| 蒋伟杰,齐晨扬,陈琦.算法创新与企业全要素生产率提升——来自专利文本的经验证据[J].数量经济技术经济研究,2026,(5):30-53 | | 算法创新与企业全要素生产率提升——来自专利文本的经验证据 | | Algorithm Innovation and Promotion of Enterprise Total Factor Productivity: Empirical Evidence from Patent Texts | | | | DOI： | | 中文关键词: 算法创新 数据要素 全要素生产率 大语言模型 | | 英文关键词: Algorithm Innovation Data Factor Total Factor Productivity Large Language Model | | 基金项目: | | | 中文摘要: | | 在数字经济加速深化的背景下，算力、算法与数据已成为支撑实体经济转型的关键资源。算法创新作为激发数据要素潜在价值的核心引擎，正逐渐成为推动企业高质量发展的新动能。本文利用 2010~2021 年中国上市公司数据，引入 MacBERT 大语言模型处理海量专利摘要文本，精准识别算法专利，并以此刻画地区算法创新水平，实证检验其对企业全要素生产率的影响。研究发现，算法创新能够显著提升企业全要素生产率。机制分析表明，算法创新主要通过提升投入端和产出端数据利用效率发挥赋能作用，一方面，算法创新能够缓解信息摩擦，纠正要素错配并提升产能利用率；另一方面，算法创新能够增强市场需求识别能力，驱动研发创新、降低销售费用并加速资金周转。异质性分析显示，在数字基础设施完善、数据开放程度高的地区，以及需求波动大的行业，算法创新的促进作用更为明显。本文为充分挖掘数据要素价值提供了微观经验证据，为构建算法创新支撑体系、优化数字化发展环境提供了重要的政策启示。 | | 英文摘要: | | Due to the accelerating and deepening digital economy, computing power, algorithms, and data are critical resources underpinning the digital transformation of the real economy. Among these, algorithm innovation, as the core engine that unleashes the latent value of data elements, is progressively becoming a new kinetic energy driving the high-quality development of enterprises.Despite its theoretical importance, quantifying algorithm innovation at the micro-level is a significant empirical challenge in the existing economic literature. To bridge this gap, this study utilizes a comprehensive panel dataset of Chinese A-share listed companies from 2010 to 2021. Methodologically, we introduce a novel approach by employing the MacBERT large language model-a cutting-edge natural language processing technique-to process and analyze massive volumes of patent abstract texts. By fine-tuning the model for text classification and semantic similarity analysis, we accurately identify algorithm-specific patents from millions of general innovation patents. Aggregating these identified patents, we construct a robust index to delineate and portray the regional level of algorithm innovation. Utilizing this index, we empirically investigate the causal impact of regional algorithm innovation on firm-level total factor productivity (TFP).Our baseline empirical findings reveal a robust and positive relationship: algorithm innovation significantly enhances the TFP of enterprises. To ensure the reliability of our estimations and address potential endogeneity issues arising from reverse causality or omitted variable bias, we employ an instrumental variable approach, leveraging historical and geographical technological infrastructure as exogenous shocks. The core conclusion remains highly significant and robust after applying this two-stage least squares estimation, as well as a series of rigorous robustness checks, including alternative TFP measurement methods, adjusting the sample periods, and incorporating high-dimensional fixed effects to control for unobservable confounding factors.Through a detailed mechanism analysis, we demonstrate that algorithm innovation exerts its empowering effect primarily by substantially improving the efficiency of data utilization on both the input and output sides of the production process. On the input side, algorithm innovation significantly mitigates information friction between a firm and factor markets. By utilizing advanced matching algorithms and intelligent production scheduling, enterprises can dynamically optimize their resource deployment, effectively correct the severe misallocation of labor and capital factors, and thus maximize capacity utilization rates. On the output side, algorithmic advancements empower enterprises with enhanced capabilities to accurately identify and predict market demand and consumer preferences. This deep market insight actively promotes targeted research and development and product innovation. Furthermore, precision marketing algorithms drastically reduce unit sales expenses and marketing costs, while intelligent supply chain management accelerates inventory monetization and overall capital turnover, thereby realizing a closed-loop enhancement of corporate value creation.Our heterogeneity analysis uncovers substantial asymmetric effects across different external environments and industrial characteristics. Specifically, the promotional effect of algorithm innovation on TFP is remarkably more pronounced in regions equipped with highly developed digital infrastructure and a higher degree of government data openness. These favorable regional conditions provide the necessary hardware support and data liquidity essential for algorithms to function optimally. Additionally, the productivity-enhancing effect is significantly stronger for enterprises operating in industries characterized by high demand volatility. In such highly uncertain market environments, the predictive power and risk-mitigation capabilities of algorithms are particularly valuable, offering firms a stronger competitive advantage in dynamic resource allocation.In conclusion, this study not only provides solid micro-level empirical evidence for fully excavating and unleashing the economic value of data elements but also deepens our theoretical understanding of how digital technologies integrate with the real economy. The findings offer profound policy implications for government authorities and corporate decision-makers. To foster high-quality economic growth, it is imperative to establish a comprehensive support system for algorithmic innovation, including subsidizing core research and safeguarding intellectual property. Furthermore, policymakers should strive to optimize the broader digital development environment by accelerating the construction of new digital infrastructure, dismantling data silos to facilitate public data sharing, and formulating differentiated, region-specific support policies to maximize the transformative potential of the digital economy. | | 查看全文 |
| | |
