaely 3000 firms wih 200 employees or less. These small and medium businesses operae in all secors of he Ausralian economy excep Governmen adminisraion, Educaion, Healh, and Uiliies which have been excluded from he daabase. The res of he paper is srucured as follows. Secion 2 provides a brief discussion on recen relaed sudies and he insiuional backgrounds (Ausralian SMEs expor and innovaion aciviies in general and case-sudy inspired illusraions of he link beween expor and innovaion from SMEs in Ausralian wine indusry and hose which have received Ausralian Exporer Award beween 2001 and 2010). Secion 3 discusses he empirical framework and he daa. Secion 4 presens and discusses he resuls. Secion 5 summarises he finding and discusses some of he policy implicaions. 3

5 II Lieraure Review (i) Expor and Innovaion The link beween expor and produciviy has been he subjec of many differen sudies for many years due o is imporan implicaions on he benefis of globalisaion and he iner-link beween indusrial and innovaion policy and rade policy. As he availabiliy of large, firm-level, longiudinal daa has improved over he las fifeen years, our abiliy o evaluae wo major compeing hypoheses (which are no muually exclusive) behind he expor-produciviy relaionship has also improved in erms of deails and sophisicaion. The firs hypohesis of ineres is called he self-selecion hypohesis and i is based on he idea ha beer or more producive firms self-selec ino expor marke because of he poenially high exra coss for enering he foreign markes. These ransacion coss of exporing may include, for examples, ransporaion coss, disribuion or markeing coss, or he coss o ailor he producs for foreign consumers. Because of such enry barriers, firms may exhibi forward looking behaviour by aking acions for improving heir produciviy before enering any foreign marke. As a resul, any cross-secional performance difference beween exporers and non-exporers can be explained by he ex ane differences beween he wo ypes of firms. The compeing hypohesis, he learning-by-exporing hypohesis, argues ha expor marke paricipaion provides he opporuniy for exporers o improve heir performance due o a higher level of marke compeiion and he poenial for knowledge flows from inernaional consumers. Wagner (2007), for example, surveys more han 40 sudies based on firm level daa from more han 30 counries and finds ha a majoriy of he sudies suppors he self-selecion hypohesis while paricipaion in he expor marke does no appear o lead o improved produciviy. The lack of suppor for he learning-by-exporing hypohesis is furher shown by a number of recen heoreical and empirical models which emphasise he role of firm heerogeneiy and R&D. For example, Consanini and Meliz (2008) endogenises firm s decision o expor and innovae and shows ha he expor-produciviy link can be explained by he decision o innovae before he expor marke enry. Recen empirical sudies, such as Aw e al. (2008), look a he relaionship in more deails by incorporaing R&D invesmen or innovaion decision and also find evidence for he self-selecion hypohesis. Oher recen sudies which also suppor he self-selecion 4

6 hypohesis include Kirbach and Schmiedeberg (2008) and Chada (2009). The laer is ineresing because i finds ha innovaion can ac as a sraegic ool o gain marke share in he world markes and hus i is imporan for firms o innovae o ener he expor marke. Finally, Long e al. (2009) explores he effecs of rade liberalizaion on he incenives for firms o innovae and on produciviy. They find ha rade liberalizaion s impac is dominaed by he self-selecion effecs and he effecs of rade on innovaion or incenive o spend in R&D depend on he coss of rade. Neverheless, oher sudies such as Crespi e al. (2008), Damijan e al. (2010), Girma e al. (2003; 2008), MacGarvie (2006) and Fernandes and Paunov (2010) provide evidence ha globalisaion may feed back ino improved domesic performance hrough he learning effecs from global marke paricipaion on innovaion. The las wo sudies menioned above show he learning effecs occur hrough impors while he oher sudies show he effecs occur hrough expor marke paricipaion. Furhermore, De Loecker (2007) and Io (2011) find ha firms which expor more o higher income and more echnologically advanced regions exhibi sronger learning-by-exporing effecs; his is consisen wih he noion ha expor marke paricipaion provides benefis hrough beer access o bes pracice echnologies (Girma e al., 2003). However, given he reliance of mos of he sudies cied above on daa from medium and large enerprises and, paricularly, from he manufacuring secor, we need furher complemenary evidence based on SMEs. For reasons such as he coss of acquiring legal proecions on innovaion and heir enforcemen, i has been argued ha SMEs may have a lower propensiy o innovae han larger firms (for examples, Acs and Audresch, 1988 and Arundel and Kabla, 1998). Furhermore, Jensen and Webser (2006) argue ha he implicaion of underinvesmen in innovaion aciviy by SMEs can poenially be more significan han realised given ha SMEs may hold a significan share of overall economic aciviy. In oher words, a beer undersanding of he innovaive paerns of SMEs is crucial for an effecive innovaion policy in order for i o generae economic growh in he mos opimal way. For example, SMEs may lack he necessary absorpive capaciy o ap from world s bes pracice echnology in he global marke, reducing he learning by exporing effecs (Io and Lechevalier, 2010). Similarly, individual SME s expor revenues may no be large enough o induce he firm 5

7 o inves in new echnologies, an imporan channel for rade liberalizaion o lead o growh following heir enrance o he expor marker argued by Busos (2011). In addiion, we also need furher analysis of firms in indusrial secors oher han manufacuring and he role of differen ypes of innovaion. Firs, he exen of marke failure in innovaion aciviies vary by indusrial secors and he effeciveness of insrumens o comba such marke failure including he provision of inellecual propery righs (IPRs) proecion also vary by secors (see, for example, Mansfield e al., 1981 as cied in Jensen and Webser, 2006). Second, he ype of innovaion aciviies may well vary across indusrial secors because of he mulifaceed naure of innovaion. Finally, Schumpeer (1934), for example, discusses innovaion in erms of produc innovaion, process innovaion, organisaion innovaion and marke innovaion. If we recognise ha he variaion in he ypes of innovaion aciviies as classified above correlae wih he characerisics of he producs or he markes in which a firm operaes, hen we may find cerain firms in cerain indusries o be more concerned wih produc innovaion while oher firms in oher indusries o be more concerned wih process innovaion. Consequenly, we expec he link beween innovaion and expor may depend on he ype of innovaion aciviy. Furhermore, IPRs proecion such as paen or rade mark may be more effecive for produc innovaion han process innovaion, leading o varying paerns of innovaive aciviies and expor-innovaion relaionships across indusrial secors as well. (ii) Insiuional Background Ausralian SMEs expor and innovaive aciviies The Ausralian SMEs presen an ineresing case o sudy he deerminan of firm level innovaive aciviies including he link beween expor and innovaion. As in mos oher counries, SMEs are an imporan componen he Ausralian economy accouning slighly more han 60 per cen of oal employmen and 50 per cen of value added (ABS, 2001). Because of hese, Ausralian SMEs have received a lo of governmen s aenion in erms of he various policies and incenives argeed a hem in order o help hem improve heir producive and innovaive performance. Naurally, he imporance of SMEs varies across indusries (ABS, 2008). The highes share of value added conribuion can be found from SMEs in he Agriculure, 6

8 foresry and fishing (97 per cen of he indusry value added in 2006/07), Renal, hiring and real esae services (90 per cen of he indusry value added), and Accommodaion, cafes and resauran (75 per cen of value added). In conras, he lowes share of value added can be found from SMEs in Reail rade (56 per cen of indusry value added), Manufacuring (45 per cen), and Informaion media and elecommunicaion (17 per cen). Similarly, he role of SMEs in Ausralia s expor aciviies also varies by indusries. Overall, in , SMEs accoun for 90 per cen of he number of Ausralian businesses which paricipae in he expor marke. However, as found in oher counries, hey accoun for less han 10 per cen of value of he goods expored (ABS, 2006). Furhermore, a recen saisical repor, ABS (2010), shows ha based on he value of goods expor, by he financial year of SMEs share is he highes in he following secors: Consrucion (37 per cen), Transpor, posal and warehousing (23 per cen) and Wholesale rade (16 per cen). Thus, we probably will no learn much on he link beween expor and innovaion of Ausralian SMEs if we only look a he manufacuring indusry. On innovaion, according o he laes ABS Innovaion Survey conduced in 2005 (ABS, 2007), here are approximaely 141,300 businesses 1 operaing in Ausralia and around 34 per cen of hem businesses underake innovaion by inroducing eiher new produc, new operaional process, and/or new organisaional processes. 2 As expeced, he exen of innovaiveness varies by business size wih around 58 per cen of very large businesses (250+ employees), per cen of medium businesses (20-99 employees), and per cen of small businesses (5-19 employees) repored as innovaors. I also varies by indusry wih he leading indusries include Elecriciy, gas and waer supply (49 per cen of businesses are innovaors), Wholesale rade (43 per cen) and Manufacuring (42 per cen). I is also worh noing ha, according o he innovaion survey, for Ausralian SMEs operaional process innovaion is he mos imporan ype of innovaion compared o he oher wo. Thus we may expec ha if 1 The ABS uses he erm Businesses while in his paper we use he erm enerprises o be consisen wih he erm (ha is, SME) used in he exising lieraure. 2 Here, new may refer o new o he business (74 per cen of produc innovaion), new o he indusry (10 per cen), new o Ausralia (10 per cen), or new o he world (6 per cen). 7

9 here is any link beween SMEs expor and innovaion aciviies, process innovaion would be relaively more imporan. Finally, in erms of he conribuion o he degree of sales urnover, 65 per cen of he innovaing businesses repor ha heir produc innovaion accoun less han 10 per cen of heir urnovers (ABS, 2007). In oher words, in general, produc innovaion is no a significan revenue generaor for Ausralian businesses. However, he conribuion of producion innovaion on oal revenues varies across indusries wih businesses in mos services indusry repored less han 10 per cen share. This resul is undersandable given he naure of he produc in he service secor. In conras, for businesses in he Mining and Manufacuring indusries, he share of produc innovaion on oal revenues range from 10 per cen o as high as 50 per cen. Also, when we look a variaion across business size, i is ineresing o noe ha none of he large businesses (100+ employees) repors ha produc innovaion conribues more han 50 per cen of business urnovers. In conras, 12 per cen of small businesses wih 5 19 employees repored ha more han 50 per cen of heir urnovers can be aribued o produc innovaion. In oher words, innovaion appears o be more imporan for he livelihood of smaller businesses, again highlighing he imporance of a sudy of SMEs o complemen he sudies based on daa on large enerprises. Case sudies Given he anonymiy of he firms in he BLD panel daa we use, o help in inerpreing he esimaion resuls we discuss briefly he case of Ausralian wine indusry and SMEs which are he recipiens of he Ausralian Exporer Award given annually. 3 The firs, based on an in deph sudy of Aylward (2004; 2006), illusraes he relaionship beween innovaion and expor for Ausralian wine producers. The second case highlighs some imporan characerisics of SMEs and how hey relae o expor performance, paricularly for SMEs in he Service secor. The mos imporan lesson for hese wo case sudies which will be discussed below is ha expor marke paricipaion appears o be driven by he firms abiliy o 3 The Ausralian Expor Awards has run for 48 years and provided recogniions and honors o excepional Ausralian exporers based on he crierion of susainable expor growh achieved hrough innovaion and commimen. See hp:// (accessed March 11, 2011) for more deails. 8

10 coninuously come up wih beer processing echnology via skill and echnology updaing o deliver heir services. 4 Thus, i appears expor marke paricipaion depends more on process innovaion han on produc innovaion. Furhermore, how business owners or managers view he imporance of innovaion maers. Ausralian Wine Indusry According o Aylward (2004; 2006), Ausralia was he fourh larges exporer of wine in erms of value sharing around fory per cen of global wine expors o he Unied Saes in The Ausralian wine indusry consiss of wo major clusers (Souh Ausralia and New Souh Wales & Vicoria). However, while he Souh Ausralian cluser only accouned for around weny five per cen of wineries, is shares of Ausralian s oal wine producion and expor were fify and sixy per cen respecively. More ineresingly, Aylward s sudies find a close link beween he Souh Ausralian wine cluser s higher produciviy and propensiy o expor and explain ha he differences beween he wo clusers in erms of he business owners or managers beliefs on he imporance of innovaion. For example, sixy six per cen of wine producers in Souh Ausralia responding o Aylward s inerview hough here was a srong link beween heir innovaion and expor performance. In conras, only fory wo per cen of he respondens from he New Souh Wales & Vicoria clusers believed he same. The inerviewees also differed in erms of how hey defined innovaion, he exen of heir firm s collaboraion, and he use of Ausralian wine indusry s research and analyical services. One oher finding o noe from he sudy is ha while here was negligible difference in how he firms in boh clusers define produc innovaion, hey differ significanly in defining wha hey hough a process innovaion enailed. This las finding poins o he possibiliy ha process innovaion is probably more imporan han produc innovaion in explaining he link beween expor and innovaion in he indusry. Ausralian Exporer Award Winners in Services Secor In he las fory eigh years, he Ausralian governmen has given awards o businesses deemed o have excepional performance in he expor marke every year. The awards 4 See he case sudies for he award winners provided by Ausralian Expor Awards websie menioned in he previous end noe. 9

11 are given o businesses belonging in various caegories such as agribusiness, ars and enerainmen, emerging exporer, and large and advanced manufacurer. For he purpose of his sudy, wo caegories of paricular ineres are he emerging exporer and small and medium businesses in services caegories. Beween 2001 and 2010, weny four businesses received he emerging exporer awards, en of which were from he services secor. In he same period, sixeen businesses received he small and medium exporer in services awards. In erms of heir produc characerisics, a majoriy of hese award recipiens in he services secor operaed in informaion echnology relaed field (en businesses), highly specialized engineering design and prooype manufacuring operaion (eigh businesses), and specialized manufacuring and indusrial consulancy services for he mining indusry (four businesses). For example, one hese businesses which operaed in informaion echnology services and employed around fify was considered as he larges specialis provider of independen informaion securiy consuling services in he region. Is consumers came from more han weny counries such as Singapore, Malaysia, Souh Korea, Japan, he Unied Saes, and France. One oher Ausralian small business ha excelled in he expor marke sold mariime simulaion, raining and consulancy services o he inernaional mariime and defence indusries. Unforunaely, here is no deailed case sudy similar o Aylward (2004; 2006) s o help us in undersanding he imporance of innovaion o hese services exporers. III Empirical Model and Daa (i) Empirical Model To address our research quesions, we need he abiliy o make causal inference as opposed o simply es for he exisence of correlaion beween innovaion and expor. We adop an empirical mehodology which can produce unbiased esimaes of he relevan reamen effecs (from being an exporer or being an innovaor) by assuming ha ha we have observables o explain firms propensiy o eiher expor or innovae o conrol for unobserved confounding facors. In paricular, we follow he approach used in similar sudies such as Becker and Egger (2010) and Damijan e al. (2010) by adoping he propensiy score maching (PSM) mehod. 10

12 As argued by an exensive lieraure on he PSM mehodo such as Dehejia and Wahba (2002), he esimaion of causal effecs based on a comparison of reamen group wih a non-experimenal comparison group may suffer from he problem of selfselecion or oher sysemaic biased relaing o he sample selecion. Under cerain assumpions, he PSM mehod correcs he sample selecion bias by pairing reamen and comparison unis according o observed characerisics. The mehod provides a naural weighing scheme ha ensures he unbiasedness of he esimaed reamen effecs. In his sudy, here are wo reamen effecs of ineres: innovaion effecs and exporing effecs. We ask if innovaors are more likely o become exporers han noninnovaors afer we conrol for heir propensiy o innovae. Similarly, we invesigae he reverse direcion of causaliy by asking if exporers are more likely o be innovaors han non-exporers who exhibi a similar propensiy o become exporer. We idenify non-innovaors which exhibi a similar propensiy o innovae and non-exporers which exhibi a similar propensiy o expor as by esimaing he following innovaion and expor propensiy models (Damijan e al., 2010): and I i 1 f X i 1 i Pr (1) Ei 1 f Z i 1 i Pr. (2) In boh equaions (1) and (2), in each period firm i s propensiies o innovae ( I ) and o expor ( PrE 1 Pr 1 i i ) are expressed as a funcion of observed (exogenous or predeermined) previous period characerisics such as produciviy, size of employmen, capial inensiy and impor saus. Based on he esimaed propensiy o innovae (equaion 1), we mach innovaors and non-innovaors a period. Similarly, based on he esimaed propensiy o expor (equaion 2) we obain a lis of mached exporers and non-exporers. These wo pairs represen our reamen and conrol groups. We invesigae he innovaion effecs on expor propensiy using all kind of innovaion, produc innovaion or process innovaion and for firms in hree separae 11

13 subsecors (Primary, Manufacuring, Services). Specifically, using he lis of mached innovaors in period, we esimae he average reamen effecs of innovaion (ha is, he average reamen of he reaed) on expor marke paricipaion by comparing heir probabiliies o become exporers in period and in period +1 separaely. We expec period +1 s comparison suffer less from he problem of unobserved conemporaneous shocks han he period s comparison. Similarly, we also invesigae he reverse case. Tha is, using he resuling mached exporers in period, we esimae he average reamen effecs of expor marke paricipaion on innovaion by comparing heir probabiliies o become innovaors (produc and/or process) in period and in period +1 separaely. We also conduc he above analyses on a subsample of our daa where we only consider exporers (innovaors) a period who were no exporers (innovaors) in period -1. Tha is, we invesigae an even sronger es of he direcion of causaliy by asking if curren period innovaion (expor marke paricipaion) leads firms o become new exporers (innovaors) in he curren or nex period. However, due o sample size limiaion, we do no esimae he effecs a he subsecor level. (ii) Daa To esimae he models described above, we use firm level daa from he firs confidenialised uni record file (CURF) ediion of he Business Longiudinal Daabase (BLD) produced by he Ausralian Bureau of Saisics. The firs CURF ediion of he BLD ha we use conains daa from wo panels wih a combined sample size of around 3,000 Ausralian small and medium businesses employing 200 or fewer workers each year. The firs panel in he daa conains annual business level daa from he financial years from o The second panel covers he period beween and Overall, he acual number of businesses covered by he BLD panel daa wih useable observaions is 1,826 for , 3,486 for and 3,314 for for a oal of 8,626 firms across he years. The broad secoral disribuion of hese firms by ype of innovaion and he firms expor saus is provided in Table 1. 12

14 TABLE 1 Disribuion of Firms by Secor, Innovaion and Expor Saus (%) Type of innovaion Expor saus Secor Primary (n=2,330) Manufacuring (n=1,324) Services (n=4,972) Produc innovaion only (7.8) Process innovaion only (10.9) Produc and process innovaion (11.3) No innovaion (70.0) Toal (100) Toal n=(8,626) Non-exporer Exporer Suboal Non-exporer Exporer Suboal Non-exporer Exporer Suboal Non-exporer Exporer Suboal Non-exporer Exporer Suboal Noe: Primary secor includes agriculure, fishing & foresry and mining. Services secor includes consrucion, wholesale rade, reail rade, accommodaion, cafes & resaurans, ranspor & sorage, communicaion services, propery & business services, culural & recreaional services, and personal & oher services. Source: Processed from pooled panel daa 2004/05, 2005/06 and 2006/07 of he CURF Business Longiudinal Daabase (ABS, 2009) by he auhor. From Table 1 above, he Services secor appears o have he larges sample size wih 4,972 businesses. However, we believe his reflecs more of he sample design of he BLD daa collecion han he acual disribuion of small and medium businesses in he Ausralian economy. In erms of expor marke paricipaion, of he 8,626 businesses in he full sample, fifeen per cen are exporers. As expeced, he proporion of exporers in he sample varies by secor wih he manufacuring secor having he highes proporion a around weny nine per cen, or double he rae of he each of he oher secors. In erms of innovaion, Table 1 also shows ha overall around hiry per cen of he sampled SMEs have eiher produc or process innovaion (7.8% produc innovaion only, 10.9% process innovaion only, and 11.3% boh produc and process innovaion). As in he case of expor, he proporion of innovaing SMEs also varies across secors. For example, from Table 1 we can infer ha businesses in he 13

15 manufacuring secor have he highes proporion of innovaors wih around fory per cen of he businesses having eiher produc or process innovaion. For esimaion, we clean he daa o make sure ha each observaion has nonmissing values in he relevan variables. The resul is an esimaing sample of around 1,800 firms for each sample year. A descripive summary of he pooled clean sample for he and financial years is provided in Table 2. TABLE 2 Descripive Saisics Variable Descripion =2005/06 =2006/07 PRODINNOV PROCINNOV INNOV EXPORT EMPSIZE 1 =1 if had goods/service innovaion a period =1 if had operaional process innovaion a period =1 if had produc/process innovaion a period =1 if had any expor income a period =number of employees a period LLABPRODVA = log of value added (sales 1 less non-capial purchases) per employee a period LINVINT 1 = log of capial purchase per employee in period -1 N Mean Sd. Dev. N Mean Sd. Dev IMPORT 1 = 1 if had any impor purchase MFG =1 if indusry division is manufacuring SERVICE =1 if indusry division is services Source: Processed from pooled panel daa 2004/05, 2005/06 and 2006/07 of he CURF Business Longiudinal Daabase (ABS, 2009) by he auhor. From Table 2, in , approximaely weny per cen of he sampled SMEs are produc innovaors and weny six per cen are process innovaors. The proporion of hose wih eiher ype of innovaion is approximaely hiry four per cen. These raes are comparable o he hiry per cen innovaion rae in he full sample described earlier and, if we include organisaional processes innovaion, our cleaned sample provides 14

16 similar rae of innovaion produced by he Ausralian Innovaion Survey daa discussed in Secion 2. Also, from he same able, he proporion of manufacuring SMEs is approximaely fifeen per cen. This is abou double he proporion of manufacuring SMEs according o he overall figure for Ausralian SMEs (ABS, 2001), indicaing ha our daa oversample he secor. Finally, in erms of he propensiy o expor, approximaely 15 per cen of he SMEs in our daa repored posiive expor income. This is significanly less han he overall proporion of exporing SMEs discussed earlier. Because of hese sampling biases, paricularly he one he underesimaion of expor propensiy, we need o be cauious in inerpreing he esimaion resuls based on he expor propensiy equaion. IV Resuls (i) Propensiy o Innovae and o Expor Tables 3 and 4 presen he esimaed coefficiens of he propensiy o innovae and o expor based on equaions 1 and 2 respecively. 5 These esimaes are based on pooled sample across all indusrial secors as well as for each of hree major indusrial divisions: manufacuring, services and resources. 6 In general, as shown in Tables 3 and 4, he esimaed coefficiens are saisically significan and of he expeced sign; and in all cases hey are joinly saisically significan. For innovaion, Table 3 shows ha he propensiy o innovae in he curren period is posiively correlaed wih he immediaely preceding period s levels of employmen, labour produciviy and capial inensiy and wheher or no he businesses had any exposure o he impor marke. The posiive relaionships wih size of employmen and labour produciviy appear o be non-linear, wih diminishing effecs. For expor propensiy, Table 4 shows ha only employmen and impor variables are saisically significan. I should be noed however ha our variable consruced using employmen 5 Marginal effec esimaes were obained using STATA s dprobi command. 6 The coefficien esimaes of he propensiy o innovae equaion esimaed a he secor level are omied for space consideraion bu hey are available upon reques. The hree secors are defined according o ANZSIC Version 1993, Primary is A (Agriculure, Foresry & Fishing) and B (Mining), Manufacuring is C (Manufacuring), and Services is E (Consrucion), F (Wholesale Trade), G (Reail Trade), H (Accommodaion, Cafes and Resaurans), I (Transpor and Sorage), J (Communicaion Services), L (Propery and Business Services), P (Culural and Recreaional Services), and Q = Personal and Oher Services. 15

17 size is limied in he sense ha he employmen size figure is only provided in hree discree inervals: 1-5, 5-19, and This migh conribue o a larger sandard error of he esimaes and he saisically non-significan coefficien esimaes. TABLE 3 Propensiy o Innovae All Secors Produc or process innovaion or boh Pr INNOV 1 Produc innovaion Pr PRODINNOV 1 Process Innovaion Pr PROCINNOV 1 Coeff. dy/dx Coeff. dy/dx Coeff. dy/dx EMPSIZE *** *** *** (0.009) (0.001) (0.010) 2 EMPSIZE *** *** *** (0.000) (0.000) (0.000) LLABPRODVA * ** (0.195) (0.226) (0.276) 2 LLABPRODVA ** (0.010) (0.011) (0.014) LINVINT *** * *** (0.019) (0.029) (0.022) IMPORT *** *** *** (0.087) (0.106) (0.104) YEAR 2006 / *** *** *** (0.087) (0.105) (0.097) CONST *** *** *** (0.997) (1.144) (0.339) Pr[Y=1] N. Obs Log pseudolikelihood Pseudo R Noe: (): sandard errors. All probi regressions were esimaed wih 1-digi ANZSIC indusry dummy variables when applicable and wih robus sandard error. ***,**,* indicaes saisically significan a he 1, 5, and 10% level of significance. 16

18 TABLE 4 Propensiy o Expor Pr Expor 1 All secors Primary Manufacuring Services Coeff. dy/ dx Coeff. dy/ dx Coeff. dy/ dx Coeff. EMPSIZE 0.004*** *** (0.002) (0.001) (0.017) (0.001) LLABPRODVA (0.052) (0.035) (0.078) (0.056) LINVINT (0.045) (0.024) (0.055) (0.033) IMPORT *** *** *** *** (0.214) (0.092) (0.173) (0.129) YEAR 2006 / (0.178) (0.105) (0.246) (0.152) CONST *** *** *** (0.510) (0.425) (0.820) (0.619) Pr[Y=1] N. Obs Log-likelihood Pseudo R Noe: (): sandard errors. All probi regressions were esimaed wih 1-digi ANZSIC indusry dummy variables when applicable and wih robus sandard error. ***,**,* indicaes saisically significan a he 1, 5, and 10% level of significance. dy/ dx (ii) The Effecs of Innovaion on Expor Marke Paricipaion Based on he esimaed coefficiens summarized in Tables 3 4 (and Tables A.1 3) and he resuling innovaion propensiy score, each SME which innovaed in period (he reaed firm) is mached o one or more of he non-innovaing firms (he unreaed firms) using he neares neighbour propensiy score maching mehodologies. 7 The resuling 7 In addiion, we also conduc he maching process using he radius maching mehod. The resuls (available upon reques) are roughly similar o he resuls presened in he paper based on he neares neighbour approach; excep, he radius approach ends o produce esimaes wih weaker saisical significance due o fewer mached samples obained. We refer o Imbens (2004) and he cied references here in for an excellen survey of he implicaions of he differen maching approaches. Unforunaely, due o daa access resricions pu on place by he Ausralian Bureau of Saisic which required us o access he daa indirecly via an online STATA do file submission and limied he ypes of STATA commands ha we could issue o only buil in saisical analysis command (programming commands 17

19 summarised in Table 5 below. 8 TABLE 5 maching esimaors for he effecs of innovaion on expor marke paricipaion are Does Innovaing Lead o Expor? Oucome: expor Average reamen effecs on he reaed Treamen: innovaion Innovaion PrExpor Expor ATT SE N ATT SE N ALL SECTORS Innovaion ype Produc 0.168*** (334/334) 0.131** (153/143) Process 0.090*** (399/399) 0.114** (201/200) Produc/process 0.104*** (655/655) 0.116*** (313/305) Innovaion Pr 1 PRIMARY Innovaion ype Produc 0.222*** (45/45) 0.174** (23/24) Process (73/73) 0.144* (42/43) Produc/process (110/110) (60/60) MANUFACTURING Innovaion ype Produc (73/73) (28/22) Process (100/100) (46/37) Produc/process 0.140** (143/143) (61/54) SERVICES Innovaion ype Produc (214/214) (104/106) Process 0.098** (225/225) 0.133* (112/111) Produc/process 0.108*** (397/397) (190/186) Noe: *,**,*** denoe saisically significan a 10%, 5%, and 1% respecively. The sandard errors are compued based on Lechner (2001) approximaion. In he parenheses are he numbers of reaed and mached conrol (possibly no unique) observaions obained using he neares neighbour crierion. In he firs par of Table 5, under he Innovaion Pr Expor heading, he esimaed effecs of curren innovaion on curren expor marke paricipaion are such as he FOR loop command and non-sandard ADO files were no allowed), we were no able o conduc oher more sophisicaed propensiy maching esimaion echniques, such as he kernel densiy based maching or boosrapping for sandard errors. 8 The balancing propery ess (omied due o space consrains bu are available upon reques) suppor he validiy of he resuls of he maching process when all secors are pooled. Despie he relaively weak esimaes of he propensiy models, he resuls of he maching process appear quie reasonable in idenifying valid mached conrol observaions. In oher words, he balancing propery ess appear o be saisfied for almos all neares neighbor maching exercises (similar covariae balance resuls, available upon reques, are also obained for he secor level esimaion). Furhermore, i appears ha he balancing propery of he resuls based on radius maching mehod is weaker compared o ha of he neares neighbor resuls. Similar covariae balance resuls are also obained for he secor level esimaion. 18

20 presened. While he esimaed rae differenials in expor marke paricipaion are based on mached innovaors non-innovaors using previous period condiions, because of heir conemporaneous naure hese esimaes may no ruly indicae direcion of causaliy. There may sill some unobserved confounding facors ha explain he posiive correlaion. The second par of Table 5 provides a clearer indicaion of wheher or no innovaion leads o expor. For example, from Table 5, if we lump all secors ogeher, curren innovaing firms have around 9 17 percenage poins higher propensiy o be in he expor marke. This effec is also significan in magniude given ha, as discussed earlier, he overall proporion of exporing SMEs in our sample is only around 15 per cen. Furhermore, he variaion in he esimaes across innovaion ype and secor suggess ha he relaionship beween innovaion and expor differs across indusrial secors as well as across he differen ypes of innovaion. For example, for he Primary secor, he conemporaneous relaionship beween produc innovaion and expor marke aciviies is he sronges (0.222). In conras, for he Service secor, he relaionship beween curren innovaion and expor is slighly sronger in erms of process innovaion (0.098) han produc innovaion (0.070). To furher invesigae he direcion of causaliy in he relaionship beween innovaion and expor, we esimae he effecs of curren period innovaion on he propensiy o have any expor income in he nex period. The resuls of he esimaion are provided in he second par of Table 5 s columns under he heading Innovaion Pr Expor 1. As can be seen from he able, when all secors are pooled ogeher, here is clear evidence ha Ausralian SMEs innovaive aciviies, eiher produc or process, leads o expor. Unforunaely, he sample size is approximaely halved when he analysis is performed a he secor level, possibly resuling in many of he secor esimaes as saisically insignifican. The esimaes for Manufacuring, for example, show reasonable posiive magniude bu none of which is saisically significan. Neverheless here is indicaion ha process innovaion leads o services expor and ha eiher produc or process innovaion leads o primary produc expor. (iii) The Effecs of Expor Marke Paricipaion on Innovaion 19

21 Table 6 provides a summary of PSM esimaes of he possibe reversed direcion of causaliy running from expor marke paricipaion o innovaive aciviies. Using he esimaed propensiy o expor presened in Table 4, we mach curren exporers ( reaed SMEs) o curren non-exporers ( unreaed SMEs) and esimae he average reamen effecs on he reaed of expor on heir propensiy o have produc innovaion, process innovaion, or a combinaion of boh ypes of innovaion. TABLE 6 Does Exporing Lead o Innovaion Oucome: expor Average reamen effecs on he reaed Treamen: innovaion Expor PrInnovaion Pr Innovaion 1 ATT SE N ATT SE N ALL SECTORS Innovaion ype Produc 0.122** (219/221) (26/24) Process 0.166*** (242/246) 0.178* (104/106) Produc/process 0.129** (299/303) (131/128) Expor PRIMARY Innovaion ype Produc (46/46) (25/24) Process (47/50) (25/24) Produc/process 0.251*** (54/58) (28/28) MANUFACTURING Innovaion ype Produc (70/69) (39/36) Process 0.264*** (91/91) (39/36) Produc/process (104/103) (39/36) SERVICES Innovaion ype Produc 0.225*** (102/102) (46/47) Process 0.279*** (104/104) 0.303** (47/49) Produc/process 0.194** (139/139) (62/66) Noe: *,**,*** denoe saisically significan a 10%, 5%, and 1% respecively. The sandard errors are compued based on Lechner (2001) approximaion. In he parenheses are he numbers of reaed and mached conrol (possibly no unique) observaions obained using he neares neighbor crierion. From Table 6, he esimaed links from expor o innovaion also vary by indusry and by ype of innovaion. The SMEs in he Services secor show he sronges and mos robus posiive relaionship beween expor marke paricipaion and curren innovaion. More imporanly, he second se of columns in Table 6 under he heading of Expor 20

22 Innovaion Pr 1 provide clear evidence ha expor leads o innovaion, bu only for process innovaion and only in Services secor. For oher ypes of innovaion and oher secors, he evidence is no as clear. In oher words, he benefis from exporing, such as he echnology upgrading effecs arising from he exposure o bes pracice echnology such as found by Girma e al. (2003), appear o be raher limied o SMEs in he Services secor. (iv) New Exporers and New Innovaors To ge an even more definiive indicaion of he direcion of causaliy beween expor and innovaion furher, we repea he propensiy maching analysis on a subsample of firms which we can idenify as new exporers or new innovaors. This allows us o invesigae how likely innovaion can lead a business o swich from having been ouside he expor marke before i innovaes o having paricipaed in he expor marke in he nex period. Similarly, we ask how likely expor marke paricipaion can lead a business o swich from being a non-innovaor before i paricipaes in he curren expor marke o becoming an innovaor. Thus, we define new exporers as firms wih no expor income in period -1; similarly, we define new innovaors as firms wihou any innovaion in period The maching esimaors of he average reamen effecs on he reaed are summarised in Table 7. However, due o he limiaion of he resuling sample size, we only perform he analysis a he overall indusry level. From he upper half of he able, i appears ha curren innovaors, especially produc innovaors, which are nonexporers in he previous period, are more likely in becoming an exporer in he curren period compared o curren non-innovaors. On he oher hand, if we look a he probabiliy of becoming a new exporer in period +1, he relaionship is sronges for he process innovaors. 10 Thus, i appears ha he processes ha resul in produc innovaion leading o a swich o become exporers work faser han he processes 9 Ideally, we would wan o condiion all oher preceding periods (-2, -3, and so on) o idenify new exporers more accuraely. However, his is no possible wih our limied daa period. Hence our analysis ress on an assumpion of ha here is enough persisence in he expor marke. 10 I should be noed, however, hese findings are no robus in erms of maching mehod, wih none of he resuls based on he radius maching mehod is saisically significan. Also, oher sudies employing a similar mehodology such as Damijan e al. (2010) are also sensiive o he maching mehods used. 21

23 associaed wih process innovaion leading o expor. To some exen his finding is quie inuiive: when a firm has inroduced a new produc, hen i migh play as a marke leader and eners he expor marke a around he same ime. On he oher hand, having innovaed in producion process is no as srongly relaed o an immediae marke leadership posiion. Oucome: expor TABLE 7 New Exporers and New Innovaors Average reamen effecs on he reaed Innovaion Innovaion Pr NewEXPORTER Pr NewEXPORTE R 1 ATT SE N ATT SE N Innovaion ype Produc 0.054*** (242/242) (114/110) Process (288/288) 0.074*** (148/147) Produc/process 0.027* (490/490) (239/225) Oucome: Expor Expor innovaion Pr NewINNOVATOR PrNewINNOVAT OR 1 ATT SE N ATT SE N Innovaion ype Produc (129/132) (57/59) Process 0.176*** (143/144) (68/71) Produc/process 0.155** (157/162) (76/77) Noe: *,**,*** denoe saisically significan a 10%, 5%, and 1% respecively. The sandard errors are compued based on Lechner (2001) approximaion. In he parenheses are he numbers of reaed and mached conrol (possibly no unique) observaions obained using he neares neighbor crierion. New exporers are defined by condiioning on EXPORT 1 0. New innovaors are defined by condiioning on INNOVATION 1 0. In he lower half of Table 7, we provide he esimaes relevan for addressing he second quesion: is curren expor paricipaion associaed wih a higher probabiliy of becoming a new innovaor in he curren or he nex period? The answer is ha, unlike in he case when innovaion is he reamen discussed above, none of he esimaed relaionships beween curren expor and he propensiy o become new innovaor in period +1 is saisically significan. However, curren expor appears o lead o a swich o becoming a process innovaor in he same period. Alogeher, our neares neighbour esimaes provide differen characerisaion of he link beween innovaion and expor for Ausralian SMEs. For example, for small 22

24 firms like Ausralian SMEs which mos produc innovaion involve producs which are no new o he world and who are more likely o be financially consrained relaive o large firms he ype of innovaive aciviies which appears o maer he mos wih regards o expor marke paricipaion is process innovaion. Also, consisen wih he argumen in Damijan e al. (2010), our resuls indicae ha he posiive effecs of curren produc innovaion on he probabiliy of becoming an exporer in he curren period shown in he upper half of Table 7 appears o suppor earlier findings such as Cassiman and Golovko (2007), Cassiman and Marinez-Ros (2007) and Becker and Egger (2010) ha produc innovaion is crucial for enering he inernaional marke successfully. On he oher hand, he srong posiive relaionship beween curren expor marke aciviy and he probabiliy of becoming a new process innovaor in he curren period shown in lower half of Table 7 appears o be consisen wih he findings ha once in he expor marke, firms need o conduc process innovaion o say compeiive. V Conclusions and Policy Implicaions In his paper we addressed he quesions of wheher or no exporing firms learned from heir paricipaion in he expor markes and hus hey became more innovaive han hose which focused only on he domesic markes (learning-by-exporing hypohesis) and wheher or no firms inroduced innovaion before hey enered he foreign markes (self-selecion hypohesis). Looking a Ausralian small and medium enerprises (SMEs) employing 200 or fewer workers, we assessed if exising evidence mosly based on daa from medium and large firms and firms in he manufacuring secor could be generalised ino he cases of small firms and firms from he primary and services secors uilising he propensiy score maching mehodology. Depending on secor and ype of innovaion, we found innovaion lead o expor and, wih weaker evidence, expor lead o innovaion. For example, perhaps reflecing he srengh of Ausralia s primary (mining and agriculure) secor in he inernaional marke, he primary secor exhibied he sronges saisical evidence ha innovaion lead expor. For he service secor, we found ha only process innovaion lead expor and only in his secor we found saisically significan evidence ha expor lead o process innovaion. 23