Cristiano Antonelli

1. THE "MATTHEW EFFECT" IN R&D PUBLIC SUBSIDIES: THE CASE OF ITALY Cristiano Antonelli Dipartimento di Economia, Università di Torino and BRICK –Collegio Carlo Alberto Francesco Crespi Dipartimento di Economia, Università Roma Tre and BRICK – Collegio Carlo Alberto IX WORKSHOP SIEPI ROMA, 27-28 GENNAIO 2011

2. Background • Significant amounts of public funds are spent to support innovative activities performed by private firms (OECD, 2007). • The actual impact of R&D subsidies on firm’s innovative activities is not obvious. Crowding-out is possible (David and Hall, 2000; David et al., 2000; Hall and van Reenen, 2000). • ‘Government failures’ might be as large or even larger than the ‘market failure’ it is supposed to correct (Nelson, 1980)

3. Sources of ‘Government failures’ • Asymmetric information: difficulty of policymakers and program officials to know which firms to favour. (Grossman, 1991; Stiglitz and Wallstain, 2000) • Interest groups: politicians try to maximise votes and to allocate subsidies optimally from a political point of view, by responding to the requests of interest groups. (Peltzman, 1976; Olson, 1982; Mitchell and Munger, 1991; Magee, 1997). • Rent seeking: the possibility of receiving public support gives industries and other interest groups an incentive to invest large resources in unproductive rent-seeking activities such as lobbying (Tollison, 1997).

4. Tax credits: the solution? • Pros: • tax credits are neutral with respect to industry or sector and the characteristics of the firm; • minimize the discretionary decisions involved in project selection; • lower administrative costs. • Cons: • risk to provide support to an array of activities that often do not consist in research activities (See Italian R&D statistics); • The risks of opportunistic behaviour; • lobbying activities to obtain changes in the definitions of what counted as R&D as to broadening allowable costs (Alt et al., 2010). • Tax credits represent an agile way of providing public support to R&D and to reduce problems related to ‘government failure’. • R&D grants are potentially better suited to fill the gap between the private and social returns to innovation (David et al., 2000). Many countries still heavily rely on discretionary incentives even though this may come at a cost.

5. A further issue (not fully explored) on selective grants • The observed persistence in grant allocation to past recipients may represent a further source of ‘government failures’. • Paper objectives: • enquire about the causes and the effects of persistence in the provision of public subsidies. • qualify the persistence, whether it is actually and necessarily dysfunctional, or it may be fruitful from a dynamic efficiency viewpoint.

6. 1. THE "MATTHEW EFFECT" IN R&D PUBLIC SUBSIDIES • We apply to research policy the notion of Matthew effect drawn from the economics of science (Merton, 1968; Arora and Gambardella, 1997; Rigney, 2010). • As Merton noted: “…eminent scientists get disproportionately great credit for their contributions to science while unknown scientists tend to get disproportionately little credit for comparable contributions” (Merton, 1968:57). • Differences in the output of scientists may depend upon four classes of factors: a) the original talent and ability of researchers, b) their reputation based upon previous publications and related citations, c) the actual amount of their current efforts and d) the competence cumulated through time by means of learning processes.

7. THE "MATTHEW EFFECT" IN THE ECONOMICS OF SCIENCE • In the economics of science it is difficult to quantify the actual amount of efforts on the one hand and the actual quality of publications on the other. • The quality of publications in fact depends upon the number of citations, but the number of citations may depend upon sheer reputation effects. • Reputation at the same time may enable to increase the command of productive inputs, so that reputation and competence interact.

8. THE "MATTHEW EFFECT" IN THE ECONOMICS OF INNOVATION • The application of the Matthew effect analysis in the economics of innovation is likely to yield better results than in its original context: • In the economics of innovation quantitative information about the actual efforts of private firms in terms of the levels of R&D is available through time. • We can study the persistence in the distribution of public subsidies and the actual dynamic behaviour of the recipients, as measured by reliable indicators of efforts that have been invested. • In the economics of innovation we can better assess whether possible Matthew effects depend on sheer reputation effects.

9. 2. Virtuous and Vicious Matthew effects • We propose the distinction between virtuous and vicious Matthew effects. • The first consists in the persistence of the provision of subsidies to firms that have been actually able to use previous subsidies to effectively use larger amount of R&D, build higher levels of competence and expertise. • The latter include the cases of persistence in the assignment of public subsidies based on sheer reputation.

10. Vicious ‘Matthew Effect’ • A further source of ‘government failure’ • Members of the selection committees would be too much influenced by the scientific and technological reputation of the candidates, rather than by the quality of the projects. • Dysfunctional persistence effects in the probability of gaining access to public funding, even independently from their actual innovative efforts. • Valiant research programmes presented by unknown firms risk to be deprived of the deserved public support: • waste of resources; • misallocation of public money; • losses associated with the delay and the possible decay of relevant research programmes.

11. Virtuous ‘Matthew Effect’ • A virtuous persistence effect can be justified by the economics of knowledge (Arrow, 1962a, 1962b, 1969; David, 1994). • Committees members may be right in confirming their preferences for scientists and firms that have taken advantage of previous awards simply because their products embody a larger amount of inputs and higher levels of competence: • authors (firms) who have been selected in previous tournaments are the persistent recipients of allocations because they had the opportunity to enlarge their role as investigators in terms of increased access to scarce research resources and the opportunity to concentrate and specialize in conducting their research. • past recipients had the opportunity to learn to learn (Stiglitz, 1987). • Knowledge exhibits intrinsic cumulability. The larger the knowledge base under the command of each firm and the larger the chances to generate new technological knowledge (Weitzman, 1996 and 1998).

12. 3. The ‘picking-the-winner strategy’ • In the absence of asymmetric information and opportunistic behaviours best projects are selected. However, ‘government failures’ might be relevant! • A possible way to reduce ‘government failures’ in the allocation of subsidies and to increase the efficiency of public support to private companies is to follow a ‘picking-the-winner strategy’ (Shane, 2009; Cantner and Kösters, 2009). (Sort of second best strategy) • Evidence for a policy focus on the most promising and best-equipped firms has been recently found for example in the German case (Aschhoff, 2010; Hussinger, 2008; Cantner and Kösters, 2009)

13. The ‘picking-the-winner strategy’ • The general assessment of firms’ quality seems to be an easier task than the evaluation of the potential outcome of a specific project. • Regular R&D or patenting activities, high level of human capital might represent crucial, objective indicators that public agencies may consider in taking their decisions. • If these firms have been subsidized in the previous years, this overwhelming performance may also be an effect of the previous support.

14. The ‘picking-the-winner strategy’ • The adoption of objective criteria based on firm’s past performance, reduce the tendency of assuming totally arbitrary choices that might be affected by lobbying activities of interest groups. • In this context persistence in the access to R&D grants should be associated with a virtuous Matthew effect instead of a vicious one. The selected firms are currently able to perform more R&D and to support higher levels of talents of the scientific personnel at work within the firm. • Matthew effects would be consistent and would complement a strategy of ‘picking the winners’ in the provision of public subsidies to R&D. • This would replace pure arbitrary criteria that might be adopted by selection committees in the absence of such a constraining strategy, hence increasing the efficiency of public support to firms’ innovative activities.

15. Research Hypotheses • H1: Matthew effects are relevant. We expect that significant persistence is at work in the allocation of public subsidies. • H2: Matthew effects can be of two types. If a pure reputation vicious effect is at work we expect that in the allocation of R&D subsidies only the achievement of past grant is relevant in explaining the current access to public funds. On the contrary we expect that in the virtuous case such persistence can be explained by the accumulation of expertise, tacit and codified knowledge by firms that had access to larger resources because of the allocation of public subsidies in the past. In this context Matthew effects would be consistent with a ‘picking the winners’ strategy, with potential benefits in the effectiveness of the adopted policy instrument.

16. Empirical Strategy • Descriptive analysis based on TPMs • Econometric analysis on the determinants of the access to R&D subsidies • Evaluation impact based on Propensity Score Matching Estimator

17. Database description • Questionnaire surveys developed originally by Mediocredito Centrale (MCC, now Unicredit), regarding a representative sample of Italian manufacturing firms with no less than 11 employees. • We merged two waves (covering years from 1998 to 2003). • We restricted the sample to firms which invest in R&D activities and which have been observed in both the two waves of the survey. • We ended up with a balanced dataset of 752 manufacturing firms observed two times over a 6-year period.

18. Summary statistics

19. The analysis of TPMs • In the case of a 2-dimensional matrix there is evidence of persistence if the sum of the main diagonal terms is more than 1. • Strong persistence is identified if the sum of the main diagonal terms is more than 1 and all the main diagonal terms are larger than 1/n (in this case 0.5)

20. Transition probabilities between period T and T-1 along years 1998-2003. Full sample T T-1 The probability of obtaining R&D subsidies in period t for subsidized firms in period t-1 is more than the double than for non sub.firms in t-1 Still not a sufficient evidence of true state persistence!!

21. Transition probabilities between period T and T-1 along years 1998-2003 by size classes

22. Transition probabilities between period T and T-1 along years 1998-2003 by Pavitt Classes

23. Transition probabilities between period T and T-1 along years 1998-2003 by class of R&D personnel intensity

24. Probit model Probit model of the event (Y=1) of receiving a public R&D subsidy Pr(Yit = 1| Xit-1 , Yit-1) Where: Xi,t-1 vector of observable firm i’s characteristics at t-1 Yi,t-1 the event of being subsidized or not at time t-1

25. Control variables Control variables have been selected according to the relevant literature (Busom, 2000; Wallsten, 2000; Arvanitis et al., 2002; Almus and Czarnitzki, 2003; Duguet, 2004; Blanes and Busom, 2004; Görg, H. and E. Strobl, 2007; Hussinger, 2008)

26. Probit model, Dependent variable: Access to public R&D subsidies (t) (R&DSUB)

27. Econometric results • Evidence of persistence in the provision of public subsidies. • Result is robust to the introduction of a number of relevant control variables. This indicates state dependence suggesting that some mechanism related to Matthew effects is at work. • Larger and R&D intensive firms are perceived as more promising to be successful with their R&D projects. • We interpret this result as evidence that public authorities followed a “picking the winner” strategy. • The joint significance of R&D capabilities and past R&D grants can be interpreted as a first indication that a virtuous Matthew effect might have prevailed.

28. Propensity Score Matching Estimators • THE PROBLEM OF CAUSAL INFERENCE IN IMPACT EVALUATION: • impossible to observe the individual treatment effect; • Need to construct the counterfactual: the outcome participants would have experienced, on average, had they not participated.

29. Propensity Score Matching Estimators • Evaluating the causal effect of some treatment on some outcome Y experienced by units in the population of interest. • Y1i → the outcome of unit i if i were exposed to the treatment • Y0i → the outcome of unit i if i were not exposed to the treatment • Di ∈{0, 1} → indicator of the treatment actually received by unit i

30. Propensity Score Matching Estimators • assume that all relevant differences between the two groups are captured by their observables X; • select from the non-treated pool a control group in which the distribution of observed variables is as similar as possible to the distribution in the treated group; • Calculate Average Treatment Effect on the treated

31. Some good properties of the database(Heckman, Ichimura and Todd, 1998) • Information on both supported and not supported firms come from the same survey. • Large set of information on firms structure and behaviour in order to properly select the control sample. • Presence of a large number of non-treated companies in the sample. • The use of lagged variables as controls allows to avoid problems of endogeneity in the selection equation.

32. Impact Evaluation Analysis Results based on selection Model 4 Additionality but not complementarity

33. Conclusions • Both the descriptive and econometric evidences suggest the presence of persistence in R&D subsidies. • Matthew effects are at work: after controlling for a large number of factors the lagged variable related to the access of R&D subsidies is still significant. • The persistent character of R&D subsidies is associated with larger amount of total R&D activities. • Our results reject the claim that discretionary procedures of allocation engender automatically perverse effects of persistence and exclusion. • Empirical results provide evidence on the working of a positive persistence, i.e. virtuous Matthew effects in the Italian experience. • The magnitude of costs associated to discretionary procedures can be limited by adopting a ‘picking the winner strategy’ which is coherent with the case of virtuous Matthew effect.