Globalization and Knowledge Spillover: International Direct Investment, Exports and Patents*

Transcription

1 Globalzaton and Knowledge Spllover: Internatonal Drect Investment, Exports and Patents* Cha-Ln Chang Department of Appled Economcs Natonal Chung Hsng Unversy Tachung, Tawan Sung-Po Chen Department of Appled Economcs Natonal Chung Hsng Unversy Tachung, Tawan Mchael McAleer Econometrc Instute Erasmus School of Economcs Erasmus Unversy Rotterdam and Tnbergen Instute The Netherlands and Instute of Economc Research Kyoto Unversy Japan EI August 2010 * For fnancal support, the frst author acknowledges the Natonal Scence Councl, Tawan; and the thrd author acknowledges the Australan Research Councl, Natonal Scence Councl, Tawan, and the Japan Socety for the Promoton of Scence. 1

2 Abstract Ths paper examnes the mpact of the three man channels of nternatonal trade on domestc nnovaton, namely outward drect nvestment, nward drect nvestment (IDI) and exports. The number of Tradc patents serves as a proxy for nnovaton. The data set contans countres that are consdered to be hghly competve n the world market, coverng the perod 1994 to The emprcal results show that ncreased exports and outward drect nvestment are able to stmulate an ncrease n patent output. In contrast, IDI exhbs a negatve relatonshp wh domestc patents. The paper shows that the mpact of IDI on domestc nnovaton s characterzed by two forces, and the posve effect of cross-border mergers and acqusons by foregners s less than the negatve effect of the remanng IDI. Keywords: Internatonal drect nvestment, Export, Tradc Patent, Outward Drect Investment, Inward Drect Investment, R&D, negatve bnomal model. JEL: F14, F21, O30, O57. 2

3 1. Introducton Globalzaton means dfferent thngs n dfferent contexts. From an economc perspectve, refers to the cross-border movements of goods, funds, personnel and nformaton. The more easly do such movements take place, the hgher s the degree of globalzaton. In ths context, the movement of goods refers to trade, whle movements of funds refer to nternatonal drect nvestment (Lu et al., 2005). Furthermore, based on the drecton of the flow of capal, foregn drect nvestment can be broken down nto outward drect nvestment (hereafter ODI) and nward drect nvestment (hereafter IDI). Accordng to the World Investment Report by the Uned Natons and the World Development Indcators by the World Bank, n 2008 the amounts of ODI and exports for the world as a whole over the years have generally exhbed an upward trend. They have both accounted for an mportant share of GDP, as shown n Fgures 1, 2 and 3, have played an essental role n the process of globalzaton, and have been regarded as the man channels for technology spllover (Branstetter, 2006; Lu & Zou, 2008). Furthermore, accordng to endogenous growth theory, technologcal nnovaton s mportant to the sustaned growth of an economy (Romer, 1986; Lucas, 1988). The man reasons for the economc growth of the late-ndustralzng economes have been the acquson of knowledge and ntellgence, technologcal nnovaton and human capal accumulaton (Bassann and Scarpetta, 2002; Hu & Mathews, 2005; Mueller, 2006). From the above, can be seen that countres that frequently engage n ODI also attract IDI and promote export actves. Such behavor rases the ssue as to why countres wsh to engage n such actves. For ths reason, we propose the followng questons: are ODI, IDI and exports capable of enhancng a country s nnovaton and 3

4 technologcal depth? Conversely, needs to be asked whether these actves can lower a country s wllngness to engage n nnovaton actves, and over the long term cause that country to lack competveness n nternatonal markets, wh the result that the economy s growth can not be sustaned, so that the country falls nto recesson. Ths paper examnes, whn the context of globalzaton, the mpact of the three man channels of nternatonal trade, namely ODI, IDI and exports, on domestc nnovaton, where the number of patents serves as a proxy for nnovaton. By examnng the mpact between countres, whn the context of globalzaton, of ODI, IDI and exports on patents, s hoped that the results of ths paper can serve as a reference for publc and prvate polcy, and that approprate nternatonal trade polces may be formulated to ncrease a country s nnovaton actves, upgrade s ndustral technology, and ultmately promote economc sustanably and stable growth. The remander of the paper s organzed as follows. Secton II provdes a revew of the lerature on exports and nnovaton, and nternatonal drect nvestment and nnovaton. Secton III presents the varables, data and sample statstcs for the emprcal analyss. Secton IV dscusses the research methods. Secton V ntroduces the emprcal model and analyzes the results. Secton VI provdes the concluson and some suggestons for further research. 2. Exports, Internatonal Drect Investment, and Innovaton Tradonal economc theory on the relatonshp between nnovaton and exports largely focuses on the topc of whether nnovaton nfluences exports. For example, the technologcal gap theory (Posner, 1961) and the lfe-cycle theory (Vernon, 1966) both argue that nnovaton wll gve manufacturers a greater comparatve advantage, so that they wll become net exporters. Moreover, n early studes, the focus was on whether a 4

5 country that s engaged n nnovaton actves can provde a boost to exports (Roper & Love, 2002; Gourlay & Seaton, 2004; Yang et al., 2004; Gourlay et al., 2005; Roper et al., 2006; Tomura, 2007). In contrast, the mpact of exports on nnovaton has not yet led to consstent results (Keller, 2009). In recent years, studes on these two ssues have focused manly on dscussng the mpact of exports on nnovaton. For example, Ln & Yeh (2005) found that the exports of Tawan s electroncs ndustry exhbed a sgnfcantly posve relatonshp wh that ndustry s R&D, but that such a relatonshp only exsted n the case of manufacturers engaged n foregn drect nvestment. For South Korean manufacturers as an example, Han & Lee (2007) used the number of patents approved by the U.S. Patent Offce and the South Korean Intellectual Property Offce (KIPO) as the explaned varables to examne the mpact of the export rato on nnovaton. Ther results ndcated that the proporton of exports only exhbed a sgnfcant and posve relatonshp n regard to those patents approved by the U.S. It was argued that ths dfference arose because the dfferent strateges adopted by manufacturers n applyng for patents eher at home or abroad. The dscusson to date has not taken nto consderaton the possbly that the mpact of exports on nnovaton was subject to a tme lag. In fact, the mpact of exports on nnovaton s lkely to have a deferred effect. For example, Salomon & Shaver (2005) dscovered that for Spansh manufacturers there exsted a sgnfcant and posve relatonshp between the export behavor of Spansh manufacturers, wh a tme lag of one or two perods, and the number of products nnovated n the domestc economy, as well as the number of patent applcatons. Through exports, was possble to acqure knowledge that was lackng n the domestc market, and thereby to promote nnovaton. In other words, the effect of learnng by exportng was found to exst. In order to verfy ths learnng effect, Grma et al. (2008) analyzed manufacturers n the U.K. and Ireland. Ther results showed that Irsh exports were able to ncrease nnovaton actves wh a 5

6 tme lag of one perod, whereas n the U.K. there was no compellng evdence to show that exports could ncrease nnovaton actves. The authors concluded that ths dfference was due to the domestc markets, the szes of these two countres, and the destnatons of ther exports not beng the same. Based on Salvatore s (2007) defnon, nternatonal drect nvestment refers to real nvestment engaged n overseas, and ncludes the acquson and control of factores, capal goods, land, nventory and management. It frequently nvolves eher the settng up of overseas subsdares or purchasng large quantes of shares n order to obtan the rght to operate. Accordng to the drecton n whch the funds for nvestment flow, a dstncton may be made between ODI and IDI. Two man types of result may be nferred from past emprcal studes: n terms of s mpact on nnovaton, nternatonal drect nvestment has (1) posve spllover effects, and (2) negatve or uncertan spllover effects. The posve spllover effects refer to the dscovery of knowledge spllovers, such as technology or management, when engagng n IDI or ODI. Ths leads to an ncrease n nnovaton actves n the host country and/or the home country. For example, Ln & Yeh (2005) nterpret IDI and R&D as beng mutually dependent, so that f the quanty of nput of one ncreases, the amount of expendure on the other s also ncreased. Blnd & Jungmtag (2004) argued that externally-nduced competon has a tranng effect on the host country s domestc market whch, n turn, has a posve effect on the host country s nnovaton actves. However, Ln et al. (2009) adopted a que dfferent approach to analyzng ODI. They argued that ODI benefs the home country s nnovaton. Branstetter (2006) also advanced a smlar vew, and proceeded to consder that nternatonal drect nvestment emboded a feedback effect, so that ODI not only caused the nnovatve behavor of the host country to ncrease, but also led to a posve effect on nnovaton actves n the home country. 6

7 The negatve spllover effect refers to nflows of foregn capal whch result n the host country becomng excessvely dependent on technology, thereby leadng to a reducton n nnovaton actves (Kumar, 1987). However, the uncertan spllover effect refers to foregn capal mpactng the host country s nnovaton actves both posvely and negatvely. Frst, dfferent measurement ndcators of IDI are used smultaneously, as n Grma et al. (2009), who used the proporton of foregn nvestment and the amount of foregn nvestment sold domestcally as the IDI ndcator. The emprcal evdence showed that the proporton of foregn nvestment has a posve and sgnfcant effect on product nnovaton, whle the amount of foregn nvestment s characterzed by a sgnfcant and negatve relatonshp wh product nnovaton. Second, dfferent studes have focused on dfferent ndustres. For nstance, Deolalkar & Evenson (1989) sought to estmate the patent demand functon for Inda. Ther emprcal results ndcated that the hgher the proporton of foregn nvestment n the chemcal ndustry, the smaller the number of patents. In contrast, n the lght engneerng and other engneerng ndustres, IDI was found to be posvely related to patents. Fnally, varous results have also been obtaned when nternatonal drect nvestment s decomposed nto ODI and IDI. For nstance, Pottelsberghe & Lchtenberg (2001) used a sample of 13 OECD countres to examne whether FDI led to a technology transfer effect, and concluded that transfers of technology across borders as part of FDI should not be consdered n one drecton only. Thus, they decomposed FDI nto ODI and IDI n order to vew capal flows as movng n two drectons. Ther emprcal evdence showed that ODI s a technology spllover channel that has both a sgnfcant and posve effect on the domestc country s total factor productvy. In contrast, IDI dd not help to mprove the technology of the host country, and even adversely affected. The reason for ths was that IDI had a tendency to acqure 7

8 technology from the host country, and then to gve ths technology, whch dd not dssemnate n s own country (the home country), to another country (the host country). In addon, as the proporton of cross-border mergers and acqusons (hereafter cross-border M&As,) n nternatonal drect nvestment has been ncreasng annually (UNCTAD, 2007), n emprcal research the topc of IDI has also been examned together wh that of cross-border M&As. For example, Lu & Zou (2008) found that the sgnfcant and posve relatonshp between cross-border M&As actves n Chna s hgh-technology ndustry and nnovaton only exsted among dfferent ndustres, and that whn a partcular ndustry the relatonshp was posve but not sgnfcant. The reason for ths was that mergers and acqusons actves may ncrease the degree of ndustral concentraton and monopoly power, so that ndustres whn the same domestc sector wll be characterzed by relatvely ltle nnovatve behavor. Usng a sample of 14 OECD member states, Bertrand & Zunga (2006) examned the relatonshp between cross-border M&As by foregners and R&D, and ther emprcal results showed that the overall relatonshp was posve but not sgnfcant. By focusng on the ndustres technology ntensy, the relatonshps between these mergers and acqusons actves and R&D for hgh, medum and low levels of technology were found to be negatve and not sgnfcant, sgnfcant and posve, and posve and not sgnfcant, respectvely. Thus, can be seen that the sgnfcant and posve relatonshp between cross-border M&As by foregners and nnovaton exsts only n the local context. 3. Data Ths paper uses panel data for countres coverng the perod 1994 to

9 There are three crera for selectng the sample, as follows: (1) globalzaton, (2) OECD member states, and (3) upper-mddle or hgh ncome countres. The reason for usng these three crera s that the emphass n ths paper s on globalzaton, so that the most mportant economes on the fve contnents, namely Europe, Asa, Amerca, Oceana and Afrca, are ncluded n the sample. Second, the motvaton for ncludng the OECD countres n the sample s that more than 90% of the world s foregn drect nvestment orgnates n OECD countres (Ou Yang & Hwang, 2006), so that the vast majory of OECD countres are engaged n cross-border drect nvestment actves. Fnally, the reason for selectng upper-mddle or hgh ncome countres s that when relatvely hgh ncome countres are compared wh low ncome countres, the hgher ncome countres wll tend to engage n R&D actves, and wll attach greater mportance to patents, whch are the embodment of ntellectual property rghts. As shown n Table 1, the sample s subdvded nto contnents, wh 22 countres n Europe, eght countres n Asa, four countres n the Amercas, two countres n Oceana, and one country n Afrca. Second, as shown n Table 2, there are 28 OECD member countres and nne countres that are not members of the OECD. Fnally, when countres are categorzed accordng to the World Bank s ncome classfcaton, 28 countres are hgh ncome countres, eght are upper-mddle ncome countres, and one country s a low-mddle ncome country 1. The only low-ncome country s Chna. As Chna has a sgnfcant nfluence on the world economy, and as one of the man countres nto whch foregn capal flows, n dscussng the topc of globalzaton the Chnese economy should be ncorporated nto the sample. Overall, the sample comprses advanced countres, that s, a hghly competve group of countres n the world market that encompasses the world s major economes on all fve contnents. 1 Accordng to the World Bank s ncome classfcaton, ncome can be subdvded nto the followng four levels: hgh, upper-mddle, lower-mddle, and low. 9

10 Although we do not examne each country n the world n detal, the sample serves as a bass for dscussng the topc of globalzaton n the broader sense. In addon, the data for all the varables are obtaned prmarly from three sources: OECD s Man Scence and Technology Indcators (MSTI) for 2007, the Uned Natons World Investment Report (hereafter WIR) publshed n 2008, and the World Bank s World Development Indcators (hereafter WDI), also publshed n Due to the omsson of data on Sngapore s exports (the data were not avalable from 1994 to 2000), the data for Sngapore s exports were obtaned from the Sngapore Trade Development Board. In addon, as Tawan s not ncluded n the above publcatons, the data for Tawan were obtaned from the Drectorate-General of Budget, Accountng and Statstcs, Tawan. Ths paper uses patents as a proxy for nnovaton, as shown n Table 3. In selectng the number of patents, prevous studes have frequently used the number of patent applcatons submted to a specfc patent offce as the nnovaton ndex (Branstetter, 2006; Deolalkar & Evenson, 1989). Alternatvely, they have used the correspondng numbers of patents appled for to the patent offces of two countres to represent ths number, and thereby faclate a comparson (Han & Lee, 2007). In ths paper, s argued that comparng the dfferences n nnovaton output between countres wll lead to bas, due to the host country s home advantage, f only the applcatons for patents to a sngle patent offce are used. When an nventor apples for a patent, as compared wh applyng to a patent offce n another country, they are more lkely to prefer to apply to ther own country s patent offce for a patent. For ths reason, we use a tradc patent that s based on smultaneous applcatons for approval to patent offces n Europe, USA and Japan as our ndcator. In ths way, we can reduce the bas that s generated due to the host country s advantage. In addon, as a patent s the outcome of nnovaton, f we can presume that a 10

11 hgher economc value of an nnovaton s mpled by a tradc patent, the greater wll the patent be able to reflect economc growth. In addon, ths paper uses patents appled for n one drecton to the European patent offce, that s, unlateral patents, so that we can further compare whether dfferences n the numbers of patents are sgnfcant n relaton to the coeffcents of the explanatory varables. Exports and nternatonal drect nvestment constute the man channel of technology spllover among countres (Branstetter, 2006; Lu & Zou, 2008). Of these, nternatonal drect nvestment s bdrectonal, and can be dvded nto ODI and IDI (Pottelsberghe & Lchtenberg, 2001). Meanwhle, based on the defnon provded n the WIR for 2008, Foregn Drect Investment comprses three parts, namely, equy capal, 2 renvested earnngs, 3 and ntra-company loans. 4 Thus, we further decompose IDI nto two parts, namely cross-border M&As by foregners, and other drect nvestment. In addon, R&D actves requre nputs over a long perod of tme to produce results, the nputs n the current perod wll be separated from the benefs not yet seen by a tme lag extendng to future perods (Tsou & Lu, 1997). For ths reason, the lag of R&D s taken nto consderaton (Han & Lee, 2007). The above explanatory varables are shown n Table 3. As the sample ncludes data for countres, to remove dfferences n the amounts expended on R&D varyng from country to country, all of the explanatory varables are dvded by ther own-country GDP. From Table 4, whch gves the descrptve statstcs for the emprcal varables, three phenomena may be observed. Frst, regardless of whether the tradc patent 2 Ths refers to shares of companes bought by foregn drect nvestors outsde the countres n whch they resde. 3 Ths refers to the porton of the surplus on the drect nvestors nvestments that s retaned and used for renvestment. 4 Ths refers to the loans n the form of short-term or long-term funds between drect nvestors and afflated companes. 11

12 (TRI_PATENT) or the unlateral patent (EPO_PATENT) s used, the respectve standard devatons are both twce as large as ther means. From ths, we see that the patent nformaton s characterzed by over-dsperson. Second, by addng the means of total exports (EXP) and ODI, ther combned share of GDP exceeds 40 percent ( = 0.415). In other words, exports and ODI together account for a hgh proporton of nternatonal trade. Thrd, wh cross-border M&As by foregners (M & A_SALE) accountng for more than one-half of IDI ( / = 0.54), s clear that cross-border M&As are the man component of nternatonal drect nvestment. 4. Models 4.1 Negatve bnomal model As a patent s a non-negatve dscrete varable, ths paper uses the count data model. The two types of count data model commonly used are the Posson model and the Negatve bnomal model. The Posson model s probably densy functon s gven n equaton (1), where y s the number of patents n country n year t, and s the average number of patents n country n year t, namely the un frequency of patent applcatons. In ths model, the mean and varance are equal, as shown n equaton (2). However, n emprcal research patent data are often characterzed by over dsperson (Aggarwal, 2004), that s, the varance s greater than the mean. Thus, the Posson model may be napproprate, so that the negatve bnomal model s commonly used to resolve the shortcomngs of the Posson model: Prob y λ e λ y! y y 0,1,2,..., n (1) E y Vary (2) 12

13 Accordng to Hausman et al. (1984), the negatve bnomal model has an ndvdual and unobserved effect, whch generalzes the Posson model. It assumes that the Posson model parameter,, conforms to a Gamma, dstrbuton, where does not change across countres or over tme. The basc negatve bnomal model s gven n equaton (3) (for a detaled dervaton, see Hausman et al. (1984)): Prob -λ y y e λ f λ dλ Γ 0 1 y! Γγ y γ Γy 1 δ 1 δ γ 1 δ y (3) The negatve bnomal model relaxes the assumpton n the Posson model that the mean and varance are equal, so that allows the number of patents to be characterzed by over dsperson, as n (4): Var E y y y Var y E 1 2 δ δ 1 1 δ (4) As ths paper uses panel data, we can use the fxed effects and random effects models, each of whch s explaned below. 4.2 Fxed effects negatve bnomal model Frst, we confgure the model parameters, and, as shown n equatons (5)-(8) below: βx μ γ α e e (5) 13

14 (6) e γ βx E(y ) e (7) e (8) where y s the number of patents n country n year t, s the expected value of y, X denotes the explanatory varables, s a parameter to be estmated, and s the fxed effect of an ndvdual country that does not change over tme. Under the condons of the sum of the patents, t densy functon of y y..., y, 1 y, the condonal probably s gven n equaton (9). From equaton (10), can be seen that the varance s greater than the mean, so that ths model allows the explaned varable to be characterzed by over dsperson: Pr Γ Γ y 1 t t Γ γ Γ y 1 Γ y t t t t Γ γ y y y (9) Var E e e y 1 y e X X y Var y E e 1 Var y Ey (10) Fnally, we can derve the lkelhood functon, as shown n equaton (11). After γ e X s substuted, maxmum lkelhood estmaton can be used to obtan estmates of the parameters: 14

15 ln L ln ln y 1 Γ t t t t t ln Γ γ y ln Γγ ln Γy 1 y (11) 4.3 Random effects negatve bnomal model The steps for nferrng the random effects model are essentally the same as those for the fxed effects model dscussed above. It s necessary to confgure the parameters, and, as shown n equatons (12)-(15). The dfference from the fxed effects model s that the random effects model assumes that s randomly dstrbuted, n whch case the jont probably densy functon for y y, y,... 1 and s gven n equaton (16) below: e X e (12) (13) e γ X Ε( y ) e (14) e (15) Pr( y, δ ) Pr( y ) g ( δ ) (16) In order to obtan the probably densy functon of y, s necessary to use ntegraton to remove from the jont probably densy functon. For ths reason, s necessary to select an approprate dstrbuton for, as shown n equaton (17), where z conforms to a Beta ( a, b ) dstrbuton. The probably densy functon s gven n equaton (18): 15

16 z 1 1 e 1 (17) 1 a1 b1 B( a, b) z (1 z) f ( z ) (18) Through substuton of the above condons and usng ntegraton, we can obtan the probably densy functon of y, as shown n equaton (19), and obtan s lkelhood functon, as n equaton (20). After γ e X s substuted, by usng maxmum lkelhood estmaton, we can obtan the estmates of the parameters: Pr y 1 T 0 t 1 y y 1 z y 1 z f z dz a ba b y aba b y t Γ Γγ y γ Γ y 1 (19) ln L ln a b lna lnb y lna lnb a b y ln Γγ y ln Γγ ln Γy 1 ln t (20) The basc model n ths paper examnes the mpact of exports, ODI, IDI and R&D expendure, usng a tme lag of one perod on tradc patents, wh the emprcal model gven n equaton (21). In (21), represents the expected value of the tradc patent, EXP denotes exports, ODI s outward drect nvestment, IDI s nward drect nvestment, L 1 _ GERD s domestc R&D expendure, wh a tme lag of one perod, s the error term, and 1, 2, 3 and 6 explanatory varables: ( are the parameters assocated wh the γ exp EXP ODI IDI L1_ GERD ) (21)

17 As the proporton of cross-border M&As n nternatonal drect nvestment ncreases annually (see UNCTAD, 2007), we decompose nward drect nvestment (IDI) nto cross-border M&As by foregners and other drect nvestment, wh a vew to examnng the mpact of these two forces on tradc patents. The emprcal model s gven n equaton (22), where MA _ SALE denotes cross-border M&As by foregners, PRIVATE s other drect nvestment,, EXP, ODI, L 1 _ GERD and s as defned above, and 1, 2, 4, 5 and 6 are the parameters: γ exp ( EXP 2 ODI 4 MA_ SALE 5 PRIVATE 6 L1_ GERD ) 1 (22) In order to make the emprcal results reflect more accurately the source of most R&D expendure, we change the data on R&D expendure from total domestc R&D expendure (GERD) to R&D expendure for the domestc busness sector (BERD). The emprcal model s gven n equaton (23), where L 1 _ BERD represents R&D expendure n the domestc busness sector, wh a tme lag of one perod. The defnons of the rest of the varables are as descrbed above, and 1, 2, 3, 4 and 5 are the parameters: γ exp ( EXP 2 ODI 4 MA_ SALE 5 PRIVATE 7 L1_ BERD ) 1 (23) Fnally, the patent data are changed from tradc patents to unlateral patents, and the emprcal model s gven n equaton (24), where represents the expected values of the unlateral patents, the defnons of the remanng varables are as descrbed above, and 1, 2, 4, 5 and 6 are the parameters: exp ( EXP 2 ODI 4 MA_ SALE 5 PRIVATE 6 L1_ GERD (24) ) 1 17

18 5. Emprcal Results The basc model used n the emprcal analyss examnes the mpact of four varables, namely exports, ODI, IDI and R&D expendure on patents. In order to mantan consstency, s necessary to determne the number of perods for whch R&D expendure s deferred n order to establsh the basc model. In order to enhance effcency n estmaton, we use bootstrappng methods to estmate the varances. Tables 5-10 report the t-values both wh and whout bootstrappng. Table 5 reports the results of determnng the number of perods by whch the R&D expendure should be deferred usng the negatve bnomal model based on both fxed and random effects. The emprcal results ndcate that the nfluence of R&D expendure deferred one perod mproves explanatory power. The fndng that the mpact of the R&D nput on patents has a one-perod lag effect s consstent wh that of Tsou & Lu (1997). For ths reason, n the subsequent dscusson, R&D expendure deferred one perod (L1_GERD) wll serve as the R&D expendure varable, such that columns (2) and (6) n Table 5 wll be the basc model. In Table 6, we test the model usng the Hausman test, wh the null hypothess as the random effects model, and the alternatve hypothess as the fxed effects model. As the test does not reject the null hypothess, the subsequent analyss s explaned usng the random effects model. 5 From the basc model, we can draw the followng four conclusons: () Exports (EXP) exhb a sgnfcant posve relatonshp wh patents at the 1% level. Ths result explans the strong competon facng world markets. For a country s exporters to gan a foothold n nternatonal markets, s necessary to mprove the qualy of ther exports. For ths reason, they have 5 Although ths paper dscusses the random effects model, each of the tables also lsts the results for the fxed effects model. 18

19 an ncentve to engage n R&D, and to apply for patents to protect ther nnovatons, thereby enhancng ther export competveness. () Outward drect nvestment (ODI) also exhbs a sgnfcant and posve relatonshp wh patents at the 1% level, whch ndcates that a country that s engaged n ODI s able to access knowledge, technology and other addonal products from the host country, mport ths to the home country, to engage n nnovatve R&D to enhance the level of technology, and n turn apply for a patent. In contrast, nward drect nvestment (IDI) by foregners exhbs a negatve, but nsgnfcant, relatonshp wh the home country s patents. Ths ndcates that nflows of foregn nvestment not only do not posvely benef the nnovaton n the home country, but negatvely mpacts nstead. The results of the mpact of the two-way drect nvestment (ODI and IDI) effect on nnovaton are consstent wh the concluson reached by Pottelsberghe & Lchtenberg (2001). () Domestc R&D expendure wh a lag of one perod (L1_GERD) exhbs a posve relatonshp wh the patent at the 10% level. The results suggest that further dscusson on ths ssue s requred, as R&D expendure and patents are nnovatve nputs and outputs, and hence should be characterzed by a hghly sgnfcant relatonshp. For ths reason, we dscuss ths ssue at greater length below. (v) By comparng three behavoural coeffcents, namely exports, ODI and R&D expendure, that can be determned n the home country, s found that the mpacts of all three coeffcents on patents, from the largest to the smallest, are as follows: R&D expendure (10.364), exports (1.476), and ODI (1.411). Table 7 presents the emprcal results for equaton (22), wheren IDI s 19

20 decomposed nto cross-border M&As, and other drect nvestment by foregners, and the mpact of each on patents s then tested. The emprcal results are presented n Table 7, and there are two conclusons: () the relatonshps and sgnfcance between exports, ODI, R&D expendure wh a lag of one perod and patents are all consstent wh the results of the basc model; () cross-border M&As by foregners exhb a posve, but not sgnfcant, relatonshp wh the patents. Ths result s consstent wh that of Bertrand & Zunga (2006), who use 14 OECD member countres as ther sample. Other drect nvestment has a negatve relatonshp wh patents at the 10% level. From these two mpacts on patents, one negatve and one posve, we can ndrectly explan why IDI exhbs a negatve but nsgnfcant relatonshp on patents. The reason s that the negatve effect of other drect nvestment on patents s stronger than the posve effect on patents of cross-border M&As by foregners. In other words, nflows of foregn capal, n general, are of ltle or no benef to domestc nnovaton. However, f these nflows are decomposed nto two parts, namely, cross-border M&As by foregners and other drect nvestment, then s not the case that they have no effect on domestc nnovaton, but rather that there s only a lmed posve effect on nnovaton. The results of the basc model show that R&D expendure does not have a sgnfcant mpact on patents, and so ths secton examnes the nnovaton nput and output sdes, as represented by R&D expendure and patent data. Frst, we change the R&D expendure data from the overall domestc R&D expendure (GERD) used n the basc model nto the domestc busness sector R&D expendure (BERD), and examne whether ths replacement s able to change the sgnfcance of R&D expendure on patents. By comparng columns (2) and (4) n Table 8, can be seen that, n the random 20

21 effects model, R&D expendure wh a lag of one perod s stll not sgnfcant for patents, but the estmated coeffcents are sgnfcantly dfferent from each other. 6 Second, the patent data based on tradc patents (TRI_PATENT) are replaced wh data based on unlateral patents (EPO_PATENT). From columns (6) and (8) n Table 9, can be seen from the random effects model that the sgnfcance of the three explanatory varables, namely R&D expendure wh a lag of one perod, other drect nvestment and cross-border M&As by foregners, markedly ncreases for unlateral patents when compared wh that for tradc patents. For the frst two varables, ths sgnfcance ncreases from the 10% to the 1% level, whle n the case of cross-border M&As by foregners, ths sgnfcance ncreases less markedly to the 5% level. Thus, can be seen that the dfferences n the patent data dffer markedly n relaton to the sgnfcance of the coeffcents of the explanatory varables. In other words, when comparng the results of nnovaton across countres, the selecton of patents s mportant. Takng the present paper as an example, because the sample encompasses fve contnents, f unlateral patents are used as the nnovaton ndcator, such a choce s clearly not objectve and easly leads to bas. Thus, s suggested that usng the tradc patent as a proxy for nnovaton s more approprate. The emprcal results above have shown that nflows of foregn capal are of ltle or no benef to domestc nnovaton, but nstead lead to a negatve mpact. In contrast, by engagng n autonomous behavour through exports, ODI and R&D, s possble to promote nnovaton actves domestcally. In order that excessve relance and expectatons are not placed on nflows of foregn capal, the best polcy for the promoton of nnovaton s to mantan a frm grasp on the domestc country s affars. It s only n ths way that the level of technology can be enhanced, techncal standards 6 Based on the random effects model, the L1_GERD coeffcent has a value of , and the L1_BERD coeffcent a value of

22 upgraded, and ultmately the promoton of economc growth sustaned. The reason why nventors apply for patents s to protect ther nnovatons. However, behnd the results of nnovaton, there s usually a perceved economc value. The hgher s ths economc value, the greater s the ncentve for these nventors to apply smultaneously for patents n dfferent countres. From the defnons of the varables descrbed above, tradc patents represent smultaneous applcatons for patents n Europe, USA and Japan, whle unlateral patents refer to applcatons for patents made to the European patent offce. For ths reason, can be assumed that tradc patents are superor to unlateral patents n terms of representng economc benefs. Columns (5) and (7) n Table 9 present the regresson results n relaton to tradc and unlateral patents, and these suggest the followng phenomena. The frst phenomenon s that the contrbuton of ODI to products of low economc value n the home country s greater than that of products of hgh economc value. In ths paper, we propose two possble reasons. Frst, although ODI results n the capal-recevng country obtanng factors such as technology and knowledge that are helpful to the home country s nnovaton, ndvduals are ratonal, as are countres, n general. For ths reason, the host country wll retan a number of key nnovaton factors (products wh relatvely hgh economc value), so that wll not be easy to obtan technology n relaton to ODI that has a correspondngly hgh economc value. Second, the samples used n ths paper are prmarly for developed countres, wh upper-mddle levels of ncome or hgher, and the bds for ODI for the countres sampled are mostly drected towards countres wh levels of technology lower than ther own. Thus, the level of technology that can be obtaned s lmed, so that s not easy to obtan technology wh a hgh economc value. The second phenomenon s that the mpact of IDI on the absolute value of the 22

23 coeffcent of the tradc patent coeffcent s greater than the absolute value of the coeffcent of the unlateral patent. 7 That s, the harm done to the home country s products of hgh economc value s greater than that to s products of low economc value. Ths paper proposes two possble reasons for ths phenomenon. In relaton to foregn nvestment, the nflows of such nvestment nto the domestc country wll not only generate substantal economc profs, but wll also result n some factors beng obtaned that can help domestc nnovaton. The more mportant are these nnovaton factors, the better, that s, the greater s the economc value of the products, the better. For ths reason, the harm done to the domestc country s products of hgh economc value wll be greater. Fnally, n order to verfy the two observed phenomena dscussed above, we change the R&D expendure varable from total domestc R&D expendure (GERD) to domestc busness sector R&D expendure (BERD). We test the dfferences between these R&D expendure varables n order to examne whether we can obtan the same phenomena. The emprcal results presented n columns (5) and (7) n Table 10 suggest that the above two phenomena stll exst, so that, regardless of whether GERD or BERD s used for R&D expendure: (1) the contrbuton of ODI to the domestc country s products of low economc value s greater than that to products of hgh economc value; and (2) the harm caused by IDI to the domestc country s products of hgh economc value s greater than that done to products of low economc value. 6. Conclusons and Recommendatons 7 As IDI has a negatve relatonshp wh the patents, our goal s to compare whch of the two mpacts of IDI on tradc and unlateral patents has the greater nfluence. The exstence of a negatve value wll nfluence the results, and so s necessary to use absolute values. 23

24 In ths paper, we used panel data for countres coverng the perod 1994 to 2005, and performed an emprcal analyss usng a negatve bnomal model. The man purpose was to examne the mpact of the two man technology spllover channels, namely nternatonal drect nvestment (IDI) and exports on nnovaton. IDI was decomposed nto outward drect nvestment (ODI) and nward drect nvestment (IDI) based on the drecton of capal flows, and because of the ncreasng mportance over the years of cross-border mergers and acqusons (cross-border M&As) by foregners. For these reasons, we decomposed nward drect nvestment nto cross-border M&As by foregners and other drect nvestment. In addon, as R&D expendure and patents are nnovaton nputs and outputs, we also ncluded R&D expendure n the model as an explanatory varable. Moreover, among the explaned varables, we used the number of patents as a proxy varable for nnovaton, and also used data on both tradc and unlateral patents. In summary, whn the context of globalzaton, ths paper nvestgated the mpact of the exports, ODI, IDI, cross-border M&As by foregners and R&D expendure on the number of patents for countres that are consdered to be hghly competve n world markets. Based on the emprcal results, we can draw the followng conclusons: 1. Among the man channels of nternatonal trade, exports and ODI exhbed a posve relatonshp wh the domestc country s patents, that s, ncreased exports and ODI are able to stmulate an ncrease n patent output. In contrast, IDI exhbs a negatve relatonshp wh domestc patents. 2. R&D expendure deferred one perod has a sgnfcant and posve mpact on the number of patents, whch explans the deferred nature of the mpact of the R&D nput on the patents. The nput n the current perod wll only exhb a sgnfcant outcome n the followng perod. 3. The respectve mpacts of cross-border M&As by foregners and other drect 24

25 nvestment on patents are found to be posve and negatve, respectvely. Ths fndng explans the mpact of IDI on domestc nnovaton as beng characterzed by two forces, and the negatve effect of other drect nvestment s greater than the posve effect of cross-border M&As by foregners. 4. Dfferences n patent data can lead to substantal dfferences n the estmated results. Fnally, when comparng the results of nnovaton among countres whn the framework of globalzaton, f unlateral patents are used as the explaned varable, the results obtaned are lkely to be much mproved. However, s also lkely that, because of the presence of a host country advantage, the emprcal estmates are lkely to be based. Thus, s necessary to use tradc patents as a proxy for nnovaton as they are a more approprate varable, and also closer to the actual nnovaton results for each country. In an era of globalzaton, although countres are frequently engaged n exchange wh each other, each government has sought to attract foregn nvestment, and has been wary of outflows of capal from the domestc economy. However, the emprcal results ndcate that the competve behavour between countres s smlarly ratonal and ndvdualstc, so that nflows of foregn capal are of ltle or no benef to domestc nnovaton, and nstead can lead to a negatve mpact. In contrast, countres that engage n autonomous behavour related to exports, ODI and R&D nput are then able to encourage nnovaton actves. In other words, there are not excessve expectatons of, or relance on, nflows of foregn funds from abroad, so that nflows of foregn capal may contrbute to a country s GDP n the short term. However, when vewed from a long-term perspectve, the key to sustaned and stable economc growth s stll nnovaton, and the promoton of such nnovaton must stll be grasped. Only n ths way can the level of technology be rased and the sustaned growth of the economy 25

26 ultmately enhanced. 26

27 References Aggarwal, A. (2004), Macro Economc Determnants of Antdumpng: A Comparatve Analyss of Developed and Developng Countres, World Development, 32, pp Bassann, A. and S. Scarpetta (2002), Does human capal matter for growth n OECD countres? A pooled mean-group approach, Economcs Letters, 74, pp Bertrand, O. and P. Zunga (2006). R&D and M&A: Are Cross-border M&A Dfferent? An Investgaton on OECD Countres, Internatonal Journal of Industral Organzaton, 24, pp Blnd, K. and A. Jungmtag (2004). Foregn Drect Investment, Imports and Innovatons n the Servce Industry. Revew of Industral Organzaton, 25, pp Branstetter, L. (2006). Is Foregn Drect Investment a Channel of Knowledge Spllovers? Evdence from Japan s FDI n the Uned States. Journal of Internatonal Economcs, 68, pp Deolalkar, A.B. and R.E. Evenson (1989). Technology Producton and Technology Purchase n Indan Industry: An Econometrc Analyss. The Revew of Economcs and Statstcs, 71, pp European Patent Organsaton (2010), accessed at Grma, S., Y. Gong, and H. Görg (2009). What Determnes Innovaton Actvy n Chnese State-owned Enterprses? The Role of Foregn Drect Investment. World Development,, pp Grma, S., H. Görg, and A. Hanley (2008). R&D and Exportng - A Comparson of Brsh and Irsh Frms. Revew of World Economcs, 144, pp Gourlay, A., and J. Seaton (2004). Explanng the Decson to Export: Evdence from UK Frms. Appled Economcs Letters, 11, pp Gourlay, A., J. Seaton, and J. Suppakjarak (2005). The Determnants of Export Behavour n UK Servce Frms. The Servce Industres Journal, 25, pp Han, Y.J., and W.Y. Lee (2007). The Effects of the Characterstcs of Korean Frms on the Patent Producton Functon. Economcs of Innovaton and New Technology, 16, pp Hausman, J.A., B.H. Hall, and Z. Grlches (1984). Econometrc Models for Count Data 27

28 wh an Applcaton to the Patents-R&D Relatonshp. Econometrca, 52, pp Hu, M. C., and J. A. Mathews (2005). Natonal Innovatve Capacy n East Asa. Research Polcy, 34, pp Keller, W. (2009). Internatonal Trade, Foregn Drect Investment, and Technology Spllovers, NBER Workng Paper Cambrdge, MA: Natonal Bureau of Economc Research. Kumar, N. (1987). Technology Imports and Local Research and Development n Indan Manufacturng. The Developng Economes, 25, pp Ln, H.L., and R.S. Yeh (2005). The Interdependence between FDI and R&D: An Applcaton of an Endogenous Swchng Model to Tawan s Electroncs Industry. Appled Economcs,, pp Ln, H.L., R.S. Yeh, and C.F. Chung (2009). The Effect of Outward Investment to Chna on Domestc R&D: A Two-hurdle Model wh Endogenous ODI. Appled Economcs, 41, pp Lu, B.-J., T.-J. Chen and Y.-H. Weng. (2005). Trade Theory and Polcy. Tawan: Yeh-Yeh Book Gallery. (n Chnese) Lu, X. and H. Zou (2008). The Impact of Greenfeld FDI and Mergers and Acqusons on Innovaton n Chnese Hgh-tech Industres. Journal of World Busness, 43, pp Lucas, R.E. (1988), On the Mechancs of Economc Development, Journal of Monetary Economcs, 22, Mueller P. (2006), Explorng the knowledge flter: How entrepreneurshp and unversy ndustry relatonshps drve economc growth, Research Polcy, 35, OECD. (2007). Man Scence and Technology Indcators. OECD. Ou Yang, H. and J.-T. Hwang. (2006). Internatonal Trade Theory and Polcy. Tawan: San Mn Books. (n Chnese) Posner, M.V. (1961). Internatonal Trade and Techncal Change. Oxford Economc Papers, 13, pp Pottelsberghe, B.P. and F. Lchtenberg (2001), Does Foregn Drect Investment Transfer Technology across Borders? Revew of Economcs and Statstcs, 83, pp Romer, P.M. (1986), Increasng Returns and Long-Run Growth, Journal of Polcal Economy, 94,

29 Roper, S. and J.H. Love (2002). Innovaton and Export Performance: Evdence from the UK and German Manufacturng Plants. Research Polcy, 31, pp Roper, S., J.H. Love, and D.A. Hígon. (2006). The Determnants of Export Performance: Evdence for Manufacturng Plants n Ireland and Northern Ireland. Scottsh Journal of Polcal Economy, 53, pp Salomon, R.M. and J.M. Shaver (2005). Innovaton, Learnng by Exportng: New Insghts from Examnng Frms. Journal of Economcs & Management Strategy, 14, pp Salvatore, D. (2009). Internatonal Economcs. New York: Wley. Tomura, E. (2007). Effects of R&D and Networkng on the Export Decson of Japanese Frms. Research Polcy, 36, pp Tsou, M.-W. and J.-T. Lu. (1997). The Spllover Effect from Foregn Drect Investment: Emprcal Evdence from Tawan Manufacturng Industres. Tawan Economc Revew, 25(2), pp (n Chnese) UNCTAD (2007), World Investment Report: Transnatonal Corporatons, Extractve Industres and Development. New York and Geneva: Uned Natons. Vernon, R. (1966). Internatonal Investment and Internatonal Trade n the Product Cycle. Quarterly Journal of Economcs, 80, pp World Bank (2010), Country Classfcaton by Income, accessed at s Yang, C.H., J.R. Chen, and W.B. Chuang, (2004). Technology and Export Decsons. Small Busness Economcs, 22, pp

30 Table 1. OECD Countres Europe Asa Amerca Oceana Afrca Country Uned Kngdom, Austra, Czech Republc, Denmark, Fnland, France, Germany, Greece, Hungary, Iceland, Ireland, Italy, Sweden, Swzerland, Netherlands, Portugal, Slovak Republc, Span, Norway, Slovena, Poland, Romana Japan, Korea, Sngapore, Israel, Turkey, Chnese Tape, Russan Federaton, Chna USA, Canada, Mexco, Argentna Australa, New Zealand, South Afrca Total Table 2. OECD Countres by Income Income Level OECD Member Non-OECD Member Total Hgh 25 a 3 b 28 Upper-Mddle 3 c 5 d 8 Lower-Mddle 0 1 e 1 Total 28 9 Source: World Bank Notes: a: Uned Kngdom, Austra, Czech Republc, Denmark, Fnland, France, Germany, Greece, Hungary, Iceland, Ireland, Italy, Sweden, Swzerland, Netherlands, Portugal, Slovak Republc, Span, Norway, Japan, Korea, Uned States, Canada, New Zealand, Australa; b: Slovena, Sngapore, Israel; c: Poland, Turkey, Mexco; d: Romana, Tawan, Russan Federaton, Argentna, South Afrca; e: Chna 30

31 Table 3. Varables Dependent Varable TRI_PATENT Tradc Patent a Number of tradc patent famles. (that s, a patent s fled at the EPO, the JPO and s granted by the USPTO. EPO_PATENT European Patent Offce Patent b Number of patent applcatons to the EPO c. Independent Varable EXP Export Rato of Export dvded by GDP ODI Outward Drect Investment Rato of ODI dvded by GDP IDI Inward Drect Investment Rato of IDI dvded by GDP M&A_SALE Inward Cross-Border M&As Rato of Inward Cross-Border M&As dvded by GDP PRIVATE Other Drect Investment Varable IDI mnus Varable M&A_SALE GERD Gross Domestc Expendure on R&D Rato of Gross Domestc Expendure on R&D by GDP BERD Expendure on R&D n the Busness Enterprse Sector Rato of Expendure on R&D n the Busness Enterprse Sector by GDP L0, L1, L2, L3 Three-year tme lag: Current, one year, two year and three year tme lag n sequence Notes: a: The prmary dependent varable, provded wh the ntenton of mprovng nternatonal comparably (the home advantage s suppressed, the values of the patents are more homogeneous) (OECD, 2007). b: The mnor dependent varable that s compared wh the tradc patent. c: The European Patent Offce provdes a unform applcaton procedure for ndvdual nventors and companes seekng patent protecton n up to 40 European countres, and s the executve arm of the European Patent Organzaton. The European Patent Organzaton s an ntergovernmental organzaton that was establshed n 1977 on the bass of the European Patent Conventon (EPC), sgned n Munch n

32 Table 4. Summary Statstcs Varables Mean Std. Error Mn. Max. TRI_PATENT EPO_PATENT EXP ODI IDI M&A_SALE PRIVATE L1_GERD L1_BERD

33 Table 5. Tme Lags of R&D TRI_ Fxed Effects Random Effects PATENT (1) (2) (3) (4) (5) (6) (7) (8) EXP (2.77)*** (3.02)*** (1.98)** (1.76)* (2.99)*** (3.02)*** (1.92)* (1.81)* [9.07]*** [8.57]*** [7.85]*** [6.44]*** [8.73]*** [8.24]*** [7.55]*** [6.16]*** ODI (2.93)*** (2.99)*** (2.57)*** (2.22)** (2.91)*** (3.03)*** (2.55)** (2.29)** [2.95]*** [3.01]*** [2.61]*** [2.30]** [2.99]*** [3.08]*** [2.68]*** [2.]** IDI (-1.21) (-1.60) (-1.42) (-1.43) (-1.26) (-1.62) (-1.44) (-1.44) [-1.92]* [-1.89]* [-1.65]* [-1.49] [-1.96]** [-1.93]* [-1.66]* [-1.49] L0_GERD (1.33) (1.47) [3.83]*** [4.13]*** L1_GERD (1.56) [3.15]*** (1.68)* [3.45]*** L2_GERD (1.08) [2.21]** (1.17) [2.48]** L3_GERD (0.97) [1.78]* (0.96) [2.03]** CONSTANTS (2.73)*** (2.61)*** (2.35)** (1.90)* (2.72)*** (2.62)*** (2.29)** (1.98)** [14.27]*** [15.85]*** [17.31]*** [18.13]*** [14.23]*** [15.80]*** [17.26]*** [18.08]*** Log lkelhood Wald ch2 Prob ch Groups Observatons Note: Bootstrap t-statstcs are n parentheses, and t-statstcs whout bootstrappng are n brackets. ***, ** and * denote sgnfcance at the 1%, 5% and 10% levels, respectvely. 33

34 Table 6. Basc Model TRI_PATENT Fxed Effects Random Effects EXP (3.02)*** [8.57]*** (3.02)*** [8.24]*** ODI 1.7 (2.99)*** [3.01]*** (3.03)*** [3.08]*** IDI (-1.60) [-1.89]* (-1.62) [-1.93]* L1_GERD (1.56) [3.15]*** (1.68)* [3.45]*** CONSTANTS (2.61)*** [15.85]*** (2.62) [15.80]*** Log lkelhood Wald ch Prob ch Groups Observatons Hausman test Note: Bootstrap t-statstcs are n parentheses, and t-statstcs whout bootstrappng are n brackets. *** and * denote sgnfcance at the 1% and 10% levels, respectvely. 34