Specialization, Diversification, and Environmental Technology Life Cycle

abstract

The article analyzes whether and to what extent regional related and unrelated variety matter for the development of green technology, and whether their influence differs over the technology life cycle. Using patent and socioeconomic data on a thirty-year (1980–2009) panel of US states, we find that unrelated variety is a positive predictor of green innovative activities. When unpacked over the life cycle, unrelated variety is the main driver of green technology development in the early stages, while related variety becomes more prominent as the technology enters into maturity.

Key words:

• green technology
• technology life cycle
• regional diversification

JEL codes:

• O33
• Q55
• R11

Acknowledgments

We are indebted to the editor and four anonymous referees for constructive feedback. We are also grateful for comments received at the following events: SEEDS Workshop (Ferrara, 2018); GeoINNO Conference (Barcelona, 2018); IAERE Conference (Turin, 2018); INGENIO PhD Days (Valencia, 2018); EU-SPRI Conference (Paris, 2018); Technology Transfer Conference (Valencia, 2018); Spanish Regional Sciences Association (Valencia, 2018); IAERE Conference (Udine, 2019); Rethinking Clusters Workshop (Padua, 2019); EMAEE Conference (Brighton, 2019). Davide Consoli acknowledges the financial support of Consejo Superior de Investigaciones Científicas (CSIC) “Ayuda de Incorporación para Científicos Titulares” (OEP2014) and of the Ministerio de Economía, Industria y Competitividad “Programa Estatal de I+D+i Orientada a los Retos de la Sociedad, 2017” (ECO2017- 86976R). François Perruchas acknowledges the financial support of European H2020 Project RISIS2 (grant agreement nº 824091). All errors and omissions are our own.

Data availability statement

The data described in this article are openly available in the Open Science Framework at DOI:10.17605/OSF.IO/TPA6U.

Supplementary material

Supplemental data for this article can be accessed here.

Notes

¹ For instance, Norton and Rees (1979) find that the decline of the US Manufacturing Belt during the late 1960s was essentially a core-periphery realignment, with theoretical roots in the product life cycle framework. The decentralization of production toward peripheral southern and western states followed the dispersion of innovative capacity and the rise of new, high-tech sectors at the beginning of the life cycle.

² Patent offices use IPC and CPC to classify patent documents. Both classification systems exhibit a hierarchical structure based on the technical content of the patents through codes. At the lowest level, that is, full-digit, the codes are very specific and refer to narrow technological fields, for example, IPC full-digit G06F9/02—“Arrangements for program control using wired connections.” At the highest level, that is, one-digit, the codes refer to broad technological domains, for example, IPC one-digit G —“Physics.”

³ In an intermediate step, we convert the IPC codes listed in the Env-Tech into CPC codes using a correspondence table provided by the EPO and the US Patent and Trademark Office. This allows us to use a unique classification system.

⁴ Specifically, Env-Tech classification codes cover eight technology groups: environmental management, water management, energy production, capture and storage of greenhouse gases, transportation, buildings, waste management and production of goods.

⁵ The GeoNames database provides geographic coordinates of a wide range of features such as mountains, lakes, countries borders, etc. In particular, it includes information on the latitude and longitude of the majority of the cities around the world. See http://www.geonames.org for more information.

⁶ When the postcode is missing, we identify the city in the address string using GeoNames, that is, we split addresses into several elements in order to isolate the street, city, etc. Then, since the city is usually provided at the end of the address, we browse the address string from right to left. Our algorithm compares each element of the address with the city name information included in GeoNames. We repeat this process for all the elements of the address string, moving from the end to the beginning, and associate the address to the city name in case of matching. For example, in the address: John Smith, 1 West 72nd Street, New York, NY, there are four elements to check: “John Smith,” “1 West 72nd Street,” “New York,” and “NY.” Starting from the right, the city will be detected in the second loop of the algorithm, that is, New York.

⁷ Daily search limits and costs did not enable us to use Google Maps API to search for the geographic coordinates of all addresses.

⁸ For more details, see https://github.com/cortext/patstat.

⁹ IPC three-digit classes capture generic domains of application while higher disaggregation, IPC eight digits, refers to specific applications. The first four digits of the code indicate the class and subclass, whereas the last four digits are the groups and subgroups. To illustrate, IPC code “A61B 5/022” identifies inventions that allow the “measurement of pressure in heart or blood vessels by applying pressure to close blood vessels,” while subclass “A61B” describes inventions related to “diagnosis, surgery and identification,” and finally class “A61” is associated with “Medical or Veterinary Science; Hygiene.”

¹⁰ The Env-Tech classification OECD (2016) groups green technologies at different digits (up to four). In the present article we exploit the two-digit level, which is a compromise between narrow (three digits) and broad (one-digit) technological fields. Moreover, the two-digit level guarantees coverage of all the technologies listed in the classification since some of them (e.g., four- and three-digit codes) are not provided for all the technologies. Finally, the two-digit level ensures that all the classes of the Env-Tech classification have at least one patent family over the period 1980–2009. Table 2 provides the list of green technological domains employed to define technology life cycle stages.

¹¹ We employ standard scores to normalize the values: $\frac{X-\mu }{\sigma }$ where $X$ are the values of patenting intensity or ubiquity, $\mu$ is the mean and $\sigma$ the standard deviation. See Figure B2 and B3 in the online materials.

¹² An exhaustive description of the yearly patterns is provided in Appendix B in the online materials.

¹³ Alternative patent indicators have been used to test the robustness of our results. The findings described in the following section hold, even if we consider just patent families with at least one granted patent. Results are available upon request.

¹⁴ Neighboring states are those that share a border.

¹⁵ In a set of ancillary regressions, we control for concentration of inventive activities in some areas within states. Although data availability restricts our analysis at the state level in order to control for R&D, human capital, neighboring states, and environmental policy, we estimate a model in which the units of analysis are the core-based statistical areas (CBSAs). In this model R&D, human capital, and environmental policy proxies are at the state level, whereas inventive activity is at the CBSA, which is a combination of metropolitan and micropolitan areas. The results—including fixed effects at the CBSA level estimations—confirm the main findings. In an additional specification, we also estimate the model in Equation 1 focusing on inventions developed in metropolitan statistical areas within each state. Once again, the robustness of our finding is confirmed. The results of these alternative empirical strategies are available upon request.

¹⁶ Energy intensity is measured using data on energy consumption by all the sectors of the US economy (i.e., residential, commercial, transportation, and industrial sectors). The results are robust if we employ data on energy consumption by the industrial sector. Moreover, we adopt an alternative strategy that accounts for emission intensity (CO2 emissions from fossil fuel consumption per unit of GDP). This proxy allows controlling for the use of fossil fuels not merely employed for energy generation (Mazzanti et al. 2015). Using this alternative strategy does not change the main results. Tables are available upon request.

¹⁷ The null hypothesis is rejected at 5 percent.

¹⁸ In Table C1 of Appendix C, in the online materials, we test the robustness of the assignment of green technologies to the life cycle stage. Using an alternative methodology to assign Env-Tech two-digit technologies close to the mean value, the coefficient of RV in the diffusion phase is not significantly different from zero.

¹⁹ See, for example, https://s3platform.jrc.ec.europa.eu/what-is-smart-specialisation.

²⁰ The data are freely available at https://www.greentechdatabase.com/.