Georg-August-Universität Göttingen (founded in 1737) Diskussionsbeiträge Documentos de Trabajo Discussion Papers. Nr. 199

Transcription

1 Ibero-Amerka Instut für Wrtschaftsforschung Instuto Ibero-Amercano de Investgacones Económcas Ibero-Amerca Instute for Economc Research (IAI) Georg-August-Unversät Göttngen (founded n 1737) Dskussonsberäge Documentos de Trabajo Dscusson Papers Nr. 199 The Long-Run Relatonshp between Outward FDI and Total Factor Productvy: Evdence for Developng Countres Derk Herzer January 2010 Platz der Göttnger Seben Goettngen Germany Phone: +49-(0) Fax: +49-(0) e-mal: uwa@gwdg.de

2

3 The Long-Run Relatonshp between Outward FDI and Total Factor Productvy: Evdence for Developng Countres Derk Herzer * Johann Wolfgang Goethe-Unversy, Grüneburgplatz 1, Frankfurt am Man, Germany Abstract Ths paper examnes the long-run relatonshp between outward foregn drect nvestment (FDI) and total factor productvy for a sample of 33 developng countres over the perod Usng panel contegraton technques, we fnd that: () outward FDI has, on average, a posve long-run effect on total factor productvy n developng countres, () ncreased factor productvy s both consequence and a cause of ncreased outward FDI, and () there are large dfferences n the long-run effects of outward FDI on total factor productvy across countres. Cross-sectonal regressons ndcate that these cross-country dfferences n the productvy effects of outward FDI are sgnfcantly negatvely related to cross-country dfferences n labor market regulaton, whereas there s no statstcally sgnfcant assocaton between the productvy effects of outward FDI and the level of human capal, the level of fnancal development, or the degree of trade openness n the home country. JEL classfcaton: F21; O11; F23; C23 Keywords: outward FDI; total factor productvy; developng countres, panel contegraton 1. Introducton Foregn drect nvestment (FDI) outflows from developng countres have grown faster n recent decades than those from developed economes. The share of developng countres n total world FDI outflows ncreased more than twenty-fold from approxmately 0.5% n the early 1970s to about 14% n the md-2000s. In 2007, FDI outflows from developng countres reached 253 bllon US$, whch s about three tmes the value of world FDI outflows n The stock of outward FDI by developng country frms actually exceeded $2 trllon n 2007, accountng for about 15% of the world outward FDI stock n that year. 1 * Correspondng author. Tel.: ; fax: E-mal address: DerkHerzer@gmx.de. 1 The fgures are based on data from the UNCTAD FDI database ( 1

4 Despe the enormous ncrease n FDI outflows from developng countres, there s surprsngly ltle evdence on the potental economc consequences of outward FDI for these countres. The emprcal lerature regardng outward FDI conssts manly of frm and ndustry-level studes on the effects of outward FDI on employment, exports, nvestment, and productvy n developed economes, whle only a few such mcro studes are avalable on developng countres. Macroeconomc studes of the overall mpact of outward FDI on the home developng countres have not yet been performed. Ths paper attempts to fll ths gap by examnng the long-run relatonshp n developng countres between outward FDI and total factor productvy as a key to economc growth. Specfcally, we am to answer the followng questons: (1) Does outward FDI reduce or ncrease total factor productvy n developng countres on average over the long run? (2) Is outward FDI an exogenous nfluence on total factor productvy or does the causaly run n both drectons? (3) Are the productvy effects of outward FDI constant or do they dffer across countres? To answer these questons, we apply panel contegraton technques to a sample of 33 developng countres over the perod Panel contegraton estmators are robust under contegraton to a varety of estmaton problems that often plague emprcal work, ncludng omted varables, endogeney, and measurement errors (see, e.g., Banerjee, 1999; Baltag and Kao, 2000; Pedron, 2007). Moreover, panel contegraton methods can be mplemented wh shorter data spans than ther tme-seres counterparts. Our man results are: (1) outward FDI has, on average, a posve long-run effect on total factor productvy n developng countres, (2) ncreased factor productvy s both consequence and a cause of ncreased outward FDI, and (3) there are large dfferences n the long-run effects of outward FDI on total factor productvy across countres. Gven ths latter fndng, a further queston arses: (4) How can the cross-country dfferences n the long-run productvy effects be explaned? As an addonal contrbuton, we attempt to answer ths queston by examnng whether the observed cross-country dfferences n the long-run effects of outward FDI are lnked to dfferences n country characterstcs, such as the levels of human capal, fnancal market development, trade openness, and labor market regulaton. Usng cross-sectonal analyss, we fnd that the crosscountry dfferences n the productvy effects of outward FDI are negatvely and sgnfcantly related to cross-country dfferences n labor market regulaton, whereas there s no statstcally sgnfcant assocaton between the productvy effects of outward FDI and the level of human capal, the level of fnancal development, and the degree of trade openness n the home country. 2

5 The remander of ths paper s composed of four sectons. In Secton 2, we dscuss the theoretcal background and related emprcal lerature. Secton 3 sets out the basc emprcal model and descrbes the data. The econometrc mplementaton and the estmaton results are presented n Secton 4, whle Secton 5 concludes. 2. Theoretcal background and related emprcal lerature 2.1. Theoretcal background As a background for the dscusson of the potental theoretcal effects of outward FDI on domestc productvy n developng countres, s useful to brefly descrbe the characterstcs of outward FDI from these countres and to contrast them wh the pattern of outward nvestment from developed economes. Table 1 Outward FDI from developng countres by sector and destnaton, percentage shares Outward FDI stock by sector, 2006 a Developng countres Developed countres Prmary Manufacturng Servces Unspecfed Outward FDI flows by destnaton, b Destnaton Developng countres Developed countres Developng countres c Developed countres Notes: a Source: UNCTAD (2008), own calculatons; b averages over the perod , source: UNCTAD (2006), own calculatons; c FDI from developng and transon economes excludng offshore fnancal centers to developng and transon economes A strkng feature of outward FDI from developng countres s the domnance of the servce sector, as shown n Table 1. In 2006, 83.7% of the stock of ther outward FDI was n servces, such as trade, fnance, and busness actves (see, e.g., UNCTAD, 2006); n developed countres, the correspondng share was 61.7%. The second largest sector s manufacturng, whch accounted n 2006 for 9.4% of the outward FDI stock of developng countres and 28.5% of the outward FDI stock of developed countres. Outward FDI n the prmary sector, n contrast, s of mnor mportance. In 2006, only 3.4% of the stock of outward FDI from developng countres was n ths 3

6 sector, whle the correspondng fgure for developed-country outward FDI was 7.9%. Perhaps most strkng, however, s that developng countres heavly nvest n other developng countres, mostly from the same regon (see, e.g., UNCTAD, 2006). In the perod , approxmately 75% of FDI outflows from developng countres went to the developng world. The oppose s true for developed countres that tend to predomnantly drect ther nvestment towards other developed countres. Fnally, s also worth mentonng that, although the number of developng-country multnatonals on the top-100 lst has ncreased n recent years, most of them are relatvely small compared to multnatonals from developed countres (see, e.g., UNCTAD, 2006). Wh ths background, we begn our dscusson of the potental productvy effects of outward FDI n developng countres by consderng possble nteractons between domestc and foregn actves of multnatonal frms. To represent these nteractons, suppose that the producton functon of a multnatonal frm s gven by Q f D, F,, ), where Q s the representatve ( d f frm s worldwde output, D s domestc nput, F s foregn nput, nfluence domestc producton (such as domestc productvy), and d s a vector of factors that f represents factors that nfluence foregn producton. Assumng that domestc nput producton s a functon of domestc capal, D(K d ), and that foregn nput producton s a functon of foregn capal, F(K f ), the frstorder condon that characterzes the frm s prof-maxmzng choce of domestc nputs s Q [ D( Kd ), F( K f ), d, f ], (1) D( K ) d where s total nput cost the frm s cost of capal. From Equaton (1) can be seen that domestc and foregn producton (or nvestment) of the multnatonal frm can be related eher through the cost of capal, and thus through the fnancal sde of the frm, f s somehow a functon of F, or through the producton process, f Q[ D( K ), F( K ),, ]/ D( K ) F( K ) s nonzero (see also Desa et al., 2005). Interactons between foregn and domestc actves operatng through the fnancal sde of the frm occur n a suaton where fxed nvestments n dfferent locatons compete for funds due to costly external fnancng, as dscussed n more detal by Stevens and Lpsey (1992). In such a scenaro, the decson to nvest scarce resources abroad nevably reduces the lkelhood of concurrent nvestments at home, mplyng that each dollar of outward FDI dsplaces a dollar of domestc nvestment (see also, e.g., Feldsten, 1994; Desa et al., 2005; Herzer and Schrooten, 2008). Ths substuton of domestc for foregn nvestment, n turn, s lkely to also reduce domestc productvy. In partcular, when the nvestments abroad come at the expense of 2 d f d f d f 4

7 nvestments necessary to sustan productvy at home (such as new machnery, worker tranng, and research and development (R&D)), outward FDI may reduce the domestc productvy of the nvestng frm n the long run. Some studes, however, suggest that the suaton of fxed resources appears to be rather atypcal for multnatonal frms, at least for developed-country multnatonals. Desa et al. (2004), for example, analyze how US multnatonals capalze afflates around the world and fnd that US multnatonal afflates substute nternal borrowng for costly external fnance stemmng from adverse capal market condons. Smlarly, Desa et al. (2008) show that US parents provde afflates wh addonal equy to fnance profable nvestment opportunes durng currency crses. However, these fndngs cannot necessarly be generalzed to developng-country multnatonals. Snce multnatonals from developng countres are relatvely small compared to ther developed-country counterparts, they often do not have nternal capal markets and must therefore rely on external fnancng to fnance ther nvestment projects. In addon, domestc fnancal markets are undeveloped n many developng countres. Thus, although some developngcountry multnatonal companes have access to foregn capal markets (see, e.g., UNCTAD, 2006), they are generally more lkely to face fnancal constrants than developed-country multnatonals. The concluson from ths s that s also more lkely for developng-country than for developedcountry multnatonals that outward FDI leads to a reducton n domestc productvy due to potental nteractons between domestc and foregn actves through the fnancal sde. The fnancal sde s not the only source of possble nteractons between domestc and foregn actves of multnatonal frms, as shown above. A second and perhaps more mportant source of nterdependence s the producton process. Because of producton nterdependence, outward FDI can affect domestc productvy n several ways, each of whch depends on the multnatonal frm s nvestment motve and the respectve nvestment type. In the followng, we dstngush four key types of nvestment: horzontal FDI, vertcal FDI, strategc asset-seekng FDI and resource-seekng FDI. Horzontal or market-seekng FDI s motvated by market access and avodance of trade frctons such as transport costs and mport protecton n the host country (for models of horzontal FDI see, e.g., Markusen, 1984; Horstmann and Markusen, 1987; Markusen and Venables, 1998). The decson to engage n horzontal FDI s guded by the proxmy-concentraton tradeoff n whch proxmy to the host market avods trade costs but ncurs the added fxed cost of buldng a second producton facly. FDI of ths type thus occurs when a frm decdes to serve foregn markets through local producton, rather than exports, and hence to produce the same product or servce n multple countres. Consequently, horzontal FDI may substute for exports of the goods that were 5

8 prevously produced n the nvestor s home country. Ths decrease n domestc export producton, n turn, may be accompaned by a decrease n domestc productvy, snce export ntensy and frm productvy may be lnked, as some studes suggest (see, e.g., Castellan, 2002; Baldwn and Gu, 2003). However, such effects occur only f the produced good s tradable. As dscussed above, the overwhelmng majory of outward FDI from developng countres s n servces, most of whch are non tradable, mplyng that the bulk of horzontal developng-country FDI cannot substute for home-country exports. But even those horzontal nvestments that are drected to tradable sectors do not necessarly reduce domestc exports and productvy. The reason s because there s rarely a pure case of horzontal producton n the sense that there s nevably some vertcal component to a frm, horzontal FDI can boost exports of ntermedate goods and servces from the home to the host country. For example, headquarters n the home country provde specalzed servces to foregn afflates (such as R&D, desgn, marketng, fnance, strategc management) even f the same fnal goods are produced n both the home and foregn country (see, e.g., Kokko, 2006). Thus, n general terms, multnatonal frms combne home producton wh foregn producton to ncrease ther productvy and hence competveness both nternatonally and domestcally (see, e.g., Herzer, 2008; Desa et al., 2009). Furthermore, n the long run, horzontal FDI may allow the frm to rase s competveness through access to new markets or successful penetraton of exstng markets, thereby addonally ncreasng domestc productvy. The horzontal motve s the most mportant for outward FDI from developng countres, followed by the vertcal motve (see, e.g., UNCTAD, 2006). Vertcal or effcency-seekng FDI s drven by nternatonal factor prce dfferences (for models of vertcal FDI see, e.g., Helpman, 1984; Helpman and Krugman, 1985). It takes place when a frm fragments s producton process nternatonally, locatng each stage of producton n the country where can be done at the lowest cost. Such relocatons reduce domestc producton, at least n the short run (as wh horzontal FDI). However, n the longer run, vertcal nvestment may allow the frm to mport cheaper ntermedate nputs from foregn afflates and/or to produce a greater volume of fnal goods abroad at lower cost, thereby stmulatng exports of ntermedate goods used by foregn afflates (see, e.g., Herzer, 2008). The new structure of the producton chan may thus be assocated wh ncreased effcency, and, as a result, the frm may be able to mprove s competve poson, thus rasng s domestc productvy over the long run (see, e.g., Kokko, 2006). On the other hand, f the frm s not able to adjust over the longer term to the reducton n domestc producton by falng to rase s competveness (e.g., due to labor market rgdes), both vertcal and horzontal FDI wll substute foregn actves for domestc actves over the long run, whch may also lead to a longterm decrease n domestc productvy (see, e.g., Bzer and Görg, 2009). 6

9 The thrd most mportant motve for outward FDI from developng countres s the strategc asset-seekng motve (see, e.g., UNCTAD, 2006). As the name mples, frms undertake strategc asset-seekng FDI to acqure assets that are not avalable n ther own country. Strategc assetseekng FDI s made, for example, when nvestors attempt to gan access to nternatonally recognzed brand names and local dstrbuton networks n order to strengthen ther nternatonal competve poson. Strategc asset-seekng FDI also occurs n the form of technology-sourcng FDI when frms attempt to gan access to foregn technology by eher purchasng foregn frms or establshng R&D facles n foregn centers of excellence (for models of technology-sourcng FDI, see, e.g., Neven and Sots, 1996; Fosfur and Motta, 1999; Bjorvatn and Eckel, 2006). If foregn afflates then acqure new knowledge n terms of technologcal know-how, management technques, knowledge of consumer tastes, etc., ths knowledge can be transferred back to the parent company, thus ncreasng domestc productvy n the long term (see, e.g. Van Pottelsberghe and Lchtenberg, 2001). However, snce the ably to absorb knowledge from abroad depends on the absorptve capacy of the nvestng frm, frms wh low levels of technologcal capacy are lkely to be unable to effectvely access and explo foregn knowledge through outward FDI. The potental productvy gans for outward nvestors from developng countres may therefore be smaller than s the case for ther developed-country counterparts. Another consderaton related to knowledge spllovers through outward FDI s that a substantal percentage of outward FDI from developng markets goes to other developng countres (as shown n Table 1). Ths South-South outward FDI may not generate sgnfcant knowledge spllovers because the bulk of FDI s not located n clusters of specfc technologcal expertse. On the other hand, narrower technologcal gaps between home and host frms may faclate absorpton of technologcal knowledge, mplyng that South-South FDI may generate more spllovers than South-North FDI (see, e.g., Moran, 2008). In addon, developng-country multnatonals have a greater propensy to establsh lnkages wh local frms than do ther counterparts from developed countres (see, e.g., UNCTAD, 2006), whch n turn enables them to more deeply ntegrate nto the host economes, and ths deeper ntegraton could be partcularly benefcal n terms of reverse knowledge and technology flows back to the home country (see, e.g., Globerman and Shapro, 2008). Fnally, the least mportant, yet stll sgnfcant, motve for nvestors from developng countres s the resource-seekng motve (see, e.g., UNCTAD, 2006). Resource-seekng FDI occurs when frms dentfy specfc host country locatons as attractve source of natural resources at the lowest cost. Such FDI s usually assocated wh exports of resource-based products from the host country and should mprove the productvy of domestc producton whch uses the mported resources as low cost, hgh qualy nputs. 7

10 An mportant pont s that outward FDI may not only affect the productvy of the nvestng frms, but also that of the economy as a whole through productvy spllovers to local frms (see, e.g., Blomström and Kokko, 1998). For example, local frms may mprove ther productvy by copyng technologes used by domestc multnatonals, or domestc producers may benef from the knowledge and expertse of the outward-nvestng frms through labor turnover. Moreover, the ncreased competon between nternatonal frms and ther domestcally orented counterparts may force the latter to use ther exstng resources more effcently. Outward-nvestng frms, due to the ncreased productvy, may be also able to provde hgher qualy nputs at lower prces to local producers. In addon, f outward FDI allows the nvestng frms to grow larger than would be possble wh producton n just one country, both the nvestng companes and ther local supplers may benef from economes of scale. Outward FDI may thus enable domestc supplers to move down ther learnng curves and, therefore, to realze substantal productvy gans. Snce, however, FDI may act as an mportant vehcle for the transfer of technologcal and manageral know-how, s lkely to ncrease the competveness of the host economy as well. Ths may lead to reductons n domestc output and productvy when domestc consumers prefer the foregn competors. Furthermore, the ncreased competveness may allow domestc frms n the host country to challenge the foregn frms and thereby to capture market shares from the foregn afflates of the home country s multnatonals. Outward FDI may therefore enable competors n the host country to attract demand away from the home country frms, forcng them to reduce ther producton and to move up ther average cost curve, resultng n productvy losses n the home country. In addon, outward FDI can reduce domestc capal accumulaton and thus domestc productvy when outward nvestors clam scarce domestc resources, such as domestc fnancal capal, that could otherwse have been used by domestc nvestors for nvestment n ther home country (see, e.g., UNCTAD, 2006). Thus, the net effect of outward FDI on aggregate productvy n developng countres s theoretcally ambguous and must be determned emprcally. Although there are no emprcal nvestgatons on ths overall macroeconomc effect on developng countres, some studes do exst on the frm- and ndustry-level effects of outward FDI on domestc productvy for both developng and developed countres. Also, there s some evdence of cross-border R&D spllovers through outward FDI at the macro-level for developed economes. 8

11 2.2. Avalable evdence Van Pottelsberghe and Lchtenberg (2001) use country-level macro data for a panel of 13 developed countres over the perod to examne whether technology-sourcng FDI affects domestc productvy through foregn R&D spllovers. They fnd a posve long-run relatonshp between the foregn R&D capal stock weghted by outward FDI and domestc total factor productvy, mplyng that outward FDI nto R&D-ntensve countres ndeed has benefcal effects upon home-country productvy by transferrng technologcal knowledge from the host country. However, Bzer and Kerekes (2008) reach a dfferent concluson. Ther fndngs, based on ndustry-level data for 17 OECD countres between 1973 and 2000, suggest that the nteracton between foregn R&D capal and outward FDI s negatvely assocated wh domestc productvy n non-g7 countres; for the G7 the evdence of R&D spllovers through outward FDI s not sgnfcant. Both studes nvestgate only whether outward FDI nto major R&D-performng countres acts as a channel for R&D spllovers, thus neglectng all other potental productvy effects of outward FDI. Braconer et al. (2001), n contrast, nvestgate both the effect of the outward-fdi-weghted foregn R&D capal stock and the effect of pure outward FDI on domestc productvy. Usng frm- and ndustry-level data for Sweden over the perod , they fnd neher evdence of FDI-related R&D spllovers, nor any correlaton between outward FDI per se and domestc productvy for Sweden. These results dffer from those of Drffeld et al. (2009). In an ndustry study for the UK coverng the perod , the authors dstngush between outward FDI n hgh-cost, hgh-r&d-ntensve and outward FDI n low-cost, low-r&d-ntensve countres. They fnd that both types of FDI generate productvy growth n the UK, suggestng that technologysourcng and effcency-seekng FDI ncrease domestc productvy. A smlar result s obtaned by Drffeld and Chang (2009), who nvestgate the effects of outward FDI from Tawan to Chna. Based on ndustry data for , they report a posve assocaton between outward FDI to Chna and labor productvy n Tawan. Gven the fact that labor costs n Tawan are sgnfcantly above those n Chna, the authors conclude that ths productvy effect s due to vertcal (effcencyseekng) FDI. Vahter and Masso (2007), on the other hand, use frm-level panel data from Estona between 1995 and 2002 to examne the effects of outward FDI on total factor productvy of the nvestng frms and the rest of the ndustry. They fnd that outward FDI s posvely related to the productvy of the parent companes, whereas there s no robust evdence of productvy spllovers to other frms. Snce the overwhelmng majory of FDI by Estonan frms s horzontal (marketseekng) (see, e.g., Masso et al., 2008), the posve productvy effects of Estonan outward FDI appear to be prmarly assocated wh ths type of FDI. 9

12 In another study, Kmura and Kyota (2006) analyze Japanese frm-level data for the perod One of ther fndngs s that outward FDI ncreases frm productvy. More specfcally, ther results suggest that frms engagng n outward FDI experence, on average, productvy growth 1.8% hgher than domestc frms not engagng n outward FDI. Hjzen et al. (2006a), however, crcze ths study for falng to control for the endogeney bas that arses when more productve frms self-select nto nvestng abroad. To deal wh ths endogeney problem, they apply matchng and dfference-n-dfferences analyss to data of Japanese frms for the perod The evdence n ther study suggests that the effect of outward FDI on Japanese frm productvy s not sgnfcant. Propensy score matchng and dfference-n-dfference technques are also used n other studes. Barba Navarett and Castellan (2004) apply these methods to Italan frm-level data for the perod They fnd that multnatonal frms have hgher total factor productvy growth after nvestng abroad than does a counterfactual of natonal frms. Klenert and Toubal (2009), n an analyss of German frm-level data for years , fnd no sgnfcant effect from the establshment of a foregn afflate on frm productvy growth. Hjzen et al. (2006b), usng French frm-level data between 1984 and 2002, report that frms that nvest n developed countres ncrease ther productvy, whle frms that nvest n developng countres experence no productvy effects, whch could suggest that productvy effects of outward FDI are prmarly assocated wh horzontal nvestment rather than vertcal nvestment. Barba Navarett et al. (2009) obtan the same result n a sample of French frms n the perod ; for Italy, however, they fnd exactly the oppose pattern: frms that nvest n developng countres experence an ncrease n total factor productvy, whereas FDI nto developed countres has no productvy effects. Fnally, Bzer and Görg (2009) examne the effect of outward FDI on domestc total factor productvy usng ndustry-level panel data for 17 OECD countres over the perod Ther results suggest that outward FDI has, on average, a negatve effect on total factor productvy, but that there are large dfferences across countres. Outward FDI has the largest negatve effect on total factor productvy n South Korea the only developng country n the sample. In France, Japan, Poland, Sweden, the Czech Republc, the UK, and the US, n contrast, ncreased outward FDI s assocated wh hgher total factor productvy. Gven the mxed results, perhaps the only conclusons that can safely be drawn from these studes are that outward FDI can have posve, as well as negatve, effects on domestc productvy, that the domestc productvy effects of outward FDI do not necessarly depend upon the nvestment motve, and that the effects of outward FDI can dffer sgnfcantly from country to country. The latter may apply n partcular to developng countres, whch dffer wdely n terms of 10

13 country sze, ncome level, economc structure, natural resources, technologcal capables, trade openness, government polces, and other characterstcs. Unfortunately, the studes do not provde any nformaton on how outward FDI could affect aggregate productvy n developng countres on average over the long run. 3. Emprcal model and data The analyss wll examne the long-run relatonshp between outward FDI and total factor productvy n developng countres. In ths secton, we present the basc emprcal model, dscuss some econometrc ssues, and descrbe the data Emprcal specfcaton and econometrc ssues We assume that the correct specfcaton of the functonal form of the long-run relatonshp between total factor productvy and outward FDI s gven by log( TFP ) a t blog( OFDI ), (2) where 1, 2,..., N s the country ndex, t 1, 2,..., T s the tme ndex, log( TFP ) represents the log of total factor productvy, and log( OFDI ) s the log of outward FDI. Followng Bzer and Görg (2009), we use outward FDI stocks rather than outward FDI flows, because stocks, due to the accumulaton of flows, may more effectvely capture long-run effects. The sze of the long-run effect of outward FDI on total factor productvy s measured by the coeffcent b, whch can be nterpreted as the long-run elastcy of total factor productvy wh respect to outward FDI. Fnally, any country-specfc omted factors whch are relatvely stable n the long run or evolve smoothly over tme are captured by country-specfc fxed effects, a, and country-specfc tme trends, t. Equaton (2) assumes a long-run bvarate relatonshp between permanent movements n the log level of outward FDI and permanent movements n the log level of total factor productvy. Necessary condons for ths assumpton to hold (and thus for our model to be a correct descrpton of the data) are that both the ndvdual tme seres for the log of total factor productvy and the ndvdual tme seres for the log of outward FDI are nonstatonary or, more specfcally, ntegrated of the same order and that log( TFP ) and log( OFDI ) form a contegrated par. A regresson consstng of two contegrated varables has a statonary error term,, n turn mplyng that no 11

14 relevant ntegrated varables are omted; any omted nonstatonary varable that s part of the contegratng relatonshp would enter the error term, thereby producng nonstatonary resduals and thus leadng to a falure to detect contegraton. Another assumpton nherent n Equaton (2) s that total factor productvy s endogenous n the sense that, n the long run, changes n outward FDI cause changes n total factor productvy. However, although the exstence of contegaton mples long-run Granger-causaly n at least one drecton, long-run causaly may also run from total factor productvy to outward FDI. The ratonale s that recent theoretcal work on frm heterogeney and FDI suggests that the establshment or acquson of foregn afflates nvolves addonal costs of overcomng legal, cultural and socal barrers, so that only frms above a certan productvy threshold can cope wh these fxed costs and thus engage n outward FDI (see, e.g., Helpman et al., 2004; Aw and Lee, 2008). That s, only the most productve frms self-select nto nvestng abroad. 2 Snce an ncrease n aggregate productvy s generally assocated wh an ncrease n average frm productvy and, consequently, wh an ncrease n the number of frms reachng the crcal productvy level necessary for FDI, a macroeconomc mplcaton of heterogeneous-frm models s that the aggregate amount of outward FDI should ncrease as total factor productvy ncreases. On the other hand, gven that total factor productvy growth s generally assocated wh domestc output growth, hgher demand, and hence better prof opportunes for domestc nvestment, an ncrease n total factor productvy can also lead to a reallocaton of scarce funds to more profable domestc nvestment opportunes n place of less profable outward nvestment. Consequently, ncreased factor productvy may be both the cause of reduced and the cause of ncreased outward FDI actvy. The emprcal mplcaton s that s not only crucal to examne the tme-seres propertes of the varables and to test whether the varables are contegrated, but s also mportant to deal wh ths endogeney problem and to nvestgate the drecton of causaly. Further, an econometrc ssue s the potental cross-country heterogeney n the relatonshp between outward FDI and total factor productvy. As dscussed n Secton 2.2, ndvdual country studes based on frm- and ndustry-level data tend to fnd dfferent results for dfferent countres. In partcular, the study by Bzer and Görg (2009) suggests that the productvy effects of outward FDI are not constant across countres. Thus, we face a dlemma regardng the optmal estmaton strategy. On the one hand, effcency gans from the poolng of observatons over the cross- 2 Models of frm heterogeney typcally predct a productvy rankng n whch foregn nvestng frms are the most productve, followed by exporters and non-exporters. Ths s confrmed by emprcal evdence whch shows that multnatonals are the most productve among the three types of frms, see, e.g., Head and Res (2003) for Japan, Grma et al. (2004) for Ireland, Arnold and Hussnger (2005) for Germany, and Grma et al. (2005) for the UK. 12

15 sectonal uns can be acheved when the ndvdual slope coeffcents are the same, b b. On the other hand, pooled whn-dmenson estmators produce nconsstent and potentally msleadng pont estmates of the sample mean of the heterogeneous contegratng vectors when the true slope coeffcents are heterogeneous (see, e.g., Pesaran and Smh, 1995). Although a comparatve study by Baltag and Grffn (1997) concludes that the effcency gans from poolng more than offset the bases due to ndvdual country heterogeney, we try to solve ths dlemma by usng both homogeneous (whn-dmenson-based) and heterogeneous (between-dmenson-based) estmators. The latter allows us, n addon, to explcly analyze the potental heterogeney n the effects of outward FDI across countres. A fnal econometrc ssue s the potental cross-sectonal dependence among the varables. In partcular, the total factor productves are lkely to be cross-sectonally dependent, as suggested by both theoretcal models and emprcal studes. Neoclasscal growth models, for example, mplcly assume that all countres have access to the same global stock of knowledge, whle many endogenous growth models predct that knowledge dffuses across natonal boundares through channels such as trade and FDI. Snce the total factor productvy of a country depends upon the avalable (local and foregn) technologcal knowledge, follows that factor productves are globally nterdependent. Ths concluson s supported by several studes showng that a country s productvy depends not only upon s own stock of knowledge but also upon the knowledge stocks, and thus the total factor productves, of s tradng partners (see, e.g., Coe and Helpman, 1995; Van Pottelsberghe and Lchtenberg, 2001; Luntel and Khan, 2004). Gven that standard panel un root and contegraton tests may be based n the presence of such cross-sectonal dependence, we also use recent advances n panel data econometrcs to account for ths ssue Data and descrptve statstcs We now descrbe the data used n the emprcal analyss. Total factor productvy s defned ( 1 ) L n the usual way as TFP Y /[ K ], where Y s output, K denotes the capal stock, L stands for labor nput, 1 s the capal share of ncome, and s the labor share of ncome. We assume a constant of , whch can be justfed as follows: Frst, s common practce n the lerature to assume (or use) a constant labor share of two-thrds; and second, the evdence of Golln (2002) suggests that the labor share s, n fact, approxmately constant across tme and space wh a value of about two-thrds. Output s measured by real GDP (n 2000 US$), whle L s represented by the labor force (.e., the number of persons of workng age, defned as years). Whout queston, a better measure of labor nput would be employment tmes average hours, but relable data on 13

16 employment and hours worked are generally not avalable for developng countres over a long enough perod of tme. Therefore, we follow the common practce n developng country studes and use the labor force as our measure of labor nput (see, e.g., Abu-Qarn and Abu-Bader, 2007). Unfortunately, offcal capal stock data are also not avalable for most developng countres. We therefore construct the physcal capal stock from real nvestment data (gross capal formaton n 2000 US$) usng the perpetual nventory equaton Kt It ( 1 ) Kt1, where I t s nvestment and s the deprecaton rate. Consstent wh the lerature, we set the nal value of the capal stock equal to K 0 I0 /( g ), 3 where I 0 s the value of the nvestment seres the frst year s avalable, and g s the average growth rate of the nvestment seres between the frst year wh avalable data and the frst year of the estmaton perod (see, e.g., Casell, 2005). In cases where the tme nterval between the frst year wh avalable data and the frst year of the estmaton perod s less than fve years, we follow Luntel and Khan (2004) and use the average growth rate of nvestment over the estmaton perod. As s standard n the lerature, a deprecaton rate of 6% s assumed. All data used to calculate total factor productvy are from the World Development Indcators (WDI) 2008 CD-Rom. The data on outward FDI stocks are obtaned from the UNCTAD FDI database ( UNCTAD (2008) defnes FDI as an nvestment nvolvng a long-term relatonshp and reflectng lastng nterest and control of a resdent enty n one economy n an enterprse resdent n an economy other than that of the foregn drect nvestor. The stock of FDI s defned as the value of the share of the foregn enterprse capal and reserves (ncludng retaned profs) attrbutable to the parent enterprse plus the net ndebtedness of afflates to the parent enterprse. Snce UNCTAD reports outward FDI stocks as shares of GDP, we multply the outward FDI-to-GDP rato by real GDP (from the WDI) to construct real outward FDI stocks (n 2000 US$). Gven that the UNCTAD data start n 1980 whle the WDI 2008 data end n 2005, the emprcal analyss covers the perod Our sample ncludes all countres for whch data n ths perod are avalable. Of these countres, four are n North Afrca (Algera, Egypt, Morocco, and Tunsa), nne are n sub-saharan Afrca (Benn, Botswana, Burkna Faso, Gabon, Kenya, Mal, Senegal, South Afrca, and Swazland), three are n Central Amerca (Costa Rca, Mexco, and Panama), nne are n South Amerca (Argentna, Bolva, Brazl, Chle, Colomba, Ecuador, Paraguay, Peru, and Venezuela), fve are n East Asa (Hong Kong, Indonesa, Malaysa, South Korea, and Thaland), and three are 3 The ratonal for ths choce s that I / (g + δ) s the expresson for the capal stock n the steady state of the Solow model. 14

17 n South Asa and the Mddle East (Inda, Jordan, and Pakstan). Thus, our sample conssts of 33 countres from all developng regons of the world and should therefore be reasonably representatve of developng countres. Table 2 lsts the countres along wh the average values for TFP and OFDI over the perod of observaton. As can be seen, there are consderable dfferences n the values of these parameters across countres. Hong Kong s the country wh the hghest productvy level, followed by Argentna, South Korea, and Mexco, whle Burkna Faso ranks at the bottom of the productvy scale. Total factor productvy n Hong Kong exceeds the productvy ndex of Burkna Faso by almost 20-fold. Perhaps even more strkng are the cross-country dfferences n the outward FDI stocks. Brazl s FDI stock s almost tmes larger than the FDI stock of Burkna Faso. Brazl heads the lst of foregn nvestors, followed by Hong Kong, South Afrca, and Argentna. The countres wh the lowest stocks of FDI are (n ascendng order) Burkna Faso, Benn, Bolva, and Tunsa. Altogether, appears that countres wh hgher factor productves tend to have larger outward FDI stocks, whle those wh lower productves correspondngly have smaller FDI stocks, suggestng a posve relatonshp between these two varables. Table 2 Countres and country summary statstcs Average of TFP Average of OFDI Average of TFP Average of OFDI Algera Jordan Argentna Kenya Benn Malaysa Bolva Mal Botswana Mexco Brazl Morocco Burkna Faso Pakstan Chle Panama Colomba Paraguay Costa Rca Peru Ecuador Senegal Egypt South Afrca Gabon South Korea Hong Kong Swazland Inda Thaland Indonesa Tunsa Venezuela Fgure 1 llustrates ths cross-country relatonshp graphcally. It shows the scatter plot of the average log of total factor productvy versus the average log of outward FDI for the 33 countres n our sample along wh the regresson lne. The slope of the regresson lne s posve (and statstcally sgnfcant wh a t statstc of 5.94), ndcatng that ncreased outward FDI s ndeed 15

18 posvely assocated wh ncreased factor productvy. In the next secton, we examne ths relatonshp n more detal usng panel contegraton and causaly technques. Fgure 1 Scatter plot of the log of total factor productvy versus the log of outward FDI Log of total factor productvy Log of outward FDI 4. Emprcal analyss The pre-tests for un-roots and contegraton, whch are reported n the Appendx, suggest that the varables are nonstatonary and contegrated, as assumed n Equaton (2). In ths secton, we provde estmates of the contegratng relatonshp between outward FDI and total factor productvy and test the robustness of the estmates. We also nvestgate the drecton of causaly between the two varables, examne the degree of heterogeney n the effects of outward FDI on total factor productvy across countres, and search for explanatons for the cross-county heterogeney Long-run relatonshp In order to estmate the long-run elastcy of total factor productvy wh respect to outward FDI, we use the dynamc ordnary least squares (DOLS) estmator. Ths estmator s asymptotcally unbased and normally dstrbuted even n the presence of endogenous regressors 16

19 (see, e.g., Stock and Watson, 1993), thus allowng us to control for the potental endogeney of outward FDI dscussed n Secton 3.1. Furthermore, the DOLS estmator performs well n fne samples compared wh other contegraton estmators (such as the fully modfed estmator) both n tme-seres and panel data (see, e.g., Stock and Watson, 1993; Kao and Chang, 2000). The whndmenson-based DOLS model used n ths paper and followng Kao and Chang (2000) s k log( TFP ) a t b log( OFDI ) log( OFDI, (3) jk where Φ j are coeffcents of current, lead and lag dfferences, whch account for possble seral correlaton and endogeney of the regressor(s), thus yeldng unbased estmates. The results of ths estmaton procedure are presented n the frst column of Table 3 where, for brevy, we report only the estmated b coeffcents. The estmated coeffcent s hghly sgnfcant and posve. More precsely, the elastcy of total factor productvy wh respect to outward FDI s estmated to be 0.024, mplyng that, n the long run, a 1% ncrease n the outward FDI stock s assocated wh an ncrease n total factor productvy by 0.024%. From ths can be concluded that developng countres benef n general or on average from outward FDI due to the ncreased productvy of the nvestng companes and assocated productvy spllovers to local frms. j j ) Table 3 Estmates of the long-run effect, b, of outward FDI on total factor productvy Whn-dmenson DOLS estmator (Kao and Chang, 2000) DOLS mean group estmator (Pedron, 2001) CCE mean group estmator (Pesaran, 2006) 0.024** (2.97) 0.014** (3.42) 0.050** (4.61) Notes: The dependent varable s log(tfp ). t statstcs n parenthess. ** ndcate sgnfcance at the 1% level. The DOLS regressons were estmated wh one lead and one lag. To assess the robustness of ths concluson, we perform several sensvy checks. Frst, we nvestgate whether the posve relatonshp between outward FDI and total factor productvy s robust to alternatve estmaton methods. Specfcally, a potental problem wh the above estmaton procedure could be that assumes a homogeneous b, whch s lkely to be emprcally ncorrect (as we wll show n Secton 4.3). To allow the slope coeffcents to vary across countres, we use the between-dmenson, group-mean panel DOLS estmator suggested by Pedron (2001). Ths estmator nvolves estmatng separate DOLS regressons for each country and averagng the longrun coeffcents, ˆ 1 N N bˆ. The t statstc for the average s the sum of the ndvdual t statstcs b 1 dvded by the root of the number of cross-sectonal uns, N t bˆ 1 bˆ t / N. We present the DOLS 17

20 group-mean pont estmate of the effect of outward FDI on total factor productvy n the second column of Table 3. Snce, however, the DOLS estmates may be based n the presence of crosssectonal dependence, we also report (n the thrd column) the result of the common correlated effects (CCE) mean group estmator suggested by Pesaran (2006). Ths estmator allows for crosssectonal dependences that potentally arse from multple unobserved common factors and s the smple average of the ndvdual common correlated effects estmators gven by Equaton (A.6) n the Appendx. As can be seen, all three estmators produce smlar results, suggestng that the posve relatonshp between outward FDI and total factor productvy s neher due to potentally restrctve homogeney assumptons nor to possble cross-sectonal dependence. However, gven the relatvely short tme dmenson of our data, the mean group results (whch are based on ndvdual tme-seres regressons) should be nterpreted wh cauton. In addon, the CCE mean group estmator s ntended for the case n whch the regressors are exogenous, so that we lose the ably to account for the lkely endogeney of outward FDI. Therefore, we contnue our robustness analyss wh the pooled whn-dmenson panel DOLS estmator. As noted n Secton 3.1, there s evdence to suggest that the effcency gans from poolng are lkely to more than offset the potental bases due to ndvdual heterogeney (see, e.g., Baltag and Grffn, 1997). Fgure 2 DOLS estmaton wh sngle country excluded from the sample Coeffcents on log(ofdi ).036 t statstcs of the coeffcents We re-estmate the DOLS regresson excludng one country at a tme from the sample to verfy that the posve effect of outward FDI s not due to ndvdual outlers. The sequentally estmated long-run coeffcents and ther t statstcs are presented n Fgure 2. As they are relatvely 18

21 stable between and and always sgnfcant at the 5% level, we conclude that the posve productvy effect s not the result of ndvdual outlers. Next, we examne whether the posve long-run relatonshp between outward FDI and total factor productvy s due to sample-selecton bas. Sample-selecton bas occurs when the selected sample s not random and thus not representatve. A potental problem wh our sample could be that ncludes only 6 low-ncome countres (Benn, Burkna Faso, Kenya, Mal, Pakstan, Senegal), whle the rest of the countres are classfed by the World Bank (WDI 2008 CD-Rom) as mddle- or hgh ncome countres. Another possble problem s the dstrbuton of outward FDI among the countres. In fact, our sample s domnated by 6 countres (Argentna, Brazl, Hong Kong, Mexco, Panama, South Afrca) that nvest more than the sample average. We therefore re-estmate the DOLS regresson for four subsamples: low-ncome countres, mddle- and hgh-ncome countres, countres wh outward FDI above the sample average, and countres wh outward FDI below the sample average. The resultng coeffcents are lsted n Table 4. Regardless of whch subsample s used, the long-run relatonshp between outward FDI and total factor productvy remans posve and sgnfcant at least at the 5% level. From ths, can be concluded that the posve coeffcent on log( OFDI ) s not due to sample-selecton bas. Clearly, would be desrable to also assess whether there are sgnfcant dfferences n the effects of outward FDI on total factor productvy between low-ncome and mddle- to hgh-ncome countres or between countres wh low levels of outward FDI and those wh hgh levels of outward FDI. However, the small sample szes do not allow statstcally meanngful comparsons. Table 4 DOLS estmates for subsamples log(ofdi ) Number of countres n the subsample Low-ncome countres 0.079* (2.16) 6 Mddle- and hgh-ncome countres 0.028** (3.08) 27 Countres wh outward FDI above the sample average 0.056* (2.19) 6 Countres wh outward FDI below the sample average 0.021* (2.38) 27 Notes: ** (*) ndcate sgnfcance at the 1% (5%) level. t statstcs n parentheses. Fnally, we examne whether the estmated coeffcent on outward FDI s sensve to the model specfcaton. To ths end, we combne the defnon of total factor productvy, TFP Y /[ K ( 1 ) L Y ], wh Equaton (1) to obtan (after some rearrangements) K log a b OFDI L (1 )log L log( ), (4) 19

22 where Y / L ) s labor productvy, K / L ) s capal per worker, and s (as before) the labor ( ( share of ncome, whch was assumed to be Snce the elastcy of labor productvy wh respect to outward FDI, b, s equal to the elastcy of total factor productvy wh respect to outward FDI, Equaton (4) allows us to test whether there are systematc bases n the calculated factor productvy and the estmated coeffcent on outward FDI. If there are no systematc bases, Equaton (4) should, on the one hand, produce approxmately the same outward FDI coeffcent as Equaton (2) and thus a b value of about and, on the other hand, a coeffcent on log capal per worker of about ( =) The DOLS estmates for these two parameters are gven n Table 5. As can be seen, the coeffcent on log capal per worker s ndeed close to 1/3 and the estmated b coeffcent s close to In addon, the last row of Table 5 shows that smple Wald tests do not reject the restrctons that ( 1 ˆ ) = and b = Thus, can be concluded that the posve productvy effect of outward FDI s robust to dfferent estmaton technques, potental outlers, sample selecton, and the specfcaton of the emprcal model. Table 5 DOLS estmates of the coeffcents on log capal per worker, (1 ), and log outward FDI, b log(k /L ) log(ofdi ) 0.368** (14.14) 0.019** (2.99) Wald tests Restrcton: ( 1 ˆ ) = Restrcton: bˆ = 0.24 χ 2 (1) [p-values] 1.81 [0.18] 0.52 [0.47] Notes: The dependent varable s log(y /L ). ** ndcate sgnfcance at the 1% level. t-statstcs n parentheses. The DOLS regresson was estmated wh one lead and one lag. The number of degrees of freedom υ n the χ 2 (υ) tests correspond to the number of restrctons Long-run causaly Recent theores on frm heterogeney and FDI suggest that only those frms wh productves above a certan threshold fnd profable to locate producton or other actves abroad (see, e.g., Helpman et al., 2004; Aw and Lee, 2008). Snce s reasonable to assume that an ncrease n aggregate productvy s assocated wh an ncrease n the number of frms reachng ths threshold, follows that economy-wde productvy gan should lead to ncreased FDI outflows. Consequently, the posve coeffcent on log( OFDI ) does not necessarly reflect a causal effect of outward nvestment on total factor productvy, but causaly may also run from log( TFP ) to log( OFDI ). To test the drecton of causaly, we use a two-step procedure. In the frst step, we employ the DOLS estmate of the long-run relatonshp to construct the dsequlbrum term 20