Effects of targeted R&D support: European evidence

Transcription

1 SIMPATIC working paper no. 28 December 2014 Effects of targeted R&D support: European evidence Dirk Czarnitzki, Elena Huergo, Mila Köhler, Pierre Mohnen, Sebastian Pacher, Tuomas Takalo and Otto Toivanen The SIMPATIC project is coordinated by Bruegel (Belgium) and involves the following partner organisations: KU Leuven (Belgium), UNU-Merit (Netherlands), SEURECO (France), E3MLab (Greece), Univesidad Complutense de Madrid (Spain), Federal Planning Bureau (Belgium), Imperial College (United Kingdom), Institut za ekonomska raziskovanja (Slovenia). Project website: LEGAL NOTICE: The research leading to these results has received funding from the Socio-economic Sciences and Humanities Programme of the European Union's Seventh Framework Programme (FP7/ ) under grant agreement no The views expressed in this publication are the sole responsibility of the authors and do not necessarily reflect the views of the European Commission.

2 Effects of targeted R&D support: European evidence * Dirk Czarnitzki Elena Huergo Mila Köhler Pierre Mohnen Sebastian Pacher Tuomas Takalo Otto Toivanen* December 28 th 2014 Abstract This paper provides a project and firm-level analysis of the effects R&D subsidies have on private R&D investment. We use data from Finland, Flanders and Germany. We find that the effects of subsidies at the project level are not strong: We find complete crowding out or worse for all data sets. At the firm level, the results are very different: We find strong crowding in / additionality in Belgium and Finland and Flanders, and rowding out for Germany. At least for Finland and Flanders, the results should be interpreted with caution. JEL: D04, G38, H25, L59, O31, O38. Keywords: R&D subsidies, applications, SME, subsidy rate. * This paper is an updated version of a paper by the same title, dated July 31 st Czarniztki: KU Leuven, ZEW and ECOOM, dirk.czarnitzki@kuleuven.be; Huergo: GRIPICO-Universidad Complutense de Madrid, ehuergo@ccee.ucm.es; Koehler: KU Leuven and ZEW, mila.koehler@kuleuven.be; Mohnen: Maastricht University, UNU-MERIT, p.mohnen@maastrichtuniversity.nl Pacher: UNU-Merit, s.pacher@maastrichtuniversity.nl; Takalo: Bank of Finland and KU Leuven, tuomas.takalo@gmail.com; and Toivanen: KU Leuven and CEPR, otto.toivanen@econ.kuleuven.be. We thank the various government agencies for access to the application data. We also wish to thank the EU Seventh Framework Programme for SIMPATIC and TEKES for financial support Any remaining errors are ours. 1

3 1 Introduction Economic research, at least since Nelson (1959) and Arrow (1962), has for a long time suggested that public support to private R&D may be socially beneficial because of the appropriability problems linked to knowledge and its creation. Politicians throughout the world have responded and various forms of public support to private R&D are proliferating as Figure 1 (from OECD 2011) shows. Despite the widespread use of such policies and a large amount of research studying their effectiveness (see recent surveys by Zúñica-Vicente et al. 2012, Cerulli 2010) in affecting private R&D investment, no clear consensus has emerged as to how useful these policies are. One reason for this state of affairs is that results across studies are difficult to compare because of differences in both data and in methods. The objective of this paper is to provide comparable results by using data collected in different countries but using the same approach, and by studying these data using the same methods. 2

4 For this paper, we use data from Finland, Flanders (Belgium) and Germany. Our data on R&D investments is mostly at the level of individual R&D projects, though for a subset of countries we have R&D investment data also at the firm level. The period of observation varies across countries, being longest for Germany ( ) and Finland ( ). We observe the targeted support also at the project level. For Finland, we observe not only project level investment and support, but also grades by the support granting agency, given to an application. The plan of the paper is as follows: In the next section we briefly describe the support schemes to which we have access to in each country. The third section is devoted to explaining our data collection, and to a descriptive analysis of the data. In section four we provide our econometric analysis. In our project level estimations we rely on the methods developed by Takalo, Tanayama and Toivanen (2013a,b) though we do not 3

5 interpret our results as tightly in a structural way. The fifth section provides conclusions. 2 Subsidy schemes Subsidy schemes are uniform neither across countries nor across time. Simplifying, one can categorize R&D subsidies into targeted (e.g., subsidies) and untargeted (e.g. R&D tax incentives) and/or national and local. Targeted aid can either be available to all firms, or e.g. specific industries can be chosen / emphasized. Finland for example (see Takalo and Toivanen 2013) had mostly targeted funding, but within that form of aid, the share of funding channeled to specific industries and/or technologies has increased over time. Germany is similar to Finland in this respect (see Beyer, Czarnitzki and Toivanen 2013). This heterogeneity in the institutional setting, both across countries and across time, naturally partly explains the observed heterogeneity in both how firms behave and in how the agencies make decisions. Despite this heterogeneity, some common trends emerge: the maximum support percentages are high (up to 80%); all countries engage (mostly internal) experts in the evaluation of the applications; these evaluations play an important role in the decisionmaking process; there is wide heterogeneity in the actual subsidy rates. There are some notable differences, in the policy tools, summarized in Table 1. 4

6 Table 1 Description of the main support policies Variable Belgium/Flanders Finland Germany The Netherlands Spain Subsidies YES YES YES YES YES max (max SME) 0.7 (0.8) 0.6 (0.7) thematic/generic NO/YES YES/YES YES/YES YES/NO YES/YES basic/applied YES/YES YES/YES YES/YES NO/YES YES/YES soft loans NO YES NO NO YES interest rate tax credits YES NO NO YES YES only central gov. YES YES YES YES NO NOTES: in the case of Flanders, "central gov." refers to the Flemish (regional) government. In the case of "thematic/generic"and basic/applied, the X in the entry X/Y refers to whether there are thematic grants and whether basic research is supported; the Y to whether there are generic (unsolicited) grants and whether applied research is supported. The policies refer to those in place during our observation period. We next briefly describe the R&D subsidy schemes in the different countries, but refer the reader interested in more detail to the aforementioned discussion papers. Finland (see Takalo and Toivanen 2013) Finland has a single government agency, Tekes, that distributes R&D subsidies. Tekes mission is to promote the development of industry and services by means of technology and innovations. This helps to renew industries, increase value added and productivity, improve the quality of working life, as well as boost exports and generate employment and wellbeing. (Tekes 2011 strategy). Tekes grants both subsidies and (soft) loans. Tekes states that funding is based on the novelty of the project, market distance, and the size of the company. 1 An integral part of the decision-making process is the grading of applications by technical and economic experts of Tekes. The maximum subsidy varies between 50 and 70% in our data depending on the year, type of firm, and type of project. Flanders (see Czarnitzki and Toivanen 2013) 1 Tekes funding services for small and medium-sized enterprises slides (rahoitusesite_engl.pdf, available from 5

7 The agency for Innovation by Science and Technology in Flanders ( Innovatie door Wetenschap en Technologie in Vlaanderen (IWT)) is a governmental agency, established by the Flemish Government in It was established to give shape to the new competences in science and technology that were transferred from the federal to the regional governments in Belgium. Since this transfer of competences made innovation policies a regional matter, the IWT has been created as the key organization for support and promotion of R&D and innovation in Flanders. IWT grants subsidies, with the maximum subsidy being 70%. In order to encourage smaller firms to perform R&D, a special program for SMEs has been put in place (the KMO programma ). Germany In Germany R&D subsidies for project funding are the main policy tool to support industrial R&D (BMBF 2012). R&D subsidies to the industry are distributed either via thematic or generic R&D programs (BMBF 2012). With the thematic R&D programs the Federal Government aims to increase Germany s competitiveness by targeting R&D subsidies to R&D projects in technological areas that the Federal Government views as particularly important for the future technological competitiveness of Germany. The generic R&D programs are not directed to specific technological areas and allow to support e.g. R&D infrastructure investments, research cooperation, and innovation networks. The generic R&D programs are thus more intended to support SMEs. Bank loans, equity investments, and guarantees also exist but play a minor role in Germany. The Federal Ministries mainly the Federal Ministry of Education and Research (BMBF) and the Federal Ministry for Economics and Technology (BMWi) design the R&D programs, i.e. they decide about the program content, duration of the support, subsidy requirements, and the maximum subsidy amounts distributed to an R&D project. Usu- 6

8 ally the Federal Ministries pass on the administrative and scientific-technical work to the currently 14 different Program Administrating Agencies that have the scientific and technical expertise to administer and guide the R&D program (BMBF 2013). The Program Administrating Agencies are commissioned to advise applicants, prepare funding decisions, process the projects and monitor their success (BMBF 2010, p.24). Regarding the amount of subsidies provided to industrial research, Germany follows the Community frameworks for State Aid for R&D (EC 1996, EC 2006). Between the maximum subsidy rate was 75% and 80% from 2006 on. 3 Data 3.1 Data collection Our data was, due to confidentiality requirements, collected and analyzed country (region) by country. The common features are: collection of project level data from the support granting agency; linking of these data with R&D survey and other registry based firm-level data which also includes information on non-applicants (and recipients). The project level data include information on the planned R&D investment; the R&D investment deemed eligible ( accepted R&D ) for support by the granting agency; (for only some of the countries and years) the actual R&D investment for those projects that get support; the maximum and actual subsidy rate (including thus information on an application being rejected); and for some countries, information on the grading process of the agency. 3.2 Descriptive statistics Looking first at subsidy and application data, Table 2 reveals that German firms projects are on average an order of magnitude larger than those of Finnish and Flemish firms, but there is clearly more variation in project size in Flanders than in either Fin- 7

9 land or Germany. For Finnish firms we know the size of the project at the application stage ( Applied amount ), the part of the planned investment deemed eligible by the agency ( Accepted amount ) and the actual amount invested. Table 2 Descriptive statistics Descriptives statistics of the support policy Variable Statistic Belgium/Flanders Finland Germany The Netherlands Spain Max subsidy rate Subsidy rate avg sd Accepted amount avg sd Application prob. avg sd Observation period 2004, 2006, 2008, NOTES: amounts are in thousands of year 2005 euro. We deflate by country-specific consumer price indices obtained from Eurostat. The mean subsidy rate varies between 0.30 in Finland and 0.48 in Germany. The differences in the probabilities of applying for a subsidy surprisingly large, with one in four firms in the Finnish sample applying for a subsidy, whereas the respective probability in Flanders is only Here one has to recall that Takalo, Tanayama and Toivanen (2013) report an application probability of 0.08 for Finland; therefore, the high Finnish probability may be explained by the fact that we use the R&D survey sample. In Germany the application probability is with low, as rejected applications are not observed. Moving then to a comparison of firm characteristics, and comparing first the nonapplicants, we find that Finnish firms are younger than German and Flemish firms while German firms are biggest (using the mean). There are many more non-applicant firms in regions receiving EU regional aid in Germany than in Finland where there are none, and of course in Flanders as Flanders has no such region(s). 8

10 Table 3 Descriptive statistics of firms Non-applicants Variable Statistic Belgium/Flanders Finland Germany The Netherlands Spain age avg sd emp avg sd salesemp avg sd sme avg sd region avg sd nobs Applicants Variable Statistic Belgium/Flanders Finland Germany The Netherlands Spain age avg sd emp avg sd salesemp avg sd sme avg sd region avg sd nobs NOTES: Sales are in Millions of 2005 euros, deflated by country-specific consumer price indices from Eurostat. Region is a Comparing the applicants to non-applicants (in the German case, the recipients, too, as there are no rejected applicants), we find them to be younger than non-applicants in Finland and Germany, but older in Flanders; Finnish (non-)applicants are younger than the Flemish and German ones. Finnish applicant firms are larger than non-applicants while the reverse is true for Flanders and Germany. The mean applicant is roughly of equal size Finland and Flanders, but a lot larger in Germany. Again, many more German applicants are from regions receiving EU regional aid than in Finland. Finally, the proportion of SMEs is higher among non-applicants than applicants in all countries. 4 Econometric analysis Our econometric approach is to estimate the R&D investment decision of the firm at two levels: First, at the level on an individual R&D project, utilizing the data obtained from the support-granting agencies. As discussed by Takalo, Tanayama and Toivanen, (2013a, henceforth TTT), these equations are subject to a selection problem in that 9

11 those projects for which firms apply support, and to which an agency grants support, may by systematically different from other projects. TTT also show that under the appropriate assumptions, the agency decision i.e., the subsidy rate is not correlated with the project level profitability shock. We maintain these assumptions and therefore estimate a selection model of the following form: (1) ln R * i ( si ) = X i β + δ ln(1 si ) + εi * Where the project level R&D investment R ( ) is observed if a firm 1) applies and 2) is granted a subsidy: (2) d 1{ + ν 0} i = Y i i θ. i s i In equation (1), X i are firm (possibly project) characteristics, s i the project level subsidy rate, and ε i, β is a vector of parameters, and δ is the key parameter of interest, characterizing how the subsidy rate affects project level investment. Using the definition of additionality from and applying suitably the derivations in Takalo, Tanayama and Toivanen (2013b), the following holds: A value for δ of less than minus one suggests crowding in or additionality. A value of minus unity would suggest that project level R&D is expanded by exactly as much as the firm receives subsidies (i.e., no crowding out, no crowding in). Any value between minus unity and zero would suggest partial crowding out. If δ takes the value zero, we have complete crowding out. Finally, for completeness, any value above zero would imply that R&D investments actually decrease as the public support increases. In equation (2), Y i are firm characteristics (including the SME dummy and possibly its interactions with other firm characteristics, excluded from the 2 nd stage), θ is a vector of parameters and ν i is the shock affecting the application and granting decision. 10

12 TTT show that the shock to the firm s decision to apply for a subsidy may be correlated with R&D investment; this is the basis for estimating a selection equation. The SME dummy is excluded from the 2 nd stage as it is an administrative definition, not directly related to R&D investment. At the same time, it should affect a firms decision to apply for a subsidy as the maximum subsidy rate is usually higher for SMEs than non-smes. The firm level R&D investment is of the following form: (3) ln R * i ( si ) = X i β + δf ( si ) + εi, where, with some abuse of notation, we utilize the same notation as with the project level investment. We now use the function f s ) to measure the effect of public support on firm level R&D. The reason for this is that there is no equivalent firm level expression as that in (1) which would be derived from a theoretical model. 2 We will use three different forms of the function to measure the impact of subsidies: First, following much of the literature, a dummy variable for a firm obtaining a subsidy in a given year; second, the natural log of the implied subsidy, measured as the subsidy rate times the accepted project level investment, calculated over all applications of a given firm in a given year; and finally, the average subsidy rate, calculated over all the projects granted a subsidy in a given year. All of these imply some measurement error. Also, there are reasons to think that the firm level R&D shock is correlated with the subsidy decision e.g. through the application behavior of the firm, and therefore on needs to instrument f s ). For consistency with the project level equations, we use ( i the SME dummy (and possibly interactions with other firm level characteristics; for ( i 2 Unless one interprets the TTT model as a firm level model see also Gonzales, Jaumandreau and Pazo (2005) in which case the same functional form as in (1) would apply. 11

13 Flanders, we end up splitting the SME dummy into separate small firm and mediumsized firm dummies) as an instrument. There are at least two reasons why the effects of public support could differ at the firm and at the project level: First, there is the issue of aggregation. For example, the TTT project level model does not yield a firm level R&D equation of the form in (3). Second, there could be within-firm spillovers between projects. In other words, obtaining subsidized funding for one project may lead to the optimal project level R&D for some of the firm s other R&D projects to increase. Reasons for this include complementarities between projects, and the possible signal value of obtaining support for one (or more) projects. 4.1 Project level R&D investment We report the results using (the natural log of) project level R&D investment in Table 4. Recall that the subsidy variable takes the form ln(1- subsidy rate) and thus an it should obtain a negative coefficient in the case subsidies do not cause complete crowding out. 12

14 Table 4 Project level R&D investment Belgium/Flanders Finland Germany The Netherlands Spain (1) (2) (3) (4) (5) lnsubs_rate 1.681*** 1.653*** *** lnage ** * *** lnage * ** 0.061** salesemp salesemp lnemp ** lnemp * * region constant 8.634** *** *** *** *** mills lamda * rho sigma nobs censored obs uncensored obs Notes: Reported numbers are coefficient and standard error. ***, **, and * denote significance at 1%, 5%, and 10% level. For Belgium, Finland and Spain, the coefficients of the subsidy term suggest not only complete crowding out, but actually a negative effect of subsidies on project level R&D. This is most likely an anomaly: A potential explanation is that high subsidy rates are granted to smaller projects (of smaller firms) and our sample selection equation and control variables do not capture this to the needed extent. These results should be treated with caution and warrant further study For Germany and the Netherlands, we cannot reject the Null hypothesis of complete crowding out: the coefficient of ln(1- subsidy rate) is positive but not significantly different from zero. As found for example by TTT, when using project level R&D investments, many firm level controls obtain insignificant coefficients. This is the case with our estimations, too. 13

15 Finally a couple of comments on the selection into getting a subsidy. In all countries but the Netherlands we find a strong correlation between the first stage (applying for and getting a subsidy) and the second stage (R&D investment) shocks. The correlations are highly statistically significant, very close to unity for Belgium, Germany and Spain, but the sign varies. In Belgium and Spain, there is positive correlation, in Germany, negative correlation (as TTT found for Finland for the period 1/2000 6/2002). A positive correlation implies that those firms who are about to invest a lot in R&D those that have good ideas from a profitability point of view are more likely to apply. This means that the firms who apply would have invested more than the average firm even without a subsidy. A negative correlation naturally implies the opposite: It is firms with below average ideas from a profitability point of view that apply. Hence the R&D budgets of the applicant firms are lower than the budgets of those firms that do not apply. It is important to note that while the level of R&D is an important determinant of the level of spillovers and hence it is important to understand selection into applying for (and getting subsidies), the sign of spillovers is not necessarily related to this. Firms decide their R&D levels based on profitability considerations. The level of spillovers from a given R&D project may or may not be related to this. 4.2 Firm level R&D investment The firm level R&D investment results, estimated using 2sls, reveal a very different picture than the program level results. For Finland and Flanders we find evidence of crowding in / additionality, with the (weak) exception of the subsidy dummy results for Flanders. For Germany, results are more mixed, suggesting either partial crowding out. 14

16 Table 5 Firm level R&D investment Subsidy variable Statistic Belgium/Flanders Finland Germany The Netherlands Spain 1(subsidy) coeff ** * ** s.e st stage F-stat instruments sme sme sme sme sme ln(subsidy, euros) coeff ** * ** s.e st stage F-stat instruments sme sme sme sme sme ln(1-subs_rate_avg) coeff *** *** * 4.591* ** s.e st stage F-stat instruments sme sme sme sme sme ln(1-subs_rate_avg) coeff *** *** 3.877* *** s.e st stage F-stat instruments sme, sme*lnage sme, sme*lnage sme, sme*lnage sme, sme*lnage sme, sme*lnage nobs Notes: ***, **, and * denote significance at 1%, 5%, and 10% level. 1st stage F-stat is the robust F-statistic on the instrument(s) in the first stage. Compared to the project level results, the results using firm level data are more mixed. Looking at the first row of Table 5 which uses as the key explanatory variable a dummy for obtaining R&D support (a modeling choice found often in the literature), we find strong additionality for Belgium and Spain, negative but not significant (or only marginally significant) coefficients for Finland and Germany, and a large positive but insignificant coefficient for the Netherlands. Notice that the 1 st stage F-tests reveal that our instrument is very weak for both Finland and the Netherlands, rendering their results meaningless. For Belgium, the instrument can be considered weak, but for Germany and Spain strong. In the second part of the table the key explanatory variable is the natural log of the subsidy. Again our instrument is too weak to produce meaningful results for Finland and Spain. For Belgium and Spain we find a positive coefficient which has a value less than one, suggesting partial crowding out (in the case of Spain, we cannot reject the hypothesis that it is one). In contrast, for Germany, we find a puzzling negative coefficient. 15

17 In the 3 rd part of the table we again change the dependent variable, now to reflect the subsidy rate. Our instrument is now much stronger. For Belgium and Spain we find the expected negative coefficient (notice, the explanatory variable is the natural log of one minus the subsidy rate), suggesting (strong) additionality. For the other three countries we find a puzzling positive coefficient, suggesting more than complete crowding out. In the last part of the table we change the instrument slightly by introducing the interaction between the SME status of a firm and (log) firm age. As can be seen, the Belgian and Dutch results are stable (though the latter suffers from a weak instrument problem), while for the other countries, the point estimate changes, and considerably so in the case of Finland. This suggests that there is strong heterogeneity in the treatment effect of subsidies. 5 Conclusions In this paper, we provide consistent and comparable European evidence on the effects that public support has on private R&D. We use project and firm level information on R&D investments with data on project level R&D support from Finland, Flanders, and Germany. While the estimation period is country-specific, it covers the much of the 2000s for all countries. We find some puzzling results which call for further investigation. For Belgium, Finland and Spain our project level results on the effects of subsidies are anomalous, suggesting that firms decrease investments as a response to receiving a subsidy. For Germany and the Netherlands, we find complete crowding out which also is somewhat puzzling as earlier research has found crowding in for Germany. The earlier empirical literature for Belgium and Germany, summarized in Lopes Bento and Czarnitzki (2012), has found additionality or partial crowding out. Our firm level results strongly suggest large het- 16

18 erogeneity in how firms respond to R&D support, and thereby call for researchers to employ methods that not only take this heterogeneity into account, but also allow the researcher to measure it. 17

19 References Arrow, Kenneth J., Economic Welfare and the Allocation of Resources for Invention (pp ), in Richard R. Nelson (Ed.), The Rate and Direction of Inventive Activity (Princeton: Princeton University Press, 1962). Beyer, M., D. Czarnitzki and O. Toivanen, 2013, Applying for R&D Subsidies: Evidence from Germany. Cerulli, G., 2010, Modelling and measuring the effect of public subsidies on business R&D: a critical review of the economic literature. Economic Record 86, pp Czarnitzki, D. and O. Toivanen, 2013 Applying for R&D Subsidies: Evidence from Flanders. Einiö, E., 2013, R&D Subsidies and Company Performance: Evidence from Geographic Variation in Government Funding Based on the ERDF Population-Density Rule, Review of Economics and Statistics, forthcoming. Huergo, E., 2013 Applying for R&D Subsidies: Evidence from Spain. Czarnitzki, D., and C. Lopes Bento, 2012, Value for Money? New Microeconomic Evidence on Public R&D Grants in Flanders, CEPS/INSTEAD working paper González, Xulia, Jordi Jaumandreu, and Consuelo Pazó, Barriers to Innovation and Subsidy Effectiveness RAND Journal of Economics 36 (2005), Mohnen, P. and B. Lokshin, 2010, What does it take for an R&D tax incentive policy to be effective? In Ghosal, V. (ed), Reforming Rules and Regulations, pp Cambridge, MA: MIT Press. Nelson, Richard R., The Simple Economics of Basic Scientific Research, Journal of Political Economy 67 (June, 1959), Pacher, S. and P. Mohnen 2013, Applying for R&D Subsidies: Evidence from The Netherlands. Takalo, T., T. Tanayama, and O. Toivanen, 2013a, Estimating the benefits of targeted R&D subsidies. Review of Economics and Statistics 95, 1, Takalo, T., T. Tanayama, and O. Toivanen, 2013b, Market failures and the additionality effects of public support to private R&D: Theory and empirical implications, International Journal of Industrial Organization, forthcoming. Takalo, T. and O. Toivanen, 2013, Applying for R&D Subsidies: Evidence from Finland Zúñica-Vicente, J., C. Alonso-Borrego, F. Forcadell, and J. Galan, 2012, Assessing the effect of public subsidies on firm R&D investment: a survey. Forthcoming in Journal of Economic Surveys 18