Association for Public Policy Analysis & Management

Innovation spillovers is a central concept in theories of technological change, but detailed empirical studies have been limited. Literature suggests that breakthrough innovations arise from novel combinations of more technologically “distant” prior inventions. How can we systematically understand the importance of spillovers from more distant domains in the development of a technology?

Utilizing advances in multiple academic disciplines (computer science, econometrics, and network analysis), this paper develops a new methodology to identify the role of spillovers in the historic advancement of a technology domain. Our methodology consists of three unsupervised steps. First, we develop definitions of solar component technology fields that identify the set of patents that represent innovations in key solar sub-technologies. Second, we utilize natural language processing to quantitatively measure the “technological distance” between solar patents. Specifically, we apply the Latent Dirichlet Allocation (LDA) algorithm, a topic modeling method, to categorize patent abstracts and to create a reduced-dimension representation of documents within a corpus based on word co-occurrence within patent abstracts. This approach allows us to calculate technological distance with simple measures and decompose distances within "technology space" to characterize the novelty of innovations relative to prior innovations in the field. Finally, we use econometric techniques to identify patterns in the technological distance between cited and citing patents at points in the citation network that suggest particular significance for the sector. These patterns quantify the dependence of high network importance on citation relationships with more distant prior art.

We demonstrate our method by studying the development of solar photovoltaic (PV) technology over the period 1901– 2018 using PATSTAT database. By applying our method to PV patent data, we will display visualizations of the trajectories of PV innovations over time and specific technological breakthroughs and spillovers in the history of PV development. Our econometric models will allow us to estimate the extent to which significant shifts along the main trajectory of solar PV innovation can be attributed to the incorporation (as identified in prior art citations) of technologically distant patents (i.e. spillovers).

The results of applying this method to solar PV demonstrate the contribution of other technology sectors to the development of technology critical for addressing energy and environmental challenges. The results can also help identify possible new roles and types of activities for public policy and research organization design, complementing other insights on the significance of public and private R&D funding and deployment support. This method is fully unsupervised and could be applied to climate-related technologies in other domains. Further, our research group will be combining the fully unsupervised methodology developed here with qualitative interviews with experts and economic modeling of technology cost and performance to assess the validity and implications of this methodology.