Pith. sign in

REVIEW 3 major objections 4 minor 34 references

Can knowledge reclassification accelerate technological innovation?

T0 review · 3 major / 4 minor · reviewed 2026-08-07 · deepseek-v4-flash

Pith's one-line read Reclassifying existing inventions accelerates technological growth, and the apparent drop in patent counts is a consequence of reclassification, not a slowdown in invention.

desk verdict A genuinely new growth model with reclassification at its core, but the cross-sectional validation leans on an assumption (constant W0) that is shakier than the prose admits. read the letter →

arxiv 2506.08656 v1 pith:RQR5Y34I submitted 2025-06-10 econ.GN q-fin.EC

classification econ.GNq-fin.EC
keywords technologicalreclassificationpatentclassificationknowledgegrowthinnovationdynamicstwo-parametermodeltruncationeffectCooperativereinterpretationof
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

Technological knowledge grows in two ways: new inventions are added, and existing inventions are reclassified as their meaning shifts — a rooftop solar panel is still a satellite-era invention, but it now belongs to a different technological class. This paper argues that reclassification is itself an engine of growth. From patent data it extracts two regularities — recent patents are more likely to be reclassified, and larger classes attract proportionally more reclassified patents — and folds both into a two-parameter model with a triggering rate $\alpha$ and a reclassification rate $\beta$. The model's characteristic equation $(1-1/g)^{1+\beta}=\alpha/g$ ties together the growth factor $g$, the timing of the apparent decline in recent patent counts, the share of reclassified patents, and the number of classifications per patent, and its predictions match observations across all major technology domains. If the paper is right, the widely reported drop in recent patent counts is a reclassification artifact, and improving classification systems is a genuine lever on the pace of innovation.

What carries the argument

The central object is a two-parameter recurrence for the cohort counts $n_\tau(t)$ of a technological class: $\Delta_t n_\tau(t)=0$ for $t<\tau-1$, $\Delta_t n_\tau(t)=\alpha\, n(t)$ for $t=\tau-1$, and $\Delta_t n_\tau(t)=\beta\, n_\tau(t)/(t-\tau+1)$ for $t>\tau-1$, where $\alpha\ge 0$ is the triggering rate and $\beta\ge 0$ the reclassification rate. Solving this recurrence with binomial identities and the generating function $G(z)=1/((1-z)^{1+\beta}-\alpha z)$ yields the characteristic equation $(1-1/g)^{1+\beta}=\alpha/g$ for the growth factor $g$, and from that root the paper derives closed predictions for the decline time $T$, the reclassification proportion $V$, and classifications per patent $W$. The two empirical regularities that carry the argument are that reclassification probability is inversely proportional to time since filing and that reclassifications into a class scale linearly with class size.

What would settle it

Estimate a reclassification rate $\beta_k$ for each CPC subclass directly from the 2019/2023 regrouping and test the timing relation per class: the filing year of peak patent count in subclass $k$ should sit roughly $\beta_k/(g_k-1)$ years before the reclassification moment, so subclasses with higher measured $\beta_k$ should show systematically later peaks. A sharper version exploits the 2013 introduction of the CPC system, which reclassified essentially the whole patent corpus at once: if reclassification is causal, classes that were heavily reclassified in that event should show a growth-rate discontinuity against otherwise similar classes that were left alone.

Watch

Extended reading notes

Core claim

The central claim is that reclassification accelerates technological growth. The author models a technological class through $n_\tau(t)$, the number of patents with filing year $\tau$ present in the class at time $t$, and lets each class evolve under two forces: new patents trigger further new patents at rate $\alpha$, and older patents are reclassified into or out of the class at rate $\beta$, with reclassification probability falling inversely with time since filing and scaling with class size. The exact solution grows exponentially, $n(t)\simeq n_0 g^t$, where the growth factor is the real root of $(1-1/g)^{1+\beta}=\alpha/g$; from the same solution the paper derives a decline time $T\approx\beta/(g-1)$, a reclassification proportion $V=g-1-\alpha$, and a classifications-per-patent relation $W\approx W_0(g-1)/\alpha$. Using $\alpha\approx 0.024$–$0.027$ and $\beta\approx 0.4$, the predicted growth factor $1.07<g<1.08$ matches the measured $1.079$; predicted decline times bracket the observed 4–9 year peaks; and the predicted classifications per patent (3.7–4.1 at group level, 2.0–2.13 at subclass level) sit on top of the measured 3.6 and 2.14. The paper also finds the predicted cross-sectional correlation between growth rates and classifications per patent within every CPC section, after controlling for group size and recent patent counts.

Load-bearing premise

The load-bearing premise is that $W_0$, the average number of classifications a brand-new patent receives when it first enters the system, stays constant or changes only very slowly across classes and over time; if $W_0$ varies, the observed correlation between classifications per patent and growth could reflect how classes are initially labelled rather than the effect of reclassification.

Editorial extensions

If this is right

  • Classes that are reclassified more often are predicted to grow faster, so keeping classification systems current — updating them frequently and consistently — becomes a direct, knowledge-intrinsic lever on the rate of innovation.
  • The apparent decline in recent patent counts, sometimes read as a real fall in invention or as a purely administrative truncation effect, is predicted to be a natural by-product of reclassification whose timing follows $T\approx\beta/(g-1)$; the model reproduces the observed 4–9 year lags.
  • The same two parameters $\alpha$ and $\beta$ predict four seemingly unrelated observables — growth factor, decline time, reclassification share, and classifications per patent — so knowledge-intrinsic factors can explain patterns such as the green-technology decline that are usually attributed to external shocks like resource prices and financial crises.
  • Reinterpretation enters the model as a routine, ongoing process rather than a rare revolutionary event, giving paradigm-shift theories of science and technology a quantitative, testable form.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Editorial inference: if reclassification itself causes growth, the same mechanism should show up wherever knowledge is retagged — scientific paper reclassification by subject codes, gene-ontology revisions, or library re-cataloguing — so the prediction that more active re-tagging yields faster measured growth is directly testable outside patents.
  • Editorial inference: the paper treats $\beta$ as a constant per technology, yet its own fits vary across reclassification moments ($\beta\approx 0.3$, $0.4$, $0.6$); letting reclassification effort respond to class size or search difficulty would make the growth-acceleration result endogenous and could change its magnitude.
  • Editorial inference: a longitudinal test that follows individual CPC classes through the 2013/2016 and 2019/2023 reclassification moments could separate reclassification-driven growth from selection of which classes get reclassified in the first place.
  • Editorial inference: the relation $W\approx W_0(g-1)/\alpha$ implies finer-grained classification schemes accelerate innovation, a prediction that could be checked by comparing growth under the CPC against growth under coarser national schemes such as the International Patent Classification.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 4 minor

Summary. This paper studies patent reclassification as a mechanism of technological knowledge growth. It documents two empirical regularities—recent patent cohorts are reclassified at higher rates, and reclassification flows into a class are proportional to class size—and builds a two-parameter model (triggering rate α, reclassification rate β) whose exact generating function yields predictions for the growth factor g (Eq. 7), the decline time T≈β/(g−1), the reclassification proportion V=g−1−α, and classifications per patent W≈W0(g−1)/α. The model is validated with four Patstat editions and CPC data at multiple levels; the paper concludes that reclassification accelerates innovation and explains apparent recent declines in patent counts.

Significance. If the causal interpretation held, this would be a useful contribution to innovation economics: it connects classification-system mechanics to growth, offers a knowledge-intrinsic explanation of patent-count declines, and generates quantitatively testable relations. The analytical derivation is careful and transparent, and the paper reports multiple robustness checks (Tables 1–6) that address group size, recent-patent bias, and double counting. These strengths are real. However, the empirical validations are weaker than the prose suggests, and the cross-sectional evidence that reclassified classes grow faster rests on an assumption (constant W0) that the paper itself shows to be suspect.

major comments (3)
  1. [§4, Eq. (15) and Figure 6] The cross-sectional test in Figure 6 is the paper's only direct evidence that classes with more reclassification grow faster, but it relies on Eq. (15), which assumes W0 is constant or changes very slowly. The author's own back-calculation using Eq. (53) gives 0.6 < W0 < 0.8 at subclass level, which he calls counter-intuitive and which is impossible if every patent has at least one subclass classification; this indicates the estimation is biased and that W0 may vary across groups. If W0,k correlates with g_k−1, the positive slopes in Figure 6 (R² 0.02–0.47) could reflect heterogeneous initial classification breadth rather than reclassification-driven growth. The quantitative prediction that the slope equals α/W0 is also not met: predicted ≈0.02 (α≈0.024, W0≈1.25) while observed slopes range from 0.002 to 0.016, an order-of-magnitude spread the paper leaves unexplained. This fragility is load-bearing for the central claim.
  2. [§4, Eqs. (10) and (53)] The validations of T and V are not independent tests. β is estimated from the reclassification fractions in Figure 2, and α is estimated from Eq. (53) using the same β to back out initial classification counts. The decline-time T is then predicted from Eq. (10) with that β, and V is computed as g−1−α; both are checked against quantities derived from the same reclassification moments (peak years of the cumulative classification curves and net reclassification proportions). For the 2023 dataset the paper changes β from 0.4 to 0.6 to bring the predicted T close to the observed 9 years (T≈7.6), which is a post hoc fit rather than a prediction. At minimum, out-of-sample validation—estimating β on one reclassification moment and testing T and V on another—would be needed.
  3. [Figure 6, SI Figure 9] The claim that the model's predictions are supported across all major technology domains is contradicted by the Chemistry/metallurgy panel: the main-text fit reports R²≈0.02, which is equivalent to no linear relationship. The SI improves this to R²≈0.12 only after excluding groups with w_k>9 from four subclasses, and the paper does not provide a principled reason for the exclusion beyond the empirical observation. This weakens the cross-domain universality claim.
minor comments (4)
  1. [Abstract and §4] Please qualify 'supported across all major technology domains' in light of the Chemistry R²=0.02 result and the post hoc exclusion of groups with w_k>9.
  2. [Figure 2] The fits use different β values (0.3, 0.4, 0.6) for different panels; please clarify whether these are fixed by the data or chosen to match each asymptote, and report uncertainty around β.
  3. [Eq. (14)] The definition of W in Eq. (14) should state explicitly that W0 is assumed constant across filing cohorts, not merely across time, since the derivation of Eq. (15) requires both.
  4. [Decline-time paragraph] The observed decline times are 7, 4, 6, and 9 years against a predicted value of 5.1; the text singles out 2023 as the only exception, but the spread across datasets is large enough to warrant a discussion of prediction intervals or measurement error.

Circularity Check

2 steps flagged · score 4.0 of 10

The reclassification-proportion 'prediction' reduces to an accounting identity, and the decline-time validation reuses the fitted β on the same datasets; the central growth-factor prediction retains independent content.

  1. fitted input called prediction [Model validations, third validation; Equations 11–13 and the paragraph beginning 'As a third validation']
    "Dividing left and right by n(t), using the approximation for n(t) in Equation 6 and summing over t, this becomes (g−1)t≃αt+Σ_{t'=1}^t v(t')/n(t'). For the left- and right-hand side to agree for large t, I conclude that v(t)≃n(t)(g−1−α). ... As a third validation, I use the estimated (lower) values for g and α in the expression V=g−1−α to predict a higher estimate for the reclassification proportion V≈0.056."

    Equation 12, Δ_t n(t)=αn(t)+v(t), is the sum of the model's own defining relations in Equation 1. Once g is measured as Δn/n and α is estimated as the new-classification fraction, V=g−1−α is an algebraic restatement of that accounting identity, not an independent prediction. The subsequent 'validation' measures V directly as net reclassifications over prior classifications—the same v(t)/n(t) quantity—using the same reclassification moments (2016/2019 and 2019/2023) that supplied the fits for β and the reconstructions for α in SI Equation 53. The agreement is therefore a bookkeeping consistency check rather than a test of the reclassification-growth mechanism.

  2. fitted input called prediction [Model validations, second validation; Equations 9–10 and β estimates from Figure 2]
    "As a second validation, I use g=1.079 and β=0.4 in Equation 10 to predict the decline-time T≈5.1. Figure 5 plots the total number of classifications of all patents by their earliest filing year and clearly demonstrates the apparent declines in recent years. ... A closer examination points out that classifications in the 2013, 2016, 2019, and 2023 data peak respectively in 2006, 2012, 2013, and 2014. This leads to decline-times in these datasets of 7,4,6 and 9 years."

    The β inserted into T≈β/(g−1) is not an independent parameter for the decline-time: it is fitted from the net-reclassification fractions r(τ,t)=β/(t−τ) in Figure 2, which are computed from the same four Patstat editions and the same families whose cumulative classifications produce the peaks in Figure 5. The peak condition leading to Equation 10 is β/(t−τ)≥g−1, a model restatement of that fitted inverse-time decay. Thus the T-validation is an in-sample consistency check rather than an out-of-sample prediction; although the peak years are not directly fit, the predicted and measured quantities are generated by the same reclassification moments and the same estimated β.

full rationale

The central derivation is not circular: Equation 7, g satisfying (1−1/g)^(1+β)=α/g, is obtained from the generating function of the model, and the first validation compares it with a separately measured OLS growth factor (g≈1.079 from 1980–2015 classifications). The cross-sectional W–g test in Figure 6 is also an independent-looking derived relation, though it depends on the fragile assumption that W0 is constant; the author's own SI back-estimate 0.6<W0<0.8 on the subclass level is counter-intuitive and indicates estimation bias, but that is a robustness problem, not a circularity. The two flagged validations are the ones that reduce to internal consistency: V=g−1−α is the accounting identity Δn=αn+v rewritten, and T reuses the β fitted from the same reclassification data that produces the apparent decline. The self-citations (footnote 4 and reference [24]) are acknowledgments and an application to green technology, not load-bearing support for the main claim. Overall, the central claim retains independent content, so the circularity score is moderate rather than severe.

Assumptions & free parameters 3 free parameters · 6 assumptions · 0 invented entities

The model rests on two fitted rates, alpha and beta, plus an assumed constant initial classification count W0. It also relies on domain assumptions about patents proxying knowledge and about the inverse-age reclassification law. No new unobserved entities, forces, or particles are introduced.

free parameters (3)
  • alpha (triggering rate) = 0.024 to 0.027 (CPC group level); 0.021 to 0.023 (subclass level)
    Estimated from classification counts corrected for reclassifications (SI, Figure 7); entered in Equation 1 and used to predict g, V, and W.
  • beta (reclassification rate) = about 0.4 overall; 0.3, 0.4, and 0.6 for the three reclassification moments
    Fit to net reclassifications per classification versus filing year in Figure 2; used in Equations 7, 10, and throughout the validations.
  • W0 (initial classifications per family) = 1<W0<1.5 on group level; 0.6<W0<0.8 on subclass level
    Estimated by dividing corrected initial classifications by unique families in the SI; assumed constant to derive W approximately W0(g-1)/alpha.
assumptions (6)
  • domain assumption Patents and their CPC classifications are a valid proxy for technological knowledge and for knowledge reclassification.
    Stated in the Phenomenology section; the author notes limitations but proceeds with patent data as the empirical definition of technology.
  • domain assumption The triggering rate of new inventions in a class is proportional to current class size n(t), with constant alpha.
    Second line of Equation 1; a standard endogenous-growth assumption, not derived in the paper.
  • domain assumption The probability of a patent being reclassified at time t is inversely proportional to its age t-tau, with constant beta, and only positive net reclassifications are modeled.
    Third line of Equation 1; fit to the three reclassification moments in Figure 2 and stated as empirical pattern 1. Deletions are not included in the model.
  • domain assumption The set of technologies is large enough that there can always be families reclassified to other classes.
    Footnote 9 in Equation 1; needed to avoid saturation of the reclassification source pool.
  • ad hoc to paper W0, the number of classifications a new patent receives upon introduction, is constant or changes very slowly.
    Used to derive Equations 14-15; authors admit subclass-level estimates give 0.6<W0<0.8, which they call counter-intuitive.
  • standard math The asymptotic theorem for rational generating functions used to extract n(t) approximately n0 g^t.
    Invoked in the model section with reference to Graham, Knuth, and Patashnik, page 341; a standard result taken as given.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Can knowledge reclassification accelerate technological innovation?." pith.science (2026). https://pith.science/paper/RQR5Y34I

@misc{pith2026250608656,
  author       = {Pith},
  title        = {Pith review of: Can knowledge reclassification accelerate technological innovation?},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/RQR5Y34I}},
  note         = {Machine review of arXiv:2506.08656}
}
read the original abstract

Technological knowledge evolves not only through the generation of new ideas, but also through the reinterpretation of existing ones. Reinterpretations lead to changes in the classification of knowledge, that is, reclassification. This study investigates how reclassified inventions can serve as renewed sources of innovation, thereby accelerating technological progress. Drawing on patent data as a proxy for technological knowledge, I discuss two empirical patterns: (i) more recent patents are more likely to get reclassified and (ii) larger technological classes acquire proportionally more reclassified patents. Using these patterns, I develop a model that explains how reclassified inventions contribute to faster innovation. The predictions of the model are supported across all major technology domains, suggesting a strong link between reclassification and the pace of technological advancement. More generally, the model connects various, seemingly unrelated knowledge quantities, providing a basis for knowledge intrinsic explanations of growth patterns.

Figures

Figures reproduced from arXiv: 2506.08656 by the authors.

Figure 1
Figure 1. The time development of a class of knowledge mixes two dynamics: the introduction of new inventions each year (triangles) and the inclusion/exclusion of older inventions as a result of reclassification (squares/crosses). that the two types strongly interact, but how exactly largely remains an open problem. Earlier contributions have focused on either the process of new idea generation [1, 2, 3, 4], or the process of… view at source ↗
Figure 2
Figure 2. For three reclassification moments (2013/2016, 2016/2019, and 2019/2023), I calculate the net reclassifications and divide this by the total number of classifications prior to reclassification. I plot this fraction for the earliest filing year of the corresponding patent families. As the filing years τ get close to the moment in time of reclassification t, the net reclassifications per classification r(τ, t) sharply… view at source ↗
Figure 3
Figure 3. I plot the number of families added and deleted for all subclasses for the reclassification moment 2019/2023. Note that the axes are log-transformed, hence the fits with slope 1.0 suggest the number of reclassifications is proportional to subclass size. The offset (the y-coordinate of the fit intersection with the y-axis) that maximizes R2 for the positive reclassifications is −1.31. Plotting the fit without transfo… view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: Here is an example plot of nτ (t) (see Equation 3) for the filing years τ where 0 < τ < t for t = 30, 35, 40, choosing α = 0.05 and β = 0.5. The number of patents increases exponentially until the apparent ’dip’ as τ gets close to t, that is, in recent years. In [PITH…
Figure 5
Figure 5. Figure 5: I plot the total number of classifications of patents by earliest filing year according to 4 different datasets. I include a fit of the 2023 data (R2 = 0.98), which indicates that classifications grow exponentially (the y-axis is logarithmic). Note the similarity with …
Figure 6
Figure 6. Figure 6: For each CPC group k, I plot the growth rate gk − 1 for the average classifications per patent wk. I plot the groups for each CPC section separately (each group belongs to one section). As predicted, most slopes are of the same order of magnitude as the estimated α ≈ 0…
Figure 7
Figure 7. Figure 7: Estimated values for α between 2011 and 2016 based the number of classifications of patents in the 2023 data and then subtracting the expected number of reclassifications over the years for various estimated values of β. 53, more specifically, CY ′ ,Y = CY,P  1 − β Y …
Figure 8
Figure 8. Figure 8: The number of positive and negative reclassifications per family per filing year on a log [PITH_FULL_IMAGE:figures/full_fig_p020_8.png]
Figure 9
Figure 9. Figure 9: Growth rate gk − 1 and number of classifications per patent wk for each CPC group k, except the groups in subclasses C10M, C10N, B32B, and F17C. separately 11. We anticipate three criticisms: 1. That the observed relation between Wk and gk −1 may only be an indirect ef…

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

34 extracted references · 30 canonical work pages

  1. [1]

    The dynamics of correlated novelties

    Tria F, Loreto V, Servedio VDP, Strogatz SH. The dynamics of correlated novelties. Scientific Reports. 2014;4. doi:10.1038/srep05890

  2. [2]

    Technology networks: the autocatalytic origins of innovation

    Napolitano L, Evangelou E, Pugliese E, Zeppini P, Room G. Technology networks: the autocatalytic origins of innovation. Royal Society Open Science. 2018;5(6):172445. doi:10.1098/rsos.172445

  3. [3]

    Idea engines: Unifying innovation &amp; obsolescence from markets &amp; genetic evolution to science

    Lee ED, Kempes CP, West GB. Idea engines: Unifying innovation &amp; obsolescence from markets &amp; genetic evolution to science. Proceedings of the National Academy of Sciences. 2024;121(6):e2312468120. doi:10.1073/pnas.2312468120

  4. [4]

    The Language of Innovation

    Tacchella A, Napoletano A, Pietronero L. The Language of Innovation. PLOS ONE. 2020;15(4):1–20. doi:10.1371/journal.pone.0230107

  5. [5]

    History of Classification of Patents

    Bailey MF. History of Classification of Patents. Patent Office Society; 1946. Available from: https://books.google.nl/books?id=lIM0AQAAMAAJ

  6. [6]

    Invention as a combinatorial process: evidence from US patents

    Youn H, Strumsky D, Bettencourt LMA, Lobo J. Invention as a combinatorial process: evidence from US patents. Journal of The Royal Society Interface. 2015;12(106):20150272. doi:10.1098/rsif.2015.0272

  7. [7]

    Long-run dynamics of the U.S

    Lafond F, Kim D. Long-run dynamics of the U.S. patent classification system. Journal of Evolutionary Economics. 2019;29(2):631–664. doi:10.1007/s00191-018-0603-3

  8. [8]

    Classifying patents based on their semantic content

    Bergeaud A, Potiron Y, Raimbault J. Classifying patents based on their semantic content. PLOS ONE. 2017;12(4):1–22. doi:10.1371/journal.pone.0176310

Show all 34 references
  1. [9]

    How much can we influence the rate of innovation? Science Advances

    Fink TMA, Reeves M. How much can we influence the rate of innovation? Science Advances. 2019;5(1):eaat6107. doi:10.1126/sciadv.aat6107

  2. [10]

    Quantitative Determination of Technological Improvement from Patent Data

    Benson CL, Magee CL. Quantitative Determination of Technological Improvement from Patent Data. PLOS ONE. 2015;10(4):1–23. doi:10.1371/journal.pone.0121635. 12

  3. [11]

    An Economic Theory of Technological Change

    Nordhaus WD. An Economic Theory of Technological Change. The American Economic Review. 1969;59(2):18–28

  4. [12]

    Endogenous Technological Change

    Romer PM. Endogenous Technological Change. Journal of Political Economy. 1990;98(5, Part 2):S71–S102. doi:10.1086/261725

  5. [13]

    Innovation network

    Acemoglu D, Akcigit U, Kerr WR. Innovation network. Proceedings of the National Academy of Sciences. 2016;113(41):11483–11488. doi:10.1073/pnas.1613559113

  6. [14]

    Technological Interdependencies Predict Innovation Dynam- ics; 2020

    Pichler A, Lafond F, Farmer JD. Technological Interdependencies Predict Innovation Dynam- ics; 2020. Available from:https://papers.ssrn.com/abstract=3547474

  7. [15]

    Role of design complexity in technology improvement

    McNerney J, Farmer JD, Redner S, Trancik JE. Role of design complexity in technology improvement. Proceedings of the National Academy of Sciences. 2011;108(22):9008–9013. doi:10.1073/pnas.1017298108

  8. [16]

    The structure of scientific revolutions

    Kuhn TS. The structure of scientific revolutions. vol. 111. Chicago University of Chicago Press; 1970

  9. [17]

    Technological paradigms and technological trajectories: A suggested interpretation of the determinants and directions of technical change

    Dosi G. Technological paradigms and technological trajectories: A suggested interpretation of the determinants and directions of technical change. Research Policy. 1982;11(3):147–162. doi:10.1016/0048-7333(82)90016-6

  10. [18]

    The NBER patent citation data file: Lessons, insights and methodological tools; 2001

    Hall BH, Jaffe AB, Trajtenberg M. The NBER patent citation data file: Lessons, insights and methodological tools; 2001

  11. [19]

    Truncation bias corrections in patent data: Implica- tions for recent research on innovation

    Dass N, Nanda V, Xiao SC. Truncation bias corrections in patent data: Implica- tions for recent research on innovation. Journal of Corporate Finance. 2017;44:353–374. doi:https://doi.org/10.1016/j.jcorpfin.2017.03.010

  12. [20]

    Global trends in the invention and diffusion of climate change mitigation technologies

    Probst B, Touboul S, Glachant M, Dechezleprˆ etre A. Global trends in the invention and diffusion of climate change mitigation technologies. Nature Energy. 2021;6(11):1077–1086. doi:10.1038/s41560-021-00931-5

  13. [21]

    Climate Change, Directed Innovation, and Energy Transition: The Long-run Consequences of the Shale Gas Revolution

    Acemoglu D, Aghion P, Barrage L, H´ emous D. Climate Change, Directed Innovation, and Energy Transition: The Long-run Consequences of the Shale Gas Revolution. National Bureau of Economic Research; 2023. 31657. Available from:http://www.nber.org/papers/w31657

  14. [22]

    Financial markets and green innovation

    Aghion P, Boneva L, Breckenfelder J, Laeven L, Olovsson C, Popov AA, et al. Financial markets and green innovation. European Central Bank; 2022

  15. [23]

    The Private Value of Clean Energy Innovation

    Martin R, Verhoeven D. The Private Value of Clean Energy Innovation. Author working paper; 2022. Available from:https://www.aeaweb.org/conference/2023/program/paper/ e7R598BG

  16. [24]

    The evolving boundary of green technology

    Barbieri N, Hotte K, Persoon P. The evolving boundary of green technology. arXiv e-prints. 2025;:arXiv:2503.21310. doi:10.48550/arXiv.2503.21310

  17. [25]

    www.cooperativepatentclassification.org; 2025

    Cooperative Patent Classification. www.cooperativepatentclassification.org; 2025. Available from:https://www.cooperativepatentclassification.org/cpcSchemeAndDefinitions/ table. 13

  18. [26]

    University Versus Corporate Patents: A Window On The Basicness Of Invention

    Trajtenberg M, Henderson R, Jaffe A. University Versus Corporate Patents: A Window On The Basicness Of Invention. Economics of Innovation and New Technology. 1997;5(1):19–50. doi:10.1080/10438599700000006

  19. [27]

    On limiting or encouraging rivalry in technical progress: The effect of patent scope decisions

    Merges RP, Nelson RR. On limiting or encouraging rivalry in technical progress: The effect of patent scope decisions. Journal of Economic Behavior & Organization. 1994;25(1):1–24. doi:10.1016/0167-2681(94)90083-3

  20. [28]

    Concrete mathematics: a foundation for computer science

    Graham RL, Knuth DE, Patashnik O. Concrete mathematics: a foundation for computer science. 9th ed. Addison-Wesley; 1993

  21. [29]

    Patent Citations as a Measure of Knowledge Flows: The Influence of Examiner Citations

    Alc´ acer J, Gittelman M. Patent Citations as a Measure of Knowledge Flows: The Influence of Examiner Citations. The Review of Economics and Statistics. 2006;88(4):774–779

  22. [30]

    Patent citation: A technique for measuring the knowl- edge flow of information and innovation

    Sharma P, Tripathi RC. Patent citation: A technique for measuring the knowl- edge flow of information and innovation. World Patent Information. 2017;51:31–42. doi:https://doi.org/10.1016/j.wpi.2017.11.002

  23. [31]

    Concrete Mathematics: A Foundation for Computer Science

    Graham RL, Knuth DE, Patashnik O. Concrete Mathematics: A Foundation for Computer Science. Reading: Addison-Wesley; 1989. Supporting information In this Supporting Information, I first discuss how to derive the model solutions, an expression for the growth factor, and the numb...

  24. [32]

    That the observed relation betweenW k andg k −1 may only be an indirect effect because both quantities would correlate the size of CPC groupk(in number of patents)

  25. [33]

    That the observed relation betweenW k andg k −1 may only be an indirect effect because both quantities correlate to the characteristic to have many recent patents (this would count forW k because reclassification especially targets recent patents, and many recent patents would...

  26. [34]

    This criticism suggests that if theg k −1 would be based not on the number of classifications but on the number of unique patents, there would not be a relation withW k

    That the observed relation betweenW k andg k −1 is only an ’apparent effect’ because more reclassifications results in more classifications but not necessarily in more patents (or faster growth). This criticism suggests that if theg k −1 would be based not on the number of cla...

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.