Patents, Firm Rents, and Worker Compensation: Causal Evidence from Quasi-Random Patent Allocation — with André Diegmann
This paper provides new causal evidence on how patent allowances affect firms and their employees based on quasi-random assignment of patent applications to examiners. Exploiting employer–employee records with newly linked German firm data and web-scraped patent documents, we show that patent-induced shocks reduce firm exit, improve productivity, and increase wages, with rent-sharing elasticities between 0.10 and 0.21. Wage gains are broadly observed across occupational tasks, with substantial heterogeneity: managers benefit disproportionately in publicly traded firms, whereas broader wage increases accrue to workers in non-traded firms. Our findings highlight the role of institutional features and firm organization in shaping how rents are shared.
Intangible Investment and Wage Distribution
This paper studies how intangible investment shapes the wage distribution. It combines firm-level innovation data from the Mannheim Innovation Panel (MIP) with the Linked Employer-Employee Data of the IAB (LIAB) and imputes establishment-level intangible investment. Across establishments, intangible investment is associated with wage polarization, with stronger wage increases at the bottom and top. Within establishments, however, investment spikes raise wages at the top while wages in the middle and bottom change little, widening the upper tail of the wage distribution. Two complementary mechanisms help explain this pattern. Skill-biased technological change concentrates wage gains among skilled workers, while employer market power limits gains among lower-paid workers. When estimated jointly, the two channels reinforce each other: greater employer market power particularly limits wage gains for low-skilled workers, widening the difference in gains between low- and high-skilled workers. The results show that the wage effects of intangible investment depend on both workers’ skill and employer market power.
Patenting Activity in Germany Across Inventors and Firms (with André Diegmann, Enrico De Monte, Javier Miranda, Bettina Peters, and Verena Plümpe)
We build a new linked data infrastructure for Germany that connects EPO patent applications to legally registered firms in the Mannheim Enterprise Panel (MUP) and, through a novel firm–establishment linkage, to administrative establishment and inventor records of the Institute for Employment Research (IAB). We use these PATSTAT–MUP–IAB data to document three facts on firm-level patenting between 2001 and 2021. First, patenting is highly concentrated: the top 1% of patenting firms account for about half of all applications and the top 10% for more than four-fifths, and patent output is far more concentrated than employment or firm age. Second, the extensive margin of patenting weakens over time: applications grow faster than the number of patenting firms, and the share of first-time patenting firms falls from about 30% to about 20%. Third, patenting remains concentrated in manufacturing, where the decline in first-time patenting is most pronounced. Inventor-level evidence for the 2000s complements these facts: EPO filings are a positively selected segment of German inventive activity, and leading patenting firms produce more patents per inventor than other firms.
What Wins Interest Is Not What Closes the Deal: LLM-Measured Judgment in Early-Stage Finance (with Müge Süer)
What makes investors interested in an early-stage venture may not be what closes the deal. We build a new dataset of 642 pitches from the U.S. television show Shark Tank, following the same ventures from investor interest, through an on-air handshake, to whether that agreement survives due diligence after the show, an outcome no database records. For each pitch, we distinguish realized financial figures (hard information) from three forms of soft information: founders' forward-looking claims, linguistic features captured by conventional textual analysis, and contextual judgment. To measure judgment, we develop an approach based on large language models (LLMs) that reads each pitch in context and forms a judgment of the venture. Judgment is the strongest predictor of competing investor offers, is associated with better terms for founders, and predicts which ventures survive after the show. It remains the strongest predictor at the handshake, although realized and forward-looking financial figures also become predictive at this stage. However, judgment does not predict whether an agreed deal closes. At that stage, founders' forward-looking claims become predictive instead: larger asks, projections, and implied valuations are associated with a higher probability that the deal falls through. Different information is predictive at different stages of the same investment process.
The Effect of Anticipated Discrimination on Gender Differences in Entrepreneurial Finance (with Müge Süer)