• <tr id="yyy80"></tr>
  • <sup id="yyy80"></sup>
  • <tfoot id="yyy80"><noscript id="yyy80"></noscript></tfoot>
  • 99热精品在线国产_美女午夜性视频免费_国产精品国产高清国产av_av欧美777_自拍偷自拍亚洲精品老妇_亚洲熟女精品中文字幕_www日本黄色视频网_国产精品野战在线观看 ?

    Modeling the dynamics of firms’technological impact?

    2021-12-22 06:51:54ShuqiXu徐舒琪ManuelSebastianMarianiandLinyuan呂琳媛
    Chinese Physics B 2021年12期

    Shuqi Xu(徐舒琪) Manuel Sebastian Mariani and Linyuan L¨u(呂琳媛)

    1Yangtze Delta Region Institute(Huzhou)&Institute of Fundamental and Frontier Sciences,University of Electronic Science and Technology of China,Huzhou 313001,China

    2University Research Priority Program on Social Networks,University of Zurich,CH-8050 Zurich,Switzerland

    3Beijing Computational Science Research Center,Beijing 100193,China

    Keywords: firm technological impact,patent analysis,impact dynamics

    1. Introduction

    Complexity science methods have been recently applied to understand the dynamics of impact along creative careers.[1–7]In particular, researchers have unveiled remarkable regularities through the quantitative analysis of individuals’ creative works (such as scientists’ articles and directors’ movies). These works found that the career trajectories of scientists,[1,8]artists,[8]directors,[6,8]and best-seller authors,[6,9]among others, are determined by a uniform pattern called random-impact rule: the most important works occur randomly within an individual’s work sequence. Sinatraet al.emphasized that the ability of an individual to produce high-impact works is independent of his/her career stage,[1]based on which they developed theQmodel,where the impact of an individual’s works is determined by a random factor,reflecting the potential of the idea,and a parameterQ,representing the individual’s capability to improve the work.According to the model,high-impact discoveries require a combination of an excellent idea and highQ. These methods and findings are mainly focused on individuals’creative careers,and they have not been extended to model the impact of organizations,such as firms.

    Existing evidence for firms suggests discrepancies among the impact dynamics of firms and individuals. Vast management literature acknowledged that firm age is an essential determinant of a firm’s impact.[10,11]And it is commonly recognized that young firms tend to present higher capability of innovation.[12,13]As a consequence, firm age might be negatively related to the impact of its technology. Recently,Xuet al.[14]provided initial evidence for systematic differences between temporal patterns of the highest-impact works in scientists’careers and firms’lifecycles.However,this work focused on the empirical evidence,and it did not propose a dynamical model to describe firms’impact evolution. How to modify existing impact dynamical models to accurately reproduce the observed firms’impact patterns?

    To answer this question, we analyze the patent licensing histories of worldwide firms from the U.S. Patent and Trademark Office(USPTO)and the European Patent Office(EPO).We propose a new dynamical model to capture the observed dynamic regularities. Specifically,by analyzing the evolution of average patent impact within firms’lifecycles, we observe a constantly decreasing trend, which indicates that firms tend to receive diminishing returns from their patents as they are issued more patents.We show that theQmodel is unable to capture firms’empirical impact patterns. Motivated by this finding,we generalize theQmodel by incorporating the property that patents’impact depends not only on the firm’s capability and the potential of the patent idea, but also on the patent’s position within the firm’s lifecycle. Therefore, we propose a time-varying returns model to reproduce the observed patterns of the dynamics of firms’impact. We show that for both two analyzed datasets, the proposed model matches the empirical results more accurately than the state-of-art random-impact model andQmodel. Our study sheds light on modeling the patterns of firms’impact dynamics,which can provide useful insights for managerial practices.

    2. Method

    2.1. Patent data of firms

    Patent activity is usually considered as a rich source to reflect firms’impact in terms of technology.[15–17]In this study,we investigate two large patent datasets: (i) The record of the patents granted by the EPO to 163565 firms from 1978 to 2016,covering 772127 patents. Note that the original data involves all types of applicants which are not limited to firms.In the analysis, we identify firm applicants by specific name identifiers, see Appendix A for details. (ii) The patent data from the USPTO to 7440 firms from 1926 to 2017,amounting to 2458402 patents.[18,19]Only patents whose applicant is a listed firm on the US stock market are included in the data.For each patent, the two datasets both include patent number,issuance date,names and ids of applicant firms,and a list of patent references. Therefore,we can identify the applicant firms of each patent, and based on the reference information,we can calculate the citation count of each patent at any given time. We summarize the main properties of the two studied datasets in Table 1.

    Table 1. Datasets description.

    EPO and USPTO are regarded as the most representative and highly reliable databases for patent analysis and have been used in many information science and cross-disciplinary studies.[20–23]Table 1 demonstrates that at the patent level,USPTO contains more patents than EPO. Kimet al.[20]argued that the United States is considered as one of the world’s largest commercial and technological markets,and for this reason, international firms tend to file patents in the USPTO to claim their rights to protect the innovation in the US. As for the EPO data, it contains rich information on technological innovations mostly related to firms located in European countries. At the firm level, the USPTO data involves fewer firms than the EPO data, because all firms in the USPTO data are limited to listed firms,whereas there is no such restriction for firms in the EPO data.

    We further show the cumulative distribution of the number of issued patents for firms in the two datasets in Fig.1(a).We observe that firms’ patenting activities is heavier in the USPTO than in the EPO. In the EPO data, 61.5% of firms have been issued only one patent, 93.9% of firms have been issued no more than 10 patents. For the USPTO data,there are only 13.44%of firms that have been issued one patent in total,82.34% of firms have been issued no more than 100 patents.One of the main reasons for this difference is that the listed firms usually have an advantage over ordinary firms in terms of resource and strength(for example,in terms of finance and innovation),which allows them to issue more patents.

    Table 2 reports the top five firms issued the largest number of patents in the two datasets. In the EPO data, the top firms are mostly from Japan,the United States,and European countries; as for the USPTO data, most of the top firms are from the United States and Japan.

    In the following, we only consider patents issued up to 2006. In this way, we have at least a 10-year time window to observe their impact. Furthermore, the research subjects of this study are firms, we need to focus on firms with sufficiently active patenting activities. For this reason, we limit our firm-level analysis to firms that have been issued at least 20 patents for the EPO data,which refers to 3548 firms,and at least 15 patents for the USPTO data,which filters 2819 firms(see Table 1). Our conclusions are robust with respect to alternative choices of the filter standard of analyzed firms (see Appendix C for the results of filtering firms based on their number of years of patenting activities).

    Table 2. Top firms by the number of issued patents. Time range refers to the first and last patent issuance year.

    Fig.1. Cumulative distribution of firms’patent number and technological impact. (a) Cumulative distribution of the number of issued patents for firms in the EPO and USPTO data. (b)–(c)Cumulative distribution of TI for firms in EPO(b)and USPTO(c)data,as well as the impact thresholds that distinguish the three firm groups of impact.

    2.2. Quantifying technological impact

    Patent’s technological impact. To evaluate the novelty of the innovation seeking patent protection, patent applicants are required to disclose the prior knowledge on which they relied,[24]represented by citations to prior patents. Citations are proposed by the patent applicant, and subsequently checked and corrected by the patent examiner.[25,26]Because of this, patents’ citation count is the most used metric for patents’ technological impact.[23,27]Despite its wide use, citation count suffers from a serious age bias by favoring older patents over more recent ones.[28–30]The main reason is that it takes time for a patent to accumulate citations, and the preferential attachment (also known as the “rich-get-richer”mechanism)[31]further amplifies discrepancies in the citation count of patents issued at different times.[32,33]The age bias makes it unreliable to directly compare the citation count of patents issued in different years.[34]To overcome this issue,following Ref. [14], we measure a given patent’s technological impact in terms of its normalized citation count (NC),which corresponds to its ranking by citation count when compared against all patents issued in the same year. In this way,NC ∈[0,1)close to one(zero)corresponds to high(low)impact. Normalized citation count can effectively suppress the age bias held by citation count,[14]thus allowing us to compare on the same scale the impact of patents of different ages.

    Fig.2. An illustration of two firms’lifecycles. Each dot represents a patent,and the x-axis refers to the issuance year of patents,y-axis refers to the technological impact(NC)of each patent,red dots denote breakthrough patents(patents with the highest NC). (a) Firm CANON’s lifecycle from the EPO data. (b)Firm APPLE’s lifecycle from the USPTO data.

    Firm’s technological impact and classification. Motivated by previous works,[14,35,36]a firm’s lifecycle is defined as a time-ordered sequence of its issued patents (see Fig. 2).And we define a given firm’s technological impact(TI)as the impact of its highest-impact patent, which we refer to as the firm’s “breakthrough”. Based on the technological impact of all analyzed firms,we classify firms into three groups:we consider the top 10%firms as high-impact firms,the bottom 30%firms as low-impact firms, and the remaining 60% firms as medium-impact firms, see Figs.1(b)and 1(c)for the classification details.

    2.3. Modeling impact dynamics

    To describe the temporal dynamics of the technological impact in firms’ patenting lifecycles, we resort to three models to reproduce patents’impact.

    State-of-the-art models and their shortcomings. We consider two models from previous works:[1]the randomimpact model and theQ-model. The random-impact model assumes that the impact of each of a firm’s issued patents is a random variable. Specifically,the model keeps a given firm’s total patent number unchanged, and it extracts the impact of each of the firm’s patents from the impact values of patents issued in that issuance year(without replacement). According to the random-impact model,the only difference between two firms is their patent productivity: a productive firm is more likely to have a breakthrough patent with high technological impact merely because of a larger number of attempts with the same likelihood to have high impact.

    TheQmodel assumes that each firm has an capability valueQ,which is unique to the firm and stable throughout the firm’s lifecycle.Qcaptures the capability of firms to improve(Q>1)or reduce(Q<1)the impact of a patent. Specifically,in the model, a firmαrandomly chooses a patent ideaiwith potential impactri, and it modulates the idea according to itsQ,resulting in a patent of impact,

    whereriis a patent-dependent random factor which is lognormal distributed with mean 1,identical for all firms.

    The above two models were both motivated by the empirical findings on scientists’careers,[1]in which the highestimpact works are randomly distributed among individuals’careers. For firms, the validity of the two models has not been examined.

    Time-varying returns model. In theQmodel, a firm’s early and late patents have the same probability to be the breakthrough patents, and the potential impact(r)of a patent idea itself and the firm’s capability (Q) are the only two factors that determine the impact of a patent,independently of the patent’s order within the applicant firm’s lifecycle. However,in management literature,many studies[12,13,37]have also uncovered the role of age played in firms’technological quality.These works argued that as firms become more experienced,the cost and benefits brought by innovation change,which suggests that firms may receive time-varying returns from their patents depending on the stage in their lifecycles. Inspired by these findings, based on theQmodel, we propose a timevarying returns model which assumes that a firm will enhance or diminish the potential impact of a patent based on its inherent ability combined with its age,quantified by its number of previously issued patents. In formulas, the impact of patentifrom firmαis modeled as

    whereΦis a time-varying function representing the influence of firm age on patent impact,ni ∈{1,...,Nα}is the position of patentiwithin the time-ordered sequence of firmα’s patents,andNαis the total patent number issued to firmα. In the next section,we will show that a monotonously decreasing function exhibits an accurate fit to the empirical data,which we refer to as a diminishing returns effect.

    3. Results

    3.1. Productivity and impact dynamics in firms’patenting lifecycles

    Fig. 3. Temporal dynamics of patent productivity and technological impact during firms’ lifecycles. The panels show the temporal trends of the total number of issued patents(a),(b)and cumulative technological impact(c),(d)for the three different groups of firms. Only firms with at least 20-year lifetime are involved in this analysis. Error bars refer to the standard error of the mean. The results show that there are sustained differences across the three impact groups of firms,and the growing trends of productivity and impact are well-described by sigmoidal and power-law functions,respectively,for both datasets. The fitting parameters are shown along the curves.

    3.2. Timing pattern of the firms’breakthrough patents

    We focus here on the recent finding uncovered by the recent study[14]that breakthrough patents (highest-impact patents)tend to occur early within a firm’s patenting lifecycle.This finding is revealed by the ratioN?/Nbetween the relative position of a firm’breakthrough patent,N?, and the total number of patents issued to the firm,N; a low (high) value of this ratio indicates that the breakthrough tends to be among the early(late)patents by the firm.[14]The results on the EPO and USPTO data are shown in Fig. 4. We find that the probability that a firm is issued its breakthrough patent is indeed not constant over time. A firm’s technological breakthrough is more likely to be among the firm’s earliest patents, regardless of firm’s group of impact,represented by an early peak ofP(N?/N). This is true for both the EPO and USPTO datasets;the same is not true for reshuffled lifecycles constructed from the original data by randomizing the temporal order of each firm’s patents (see the light lines in Fig. 4). These findings are consistent with the previous findings by Xuet al..[14]In addition, Fig. 4 reports that the peak for high-impact firms is slightly higher than that for the medium and low-impact firms.While through statistical tests,we find only the difference between the high-impact and low-impact firms in the USPTO data is significant,which demonstrates that the timing pattern of top-impact firms’breakthrough patents is similar to that of ordinary firms’ breakthroughs. So in the following, we will construct a unified model for all analyzed firms. In Table 3,we further provide specific impact values(TI)and relative positions of breakthrough patents(N?/N)for a few high-impact firms that are involved in both the EPO and USPTO data. For some of them,such as MOTOROLA MOBILITY and INTERNATIONAL BUSINESS MACHS, the breakthrough patents in the two datasets exhibit similar temporal positions. While there also exist firms that demonstrate gaps in terms of the timing of their breakthrough patents in the EPO and USPTO data,such as EASTMAN KODAK and EXXON MOBIL.This difference confirms firms’heterogeneity in terms of patenting activities in different patent offices, under which the uniform early timing has more credibility.

    The observed early peak ofP(N?/N) can find explanations from previous management literature which claimed that young firms tend to present a higher probability of innovation than mature ones.[12,13]The underlying reason could be that as firms age, benefits from technological progress diminish,[13]and therefore, patents from old firms are more likely to be the extension and improvement of their established technologies,[39]which tend to hold relatively less impact compared with those from new entrants.The result proves that patent impact is correlated with the patent’s temporal order among the applicant firm’s lifecycle, and provides initial evidence in favor of a time-varying mechanism,i.e.,firms will obtain varying returns from patents at different stages in their lifecycles. This is radically different from the random-impact phenomenon in scientific and artistic careers, for which the return is stable.

    Fig.4. Distribution of the temporal order of firms’highest-impact patents.We show the probability distribution of the relative order(N?/N)of firms’technological breakthroughs, and the same distribution for randomized patenting lifecycles (denoted by the light lines) for high-, medium-, and low-impact firms, for the EPO and USPTO data, respectively. The results represent the average of 200 times statistics(to take the“paralleling”breakthroughs with the same impact value into consideration) and randomized firms’ lifecycles. The early-stage peak of the distributions for all three groups of firms indicates that firms are more likely to achieve their highestimpact patent at the beginning of their patenting activity, contradicting the random-impact rule.

    3.3. Dynamical model of firms’technological impact

    To further understand and model the dynamics of firms’impact, we then analyze the average impact of patents issued to firms in different phases of lifecycles. In Fig. 5, we show patents’ average impact as a function of their temporal orderwithin firms’ lifecycles. We find an approximately linear decreasing trend, which indicates that firms tend to obtain diminishing returns from patents as they are issued more patents.Based on this finding,in the time-varying returns model shown in Eq.(2),we assume a linearly-decreasing function to model the phenomenon of the diminishing returns

    Table 3. The technological impact and timing to achieve the breakthrough patents of several high-impact firms. Here we select firms that have records in both the EPO and USPTO data.

    whereβandγare parameters to be estimated based on the empirical data. Whenβ=1,γ=0(no-varying scenario),the time-varying returns model reduces to theQ-model.

    Fig.5.Average impact of patents as a function of their temporal order within the applicant firms’lifecycle. We divide each firm’s issued patents into 20 equally-sized groups according to the patents’issuance date. By aggregating all the analyzed firms,each dot represents the average impact of firms’patents that belong to the corresponding group. For example, the first dot from the left represents the average impact of firms’earliest 5%patents. Error bars refer to the standard errors of the mean. The results show that for both the two datasets, there is a linearly decreasing trend of patent impact as the patent order increases(via linear fit),which suggests a monotonously decreasing function for the time-varying returns model.

    Assuming the proposed model, we firstly sum over the impact values of a firm’s all patents,and empirically estimate each firm’sQbased on Eq. (2) through the approximate formula

    whereIαis the total patent impact of firmα,Pαdenotes the set of patents issued to firmα. Subsequently, we generate each patent’s simulated impact according to Eq.(2),and compare the simulated temporal pattern of firms’ breakthrough patents(i.e.,the probability distribution ofN?/N)with the corresponding pattern in the empirical data.

    To calibrate the model parameters to the data, we apply a maximum likelihood approach. Specifically, we consider the Jensen Shannon divergence (DJS)[40]to measure the statistical “distance” between the observed temporal distribution,Pdata(N?/N), and the model-generated distribution,Pmodel(N?/N), which is the symmetric version of the Kullback–Leibler divergence.[1,41]It is defined as

    whereM=(Pdata+Pmodel)/2 is the mixed distribution. Then the likelihood function is defined as the fraction of modelgenerated distributions that are sufficiently close to the empirical one among 100 times simulations, i.e., the fraction of the simulated distributions such thatDJS<0.002 for the EPO data andDJS<0.005 for the USPTO data,in which the thresholds are chosen based on theDJSresults in each dataset. The threshold is a little larger concerning the USPTO data than the EPO data considering theDJSresults are overall larger for the former. The value of the likelihood function ranges from 0(no simulation fulfills the threshold) to 1 (all of the simulations fulfill the threshold). We choose the optimal parameter pair(?β,?γ)that maximizes the likelihood to generate a model whose temporal pattern is sufficiently close to the empirical one. The results are shown in Figs.6(a)and 6(b),in which the color of each box refers to the value of the likelihood function.Figures 6(a) and 6(b) demonstrate that the optimal combination of parameters(denoted by the box with green contour)is ?β=0.9, ?γ=0.06 for the EPO data, and ?β=0.9, ?γ=0.08 for the USPTO data, which are similar to the slope of the fitted line in Fig. 5. We also observe that theQmodel corresponding to the box with pink contour achieves 0 in terms of the likelihood on both datasets,namely,none of the 100 times simulations fulfills the related thresholds. This reveals that to accurately model firms’ impact dynamics, the role of the diminishing returns effect cannot be ignored.

    Fig. 6. Comparison of the empirical and model-based temporal patterns of firms’ breakthrough patents. (a), (b) The likelihood that the timing distribution of breakthrough patents(P(N?/N))generated by the time-varying returns model is sufficiently close to the empirical one(see Subsection 3.3 for details)among 100 times simulations for different parameter combinations. The parameter pairs that lead to the maximum likelihood are ?β =0.9, ?γ =0.06 for the EPO data (a), and ?β =0.9, ?γ =0.08 for the USPTO data (b), which are corresponding to the boxes with green contour. The Q model(corresponding to the box denoted by the pink contour)achieves the lowest likelihood. (c),(d)A comparison of the empirical and model-based distributions P(≤N?/N)of the breakthrough patents’temporal positions of firms. We consider two state-of-art models and the proposed time-varying returns model,which uses the optimal parameter pairs as mentioned above. The results indicate that the empirical results in the two datasets are well reproduced by the time-varying returns model.

    Figures 6(c) and 6(d) show the fitting performance of the proposed model with the optimal parameters against the random-impact model and theQmodel for the EPO and USPTO datasets. The result reveals that the proposed timevarying returns model can match the empirical curve much more accurately. While the random-impact model andQmodel assume that a patent’s impact does not depend on its temporal position among its applicant firm’s patents,thus fail to generate the concave timing distribution of breakthrough patents observed in the empirical data. The results further confirm the discrepancy of the timing pattern of the highestimpact works for firms and individuals(such as scientists and artists[1,6,8]). Besides, we also notice that although our proposed model outperforms the random-impact model andQmodel by a wide margin, it cannot fit the empirical results perfectly, especially for the USPTO data. This could be due to the heterogeneity in firms’impact dynamics and model parameters,which will be validated in future works.

    4. Conclusion and discussion

    In this work, we modeled the evolution mechanism of firms’ technological impact within lifecycles. The USPTO and EPO patenting history data of applicant firms on a long timescale enable us to quantitatively trace and analyze the dynamic patterns of diverse firms worldwide. By treating each firm as a time-ordered sequence of its issued patents,we quantified the technological impact of their patents through a timenormalized citation measure, based on which we studied the dynamics of firms’ patent productivity, technological impact,and the timing pattern of the highest-impact patents. There are three main takeaways from this work.Firstly,we provided evidence that firms tend to have their technological breakthroughs at the early stage within their lifecycles, which contradicts the random-impact rule that governing individual careers,and confirms previous findings.[14]Secondly,we found that the decrease of firms’average patent impact along with the patenting lifecycles is accurately matched by a linear function. Thirdly,most importantly, inspired by these findings, we proposed a time-varying returns model, which is proved to capture the temporal distributions of firms’ technological breakthroughs more accurately than two state-of-art dynamical models.

    A potential limitation of our approach is that our analysis assumes that all firms follow a uniform pattern. Yet different firms might have different trajectories, and the “average” behavior might not hold for all of them. Thus, our results could be not informative for specific types of firms. In future research,we will explore multiple patterns held by distinct types and industries of firms. Identifying drivers of the temporal position of firms’technological breakthroughs is also worth further research: The proposed model assumes that the time-varying functionΦis given, which leads to the important problem of identifying its determinants at the firm level.Besides,incorporating our insights into real-world investment and policy decisions would need careful integration with a variety of signals beyond patent analysis.Addressing these questions will help us identify the best path for firm development and further catch a glimpse of the secret of business success.

    To conclude,this paper enriches the understanding of firm impact dynamics.It provides the first step towards new models to describe the observed,firm-specific empirical patterns. The results can offer insights into technological innovation management for firms.

    Acknowledgment

    We thank Luciano Pietronero, Emanuele Pugliese, and Andrea Zaccaria for providing us with the EPO patent data and enlightening discussions.

    Appendix A: Filtering firm applicants in the

    EPO data

    As the research target in this paper is firm and not all applicants involved in the EPO data are firms,we remove some of the patent data that can be sure not assigned to firms. In detail,we identify four kinds of applicants based on their names and the information of their patents’inventors according to the following rules. (i) Firm applicants. If an applicant has the firm-related strings (refer to the first row in Table A1) in its name,this applicant is identified as a firm. (ii)Individual applicants. For each patent issued to a given applicant,if all its inventors also appear in the assignee field,and the applicant’s name does not contain any firm-related string,this applicant is identified as an individual applicant.[42](iii)Educational applicants. If the applicant does not have any firm-related string in its name but has the education-related strings (refer to the second row in Table A1), it is considered as an educational institution. (iv)Other applicants. If none of the above conditions is satisfied,the applicant belongs to this set. To exclude the patent data that is irrelevant to firms,we remove the patents issued to individual applicants and educational applicants,and leave the remaining data for the study. In total, we exclude 15.5%applicants and 4.3%patents,note that few of such applicants have larger than 20 patents, so the filtering has little influence on our analysis.

    Table A1. The identification strings in applicants’names that we apply to filter applicants in the EPO data.

    Fig.B1. Distribution of the parameter values(b and η)in the fitting of productivity and impact dynamics(see Fig.3)for firms in the(a),(c)EPO and(b),(d)USPTO data.

    As for the USPTO data,in view of Koganet al.[18,19]have made the extraction based on the Center for Research in Security Prices(CRSP)database, we do not implement any applicant filter on this data.

    Appendix B: Distribution of the fitting parameters in Fig.3

    Based on Fig. 3, we further show the distribution of the fitting parameters of the productivity growth and impact growth for the three groups of firms in Fig. B1. The results reveal that in terms of productivity, the development of high-impact firms (〈bhigh〉=0.33 in EPO,〈bhigh〉=0.24 in USPTO)is faster than that of low-impact firms(〈blow〉=0.36 in EPO,〈blow〉= 0.31 in USPTO). In terms of technological impact, for low-impact firms, the growth is steady, with〈ηlow〉=0.88 in EPO, and〈ηlow〉=0.82 in USPTO.The increase of other firms is much faster, for whom the averageηis bigger,see Fig.B1.

    Appendix C: Robustness check of the main results

    In the main text, we consider firms that have been issued at least a given number of patents. To guarantee the robustness of our main results, we further add analysis under a different filter criterion for firms in both datasets, i.e.,the analyzed firms need to have long-lasting active patenting activities.[1,5,14]Specifically, for EPO data, we only consider firms that have at least 10-year patenting activity and have been issued to at least one patent every two years,which ends up with 1495 firms; for USPTO data, firms need to have at least 15-year patenting activity and have been issued to at least one patent every three years,involving 1007 firms. The main results under the new filter criterion are shown in Fig. C1,which demonstrate the robustness of our results.

    Fig. C1. Robustness check of the main results, in which we restrict firms’ patenting activities to be long-lasting and active. The results are consistent with the conclusions in the main text. (a), (b) Distribution of the temporal order of firms’ highest-impact patents, which is similar to Fig. 4. (c), (d) Comparison of the empirical and model-based temporal patterns of firms’ breakthrough patents, which is similar to Fig. 6. The optimal parameters for the EPO data are ?β =0.8, ?γ =0.06,and for the USPTO data,they are ?β =0.9, ?γ =0.08.

    中文字幕人成人乱码亚洲影| 久久久久久人人人人人| 观看免费一级毛片| 美女 人体艺术 gogo| 嫁个100分男人电影在线观看| 老汉色av国产亚洲站长工具| 18禁黄网站禁片免费观看直播| 国产99白浆流出| 又黄又爽又免费观看的视频| 成人18禁高潮啪啪吃奶动态图| 热99re8久久精品国产| 听说在线观看完整版免费高清| 两个人免费观看高清视频| 无遮挡黄片免费观看| 12—13女人毛片做爰片一| 久久精品国产综合久久久| 国产v大片淫在线免费观看| 国产精华一区二区三区| 日本免费a在线| 国产精品日韩av在线免费观看| 欧美激情 高清一区二区三区| 在线观看免费日韩欧美大片| 长腿黑丝高跟| 亚洲国产精品sss在线观看| 亚洲精品国产区一区二| 国产成人精品久久二区二区免费| 婷婷丁香在线五月| 国产视频内射| 成人手机av| 日韩av在线大香蕉| 久久久国产精品麻豆| av欧美777| 韩国精品一区二区三区| 成人特级黄色片久久久久久久| 亚洲av片天天在线观看| 欧美最黄视频在线播放免费| 亚洲欧美精品综合久久99| 日韩有码中文字幕| 99国产精品一区二区蜜桃av| 久久久久久大精品| 91麻豆精品激情在线观看国产| 午夜免费成人在线视频| 欧美一区二区精品小视频在线| 在线十欧美十亚洲十日本专区| 一级毛片女人18水好多| 亚洲欧洲精品一区二区精品久久久| 狂野欧美激情性xxxx| av有码第一页| 欧美av亚洲av综合av国产av| 亚洲精品在线美女| 9191精品国产免费久久| 成人18禁在线播放| 亚洲欧美精品综合久久99| 一级作爱视频免费观看| 国产成人精品久久二区二区91| 亚洲精品粉嫩美女一区| 日韩精品免费视频一区二区三区| 午夜免费成人在线视频| 亚洲av成人不卡在线观看播放网| 亚洲一区二区三区色噜噜| 亚洲国产高清在线一区二区三 | 美女扒开内裤让男人捅视频| 久久久久久亚洲精品国产蜜桃av| 亚洲色图 男人天堂 中文字幕| 亚洲精品中文字幕一二三四区| 国产91精品成人一区二区三区| 日韩 欧美 亚洲 中文字幕| 色哟哟哟哟哟哟| 国产亚洲精品一区二区www| 欧美久久黑人一区二区| 性欧美人与动物交配| 久久这里只有精品19| 久99久视频精品免费| 青草久久国产| 久久精品成人免费网站| 免费搜索国产男女视频| 久久久精品欧美日韩精品| 十八禁网站免费在线| 一进一出好大好爽视频| 国产私拍福利视频在线观看| 国产91精品成人一区二区三区| 国产精品久久久人人做人人爽| av中文乱码字幕在线| 精品一区二区三区视频在线观看免费| 精品国产国语对白av| 国产1区2区3区精品| 观看免费一级毛片| 久久精品国产99精品国产亚洲性色| 人人妻人人澡人人看| 十八禁人妻一区二区| 少妇裸体淫交视频免费看高清 | 白带黄色成豆腐渣| 一级片免费观看大全| 美女免费视频网站| 97碰自拍视频| 无限看片的www在线观看| 性欧美人与动物交配| 在线播放国产精品三级| 一级a爱片免费观看的视频| 久久久久精品国产欧美久久久| 国产黄a三级三级三级人| 老熟妇仑乱视频hdxx| 国产99白浆流出| 成人av一区二区三区在线看| 97碰自拍视频| 人成视频在线观看免费观看| 搞女人的毛片| 麻豆一二三区av精品| 日日爽夜夜爽网站| 日韩大尺度精品在线看网址| 一区二区三区国产精品乱码| 国产激情欧美一区二区| 黄色丝袜av网址大全| 久久久久精品国产欧美久久久| 91老司机精品| 一级毛片精品| 国产精品电影一区二区三区| 高潮久久久久久久久久久不卡| 黄片播放在线免费| 人人澡人人妻人| 日本一本二区三区精品| 在线国产一区二区在线| 一本大道久久a久久精品| www日本在线高清视频| av电影中文网址| 欧美绝顶高潮抽搐喷水| 欧美日韩瑟瑟在线播放| 欧美国产精品va在线观看不卡| 在线十欧美十亚洲十日本专区| 激情在线观看视频在线高清| 久热爱精品视频在线9| 中文亚洲av片在线观看爽| 在线观看www视频免费| 嫩草影视91久久| 天天一区二区日本电影三级| netflix在线观看网站| 可以在线观看毛片的网站| 成人精品一区二区免费| 国产熟女xx| 搞女人的毛片| 777久久人妻少妇嫩草av网站| 麻豆成人av在线观看| 日本成人三级电影网站| 悠悠久久av| 中文字幕人成人乱码亚洲影| 国产一区在线观看成人免费| 久热这里只有精品99| 午夜福利一区二区在线看| 亚洲熟妇中文字幕五十中出| av在线天堂中文字幕| 久久中文字幕一级| 日本 欧美在线| 欧美成人一区二区免费高清观看 | 日日摸夜夜添夜夜添小说| www.熟女人妻精品国产| 免费在线观看日本一区| 搡老熟女国产l中国老女人| 国产精品一区二区精品视频观看| 国产精品亚洲一级av第二区| 麻豆久久精品国产亚洲av| 村上凉子中文字幕在线| 欧美av亚洲av综合av国产av| 欧美av亚洲av综合av国产av| 宅男免费午夜| 亚洲精品色激情综合| 久久中文字幕人妻熟女| 免费看美女性在线毛片视频| 18美女黄网站色大片免费观看| 国产精华一区二区三区| 免费在线观看亚洲国产| 亚洲免费av在线视频| 淫妇啪啪啪对白视频| 欧美国产精品va在线观看不卡| 在线观看66精品国产| 12—13女人毛片做爰片一| 亚洲精品在线观看二区| 国内揄拍国产精品人妻在线 | 亚洲av第一区精品v没综合| 18美女黄网站色大片免费观看| 91国产中文字幕| 两人在一起打扑克的视频| 国产亚洲欧美精品永久| √禁漫天堂资源中文www| 热re99久久国产66热| 在线观看66精品国产| 在线观看66精品国产| 亚洲 欧美 日韩 在线 免费| 日韩免费av在线播放| 免费看a级黄色片| 亚洲,欧美精品.| 老汉色av国产亚洲站长工具| 欧美激情高清一区二区三区| 欧美成人一区二区免费高清观看 | 午夜免费鲁丝| 一区二区三区国产精品乱码| 变态另类成人亚洲欧美熟女| www.自偷自拍.com| 午夜免费观看网址| 午夜福利免费观看在线| 老鸭窝网址在线观看| 亚洲五月婷婷丁香| 免费在线观看黄色视频的| www国产在线视频色| 色综合亚洲欧美另类图片| 亚洲美女黄片视频| 一区二区日韩欧美中文字幕| 亚洲熟妇熟女久久| 亚洲一码二码三码区别大吗| 国产av又大| 国产成人欧美在线观看| 婷婷丁香在线五月| 国产精品一区二区免费欧美| svipshipincom国产片| 精品久久久久久久久久久久久 | 一本综合久久免费| 一边摸一边做爽爽视频免费| 亚洲全国av大片| 欧美黑人精品巨大| 日韩三级视频一区二区三区| 国产成人精品久久二区二区91| 亚洲人成伊人成综合网2020| 日本一本二区三区精品| 一区二区三区国产精品乱码| 亚洲人成网站在线播放欧美日韩| 91字幕亚洲| 国产一卡二卡三卡精品| 最近最新中文字幕大全免费视频| av福利片在线| 黄色丝袜av网址大全| 久久久久免费精品人妻一区二区 | 在线视频色国产色| 欧美黄色片欧美黄色片| 成人午夜高清在线视频 | xxx96com| 男女那种视频在线观看| 波多野结衣巨乳人妻| 麻豆成人av在线观看| 美女免费视频网站| 久久久久久久午夜电影| 又大又爽又粗| 女人被狂操c到高潮| 精品久久久久久久久久久久久 | 国产成+人综合+亚洲专区| 欧美日韩精品网址| 久久欧美精品欧美久久欧美| 在线免费观看的www视频| 欧美日韩乱码在线| 欧美日韩福利视频一区二区| 午夜久久久久精精品| 欧美乱码精品一区二区三区| 精品少妇一区二区三区视频日本电影| 亚洲自拍偷在线| 免费av毛片视频| 久久人妻福利社区极品人妻图片| 欧美人与性动交α欧美精品济南到| 精品久久久久久成人av| 脱女人内裤的视频| 欧美日韩瑟瑟在线播放| 日韩欧美 国产精品| 国产精品久久久久久精品电影 | 看片在线看免费视频| 日本成人三级电影网站| 天天躁狠狠躁夜夜躁狠狠躁| 亚洲男人天堂网一区| 国产高清有码在线观看视频 | 国产精品香港三级国产av潘金莲| 久久久久久大精品| 久久久国产成人免费| 欧美最黄视频在线播放免费| 日韩国内少妇激情av| 欧美一区二区精品小视频在线| 国产精品久久久久久精品电影 | 欧美最黄视频在线播放免费| 少妇 在线观看| 国产又色又爽无遮挡免费看| 国产精品日韩av在线免费观看| 国产成人影院久久av| 老司机午夜福利在线观看视频| 两个人看的免费小视频| 国产乱人伦免费视频| 午夜福利欧美成人| 99在线人妻在线中文字幕| 一级作爱视频免费观看| 九色国产91popny在线| 亚洲人成网站在线播放欧美日韩| 久久亚洲精品不卡| www.999成人在线观看| 亚洲av第一区精品v没综合| 久9热在线精品视频| 日韩大尺度精品在线看网址| 久久国产精品影院| 久热这里只有精品99| 一级片免费观看大全| 夜夜夜夜夜久久久久| 91九色精品人成在线观看| 亚洲av成人不卡在线观看播放网| 亚洲国产欧洲综合997久久, | 亚洲色图av天堂| 婷婷精品国产亚洲av在线| 国产黄a三级三级三级人| 青草久久国产| 天堂影院成人在线观看| 亚洲中文字幕日韩| 他把我摸到了高潮在线观看| 两个人视频免费观看高清| 亚洲欧美精品综合久久99| 女性被躁到高潮视频| 国内揄拍国产精品人妻在线 | 午夜精品久久久久久毛片777| 国内久久婷婷六月综合欲色啪| 老司机深夜福利视频在线观看| 美女大奶头视频| 欧美日韩乱码在线| 日本撒尿小便嘘嘘汇集6| 极品教师在线免费播放| 十八禁人妻一区二区| 校园春色视频在线观看| 欧美 亚洲 国产 日韩一| 亚洲国产精品999在线| 国产精品自产拍在线观看55亚洲| 99热6这里只有精品| 日日摸夜夜添夜夜添小说| 亚洲五月色婷婷综合| 巨乳人妻的诱惑在线观看| 最近最新中文字幕大全免费视频| 久久久精品国产亚洲av高清涩受| 日日摸夜夜添夜夜添小说| 一区二区三区高清视频在线| 深夜精品福利| 欧美色视频一区免费| 免费观看精品视频网站| 久久婷婷成人综合色麻豆| 欧美激情 高清一区二区三区| 最近最新免费中文字幕在线| 一卡2卡三卡四卡精品乱码亚洲| 757午夜福利合集在线观看| 国产精品久久久av美女十八| 欧美日韩乱码在线| 色av中文字幕| 亚洲中文字幕一区二区三区有码在线看 | 成人永久免费在线观看视频| 欧美日本亚洲视频在线播放| 成人午夜高清在线视频 | 最好的美女福利视频网| 51午夜福利影视在线观看| 成人亚洲精品av一区二区| 欧美成人午夜精品| 亚洲国产精品sss在线观看| 大香蕉久久成人网| 黑人欧美特级aaaaaa片| 亚洲一区二区三区色噜噜| 亚洲精品美女久久av网站| 久久久精品国产亚洲av高清涩受| 9191精品国产免费久久| 18禁观看日本| 99国产综合亚洲精品| 国产v大片淫在线免费观看| 色婷婷久久久亚洲欧美| 国产av一区二区精品久久| 亚洲成av人片免费观看| 中文字幕久久专区| 久久精品人妻少妇| 国产一区二区三区在线臀色熟女| 久久久国产成人精品二区| 成年版毛片免费区| 国产又色又爽无遮挡免费看| cao死你这个sao货| 夜夜躁狠狠躁天天躁| 老熟妇仑乱视频hdxx| 色综合站精品国产| 在线看三级毛片| 岛国在线观看网站| 日韩欧美三级三区| 中文字幕精品亚洲无线码一区 | 999久久久精品免费观看国产| 99在线人妻在线中文字幕| 最新在线观看一区二区三区| 亚洲色图av天堂| 亚洲,欧美精品.| 99久久国产精品久久久| 窝窝影院91人妻| 国产三级在线视频| 一级片免费观看大全| 国产精品电影一区二区三区| 波多野结衣高清无吗| 亚洲国产精品成人综合色| 日本一本二区三区精品| 午夜福利在线在线| 日本a在线网址| 成人18禁高潮啪啪吃奶动态图| 伊人久久大香线蕉亚洲五| 香蕉丝袜av| 中文字幕最新亚洲高清| 国产av在哪里看| 啪啪无遮挡十八禁网站| 国产又黄又爽又无遮挡在线| 欧美亚洲日本最大视频资源| a级毛片a级免费在线| 成人三级做爰电影| 国产视频一区二区在线看| 哪里可以看免费的av片| 99热6这里只有精品| 女同久久另类99精品国产91| 长腿黑丝高跟| 很黄的视频免费| 国产乱人伦免费视频| 99热只有精品国产| 两人在一起打扑克的视频| 日韩国内少妇激情av| 亚洲男人天堂网一区| av片东京热男人的天堂| av免费在线观看网站| 亚洲久久久国产精品| 国产日本99.免费观看| 欧美日韩精品网址| 亚洲熟妇熟女久久| 国产亚洲欧美精品永久| 一进一出抽搐gif免费好疼| 最近最新中文字幕大全免费视频| 日本在线视频免费播放| 久久久久久九九精品二区国产 | 久久久精品欧美日韩精品| 高清在线国产一区| 国产av在哪里看| 日韩高清综合在线| 波多野结衣巨乳人妻| 黄色女人牲交| 很黄的视频免费| 一边摸一边做爽爽视频免费| 国产精品爽爽va在线观看网站 | 国产伦在线观看视频一区| 亚洲片人在线观看| 久99久视频精品免费| 成年版毛片免费区| 欧美性猛交黑人性爽| 欧美大码av| 国产三级在线视频| 大香蕉久久成人网| 搡老熟女国产l中国老女人| 久久天堂一区二区三区四区| 国产麻豆成人av免费视频| 亚洲专区国产一区二区| 精品国产乱码久久久久久男人| 欧美久久黑人一区二区| 久久久久久九九精品二区国产 | 午夜免费成人在线视频| 久久亚洲真实| 欧美乱码精品一区二区三区| 免费电影在线观看免费观看| 亚洲成人久久爱视频| 侵犯人妻中文字幕一二三四区| 国产欧美日韩精品亚洲av| 美女扒开内裤让男人捅视频| 亚洲三区欧美一区| 免费观看人在逋| 91老司机精品| 亚洲国产看品久久| 亚洲成人久久爱视频| 欧美日韩精品网址| 国产精品av久久久久免费| 午夜免费鲁丝| 久热这里只有精品99| 欧美色欧美亚洲另类二区| 日韩中文字幕欧美一区二区| 一夜夜www| 看片在线看免费视频| 少妇被粗大的猛进出69影院| 亚洲免费av在线视频| 999精品在线视频| 69av精品久久久久久| 国产1区2区3区精品| 国产真人三级小视频在线观看| 国产极品粉嫩免费观看在线| 一二三四在线观看免费中文在| 成人亚洲精品一区在线观看| 91大片在线观看| 亚洲国产精品sss在线观看| 亚洲一卡2卡3卡4卡5卡精品中文| 在线观看66精品国产| 中文字幕另类日韩欧美亚洲嫩草| 国产激情欧美一区二区| 日韩成人在线观看一区二区三区| 午夜亚洲福利在线播放| 中文字幕av电影在线播放| 久久精品国产清高在天天线| 国产亚洲欧美98| 日韩免费av在线播放| 国产色视频综合| 777久久人妻少妇嫩草av网站| 国产精品av久久久久免费| 中文字幕久久专区| 久久亚洲精品不卡| 国产视频内射| 老熟妇仑乱视频hdxx| 高清毛片免费观看视频网站| 国产精品国产高清国产av| 亚洲av日韩精品久久久久久密| 黄色a级毛片大全视频| 精品国产一区二区三区四区第35| 国产极品粉嫩免费观看在线| 中文亚洲av片在线观看爽| 亚洲一码二码三码区别大吗| 91在线观看av| 国产三级黄色录像| 免费观看精品视频网站| 宅男免费午夜| 成人国语在线视频| 成人av一区二区三区在线看| 男人的好看免费观看在线视频 | 日日摸夜夜添夜夜添小说| 午夜a级毛片| 中文字幕人妻熟女乱码| 一本大道久久a久久精品| 草草在线视频免费看| 黄片大片在线免费观看| 欧美一级a爱片免费观看看 | 国产亚洲精品av在线| 人人妻人人澡欧美一区二区| 亚洲七黄色美女视频| 可以在线观看毛片的网站| 欧美激情 高清一区二区三区| 一卡2卡三卡四卡精品乱码亚洲| 精品人妻1区二区| 人人妻人人澡人人看| 国产爱豆传媒在线观看 | 99在线视频只有这里精品首页| 一区二区三区激情视频| 这个男人来自地球电影免费观看| 又黄又爽又免费观看的视频| 黄色视频,在线免费观看| 久久久久久久久中文| 视频区欧美日本亚洲| 可以在线观看毛片的网站| 国产av不卡久久| 中文字幕久久专区| 日韩欧美三级三区| 国产av一区在线观看免费| 久热这里只有精品99| √禁漫天堂资源中文www| www日本黄色视频网| 久久草成人影院| 精品久久久久久,| 欧美丝袜亚洲另类 | 久久亚洲精品不卡| 欧美日韩亚洲国产一区二区在线观看| 亚洲真实伦在线观看| 国产成人一区二区三区免费视频网站| 亚洲全国av大片| 国产精品美女特级片免费视频播放器 | 久久精品aⅴ一区二区三区四区| 欧美日韩亚洲国产一区二区在线观看| 久久国产乱子伦精品免费另类| 久久久国产成人精品二区| 国产亚洲精品第一综合不卡| 国产麻豆成人av免费视频| 日韩免费av在线播放| av福利片在线| 男人舔女人的私密视频| 99在线视频只有这里精品首页| 亚洲国产欧美日韩在线播放| 给我免费播放毛片高清在线观看| 麻豆成人av在线观看| 精品久久久久久久末码| 国产精品98久久久久久宅男小说| 亚洲九九香蕉| 亚洲国产欧美网| 午夜福利免费观看在线| 亚洲av成人一区二区三| av有码第一页| 成人精品一区二区免费| 淫秽高清视频在线观看| 韩国av一区二区三区四区| 中文字幕另类日韩欧美亚洲嫩草| 看黄色毛片网站| 叶爱在线成人免费视频播放| 无限看片的www在线观看| 国产亚洲精品久久久久5区| 99国产精品一区二区三区| 国产精品久久久久久人妻精品电影| 色老头精品视频在线观看| 午夜亚洲福利在线播放| 欧美激情久久久久久爽电影| 欧美在线一区亚洲| 免费无遮挡裸体视频| 一级片免费观看大全| 在线观看66精品国产| 一区二区三区精品91| 丝袜在线中文字幕| 99在线视频只有这里精品首页| 色哟哟哟哟哟哟| 美女 人体艺术 gogo| 91成人精品电影| 一卡2卡三卡四卡精品乱码亚洲| 国产真人三级小视频在线观看| 精品久久久久久成人av| 在线永久观看黄色视频| 欧美久久黑人一区二区| 十八禁网站免费在线| 午夜免费鲁丝| 女生性感内裤真人,穿戴方法视频| 看黄色毛片网站| 啦啦啦 在线观看视频| 久久 成人 亚洲| 欧美日韩亚洲综合一区二区三区_| 老司机深夜福利视频在线观看| 老鸭窝网址在线观看| 亚洲熟女毛片儿| 久久九九热精品免费| 亚洲专区中文字幕在线| 在线av久久热| 日韩欧美一区二区三区在线观看| 国产成人av激情在线播放| 亚洲av日韩精品久久久久久密| av电影中文网址| 成人午夜高清在线视频 |