EconBase
← All papers

Systematic Bias in Green Patent Classification: Silent Green and False Green

Hamid Bekamiri, Jan Auernhammer, Milad Abbasiharofteh, Jesper Lindgaard Christensen

arXiv 24 Aug 2026 · Econometrics

arXiv:2608.23420 · PDF · Extracted main text

Abstract

Green-patent indicators built on Cooperative Patent Classification Y02 tags are widely used in research, policy, and investment, yet their construct validity has not been audited at corpus scale. We assess whether Y02 is systematically biased and whether that bias may reinforce the ESG innovation disconnect. We introduce an Error-as-Signal framework that treats disagreement between an administrative label and an independent model as diagnostic evidence of measurement error. Screening 9,075,421 USPTO granted patents (1962-2024) with a fine-tuned domain model against Y02 yields 517,772 disagreements. Two independent open-weight large language models then judge by consensus whether each flagged invention has a direct climate-mitigation or adaptation function. We identify 180,384 administrative Type I errors (False Green), concentrated in digital data processing, wireless networks, digital communications, and semiconductors, and 29,465 Type II errors (Silent Green), concentrated in separation processes, catalysis, exhaust control, heat pumps, power systems, and batteries. Correcting consensus-attributed errors reduces the measured green-patent population by 25.5% (592,387 to 441,468 patents; sensitivity bounds 390,540-508,126), with ICT energy efficiency (Y02D) falling 67.6%. Omission follows an inverted-U relationship with technological atypicality, making complex and unconventional inventions especially likely to go unlabelled. Event tests show no discrete jump in misclassification when green classification became salient and only a small rise in green framing after the 2013 CPC launch. The bias primarily reflects classification capacity rather than applicant strategy and structurally under-recognizes heavy industry.

Citation extraction

43
references
92
in-text mentions
43
distinct cited
1
self-citations
15,700
main-text words

appendix boundary found by appendix_command · 81% of the source is main text. Read the extracted text to check this.

Most heavily cited references

The works this paper leans on most, across its whole bibliography — not restricted to papers in our corpus. Ranked by composite intensity, which combines how often a work is mentioned, how many sections mention it, and how much of that falls in the main text rather than the appendix.

ReferenceIntensityMentionsSectionsMain text
1Cohen, Lauren H. and Gurun, Umit G. and Nguyen, Quoc H (2026) The ESG–Innovation Disconnect: Evidence from Green Patenting1.00084100%
2Jacobs, Abigail Z. and Wallach, Hanna (2021) Measurement and Fairness1.00073100%
3Santarlasci, Lapo and Rungi, Armando and Zinilli, Antonio (2023) Seeing Through Green: Text-Based Classification and the Firm's Returns from Green Patents1.00063100%
4Veefkind, Victor and Hurtado-Albir, Javier and Angelucci, Stefano an… (2012) A New EPO Classification Scheme for Climate Change Mitigation Technologies0.87462100%
5Adcock, Robert and Collier, David (2001) Measurement Validity: A Shared Standard for Qualitative and Quantitative Research0.73732100%
6Cronbach, Lee J. and Meehl, Paul E (1955) Construct Validity in Psychological Tests0.73732100%
7Griliches, Z (1990) Patent Statistics as Economic Indicators: A Survey0.73732100%
8Lan, Yuxuan and Yuan, Ziyue and Tang, Ruiqi and Hsu, Shu-Chien and W… (2025) Green Innovation and the ESG Disconnect: Evidence from Green Patenting in the Construction Industry in China0.73732100%
9Rainville, Anne and Dikker, Irma and Buggenhagen, Magnus (2025) Tracking Innovation via Green Patent Classification Systems: Are We Truly Capturing Circular Economy Progress?0.73732100%
10Bekamiri, Hamid and Hain, Daniel S. and Jurowetzki, Roman (2024) PatentSBERTa: A deep NLP based hybrid model for patent distance and classification using augmented SBERT self0.64422100%

Showing the top 10 of 43 scored citations.