Predictive Patentomics: Forecasting Innovation Success and Valuation with ChatGPT

Abstract

Conventional approaches to analyzing structural data have historically limited our economic understanding of innovation. This paper pushes the boundaries, taking an LLM approach to patent analysis with the novel ChatGPT technology. I develop deep learning predictive models that incorporate OpenAI’s textual embedding features to access complex, intricate information about the quality and impact of each invention. These models achieve an R-squared score of 42% predicting patent value, 23% for patent citations, and clearly isolate the worst and best applications. My techniques also enable a revision to the contemporary Kogan, Papanikolaou, Seru, and Stoffman (2017) valuation of patents with a median deviation of 1.5 times, accounting for potential institutional anticipation and generating substantial incremental value for economic applications. Furthermore, the application-based measures provide previously inaccessible latent information regarding corporate innovative productivity; a long-short portfolio based on predicted acceptance rates achieves significant abnormal returns of 3.3% annually. The models provide an opportunity to reinvent startup and small-firm corporate policy vis-à-vis patenting.