How do innovation dynamics work? Is it possible to predict which kind of technological innovation will emerge in the medium-term future?

Researchers at CREF Matteo Straccamore, Luciano Pietronero, and Andrea Zaccaria, tried to reconstruct the innovation dynamics of about two hundred thousand companies following their patenting activity for about ten years.

They quantified the relatedness between a firm and a technology sector in different ways, namely using standard methods based on co-occurrences networks and supervised machine learning algorithms (Tacchella et al. (2021); Albora et al. (2021)).

To compare such assessments, they developed an out-of-sample prediction framework based on the assumption that, on average, the next technology sector in which a firm will patent will be among the ones that are more related to its present patenting portfolio. In this way, they can build and study the technological adjacent possible of innovative firms. The concept of adjacent possible was originally introduced by Kauffman (1996) and subsequently mathematically formalized by Tria et al. (2014); Loreto et al. (2016).

They found that machine learning algorithms not only show better prediction performances but allow for a two-dimensional representation of technology sectors that we call Continuous Technology Space (CTS). The CTS can be used to visualize the patenting portfolio of companies and to design strategic investments and acquisitions.

You can read the full article here