R&D Subsidies and Incremental Innovation Lock-in:A Machine Learning-based Patent Text Analysis
Government R&D subsidies may primarily encourage incremental innovation rather than promote disruptive breakthroughs,which might be beneficial in explaining the long-standing"innovation paradox"in China.To explore this issue,we use machine learning-based text-mining techniques to analyze the similarity of patent content,quantify the de-gree of incremental innovation within firms,and further examine the impact of R&D subsidies on incremental innovation and its mechanisms.We first construct a heterogeneous innovation choice model to simulate how firms allocate innovation resources be-tween incremental innovation(improvements in existing production technologies)and breakthrough innovation(develop-ment of new technological paths).By introducing quantity-based subsidies,we deduce the potential effects of innovation policy on firms'innovation behavior.The model reveals that,although R&D subsidies could expand the overall scale of innovation,they also distort the structure of innovation effort,gradually shifting R&D activities toward incremental inno-vation.In the empirical analysis,based on a dataset of listed companies from 2010 to 2022 in China,which includes mil-lions of patent abstract texts,we employ text similarity analysis to measure the degree of incremental innovation.Specifi-cally,we first vectorize patent abstract texts and calculate the patent-to-patent cosine similarity between patent applica-tions in the current year and other patent applications in the past three years.Then,we obtain the similarity indicator at the firm-year level by calculating the mean of patent-to-patent similarity.Higher similarity indicates that the firm's inno-vation relies more on existing technological paradigms than on exploring new technological domains,namely,a higher de-gree of incremental innovation.In addition,we examine the impact of R&D subsidies on incremental innovation using a panel fixed-effects model,with a one-period lag on the R&D subsidy variable to account for the time lag for firms to achieve innovation outcomes.To address the potential endogeneity,we employ an instrumental variable(IV)approach to support our main arguments,using the number of modem primary schools established in the late Qing Dynasty and the implementation of the"Independent Audit Supervision Reform"in the region as instruments.The benchmark results indicate that receiving R&D subsidies significantly contributes to firms'incremental innova-tion,which is consistent with the conclusions of our theoretical model.We further conduct a mechanism analysis and find that R&D subsidies can help firms increase returns to scale within existing technological paradigms by accumulating R&D experience,technical knowledge,and market reputation,thereby reinforcing adaptive expectations to the current technological path.Meanwhile,R&D subsidies encourage firms to keep investing in specialized assets within the existing paradigm,raising the transition costs to new technological paradigms,which leads to incremental lock-in to the current technological path.Due to lower investments,shorter cycles,and lower risks,incremental innovation becomes a strategic choice for firms to achieve non-innovation objectives.In order to secure favorable evaluations in project assessments and cater to the innovation preferences of local officials,firms need to increase the quantity of their innovative outputs through incremental innovation to maximize their policy benefits.The policy implications are as follows.First,the collaboration among industry,academia,and research institutions should be strengthened to support breakthrough innovation and innovation incentives should be designed in a mixed way rather than relying solely on subsidies.Second,firms'path dependence on incremental innovation could be weakened through policy measures like breaking technological monopolies and promoting shared use of specialized assets.Third,in-cremental innovation in cutting-edge technologies and supporting technologies for potential breakthroughs should be en-couraged,fostering a complementary relationship between incremental and breakthrough innovations.Fourth,the innova-tion evaluation system should be reformed,and external supervision should be introduced to reduce firms'strategic moti-vations for engaging in incremental innovation.This paper offers a new explanation of the pathways through which public innovation policies could influence corpo-rate R&D activities from a perspective of innovation path dependence.This study also provides a novel measurement ap-proach for incremental and breakthrough innovations using machine learning-based text-mining techniques.
万方数据
维普
