A Methodology for Disentangling Public Funded R&D ( GBARD ) by Industries

1. A Methodology for Disentangling Public Funded R&D (GBARD) by Industries Matilde Mas (University of Valencia and Ivie) Eva Benages (Ivie) Juan Fernández de Guevara (University of Valencia and Ivie) Laura Hernández (Ivie) NTTS Conference, Brussels. 2017

2. Introduction • Introduction • Public transfers in R&D • Disaggregation of pubic funded R&D (GBARD) by industries • From GBARD to ICT GBARD by industries • Results • Main issues

3. Introduction • SPINTAN Project aims (EU’s Seventh Framework Programme, grant agreement no: 612774), among other goals, to build a database that comprises investment and capital stock of intangibles in the public sector. • The project complements previous projects (INTAN-Invest, COINVEST and INNODRIVE) which covered the market sector (8 industries): SPINTAN covers the non-market sector. • SPINTAN conceptual framework (Corrado, Haskel and Jona-Lasinio, 2015) clarifies: • Definition of the non-market sector, assets boundaries, return of non-market capital, functions of the government related to SPITAN… • Which government expenditures have to be capitalized?

4. Introduction • Which part of the government expenditures have to be capitalized? • This question is particularly relevant given the fact that public sector transfers (current and capital expenditures) are not included as government GFCF. • They are habitually considered expenditures of the industries that receive the transfer. • Need for measurement of the public transfer for the acquisition of intangible assets, particularly in the case of R&D.

5. Introduction • The relevance of knowing the contribution of the public sector to R&D goes beyond of the SPINTAN project. • Target 2 of Europe 2020 strategy (EC) is aimed to reach 3% of the EU's GDP invested in R&D. • Action 55 of the Digital Agenda for Europe (DAE,first flagship initiative of E2020) aims for Member States to double their annual public spending on ICT R&D. • However, monitoring trends in the public funding of ICT R&D is not straightforward given the lack of readily data available. • PREDICT Project (Prospective Insights in ICT R&D, CNECT & JRC B6) monitors the ICT sector for the European Semester (European Digital Progress Report) and for the Digital Single Market.

6. Introduction • One of the indicators monitored in PREDICT is the public funding of the ICT R&D. • In 2012 JRC unit B6 developed a methodology (Stančik, 2012) for disentangling the part of the public funding of R&D was devoted to ICT assets. • This methodology only offered data by NABS chapters. • Ivie began to produce the data of the ICT GBARD included in the PREDIC in 2013 . • In the SPINTAN Project Ivie have developed a new methodology (SPINTAN Working Paper no 23)based on Stančik (2012) for • calculating GBARD by NACE industries; • The part being devoted to ICT assets in each NACE industry. • The methodology developed in SPINTAN is being further developed jointly by JRC B6 and Ivie for PREDICT 2017 database.

7. Introduction • The aim of this communication is: • To present themethodology for estimating GBARD by industry, • High level of industry disaggregation NACE Rev. 2: 37 industries, of which 7 ICT industries (following OECD, 2007, ICT sector definition) • For the 2006-2015 period • For Member States and EU28 • To disaggregateGBARD by industries into the ICT and non-ICT components. • To analyze the results of applying the proposed methodology • GBAORD by industry and country • ICT GBAORD by industry and country • The final methodology presented here is the basic method developed in SPITAN project and the developments implemented by Ivie and JRC B6.

8. Public transfers in R&D • How to measure public transfers in R&D, i.e. the resources the government endow to R&D? • FrascatiManual (OECD, 2015): • Performance-based approach: Gross expenditure on R&D (GERD): aggregation of the amount each industry receive for their intramural R&D. • Funder-based approach. GBARD (Government budget allocations for R&D): aggregation of the sums committed by government to fund other industries R&D. • Useful to increase the timeliness of R&D statistics. • Released by socioeconomic objectives (NABS chapters). • None of the two approaches suffice either SPINTAN or PREDICT objectives • Our aim is to know which part of public funding is devoted to each industry and the part devoted to ICT assets.

9. Public transfers in R&D • Available GBARDdata from Eurostat: • EU28 and Member States • 2004-2015for most Member States and 2006-2014 for the EU aggregates • Classified by socio-economic objective (SEO), using the NABS 2007 classification: • * Nomenclature for the Analysis and Comparison of Scientific Programmes and Budgets.

10. Public transfers in R&D • Estimation of missing NABS chapters • In some countries Eurostat data include missing NABS chapters, although these chapters are included in the total (NABS 99). Additionally, the aggregation of NABS chapters does not always coincide with the total GBARD (NABS 99). • Given the fact that the methodology developed needs to disentangle each NABS’ GBARD into its correspondent NACE industry, we need to impute the missing NABS chapters. • Therefore: • Missing NABS chapters have been estimated. • We have produced an additional adjustment for countries with differences between their NABS99 (total GBARD) and the sum of NABS chapters. The structure of GBAORD by NABS chapters has been applied to the total (NABS99). • We have adjusted all the countries-years estimated data in 1) and 2) so that • Country aggregations coincide with Eurostat’s total GBARD by country and • 2) the aggregation of NABS by countries coincide with EU GBARD by NABS. • This adjustment uses the RAS method.

11. Public transfers in R&D • Eurostat GBARD data by NABS show some missing NABS chapters • Official GBARD data for France and imputed GBARD data

12. Disaggregation of Public Funded R&D (GBARD) • What we are looking for is a correspondence between NABS chapters and NACE industries, in order to break down GBARD by NACE industries. • For this reason, we need: • A correspondence between NABS and NACE classifications • Weights to distribute GBARD of each NABS chapter among NACE industries since each NABS is generally distributed in more than one industry. GBAORD by NABS GBAORD by NACE NABS 1 NABS 2 … … … NABS 14 NACE 01-03 NACE 05-09 NACE 10-12 … … …

13. Disaggregation of Public Funded R&D (GBARD) • Stančik, J. (2012)(http://ftp.jrc.es/EURdoc/JRC69978.pdf) offered a first NACE/NABS correspondence (including NACE Rev. 2 and NACE Rev.1.1). • This correspondence was used in the SPINTAN WP 23. • However, we have to take the other way around since we need a NABS → NACE. • JRC unit B6 and Ivie have revised the correspondence in 2016 Stančík, J. (2012): What we need: NABS classification NABS classification NACE classification NACE classification NACE 01-03 NACE 05-09 NACE 10-12 … … … … NACE 99 NABS 1 NABS 2 … … … … … NABS 14 Aggregation of NACE Disaggregation of NABS NABS 1 NABS2 … … … … … NABS 14 NACE 01-03 NACE 05-09 NACE 10-12 … … … … NACE 99

14. Disaggregation of Public Funded R&D (GBARD) Weights: • Since each NABS corresponds to more than one NACE, we need a proxy variable to estimate the distribution weights of each NABS among NACE (Rev. 1.1 and Rev. 2) industries. • We follow the spirit of Stančík, (2012): • Proxy selected: total labour costs (salaries times hours worked) of employees with higher education (ISCED codes 5a, 5b and 6) • Assumption: the distribution of government R&D expenditures by industries is similar to the distribution of labour costs of the most qualified employees. Labour costs by NACE (higher education) NABS 1 NABS 2 … … … … … … NABS 13 NABS 14 Table 1 Example: NABS 1 correspondence and estimation of weights NACE 011 … … NACE 122 … NACE 221 … NACE 325 … NACE 990

15. NABS-NACE correspondence • However, there are some NACE codes that are assigned to more than one NABS. • We have to breakdown labour costs of the NACEs affected to calculate the weights. • In these cases, we use the weight of the GBARD in the NABS involved to assign the labour costs of each NACE code to the corresponding NABS chapters • For example, NACE Rev.2 code 512 (Freight air transport and space transport) is assigned to NABS 3 (Exploration and exploitation of space) and 4 (Transport, telecommunication and other infrastructures), but we don’t have information about how much of the labour costs of higher education employees in this industry should be assigned to each NABS. • Therefore, we take as weights to distribute NACE 512 sector labour costs among NABS the weight of GBARD figures of each NABS (differentiating by country and year). Table 2 Labour costs of employees with higher education assigned to NABS 3 NABS 1 … NABS 3 NABS 4 … NABS 14 NACE 011 … NACE 512 … … NACE 990 ? Labour costs of employees with higher education assigned to NABS 4

16. Public Funded R&D (GBARD) by industry • Once the labour costs of employees with higher education at 3 digits NACE are assigned to each NABS, the weight of these costs are calculated by NACE within the same NABS • This weight will be used to estimate GBARD figures by industries. • The calculations are made for each country and year. • For each NACE industryk (in a given year and country): where j= NABS chapter, k= NACE industry and labour costs are hours worked times wages for employees with higher education. * See in Tables 1, 2 and 3 examples of how these weights are computed

17. Example of the estimation procedure Assignment of labour costs (higher education) Table 3 Example: NABS 3 and 4 distribution by NACE industries and estimation of weights NABS 1 NABS 2 NABS 3 NABS4 NABS 5 … … … … … … … NABS 13 NABS 14 … NACE 301 NACE 302 NACE 421 … NACE 511 NACE 512 … NACE 611 … NACE 711 NACE 712 … Procedure for NACEs assigned to more than one NABS Table 2 Example: Distribution of NACE 512 between NABS 3 and 4 using GBARD weights GBAORD NACE 512=2,667+1,500= 4,167

18. Disaggregation of Public Funded R&D (GBARD) • Statistical information: • GBARD by NABS from Eurostat • LFS (Labour Force Statistics): Hours worked by employees with higher education by 3 digits NACE industries. • Tailored request to Eurostat. • SES (Structure of Earnings Survey): Wages of employees with higher education by 3 digits NACE industries 2006 (NACE Rev. 1.1) and 2010 and 2014 (NACE Rev. 2). Data for the remaining years are extrapolated. • Tailored request to Eurostat.

19. Several adjustments: Correspondence between NACE Rev.1.1 and NACE Rev.2 • LFS and SES data are classified following different revisions of NACE classifications: • NACE Rev.1.1 for the period 2004-2007 • NACE Rev.2 for the period 2008-2013 • For this reason we apply thecorrespondence between NACE Rev.1.1 and NACE Rev.2 to convert GBARD by NACE Rev. 1.1. • Final NACE Classification: 37 2-digits NACE Rev.2 industries comprising 7 ICT sectors (OECD classification). • Data on GBAORD by NACE has been smoothed to take into account the effects on the dataset of the NACE revision in 2008. • Data from 2006-2012 are smoothed based on the estimation of a linear regression with a trend and a step dummy for 2008-2013.

20. Final NACE Rev.2 industry classification • Final NACE Classification: 37 2-digits NACE Rev.2 industries, differentiating ICT sectors:

21. From GBARD to ICT GBORD 2nd Step: Once GBARD by NACE industries has been computed, we take a further step: wesplit GBARD of each industry and year into ICT and non-ICT “assets”. • We again follow Stančík (2012): we apply theICT share to total GBARD for each country, industry and year: where k= NACE industry and • In the definition of the ICT occupations we use Eurostat’s taxonomy (ICT specialists in employment) ISCO-88 codes: 1236, 213, 2144, 2359, 3114, 312, 313, 7242, 7243 and 8283 ISCO-08 codes: 133, 25, 35, 2152, 2153, 2166, 2356, 2434, 3114, 7421 and 7422 • The ICT shares are smoothed (double exponential smoothing) due to the fact that detailed information (3-digit industries crossed by occupations) from the LFS and the SES is not always complete.

22. Results 1. GBARD (Eurostat data):

23. Results GBARD by Member State (2015). Millions of euros Total EU-28 GBARD: 94,450 million euros Source: Eurostat

24. Results GBARD intensity (GBARD / GDP) by Member State (2015). Percentage Source: Eurostat

25. Results Distribution of GBARD by NABS chapters (2015). EU. Percentage Source: Eurostat

26. Results GBARD and ICT GBARD by NACE (our estimation):

27. Results GBARD by NACE Rev. 2 industries. EU28 (2006-2015). Percentage 75% GBARD in 2015

28. Results ICT GBARD by NACE Rev. 2 industries. EU28 (2006-2015). Percentage

29. Results ICT GBARD/Total GBARD by NACE industries. EU28 (2015). Percentage

30. Mainissues • This paper proposes a methodology to estimate the R&D public sector transfers by NACE industries and which part correspond to ICT assets. • The methodology is based on GBARD data, which account for the sums committed by government to fund other industries R&D. • The methodology closely follows Stančík, J. (2012). • A correspondence between the NACE and NABS classification and the adequate weights to spilt different GBARD by NABS into NACE industries are used. • Two key assumptions: • the distribution of government R&D expenditures by industries is similar to the distribution of labour costs of the most qualified employees. • the percentage of ICT R&D assets in each industry is proportional to the share of labour costs of employees with higher education in ICT occupations over total labour costs with higher education. • The methodology is tested for the EU member states for the period 2006-2015. It offers a disaggregation of 37 2-digits NACE industries (comprising 7 ICT sectors).

31. Mainissues • Results show that R&D transfers are concentrated in a small number of industries: • 4 industries concentrate 75% of total GBARD: • Scientific Research and development; Health; Public Administration (including defense); and Legal, accounting, technical testing and analysis, architectural and engineering activities. • As expected, ICT GBARD intensive industries (in terms of the % of total GBARD) are the ICT sectors.

32. A Methodology for Disentangling Public Funded R&D (GBARD) by Industries Matilde Mas (University of Valencia and Ivie) Eva Benages (Ivie) Juan Fernández de Guevara (University of Valencia and Ivie) Laura Hernández (Ivie) NTTS Conference, Brussels. 2017