(Green) Knowledge spillovers and regional environmental support: do they matter for the entry of new green tech-based firms?

NUTS (Nomenclature of Territorial Units for Statistics). The European Commission divides its territory according to the classification established by Eurostat. It is based on national administrative units.

Due to the data unavailability, in the REGPAT database, on the marketability of the patented objects and their economic value, we use the terms of invention and innovation interchangeably.

Alternative definitions for the entry of new green tech-based firms (i.e., at least one green patent, more than 75% or 100% of green patents over the period observed, without considering the year of their first green patent application within the first 5 years of existence, etc.) have been applied, but results do not significantly vary. Still, results are available upon request.

Each patent can be assigned to several technological classes. For example, let us assume for simplicity that we have two patents. The patent EPO48213 includes four technology classes (at 3-digit level): H02 (Generation and distribution of electric power), B05 (Spraying or atomising in general), Y04 (Information or communication technologies having an impact on other technology areas) and Y02 (Technologies for mitigation or adaptation against climate change). The patent EPO5579 includes three technology classes C08 (Organic macromolecular compounds), B60 (Vehicles in general) and B29 (Working of plastics). When we compute the variety and green variety indicators the co-occurrences change. In fact, when we calculate green unrelated variety (UKV_GREEN) the co-occurrence of the EPO5579 patent is not considered because it is not assigned to a green technological class. Similarly, for the calculation of green related variety (RKV_GREEN), we consider only the co-occurrences within the macro-field Y. This means that we only account for the co-occurrence of Y02 with Y04 in EPO48213 patent. Co-occurrences within the nongreen technological macro-fields (e.g., B60 and B29) are not considered.

To elaborate the variable VOTE_ECO it is assumed that a given election is only relevant for a given 5-year period if the year of the election is lagged with at least 3-years from the start of a 5-year period (Santoalha and Boschma 2019). According to that, results in the first round of the French national elections of 2002 are associated with the period 2003–2009; results in 2007 are associated with the period 2010–2013; and results in 2012 are associated with the period 2014–2015.

The data comes from a survey whose aim is to assess the environmental protection expenditure in the industry. Expenditures concern investments (equipment entirely dedicated to environmental protection, purchases of more efficient equipment in terms of the environment, etc.), current expenditures to protect environment (operating and maintenance costs dedicated to environmental protection), as well as the cost of feasibility studies (return on investment, evaluation of regulatory impact, audits to obtain an environmental certification) (See https://​www.​insee.​fr/​fr/​metadonnees/​source/​serie/​s1232, for more details). The data are available at NUTS 2 level so, to compute the NUTS 3 corresponding variables, we weight the NUTS 2 variables by the NUTS 3 share of regional GDP.

The focus on the entry of new green tech-based firms implies that in a year t and a département i the mean maximum number of entries would be equal to 3. The nature of this data implies that by using our dichotomic variable we can, to a great extent, explain the capacity of the region to incentivise the successful green technological entry, as Corradini (2019) also suggested.

Other spatial weighting definitions, such as 5-nearest neighbours or an inverse distance-based matrix, were considered. Even so, the best fit of the model is obtained when we rely on a row-standardised contiguity weighting matrix, an approach that has already been used in previous contributions for the case of French metropolitan départements. See, for instance, Elhorst and Fréret (2009).

More details on the test procedure to find out the most appropriate model can be found in Sect. 5.1.

See Soete and Patel (1985) for similar approaches. Alternative rates of obsolescence have been applied with no significant changes from the final calculations.

See Saviotti (1988), Frenken and Nuvolari (2004) and Stirling (2007) for further details on the definition and properties of this index.