Public R&D Policies and Private R&D Investment: A Survey of the Empirical Evidence

Transcription

1 Public R&D Policie and Private R&D Invetment: A Survey of the Empirical Evidence Bettina Becker* Aton Buine School Abtract The importance of R&D invetment in explaining economic growth i well documented in the literature. Policie by modern government increaingly recognie the benefit of upporting R&D invetment. Government funding ha however become an increaingly carce reource in time of financial crii and economic auterity. Hence it i important that available fund are ued and targeted effectively. Thi paper offer the firt ytematic review and critical dicuion of what the R&D literature ha to ay currently about the effectivene of major public R&D policie in increaing private R&D invetment. Public policie are conidered within three categorie, R&D tax credit and direct ubidie, upport of the univerity reearch ytem and the formation of high-killed human capital, and upport of formal R&D cooperation acro a variety of intitution. Crucially, the large body of more recent literature oberve a hift away from the earlier finding that public ubidie often crowd-out private R&D to finding that ubidie typically timulate private R&D. Tax credit are alo much more unanimouly than previouly found to have poitive effect. Univerity reearch, high-killed human capital, and R&D cooperation alo typically increae private R&D. Recent work indicate that accounting for non-linearitie i one area of reearch that may refine exiting reult. Keyword: Reearch and Development; Reearch and Development Policy; Innovation Policy; Public Funding. * Economic and Strategy Group, Aton Buine School, Aton Univerity, Aton Triangle, Birmingham B4 7ET, UK. Tel.: +44 (0) b.becker@aton.ac.uk. 1

2 1. INTRODUCTION In the light of the importance of R&D invetment in explaining economic growth, it come a no urprie that analyi of the driving factor of R&D remain a ubject of key methodological and empirical concern to economic reearcher. The poitive impact of R&D on growth and productivity ha been predicted by a coniderable number of theoretical contribution, and a broad corpu of empirical work ha upported thi reult at the firm, indutry and country level (ee, inter alia, Arrow, 1962a; Romer, 1986, 1990; Groman and Helpman, 1991; Aghion and Howitt, 1998; Proudman and Redding, 1998; for the theory and, inter alia, Cameron, Proudman and Redding, 2005; Kafouro, 2005; Coe, Helpman and Hoffmaiter, 2009; O'Mahony and Vecchi, 2009; Bravo-Ortega and Marin, 2011; for recent empirical evidence). Policie by modern government increaingly recognie the benefit of upporting R&D invetment. In part thi i tetimony to the importance of Nelon' (1959) and Arrow' (1962b) early inight into the motive underlying R&D invetment. In eence, the argument proceed from the obervation that indutrial R&D exhibit a claic public good problem in that it i both non-rivalrou and not (completely) excludable. If the private rate of return thu i below the ocial rate of return, a firm are unable to fully appropriate the return from their R&D, private R&D invetment ha poitive externalitie and could be lower than ocially optimal. The empirical evidence provided in Griliche (1979, 1998) confirm that the private rate of return to indutry R&D typically i below the ocial rate of return. Thi mimatch of return provide economic jutification for government upport of private R&D. Government funding ha however become an increaingly carce reource in time of financial crii and economic auterity. Hence it i important that available fund are ued and targeted effectively. The objective of thi paper therefore i to offer the firt ytematic review and critical dicuion of what the R&D literature ha to ay currently about the effectivene of major public R&D policie in increaing private R&D invetment. Thi review conider direct and indirect effect of policie, different channel through which policie take effect, and type of firm or indutrie that tand to benefit mot from different policie. Baed on the review, remaining challenge for future reearch are identified. 2

3 Public policie are conidered within three categorie, R&D tax credit and direct ubidie, upport of the univerity reearch ytem and the formation of high-killed human capital, and upport of formal R&D cooperation acro a variety of intitution. There i to date no urvey that draw together the exiting evidence on the effect of thee major type of direct and indirect government upport of private R&D. Moreover there are no individual urvey of the fat growing empirical literature on the econd and third R&D policy categorie. Survey of the likewie rapidly advancing literature on the firt category, R&D tax credit (Hall and Van Reenen, 2000) and direct ubidie (David, Hall and Toole, 2000; García-Quevedo, 2004), exit eentially only for the early, mainly pre-2000 work. Crucially, the large body of more recent literature oberve a hift away from the earlier finding that public ubidie often crowd-out private R&D to finding that ubidie typically timulate private R&D. The recent evidence on the effectivene of tax credit alo importantly ugget much more unanimouly than concluded in the earlier work that there are poitive R&D effect. Thi review focue on the recent empirical evidence. 1 With regard to univerity reearch, pecific and general meaure of high-killed human capital, and R&D cooperation, the more recent empirical evidence alo find a number of poitive effect on private R&D invetment. The paper i tructured a follow. A the focu i on the empirical literature, ection 2 provide a brief overview of the methodological iue involved in etimating model of R&D invetment. Section 3 firt preent the key prediction from theory regarding the R&D effect of each type of public policy. It then review the exiting empirical literature and link the reult to the theory. Section 4 dicue ome remaining quetion and challenge for future reearch. In concluding, ection 5 review the main reult. 1 A the focu i on the empirical literature, the paper concentrate on the eminal thu typically early contribution regarding the theoretical literature. 3

4 2. METHODOLOGICAL ISSUES 2.1 Data and meaurement of R&D The firt iue i how to meaure and compare R&D acro firm, indutrie and countrie. Input meaure, uch a R&D expenditure or R&D intenity, a well a output meaure, uch a patent or innovation count, have been ued. One advantage of input meaure i that their economic (monetary) value may be taken a homogenou, while the economic value of output meaure uch a patent count i heterogeneou. Furthermore, the propenity to patent varie coniderably acro indutrie and countrie, and even a high patent count need not imply a high level of innovation a ome patent may never be implemented. However, becaue of it input character, higher R&D pending need not necearily imply higher innovative output either. In practice, input and output meaure appear to be correlated (e.g. Ac and Audretch, 1988; Bound, Cummin, Griliche, Hall and Jaffe, 1984). Mot of the empirical work on the determinant of R&D focue on R&D expenditure or R&D intenity. The comparability of the reult from tudie uing different dataet may nonethele be impeded by the difficulty in meauring the level of R&D expenditure accurately. Firm are given coniderable latitude in what they chooe to claify a R&D, and the definition ued may differ between dataet. Cohen and Mowery (1984), for intance, find that for the ame US firm and year, Standard and Poor' Computat data reported an average of 12% more R&D than the Federal Trade Commiion' Line of Buine Program data, with the difference reulting from the definition ued. The Fracati Manual publihe internationally agreed tandard defined by the OECD (OECD, 2002). However, it i not alway obviou from the literature whether the data definition ued follow the Fracati Manual. Hall (2006) provide a concie overview of the meaning of the term `R&D', it economic analyi and it attribute a an invetment. 2.2 The R&D equation form Mot tudie ue a a tarting point a imple panel data model of R&D invetment of the rit = a + β' Xit + εit (1) 4

5 where i indexe the cro-ection unit, uually firm, indutrie or countrie, t indexe the unit of time, uually year, r denote R&D expenditure, X denote the vector of explanatory variable, α i a contant, and εit i the error term. Other than a a convenient empirical tarting point for an analyi of the determinant of R&D, (1) may be conidered to be the tochatic form of the demand equation for R&D capital a derived from a CES production function where R&D and the flow of R&D invetment are proportional to each other in teady tate. Unoberved heterogeneity between the cro-ection unit, a long a thi i broadly table over time and additive, can be controlled for by including fixed effect in the regreion. Example of thee effect are managerial ability, language or culture. Model (1) can thu be re-written a the conventional within-group or leat-quare dummy variable etimator: rit = γ' Xit + fit + εit (2) where f denote the fixed effect. Some tudie capture common technology hock and other time-variant common effect by including time dummie in (2). There are in principle two way to meaure the impact of an R&D tax credit in an R&D equation uch a (2). The firt i a dummy variable equal to one if a credit i available and zero otherwie. While thi i imple to ue, diadvantage include it relative impreciion, a different firm may face different credit level, and, if it varie over time, that it i not eparately identifiable from time dummie. The econd meaure, much ued in recent tudie, i a price variable uch a the uer cot of R&D, that capture the marginal cot of R&D, whereby the etimated R&D repone i converted to a price elaticity. Thi meaure i omewhat more accurate a it etimate the repone directly. 2 R&D ubidie are imilarly meaured a a dummy variable or by their financial amount. More recently, ubidy effect have increaingly been inferred from treatment effect analye comparing `treated', i.e. ubidy-receiving, and `untreated' firm. One advantage of uch a non-parametric methodology i the availability of a counterfactual. Meaure of geographically localied pillover from univerity reearch to private R&D include reearch pending by department and the number of cience-pecific department of different quality within a given region and indutry. Meaure of pecific or general high- 2 For a detailed aement, ee Hall and Van Reenen (2000). 5

6 killed human capital include the number of cientit and engineer in a firm, the hare of the total number of worker with higher or tertiary education, and year of formal chooling. Dummy variable are typically ued to meaure whether or not a firm i a member of a joint venture or a formal R&D cooperative agreement. There are a number of characteritic of R&D that ugget thi type of invetment hould not be analyed in a tatic framework a in (2) but in a dynamic framework. One uch characteritic i that R&D typically behave a though it ha high adjutment cot. Theory ugget thee are important becaue of the high cot of temporary hiring and firing of highly killed employee with firm-pecific knowledge. Firm therefore tend to mooth their R&D invetment over time (Hall, Griliche and Hauman, 1986; Lach and Schankerman, 1988). Hall (1993) report that at leat 50% of R&D budget typically conit of the wage and alarie of highly qualified cientit and engineer, and the more recent figure of 60% reported in Bond, Harhoff and Van Reenen (2005) ugget that thi hare ha rien omewhat over time. Another characteritic of R&D that call for a dynamic approach i that there i typically a high degree of uncertainty aociated with the output of R&D invetment, and utained commitment to R&D i often required for project to be ucceful. The role of uncertainty i implicit in the early adjutment cot literature in the context of capital invetment (Einer and Strotz, 1963; Luca, 1967), which capture the role of backward-looking expectation formation through lagged variable. More recently, part of the growing literature on irreverible invetment ha criticied thi eentially ad-hoc approach to the pecification of adjutment cot function. Tobin (1969) made explicit the role of future expectation in q- model of invetment. Mot R&D tudie ue tandard invetment equation methodology to incorporate adjutment cot dynamic into the tatic R&D model (2), where the two main approache are a neoclaical accelerator model with ad-hoc dynamic and an Euler equation a derived from forward-looking dynamic profit maximiation by firm. 3 A Euler equation repreentation of R&D invetment tend to be little robut or informative (e.g., Hall, 1991; Bond, Harhoff and Van Reenen, 2005), mot tudie ue the former approach to model dynamic by introducing a lagged dependent variable into (2): 3 Hall (1991) and Mairee, Hall and Mulkay (1999) provide detail on the econometric etimation of thee model. 6

7 rit = ρri,t-1 + δ' Xit + fit + εit (3) 2.3 Endogeneity Incluion of lagged R&D in (3) require an intrumental variable etimator in order to avoid the downward bia that can reult when uing a fixed effect etimator in panel where the number of time period i mall (Nickell, 1981). If the firm or indutry repond to expectation of future technological hock, thi may alo reult in endogeneity bia. In general, trict exogeneity would imply that hock to current R&D, εit, do not affect the future value of the explanatory variable, and thi aumption clearly doe not hold for a dynamic model that include a lagged dependent variable. Intrumental variable can further control for endogeneity or imultaneity bia that may plauibly arie from mot type of R&D policy meaure in X: Government ubidie provided to the private ector may be endogenou, a the ucce of an application for funding depend on the characteritic of the firm and the application. Tax competition alo implie that government policy and R&D tax credit may be endogenou. The uer cot of R&D, for example, i a function of both the tax ytem and a range of other economic variable, uch a the economy' real interet rate. When uing a meaure of highly qualified human capital a an explanatory variable in an R&D equation, etimation alo need to take account of potential double counting. Indeed, R&D peronnel and R&D pending have ometime been ued a alternative dependent variable of R&D regreion. One intrumental variable technique that ha been applied relatively widely in the R&D panel data literature i the firt-difference generalied method of moment (GMM) etimator (Anderon and Hiao, 1982; Arellano and Bond, 1991). The firt-differencing tranformation eliminate the individual fixed effect from the model and in contrat to the fixed effect etimator doe not rely on aymptotic conitency in the time dimenion. However, thi etimator may be ubject to large finite ample bia in cae where the intrument available have weak predictive power. Thi applie in particular when a time erie i highly peritent, a i R&D, becaue lag will be poor predictor of future 7

8 outcome. 4 Bloom, Griffith and Van Reenen (2002) experiment with both technique and find that the point etimate are imilar, but that the GMM etimate are much more imprecie. Blundell and Bond (1998) how that efficient GMM etimation in the cae of very peritent erie may be achieved by uing the ytem approach developed by the author. Thi approach i increaingly ued in the more recent literature. 2.4 Parameter heterogeneity Although etimation of R&D equation have been conducted at different level of aggregation, the majority of the exiting work ue firm level (panel) dataet. The model typically aume, and contrain, the R&D effect of the factor under conideration to be homogeneou acro the cro-ection dimenion of firm, indutrie or countrie. Under thi aumption, the etimated coefficient reflect average effect within the ample. While average effect reveal important information, they do not provide any information about potential cro-ectional difference of the R&D effect. Set againt thi poible lack of information i the diadvantage of a maller ample ize when etimating ub-ample of firm or indutrie. Sub-ample etimate may thu be le precie, and interpretation may need to be more cautiou. However, relatively low degree of freedom need not necearily imply a lack of preciion, and author who have plit their ample into uch ub-ample have found important difference in the R&D effect of government policy. Lach (2002), for intance, find that the effect of ubidie differ between mall and large firm. González and Pazó (2008) and Hall, Lotti and Mairee (2009) report different R&D effect for hightech veru low-tech firm, and Becker and Hall (2013) confirm the exitence of uch difference at the indutry level. Conidering full-ample a well a ub-ample etimate may thu bear ueful concluion for R&D policie. 2.5 Model uncertainty Difference between tudie may alo reult from the et of control variable included in the regreion. For intance, the precie etimated long-run elaticity of R&D with repect to 4 See, e.g., Hall, Griliche and Pake (1986); Lach and Schankerman (1989); Bound, Jaeger and Baker (1995); Blundell and Bond (1998); Blundell, Bond and Windmeijer (2000). 8

9 it uer cot, while being broadly around -1.0, varie depending on the model pecification. Many alternative empirical equation have equal theoretical tatu, however, o that difference in reult from model with different control variable need to be interpreted with care. 2.6 Non-linearitie Little attention ha o far been paid to potential non-linearitie in the relationhip between private R&D invetment and policy meaure. If non-linearitie are preent, then traditional linear model may be mipecified. Guellec and Van Pottelberghe de la Potterie (2003) find that the elaticity of private R&D with repect to a government ubidy ha an inverted U- hape for a multi-country OECD ample, which enable them to identify threhold level of ubidie at which the effect of the ubidy on private R&D change ign. Görg and Strobl (2007) provide imilar evidenempirical evidence i provided in Woodward, Figueiredo and Guiamare (2006). 3. PUBLIC POLICIES IN SUPPORT OF PRIVATE R&D INVESTMENT 3.1 R&D tax credit and direct ubidie The claical public finance olution to the problem that private R&D expenditure ha poitive externalitie and may therefore be lower than ocially optimal would be to ubidie the economic activity which create the poitive externality, i.e. private R&D invetment. Two policy tool available to government are R&D tax credit and direct ubidie of private R&D project. The former i a more market-oriented approach, leaving deciion on the level and timing of the invetment to the private ector. 5 5 Of coure, even when effective, any judgement a to the deirability of a tax credit would need to be baed on a cot-benefit analyi that included deadweight cot and the relabelling of activitie a R&D within corporate account (Hall and Van Reenen, 2000). For a detailed microeconomic evaluation of the effect of a tax credit, ee Klette, Møen and Griliche (2000). See alo Jaffe (2002). For an aement of the efficiency of public R&D upport at the macroeconomic level, ee Cincera, Czarnitzki and Thorwarth (2011). 9

10 Generally, when aeing the repone of private R&D pending to a change in it price, it i important to bear in mind the implication of what i known a the `relabelling' problem (Hall and Van Reenen, 2000): The true impact of a change in the tax credit on companie' R&D expenditure may be overetimated when uing reported R&D data, a in repone to an introduction of or an increae in the tax credit, firm have an incentive to maximie the hare of R&D qualifying for the credit. They may thu move expene within their account o a to enure correct claification, wherea before the preferential tax treatment indifference with repect to the labelling of R&D expene may have led to incorrect claification of at leat part of the R&D pending. Hall and Van Reenen (2000) review ome evidence in favour of thi hypothei. Overall Hall and Van Reenen (2000) conclude in their urvey of the pre-2000 literature that tax credit have a ignificant poitive effect on R&D expenditure, although there i coniderable variation in the finding of different tudie. The more recent literature more unanimouly find a poitive effect, whereby the precie etimated elaticitie vary depending on the data, etimation method and model pecification. In a panel data tudy on the manufacturing ector of nine OECD countrie for , Bloom, Griffith and Van Reenen (2002), for intance, etimate a long-run elaticity of R&D with repect to it uer cot of around Applying a imilar etimation approach, Li and Trainor (2009) obtain a long-run elaticity of around -1.4 for a panel of manufacturing plant in Northern Ireland for , and an elaticity of between -1.5 and -1.8 i reported in Parii and Sembenelli (2003) for a panel of Italian firm for Lokhin and Mohnen (2012) and Koga (2003), repectively, find omewhat lower elaticitie of -0.8 for firm in the Netherland during , and -0.7 for firm in Japan during Mulkay and Mairee (2013) report a long-run uer cot elaticity of -0.4 for a recent ample of French firm. For the US and the Canadian manufacturing ector, repectively, Berntein and Mamunea (2005) etimate R&D own price elaticitie of -0.8 and The author ugget that one reaon why the latter elaticity i o low i that much of Canadian R&D i performed by foreign firm which are not a uceptible to change in Canadian economic condition a are dometic firm. Baghana and Mohnen (2009) confirm the long-run elaticity of for firm in the Canadian province Québec. Uing a non-parametric matching approach, Czarnitzki, Hanel and Roa (2011) conclude that R&D tax credit alo have a poitive impact on Canadian firm' deciion whether to conduct any R&D at all. In a rare 10

11 tudy on a newly indutrialied economy, Yang, Huang and Hou (2012) confirm the evidence of a poitive R&D effect of tax credit for Taiwan. One policy concluion that can be drawn from all of thee tudie i that fical policy meaure that reduce the uer cot may be expected to increae private R&D expenditure. Overall, the average negative elaticity acro the variou tudie appear to be around unity. Regarding the effect of direct R&D ubidie, the urvey of earlier tudie conclude that the econometric evidence i ambivalent and that there are additionality effect of public R&D on private R&D a well a crowding-out effect (David, Hall and Toole, 2000; García- Quevedo, 2004). David, Hall and Toole (2000), for example, find that a third of the 33 tudie under review report ubtitution effect. However, the more recent reearch much more unanimouly reject crowding-out and tend to find additionality effect. One criticim of much of the earlier work ha been that it neglect the problem of ample election bia, in that R&D intenive firm may be more likely to apply for a ubidy (David, Hall and Toole, 2000). Since it i likely that thee firm would have undertaken at leat part of the R&D even in the abence of the ubidy, the reult may have been biaed toward finding crowding-out effect. The availability of new econometric technique that control for the election bia i thu likely one reaon for the hift away from finding crowding-out effect. In thi vein, applying a matching framework to ample of French and Italian firm, repectively, Duguet (2004) and Carboni (2011) reject crowding-out and find that public ubidie on average increae private R&D. Czarnitzki and Huinger (2004) confirm thee reult for the German buine ector. Aert and Schmidt (2008) provide imilar reult for firm in Flander and in Germany, uing a conditional difference-in-difference etimator with repeated cro-ection. Employing parametric and emiparametric two-tep election model, Huinger (2008) further find evidence of additionality effect of publicly funded R&D on private R&D invetment per employee in German manufacturing. Comparing the reult of even matching method, a election model and a difference-in-difference etimator for a dataet of Italian firm, Cerulli and Potì (2012) alo reject full crowding-out of private R&D on average. Conducting a treatment effect analyi for manufacturing firm in Turkey a a developing country, Özçelik and Taymaz (2008) further corroborate the evidence of additionality effect. Regarding recent panel data regreion analye, the reult of additionality effect from the treatment effect analyi in Özçelik and Taymaz (2008) hold alo for the variou regreion 11

12 analye conducted in the tudy. Klette and Møen (2012) compare the reult of two panel fixed effect tudie on imilar Norwegian firm data for the pre-2000 and pot-2000 time period and The author conclude that the fact that their tudy on the earlier period doe not find any ignificant degree of additionality, while the tudy by Henningen, Haegeland and Møen (2012) on the later period doe find additionality, ugget that the effectivene of thi policy tool ha improved over time. Uing a dynamic panel fixed effect intrumental variable etimator in an analyi of UK manufacturing indutrie for , Becker and Pain (2008) find a poitive effect of the hare of buine R&D funded by the government on the level of R&D. The tudy thu indicate that the decline in the hare of manufacturing R&D financed by the government between 1992 and 1997 play an important role in the explanation of the comparatively poor R&D performance of the UK een over the In a further dynamic panel data regreion analyi, Bloch and Graveren (2012) obtain additionality effect of public R&D funding for a ample of Danih firm. For the buine enterprie ector of 21 OECD countrie, Falk (2006) doe not find a ignificant effect in dynamic panel data model. Nonethele, the general concluion from the pot-2000 empirical evidence mut be that public R&D ubidie ucceed in ignificantly timulating private R&D invetment. There i growing evidence that public ubidie are particularly effective in increaing R&D of mall firm, which are likely to be more financially contrained. Small firm have, for intance, le collateral in term of exiting aet to be ued for obtaining loan, and a a group they are likely to include more young firm. 6 Related to thi, large firm' greater ability to ecure funding for riky project given capital market imperfection i one of the argument put forward in upport of the hypothei that R&D increae more than proportionately with firm ize, following Schumpeter (1939, 1942). One relevant tudy i Lach (2002) which ue a difference-in-difference etimator and find for a ample of firm in Irael that ubidie for the mall firm temporarily crowd out thee firm' R&D, but have a trong timulative effect after the firt year of the ubidy. The author argue that thi may reflect the fact that firm which receive the ubidy are committed to implement the ubidied project, but that thi commitment may have led firm to temporarily cale down non-ubidied project due to the eriou killed labour hortage that characteried the economic environment in Irael over the ample period Subidie for the large 6 For a urvey of the empirical evidence on financial contraint for R&D by mall veru large and young veru mature firm, ee Hall (2002) and Hall and Lerner (2010). 12

13 firm in the ample are tatitically inignificant. Mot ubidie are granted to the large firm, however, which may explain the reult that the average effect for the pooled ample, while poitive, alo i inignificant. Thee finding are interpreted a indicating that large firm get ubidie for project that would alo have been undertaken in the abence of the ubidy, wherea mall firm ue the ubidy to finance additional project. From a policy point of view, the ubidy fund hould therefore be redirected to the maller firm. Uing Finnih data, Hyytinen and Toivanen (2005) further how that when there are economically ignificant capital market imperfection, mall and medium-ize firm in indutrie that are more dependent on external finance invet relatively more in R&D when more public funding i (potentially) available. Overall thee reult ugget that, on the one hand, ubidie targeted at financially contrained firm may raie overall private R&D pending, and that, on the other hand, policie deigned to improve thee firm' acce to external finance might reduce the need for R&D ubidie. 7 Thee concluion are alo compatible with evidence found by Almu and Czarnitzki (2003) for a tranition economy, for which capital market imperfection may a priori alo be expected to be relatively more pronounced. Applying a non-parametric matching approach to pot-reunification cro-ection from the 1990 for Eat Germany, the tudy find additionality effect of all public R&D ubidie on average. Czarnitzki and Licht (2006) moreover find that the additionality effect on firm' R&D and innovation input wa more pronounced in Eatern Germany during the tranition period than in Wetern Germany. One concluion drawn by Czarnitzki (2006), however, i that while the ubidie were initially intended to accelerate the tranformation proce of Eat Germany from a planned to a market economy, the continuing high level of ubidiation may imply inefficiencie a market force are weakened. 8 González, Jaumandreu and Pazó (2005) model firm' deciion about performing R&D when ome government upport can be expected. Applying a emitructural framework to 7 There i ome firt evidence that award of a government ubidy may provide a poitive ignal about firm quality and thu help a firm attract additional private funding, hence eaing the advere effect of capital market imperfection. Meuleman and De Maeeneire (2012) provide compelling evidence that obtaining an R&D grant reult in better acce to long-term debt and to a leer extent hort-term debt for mall and medium-ize firm in Belgium. Feldman and Kelley (2006) find that R&D grant facilitate attracting venture capital for US firm that participate in the Advanced Technology Program. 8 One novelty of thi tudy i that it take into account non-r&d performing firm and the endogeneity of their deciion a to whether or not to invet in R&D. The tudy thereby explicitly conider the fact that a large hare of mall firm do not invet in R&D due to a lack of financial reource. 13

14 Spanih firm data, the author find that public ubidie timulate private R&D pending. However, there i only a very light increae for thoe firm that would perform R&D anyway, wherea ome mall firm would not perform any R&D in the abence of (expected) public funding. Similarly to Lach (2002) the author point out that ubidie are mainly awarded to firm that would have performed the R&D anyway, and that thi ugget that public policy tend to neglect the inducing dimenion of public funding. The importance of thi dimenion i alo underlined by the reult in Hall, Lotti and Mairee (2009) who find for mall and medium-ize Italian firm that non-r&d performing firm are more likely to tart inveting in R&D if they receive a ubidy. In a cro-country analyi, Czarnitzki and Lope Bento (2012) conclude that private R&D in Belgium, Germany, Luxembourg and Spain would benefit from an extenion of public R&D ubidie to currently non-ubidied firm. Applying a matching approach to the ame dataet a González, Jaumandreu and Pazó (2005), González and Pazó (2008) moreover find that, imilarly a for mall firm, R&D ubidie are more effective for firm operating in low-tech ector, which the author argue i probably alo due to the inducement effect. In a tudy of UK manufacturing indutrie, Becker and Hall (2013) find that a higher hare of government-funded R&D ha a poitive effect only for the low-tech indutry group while being inignificant for the high-tech indutry group. Thee reult alo ugget that high-tech firm ubtitute incremental public funding for internal funding. Comparing the private R&D effect of EU veru national grant, uing a ample of German firm, Czarnitzki and Lope Bento (2011) conclude that the former yield higher effect than the latter, if funding i received from only one of the two ource. Two reaon are uggeted: Firt, EU grant may on average ditribute larger ubidie, or, econd, their requirement might be uch that only thoe firm that are mot likely to top up the grant with private funding more ubtantially comply. The larget R&D effect are obtained through imultaneou funding from both ource, and Czarnitzki and Lope Bento (2013) confirm that imultaneou receipt of multiple grant doe not caue crowding-out. There i both early and recent evidence of a different time pattern of the effect of tax credit and direct ubidie. Tax credit have a ignificant effect on R&D expenditure mainly in the hort run, but only little in the long run, wherea ubidie have a poitive effect in the medium to long run, but le o in the hort run. Hence the effect of tax credit i quicker than that of direct ubidie (David, Hall and Toole, 2000; Guellec and Van Pottelberghe de la 14

15 Potterie, 2003). The earlier tudy ugget that thi time pattern at leat in part likely reflect the fact that the tax offet againt earning occur for R&D project choen by firm themelve, the incentive of which probably favour project that will generate greater private profit in the hort run. By contrat, ubidie apply to project elected by the government, which may be of a long-term nature and create new opportunitie that may induce firm to tart further project with internal funding at a later tage, a pointed out in the more recent tudy. The tudy further ugget that tax credit and direct ubidie are ubtitute in that an increae in one dampen the effect of the other on buine R&D. Thee reult indicate that the deign and implementation of the two policy tool may be more effective if performed in a coordinated way. There i ome firt evidence that the effect of a public ubidy on private R&D may have an inverted U-hape. Guellec and Van Pottelberghe de la Potterie (2003) obtain the tronget private R&D effect for medium average ubidiation rate of 4-11%, while rate above 20% are found to be aociated with the ubtitution of government fund for private fund. Görg and Strobl (2007) imilarly find an inverted U-curve effect for indigenou Irih manufacturing firm. Thee tudie thu indicate that large grant may more likely act to finance private R&D activity that would have been undertaken anyway. From a policy point of view, the non-linear effect ugget that for any given public R&D budget, it may be more effective to grant ome intermediate level of upport to a larger number of firm than to provide a large amount of upport to fewer firm. With repect to the potential importance of international tax difference, Bloom and Griffith (2001) conclude in a cro-country analyi of eight OECD countrie that R&D in one country repond to a change in the R&D tax credit in another country. Thi reult ugget that at leat part of the reaon for the international mobility of R&D may be related to the increaing tax ubidie to R&D offered in many countrie. One implication for R&D tax policy then i that the poitive R&D effect of tax credit may be higher than previouly etimated and increaing over time. Foreign tax competition may moreover become increaingly important a impediment to capital mobility come down. Concluding, economit have generally been ceptical regarding the efficacy of tax credit, one reaon being the view that R&D wa not very enitive to change in it price. The recent evidence ugget much more unanimouly than concluded in urvey of the earlier work that R&D tax credit have a poitive effect on private R&D invetment. Generally, the negative 15

16 demand elaticity of R&D with repect to it own tax price i etimated to be broadly around unity, at leat in countrie with a tax credit. The recent evidence predominantly alo ugget that public R&D ubidie ucceed in timulating private R&D. The additionality effect ha been hown to be particularly prevalent for mall firm, which are more likely to experience external financial contraint. Moreover, thee firm are more likely to tart inveting in R&D if they receive a ubidy. On the one hand, thee reult provide trong upport of uch government funding cheme. On the other hand, a number of tudie report that mot of the funding i awarded to larger firm that would have performed the R&D even in the abence of the public ubidy, which ugget that in thee cae ubidie could be targeted more effectively. Indeed, Czarnitzki and Eberberger (2010) report that government in general prefer to grant R&D ubidie to larger firm and that thi i a common general criticim of the ditribution of ubidie: Thi kind of ditribution may contribute to a higher concentration of R&D, the peritence of leaderhip in market and higher barrier to entry, and thu eventually reduce competition. It may alo be the cae that a tax credit rather than a ubidy could be the more effective policy intrument for firm that are likely to imply ubtitute incremental public funding for internal funding, a the tax credit upport the private R&D that i actually expended by the firm. There i ome evidence that both policy tool may be more effective if performed in a coordinated way and that tax credit are the more effective hort-run policy option, while direct ubidie are the more effective medium to long-run policy. There i alo ome indication that the effect of a ubidy may have an inverted U-hape, o that ubidy level that are too high crowd out private R&D, while intermediate level timulate private R&D. Thi could imply that it may be more effective to grant ome intermediate level of upport to a larger number of firm than to provide a larger amount of upport to fewer firm. To date there are, however, only very few tudie that invetigate the relative effect of both tool or allow for a potential non-linear effect. Table 1 ummarie the key feature of tudie that repreent the main reult from the literature. <Table 1 about here> 16

17 3.2 Support of the univerity reearch ytem and the formation of high-killed human capital A growing body of evidence indicate that private R&D benefit from geographically localied knowledge pillover from univerity reearch and from the availability of highkilled human capital reource. The notion that knowledge pillover be localied goe back at leat to Marhall' (1920) concept of the external economie. He illutrate thi with the example of indutry localiation and identifie three reaon for localiation, which can alo be found in mot of the more recent literature on regional economic and economic geography: A pooled market for worker with indutry-pecific kill, upport of the production of non-tradable pecialied input, and informational pillover between firm that give clutered firm a better production function than iolated firm. The theoretical foundation of the geography of innovation i provided in Krugman (1991a, b) who how how a country can endogenouly develop into an indutrialied `core' region and an agricultural `periphery' region. An early cae tudy which indicate that knowledge pillover from univerity reearch may be a driving factor of firm' choice of location wa provided by Dorfman (1983). Her reult indicate that high-technology firm ought to locate cloe to univeritie, pointing to the importance of the MIT for the development of Boton' `high technology' Route 128 and of Stanford Univerity for the location of `Silicon Valley'. Related to thi, Nelon (1986) argue in the firt formal indication of localied knowledge pillover from univeritie to firm, that univerity reearch rarely in itelf generate new technology, it rather enhance technological opportunitie and the productivity of private R&D. In accordance with thi, the much-cited tudy by Jaffe (1989) provide evidence of a large ignificant poitive effect of univerity reearch on indutry R&D pending within US tate and conclude that a tate that improve it univerity reearch ytem will increae local innovation by attracting indutrial R&D. In upport of Dorfman' (1983) early reult, Woodward, Figueiredo and Guimarae (2006) more recently find for the US that R&D expenditure at univeritie poitively affect the location deciion of new high-tech plant in a county. Thi poitive effect 17

18 extend up to a maximum ditance of approximately 145 mile between the univerity and the new plant. 9 More recently there ha been a growing number of tudie uing data for countrie other than the US. Applying a modified verion of Jaffe' (1989) R&D model to French data, Autant-Bernard (2001), for intance, find poitive externalitie from public reearch to private R&D expenditure, and that thee externalitie are tronget within the ame geographical area. Karlon and Anderon (2009) provide evidence that indutrial R&D in Sweden tend to increae in location that offer high acceibility to univerity R&D. Abramovky, Harrion and Simpon (2007) examine the relationhip between the co-location of the average number of private ector R&D firm and univerity reearch department in 111 potcode area in the UK. The author match data for R&D lab in ix product group with data from the Reearch Aement Exercie on the quality of univerity reearch. The tronget evidence for co-location i found for pharmaceutical buine R&D and the mot highly ranked chemitry univerity department. Overall the reult raie the poibility that private ector R&D may benefit from proximity to both, frontier baic univerity reearch and alo more applied univerity reearch. The author note that while the latter may be conidered a low-quality reearch in term of the Reearch Aement Exercie and conequent univerity funding allocation, it may be relevant in ome area of technology tranfer and in attracting foreign-owned R&D. In a related analyi, Abramovky and Simpon (2011) confirm that pharmaceutical firm locate their R&D facilitie cloe to frontier chemitry univerity department. In addition, the reult for the chemical and vehicle indutrie potentially indicate the preence of knowledge flow between R&D activity and production activity. Conditional on location, the evidence for thee latter two indutrie i again conitent with geographic proximity facilitating knowledge flow from univeritie. Roa and Mohnen (2008) meaure knowledge tranfer from univeritie to firm by the amount of R&D payment made by firm to univeritie. The author' empirical reult for Canadian data corroborate the mounting evidence that a decreae in ditance increae pillover. The exiting literature on the US and a variety of other countrie hence predominantly conclude that private R&D benefit from geographically localied knowledge pillover 9 After controlling for other determinant of high-tech tart-up, univerity R&D i found to have only a mall marginal effect on county location probabilitie. Thi reult might at leat in part be due to the high-technology boom of the 1990 ample period, which exhibited it own pecific tart-up dynamic. 18

19 from univerity reearch. Thi ha important implication for regional economic and development policie, and for the evaluation and funding of univerity reearch. One role thi literature acribe to regional R&D policy i to facilitate and upport the formation of regional cluter of univerity and private R&D activity in order to exploit agglomeration economie. Supporting univerity reearch i likely to enhance regional technological opportunitie and the productivity of private ector R&D. Improving the univerity reearch ytem and facilitating pillover to the private ector ha been hown to raie local private R&D pending. There i moreover ome evidence that proximity to univerity reearch matter epecially in high-tech ector, which indicate that at leat part of the pillover are ector-pecific and not jut the diffue effect of a large reearch univerity. Hence it could be effective for government upport of univerity reearch to target in particular thoe ector in which pillover are found to be larget. The tranmiion channel of thee knowledge pillover a identified in the literature include direct peronal interaction, univerity pin-off firm, conultancy, and univerity upply of a pool of highly-trained graduate for employment in indutry. 10 Thi lat channel ugget that R&D conducive government upport of the univerity ytem extend from the reearch ide to the education ide. Conitent with thi, there i growing evidence that confirm important poitive R&D effect of high-killed human capital reource. Thee include highly qualified cientit and engineer (Adam, Chiang, Starkey, 2001; Adam, Chiang, Jenen, 2003; Becker and Pain, 2008), and more generally the hare of the number of worker with higher education in the total number of worker (García and Mohnen, 2010), the hare of the population with tertiary education in the total working age population (Wang, 2010) and year of formal chooling (Kanwar and Evenon, 2003). Thi trand of the literature thu ugget a role for education policie and human capital invetment in increaing private R&D. Table 2 ummarie the key feature of tudie that repreent the main reult from the literature. <Table 2 about here> 10 Another channel i formal cooperation agreement, which are dicued in ection

20 3.3 Support of formal R&D cooperation There i a growing literature that ugget poitive private R&D effect of R&D cooperation between a variety of intitution. A firm' memberhip of a reearch joint venture ha the obviou advantage that it may enable the firm to overcome a cot-of-development barrier that may otherwie prevent R&D invetment, if R&D require a minimum threhold invetment to be effective at all. Another benefit i the reduction of wateful duplication of R&D. Thee benefit are et againt the potential advere outcome that participant will tend to free-ride on each other' R&D invetment in cae of ufficient poitive externalitie from each firm' R&D effort, or curtail competition in other tage of the firm' interaction, a emphaied in the theoretical contribution by Kamien, Mueller and Zang (1992). 11 Imperfect ability to aimilate the return from R&D and innovation increae the incentive to free-ride (Shapiro and Willig, 1990; Keteloot and Veugeler, 1995). In their pioneering work on cooperative and noncooperative R&D in duopoly, D'Apremont and Jacquemin (1988) how that in the preence of large pillover, R&D cooperation lead to higher R&D pending by duopolit compared to the competitive cae. A ymmetric reult i that for mall pillover, R&D cooperation reduce R&D pending by the firm. Regarding empirical teting and public policy, thee reult acribe a central role to the degree of R&D externalitie in the indutry. When attempting to ae the welfare effect of R&D cooperation, one challenge for reearch i to take into account the factor that affect the level of pillover through time. In an extenion of the model by D'Apremont and Jacquemin (1988), Kamien, Mueller and Zang (1992) how that if firm create a reearch joint venture and alo hare information in that they cooperate on their R&D expenditure to maximie combined profit, i.e. they are cartelied in the R&D tage, conumer plu producer urplu are maximied. Caiman and Veugeler (2002) point out that little i known today about the complementaritie between a firm' own R&D programme, cooperative agreement in R&D, and external technology acquiition, and that a better undertanding of thee iue may enhance firm' ability to appropriate potential pillover from R&D cooperation. R&D cooperation ha played an increaing role in firm' innovative activitie (Hagedoorn, 2002). Survey of the indutrial organiation and trategic management literature on partner motive and outcome of reearch joint venture, or more generally on the theory of R&D cooperation, are provided by 11 Dixit (1988) provide an analyi within the framework of international competition. 20

21 Hagedoorn and Narula (1996), Hagedoorn, Link and Vonorta (2000), Caloghirou, Ioannide and Vonorta (2003) and Sena (2004). In an empirical analyi, Irwin and Klenow (1996) conclude that among firm who participated in Sematech, a joint R&D conortium of US emiconductor producer that wa formed to develop new technologie for the production of computer chip, there wa a drop in the total level of R&D expenditure. They interpret thi a upporting the `haring' hypothei, with information flow reducing duplicative R&D, allowing member to pend le on R&D than before. Thi i contrated with the `commitment' hypothei of higher joint R&D expenditure on high-pillover type of R&D. Adam, Chiang and Jenen (2003) find that cooperation between federal laboratorie and firm ha a poitive impact on private R&D expenditure and that no other channel of technology tranfer from federal laboratorie exert a comparable effect. The author point out that arrangement that trive to enure effort by both, firm and federal laboratorie, are required for technology tranfer to be ucceful. A firt empirical inight into the potential effect of indutry-univerity cooperative reearch centre on indutry R&D expenditure i provided by Adam, Chiang and Starkey (2001) for the US. Thee centre are defined a " mall academic center to foter technology tranfer between univeritie and firm" (op. cit. p. 73). The reult ugget that the development of thee centre ha fotered knowledge pillover between univeritie and member firm. When the author differentiate between National Science Foundation cooperative reearch centre and other, the effect i ignificant only for the former. However, the author note that the coefficient may be biaed upward a the centre are matched to larger and more productive laboratorie. Two interpretion given by the author are, firt, that indutryuniverity cooperative reearch centre provide new project and timulate indutrial reearch, and, econd, that larger laboratorie are attracted to them. Hall, Link and Scott (2003) examine the performance of 54 indutry-univerity project funded by the US Advanced Technology Program which combine public fund with private invetment for the creation and application of generic technology needed to commercialie new technology rapidly. The tudy find that project with univerity involvement are more likely to be in new technological field where R&D i cloer to cience and that, therefore, uch project experience more difficulty and delay, but are more likely not to be aborted prematurely. The author' interpretation i that univeritie are contributing to baic reearch awarene and inight among partner in the funded project. With repect to geographic proximity, Pond, 21