Tasks to TFP: The Micro Origins of Aggregate Productivity

Department of Economics and Decision Sciences

Speaker : Joseba martinez (LBS)

Room : T-027

Abstract :

We introduce a new measure of innovation, task relevance, which links the technology covered by patents to the production tasks performed in the U.S. economy over the past century. We first create a representation of the task space based on textual descriptions of work, and second, measure the semantic relevance of U.S. patents to tasks using natural language processing tools. Unlike existing metrics, this relevance measure can be computed for any patent upon publication, while still correlating strongly with established measures of patent quality, such as citations, importance, and value. To demonstrate the explanatory power of task relevance as a measure of technological change, we revisit two classic questions. Cross-sectionally, we find that relevance explains a substantial share of labor market polarization in task wages and hours, in line with \cite{AutorDorn2013}. At the aggregate level, we demonstrate that: i) relevance shocks transmit in a manner consistent with classic aggregate technology shocks, raising output, lowering prices, and have persistent positive effects on TFP, explaining a sizable fraction of its variance; and ii) in out-of-sample forecasts, task relevance significantly improves predictions of medium- to long-run TFP growth relative to benchmarks.