54866 THE WORLD BANK Research Observer EDITOR Emmanuel Jimenez. World Bank CO-EDITOR Gershon Feder. World Bank EDITORIAL BOARD Susan Collins. Georgetown University Angus Deaton. Princeton University Barry Eichengreen. Cniversity of California-Berkeley Emmanuel Jimenez. World Bank Benno Ndulu. World Bank Howard Pack, University of Pennsylvania Luis Serven. World Bank Sudhir Shetty. World Bank Michael Walton. World Bank The World Bank Research Observer is intended for anyone who has a professional interest in development. Observer articles are written to be accessible to nonspecialist readers; con- tributors examine key issues in development economics, survey the literature and the lat- est World Bank research. and debate issues of development policy. Articles are reviewed by an editorial board drawn from across the Bank and the intern ational community of econo- mists. Inconsistency with Bank policy is not grounds for rejection. The journal welcomes editorial comments and responses. which will be considered for pub- lication to the extent that space permits. On occasion the Observer considers unsolicited contributions. Any reader interested in preparing such an article is invited to submit a proposal of not more than two pages to the Editor. Please direct all editorial correspon- dence to the Editor. The World Bank Research Observer, 1818 H Street. NW, Washington. DC 20433. USA. The views and interpretations expressed in this journal are those of the authors and do not necessarily represent the views and policies of the World Bank or of its Executive Directors or the countries they represent. The World Bank does not guarantee the accuracy of data included in this publication and accepts no responsibility whatsoever for any consequences of their use. When maps are used, the boundaries. denominations. and other information do not imply on the part of the World Bank Group any judgment on the legal status of any territory or the endorsement or acceptance of such boundaries. For more information, pLease visit the Web sites of the Research Observer at www.wbro.oxfordjournaLs.org, the WorLd Bank at www.worldbank.org, and Oxford University Press at www.oxfordjournals.org. THE WORLD BANK Research Observer Volume 23 · Number 2 · Fall 2008 Symposium on FOI CoLLecting the Pieces of the FDI Knowledge Spillovers Puzzle Roger Smeets 107 Can Survey Evidence Shed light on Spillovers from Foreign Direct Investment? Beata S. Javorcik 139 The Growing Phenomenon of Private Tutoring: Does It Deepen Human Capital, Widen Inequalities, or Waste Resources? Hai-Anh Dang and E Halsey Rogers 161 Can Biological Factors Like Hepatitis B Explain the Bulk of Gender Imbalance in China? A Review of the Evidence Monica Das Gupta 201 Industrial location in Developing Countries Uwe Deichmann, Somik V. Lall, Stephen J. Redding, and Anthony J. Venables 219 Subscriptions A subscription to The World Rank Research Observer (JSSN 0257-3032) comprises 2 issues. Prices include postage; for subscribers outside the Americas, issues are sent air freight. Annual Subscription Rate (Volume 23, 2 issues, 2008) Academic libraries Print edition and site-wide online access: US$140/J:93/140 Print edition only: US$133/J:S9/133 Site-wide online access only: US$133/J:89/133 Corporate Print edition and site-wide online access: US$208/fl39/20S Print edition only: US$198/fl32/198 Site-wide online access only: US$198/fl32/198 Personal Print edition and individual online access: US$50/f33/50 Please note: US$ rate applies to US & Canada, Euros applies to Europe, UKJ: applies to UK and Rest of World. Readers with mailing addresses in non-OECD countries and in socialist economies in transition are eligible to receive complimentary subscriptions on request by writing to the UK address below. There may be other subscription rates available; for a complete listing, please visit www.wbro.oxfordjournals.org/subscriptions. Full pre-payment in the correct currency is required for all orders. Payment should be in US dollars for orders being delivered to the USA or Canada; Euros for orders being delivered within Europe (excluding the UK); GBP sterling for orders being delivered elsewhere (i.e. not being delivered to USA, Canada, or Europe). All orders should be accompanied by full payment and sent to your nearest Oxford journals office. Subscriptions are accepted for complete volumes only. Orders are regarded as firm, and payments are not refundable. Our prices include Standard Air as postage outside of the UK. Claims must be notified within four months of despatch/order date (whichever is later). Subscriptions in the EEC may be subject to European VAT. If registered, please supply details to avoid unnecessary charges. For subscriptions that include online versions, a proportion of the subscription price may be subject to UK VAT. Subscribers in Canada, please add GST to the prices quoted. Personal rate subscriptions are only available if payment is made by personal cheque or credit card, delivery is to a private address, and is for personal use only Back issues: The current year and two previous years' issues are available from Oxford University Press. Previous volumes can be obtained from the Periodicals Service Company, 11 Main Street, Germantown, NY 12526, USA. E-mail: psc@periodicals.com. Tel: (518) 537-4700. Fax: (518) 537-5899. Contact information: journals Customer Service Department, Oxford University Press, Great Clarendon Street, Oxford OX2 6DP. UK. E-mail: jnls.cust.serv@oxfordjournals.org. Tel: +44 (0)1865353907. Fax: +44 (0)1865353485. In the Americas, please contact: journals Customer Service Department, Oxford University Press, 2001 Evans Road, Cary, NC 27513. USA. E-mail: jnlorders@oxfordjournals.org. Tel: (800) 852-7323 (toll-free in USA/Canada) or (919) 677-0977. Fax: (919) 677-1714. In Japan, please contact: journals Customer Service Department, Oxford University Press, 4-5-10-8F Shiba, Minato-ku, Tokyo, 108-8386, japan. E-mail: oupjnl@po.iijnet.or.jp. Tel: +81 3 5444 5858. Fax: +81 3 34542929. Postal information: The World Rank Research Observer (ISSN 0257-3032) is published twice a year, in Feb. and Aug., by Oxford University Press for the International Bank for Reconstruction and Development/THE WORLD BANK. Postmaster: send address changes to The World Rank Research Observer. journals Customer Service Department, Oxford University Press. 2001 Evans Road, Cary, NC 27513-2009. Communications regarding original articles and editorial management should be addressed to The Editor, The World Rank Research Observer, The World Bank, 1818 H Street. NW. Washington, D.c. 20433, USA. Oxford Journals Environmental and Ethical Policies: Oxford journals is committed to working with the global community to bring the highest quality research to the widest possible audience. Oxford journals will protect the environment by implementing environmentally friendly policies and practices wherever possible. Please see http://www.oxfordjournals.org/ ethicalpolicies.html for further information on Oxford journals' environmental and ethical policies. Digital Object Identifiers: For information on dois and to resolve them, please visit www.doi.org. Permissions: For information on how to request permissions to reproduce articles or information from this journal. please visit www.oxfordjournals.org/jnls/permissions. Advertising: Inquiries about advertising should be sent to Helen Pearson. Oxford journals Advertising, PO Box 347. Abingdon OX14 IGJ, UK. E-mail: helen@oxfordads.com. Tel: +44 (0)1235 201904. Fax: +44 (0)8704 296864. Disclaimer: Statements of fact and opinion in the articles in The World Bank Research Observer are those of the respective authors and contributors and not of the International Bank for Reconstruction and Development/THE WORLD BANK or Oxford University Press. Neither Oxford University Press nor the International Bank for Reconstruction and Development/THE WORLD BANK make any representation, express or implied, in respect of the accuracy of the material in this journal and cannot accept any legal responsibility or liability for any errors or omissions that may be made. The reader should make her or his own evaluation as to the appropriateness or otherwise of any experimental technique described. Paper used: The World Bank Research Observer is printed on acid-free paper that meets the minimum requirements of ANSI Standard Z39.48-1984 (Permanence of Paper). Indexing and abstracting: The World Bank Research Observer is indexed and/or abstracted by AHliINPOHM, CAB Abstracts, ClIrrent Contents/Social and Behavioral Sciences, Journal of Economic LiteratlIre/EconLit, PAIS International, RePEc (Research in Economic Papers), Social Services Citation Index, and Wilson Business Abstracts. Copyright D 2008 The International Bank lor Reconstruction and Development/THE WORLD BANK All rights reserved; no part of this publication may be reproduced, stored in a retrieval system. or transmitted in any form or by any means, electronic, mechanical, photocopying, recording, or otherwise without prior written permission of the publisher or a license permitting restricted copying issued in the UK by the Copyright Licensing Agency Ltd. 90 Tottenham Court Road. London WIP 9HE, or in the USA by the Copyright Clearance Center, 222 Rosewood Drive, Danvers, MA 01923. CoLlecting the Pieces of the FDI Knowledge Spillovers Puzzle Roger Smeets Recent surveys of the empirical literature have concluded that the evidence is mixed on the magnitude. direction. and even existence of knowledge spillovers from foreign direct investment (PDI). This article reviews the recent theoretical and empirical literature that responds to these inconclusive results and considers three main issues: spillover channels. mediatinfJ factors. and PDI heterogeneity. Studies that take into account individual spil- lover channels .find robust evidence of knowledge spillovers from PDI. Studies on the importance of mediating factors and PDl heterogeneity are less conclusive and could benefit from greater convergence in methodologies and greater specificity in the spillover channels of interest. More generally. many studies do not properly distinguish between knowledge spillovers and knowledge transfers. and empirical studies seem to greatly out- /lumber theoretical studies. JEL codes: F23. 033 In the face of difficulties associated with capturing spillover effects and the multitude of factors that can influence the extent of spillovers in each economy, we caution researchers about drawing generalized conclusions about the existence of externalities associated with [foreign direct invest- ment] .... (Javorcik and Spatareanu 2005. 47) Over the past decade or so a large body of research has examined knowledge spillovers from foreign direct investment (FDI). At several points along the way scholars have paused to take stock of the evidence (Blomstrom and Kokko 1998; Saggi 2002; Gorg and Greenaway 20(4). The verdict has largely been inconclu- sive. Indeed, the empirical inconclusiveness has become so infamous that virtually every study reviewed here begins with this observation as its main motivation. Explanations for the lack of conclusive results have focused on methodological and measurement issues (Gorg and Strobl 2001). but this sort of approach has recently been disputed (Lipsey and Sjoholm 2(05). ~l' The Author 2008. Published by Oxford University Press on behalf of the rnternational Bank for Reconstruction and Development! THE WORlD BA~". All rights reserved. For permissions. please e-mail: journals.permissions@oxfordjournals.org doi:1O.1093 wbro/lknO()3 Advance Access publication March 19. 2008 23:107-138 The literature has developed in several directions to account for the ambiguity in earlier work. This study reviews these contributions, both theoretical and empirical. To provide some structure in a rapidly expanding field and to identify which approach or combination of approaches is likely to yield the most promi- sing results, the study is structured around three themes. 1 More insight into the conditions under which knowledge spillovers from pm are most likely to arise is especially important for developing countries. The highly ambiguous evidence to date on the existence of knowledge spillovers from pm does not seem to warrant the large sums of money spent by governments to attract PDI. 2 After setting the stage in the follOWing section, the article is then structured around figure 1, a representation of the pm knowledge spillover process and the pieces of the puzzle that may affect it. 3 The section on opening the black box of PDl knowledge spillovers discusses the research on vertical linkages, worker mobi- lity. and demonstration effects. This is followed by a review of the evidence on the influence of mediating factors, focusing on the role of absorptive capacity and spatial proximity. The next section analyzes the effect of pm heterogeneity, exami- ning studies on the role of ownership structure, parent-firm nationality. and motives for PDl as factors influencing the extent of knowledge spillovers. The last section points to some directions for future empirical and theoretical research. Setting the Stage Much econometric work has been done in this area [on knowledge spillovers from PDI]. but the results on the importance of spillovers are mixed at best (Gorg and Greenaway, 2004. 172). Inward pm stocks increased in all regions of the world between 1980 and 2006, especially during the early 2000s (figure 2). Developed countries were the most Figure 1. FDI Knowledge Spillovers Framework Source: Author's schematization, 108 The WorM Bank Research Observer. vol. 23, no, 2 (Fall 2(08) important benefactors. Developing countries. especially in Africa. lagged far behind. The picture is similar for FDI as a share of GDP (figure 3). On this measure countries in Latin America were catching up somewhat during the early 2000s. but Africa is still far behind. The overall increase in inward FDI may partly explain the rising interest scho- lars have shown in knowledge spillover effects. However. as the countries that stand to gain most from such spillovers are also those for which inward FDI is still a small part of their economic activity. one could wonder whether the attention devoted to knowledge spillovers from FDI as a (crucial) factor for economic devel- opment has not been disproportionate. Following Javorcik (2004b), this survey defines knowledge spillovers at the firm level as knowledge created by one firm (a multinational enterprise) that is used by a second firm (a host-country firm) for which the host-country firm does not (fully) compensate the multinational enterprise. This definition does not include pecuniary spillovers (nominal gains resulting from quality increases that are not fully reflected in prices) or competition effects (changes in market structure caused by the entry of a multinational enterprise). It does distinguish between knowledge spillover and knowledge transfer (the purposeful or intended diffusion of knowledge from one firm to the other, which creates no externality). The literature has identified three main channels along which knowledge may spill over from a multinational enterprise to a local firm (Saggi 2006) (see figure]). Demonstration effects involve the imitation. or reverse-engineering, by host-country firms of the products or practices of multinational enterprises. Figure 2. Total Inward FDI Stocks, by Region, 1980-2006 6.000 5,000 ~ 4.000 -4-elScpe 4ft o - - Nat! Nnsica ~ 3.000 - - JIIil ia & Oceania c ~ - latinNnsica iii 2.000 -Africa 1,00: ~~;~==~:~~~~~~~;;;;;;;;;~ $>~ , ~~ .I' . '. .riP ~q;. . ; " .# ~~ #4 #' .I' . Source: UNCTAD 2007. Roger Smeels 109 Figure 3. Inward PDI Stocks as Percent of GDp, by Region, 1980-2004 iii 19805 819905 82000, (2000-04) ~.------------------------------------------------------- 45+_--- 40+_--- ~35+--- " 30-1---- ~ 25+--....J i 21 ~ :. 15 10 5 o elScpe Oceania NathAmerica LatinAmelica Alia Atica Source: UNCTAD 2007; Heston, Summers, and Aten 2006. Worker mobility allows employees trained by the multinational enterprise to apply their knowledge in the local firm. Upstream and downstream vertical (interindus- try) linkages involve the spillover of knowledge from the multinational enterprise to its suppliers and customers. Much empirical research has tried to identify the direction, size, and scope of knowledge spillovers from multinational enterprises to local firms. One of the major challenges these studies face is measuring knowledge spillovers. The usual approach has been to assume that the major knowledge spillover effect is on the receiving firm's productivity, often measured by changes in the receiving firm's productivity following entry of the multinational enterprise, controlling for other observable determinants of productivity. (This survey does not address issues of measurement.) The first major review of this empirical literature appeared in 1998 (Blomstrom and Kokko 1998). It shows that most studies (many of them multiple case studies) consider the effects of knowledge spillovers from multinational enterprises through backward linkages (linkages to supplier industries). Multiple case studies tend to find evidence of the existence of knowledge spill- overs more often than econometric studies do. Gorg and Strobl (2001) conduct a meta-analysis of 21 econometric studies of the knowledge spillover effects of FDI to determine whether differences in research design, methodology, and data can at least partially explain the ambiguity of the results. The econometric studies included in the analysis estimate models of the following form: (1) where Yijt is some measure of productivity of firm i active in sector j at time t; PDI is a measure of the presence of PDI; X is a vector of firm-level control variables that are known 110 The World Bank Research Observer, vol. 23, no. 2 (Fall 2008) to affect productivity (such as own investments in R&D and human capital): Z is a vector of industry-level control variables (for example, market concentration): and e is an error term. The /3's are the parameters to be estimated, and /31 is the parameter of interest. Two problems affect models of this type. First. the measures of productivity vary across studies. making comparisons difficult. Some look at total factor pro- ductivity (TFP), while others consider labor productivity. Second, the endogeneity of FDI (the fact that FDI may be attracted to more-productive countries, regions, or sectors, reversing the causal mechanism) is not always properly accounted for. which could bias the estimation results. G6rg and Strobl (2001) find that cross-section studies find more significant evidence of positive knowledge spillovers than panel studies do. This suggests that unobserved firm heterogeneity may be present. Their results also indicate that the way FDI is measured may influence the results and that there may be publication bias in favor of studies that find evidence of significant positive knowledge spillovers. Yet Lipsey and Sjoholm (2005) show that results for different countries tend to diverge even when similar estimation techniques are used on similar data over similar time periods. They conclude that heterogeneity in host-country factors are the most likely source of the inconclusiveness of empirical research. Gorg and Greenaway (2004) survey more than 40 econometric studies, mainly at the microeconomic level. Their review indicates that the empirical evidence is at best ambiguous. with 20 cases finding evidence of positive spillovers. 17 cases finding insignificant results, and 8 cases finding evidence of significant negative knowledge spillovers. The studies they review cover different periods and countries and use both cross-sectional and panel designs. 4 Opening the Black Box of FDI Knowledge Spillover Mechanisms One of the drawbacks of these [empirical FDI spillover] studies is that they treat the specific mechanisms by which the spillovers are supposed to occur as a "black box" (Gorg and Strobl 2005, 695). The empirical literature for a long time has not explicitly considered spillover channels other than knowledge spillovers from PDI through backward linkages. Indeed. the general empirical model specified in model 1 is the most frequently encountered in econometric tests of knowledge spillovers from FDI. As Gorg and Strobl (2005) argue, such an empirical specification disregards the existence and importance of knowledge spillover channels. It could very well be that {31 picks up the net effect of PDI (including adverse competition effects. for example) (Aitken Roger Smeets 111 and Harrison 1999).5 Empirical research has increasingly been trying to explicitly take into account the different spillover channels. Vertical Linkages Many of the studies opening the black box of knowledge spillovers from FDI have focused on knowledge spillovers through vertical linkages (Hoekman and Javorcik 2006; for brief reviews see Lin and Saggi 2005; Saggi 2006). Two early theo- retical contributions in this field are Rodriguez-Clare (1996) and Markusen and Venables (1999). Rodriguez-Clare focuses on the input-demand effects of multinational enter- prises. He constructs a model with monopolistic competition in the intermediate goods sector, which national firms and multinational enterprises use as inputs in the production of final goods. He assumes that multinational enterprises' final goods are more complex (that is, require a larger variety of inputs) than those of national firms, and that all firms have a "love of variety" for intermediate inputs. Accordingly, the entry of a multinational enterprise increases demand for inter- mediate inputs, which establishes the backward linkage. Because of monopolistic competition in the intermediates sector, the arrival of the multinational enterprise leads to an increase in the variety of available inputs. Final goods producers benefit because of the love of variety for inputs, which establishes the forward linkage effect. The Markusen and Venables (1999) model has a similar setup. However, they explicitly consider the intraindustry competition effect a multinational enterprise induces upon entry. Rodriguez-Clare (1996) effectively ignores this effect, con- sidering situations in which multinational enterprises are the only firms produ- cing in one of the two countries. These two studies thus look only at pecuniary spillovers and competition effects of FDt not at knowledge spillover effects. Lin and Saggi (2007) explicitly consider vertical technology transfer through backward linkages (from multinational enterprises to their local suppliers). They assume that upon entry a multinational enterprise can negotiate an exclusivity contract with a number of local suppliers. Only then will the multinational enter- prise engage in vertical technology transfer. This model does not consider know- ledge spillovers as in the definition of being an externality considered here. A number of empirical studies have been conducted in this area (table I), all of them estimating a modified version of model 1: Yijt = f3(J + f31FDIjt + f32 2)a?kt' FDIkt) + f33 2)a~kt' FDht) k#j k#j (2) + f32 X it + f33 Z jt + eijt where a~ is the output share flowing from industry j to industry k: aJk is the share of inputs used by industry j from industry k: i indexes the firm: j and k index the industry: t 112 The World Bank Research Observer, vol. 23, no. 2 (Fall 2008) Table 1. Empirical Results on Effects of Vertical Linkages, Worker Mobility, and Demonstration Effects on PDI Knowledge "UtHVV...,'" Channel Study Sample Effect Vertical linkages Javorcik (2004b) 4,000 firms in Lithuania, Positive effects through backward 1996-2000 linkages: no effects through forward linkages Javorcik and 13,129 firms in Romania, Positive effects through backward Spatareanu 1998-2003 linkages (2008) Kugler (2006) All manufacturing plants in Positive effects through backward Colombia. 1974-1998 linkages; no effects through forward linkages Bwalya (2006) 125 Zambian manufacturing Positive effects through backward firms. 1993-1995 linkages; no effects through forward linkages Schoors and van 1,084 firms in Hungary, Positive effects through backward der Tol (2001) 1997-98 linkages; negative effects through forward linkages Worker mobility Markusen and 304 manufacturing Positive Trofimenko establishments in Colombia. (2007) 1977-91 Gorg and Strobl 228 manufacturing firms in Positive (2005) Ghana. 1991-97 Poole (2007) Formal sector workers in Positive Brazil,1996-2001 Hale and Long 1. 500 firms in China, 2000 Positive (2006) Demonstration Cheung and Lin 26 provinces in China,1995- Positive effects (2004) 2000 Hale and Long 1, 500 firms in China. 2000 Positive (2006) ----- ---------------------- ---------- Source: Author's compilation. indexes time; and y. X, Z. and e are defined as in model 1. In this model {3r measures the effect of PDI in firm i's own sector, which can be interpreted as a demonstration effect; {32 captures the effect of FDI in sector k on the productivity of firm i in sector j. weighted by the share of output flOWing from sector j to k (that is. backward linkages); and {33 captures forward linkages. Javorcik (2004b) analyzes knowledge spillovers from multinational enterprises through backward and forward linkages in a panel of about 4.000 Lithuanian firms. She finds evidence of positive knowledge spillovers through backward but Roger Smeets 113 not forward linkages. Javorcik and Spatareanu (2008) also find evidence of positive knowledge spillovers through backward linkages, although only from multinational enterprises that share ownership with local firms. Kugler (2006) analyzes interindustry spillovers from FDI in eight Colombian manufacturing sectors. He finds strong and robust evidence of backward linkages and no evi- dence of forward linkages. Bwalya (2006) obtains a similar result for a sample of 125 Zambian manufacturing firms. Schoors and van der Tol (2002) find evidence of positive knowledge spillovers through backward linkages in Hungary but nega- tive spillovers through forward linkages. Moreover, they find that these inter- sectoral effects are statistically more important than the intrasectoral effect ({3d. It is questionable whether these empirical studies actually measure knowledge spillovers and not knowledge transfer. Indeed, in a study of more than 100,000 Indonesian manufacturing establishments, Blalock and Gertler (2008) refer to the evidence they find of local firm productivity increases through vertical linkages with multinational enterprises as knowledge transfers rather than spillovers. Javorcik and Spatareanu (2005) use survey data on the perceptions of managers in local Latvian and Czech firms. They find that intentional multinational enter- prise assistance is an important factor influencing local firms' productivity. Pack and Saggi (2001) provide a theoretical treatment of vertical technology transfer. These studies clearly demonstrate the importance of knowledge transfer instead of knowledge spillovers. Worker Mobility A second channel through which knowledge spillovers can flow is worker turn- over. The multinational enterprise is likely to provide some host-country workers with better training, education, and work experience than the average local firm does. If its workers eventually move to a local firm or start their own companies, they can apply the knowledge acquired from the multinational to the local firm's benefit. As the multinational enterprise is not compensated for this, this know- ledge constitutes a knowledge spillover. Fosfuri. Motta, and Ronde (2001) were among the first researchers to formally model this channel of multinational enterprise knowledge spillovers. In their model a firm must choose between FDI and exports to serve a foreign market. If it chooses FDI it must train host-country workers. When training is completed. both the multinational enterprise and local firms can bid to acquire the services of the trained local workers. Knowledge spillovers occur if the local firm wins the bid. Such a situation is most likely to occur if market competition is low and know- ledge easily transferable. because in this case the local firm has much to gain by obtaining the knowledge, and the cost of training an additional worker for the multinational enterprise is relatively low. Markusen and Trofimenko (2007) 114 The World Bank Research Observer, vol. 23. no. 2 (Fall 2008) model worker mobility as a channel for knowledge spillovers in a general equili- brium setting in which changes in the wages paid by firms attract experts from multinational enterprises. Two models are used in empirical research on knowledge spillovers through worker mobility. The first is a straightforward extension of model 1: (3) where SM denotes some measure of the presence of foreign workers (workers previously employed by a multinational enterprise's subsidiary). If knowledge spillovers diffuse through worker mobility. f31 should be positive. A second empirical specification analyzes knowledge spillovers through worker mobility at the individual level. by looking at wages: (4) where w denotes the (log of the) individual wage level. i in this case subscripts individuals and j subscripts firms. The underlying assumption is that wages are strongly correlated with marginal labor productivity. Hence positive knowledge spillovers through worker mobility again imply that f31 is positive. ~larkusen and Trofimenko (2007) test their model using plant-level data on a sample of 304 Colombian manufacturing establishments, employing at least 10 workers. Their results show that hiring foreign experts increases real wages at the hiring plant. This effect is both instantaneous (it occurs during hiring) and persistent (it remains even after the foreign expert has left the plant). Gorg and Strobl (2005) estimate a model similar to model 3 in a panel of 228 Ghanaian manufacturing firms. Their results indicate that a local firm's owner's previous experience in a multinational enterprise increases the local firm's pro- ductivity but only if the multinational enterprise is operating in the same sector as the local firm. Having an owner that received explicit training in the multi- national enterprise does not contribute Significantly to firm-level productivity. (The extent to which this result reflects the more general situation in which any foreign employee, not just the owner, hired by a local firm can spill over know- ledge remains unclear.) Poole (2007) analyzes knowledge spillovers through worker mobility at the worker level, using data on formal sector workers in Brazil in a model similar to model 4. She finds that an increase in the presence of foreign workers (5',\;1) increases wages, indicating that knowledge is spilling over from former multina- tional enterprises' employees to national firms. Hale and Long (2006) investigate spillovers from PDI in a sample of 1,500 firms in five Chinese cities. They find evi- dence that an increase in SM (as defined by Poole) increases firm productivity. Roger Smeets 115 Demonstration Effects Various definitions of demonstration effects can be found in the literature (Cheung and Lin 2004; Moran, Graham, and Blomstrom 20(5). Saggi (2002) defines demonstration effects as occurring through the imitation and reverse engineering of a multinational enterprise's products and practices by local (host country) firms. This definition largely fits the definition here of knowledge spillovers. Many of the studies reviewed by Gorg and Strobl (2001) and Gorg and Greenaway (2004) implicitly deal with knowledge spillovers through demon- stration effects, as the majority look for horizontal (intraindustry) knowledge spill- overs. By (Saggi's) definition demonstration effects occur mainly through these horizontal spillovers. Hence, the general empirical specification looks like the one in model 1. None of these studies hypothesizes or specifies how demonstration effects take place. Cheung and Lin (2004) shed some light on this issue. They study the effect of FDI on three types of patent applications in 26 provinces in China: invention patents (patents for new technical solutions). utility patents (patents for new technical solutions relating to the shape or structure of a product). and design patents (patents for new designs of shapes or patterns). They show that increased FDI in a province has a positive effect mainly on design patents. Since the content of such patents is most easily copied. they interpret this as evidence of demon- stration effects. Hale and Long (2006) also find some circumstantial evidence of demonstration effects through network externalities. Taking Stock The work on opening the black box of knowledge spillovers from FDI seems a promising strand of research. In addition to obtaining more detailed insights into the exact mechanisms along which knowledge spillovers may come about, it yields more consistent empirical results than previous black box research. A few concerns nonetheless remain. First. theoretical work on knowledge spillovers through vertical linkages is vir- tually absent. Most studies consider only pecuniary spillovers or knowledge trans- fer. Contributions in this field are much needed. It is not always clear that empirical studies are actually measuring knowledge spillovers and not knowledge transfers. Although the distinction may seem irrelevant from the host country's perspective, the policy implications of each are very different (Blalock and Gertler 2005, 2008). Empirical researchers in this field should at least be aware of this potential bias. Second. much of the inferred effects of knowledge spillovers in the worker mobility literature are based on changes in wages. This assumes a very strong 116 The World Bank Research Observer. vol. 23. no. 2 (Fall 20GB) relation between marginal worker productivity and wages. If workers are able to collectively bargain over their wages, changes in wage structure may be a mis- leading indicator of productivity and knowledge spillovers. Moreover. to the extent that local firms are explicitly hiring and paying former employees of multinational enterprise to provide training to their own employees, any subsequent pro- ductivity effect cannot be considered a knowledge spillover according to the defi- nition adopted here (Castellani and Zanfei 2006). Finally, research on the existence of demonstration effects is less developed-that is, without considering the extensive black box literature on intraindustry know- ledge spillovers from FDI. The absence of theoretical contributions in this field and the multiplicity of definitions of demonstration effects make empirical assessment difficult. because it is not clear ex ante through which mechanisms such demon- stration effects should take place. More conceptualization on this topic seems necessary before substantial results can be expected from empirical research. Mediating Factors An explanation [Jor the diverse conclusions in PDI spillover studies] that seems plausible is that countries and firms within countries might differ in their ability to benefit from the presence of foreign-owned firms and their superior technology (Lipsey and Sjoholm, 2005, 23). One strand of literature has tried to identify the mediating factors required for the effective transmission of knowledge spillovers. Such factors can be seen as neces- sary conditions for knowledge spillover potential to turn into actual knowledge spillovers. The absence (or presence) of these factors may crucially influence observed knowledge spillovers; not taking them into account can bias empirical results. These factors usually pertain either to the receiving party (the host country, sector, region. or firm) or to the relations among the parties involved. Probably the best-known concepts in this field are absorptive capacity and spatial proximity. These are discussed below, followed by a brief review of two other mediating factors. intellectual property rights and host-country competition. Absorptive Capacity and Backwardness Two views exist in the literature on the role of the technology or productivity of a firm, region, industry, or country in capturing knowledge spillovers. Some researchers claim that technological backwardness should enhance knowledge spillovers. because the potential for improvement is large (Findlay 1978; Roger Smeets 117 Wang and Blomstrom 1992). Others argue that firms need some mmlmum amount of absorptive capacity to be able to capture knowledge spillovers (Cohen and Levinthal 1989. 1990; Glass and Saggi 1998). Such absorptive capacity, created by investments in R&D or human capital, provides the basis of funda- mental knowledge or technology necessary to assimilate and exploit external knowledge. Some early contributors to this field (implicitly) suggest a complementary relation between backwardness and absorptive capacity. Findlay (1978, 2) argues that "the greater the backlog of available opportunities ... the greater the pressure of change within the backward region .... Of course, the disparity must not be too wide for the thesis to hold." This remark hints at the import- ance of some minimum level of absorptive capacity. Abramovitz (1986. 388) argues that "a country's potential for rapid growth is strong not when it is back- ward without qualification. but rather when it is technologically backward but socially advanced." He conditions the benefits of backwardness on the presence of social capabilities. hinting at the importance of some form of absorptive capacity. In the empirical literature on knowledge spillovers from FDt the following general model is encountered: (5) where AC measures absorptive capacity and BW backvllardness (both variables are not always included simultaneously). Griffith, Redding, and Simpson (2002) consider the mediating effect of back- wardness on knowledge spillovers from FDI in a sample of 13,000 manufacturing establishments in the United Kingdom (table 2). They measure backwardness as frontier-level TFP relative to the TFP of the local establishment, where frontier- level TFP is defined either as the highest establishment-level TFP at the four-digit industry classification level at time t or as the average TFP of the top three estab- lishments with the highest TFP. Hence. an increase in BW implies that establish- ment i is becoming more backward. In modelS the effect of BW (/32) is positive and significant for both measures of backwardness. illustrating the importance of backwardness. Griffith. Redding, and van Reenen (2004) use a similar measure of backward- ness at the country-industry level. Their research yields positive and significant results for both backwardness and absorptive capacity. Castellani and Zanfei (2003) use a slightly different measure of backwardness: the ratio of the average TFP level of foreign firms in two-digit industry j over firm i's TFP level. Absorptive capacity is measured as the TFP level of firm i. They find that only /32 is positive and significant in a model similar to modelS. Peri and Urban (2006) obtain a similar result using a panel of German and Italian firms. 118 The World Bank Researcfl Observer, vol. 23, no. 2 (Fall 2008) Table 2. Empirical Results on Effects of Absorptive Capacity/Backwardness and Geographic Proximity on PDI Knowledge Spillover Factor Study Sample Effect Absorptive Griffith. Redding. 13.000 manufacturing firms in Backwardness: Positive capacity! and Simpson United Kingdom. 1980-92 backwardness (2002) Castellani and 3.932 firms in France. Italy. and Backwardness: Positive Zanfei (2003) Spain. 1992-97 Absorptive capacity: No effect Girma (2005) 7,516 firms in United Kingdom. Absorptive capacity: Inverted 1989-99 U-shaped effect Peri and Urban 40.000 firms in Italy Backwardness: Positive (2006) 800 firms in Germany. 1993-99 Girma and Giirg 2.100 electronic firms in United Absorptive capacity: (20(7) Kingdom. 1980-92 U-shaped effect 4.800 engineering firms in United Kingdom. 1980-92 Geographic Barrios. 338 plants in 26 counties in Ireland. Positive proximity Bertinelli. and 1983-98 Strobl (2006) Girma and 11.000 plants and 10 Nomenclature Positive Wake lin (2007) of Territorial Units for Statistics (NUTS) in 1 region in United Kingdom. 1980-92 Nicolini and 26 sectors and 30 NUTS in 2 regions Positive Resmini (2007) in Bulgaria. Poland. and Romania, 1998-2003 Source: Author's compilation. Using a sample of 7.516 British companies Girma (2005) investigates the role of absorptive capacity in capturing knowledge spillovers from PDI. His measure of absorptive capacity is a firm's TFP level at time t 1 relative to the highest level of TFP in the firm's industry at the four-digit classification level. (This measure is roughly the inverse of the backwardness measure used in the three studies cited above.) Applying threshold regression analysis, among other methods, he finds an inverted U-shaped effect of absorptive capacity on PDI: the knowledge spillover mediating effect is maximized at intermediate levels of absorp- tive capacity. Using the same measure of absorptive capacity in a panel of British firms in the electronics and engineering industries, Girma and Garg (2007) instead find evidence of a U-shaped effect of absorptive capacity. 6 Griinfeld (2006) corroborates this result theoretically. What, then, is the general conclusion regarding the mediating effect of back- wardness and absorptive capacity on knowledge spillovers from FDI? Comparing Roger Smeets 119 studies is difficult, because they use different empirical specifications and employ different measures of backwardness and absorptive capacity. Moreover, many of these studies disregard the relation behveen backwardness and absorptive capacity. An exception is Castellani and Zanfei (2003), who consider the correlation between backwardness and absorptive capacity at the industry leveL In their specification absorptive capacity (AC) is the denominator of backwardness (BW), Not surprisingly, they find an overall negative relation between backwardness and absorptive capacity, However, in their empirical estimation they ignore this relation, In terms of model 5 this implies that the marginal effect of ACuon Yijtis given by This total derivative of Yijt with respect to AC it shows that the marginal effect of a firm's absorptive capacity (AC it ) on its productivity (Y/jt) has both a direct component (/31) and an indirect component (through its effect on BWil ), Given that AC is the denominator of BW. this implies that an increase in absorptive capacity will reduce backwardness (dBWu/dAC it < 0). Thus, even if the direct effect of AC (/31) is positive, its indirect effect through BW (/32·[dBWjt/dAC it ]) is clearly negative. The empirical disregard for the relationship between backwardness and absorp- tive capacity applies to all studies that simultaneously include both measures. In generaL if backwardness is measured in terms of relative TFP levels and absorptive capacity is measured in terms of absolute TFP levels, R&D stocks. human capitaL and so forth the knowledge production function literature (Griliches 1979) suggests that a relation probably exists between backwardness and absorptive capacity. which should be taken into account empirically. A simple way out of this problem is to use the AC measure of Girma (2005) and Girma and Gorg (2007). who measure absorptive capacity as the inverse of backwardness: an increase in backwardness implies a simultaneous and pro- portional decrease in absorptive capacity and vice versa. Absorptive capacity as a relative concept also seems to make sense intuitively: as Castellani and Zanfei (2003) show. high absolute levels of TFP(AC) may still be accompanied by large technology gaps if foreign firms in the sector also exhibit extremely high (average) TFP levels. In such a situation absolute measures of absorptive capacity probably do not capture actual absorptive capacity. Finally. some studies estimate backwardness relative to frontier-level TFP, where the frontier is the highest (average) TFP level of the relevant sector in general. Because knowledge spillovers from FDI are investigated, however, it seems more appropriate to consider the TFP of the relevant multinational enterprises as the frontier. 7 120 The World Bank Researcfl Observer, vol. 23, no, 2 (Fall 2008) Spatial Proximity A well-established body of empirical literature suggests that spatial proximity (being geographically close to the knowledge source) is an important condition for capturing knowledge spillovers. Reasons for the purported relevance of geogra- phy can be traced to the individual knowledge spillover channels examined above. Researchers such as Girma and Wakelin (2007) argue that many of these chan- nels have a clear spatial component. The limited geographic mobility of labor, for example, implies that knowledge spillovers through worker mobility are highly localized. Theoretical work on the spatial dimension of knowledge spillovers from Fm is sparse. Martin and Ottaviano (1999) and Baldwin, Martin, and Ottaviano (2001) introduce spatially bounded knowledge spillovers in a new economic geography setting. Combining a two-region new economic geography model (Krugman 1991) with a Romerian-type endogenous growth model (Romer 1990), they investigate the influence of spatially bounded knowledge spillovers on growth rates in the two regions. Their results show that geography (firm location) matters for growth only when knowledge spillovers are spatially bounded. If spill- overs are global, both regions grow at similar rates in long-run equilibrium. Knowledge spillovers from multinational enterprises are absent in these frameworks. Jaffe, Trajtenberg, and Henderson (1993) and Jaffe and Trajtenberg (2002) make seminal empirical contributions on the spatial dimension of knowledge spill- overs (not necessarily from Fm). By looking at patent citations while controlling for the fact that innovation activity itself may be localized, they show that know- ledge spillovers are localized at various levels (country, state, and metropolitan statistical areas). Audretsch and Feldman (1996) show that geographic clustering of innovative activity is more pronounced in knowledge-intensive industries. Keller (2002) attaches a number to the spatial decay of knowledge spillovers from R&D in the Group of Five large industrial countries to nine European countries. He finds the "half-life" of knowledge spillovers (the distance within which half of total knowledge spillovers are eroded) to be about 1,200 kilometers. Bottazzi and Peri (2003) find an even stronger localization effect of knowledge spillovers in the EU-IS, where the effect of regional R&D (inputs) on the number of patents (outputs) vanishes beyond 300 kilometers. Although a wide body of literature exists on the spatial dimension of knowledge spillovers, specific applications to knowledge spillovers from Fm are still relatively limited. The empirical specification can be extended to incorporate a regional effect: Roger Smet'ts 121 where rand s index the regions. Hence f31 measures the effect of PDI within firm 1's region, and f32 measures the effect of PDI in other regions. Sometimes the effect on firms in other regions is weighted by a matrix, w, that incorporates the distance between regions rand s. Region-specific characteristics (such as region size relative to population or GDP) are captured by Zrt. If knowledge spillovers from PDI are spatially bounded, one would expect f3] to be positive and f32 to be insignificant. Barrios. Bertinelli, and Strobl (2006) construct an index that measures the extent to which local firms and multinational enterprises coagglomerate within counties. They find that productivity effects of PDI in Ireland are positive and sig- nificant only in counties that show a positive and significant degree of coaglom- meration (see table 2). Girma and Wakelin (2007) distinguish 10 regions that roughly correspond to the Nomenclature of Territorial Units for Statistics 1 (NUTS 1) classification in the European Union. 8 Their results indicate that the productivity of domestic plants is positively affected by PDI within but not outside the region (both weighted and unweighted by distance). Nicolini and Resmini (2007) document positive knowledge spillover effects on regional (domestic) TPP from multinational enterprises located in the same region and negative spillover effects from the presence of multinational enterprise in other regions. Intellectual Property Rights Two offsetting effects make the relation between the strength of intellectual pro- perty rights and the extent of knowledge spillovers from PDI ambiguous. Strong intellectual property rights induce multinational enterprises to transfer more and higher quality knowledge to their subsidiaries. thereby increasing knowledge spill- over potential. but they make it more difficult to capture knowledge spillovers (for example, through imitation). The net effect is not clear a priori. Markusen (2001) studies the effect of changes in intellectual property rights protection on welfare and spillovers in a host developing country. He finds that if the multinational enterprise cannot write an enforceable contract with a local agent. increased intellectual property right protection makes spillovers less likely. Glass and Saggi (2002) show that increased intellectual property right protection in developing countries has a similar effect on multinational enterprises and national firms in industrial countries, so that PDI does not become relatively more attractive. Most empirical research considers only the effect of intellectual property rights on the volume or composition of PDI or on the incentives for intrafirm technology transfer. Javorcik (2004a) investigates the effect of intellectual property rights on 122 The World Bank Research Observer, vol. 23, /10. 2 (f'all 2008) the composition of inward FDI in the Russian Federation and five countries in Central and Eastern Europe. Branstetter. Fisman, and Foley (2006) analyze the effect of intellectual property rights protection on technology transfer from 1.000 U.S. multinational enterprises to about 5,000 of their foreign affiliates in 16 deve- loping countries. The implication of their results for FDI knowledge spillovers are not clear. Feinberg and Majumdar (2001) analyze the knowledge spillover effects of FDI in a sample of 65 domestic firms and 30 multinational enterprises operating in the pharmaceuticals sector in India during the 1980s and early 1990s, when intellectual property rights protection in the sector was reportedly weak. They find virtually no evidence of knowledge spillovers. The finding could be considered circumstantial evidence that weak intellectual property rights protection does not stimulate knowledge spillovers from FDI. Indeed, Allred. and Park (2007) con- clude that there exists an optimal and positive degree of intellectual property rights protection that stimulates diffusion of knowledge from multinational enterprises. Competition in the Host Country or Sector Blomstrom, Globerman, and Kokko (2001) argue that greater competition may induce multinational enterprises to transfer more (high-quality) technology to their subsidiaries, increasing the potential for knowledge spillover. Theoretical models by Glass and Saggi (1998), Wang and Blomstrom (1992), and others show that this may be the case. Empirical studies do not appear to have explicitly studied the effect of host-sector competition on knowledge spillovers from FDI. 9 Taking Stock Research on the knowledge spillover-mediating roles of absorptive capacity and technology gaps remains inconclusive. Comparing studies is difficult because of differences in methodologies and measurement. Future empirical research might benefit from convergence in definitions of absorptive capacity and backwardness. It may also be useful to start thinking about absorptive capacity as a relative concept (Girma 2005; Girma and Gorg 2007). Investigating the nonlinear med- iating effects of these factors also seems to be a promiSing direction for future research (Girma 2005; Girma and Gorg 2007; Falvey. Foster, and Greenaway 2007). Specific applications regarding the spatial dimension of knowledge spillovers from PDI remain limited; more theoretical work on this topic is needed. Are there reasons to believe that the spatial dimension of knowledge spillovers from FDI will differ from that of knowledge spillovers in general? The answer hinges on the Roger Smects 123 specific spillover channels being considered. Knowledge spillovers transmitted through worker mobility are bound to be restricted geographically. The impli- cations are less obvious for knowledge spillovers through vertical linkages and demonstration effects, because both supplier and customer relations and imitation and reverse engineering may easily cross national or regional borders. Studies investigating the spatial dimension of knowledge spillovers from FDI might benefit from clearly spelling out the spillover channels of interest and carefully consider- ing their spatial dimension. The influence of intellectual property rights regimes on FDI knowledge spill- overs seems to be an important but neglected issue. More theoretical and empiri- cal research is needed that analyzes the impact of intellectual property rights regimes directly on knowledge spillovers rather than indirectly through intrafirm technology transfer. Since the effect of intellectual property rights on knowledge spillovers is not clear a priori (because of offsetting mechanisms on spillover potential and technology access), a great deal of insight can still be gained. FDI Heterogeneity To advance the literature on FDI spillovers, the questions "What kind of FDI?" and "What is the nature of [multinational corporation] activity in the local market?" need to be addressed. (Feinberg and Keane, 2005: p. 269) A third stream of research acknowledges the heterogeneity of multinational enter- prises' foreign activities and the effect on FDI knowledge spillovers (table 3). Some studies examine the relation between multinational enterprise ownership and knowledge spillovers. Others examine the relation between the nationality of the foreign investor or PDI motives and knowledge spillovers. Ownership of the Multinational Enterprise Muller and Schnitzer (2006) study the theoretical relation between knowledge spillovers and multinational enterprise ownership when the multinational enter- prise engages in an international joint venture with the host-country firm. They document a tradeoff in which a larger ownership share induces the multinational enterprise to transfer more technology to its subsidiary. increasing spillover potential but reducing the extent to which the host-country firm is exposed to the technology. The actual relation between multinational enterprise ownership and knowledge spillovers may turn out to be an empirical matter. 124 The World Bank Researcll Observer. vol. 23. no. 2 (Fall 2008) Table 3. Empirical Results on Effects of PDI l'l.llUWICL"'''' Spillover Factor Study Sample Effect Multinational Blomstrom and 13.663 manufacturing Minority and majority FOI shares enterprise Sjoholm (1999) firms in Indonesia. had equal spillover effects ownership 1991 Dimelis and Loud 4.056 manufacturing Minority FOI shares had greater (2002) firms in Greece. 1997 spillover effect than majority FOI shares Javorcik (2004b) 4.000 firms in Shared foreign and domestic Lithuania. 1996- ownership had positive spillover 2000 effect Javorcik and 13.129 firms in Shared foreign and domestic Spatareanu (2008) Romania. 1998- ownership had positive vertical 2003 spillover effect and negative horizontal spillover effect Abraham. Konings. 17.645 plants in Minority FOI shares had greater and Slootmaekers China. 2000-04 spillover effect than majority FOI (2007) shares Nationality of Buckley. Clegg. and 1 30 industries in · No effect for FOI from Hong parent company Wang (2007b) China. 1995 Kong. China; Macau, China; and Taiwan. China · Positive effect for FOI from other countries in high-technology sectors Buckley. Clegg. and 158 industries in · Positive effect for FOI from Hong Wang (2007a) China. 2001 Kong, China; Macau. China; and Taiwan. China. in labor-intensive industries · Positive effect for FOI from other countries in technology-intensive industries Abraham. Konings, 17,645 plants in · For locally owned enterprises. and Slootmaekers China. 2000-04 greater effect for FOI from Hong (2007) Kong. China; Macau. China; and Taiwan. China. than for FOI from other countries · For foreign-owned enterprises. the effect was opposite Javorcik, Saggi. and 50,957 firms in · I'OI from Asia and America had Spatareanu (2004) Romania. 1998- positive upstream spillover effect 2000 · FOI from the European Union had negative effect Continued Roger Smeets 125 Table 3. Continued Factor swdu Sample EJleet Girma and Wakelin 11.000 plants and 10 · POI from Japan and other (2007) Nomenclature of countries had positive effect Territorial Units for · FOI from United States had no Statistics (NUTS) in 1 effect region in United Kingdom. 1980-92 Motive for POI Girma (2005) 7,516 firms in United · Exploiting POI had positive effect Kingdom, 1989-99 · Sourcing POI had no effect Driffield and Love 11 manufacturing · Exploiting FOI had positive effect (2007) sectors in United · Sourcing POI had no effect Kingdom. 1987-97 Protsenko (2003) 256 German FOI · Vertical FOI had positive effect projects in Czech · Horizontal FOI had negative effect Republic. 1993-99 Source: Author's compilation. Empirical research usually distinguishes between minority PDI (the foreign investor holds a minority share in the foreign affiliate) and majority PDI (the foreign investor holds a majority share in the foreign affiliate): (7) where Min_FDT and MaLFDT measure the amount of minority and majority FDT in sector j. Some empirical studies distinguish between wholly owned subsidiaries and shared subsidiaries. Sometimes the intersectoral spillover effects of different types of PDI are investigated as well. Blomstrom and Sjoholm (1999) were among the first researchers to consider this relation empirically. Their study of 13,663 Indonesian manufacturing firms reveals that both minority and majority PDI lead to spillovers, with no statistical differences between the estimated effects. Dimelis and Louri (2002) consider a sample of 4,056 Greek manufacturing firms. In separate regressions they analyze the relation between multinational enterprise ownership and knowledge transfer (to the local affiliate) and the relation between multinational enterprise ovvnership and knowledge spillovers (to other local firms). The results broadly confirm the intuition in Muller and Schnitzer (2006): only majority-owned foreign affiliates experience increases in productivity as a result of knowledge transfer, and minority POI is more likely than majority PDI to produce knowledge spillovers. Javorcik (2004b) analyzes a panel of about 4.0()() firms in Lithuania, distinguishing between horizontal (intraindustry) and vertical (interindustry) 126 The World Bank Research Observer. vol. 23. no. 2 (Fall 2008) spillovers. She finds that firms that are owned by both the foreign investor and a local firm create backward knowledge spillovers (to supplying industries). while wholly owned subsidiaries do not. She finds no evidence of horizontal or forward knowledge spillovers or statistical differences between the effects of minority and majority FDI. Javorcik and Spatareanu (2008) analyze a panel of 13.129 Romanian firms. They find that shared foreign and domestic ownership induces positive vertical spillovers and negative horizontal spillovers. Wholly owned subsidiaries do not induce vertical spillovers and induce larger negative horizontal spillovers. These negative effects are explained by adverse competition effects. Abraham. Konings. and Slootmaekers (2007) analyze the relation between minority- and majority-owned FDI and knowledge spillovers in an unbalanced panel of 17.645 plants in China. Their results show that minority FDI has a negative (competition) effect on locally owned enterprises' productivity and that majority FDI has no effect. The effect of minority FDI on foreign-owned enter- prises is positive and larger than that of majority FDt Nationality of the Parent Company Some recent studies argue that the nationality of the foreign investor affects the knowledge spillover effects of FDI (see table 3). Most studies in this field consider PDI in China, comparing the effects of FDI from Hong Kong. China; Macau. China; and Taiwan. China, (HMT_FDI) on the one hand and from Western countries (OTHER_FDI) on the other hand. The specification is similar to the one in model 7. with Min_FDI and MaLFDI replaced by HMT_FDI and OthecFDI. Buckley, Clegg, and Wang (2007b) argue that FDI from Hong Kong. China; Macau, China; and Taiwan, China, is less technologically advanced than that of investors from outside China. As a result. although initial increases in such FDI may induce positive spillover effects, beyond some threshold level the negative competition effect starts to dominate. They therefore predict a nonlinear spillover effect of increased FDI from these sources. This contrasts with the positive linear effect of FDI from Western countries (the knowledge spillover effect is expected to dominate, because it carries more advanced technology). Their empirical analysis of 130 Chinese industries confirms their expectations: FDI from outside China has the expected (linear) positive effect (albeit only in high-technology sectors). Buckley, Clegg. and Wang (2007a) investigate the relation between FDI from Hong Kong. China; Macau. China; and Taiwan, China. and from outside China in a sample of 158 Chinese industries. taking into account receiving firms' and industries' characteristics. They find that such FDI generates more knowledge spil- lovers in labor-intensive industries and that FDI from outside China generates more knowledge spillovers in technology-intensive industries. Roger Smel'ts 127 Abraham. Konings. and Slootmaekers (2007) show that the spillover effects of both minority and majority Fm from Hong Kong. China; Macau. China; and Taiwan. China. on locally owned enterprises are larger than those from Fm from other countries. The opposite holds for knowledge spillovers to foreign-owned enterprises. Javorcik, Saggi, and Spatareanu (2004) compare the upstream knowledge spil- lover effects of Fm from Asian, European. and American (North and South) enter- prises in a panel of 50.957 Romanian firms. They posit three reasons to expect weaker knowledge spillover effects from Fm from the European Union: the European Union is located closer to Romania, Romania was engaged in a prefer- ential trade agreement with the European Union during the period of investi- gation, and inputs sourced from home-country suppliers by EU subsidiaries comply with Romania's rules of origin, which is not the case for Asian or American subsidiaries. All these mechanisms make knowledge spillovers through vertical linkages less likely for EU subsidiaries, because they stimulate imports of intermediate inputs from the European Union. The results confirm their expec- tations: Fm from Asia and America has positive vertical (upstream) knowledge spillover effects on Romanian firms. The effect is negative for Fm from the European Union, which the authors explain by pointing to increased competition in the downstream sector in which multinational enterprises are operating. Girma and Wakelin (2007) distinguish between inward Fm into the United Kingdom from Japan, which accounts for the majority of R&D-intensive inter- national companies in the electronics industry; from the United States. which has long invested in the British manufacturing industry; and from other countries. Their results indicate that Japanese and other international firms produce signifi- cant and positive knowledge spillover effects. whereas u.s. firms do not have a dis- cernible spillover effect. The authors hint at the relative high R&D-intensity of Japanese Fm as an explanation for this result. Motives for FDJ Most of the studies discussed above assume that Fm has knowledge spillover potential. that the firms engaging in Fm do so to exploit a technological or other ownership advantage abroad, part of which may spill over to the host country. This type of PDI is known as technology-exploiting PDI (Kuemmerle 1999; Le Bas and Sierra 2002). Most of the traditional literature on pm refers to this type of investment (Hymer 1960; Dunning 1977; Markusen 2002). Scholars have recently pointed out a different type of PDI-technology-seeking PDI-which is motivated by a desire to source or seek external foreign knowledge (Dunning and Narula 1995; Kuemmerle 1999; Posfuri and Motta 1999; Siotis 1999; Le Bas and Sierra 2(02). Firms engaging in technology-seeking PDI try to 128 The World Bank Research Observer. vol. 23. no. 2 (Fa112008) capture knowledge spillovers from firms in the host countries in which they invest. Knowledge spillovers are expected to flow from local firms to the multi- national enterprise instead of the other way around. A few studies investigate knowledge spillovers by distinguishing between these types of FDL The empirical model is similar to that in model 7, with technology- exploiting and technology-seeking FDI substituted for Min_FDI and MaLFDI. In a panel study of 11 manufacturing sectors in the United Kingdom, Driffield and Love (2007) find that technology-sourcing FDI does not generate knowledge spill- overs. whereas technology-exploiting FDI does. Girma (2005) obtains similar results. 10 FDI can also be classified as horizontal (Markusen 1984). vertical (Helpman 1985). or export platform (Ekholm, Forslid, and Markusen 20(7). Horizontal FDI is usually motivated by market-seeking incentives, vertical FDI by efficiency- or resource-seeking incentives, and export-platform FDI by the desire to find an effi- cient location from which to more profitably export to third countries. The extent of knowledge spillovers from these types of FDI may differ (Javorcik and Spatareanu 2005; Driffield and Love 2007; Beugelsdijk, Smeets, and Zwinkels forthcoming). Protsenko (2003) examines the spillover effects of horizontal and vertical German PDI in the Czech Republic. He finds that vertical FDI generates positive knowledge spillovers. whereas horizontal FDI has effects largely through increased competition. These results suggest that the distinction between horizontal. verti- caL and export-platform FDI is potentially important in determining the extent of knowledge spillovers. Taking Stock The work on the relation between multinational enterprise ownership and know- ledge spillovers has strong intuitive appeal. because it seems likely that not all types of subsidiaries (minority. majority) generate the same knowledge spillovers. Theoretical work in this area is scant, however; more insights are needed to guide empirical work. The empirical results obtained so far are difficult to compare. because they take slightly different approaches. A fruitful extension in this area would be to consider the influence of multinational enterprise ownership along a continuum. Instead of analyzing the spillover effect of different categories of subsidiaries (minority; majority), researchers might analyze the influence of actual ownership shares (0-100 percent) on local firms' productivity. Such an approach would allow researchers to analyze nonlinear effects. Studies distinguishing between the country origin of FDI often do so based on a variety of economic rationales (such as differences in expected R&D intensities, or Roger Smeets 129 differences in local input sourcing). Future research should investigate whether these more general underlying economic rationales can be used to distinguish different types of FDt instead of the more specific country of origin. Such an approach may stimulate the development of both more theoretical research in this area as well as a more general empirical application. Distinguishing FDI motives may contribute to a better understanding of the likelihood of knowledge spillovers from FDI. Theoretical models in this field have looked only at the relation between FDI motives and firm heterogeneity. A useful extension would be a model in which the extent of knowledge spillovers is endogenously determined by firms' motives in pursuing FDI. Also more empirical research is needed that directly investigates this relation. Although the few studies reviewed above indicate that technology-seeking FDI does not generate knowledge spillovers, more recent empirical research indicates that this type of FDI may at least have a large potential of doing so (Feinberg and Gupta 2004; Cantwell and Mudambi 20(5). More research investigating the differential knowledge spillover effects of horizontal. vertical. and export-platform FDI is also warranted. Conclusion If country and industry differences are important to the impact of inward FDI on host countries, the main lesson might be that the search for uni- versal relationships is futile (Lipsey and Sjoholm, 2005, 40). With so many dimensions and so many factors at the country, sector, regional. and firm level influencing the relation between FDI and knowledge spillovers, the search for universal relations may well be futile. This does not imply that the search for knowledge spillovers from FDI is futile, however. The studies surveyed in this article that explicitly investigate the individual knowledge spillover channels identified in figure 3 (and summarized in table 1) all seem to conclude that knowledge spillovers from FDI do occur through these channels (except through forward linkages). Explicitly taking into account these knowledge spillover channels seems to be an important step forward in this literature. The literature on mediating factors and FDI heterogeneity is inconclusive, at least partly because of the lack of comparability across studies caused by differ- ences in methodologies and measurement. Several changes could improve results. First, researchers could move toward convergence. for example, by uniformly measuring absorptive capacity as a relative concept or measuring multinational enterprise ownership along a continuum rather than as a categorical variable. Second, any study of knowledge spillovers should specify the channels ana- lyzed. Such an approach would clearly delineate the possible role of mediating 130 The World Bank Research Observer, vol. 23. no. 2 (Fall 2008) factors or PDI heterogeneity. Por example. the relevance of the spatial dimension as a mediating factor for knowledge spillovers strongly depends on the spillover channels considered; also, different types of PDI may spill over knowledge through different channels to different extents. Third, deeper insight into the (conditional) existence of knowledge spillovers from PDI is not likely to come from any of the outlined approaches indiVidually. Spillover channels. mediating factors. and PDI heterogeneity coexist and interact in determining the extent of knowledge spillovers. Theoretical and empirical research should therefore try to address them simultaneously (Wei and Liu 2006; Liang. 2008). Does the importance of absorptive capacity for capturing knowledge spillovers through demonstration effects vary with the degree of multinational enterprise ownership? Is the spatial dissipation of knowledge spillovers through backward linkages different for horizontal and vertical PDI? These kinds of inter- related questions should guide future work on this topic. Two important overarching issues need to be noted. Pirst, empirical work too often ignores the conceptually important distinction between intentional know- ledge transfers and unintentional knowledge spillovers. As Blalock and Gertler (2007, 2(08) and Javorcik and Spatareanu (2005) clearly show. many of the esti- mated effects are more likely related to knowledge transfer than knowledge spil- lover. Prom a policy perspective this distinction is very important: whereas the existence of knowledge spillovers (which are externalities) clearly warrants inter- ventionist government policy, the existence of knowledge transfer (which takes place through market mechanisms) clearly does not. u Mistakenly assigning the beneficial productivity effects of PDI to knowledge spillovers may convince govern- ments of many developing countries to undertake costly and wasteful PDI policies. Future empirical work on this topiC should be very careful in labeling estimated positive effects of PDI as spillovers and even more careful in deriving far-reaching (costly) policy implications from them. Second, a wide gap remains between theoretical and empirical research (one exception is Alfaro and Rodrfguez-Clare 20(4). Theory and empirics have devel- oped more or less independently. In many of the areas reviewed above, more theoretical work is needed. The definition and functioning of demonstration effects. the spatial dimension of knowledge spillovers from FDI in interaction with different spillover channels, and the relation between various motives for FDI and knowledge spillovers are just a few of the areas in which theory to guide future empirical work has been lacking. Given advances in the literature on firm hetero- geneity (Melitz 2003; Helpman, Melitz, and Yeaple 2004), the relation between PDI heterogeneity and knowledge spillovers seems a particularly promising field for future theoretical research. The result of these recommendations may be to further highlight the weak gen- eralizability of research results. Sacrificing some generalizability in order to obtain Roger Srneets 131 more detailed and conclusive results is preferable. however. to losing sight of important nuances in order to obtain more general results-an approach that has not, to date. yielded consistent results. Notes Roger Smeets is a PhD student in the Department of Economics at Radboud Ilniversity Nijmegen. the Netherlands; his email addressisr.smeets@fm.ru.nI.TheauthorthanksAlbertdeVaaI.Eelke de long, and three anonymous referees for useful comments and suggestions on an earlier draft of this article. 1. This survey does not review methodological developments in the empirical FDI spillover litera- ture. Chapter 5 in Castellani and Zanfei (2006) includes a useful overview of recent methodological advances, 2. Other benefits of inward FDI, such as employment generation and knowledge transfers through licensing, take place through market mechanisms and thus are not arguments for active government involvement. 3. In a similar vein. Rogoff and his coauthors deal with the (macroeconomic) growth effects of capital account liberalization through foreign portrolio investment by investigating the necessary preconditions or mediating factors under which these effects arise (Prasad and others 2003: Kose and others 2(06). 4. Keller (2004) provides an excellent survey of the wider literature on international technology diffusion, including knowledge spillovers from FDI. 5. The existence of adverse competition effects assumes that the multinational enterprise goes abroad mainly to produce for the local market (that is, it assumes that FDI is mainly of the horizon- tal type). For both vertical and export-platform FDI the adverse competition effect is less likely to occur and will be less severe if it does (Protsenko 2003). 6. Falvey, Foster, and Greenaway (2007) consider the simultaneous effects of backwardness and absorptive capacity at the country level. However. they essentially estimate an empirical growth model and focus on trade-related knowledge spillovers. Their results hint at the importance of backwardness over absorptive capacity, although the results vary with the specification and estimation method. 7. I thank an anonymous referee for pointing this out. 8. NUTS provides a single nniform breakdown of territorial units for the production of regional statistics for the European Ilnion. NUTS 1 denotes the broadest level, NUTS 3 denotes the most disag- gregated one. 9. Kathuria (2002) examines the effect of liberalization of Indian industries between 1989 and 1997 on knowledge spillovers from FDI. Although liberalization increased competition in general, the reforms applied mainly to trade liberalization. The effect on knowledge spillovers occurred mainly through higher FDI. 10. An extensive body of literature investigates this issue indirectly. by considering the relation between technology-exploiting and technology-sourcing FDI and firm heterogeneity, The results are ambiguous. Some studies find that only low-productivity firms engage in technology-sourcing FDI. which would imply that the knowledge spillover potential from this type of FDI is low (Kogut and Chang 1991; Hennart and Park 1993; Almeida 1996; Neven and Siotis 1996). Other studies show that high-productivity firms are more likely to undertake technology-sourcing FIJI, which would imply that the potential for spillover is high (Cantwell and Janne 1999; Chung and Alc;,icer 2002; Berry 2006: Branstetter 2006: Grillith, Harrison, and van Reenen 2006), 1 1. If backward knowledge transfers increase competition in supplying industries, reducing prices on intermediate goods and end products, the wealth of consumers in the host country rises. so that the social returns of knowledge transfer exceed the private returns. In this case interventionist govern- ment policy could be warranted, as Blalock and Gertler (2008) note. 132 l'he World Bank Research Observer. vol. 23, 110. 2 (Fall 2008) References Abraham, Filip, Joep Konings, and Veerle Slootmaekers. 2007. "FDI Spillovers, Firm Heterogeneity and Degree of Ownership: Evidence from Chinese Manufacturing," CEPR Discussion Paper 6573. Centre for Economic Policy Research, London. Abramovitz, Moses. 1986. "Catching Up, Forging Ahead, and Falling Behind," Journal of Economic History 46(2):385-406. Aitken, Brian J., and Ann E. Harrison. 1999. "Do Domestic Firms Benefit from Direct Foreign Investment? Evidence from Venezuela." American Economic Review 89(3):605-18. Alfaro. Laura. and Andres Rodriguez-Clare. 2004. "Multinationals and Linkages: An Empirical Investigation." Economica 4(2):113-69. Allred. Brent B.. and Walter G. Park. 2007. "Patent Rights and Innovative Activity: Evidence from National and Firm-Level Data." Journal of International Business Studies 38(6):878-900. Almeida. Paul. 1996. "Knowledge Sourcing by Foreign Multinationals: Patent Citation Analysis in the U.S. Semiconductor Industry." Strategic Management Journal 17 (Winter): 1 55-65. Audretsch. David B.. and Maryann P. Feldman. 1996. "R&D Spillovers and the Geography of Innovation and Production." American Economic Review 86(3):630--40. Baldwin. Richard E.. Philippe Martin. and Gianmarco Ottaviano. 2001. "Global Income Divergence. Trade and Industrialization: The Geography of Growth Take-Off." Journal of Economic Growth 6(1):5-37. Barrios, Salvador. Luisito BertineliL and Eric Strobl. 2006. "Coagglomeration and Spillovers." Regional Science and Ur/Jan Economics 36(4):467-81. Berry, Heather. 2006. "Leaders. Laggards. and the Pursuit of Foreign Knowledge." Strategic MarUl/lement jourIlal 27(2):151-68. Beugelsdijk. Sjoerd. Roger Smeets, and Remco Zwinkels. (forthcoming). "The Impact of Horizontal and Vertical FDI on Host's Country Economic Growth." International Business Review. Blalock. Garrick. and Paul J. Gertler. 2005. "Foreign Direct Investment and Externalities: the Case for Public Intervention." In Theodore H. Moran, Edward M. Graham, and Magnus Blomstromeds .. Does Foreign Direct Investment Promote Developmellt? 73-106. Washington. D.C.: Institute for International Economics and Center for Global Development. ___ 2008. "Welfare Gains from Foreign Direct Investment through Technology Transfer to Local Suppliers." Journal of International Economics 74(2 ):402 - 21. Blomstrom. Magnus, Steven Globerman, and Ari Kokko. 2001. "The Determinants of Host Country Spillovers from Foreign Direct Investment." In Nigel Pain, ed .. Tnward Tnveslment. Technological Change. and Growth. I.ondon: Palgrave. Blomstrom. Magnus. and Ari Kokko. 1998. "Multinational Corporations and Spillovers." Journal of Economic Surveys 12 (2): 1- 31. Blomstrom. Magnus. and Fredrik Sjoholm. 1999. "Technology Transfer and Spillovers: Does Local Participation with Multinationals Matter?" European Economic Review 43(4-6):915-23. Bottazzi, Laura. and Giovanni Peri. 2003. "Innovation and Spillovers in Regions: Evidence from European Patent Data." European Economic Review 47(4):687 710. Branstetter, Lee. 2006. "Is Foreign Direct Investment a Channel of Knowledge Spillovers? Evidence from Japan's pm in the United States." Joumal of Intemational Economics 68(2):325-44. Branstetter. Lee. Raymond Pisman, and C. Pritz Foley. 2006. "Do Stronger Intellectual Property Rights Increase International Technology Transfer? Empirical Evidence from U.S. Firm-Level Panel Data." Quarterly JOllmaI ofEcol1omics 12(1):321-49. Roger 8mrets 133 Buckley. Peter J.. Jeremy Clegg, and Chengqi Wang. 2007a. "The Impact of Foreign Ownership, Local Ownership and Industry Characteristics on Spillover Benefits from Foreign Direct Investment in China." International Business Review 16(2):142-58. ___ . 2007b. "Is the Relationship between Inward Fm and Spillover Effects Linear? An Empirical Examination of the Case of China." Journal oj International Business Studies 38(3 ):44 7 - 59. Bwalya. Samuel M.. 2006. "Foreign Direct Investment and Technology Spillovers: Evidence from Panel Data Analysis of Manufacturing Firms in Zambia." Journal oj Development Economics 81(2): 514-26. Cantwell, John, and Odile Janne. 1999. "Technological Globalisation and Innovative Centres: The Role of Corporate Technological Leadership and Locational Hierarchy." Research Policy 28(2): 119-44. CantwelL John, and Ram Mudambi. 2005. "MNE Competence Creating Subsidiary Mandates." StrategiC Management Journal 26(12):1109- 28. Castellani. Davide, and Antonello Zanfei. 2003. "Technology Gaps. Absorptive Capacity and the Impact of Inward Investments on Productivity of European Firms." Economics oj Innovation and New Technology 12(6):555-76. ___ . 2006. Multinational Firms, innovation and Productivity, Cheltenham, United Kingdom: Edward Elgar. Cheung, Kui-Yin, and Ping Lin. 2004. "Spillover Effects of FDT on Innovation in China: Evidence from the Provincial Data." China Economic Review 1 5(l ):25-44. Chung, Wilbur. and Juan Alcacer. 2002. "Knowledge Seeking and Location Choice of Foreign Direct Investment in the United States." Management Science 48(12):1534-54. Cohen, Wesley, and Daniel A. Levinthal. 1989. "Innovation and Learning: the 1\vo Faces of R&D." Bconomic Journal 99(397):569-96. 1990. 'Absorptive Capacity: A New Perspective on Learning and Innovation." Administrative Science Quarterly 35(1):128-52. Dimelis, Sophia, and Helen Louri. 2002. "Foreign Ownership and Production Efficiency: A Quantile Regression Analysis." OxJord Economic Papers 54(3):449-69. Driffield. Nigel. and James H. Love. 2007. "Linking FDT Motivation and Host Economy Productivity Eflects: Conceptual and Empirical Analysis." Journal oj International Business Studies 38(3): 460-73. Dunning, John H. 1977. "Trade, Location of Economy Activity and MKE: A Search for an Eclectic Approach." In8. Ohlin, P O. Hesselborn, and P. S. Wijkmaneds., The International Allocation oj Economic Activity. 395-418. London: Macmillan. Dunning, John H.. and Rajneesh Narula. 1995. "The R&D Activities of Foreign Firms in the United States." International Studies of Management and Organization 2 5( 1- 2): 3 9- 73. Ekholm. Karolina. Rikard Forslid, and James R. Markusen. 2007. "Export Platform Foreign Direct Investment." Journal oj the European Economic Association 5(4):776-95. Falvey. Rod, Kei! N. Foster, and David Greenaway. 2007. "Relative Backwardness, Absorptive Capacity and Knowledge Spillovers." Economic Letters 97(3):230-34. Feinberg, Susan E.. and Ani! K. Gupta. 2004. "Knowledge Spillovers and the Assignment of R&D Responsibilities for Foreign Subsidiaries." Strategic Management Journal 25(8-9):823-45. Feinberg, Susan Eo, and Michael P. Keane. 2005. "Intra-Firm Trade of U.S. MNCs: Findings and Implications for Models and Policies toward Trade and Investment." InTheodore H. Moran, Edward M. Graham, and Blomstrom Magnuseds.. Does Foreign Direct Investment Promote Development? 245-71. Washington. D.C.: Institute for International Economics and Center for Global Development. 134 Tile World Bank Researcll Observer. 1101. 23. no. 2 (Fall 2008) Feinberg. Susan E., and Sumit K. Majumdar. 2001. "Technology Spillovers from Foreign Direct Investment in the Indian Pharmaceutical Industry." Journal of International Business Studies 32(3):421-37. Findlay; Ronald. 1978. "Relative Backwardness, Direct Foreign Investment. and the Transfer of Technology: A Simple Dynamic Model." Quarterly Journal of Economics 92(1):1~16. Fosfuri, Andrea, and Massimo Motta. 1999. "Multinationals without Advantages." Scandinavian Journal of Economics 101(4):617-30. Fosfuri. Andrea, Massimo Motta. and Thomas Ronde. 2001. "Foreign Direct Investment and Spillovers through Workers' Mobility." Journal of International Economics 53(1):205-22. Girma, Sourafel. 2005. "Absorptive Capacity and Productivity Spillovers from FOI: A Threshold Regression Analysis." Oxford Bulletin of Economics and Statistics 6 7(3 ):281-306. Girma, Sourafel. and Holger Garg. 2007. "The Role of Efficiency Gap for Spillovers from FOI: Evidence from the U.K. Electronics and Engineering Sectors." Open Economies Revin\} 18(2): 215-32. Girma, Sourafel. and Katharine Wakelin. 2007. "Local Productivity Spillovers from Foreign Direct Im'estment in the U.K. Electronics Industry." Regional Science and Urban Economics 37(3): 399-4]2. Glass. Amy J., and Kamal Saggi. 1998. "International Technology Transfer and the Technology Gap." Journal of Development Economics 55(2):369-98. _ _ . 2002. "Intellectual Property Rights and Foreign Direct Investment." Journal of International Economics 56(2):387-410. Carg, Holger. and David Greenaway. 2004. "Much Ado about Nothing? Do Domestic Firms Really Benefit from Foreign Direct Investment?" World Bank Research Observer 19(2):] 71-97. Garg. Holger. and Eric Strobl. 2001. "Multinational Companies and Productivity Spillovers: A Meta- Analysis." Economic JournaI1l1(475):F723-39. 2005. "Spillovers from Foreign Firms through Worker Mobility: An Empirical Investigation." Scamlinavian Journal of Economics 107(4):693 -709. Griffith, Rachel. Rupet Harrison, and John Van Reenen. 2006. "How Special Is the Special Relationship? Using the Impact of u.s. R&D Spillovers on U.K. Firms as a Test of Technology Sourcing." American Economic Reviw 96(5):1859-75. Griffith. Rachel. Stephen Redding, and Helen Simpson. 2002. "Productivity Convergence and Foreign Ownership at the Establishment Level." Institute for Fiscal Studies Working Paper 22. London. Griffith. Rachel. Stephen Redding. and John van Reenen. 2004. "Mapping the Two Faces of R&D: Productivity Growth in a Panel of OECD Industries:' Review of Economics and Statistics 86(4): 883-95. Griliches, Zvi. 1979. "Issues in Assessing the Contribution of R&D to Productivity Growth." Bell Journal of Economics 10(1):92-116. Griinfeld. Leo. 2006. "Multinational Production. Absorptive Capacity. and Endogenous R&D Spillovers." Review of International Economics 14( 5):9 2 2 -40. Hale, Galina, and Cheryl Long. 2006. "What Determines Technological Spillovers of Foreign Direct Investment? Evidence from China." Yale Economic Growth Center Discussion Paper 934. New Haven. CT. Helpman. Elhanan. 1985. "A Simple Theory of International Trade with Multinational Corporations." Journal of Political Economy 92(3):451-71. Roger Srneets 135 Helpman, Elhanan, Marc J. Melitz, and Stephen R. Yeaple, 2004, "Export versus FDI with Heterogeneous Firms," American Economic Review 94(1 );300-16. Hennart, Jean-Fran<;ois. and Young R. Park. 1993. "Greenfield vs. Acquisition: The Strategy of Japanese Investors in the United States." Management Science 39(9):1054-70. Heston, Alan. Robert Summers, and Bettina Aten. 2006. Penn World Table Version 6.2. Center for International Comparisons of Production, Income and Prices. University of Pennsylvania, Philadelphia. Hoekman. Bernard. and Beata S. Javorcik. 2006. "Lessons from Empirical Research on International Technology Diffusion through Trade and Foreign Direct Investment." InBernard Hoekman, and Beata S. Javorcikeds.. Global Integration and Technology Transfer. 1-26. Washington, D.C.: Palgrave Macmillan/World Bank. Hymer. Stephen H. 1960. The International Operations of National Firms: A Study of Direct Foreign Investment, Ph.D. diss., Massachusetts Institute of Technology, Cambridge. Mass. Jaffe. Adam B.. and Manuel Trajtenberg. 2002. Patents. Citations and Innovations: A Window on the Knowledge Economy. Cambridge Mass: MIT Press. Jaffe, Adam B., Manuel Trajtenberg. and Rebecca Henderson. 1993. "Geographic Localization of Knowledge Spillovers as Evidenced by Patent Citations." Quarterly Journal oj Economics 108(3): 577-98. Javorcik, Beata S. 2004a. "The Composition of Foreign Direct Investment and Protection of Intellectual Property Rights: Evidence from Transition Countries." European Economic Review 48(1):39-62. ___ . 2004b. "Does Foreign Direct Investment Increase the Productivity of Domestic Firms? In Search of Spillovers through Backward Linkages." American Economic Review 94(3):605-27. Javorcik, Beata S.· and Mariana Spatareanu. 2005. "Disentangling FOI Spillover Effects: What Do Firm Perceptions Tell Us?" InTheodore H. :\foran, &hvard M. Graham, and Magnus Blomstromeds., Does Foreign Direct Investment Promote Development? 45-71. Washington, D.C.: Institute for International Economics and Center for Global Development. 2008. "To Share or Not to Share: Does Local Participation Matter for Spillovers from Foreign Direct Investment?" Journal of Development Economics 8 194-217. Javorcik, Beata S.. Kamal Saggi, and Mariana Spatareanu. 2004. "Does It Matter Where You Come From? Vertical Spillovers from Foreign Direct Investment and the Nationality of Investors." Policy Research Working Paper 3449. World Bank, Washington, ne. Kathuria, Vinish. 2002. "Liberalisation, FOI and Productivity Spillovers. An AnalysiS of Indian Manufacturing Firms." Oxford Economic Papers 54(4):688-718. Keller, Wolfgang. 2002. "Geographic Localization of International Technology Diffusion." American Economic Review 92(1):120-42. 2004. "International Technology Diffusion." Journal oj Economic Literature 42(3):752-82. Kogut. Bruce. and Sea J. Chang. 1991. "Technological Capabilities and Japanese Foreign Direct Investment in the United States." Review of Economics and Statistics 73(3 ):401-13. Kose, M, Ayhan, Eswar Prasad. Wei Shang-Jin. and Kenneth Rogoff. 2006. "Financial Globalization: A Reappraisal." llvlF Working Paper 189. International Monetary Fund, Washington D.C. Krugman, Paul. 1991. "Increasing Returns and Economic Geography." Journal of Political Economy 99(3):483-99. Kuemmerle. Walter. 1999. "Foreign Direct Investment in Industrial Research in the Pharmaceutical and Electronics Industries: Results from a Survey of Multinational Firms." Research Policy 28(2- 79-93. 136 The World Bank Research Observer. vol. 23. no. 2 (J:'all 20(8) Kugler, Maurice. 2006. "Spillovers from Foreign Direct Investment: Within or between Industries?" Journal of Development Economics 80(2):444-77. Le Bas, Christian, and Christophe Sierra. 2002. "Location versus Home Country Advantages in R&D Activities: Some Further Results on Multinationals' Location Strategies." Research Policy 31(4): 589-609. Liang. Feng. 2008. "Does Foreign Direct Investment Improve the Productivity of Domestic Firms? Technology Spillovers. Industry Linkages. and Firm Capabilities." Haas School of Business. University of California. Berkeley. Lin. Ping, and Kamal Saggi. 2005. "Multinational Firms and Backward Linkages: A Critical Survey and A Simple Model." InTheodore H. Moran. Edward Graham. and Magnus Blomstrom eds .. Does ~Foreign Direct Investment Promote Development? 159-74. Washington. D.C.: Institute for International Economics and Center for Global Development. 2007. "Multinational Firms. Exclusivity and Backward Linkages." Journal of International Economics 71(1):206-20. Lipsey. Robert E.. and Fredrik Sjoholm. 2005. "The Impact of Inward FDI on Host Countries: Why Such Different Answers?" In Theodore H. Moran. Edward Graham. and Magnus Blomstrom. eds .. Does Foreign Direct Investment Promote Development? 23-43. Washington. D.C.: Institute for International Economics and Center for Global Development. Markusen. James R. 1984. "Multinationals. Multi-Plant Economies, and the Gains from Trade." Journal of lnternational Economics 16(3-4):205- 26. ___ . 2001. "Contracts. Intellectual Property Rights, and Multinational Investment in Developing Countries." Journal of International Economics 53(1):189-204. 2002. Multinational Firms and the Theory of International Trade. Cambridge, MA: MIT Press. Markusen. James R., and Natalia Trofimenko. 2007. Teaching Locals New Tricks: Foreign Experts as a Channel of Knowledge Transfer. NBER Working Paper 12872. Cambridge. Mass.: National Bureau of Economic Research. Markusen. James R., and Anthony J. Venables. 1999. "Foreign Direct Investment as a Catalyst for Industrial Development." Journal of International Economics 4 3( 2): 33 5 - 56. Martin, Philippe. and Gianmarco Ottaviano. 1999. "Growing Locations: Industry Location in a Model of Endogenous Growth." European Economic Review 43(2):281-302. Melitz, Marc J. 2003. "The Impact of Trade on Aggregate Industry Productivity and Intra-Industry Reallocations." Econometrica 71(6):1695-25. Moran, Theodore H.. Edward M. Graham. and Magnus Blomstrom eds. 2005. Does Foreign Direct Investment Promote Development? Washington, D.C.: Institute for International Economics and Center for Global Development. Muller. Thomas, and Monica Schnitzer. 2006. "Technology Transfer aud Spillovers in International Joint Ventures." Journal oflntemational Economics 68(2):456-68. Neven, Damien, and Georges Siotis. 1996. "Thchnology Sourcing and FDI in the EC: An Empirical Evaluation." International Journal of Industrial Organization 14(5):543-60. Nicolini. Marcella, and Laura Resmini. 2007. "Productivity Spillovers and Multinational Enterprises: In Search of a Spatial Dimension." DYNREG Working Paper 10. Economic and Social Research Institute. Dublin. Pack, Howard. and Kamal Saggi. 2001. "Vertical Technology Transfer via International Outsourcing." Journal of Development Economics 65(2):389-415. Peri, Giovanni. and Dieter Urban. 2006. "Catching Up to Foreign Technology? Evidence on the 'Veblen-Gerschenkron' Effect of Foreign Investments." Regional Science and Urban Economics 36(1 ):72 -98. Roger Smee[s 137 Poole, Jennifer P. 2007, "Multinational Spillovers through Worker Turnover," Department of Economics. University of California, Santa Cruz. Prasad. Eswar, Kenneth Rogoff, Wei Shang-Jing, and M. Ayhan Kose. 2003. "Effects of Financial Globalization on Developing Countries: Some Empirical Evidence." li'vlF Occasional Paper 220. Washington D.C. Protsenko, Alexander. 2003. "Vertical and Horizontal Foreign Direct Investments in Transition Countries." Ph.D. diss., Ludwig-Maximilians University, Munich. Rodriguez-Clare. Andres. 1996. "Multinationals. Linkages and Economic Development." American Economic Review 86(4):852-73. Romer, Paul. 1990. "Endogenous Technological Change." Journal oj Political Economy 98(5): S71-S102. Saggi, KamaL 2002. "Trade. Foreign Direct Investment, and International Technology Transfer: A Survey." World Bank Research Observer 17(2):191-235. 2006. "Foreign Direct Investment. Linkages and Technology Spillovers." In Bernard Hoekman and Beata S. Javorcik. eds. Global Integration and Technology TraTl5jer. 51-65. Washington. D.C.: Palgrave Macmillan/World Bank. Schoors. Koen. and Bartoldus van der Tol. 2002. "Foreign Direct Investment Spillovers within and between Sectors: Evidence from Hungarian Data." Working Paper 157. Department of Economics. Ghent University. Belgium. Siotis, Georges. 1999. "Foreign Direct Investment Strategies and Firms' Capabilities." Journal oj Economics and Management Strategy 8(2):251-70. UNCTt\D (United Nations Conference on Trade and Development). 2007. World Investment Report 2007: Transnational Corporations, Extractive Industries. and Development. Geneva. Wang. Jian Y, and Magnus Blomstrom. 1992. "Foreign Investment and Technology Transfer. A Simple Model." European Economic Review 36(1):137-55. Wei, Yingqi, and Xiaming Liu. 2006. "Productivity Spillovers from R&D, Exports and FDf in China's Man ufacturing Sector." Journal oj International Business Studies 37 (4): 5 44- 5 7. 138 The World Bank Research Observer. vol. 23, no. 2 (Fall 2(08) Can Survey Evidence Shed Light on SpiLLovers from Foreign Direct Investment? Beata S. Javorcik Although some economists remain skeptical of the existence of positive externalities associated with foreign direct investment (FDI) , many countries spend large sums attracting foreign investors in the hope of benefiting from knowledge spillovers. Data col- lected through enterprise surveys conducted in the Czech Republic and Latvia suggest that the entry of multinationals affects domestic enterprises in the same industry or in upstream or downstream sectors through multiple channels. Some of these channels represent true knowledge spillovers while others have positive or negative effects on domestic producers in other ways. The relative magnitudes of these channels depend on !lOst country conditions and the type of PDT inflows, which explains the seemingly inconsistent findings of the literature. The focus of the debate should shift from attempting to generalize about whether or not PDT leads to productivity spillovers to determining under what conditions it can do so. JEL codes: F21, F23. 024.033 In the view of many policymakers. particularly those in developing countries. foreign direct investment (FDI) is not only a source of capital and additional employment but primarily a channel through which new technologies and know- how are transferred across international borders. Policymakers hope that know- ledge brought by foreign affiliates will spill over to domestic firms and increase the competitiveness of their economies. This belief has led many countries to use externalities associated with FDI as a justification for providing fiscal and financial incentives to foreign investors. The fact that large sums are often spent attracting FDI and the importance of technology transfer have led many academics to search for evidence of knowledge The Author 2008. Published by Oxford University Press on behalf of the International Bank for Reconstruction and Development! THE WORID BANK. All rights reserved. For permissions, please e-mail: journals.permissions@oxfordjournals.org doi:1O.L093/wbro/lkn006 Advance Access publication June 4, 2008 23:139-159 spillovers from FDI. Their conclusions have been mixed. Early cross-section studies of intraindustry spillovers find a positive association between industry-level pro- ductivity and FDI; the conclusions of recent firm-level panel analyses are more ambiguous (see the literature surveys by Gorg and Strobl 2001: Lipsey 2002: Saggi 2002; and Gorg and Greenaway 2004). The meta-analysis of Gorg and Strobl (2001) shows that cross-sectional studies tend to overstate the intraindus- try spillover effects, possibly because the studies are unable to control for unobser- vable industry heterogeneity. Among firm-level panel analyses, those focusing on industrial countries are more likely to report positive findings on intraindustry spillovers than those using developing country data. More recent work examining interindustry spillovers has produced more encouraging results by providing evi- dence consistent with the existence of knowledge spillovers from multinationals to supplying industries (see Javorcik 2004 and Blalock and Gertler 20(8). A review of the case study literature concludes that while the majority of case studies support the existence of FDI spillovers, under some circumstances such spillovers are unlikely to take place (Moran forthcoming). Critics of globalization and academic skeptics have interpreted these mixed results as reflecting "extravagant claims about positive spillovers from FDI" that are not corroborated by the "sobering evidence" (Rodrik 1999, p. 37), They suggest that one dollar of FDI is worth no more than a dollar of any other kind of investment and that there is thus no case for special treatment of FDr. Despite the mixed evidence, governments all over the world have continued their efforts to attract FDI inflows. The 1990s witnessed an explosion in the number of national investment promotion agencies. Between 1990 and 2005, the number of such agencies increased from 11 to 63 in developing countries and from 3 to 20 in developed countries. In 2004 alone, 59 of 108 countries surveyed in the World Bank's Census of Investment Promotion Agencies offered some type of incentives to foreign investors (Harding and Javorcik 2007). The contrast between the views of academic skeptics and the actions of govern- ments has left observers wondering whether academics have simply failed to uncover spillovers that indeed exist or whether the generosity with which govern- ments treat foreign investors is not really warranted. This article contributes to this debate by presenting information collected through enterprise surveys con- ducted in the Czech Republic and Latvia. Enterprise surveys can provide useful evidence that complements case studies and econometric analyses. Survey evi- dence is less prone than case studies to the criticism of not being representative and difficult to generalize. And in contrast to econometric analyses, which often treat the mechanism behind spillovers as a black box. surveys can capture the multiple channels through which spillovers take place. The survey evidence presented here illustrates the myriad channels (including both real and pecuniary externalities) through which FDI inflows affect the 140 The World Bank Research Observer. vol. 23. no. 2 (Fall 2008) performance of domestic producers in a host country. Some of these channels are true knowledge spillovers, others exert a positive effect on domestic producers through demand shocks: some may have a negative impact on the observed per- formance of local firms. The methodologies employed in most econometric studies are unable to distinguish between the various channels. To complicate matters further, the relative magnitude of these channels depends on host country con- ditions and the type of FDI inflows, which may explain the seemingly inconsistent findings of the literature. The article is structured as follows. The next section briefly describes the enter- prise survey data used. Second II focusses on intraindustry spillovers. Sections III and IV discuss the effects of foreign entry on upstream and downstream indus- tries. The last two sections present suggestions for future research and some policy recommendations. The Data The article draws on three enterprise-level surveys commissioned by the World Bank: in 2003 and 2004 in the Czech Republic and in 2003 in Latvia (for more details. see FIAS 2003, 2004; and World Bank 2007). The surveys were con- ducted by professional polling companies through face-to-face interviews at respondents' workplaces. All respondents were guaranteed full anonymity. The 2003 survey in the Czech Republic included 391 local manufacturing companies and 119 multinationals. About 20 percent of respondents were located in Prague. with the rest distributed across all regions of the country. The 2004 survey covered 466 domestic firms, 266 of them in the manufacturing sector and 200 in the service sector. The survey also included 167 multinationals. Seventeen percent of respondents were located in Prague. The Latvia survey covered 407 manufacturing firms, 11 percent of which had received FDl. About half of the interviewed firms were located in Riga. with the rest distributed around the country. All of the companies surveyed were private. While one of the main goals of the Czech surveys was to learn about the impli- cations of foreign entry for local firms, the Latvian survey contained only a limited component pertaining to this issue. This article therefore relies primarily on the Czech data, supplementing them with information from Latvia when possible. Relying on survey data may be subject to the criticism that respondents may not answer the questions truthfully. This is unlikely to be a serious concern in this case, as all three surveys were conducted by highly reputable polling firms that guaranteed full anonymity to respondents. Respondents were free to decline being interviewed or to answer a particular question. Juvarcik 141 While some pitfalls are associated with reliance solely on survey data to investi- gate economic phenomena. enterprise surveys constitute an additional source of information that complements and enriches the conclusions of econometric evi- dence and the extensive case study literature on FDI spillovers. Survey data also help make sense of the seemingly contradictory evidence produced by statistical analyses. suggesting new directions for future research. Intraindustry Spillovers The entry of multinationals may affect local firms operating in the same sector through several mechanisms. The first mechanism relies on real externalities. such as the diffusion of knowledge through the demonstration effect. As local firms observe the actions of their foreign competitors. they learn about new tech- nologies (some of which can be embodied in machinery or inputs that are rela- tively easily available for purchase), new marketing techniques, and new types of products. Local firms can also hire workers trained by multinationals. By doing so, they can find out about new management strategies and benefit from the training multinationals provided to their former employees. The diffusion of knowledge should have an unambiguously positive effect on local firms. The second mechanism takes the form of pecuniary externalities and can be referred to as a competition effect. The entry of multinational firms increases the level of competition within the industry as long as some share of their output is sold in the host country. Even host countries with liberal trade regimes may experience an increase in competition. Producing locally reduces transportation costs and. in emerging markets, labor costs. It allows multinationals to reduce the price of their products relative to the prices they charged before entering the host country. In the long run. increased competition provides incentives for domestic producers to improve their performance; it also leads to exit of the worst perfor- mers and an increase in the average productivity level in the industry. In the short-to-medium run. however. weaker firms may experience a decline in observed performance as their market share shrinks. Multinationals may also poach the best workers from their local competitors and make access to credit more difficult (because they may be lower-risk bor- rowers than their local competitors).l Both channels create negative pecuniary externalities that affect local firms. While pecuniary externalities have a negative impact on the affected firms. they lead to more efficient outcomes for the economy as a whole. As a result of increased competition in product. labor. and credit markets. resources are reallocated from less efficient firms to firms that are better positioned to benefit from them. This in turn may benefit consumers through lower prices. 142 The World Balik Research Observer, vol. 23, 110. 2 (Fall 2008) The third mechanism may work by affecting demand for intermediates. As Rodriguez-Clare (1996) notes, if multinational entry increases demand for inter- mediates, it may result in the expansion of upstream industries. A greater variety of inputs available will in turn benefit downstream industries, including the industry of multinational entry. Demand for intermediates may increase even in the presence of a liberal trade regime, as local sourcing may lead to saving on transportation costs and remove uncertainty about the timing of delivery (which is particularly important in host countries with poorly functioning customs service). The converse is also possible. If expansion of multinationals forces local firms to exit and multinationals use local inputs less intensively, a negative effect would be observed in upstream sectors as well as industries using these inputs. The results of surveys from the Czech Republic (2003) and Latvia (2003) provide evidence of all the above mechanisms. Firms in both countries reported learning about new technologies and marketing techniques from multinationals (figure 1). In the Czech Republic, local firms seemed to benefit equally from direct competitors and multinationals operating in their sector with whom they were not competing. Moreover, both firms experiencing loss of a market share and those unaffected by foreign entry reported positive knowledge externalities associ- ated with FDI (table 1). Hgure 1. Perceived Effects of' FDI Inflows into the Same Industry by Survey Respondents in the Czech Republic and Latvia .~~~~~~----~--~~~----~ 60 t! 50 5i " c: 0 Q. -w " '(I! '6 II> 01 30 ~ ~ a. 20 10 0 .~ -. '" Q. E 8 " ill " ~ is Source: Javorcik and Spatareanu (200S). 143 Table 1. Knowledge Flows from Entry of Multinationals into Sector (percent of Czech firms reporting) From multinationals in tile same As a reslJlt oj Joreign entry into the same sector sector that are respondents Item Competitors Noncompetitors Lost market share Did not lose market share Learning about new 26 22 23 25 technologies Learning about new 12 11 11 12 marketing techniques Benefiting from 5 2 3 4 knowledge of employees trained by multinationals Note: The sample size included 327 answers. Data source: FIAS (2004). Survey respondents also reported benefiting from the knowledge of workers who had previously been employed by multinationals (figure 1). This channel of spillovers is less prevalent, however. because domestic firms may have a hard time competing with multinationals in wages. However. multinationals reported that when their employees leave. they usually find employment in local firms (figure 2). Local respondents also reported an increase in competition resulting from foreign entry. Among domestic producers 41-48 percent said that foreign entry increased the level of competition in their industry; a smaller, though significant. percentage (29 percent) reported losing market share to the foreign entrants. Survey respondents mentioned that they had lost employees to multinational entrants, although this phenomenon did not seem to be very widespread. Finally. some respondents believed that entry of multinationals worsened their access to credit (see figure 1). The survey suggests both positive and negative effects on the demand for upstream production from the entry of multinationals. On the one hand. 18 percent of respondents reported benefiting from foreign entry by becoming suppli- ers to multinationals operating in their sector. On the other hand, 21 percent of respondents whose Czech clients had been acquired by foreign investors stopped supplying these clients. Of those who continued the business relationship, five firms reported having to comply with higher quality requirements. An important message to take away from these results is the difference in the reported effects of FDI inflows. While domestic firms in both countries reported similar patterns with respect to increases in competitive pressures and loss of market share. the benefits of knowledge spillovers were much more prevalent in the Czech Republic. Twenty-four percent of Czech firms. but only 15 percent of Latvian 144 The World Bank Research Observer. vol. 23. no, 2 (Fall 2008) Figure 2. Place of Employment of Former Employees of Multinationals in the Czech Republic 70~-------------------------------------------------------, 60+-------~------~--------------~------~------~------~ 50 +----- 40 +----- J 30 +----- 20 +----- 10 +----- 0-'------- Czech firm Foreign firm Own business Noll': Eighty-two multinational firms responded to the question. Datil source: FIAS (2004). firms, reported learning about new technologies from multinationals. The difference in the ability to learn about marketing techniques was much less pronounced. Whether these differences stem from differences in the composition of PDI inflows or differences in local firms' ability to absorb knowledge spillovers, the key message is that host country conditions affect the extent of knowledge spillovers. The vast majority of econometric studies cannot distinguish between the mech- anisms described above: many studies do not even include rudimentary controls for the level of competition. It is therefore not surprising that the literature has produced seemingly inconsistent results on intraindustry spillovers. Rather than interpreting a study with a marginal improvement in the methodology as invali- dating all earlier findings, it would be more productive to focus on ways of isolat- ing individual mechanisms. Some progress has already been made in this direction. Gorg and Strobl (2005) use Ghanaian data on whether or not the owner of a domestic firm had previous experience in a multinational. which they relate to firm-level productivity. Their results suggest that firms run by owners who worked for multinationals in the same industry immediately before opening their own firm are more productive than other domestic firms. Crespi and others (2007) combine self-reported data on sources of new knowledge from UK innovation surveys with information on firm-level total factor productivity. They find that competitors are one of the key sources of knowledge contributing to firm performance. They also show that reported knowledge flows from competitors are positively correlated with the presence of multinationals in the same industry. Javornk 145 Spillovers To Upstream Sectors While multinationals have a strong incentive to prevent knowledge leakage to their competitors, they may want to transfer expertise and know-how to their suppliers. Passing on information about new technologies or business practices (such as quality control processes or inventory management techniques) to suppliers reduces input costs, increases input quality, and thus benefits multinationals. If the benefits of knowledge transfer are not fully reflected in lower quality-adjusted prices, the actions of multinationals result in knowledge spillovers. As Pack and Saggi (2001) show in a theoretical model. even if technology transferred by a multinational to a developing country supplier diffuses to other firms in the supplying industry and benefits competitors of the multinational. both the developing country supplier and the multinational can benefit. In the absence of knowledge diffusion in the supplying industry and entry in the buying industry, the developing country supplier and the multinational are in a bilateral monopoly. They impose a pecuniary vertical externality upon each other by char- ging a price above marginal cost; the double marginalization problem thus exists. Knowledge diffusion in the supplying industry stimulates new entry and brings the input price closer to marginal cost, benefiting the multinational. Entry into the industry of the multinational brings the downstream price closer to marginal cost, increases output, and benefits the developing country supplier. As a result. as long as the competition resulting from diffusion in the supplying industry and entry into the industry of the multinational are not too severe. both firms gain from diffusion that leads to entry in the downstream market. Several recent studies find evidence consistent with spillovers to upstream sectors. Using firm-level panel data from Lithuania, Javorcik (2004) shows that the total factor productivity of Lithuanian firms is positively correlated with the extent of potential contacts with multinational customers in downstream sectors. A one standard deviation increase in foreign presence in the buying sectors is associated with a 15 percent rise in the productivity of Lithuanian firms in the supplying industry. The productivity effect occurs from investments with joint foreign and domestic ownership but not from fully owned foreign affiliates. This finding is consistent with the evidence of an increase in local sourcing under- taken by jointly owned projects. Evidence supporting FDI spillovers to upstream sectors has also been found in other countries (see Javorcik and Spatareanu 2008 on Romania, and Blalock and Gertler 2008 on Indonesia). Even in the case of spillovers to local suppliers, however, one should not expect a uniform effect across countries or across industries within a country. for several reasons. First, the decision to purchase inputs locally will be driven by the host country's trade regime, efficiency, and the predictability of its customs service and transport costs. The choice of input source also depends on whether a multinational 146 The World BarIk Research Observer, vol. 23. no. 2 (Fall 2008) Table 2. Determinants of Sourcing Patterns by Multinationals ------------------------------ Determinant Share of multinationals responding (percent) Reason for importing inputs from abroad Using company's global suppliers 46 Implementing decision of parent company 37 Unavailability of particular products from local firms 36 Desire to purchase higher-quality inputs 30 Reason for buying inputs from other multinationals operating in host country Using firm's global suppliers 45 More competitive prices 45 Savings on transport costs 34 Benefits of proximity 30 Higher-quality products 29 Products not available from local firms 29 Reason for buying inputs from local suppliers Low prices 71 Benefits of proximity 64 Savings on transport costs 56 Savings on import duties 44 Note: The sample size included 327 answers. Data sourte: FlAS (2004). lollows a centralized sourcing arrangement in order to benefit from volume discounts or access to customized inputs (UNCTAD 2001). Indeed. using the company's global suppliers was the main reason multinationals operating in the Czech Republic reported using imported inputs. In other cases the decision reflected the fact that particular inputs or input') of sufficient quality were not available locally (table 2). Because it may take time to develop relationships with local suppliers. one would expect new PDl projects to be less likely to use locally produced inputs than would investors with longer experience in the host country. This is confirmed by Belderbos. Capanelli, and Pukao (2000), who find that the proportion of inputs sourced locally by Japanese multinationals increases with the number of years of operation in a given host country. In sum, in situations in which upstream sectors in the host country are underdeveloped or the multinational has very specialized input needs or relies on centralized input sourcing. the scope for spil- lovers to upstream sectors may be limited. Second, even if multinationals source inputs in the host country, they may buy them from other multinationals operating there. It is relatively common for pro- ducers of parts and components to follow their clients to a new host country. In such a situation. PDl inflows would affect supplying industries only by stimulating PDl inflows into upstream sectors. which in turn would lead to intraindustry effects in upstream sectors (described in the previous section). Javorcik 147 Third, in situations in which either a world-class supplying industry already exists or only basic inputs with limited technological content are needed, there is little scope for knowledge transfer. Multinationals may simply award contracts to the best local producers, and upstream benefits may be limited to increasing the demand for inputs and allowing upstream producers to benefit from economies of scale. The extent of these benefits hinges on multinationals increasing the overall demand for inputs (rather than replacing the demand from local competitors they forced to exit) and the production technology in upstream industries. The benefits of scale economies in the form of lower prices may be passed on to local input users in the sector of the multinational or other sectors. Fourth-and more interesting from a development perspective-by imposing higher standards on their suppliers for product quality, technological content, or on-time delivery, multinationals may induce local producers in upstream sectors to make improvements. Is this situation different from firms learning about buyer expectations in foreign markets? Most likely it is, because information costs are much lower. Multinationals operate in the same country; contacting them does not require knowledge of a foreign language or entail high travel or communi- cations costs. In some cases, a multinational may even be the party that initiates contact with a potential local supplier. Another difference is the prevalence of technical audits, which tend to be less widespread for export transactions. Some 20 percent (37 of 187) of Czech suppliers reported undergoing such an audit before signing a contract with a multinational; in some cases two or three audits were performed. The technical audits, while not considered by multinationals as a form of assistance, may be invaluable to local suppliers, as they may point out operational deficiencies of which they had not been aware. Before signing a purchase order, multinationals often explicitly require future suppliers to make certain improvements. In the Czech Republic this was the case for more than a quarter of all suppliers surveyed (49 of 190). Almost half of the audits took place six months or more before the contract was signed. The most frequent requirements were improvements to the quality assurance process, acquisition of a quality certification (such as an ISO 9000), improve- ments to the timeliness of deliveries, use of a new technology, or purchase of new equipment (figure 3). The fact that improvements to the product were less frequent is consistent with the evidence suggesting that having a suitable product is in most cases a precondition for starting a dialogue with a potential multina- tional client. The prospect of receiving a contract from a multinational also seems to induce local suppliers to undertake improvements on their own. Thirty-six percent of Czech suppliers reported making improvements with the explicit purpose of finding a multinational customer. These improvements included investing in new machinery and equipment. improving product quality, conducting staff training, 148 The World Bank Research Observer. vol. 23, no. 2 (Fall 2008) Figure 3. Types of Changes Required by Multinationals from Potential Suppliers in the Czech Republic ~r-------------------------------------------------------~ C II> ~ 50+-~===-------------------~--~--~--------------------~ 0- .~ II> ~ ~ 40 .s ~ :; 30 . Table 2. Continued Study Country Year Level/wade/age Main eJJect Comment of entering university 80 percent. Having a private tutor significantly reduced this probability, but this result most likely reflects the remedial character of this form of private tutoring in Japan. Buchmann Kenya 1995 13 -18-year-olds Private tutoring reduced the chance of Study does not control for school (2002) repeating grades and increased characteristics. student academic performance. Lee and others Republic 2000-01 Middle and high school Preclass tutoring (private tutoring that does not appear to account for ~ '" (2004) of Korea teaches the school curriculum at least student motivation for receiving ~ one month ahead of schedule) had no private tutoring. ::::L "'- ttl short- or long-term effects on student " ;:: ". academic performance. f ~ ;:;- Cheo and Quah (200S) Singapore Not reported Grade 8 Time spent with private tutor had negative impact on student academic [-, Ha and Harpham Vietnam 2002 Eight-year-olds Private tutoring had no significant Study does not control for school ~ (2005) effect on writing and multiplication characteristics. .-, ~ test scores but doubled reading scores. "'" !-" Source: Authors' summary based on cited data sources. ~ ,--" "' i1? ~ "' a" a ~ 2002). and Vietnam (Ha and Harpham 2005) and negative impacts in Korea (Lee. Kim. and Yoon 2004) and Singapore (Cheo and Quah 2005). The results from these studies should be interpreted with caution. however. because of the endogeneity resulting from self-selection into tutoring (as some of the studies acknowledge). In addition, two of these studies do not control for school charac- teristics. which may further bias the estimation results. Studies that control in some credible way for the endogeneity of private tutoring generally find that private tutoring boosts student academic perform- ance. Tutoring is found to increase test scores in India (Banerjee and others 2007). mean pass rates on the baccalaureate exams in Israel (Lavy and Schlosser 2005). the quality of universities students attend in Japan (Ono 2007). Scholastic Aptitude Test (SAT) and ACT test scores (except for ACT reading scores) and aca- demic performance in the United States (Briggs 2001; Jacob and Lefgren 2004).10 and student academic performance in Vietnam (Dang 2007b). The sole exception is in Indonesia (Suryadarma and others 2006). where tutoring was not associated with higher performance by fourth graders.l1 No studies appear to have exam- ined whether the estimated negative correlations between private tutoring and achievement in Korea and Singapore change when endogeneity is properly addressed. The three studies on India, Israel, and Vietnam are examined in some detail because they include cost data, making it possible to consider both the impact of tutoring on academic performance and the cost-effectiveness of tutoring. Furthermore. these studies reflect the variation in the usage and financing of tutoring. They include both low-income (India and Vietnam) and high-income (Israel) countries; they include estimation of tutoring effects on students of all academic abilities (Vietnam) and on underperforming students (India and Israel); and they include tutoring that is financed by a nongovernmental organization (NCO; India). publicly financed (Israel). and financed by households (Vietnam). NGO-financed Remedial Tuton'ng in India. Pratham, a large Indian NCO, financed the implementation of a two-year in-school randomized tutoring program that targets poor children in two large cities. This remedial education program targeted children in grades 3 or 4 who had not mastered basic skills. These stu- dents were taken out of their classrooms and given two hours of supplemental instruction each day by young women from the community. Banerjee and others (2007) find that this tutoring program improved student test scores by large and statistically significant amounts. Children randomly assigned to the treatment group improved their test scores by 0.6 standard devi- ations in the second year; control group children remaining in the regular class- room did not benefit. Overall. the test scores of children in schools participating in the program rose 0.14 standard deviations in the first year and 0.28 standard Dang and Rogers 175 deviations in the second year. These gains fell substantially one year after the program ended, however. More research is needed on the long-term impact of such programs. Banerjee and others attribute the relative success of the program to regular tea- chers' lack of motivation to help lagging students and to the common background shared by the students and the tutors. They also show that this tutoring program is cost-effective. At about $10-$15 a month. the tutors' salary is equivalent to just 6 -10 percent of the salary of a starting teacher. They calculate that scaling up the tutoring program would be much more cost-effective than hiring new tea- chers, at least in terms of raising test scores. The program. which has already reached tens of thousand of children across India, is estimated to cost about $2.25 per student a year. A second program used computers, rather than human tutors, to deliver the tutorials. This program raised math scores by 0.36 standard deviations the first year and 0.54 standard deviations the second year. However, it is much more expensive than the first program, at $15.18 per student a year. Government-financed Remedial Tutoring in Israel. Israel has been operating a reme- dial education program for underperforming high-school students since 1999. By 2004 the program was reaching about a third of all high schools (Lavy and Schlosser 2(05). The objective of the program is to increase the number of stu- dents who earn baccalaureate certificates by providing them with increased instructional time. In each school the program identifies up to five students in the 10th, 11 th, and 12th grades judged most likely to fail the exams. The classroom teachers then provide these students with after-school tutoring in the subjects in which they are weak. To examine the effects of the program, Lavy and Schlosser (2005) use a quasi- experimental difference-in-difference methodology (supplemented by instrumental variables as an alternative identification strategy). Their approach relies on the fact that the program was rolled out over time. allowing them to compare learning gains in schools that received it early on with those in schools that received it later. Lavy and Schlosser find that the program had a positive impact on both the students and the participating schools. The program increased the probability of a tutored student earning a baccalaureate certificate by 12 percentage points, an average improvement of 22 percent over the base rate. The targeted schools saw an increase of about 3.3 percentage points in mean pass rates on the baccalaure- ate exams, equivalent to an improvement of 6 percent over the base rate. The program did not appear to affect non tutored students. At $1,100 per tutored student (about 40 percent of annual expenditure per high school student in Israel) the average cost of the program is very high. Nevertheless, Lavy and Schlosser estimate the program's internal rate of return at 176 The World Bank Research Observer. vol. 23, no. 2 (FaIl 2008) 20 percent. Although this makes the remedial tutoring program less cost-effective than two other incentive-based programs in Israel examined by the authors. 20 percent is an impressive rate of return. Household-financed Tutoring in Vietnam. Private tutoring is very popular in Vietnam. with about 34 percent of households with children in school purchasing private lessons. Ninety percent of these households allocate 1- 5 percent of total household expenditure to private tutoring. Some 31 percent of primary students, 56 percent of lower-secondary students. and 77 percent of upper-secondary school students receive private tutoring (Dang 2007b). Dang addresses the endogeneity of tutoring with an instrumental variables strategy that uses tutoring prices as the instrument. Using data from the Vietnam Living Standards Surveys for 1992-93 and 1997-98 he shows that private tutoring improves student academic performance. After controlling for other indi- vidual. household. school. and community characteristics he finds that raising annual spending on private tutoring from 0 to 20,000 dong (D)-about $1.50 in 1998, equivalent to about 0.4 percent of mean consumption or 2 percent of spending on education by households with children in school-has strong positive effects on performance. For primary school students, tutoring reduces the prob- ability of obtaining a "poor" academic ranking by about 1 percentage point. reduces the probability of average performance by 4 percentage pOints. and increases the probability of good or excellent performance by 5 percentage points. For the same increase in expenditure at the lower-secondary level it reduces the probability of poor performance by about 1 percentage points, reduces the probability of average performance by 7 percentage points, and increases the probability of good or excellent performance by 8 percentage points. These estimates can be used to make a rough calculation of the cost-effective- ness of tutoring in promoting grade progression among lower-secondary school students. Students with "poor" academic rankings usually have to repeat grades in Vietnam. A year of lower-secondary schooling has a total cost (direct costs to households and the government plus assumed opportunity cost of forgone wages) of about D3 million a year. Everything else being equal, if the household's aim is for the child to complete a given level of education. a reduction of 1 percentage point in the probability of earning a "poor" academic ranking reduces the expected costs of grade repetition by about D30,000. Similar calculations for primary students also show that D20,000 worth of private tutoring reduces the expected costs of grade repetition by about D25,000.12 This means that the benefits from lower repetition rates alone exceed the costs of tutoring. The total benefits are likely to be much higher, because these calculations do not account for any economic benefits of better academic ranking categories or for avoided psychological costs associated with grade repetition. Dang and Rogers 177 Interpreting the Evidence Recent studies that have dealt with the endogeneity of tutoring and estimated cost-effectiveness find that private tutoring has strong positive returns. Some caution is needed in interpreting the evidence. however. for several reasons. First. this line of analysis is made difficult by the nature of private tutoring transactions: typically it is the household (rather than an NGO or government) that decides whether to purchase private tutoring. It is hard to imagine how that decision could be randomized across households in order to obtain clean measurement of the returns to privately purchased tutoring. At best a government could random- ize access to potential tutors-by, for example, flooding certain randomly chosen communities with unemployed graduates-and then examine whether house- holds chose to consume more tutoring and if so how it affected student performance. Second, estimation results from these studies should be considered in context and neither generalized to all students nor narrowed to specific subgroups of tutored students. Most of the studies cited in table 2 estimate only the average return for all students enrolled in tutoring. This approach implicitly assumes that all students share the same returns to private tutoring, regardless of their innate ability or socioeconomic background (exceptions include the studies by Jacob and Lefgren 2004; Lavy and Schlosser 2005; and Banerjee and others 2007.) If this homogeneity assumption is violated, the estimated benefits of tutoring will not apply to subgroups and may be biased as well (see Heckman, Lochner. and Todd 2006 and Heckman. Urzua. and Vytlacil 2006 for discussions of the heterogen- eity in returns to education in instrumental variables models). For example, while the studies for India (Banerjee and others 2007). Israel (Lavy and Schlosser 2005). and the United States (Jacob and Lefgren 2004) show that remedial private tutoring improves student performance. this result may hold only for the grades evaluated. It is not easy to take account of this heterogeneity in returns to private tutoring. Doing so requires detailed data on the student variables that may affect returns, as well as more sophisticated estimation techniques. Crafting more detailed policies demands more in-depth analysis of specific groups in particular contexts. Third. the evidence presented here on the returns to investment in private tutoring refers only to private returns. Little is known about the social returns to private tutoring. Even if tutoring is completely financed by households. any externalities of private tutoring would need to be estimated before social returns could be calculated (but see discussion below on the social costs of tutoring in a signaling equilibrium). Subject to these caveats, it appears that tutoring has strong positive private returns as a supplement to formal public school education. The programs on 178 The World Bank Research Observer. vol. 23. no. 2 (Fall 2008) which there is evidence may provide good starting points for policymakers seeking to design and implement supplementary education programs. What Stance Should Policymakers Take toward Tutoring? Efficiency and Equity Considerations How does private tutoring affect efficiency and equity? This section first analyzes this question while maintaining the assumptions that the market is perfect and the supply curve for public schooling perfectly inelastic after a certain point. It then examines how these conclusions change when these assumptions are relaxed. Efficiency Considerations in the Standard Case The micro evidence on private tutoring suggests that it generally improves student academic performance for the average tutored student. More limited evi- dence suggests that these improvements can be cost-effective. A question for pol- icy makers is whether. from a broader social perspective. the availability of private tutoring increases overall welfare. Are the societal gains from private tutoring likely to exceed its costs? No studies appear to have been conducted on the efficiency of private tutoring at the macro level. However. combining the micro evidence with the analytical framework set forth above can help answer this question. The availability of private tutoring increases efficiency and welfare, under certain assumptions revisited below (figure 1). For a household whose demand for education is represented by demand curve Db the availability of private tutor- ing raises consumption of education to Q2' This is more than the amount of public education (02) that the household consumes in the absence of private tutoring and more than the amount of private education (00) that the household can afford. (If the household's demand for education is on the low-demand curve. Dl · the household would consume no private tutoring given this supply curve.) Household consumer surplus increases by the amount represented by the triangle BeE. Tutors gain the producer surplus represented by the triangle ABE. One other effect is not shown directly in figure 1. High-demand households that in the absence of private tutoring would have chosen private schools may now enroll their children in public school. As demand for pUblic schools increases. the costs to the government might be expected to rise and the producer surplus to private schools to fall. But because public schools are assumed to be on the verti- cal portion of their supply curve---that is. they have reached their capacity-the Dang and Rogers 179 quantity of education provided by the government (and therefore government outlays) does not actually increase. Standard micro analysis makes it clear that the total gains to households and tutors should exceed the losses in the private school sector. Thus offering the opportunity to supplement public education with private tutoring increases welfare for households and society as a whole-at least in the standard model. Equity Considerations in the Standard Case Suppose policy makers are convinced that a robust private tutoring sector improves welfare but worry that it may increase inequality. There is indeed reason for concern. More privileged households-those with higher income and more education who live in urban areas-invest more in tutoring than other house- holds do. and private tutoring appears to increase learning achievements for these children. at least on average. If learning achievement translates into higher lifetime earnings. one would expect the availability of household-financed tutoring to increase social inequalities. One should not be too quick to equate tutoring with increasing inequality. however. or to assume that an equity-focused government should try to limit tutoring. for several reasons. First, when the appropriate counterfactual-what would happen in the absence of a private tutoring sector-is specified. tutoring may not increase educational inequality by as much as suggested above. Even productive tutoring may confer only a minor advantage on children from weal- thier and more-educated households. because these households already give their children educational advantages in many other ways-by providing them with more books. more learning equipment. and even full-time private schooling, for example, or by teaching their children themselves. Even if it were enforceable, a ban on private tutoring would likely simply redirect the education expenditures of better-off households into these other investments. Furthermore, access to sup- plementary private tutoring may benefit poorer households if it helps their chil- dren compete with wealthier children enrolled in private schools. Second, tutoring may emerge as an unintended result of other government education policies. including some policies aimed at promoting equity. Imagine, for example. that the government substantially increases its per student financing for public education in poor (low-demand) neighborhoods. This would shift the supply curve (S2 in figure 1) downward. If the shift is substantial enough. it will induce low-demand households to consume more education, including more private tutoring. To control educational inequality. governments may find it more effective to attack its roots than to discourage tutoring. which is in part a symptom of inequality. Korea took this tack in 1974. when it sought to control the growth of 180 The World Bank Research Observer. vol. 23. no. 2 (Fall 2008) private tutoring by adopting a secondary school equalization program (Kim 2005). That program switched to allocating secondary school entrance by lottery rather than examination, reducing the quality advantages of higher-ranking schools and the incentive for exam-preparation tutoring. While demand for tutor- ing has remained high, Korea is not generally believed to have severe inter genera- tional transmission of inequality through education. Third, governments can use tutoring to improve equity. The household- financed tutoring in the market equilibrium in figure 1 benefits children from high-demand households, which tend to be wealthier, but governments and others can target special tutoring programs at underperforming students. as Israel (Lavyand Schlosser 2005) and the United States (Jacob and Lefgren 2004) have done. In effect. the government would be segmenting the market depicted in figure 1 by driving the supply curve downward for low-demand households only. In this case the equity implications are clearly positive. as long as the subsidy is financed progressively. What if the Availability oj Tutoring Impedes Public School Improvements? The first assumption that needs to be relaxed is that public education is capacity- constrained-meaning that at the upper end of its range the supply of education is perfectly inelastic. This assumption is likely to hold only in the short term. Over the longer run, governments can and do take steps to increase the quantity of effective education, for example by expanding school capacity to allow longer schooling hours, improving teacher attendance and time on task. and ultimately hiring more teachers and building more schools. Such improvements extend the upward-sloping portion of the public supply curve in figure 1 and shift the vertical section outward. Distinguishing between the short and long runs therefore matters. Under the earlier assumptions of the short-run standardized framework model depicted in figure 1. private tutoring occupied a neutral territory unaffected by the debate that pits public schools against private ones. Public and private schools are typi- cally depicted as substitutes. which they generally are, at least at the level of the individual student (from the perspective of the school system as a whole. private schools may be viewed as a useful complement to government schools). But in the situation depicted in figure 1. tutoring is a complement to public schooling. It enables parents to invest in an optimal amount of education for their children. increasing both consumer and producer surplus. Thus private tutoring and public education appear to be complements in the short run. In the long run-----defined here as the time it takes to make substantial improvements in the quality and quantity of public schooling-private tutoring may substitute for public education. The availability of tutoring could diminish Dang (inti Rogers 181 parents' interest in lobbying for long-term improvements in public education. If urban elites find that tutoring gives their children an advantage in competitive examinations or the labor market and they fear that any future public school improvements would go primarily to schools serving disadvantaged areas, they may prefer the status quo. In Japan it has been argued that education reforms to expand public school activities have been blocked by the dependence on private tutoring to perform these tasks (LeTendre 1994). The tutoring market may serve as an outlet releasing political pressure for reform and quality improvement. In the long run, private tutoring may provide less of a spur to public quality than competition from private schools does. Some scholars (Hoxby 1994; Rouse 1998; Bishop and Wossman 2004) argue that the loss (or potential loss) of students to private schools puts pressure on public schools to improve quality. Private tutoring would likely have no such effect. because it does not cause students to abandon public schools. The question is not whether private tutoring enables or undermines the public sector's role as a provider of education. Public schooling will continue to be a part of virtually every national primary and secondary school system. The point is simply that where tutoring is widespread, it will likely have important effects on the quality and efficiency of public schools. Policy will need to take account of these effects over both the short and long terms. What if Teacher Corruption makes the Tutoring Market Uncompetitive? A second assumption underlying the standardized model is that the market for tutoring is competitive. This may not be the case. Public school teachers may have substantial market power as suppliers of private tutoring, especially in remote rural areas, where they may be the only potential suppliers of private tutoring. More worrisome is the fact that teachers who are corrupt and poorly monitored sometimes force their public school students to take private tutoring lessons from them or omit part of the curriculum during regular classroom hours and save it for their tutoring lessons (Buchmann 1999; Foondun 2002; Glewwe and Jayachandran 2006; Silova and Bray 2006a). Others give preferential treatment to particular children in return for a fee. This practice may reduce teacher time and energy in mainstream classes, or it may encourage teachers to work additional hours. Teachers' monopoly power reduces the consumer surplus of high-demand con- sumers. The dysfunctional monitoring system coupled with teacher corruption blurs the line between public education and private tutoring. Graphically, this would increase the slope of the (now ostensibly public) supply curve S2 and shift it to the left in figure I, forcing households to pay a higher price for the same amount of education. The more market power teachers have. the farther leftward 182 The World Bank Research Observer, vol. 23, no. 2 (Fall 2008) they may try to shift the supply curve. As one would expect, given teachers' mon- opoly power, consumer surplus falls by more than the gain in producer surplus to the teacher. In such cases. private tutoring is not likely to yield the substantial returns to tutoring documented in the empirical studies cited above. This outcome is worse than the no-corruption competitive private tutoring equilibrium: for households it may also be worse than having no tutoring at all. In the worst case, tutoring fees are simply a net transfer from households to tea- chers: the amount of education provided remains the same, but the teacher deli- vers part of it for a fee outside of school hours. In rural areas, where teacher governance is poor and this type of corruption is most likely to flourish, the trans- fer will usually be regressive, because teachers tend to have considerably higher incomes than the average rural resident. Moreover, evidence on service delivery suggests that it is the poorest households that suffer most from failures in service delivery (World Bank 2003) and pay the largest bribes relative to their consump- tion level (Hunt and Laszlo 2005). As a result, the transfer will likely reduce equity and overall welfare. This analysis of potential teacher corruption suggests that in the absence of mechanisms to control teacher corruption. allowing private tutoring may be counterproductive in some cases. Given the difficulties and undesirability of banning tutoring outright, it provides a rationale for measures to prevent public school teachers from tutoring their own students privately. Ukraine's education ministry imposed such a ban in 2004 in response to complaints from parents that teachers were providing "compulsory private tutoring" (Hrynevych and others 2006). What if the Purpose of Tutoring is not only to Increase Human Capital? How does the diagnosis change if tutoring is not necessarily productive from a societal perspective? The analysis assumed that an increase in education units consumed not only increases a student's future productivity (and hence wages) but also increases societal productivity by an equivalent amount. In theory this need not be the case. If tests measure student characteristics that have signaling value but no productive value, tutoring may not increase the productivity of tutored students, even though it increases their wages. This would be the extreme version of the signaling model introduced by Spence (1973). In such a signaling equilibrium policymakers would be right to worry about the social costs of the tutoring industry, which would in essence be an arms merchant in a negative- sum education arms race. While this extreme theoretical case certainly does not apply anywhere, criticism of some aspects of otherwise high-performing education systems in Korea and Japan-both of which have very large private tutoring sectors-has cited the Dang and Rogers 183 perceived uselessness of some of the material tutoring students master in pre- paration for university entrance examinations. Concerns about the heavy finan- cial burden of tutoring on parents have led to reform in Korea (Kim 2001). These concerns are consistent with the argument that the long-term financial returns do not justify the costs of private tutoring. Empirical evidence suggests a bunching of private tutoring investment immediately before school-leaving or uni- versity entrance exams, which is consistent with a signaling story. Standardized tests for admission to law school in the United States. which have given rise to a large test-preparation industry. have been criticized for being widely used despite their inability to predict applicants' performance as lawyers (Haddon and Post 2006). The fact that students in the United States prepare for the Law School Admission Test (LSAT)-supposedly an aptitude test-by acquiring test-taking skills taught in tutoring courses suggests that the test results contain an element of signaling. If tutoring were contributing only to productive human capital. it would not likely raise scores on an aptitude test, at least not as a result of the short courses offered by tutoring companies. Although it is analytically difficult to distinguish between the signaling and productive human capital stories. signaling incentives are likely to explain some tutoring in societies that make heavy use of tutoring (Rogers 1996; Chae. Hong, and Lee 2004; Lee 2007). But three points should be kept in mind. First. although a (partial) signaling story changes the situation depicted in figure 1 to a degree. it does not qualitatively change the conclusions. A signaling equilibrium makes the slope of the private tutoring supply curve steeper. by reducing the effec- tive units of education (human capital) received. but the outcome does not change fundamentally. Second. countries such as Korea that apparently make greater use of signaling are among the highest performers on well-designed. internationally normed student assessments such as the Programme for International Student Assessment (PISA) and the TIMSS. These high scores suggest that students are acquiring a large amount of real human capital even if they are also investing in signaling value. Third. the appropriate response in the signaling case is probably not to discou- rage tutoring-as Korea did in 1980 by banning it-but to address the problem at its source. The government could. for example. revise university admissions policies so that they place less reliance on a single examination. which makes for a tempting signal of a student's ability. PoLicy Implications The evidence suggests that tutoring can raise educational outcomes as a comp- lement to formal school systems. In the absence of corruption. and given the 184 The World Bank Research Observer, vol. 23. no. 2 (Fall 2008) assumptions discussed above. private tutoring increases the welfare of households and society as a whole. Private tutoring may place poorer households at some dis- advantage relative to richer households, however, particularly when corruption distorts the tutoring market. Corruption also reduces the efficiency of the tutoring equilibrium. This section explores what these drawbacks imply for education policy in developing countries. Government Policy and the ControL of Corruption Before turning to policy recommendations, it is useful to consider the policies govern- ments have implemented. Bray (2003) divides governments into four types: those that ban private tutoring (type f), those that ignore it (type II), those that recognize and regulate it (type III), and those that actively encourage it (type IV; table 3). Bray and Silova (Bray 2003, 2006; Silova and Bray 2006b) offer helpful detail on the different types of governments. They note that type I governments all failed in their attempts to ban private tutoring. Bans in Cambodia and Myanmar failed because those countries' institutions were too weak to implement the policy. In Korea and Mauritius the bans faced too much opposition from vested interests. forcing the governments to lift the bans and regulate private tutoring. Type II governments ignore private tutoring. These governments can be divided into two groups based on their reasons for ignoring private tutoring. Countries in the first group (including Nigeria and Sri Lanka) have weak institutions and little capacity to monitor private tutoring. Countries in the second group (including Canada and the United Kingdom) have stronger institutions and adequate capacity to monitor private tutoring. They choose not to regulate the sector, either because they consider it to have small and insignificant effects or because they prefer to let market forces govern the sector. Type III governments (such as Hong Kong. China; Mauritius; and Vietnam) take a more active role in controlling private tutoring. These governments recog- nize the importance of private tutoring and attempt to control it both directly and indirectly. They may prohibit private tutoring in early grades; forbid teachers from tutoring their own students; stipulate fees. class sizes. or syllabi for private tutor- ing classes; and reduce disparities across schools. Type IV governments (including Singapore, South Africa, Tanzania) actively encourage private tutoring. These governments believe that private tutoring con- tributes to human capital development and that private tutoring is an effective means of tailoring education to the needs of students. Policies in type IV countries range from offering general encouragement to providing subsidies for private tutoring. training courses for tutors, and tax incentives. It may be useful to add another dimension to this framework: control of cor- ruption (see table 3). By this measure (taken from Kaufmann. Kray and Mastruzzi Dang alld Rogers 185 ...... 00 0'\ Table 3. Government Policies toward Private Tutoring in Selected Economics Type Policy Measure Country Comment Prohibit private Total ban on private tutoring, Cambodia (7), Republic of Korea (65), Tutoring was banned, but the bans were tutoring Myanmar (1), Mauritius (67) ineffective because of government's inability to enforce them, II Ignore private Croatia (58), Georgia (45), Nigeria (6), Most of these countries have weak tutoring Mongolia (37), Sri Lanka (49) institutions and do not have the capacity to regulate private tutoring. Canada (94), United Kingdom (94) These countries have strong institutions and the capacity to regulate private tutoring. but they consider the private :;i tutoring market outside of their sphere '" of responsibility. ~ ~ III Recognize and Generally prohibit private tutoring in early Hong Kong (China) (93), Lithuania l?f regulate private grades and prohibit teachers from (60), Republic of Korea (65), ~ tutoring tutoring their own students. Regulations Mauritius (67). Ukraine (28). f ;::j :::l"' are accompanied by inspections and sanctions on private tutoring fees, class Vietnam (29) sizes, and syllabi. Regulations are placed ~ -. on infrastructure of private tutoring .., iii centers. Policies seek to reduce