Using a firm-level survey, this study examines the effects of foreign direct investment and

Transcription

1 Foregn Drect Investment, Technology Transfer and Frm Performance Sarah Yuetng Tong Hong Kong Insttute of Economcs and Busness Strateges Unversty of Hong Kong Aprl 2001 Abstract: Usng a frm-level survey, ths study examnes the effects of foregn drect nvestment and foregn technology on local Chnese frms. The emprcal results suggest that knowledge nflow carred out manly through drect foregn nvestment s an mportant condut n promotng Chnese frms' export. Moreover, foregn knowledge s lkely to ncrease local frms' total employment and producton, especally n the short-run. In addton, local non-afflate frms can beneft through busness dealngs wth those frms drectly assocated wth foregn busnesses. The study also explores the possble contrbutng factors related to foregn technology transfer, such as domestc competton and employees educaton level. Sarah Yuetng Tong Hong Kong Insttute of Economcs and Busness Strateges Unversty of Hong Kong Tel: Fax: E-mal: tongsy@hku.hk

2 I. Introducton What s the role of foregn drect nvestment (FDI) n developng countres? How does FDI affect local companes n ther efforts to grow and to become global? These have become mportant questons n recent years. On one hand, more developng countres are pursung economc polces open to trade and foregn nvestment. On the other hand, developng countres have become mportant players n the world economy, as producers, as consumers, as nvestors and as destnatons for cross-border nvestment. Numerous emprcal studes have demonstrated a postve correlaton between the openness of an economy and ts economc growth among developng countres (e.g. Syrqun and Chenery 1989, Borenszten, De Gregora & Lee 1995, and We 1993). Edwards (1993) and Harrson (1996) provde revews of the early studes. By the openness of an economy, we refer to a busness and regulatory envronment that are frendly toward trade and foregn nvestment. It s sometmes measured by the percentages of trade (or foregn drect nvestment) of a country s gross domestc product (GDP), or overall tarff level. Despte the overall enthusasm toward the postve mpact of openness and trade n recent years, there are only a lmted number of studes that analyze the economc mechansm nvolved n the process. Some suggest that economc openness affects growth by nducng more nvestment (e.g. Baldwn and Seghezza, 1996). Many others emphasze the role of technologcal progress assocated wth more trade and more foregn nvestment of an economy. Trade can promote technology progress n developng countres. For example, more trade nduces more R&D spendng n domestc frms so that they can be more compettve n the market place. In addton, frms n developng countres can acqure new technologes emboded n new machnes and new products they purchased from foregn sources. Smlarly, foregn drect 1

3 nvestment can facltate technology progress n developng countres. In fact, foregn drect nvestment carred out by multnatonal corporatons (MNCs) s beleved to be one of the most mportant vehcles for the nternatonal dffuson of technology. There are two reasons why FDI s very mportant for developng countres to acqure new technologes. Frst, MNCs are more advanced n technology. A substantal porton of the world s total research and development s carred out wthn the large MNCs. Therefore; MNCs often possess the much-needed new and advanced technologes. Second, through drect nvolvement of foregn busnesses, MNCs domestc afflates and other domestc producers can acqure new technology more drectly and more effectvely. The benefts from FDI are not lmted to new technology. Other drect benefts nclude the productvty ncreases n MNCs local afflates, new management sklls brought n by the MNCs, and a potental market expanson brought about through foregn nvestors. Foregn nvestment can also ncrease the productvty n the host economy ndrectly through ts nfluence on both the ndustral structure of the host economy and the conduct and performance of domestcally owned frms. Ths s accomplshed through ncreased competton n local economy, more nvestment n captal and human captal, tranng of labor and management, tranng of local supplers of ntermedate products, and transfer of knowledge (e.g. Blomstrom and Persson 1983, Frschtak and Newfarmer 1992, Blomstrom 1991). As a result of foregn nvestment and foregn knowledge nflow, local afflates of MNCs can acheve productvty ncrease and therefore hgher growth. At the same tme, the frms can also realze more export as they become more and more compettve. Emprcal studes suggest that the presence of MNCs n developng countres and the assocated nvestment have mportant 2

4 mpacts on the export of ther local afflates and not-afflates n the host economy (e.g. Atken, Hanson, and Harrson 1997, Lpsey 1995, and Naujoks and Schmdt 1995). Although these studes ndcate that foregn drect nvestment have a postve mpact on the growth and export by local frms, they usually don t explctly examne the effect of knowledge nflows that are often assocated wth foregn nvestment. As FDI can beneft domestc producers n varous ways, such as provdng new technology and new potental market network, the dstncton s mportant. In ths study, we take a dfferent approach by usng mcro-level frm survey data to nvestgate the role of foregn nvestment and foregn technology nflow separately. In partcular, we are nterested n the mpact of foregn nvestment and technology transfer on local frms export actvtes. We also nvestgate the effects of FDI and technology transfer on frms other actvtes, such as employment and producton. Furthermore, we also seek to dentfy factors that mght contrbute to foregn knowledge nflow. The emprcal results are consstent wth the hypothess that foregn drect nvestment ndeed contrbutes to local frms' exports and growth through technology transfer. The results also ndcate that the effects of foregn drect nvestment are carred out through both drect technology transfer and ndrect knowledge dffuson. That means both MNCs local afflates and other domestc frms beneft. Through analyzng contrbutng factors to foregn knowledge, we fnd that domestc competton s an mportant element n promotng technology transfer. Secton II brefly descrbes the data set used n the emprcal analyss and provdes some background nformaton regardng the ctes ncluded n the study. Secton III presents the analytcal framework for the emprcal study. Secton IV presents and dscusses the estmaton results. Secton V concludes. 3

5 II. Background and Data Descrpton The data used n the analyss are based on a survey conducted by the World Bank n early 1993 n eght ctes n Chna. 1 Sx of these ctes are located n two coastal provnces n Southeast Chna, Guangdong and Fujan. 2 The remanng two ctes are located n the nland provnce of Schuan. In each of the eght ctes, ffty to sxty frms are randomly chosen and a total of about 500 frms are ncluded n the survey. 3 Table 1 s a summary of basc economc ndces for the eght ctes. It s obvous that the ctes are qute dfferent n ther populaton, total ndustral output, ownershp composton, ndustry composton, and the ntensty of foregn nvolvement. For example, some of the ctes are much larger and more domnated by the state sector than the other ctes. These are ether provncal captals, such as Guangzhou, Fuzhou and Chengdu, or a cty of heavy ndustry base, such as Chongqng. In contrast, the other four ctes are relatvely small and have less presence of a state sector. In these small ctes, ndustral output s more concentrated on consumer goods such as food, textles, and electroncs. Between the coastal and the nland ctes, there are consderable dfferences n terms of foregn nvolvement n local ndustry. In the two nland ctes, frms wth foregn partcpaton account for less than 1% of all the frms and less than 2% of total ndustral output. For the coastal ctes, the numbers are 10% and 20%. In the two SEZs, foregn partcpaton s even hgher. More than 1/3 of the ndustral establshment and more than half of ndustral output n Shenzhen and Xamen are from frms wth foregn nvolvement. 1 These are Chengdu (CD) and Chongqng (CQ) n Schuan Provnce of Western nland, Guangzhou (GZ), Shenzhen (SZ), and Dongguan (DG) n Guangdong Provnce on Southeast coast, and Fuzhou (FZ), Xamen (XM) and Quanzhou (QZ) n Fujan Provnce on Southern coast. 2 Two of the coastal ctes, Shenzhen and Xamen, were chosen as Specal Economc Zones (SEZs) n the early 1980s. The other two SEZs are Zhuha and Shantou. 3 The frms n the sample are randomly chosen wthn each type of ownershp category, namely the state-owned enterprses, the collectvely owned enterprses and others. 4

6 Three factors are mportant to the contrast between the coastal and nland ctes. Frst, the dfference reflects the gradual and uneven nature of Chna s economc reform. These two provnces are among the frst to have a more open and more flexble economc envronment, they also enjoy a geographc proxmty to the fast growng Southeast economes. For example, n the early 1980s when Chna s economc reform began, 4 coastal ctes 4, ncludng two ctes n our sample (Shenzhen and Xamen), were chosen as Chna s Specal Economc Zones (SEZs). They are gven varous polcy prvleges ncludng some autonomy to create a more open and market orented busness envronment. In 1984, the government opened another 14 coastal ctes as a step to deepen the reform process. On the contrary, nland ctes are gven lower preference n ther reform process. Second, the two southeast coastal provnces enjoy varous locaton advantages. The regon has better nfrastructure endowment, ncludng a relatvely well-developed road system and multple seaports. The regon's sound transportaton facltes plus ts close proxmty to Southeast Asan economes 5 have gven the frms n the regon easer access to both domestc and nternatonal markets. They also contrbute to the regon s compettveness n attractng foregn nvestment. Thrd, the two coastal provnces enjoy close tes wth overseas Chnese 6. The majorty of overseas Chnese n many Southeast Asan countres and elsewhere, as well as Chnese n Hong Kong, are ether emgrants from the two provnces or descendants of those who emgrated from 4 In 1980, four Specal Economc Zones were set up, ncludng Shenzhen, Zhuha and Shantao n Guangdong Provnce, and Xamen n Fujan Provnce. In 1987, Hanan Island n Guangdong Provnce was gven provncal status and became the ffth specal economc zone. Wthn SEZs, foregn nvestment as well as varous forms of foregn jont ventures are encouraged through measures on tax abatement, relaxaton of foregn exchange controls and materal mports quotas, etc. 5 Guangdong Provnce s adjacent to Hong Kong and Macao. Fujan Provnce les just across the Tawan Strats from neghborng Tawan tself. 5

7 the two provnces. Durng the economc reform, oversea Chnese, together wth busness people from Tawan, have played a crucal role n brngng nvestment nto manland Chna. Untl recently, drect nvestments from outsde the border have n fact concentrated n Chna s southeast coastal regon. 7 The dfferences between the coastal and nland ctes gve us a unque opportunty to examne the role of foregn nvolvement and foregn technology n facltatng economc development. To emprcally study the effect of foregn nvolvement and foregn knowledge on local Chnese frms, we use a data set obtaned from frm survey. Next, we wll compare the ndustry and ownershp composton between the survey data the macro data obtaned for each of the 8 ctes. Ths s to gve us an ndcaton as to how far we can generalze out result. The survey ncludes a total of about 500 frms. They account for about 3% of all the ndustral establshments and 14% of total manufacturng output for the eght ctes together. That suggests to us that an average frm n the sample s larger than an average frm n the relevant ctes. Table 2 compares the ownershp composton between sample frms and all frms n the ctes. Among the three types of ownershp forms 8, collectve frms account for the most n the number of establshment and state enterprses account for the most ndustral outputs. Ths s true for most of the ctes n our sample as well as for the sum of the eght ctes. Among the sample frms, the proporton of collectve frms s generally smaller than that for the ctes and the proportons of state enterprses and other frms are generally larger than those for the ctes. 6 By usng the term overseas Chnese, we mean ethnc Chnese lvng outsde manland Chna and Tawan. 7 In 1992, nvestment from Hong Kong, Macao and Tawan amounted to approxmately 80% of total nvestment from outsde the manland n In the same year, 13% of total FDI landed n Fujan provnce, 33% n Guangdong provnce, and only less than 1% n Schuan provnce. 8 Busness establshment are categorzed nto three groups, state-owned enterprses, collectvely owned enterprses and other. Other forms of ownershp nclude prvate frms, foregn owned busnesses and varous types of jontventure establshments. 6

8 Smlar comparson s made regardng the frms ndustry composton based on output value. The results are summarzed n Table 3. To numercally assess the sample s smlarty n ownershp and ndustral composton wth those of the ctes, we calculate three correlaton coeffcents, two based on the shares of frms n dfferent ownershp categores and one based on the shares of frms n dfferent ndustres. The results are summarzed n Table 4. The correlatons for ownershp composton based on total output are 0.71 for the pooled data and 0.96 for the total of the eght ctes (0.41 and 0.95 when calculatons are based on the number of frms n each category). For the ndustry composton, the correlaton coeffcents between the sample and the ctes are 0.62 for the pooled and 0.70 for the total. We also obtaned the same correlaton coeffcents for the ndvdual ctes and the results show substantal varatons across ctes. Overall, the ownershp and ndustry composton of the sample frms reflect the overall structure for the ctes farly. Therefore, our analyss reflects the overall stuaton for the ctes nvolved relatvely well. However, we should take precautons when we apply our results to more general crcumstances. In the next secton, we study the effect of foregn drect nvestment on local Chnese frms. In partcular, we examne the role of foregn technology. III. Emprcal Study on the Effects of FDI and Knowledge Inflow Many theoretcal and emprcal studes suggest that foregn nvestment, especally foregn technology assocated wth foregn nvestment, carres postve effects to both local afflates as well as to local busnesses n general. In ths secton, we wll emprcally nvestgate the effects of foregn nvestment and foregn knowledge on local Chnese frms' actvtes. The analyss concentrates on three aspects of a frm s actvty: export, employment and producton growth, and tranng provded to domestc supplers. The prmary hypothess s whether foregn nvestment 7

9 and knowledge nflow each and together have sgnfcantly postve mpact on a frm s performance n the above three areas. The general model can be represented as followng: Y = α + βtech + γfdi + ζx + ε (1) H H 0 1 : γ : γ = 0 and/or ζ > 0 and/or ζ = 0 > 0 In ths equaton, Y s one of performance varables for the sample frms such as export amount or output growth. TECH s a bnary varable ndcatng whether a frm has obtaned substantal foregn knowledge nflow. FDI s the share of the frm s foregn ownershp. Fnally X s a vector that contans a frm s basc characterstcs such as geographc locaton, ndustry, and number of years n busness, etc. The varables ncluded n the followng estmatons wll be specfed n the subsequent sectons. In the frst set of analyses, we examne the effect of foregn nvestment and foregn knowledge on a frm s export actvtes. There are three closely related questons we want to address. Frst, whether foregn nvestment and foregn knowledge ncrease the probablty that a frm exports? Second, whether foregn nvestment and foregn knowledge ncrease the total amount of export? And thrd, whether foregn nvestment and foregn knowledge ncrease the share of total output beng exported. After nvestgatng the relatonshp between the frms foregn partcpaton and ther export actvtes, we study the frms employment and producton to see whether foregn nvolvement helps domestc to expand. And fnally, we explore the possble spllover effect, focusng on one partcular aspect. That s whether a frm s more lkely to provde tranng to ts local domestc supplers as a result of foregn knowledge nflow. 8

10 Before gettng nto the specfc emprcal models, we wll frst ntroduce some of the key varables used n the study. The defntons of all the varables relevant to ths paper are lsted n the appendx. Performance varables: EXPT_D s a bnary varable takes the value of 1 f a frm exports (n 1991) and 0 otherwse. LEXPT91 s the natural log of a frm s total export amount n 1991 (log (export91+1)). PCEXPT91 s the percentage of a frm's output beng exported n GEMPL s the growth rate of a frm s total employment from 1988 to GPRODN s the growth rate of a frm s total producton from 1988 to 1991 calculated usng output based on constant prce. GPRODR s the growth rate of a frm s total producton from 1988 to 1991 calculated usng output based on current prce. TRAINING s a bnary varable takes the values of 1 f a frm provded tranng to ts domestc nput supplers n 1990 and 0 otherwse. Foregn partcpaton and foregn knowledge varables: FRN s the share of foregn ownershp n a frm n FJV s a bnary varable takes the value of 1 f a frm has a foregn jont venture partner n 1991 and 0 otherwse. TECH s a bnary varable takes the values of 1 f the frm has n the past experenced substantal foregn knowledge nflow, and 0 otherwse. IMPORT s the percentage of the frm's machnery that s mported. 9

11 CONTACT s a bnary varable takes the value of 1 f the frm' typcal hgh-level manager n the frm havng busness lunches wth foregn buyers and supplers or foregn machnery supplers n 1991, 0 otherwse. Varables ndcatng a frm s characterstcs: SOE s the share of state ownershp n a frm n PVT s the share of prvate ownershp n a frm n LEMPL s the log of the frm's total number of employees n LPROD91N s the log of the frm's producton n 1991 n constant value. EDU s the weghted average years of educaton of a frm s employees. LSPEC s the number (n log) of a frm s full tme specal employees (ncludng qualty control, computer aded desgn, and tranng n the frm). AGE s the number of years a frm has been n operaton. OLD s a bnary varable takes the value of 1 f the frm started operaton before 1980 and 0 otherwse. COMPETITION s an ndex for the ntensty of domestc competton, frm perceved by a frm s top manager. It s represented by the absolute value of the frm's perceved prce electrcty, rangng from 0 to 3. 10

12 3.1. General descrptons of the sample In ths secton, we wll examne the general statstcs for some of the key varables. Table 5 presents the mean values of these varables broken down by ctes. It shows that a substantal porton of the frms has ether drect or ndrect assocaton wth foregn busnesses. For example, about one thrd of the frms have foregn ownershp. The average share of foregn ownershp n these frms s over 60%, whch suggests deep nvolvement of foregn owners. In addton, about 30 percent of the frms have foregn jont venture partners. 9 Assets holdng and formng a jont venture are not the only forms of foregn nvolvement n domestc economy. Frms sometmes form close relatons wth ther nput supplers, output buyers, and machne provders. These nformal assocatons can also serve as a source of now technology. Data show that n at least 30 percent of the frms, the top managers have had busness lunches wth ther foregn assocates. 10 Assocated wth varous forms of foregn nvolvement, 45% of the frms have reported substantal foregn knowledge nflow n the past. There s another measure that can also ndcate a frm exposure to new technologes, machne mport. Machne mport s mportant for domestc frms snce many beleved that nformaton regardng new technologes n desgnng, producton and so on could be emboded n the advanced machnes. Table 5 shows that mported machnes are present n more than two ffth of the sample frms. In those frms, mported machnes account for more than 60 percent of total machnery on average. 9 In most cases, a frm s foregn owner s also ts jont venture partner. There are, however, foregn venture partners that do not own any porton of the frm s assets. For example, some jont ventures are set up for mport materal processng. Jont ventures could also exst n the form of jont-management, etc. 10 The survey also asked whether a frm s top managers had busness dnner wth foregn busness assocates. 11

13 Table 5 also summarzes some of the frms performance varables such as export. Among all the sample frms, more than half exported n In these frms nearly 60% of ther output are sold to foregn market on average. Asde from the overall mean values for the key varables, Table 5 also shows the varaton among the ctes n terms of ther assocaton wth foregn busnesses and ther export performance. In the two nland ctes, Chengdu and Chongqng, only very few frms have foregn ownershp or have a foregn jont venture partner. Seemngly related to the sparse assocaton wth foregn busnesses n the two nland ctes, few frms reported havng experence foregn knowledge nflow or mported machnery. The dfferences n the frms access to foregn busnesses and foregn knowledge does seem to be reflected n frms export actvtes. In the two nland ctes, few frms export. Among the exportng frms, those located n the small coastal ctes (such as Shenzhen, Dongguan, and Quanzhou) sell more, as a percentage to total sales, to the world market. For total amount of export, however, there are no large varatons among frms n dfferent locatons. The above observaton suggests that foregn ownershp, technology dffuson and the Chnese frms export seem to be closely related. Other factors may also be attrbutve to the dfference n the frms export actvtes. For example, frms n the two nland ctes, as well as frms n Guangzhou and Fuzhou, the other two provncal captals are on average older. To study the relatons among the frms foregn assocaton, knowledge nflow, export, and other characterstcs, we frst calculated the correlaton coeffcents among the key varables. The results n Table 6 ndcate that postve and sgnfcant correlatons do exst among the three key varables: the share of foregn ownershp, technology transfer, and export. Furthermore, technology transfer s also postvely assocated wth the sze of the frm, the average educaton 12

14 level of employees, a frm s perceved competton ntensty, and a frm s nformal contact wth foregn busnesses. In the emprcal analyss that follows, we attempt to emprcally analyze the relatons among these dfferent actvtes carred out by the Chnese frms Lkelhood that a Frm Exports Analyzng the correlaton coeffcents can provde us wth very useful nformaton regardng the possble relatons among varables for the frms varous actvtes. To further analyze the mpact of the frms assocaton wth foregn frms and foregn knowledge nflow on the frms performance, we use multvarate regressons. Multvarable analyss wll enable us to concentrate on the key varables whle controllng other relevant varable such as the frms basc characterstcs. Frst, we study the factors possbly assocated wth the probablty that a frm exports. The queston nvolved s whether foregn nvestment and the assocated foregn knowledge nflow ncrease the lkelhood that a frm exports n the future. The dependent varable s a dummy varable denoted by EXPT_D. It takes the value of 1 when a frm exports n 1991 and 0 otherwse. There are four ndependent varables that we are most nterested n. Three out of the four are bnary varables. TECH takes the value of 1 when a frm reported havng receved foregn knowledge nflow that started before 1990, and zero f the frm does not receve foregn knowledge. FJV takes the value of 1 when the frm had formed a foregn jont venture wth a foregn partner before 1990 and 0 f the frm has not formed jont venture partnershp wth a foregn frm. FRN s the share of foregn ownershp n a frm. It represents the ntensty of foregn nvolvement n a frm. Fnally, CONTACT s another bnary varable ndcatng the frm s nformal assocaton wth foregn busnesses. Noted that there s no nformaton regardng the 13

15 tme when foregn ownershp began. Therefore, we would not be able to tell whch one of the two, foregn ownershp and export, precede the other from the estmated coeffcent. The equaton also has varables on the frms characterstcs such as total number of employees, share of varous ownershp, ndustry and locaton. Fnally, we nclude a varable EXPT0 representng a frm s ntal export status. Two alternatve ndces are used. One s a bnary varable, EXPT0_D, and the other s a contnuous varable, LEXPT0. EXPT0_D s calculated as the followng. If a frm has experenced technology transfer, t ndcates whether ths frm exports at the tme when technology transfer began. If a frm does not report technology nflow from foregn sources, t ndcates whether ths frm exports n 1980 (the frst year reported n the survey) or the begnnng of the frm s operaton (f the frms was establshed after 1980). Specfcally, f TECH s 1 for a frm and the year technology transfer started s T, then EXPT0_D takes the value of 1 f the frm exports n year T and 0 f the frm does not export n year T. When TECH s 0 for a frm, EXPT0_D takes the value of 1 f the frm exports n 1980 (or frst year of operaton f later than 1980) and 0 otherwse. LEXPT0 s obtaned n the same way, except t s a natural log value of the total export and thus a contnuous varable 11. Snce the dependent varable s a bnary varable, a logstc specfcaton s used to estmate the equaton. A logstc regresson assumes an S-shape curve and has the followng functonal form, y = 1/(1 + exp( X β+ ε )). In other word, the X s are lnearly related to the logt transformaton of the y s, where logt ( y = ln( y /(1 y )). Thus, the emprcal model we estmate s the followng: logt( expt91 ) ) = + β1tech + β 2FJV + β 3FRN + β 4 α CONTACT + γz + ε 11 LEXPT0 s calculated as ln (export0+1) n order to retan the frms wth zero ntal export n the sample. 14

16 Here Z s vector of varables of frms characterstcs ncludng LEMPL, IMPORT, SOE, PVT, FRN, COND, EXPT 0, as well as CITY and INDUSTRY. The estmaton results are summarzed n Table 7. The result n column A does not nclude EXPT0, a varable for the frm s ntal export condton. The results n columns B and C are obtaned usng the two ndcators for the frm s ntal export condton, LEXPT0 and EXPTO_D. For column C, we calculate the two margnal effects for each key varables, one at the sample mean and the other the average over all sample ponts. For the logstc model, the margnal effect s defned as the followng: margnal effect = where Λ(β T X) = E[y X] =Λ(β T X)[1 Λ(β T X)]β X 1 1+ exp( β T X) (2) Now we wll examne the estmaton results. If we don t control for the ntal export condton, as shown n the frst column n Table 7, the estmaton shows that knowledge nflow from foregn sources s postvely and sgnfcantly correlated wth the probablty that a frm wll export n the subsequent years. There s an appealng explanaton to ths fndng. That s, assumng the same ntal condtons among Chnese frms, foregn knowledge wll sgnfcantly ncrease the probablty of the recpents to export n the subsequent years. However, another scenaro could also generate outcome consstent wth the above postve results. Let s assume that when a domestc frm wants to export, t needs to acqure certan technology or knowledge to acheve that goal. Under ths assumpton, we wll observe that technology transfer precede export. When nterpretng the estmaton results, we would not be able to exclude the second scenaro. Nevertheless, what we can conclude that foregn knowledge nflows tend to precede the frm s export actvty, ether as a cause or possbly a necessary step. 15

17 The estmaton results also show that the share of foregn ownershp has a postve partal correlaton but the coeffcent s only margnally sgnfcant. When two frms both receved foregn knowledge, the one wth hgher foregn ownershp s more lkely to export, other thngs beng equal. Ths result suggests that jont equty wth foregn busnesses beneft local afflates not only n newer technology, but other export related benefts as well. One of such benefts mght be an easer access to a market network. On the other hand, formng a jont venture wth foregners does not provde more export related assstance n addton to TECH and FRN. Fnally, the results show that havng nformal contact wth foregn frms s postvely assocated wth the lkelhood that a frm exports. Ths fndng s consstent wth the assumpton that local frms beneft from nformal contact wth foregn busnesses. We now take a look at the coeffcents on frms basc characterstc varables. Frst of all, the sze of a frm s total employment s postvely assocated wth hgher probablty that a frm exports. Secondly, the coeffcent on CONTD s negatve and margnally sgnfcant. That suggests frms havng regular nformal meetng wth other domestc frms (buyers, supplers and so on) tend to be less export orented. Fnally, the coeffcents on some of the other basc varables such as SOE, PVT and OLD are not sgnfcantly dfferent from 0. Includng frms' ntal export status sgnfcantly mproves the model s overall estmaton (Column B and Column C). A frm s ntal export condton has a hgh predcton power as to whether the frm wll export n The estmated coeffcents on other varables mostly reman the same sgn. If we compare column C wth column A, the coeffcent for TECH and LEMPL reman postve and sgnfcant but the magntude decreases from 1.15 to 0.94 and from 0.90 to 0.71, respectvely. The coeffcent for FRN ncreases slghtly and the coeffcent for CONTACT remans postve but lost statstcal sgnfcance. 16

18 The margnal effects can provde us wth an ntutve nterpretaton for the magntude of the coeffcents. At the sample mean, havng foregn knowledge nflow (TECH=1) corresponds to a 23.4% hgher probablty that the frm wll export. Also, one percentage hgher n foregn ownershp wll ncrease the probablty that the frm wll export n 1991 by 0.4%. If we use the average of all sample frms for calculaton, the above margnal effects are about 11.3% and 0.2% respectvely. As mentoned above, ntal export condton has a hgher predcton power on a frm s future export. At the sample mean, ntal export status wll ncrease the lkelhood that the frm exports n 1991 by 74%. If calculated as the average of all sample ponts, the margnal effect of ntal export status s about 36%. There s not any sgnfcant dfference between older frms and newer frms n ther export. The lkelhood that a frm exports n 1991 s smlar for frms started before the economc reform to those started after the reform n 1978, other factors beng equal. Ths s somewhat surprsng. We would naturally expect newer frms to be more export-orented than older frms; gven that there have been dramatc changes n the economc system. There are three explanatons that mght be relevant to the fndng. Frst, old frms and new frms dffer greatly n many of ther characterstcs, such as foregn ownershp. When these other varables are ncluded n the estmaton, there may not be any addtonal effect for the varable OLD. Second, older frms may have other advantages that are not accounted for. For example, Old frms have accumulated more human captal. The last explanaton s related to the role of government n Chna. Although frms have ganed much autonomy n management decson makng durng the reform, government agences at varous level stll play an mportant role n frms exportng actvtes. Consequently, old state frms may possess advantages over prvate frms for ther longer and closer relatons wth government agences. 17

19 The sze of the frm s sgnfcantly assocated wth the frm s probablty to export n the future. On average, a one-percent ncrease n a frm s employment n 1988 ncrease s assocated wth nearly 9% hgher probablty that a frm export n At the sample mean pont, the number s approxmately 18%. Ths suggests that larger frms have the advantage n penetratng foregn market because they have more resources Total Amount of a Frm s Export The results n the prevous secton ndcate that foregn knowledge nflow sgnfcantly ncreases the probablty that a frm wll export n the subsequent years. In addton, foregn ownershp s also assocated wth hgher lkelhood that a frm exports. In the followng two sectons, we wll explore further the extent of the postve effect that technology transfer has on the export of local Chnese frms. In ths secton, we wll examne how the total amount of a frm s export s affected by foregn technology nflow. In the next secton, we wll analyze how the share of a frm s export n total output s affected. The varable we wll examne n ths secton s the natural log of a frm s total export value n We frst estmate the model usng an ordnary lnear specfcaton (OLS). However, there s a problem n applyng the log specfcaton to a censored data. Frms that do not export wll be dropped from the sample when takng the log of the total export. If we assume that when the desred level of export based on the characterstcs of a frm s zero or negatve, the observed level of export s zero, then droppng these frms may generate based estmates. Unfortunately, t s not feasble to apply the Tobt specfcaton whle mantanng the log lnear structure, as the logarthm of zero s undefned. To overcome ths problem, we use a modfed Tobt model (or threshold Tobt model) ntroduced by Eaton and Tamura (1996). Ths method has 18

20 been appled n the context of estmatng trade volume and foregn drect nvestment (e.g. Rauch 1999, We 1998). The modfed Tobt specfcaton s defned as followng: ln( EXPT91 + A) = X β + µ = ln( A) f X f X β + µ β + µ > ln( A) ln( A) (3) In the model, A s a threshold parameter to be estmated from the model, µ s a normally dstrbuted..d varable wth mean zero and varance σ 2. In ths specfcaton, when X β+ µ exceeds a threshold value, ln( A), there wll be a postve export; when X β+ s below the µ threshold value, the realzed level of a frm s export s zero. The ndependent varables nclude the same varables as n equaton (2), such as TECH, FRN, FJV, CONTACT, and EXPT0_D. The model also ncludes varables of the frms basc characterstcs such as LEMPL, SOE, PVT, CITY and INDUSTRY. In addton, we nclude the share of mported. The reason s many beleve that new and advanced technologes are emboded n the mported machnes. The threshold model s estmated usng the maxmum lkelhood method. The results for both the OLS and the modfed Tobt model are lsted n Table A total of 164 frms are ncluded n the OLS estmaton and 307 n the Tobt model. For most of the varables, the coeffcents from the two models have the same sgn. The ones obtaned from OLS are generally not sgnfcantly statstcally. In the subsequent dscussons, we wll focus on the results obtaned from the Tobt estmaton. The coeffcent on TECH s 0.77 and sgnfcant at 10% level. That suggests that the desred level of export n 1991 s at least 77% hgher for a frm that experenced foregn knowledge nflow before 1990 compared to one wth no foregn 19

21 knowledge nflow, other thngs beng equal. The coeffcent for IMPORT 13 s also postve but only margnally sgnfcant (at 13% level). Ths ndcates that more mported machnes s assocated wth hgher export from a frm, consstent wth the emboded technology assumpton. Foregn ownershp s postvely related to desred level of a frm s export. A onepercentage pont ncrease n a frm s foregn ownershp can result n a 2% ncrease n total export. Informal contact wth foregn busnesses s also assocated wth hgher level of desred export. However, FJV s not sgnfcantly related wth the desred level of a frm s export. These results are all consstent wth those obtaned n the prevous secton. Lke n the results for EXPT_D, a frm s ntal export condton s a strong predctor for ts export n In addton, a frm s total employment s postvely assocated wth ts desred level of export. These agan are consstent wth the earler results. One dfferent result s that the coeffcent on PVT, the share of prvate ownershp, s postvely and sgnfcant. The coeffcent s 4 tmes hgher than that of foregn ownershp. Assumng that prvate frms are more market orented and have less government nvolvement, ths result ndcates that frms operatng under more market orentaton are more motvated n extendng ther sale outward. To test the robustness of the model, we estmated the equaton wth alternatve dependent varables. One such alternatve s the natural log of a frm s export 4 years after foregn technology nflow began. 14 The results are consstent wth the earler outcome lsted n Table 8. They show that foregn knowledge nflow s assocated wth hgher desred level of a frm s 12 The OLS estmaton only ncludes frms wth postve export. The TOBIT estmaton ncludes all frms. In obtanng the TOBIT estmaton, we frst estmate an OLS usng ln(expt+1). The resultng coeffcents are then used as the ntal values for the Tobt estmaton. 13 IMPORT, the share of mported machnes n the frm s ntally ncluded n the prevous equatons on EXPT_D. But t s later dropped snce t has not sgnfcant effect. Droppng t does not affect the over model fttng. 14 Natural logs of a frms export 1, 2 and 3 years after foregn technology nflow started are also used as alternatve dependent varables to test the model. The results are not ncluded n ths paper. 20

22 future export, for example three or four years after the knowledge nflow began. However, foregn technology does not seem to sgnfcantly ncrease a frm s export mmedately. The results from ths secton are consstent wth those from the prevous secton. Foregn technology nflow tends to precede the frm s export. Technology transfer s also assocated wth hgher level of a frm s future export. In addton, the results show that there may be a lag between technology nflow and hgher export, suggestng that t takes tme for the domestc frms to absorb and adapt the new technology before they can ncrease ther export The Share of Export n Total Output The emprcal analyss so far suggests that frms that have receved foregn knowledge durng the 1980s not only are more lkely to export n 1991, they export more n total amount as well. In ths secton, we further study the mpact of foregn partcpaton by examnng the share of a frm s export n ts total output. We beleve that examnng the share of export n frm s total sale s mportant. Suppose that a frm s assocated wth foregn busnesses such as equty nvolvement and as a result receves technology transfer; based on the results from above, the frm wll be lkely to start exportng and export more. There are two alternatve ways to acheve ths outcome. On one hand, the frm could dvert part of ts domestc sale to the world market. On the other hand, the frm could also ncrease ts producton and ncrease ts export at the same tme. In the frst case, the share of export n total output wll ncrease dramatcally whle n the second case t need not. Both outcomes are good for the domestc market f we are more concerned wth export but the second case s more desrable because t s assocated wth economc expanson for the domestc frms. The varable concerned n ths secton s the share of a frm s export n ts total sale n 1991 (PCEXPT). Snce the value of the dependent varable s censored on both 0 and 100 percent, 21

23 we use ts logt transformaton for estmaton. 15 About 240 frms are ncluded n the regresson. More than half of the frms do not export whle about 10 percent of the frms export 100 percent of ther output. The results n Table 9 ndcate that the share of export n total output s not sgnfcantly hgher for frms that have experenced foregn knowledge nflow durng the 1980s. However, the share of mported machnes n the frm does a postve and sgnfcant assocaton wth hgher export share. The margnal effect on IMPORT calculated wth the formula n secton 3.1 shows that a 1-percentage ncrease n IMPORT corresponds to up to more than 1 percent hgher export share n a frm s total output. In addton, the ntensty of foregn partcpaton does not seem to be assocaton wth the export share of a frm s export. The coeffcents on FRN, the share of foregn ownershp, and FJV, foregn jont venture ndcator, are nsgnfcant. Once agan, a frm s ntal export status and total employment are mportant predctors for a frm s export share n Fnally, there s evdence that old frms, frms started operaton before 1980, tend to have smaller export share. The coeffcent s not sgnfcant. The results from the three sectons ndcate that foregn knowledge nflow brngs postve effects on local frms' export actvty. It seems that the results for EXPT_D are much stronger than those for LNEXPT and PCEXPT. Part of the reason s that there may be more measurement error n the latter. Ths s plausble, especally for usng survey data. It s probably easer to answer a yes or no queston such as whether the frm exports n a certan year than try to produce a precse amount or percentage. Nonetheless, the analyss does provde some support that foregn 15 Logt (PCEXPT)=Ln [PCEXPT/(1-PCEXPT)]. In the transformaton, 0 and 1 are replaced wth 1E(-15) and 1-1E(-15). In addton, both one-sded and two-sded Tobt estmaton are also used and the results are smlar. 22

24 knowledge has postve effect on Chnese frm s export. Frms that have receved foregn knowledge durng the 1980s are more lkely to export and to export more. More mportantly, ths s lkely to be the result of expanded producton rather than from dvertng sales from domestc to nternatonal markets. In the next two sectons, we wll examne the effect of knowledge nflow on the growth of a frm n terms of both employment and producton Employment, Output and Ther Growth In the prevous three sectons, we study the effect of foregn knowledge on frm s export. In ths secton, we examne the assocaton between foregn knowledge nflow and the frm s employment, producton ncludng ther growth. Frst, we examne the effect of foregn knowledge nflow on a frm s employment and ts growth rate. Then we look at the frms producton and ts growth. When studyng the frms employment, three dependent varables are used but the basc models are the same. LEMPL88 and LEMPL91 are the logs of total numbers of employees n 1988 and n 1991 respectvely. GEMPL s the growth rate of total employment from 1988 to Snce we use to calculate growth rate, the analyss ncludes only frms that ether have not experenced foregn knowledge nflow or have receved foregn knowledge begnnng before The followng are the basc models. lempl88 I = α + β1 TECH I + β 2 FJVI + γx I + ε I (4) lempl91 I =α+β 1 TECH I +β 2 FJV I +γx I +ε I (5) gempl I =α+β 1 TECH I +β 2 FJV I +γx I +ε I (6) In these three equatons, X ncludes EXPT0, SOE, PVT, FRN, IMPORT, as well as CITY and INDUSTRY. For equaton (6), LEMPL88 s also ncluded control for the effect of ntal frm sze. The results are summarzed n Table 10, Panel A. 23

25 From the frst two columns, we fnd that the results for LEMPL88 and LEMPL91 are essentally the same. TECH s postvely assocated wth a frm s future employment level. The sgnfcance level s only margnal (15%). Furthermore, foregn ownershp and foregn jont venture partnershp are also assocated wth hgher future employment. In addton, frms wth large state ownershp tend to have more employees, other thngs beng equal. For the regresson on employment growth, we obtan two estmates. In the frst estmaton, TECH enters the model alone and n the second specfcaton, an nteracton term between TECH and a tme varable, TECHYR s added. It s defned as the product of TECH and the number of years from the frst year of foregn knowledge nflow to The goal s to capture the possble decreasng effect of TECH on employment growth over tme. The results n the thrd show that, when TECH enters the equaton alone, the coeffcent s postve but nsgnfcant. But when the nteracton term s ncluded, the overall model fttng mproves sgnfcantly. The coeffcent for TECH s postve and that for TECHYR s negatve. Both are sgnfcant. These suggest that foregn technology nflow sgnfcantly ncrease the growth rate of a frm s total employment. However, the effect seems to decrease over tme. For example, foregn knowledge nflow started n 1987 corresponds to about 14% hgher employment growth between 1988 and However, the effect amounts to only 5% for the knowledge nflow started n 1985 and 0 for those started before Next, we analyze the effect of foregn knowledge on the frms producton. We reestmated the models (4)-(6) usng the level and growth of a frm s total output as dependent varables. Output and growth rate evaluated based on constant prces and current prces are both 16 About one thrd of the frms (67 out of 203) ncluded n the regresson receved foregn knowledge nflow, n about half of these frms (36 frms) the event occurred n and before

26 used n the equaton, ncludng LPROD88N, LPROD91N, GPRODN, LPROD88R, LPROD91R, and GPRODR. 17 The frst three varables are a frm s producton (n 1988 and 1991, natural log) and growth rate between 1988 and 1991 evaluated on constant prces. The last three are defned smlarly but evaluated on current prces. The estmaton results are n Panel B and Panel C of Table 10. The estmaton results are not as clear as those obtaned for employment. When we examne the estmaton result regardng future output level, TECH seems to be assocated wth hgher future producton only when output value s based on constant prces. In addton, foregn jont venture partnershp s not related a frm s future producton. For the results regardng the growth rate of producton, results smlar to that for employment growth are obtaned when current prces are used for evaluaton. That means, when TECH enters the model along, not sgnfcant coeffcent s obtaned. When nteracton term TECHYR s ncluded, the model generates postve coeffcent for TECH and negatve coeffcent for TECHYR. Both are sgnfcant. No such results are obtaned when output s valued on constant prces. In addton, foregn ownershp and formng a foregn jont venture partnershp before 1988 seem to be postvely and sgnfcantly assocated wth the output growth between 1988 and The analyss n the secton ndcate that foregn knowledge nflow not only have a postve mpact n facltatng local frms export, t also seems to have postve effects on a frm s future employment and output growth. These effects, however, decrease over tme and may only be present wthn the short run. We mght thnk of t as a short-term effect rather than a long run 17 In ths secton, we use growth rate calculated through current-valued as well as current valued producton. The reason s that durng the perod, the base year for constant prce was changed from 1980 prce (for 1988 and 1989) to 1990 prce (for 1990 and 1991) and there s no converson for the two. In ths case, both measures are somewhat flawed. We ncluded all ndustry dummes, assumng that the nflaton s the same for all frm n one ndustry and try to explan the dfference from the ndustry mean. 25

27 growth rate ncrease. In addton, frms also seem to draw addton benefts from foregn ownershp and formng a foregn jont venture wth foregn busnesses to gan employment and output growth Tranng of Domestc Supplers In the prevous sectons, we analyzed the possble benefts a Chnese frm may accumulate through drect equaton partnershp wth foregn busnesses and foregn knowledge nflow. As we dscussed earler, n addton to the benefts receved by ther drect recpents, foregn technology nflow may generate spllover effect to other domestc frms as well. The assumpton s that when a Chnese frm s nvolved wth foregn frms and has obtaned foregn knowledge, other local frms may beneft from the above process through busness relatons wth the frm. In ths secton, we analyze one aspect of such spllover effects. The basc queston s whether a frm s more lkely to provde tranng to ts domestc supplers when t has receved foregn knowledge nflow. At the same tme, we also look at the relaton between foregn ownershp and provdng tranng to domestc supplers. The dependent s a bnary varable valued at 1 when a frm does provde tranng to ts domestc suppler and 0 otherwse. Once agan, a logstc specfcaton has to be used for the estmaton. logt( TRAINING ) = α + β TECH + β FRN + β FJV + γx + ε (6) I In ths equaton, X nclude the EXPT_D, LPEOUT, LEMPL, AGE, SOE, PVT, LSPEC, EDU, as well as CITY and INDUSTRY. All the frms ncluded n the regresson have at least one domestc suppler. The estmaton results presented n the frst three column of Table 11 provde supportng evdence to our assumpton. Frms that have receved foregn knowledge nflow are more lkely to provde tranng to ther domestc supplers. The average margnal effect (thrd column) suggests that frms wth foregn knowledge nflow are 10% more lkely to provde 26

28 tranng to ther domestc supplers). That means foregn knowledge nflow beneft not only ts drect recpent but other related domestc frms as well. The result s consstent wth the assumpton that foregn knowledge has spllover effect to domestc economy. We offer one explanaton to the postve correlaton between recevng foregn knowledge nflow and provdng tranng to domestc supples, though there could be dfferent reasons. We beleve that whle a frm receved foregn technology, t s lkely to rase ts product standard and thus requre nput of hgher qualty. One soluton s to mport materal from abroad. However, f mport s too costly or prohbted (restrcted) by government regulatons, an alternatve soluton s to tran domestc supplers so hgher qualty nput could be acqured locally. The results also show that a frm s output per employee n natural log, LPEOUT, s postvely assocated wth the lkelhood that t provdes tranng to ts domestc supplers. Ths fndng probably reflects the fact that relatvely hgh captal-ntensve frms requre hgher nput qualty and thus more lkely to tran ther domestc supplers. The average margnal effect for LPEOUT s That means a 1% ncrease n per employee output corresponds to 5% hgher probablty that a frm provdes tranng to ts domestc nput suppler. From the estmaton results, we fnd that foregn ownershp and havng a foregn jontventure partner are not assocated wth how lkely a frm provdes tranng to ts domestc supplers. It s not too surprsng that hgh foregn ownershp does not ndcate more tranng to domestc supplers, as the effect s probably reflected through TECH. To further examne the effect of foregn ownershp on the probablty a frm provdes tranng, we replace the varable FRN wth fve foregn ownershp varables, each correspondng to a dfference source. Four countres/regons, the Unted State, Hong Kong and Macao, Tawan, and Japan, can be dentfed n our sample and the rest are categorzed as other. The fve varables are denoted as FRN_US, 27