Research

We examine how financial development shapes the creation and diffusion of FinTech innovation worldwide. Using a novel global dataset covering over 300,000 financial patents and patent transactions across 105 countries (1990–2020), we identify financial patents through a combined CPC-based search and PatentSBERTa text-classification framework. We distinguish between innovation (new filings) and adoption (patent acquisitions). Financial development is strongly associated with both margins, while regulatory sandboxes partially substitute for deep markets in less developed systems. Internally generated patents generate stronger knowledge spillovers, whereas acquired patents are more closely linked to subsequent firm sales growth, highlighting a divergence between public and private returns to innovation.

This paper asks whether green innovation spreads beyond directly regulated industries through inter-industry production networks. Using BEA input-output tables to construct downstream green innovation exposure for 84,700 U.S. firm-years (1997–2021), I find that the answer depends on network strength: below the 75th percentile of downstream economic importance, downstream green innovation displaces focal firm patenting (18.2% decline); above it, knowledge spillovers dominate (13.5% increase). The spillover is additive — firms add green patents without reducing non-green output — and concentrated in the same CPC technology domains as their downstream industries. Results survive predetermined IO weights and a placebo permutation test, identifying production networks as a regulation-independent channel for the green transition.

Links DealScan syndicated loans to Google Patents through Orbis ultimate-owner records, identifying AI patents with a classifier applied to the assignee corpus, to ask whether a lead arranger's own AI invention changes the contracts it writes. Measuring the lender at its time-varying bank-holding-company parent, rather than the booking entity DealScan reports, overturns several results obtained from the naive match. Across roughly 104,000 facilities (1985–2025), US bank lead arrangers holding AI patents filed in the preceding five years arrange facilities about 30% larger at slightly shorter maturities, with the effect increasing in the number of AI patents and in the breadth of the bank's early entry across AI subfields. Covenants, renegotiation and ex-post borrower performance are unchanged, pointing to a capacity channel rather than improved screening.