Working Paper: NBER ID: w24053
Authors: Josh Lerner; Amit Seru
Abstract: Patents and citations are powerful tools for understanding innovative activity inside the firm, and are increasingly use in corporate finance research. But due to the complexities of patent data collection and the changing spatial and industry composition of innovative firms, biases may be introduced. We highlight several patent-level biases induced by truncation of reported patent awards and citations, affecting estimates of time trends and patterns across technology classes and regions. We then introduce measures of patent and citation biases. When aggregated at the firm level, these survive popular methods of adjustment and are correlated with firm-level characteristics. We show that these issues can lead to problematic – and ex ante predictable – inferences, using several examples from prominent streams of finance literature that use patent data. We suggest a number of concrete steps that researchers can employ to avoid biased inferences.
Keywords: patent data; corporate finance; innovation; biases
JEL Codes: G30; O34
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| truncation of reported patent awards and citations (O34) | systematic biases (D91) |
| systematic biases (D91) | estimates of time trends and patterns across technology classes and regions (O57) |
| systematic biases (D91) | problematic and predictable inferences in published research (C90) |
| systematic biases (D91) | non-classical measurement error (C20) |
| non-classical measurement error (C20) | confound inferences about firm-level innovation (O31) |
| systematic biases (D91) | firm-level characteristics (L20) |
| firm-level characteristics (L20) | estimates of time trends and patterns across technology classes and regions (O33) |