Spreading inventions drive 40 percent of productivity decline

Analyzing claims across more than 11 million US patents from 1836 to 2023, researchers documented a secular decline in invention similarity using validated neural language models.

This trend is corroborated by a 98 percent drop in patent interference rates over 150 years, showing that independent simultaneous inventions have almost vanished.

In the paper's spatial competition model, inventors choose locations within an expanding idea space.

As the burden of knowledge drives up entry costs, inventors spread further apart to gain pricing power over local territories.

While spatial isolation encourages higher individual research investment and yields higher private patent values, it simultaneously attenuates local knowledge spillovers between researchers across the economy.

Proper measurement of distance in idea space requires rigorous model selection.

While traditional word-frequency models like TF-IDF incorrectly indicate growing similarity, validated neural embeddings confirm persistent dispersion.

A calibrated structural decomposition attributes roughly 40 percent of the long-run decline in US research productivity to these spatial forces.

As inventors move further apart, aggregate research spending escalates to maintain territorial coverage, while the attenuation of knowledge spillovers slows total factor productivity growth.