AI Investment of Chinese Firms: Findings Based on Large-Scale Job Posting Data
FANG Ying, WANG Xiangyu, YE Mengqian, ZHAO Xiliang
Wang Yanan Institute for Studies in Economics/Interdisciplinary Laboratory of Big Data Finance/Paula and Gregory Chow Institute for Studies in Economics, Xiamen University; Fujian Institute of Scientific and Technical Information
Summary:
Measuring firm-level artificial intelligence (AI) investment is central to understanding technology adoption, productivity upgrading, and market valuation, yet direct accounting measures are limited because most AI inputs are intangible and rarely separated in financial statements. Existing proxies based on annual-report wording, IT hardware spending, or patent output capture only part of the process and can blur real investment behavior. Building on the view that AI adoption is human-capital intensive, this study develops a recruitment-based measure that tracks whether firms are actually allocating resources to AI capability development. The core idea is that hiring demand for AI-related skills reflects concrete investment decisions at the implementation stage, thereby providing a scalable, behavior-based indicator that complements disclosure-based metrics. To operationalize this idea, we construct a Chinese labor-skill dictionary with 99,463 skills by integrating Lightcast, O*NET, and ESCO and then localizing the lexicon with a BERT-based named-entity-recognition pipeline trained on Chinese job texts. Using skill co-occurrence with four anchor concepts (AI, machine learning, image recognition, and natural language processing), we compute an AI relevance score for each skill, aggregate scores to the posting level, and classify a posting as AI-relatedwhen its average relevance exceeds 0.05. We then define AIRatio at the firm-year level as the share of AI-related postings in total postings. This framework follows a transparent skill-to-position-to-firm aggregation logic and can be extended to other technologies by replacing anchor skills. The empirical sample combines 7.13 million postings from theZhaopin platform (organized by CnOpenData and matched to listed firms) with annual-report disclosures from CNINFO, AI patent information identified under China National Intellectual Property Administration (CNIPA) rules, and accounting and market variables from CSMAR, covering the period from 2015-2022. We validate AIRatio through three complementary tests. First, benchmark validation shows strong discrimination: firms on the CEIDI Top 100 AI list have a much higher mean AIRatio than other firms (0.0617 versus 0.0097). Second, patent validation shows a significantly positive association between AIRatio and AI patent applications. Third, a capital-market validation based on the release of ChatGPT on November 30, 2022, shows that high-AIRatio portfolios outperform low-AIRatio portfolios in China's event window (form December 30, 2022 to April 1, 2023), and firm-level cumulative abnormal returns are positively related to AIRatio. UsingAIRatio, we document a substantial increase in AI investment among Chinese listed firms from 0.0023 in 2015 to 0.0157 in 2020, followed by a mild normalization to 0.0130 in 2022. Industry heterogeneity is substantial, with information services and finance leading, while agriculture, mining, and hospitality lagging behind. We then compare investment-based and disclosure-based AI measures and find only a modest correlation with frequent firm-level mismatch, including many cases of high disclosure but low contemporaneous hiring. Mechanism analysis indicates that mismatch is not purely cosmetic: disclosure contains forward-looking signals that predict next-period hiring, and it also captures technology positioning and business-AI relatedness beyond current input. In asset pricing, a disclosure premium is broad, while an investment premium is strongest when disclosure is already high, indicating a disclosure-first information hierarchy. The study contributes in three ways.First, it introduces a scalable and behavior-based AI investment metric that is less dependent on voluntary narrative disclosure. Second, it clarifies the different information content of disclosure and investment indicators. Third, it provides evidence on how AI adoption in the real economy is transmitted into equity valuation under a major technology shock. Policy implications are to support AI diffusion in traditional sectors through application-oriented incentives, strengthen AI talent training and mobility, monitor widening inter-firm AI gaps, and improve verifiable AI disclosure standards by linking narrative claims to recruitment, R&D, and patent evidence. Future research can extend the framework to non-listed firms, add new input channels such as cloud and compute spending, and develop stronger causal designs for productivity, labor, and competition effects of AI diffusion.
方颖, 王翔宇, 叶梦芊, 赵西亮. 中国企业的人工智能投入:来自招聘大数据的发现[J]. 金融研究, 2026, 551(5): 152-169.
FANG Ying, WANG Xiangyu, YE Mengqian, ZHAO Xiliang. AI Investment of Chinese Firms: Findings Based on Large-Scale Job Posting Data. Journal of Financial Research, 2026, 551(5): 152-169.
[1]方明月、林佳妮和聂辉华,2022,《数字化转型是否促进了企业内共同富裕?——来自中国A股上市公司的证据》,《数量经济技术经济研究》第11期,第50~70页。 [2]方颖、王翔宇、叶梦芊和赵西亮,2025,《中国上市企业的“人工智能+”创新:理论分析与实证证据》,工作论文。 [3]何晓斌和石一琦,2019,《人工智能的发展和我国人工智能文科人才的培养》,《清华大学教育研究》第4期,第32~38页。 [4]金星晔、左从江、方明月、李涛和聂辉华,2024,《企业数字化转型的测度难题:基于大语言模型的新方法与新发现》,《经济研究》第3期,第34~53页。 [5]李青原和肖泽华,2020,《异质性环境规制工具与企业绿色创新激励——来自上市企业绿色专利的证据》,《经济研究》第9期,第192~208页。 [6]李哲、李心武、焦焰和张文婷,2024,《“多言寡行”的数字化转型披露与分析师预测行为》,《会计研究》第9期,第61~75页。 [7]刘飞和田高良,2019,《信息技术是否替代了就业——基于中国上市公司的证据》,《财经科学》第7期,第95~107页。 [8]马慧和陈胜蓝,2022,《企业数字化转型、坏消息隐藏与股价崩盘风险》,《会计研究》第10期,第31~44页。 [9]潘越、林淑萍、张鹏东和戴亦一,2021,《语言将来时态标记特征与公司股利政策——基于投资者语言认知效应的跨国研究》,《经济研究》第7期,第127~143页。 [10]祁怀锦、曹修琴和刘艳霞,2020,《数字经济对公司治理的影响——基于信息不对称和管理者非理性行为视角》,《改革》第4期,第50~64页。 [11]邵红霞和方军雄,2006,《我国上市公司无形资产价值相关性研究——基于无形资产明细分类信息的再检验》,《会计研究》第12期,第25~32页。 [12]陶锋、王欣然、徐扬等,2023,《数字化转型、产业链供应链韧性与企业生产率》,《中国工业经济》第5期,第118~136页。 [13]吴非、胡慧芷、林慧妍和任晓怡,2021,《企业数字化转型与资本市场表现——来自股票流动性的经验证据》,《管理世界》第7期,第130~144页。 [14]肖土盛、孙瑞琦、袁淳等,2022,《企业数字化转型、人力资本结构调整与劳动收入份额》,《管理世界》第12期,第220~237页。 [15]许宪春、雷泽坤和胡亚茹,2025,《中国数据资本形成总额与数据资本存量测算》,《经济研究》第3期,第54~68页。 [16]姚加权、张锟澎、郭李鹏和冯绪,2024,《人工智能如何提升企业生产效率?——基于劳动力技能结构调整的视角》,《管理世界》第2期,第101~116、117~122、133页。 [17]张俊芳、郭戎和郭永济,2018,《中国无形资产测度及其对科技进步贡献率影响的研究》,《科学学与科学技术管理》第1期,第46~54页。 [18]赵宸宇、王文春和李雪松,2021,《数字化转型如何影响企业全要素生产率》,《财贸经济》第7期,第114~129页。 [19]Acemoglu, D. and D. Autor, 2011, “Skills, Tasks and Technologies: Implications for Employment and Earnings”,Handbook of Labor Economics, Vol. 4, pp.1043~1171. Amsterdam: Elsevier. [20]Autor, D., C. Chin, A. Salomons and B. Seegmiller, 2024, “New Frontiers: The Origins and Content of New Work, 1940–2018”,The Quarterly Journal of Economics, 139(3), pp.1399~1465. [21]Babina, T., A. Fedyk, A. He and J. Hodson, 2024, “Artificial Intelligence, Firm Growth, and Product Innovation”, Journal of Financial Economics, 151, 103745. [22]Brynjolfsson, E., D. Rock and C. Syverson, 2021, “The Productivity J-Curve: How Intangibles Complement General Purpose Technologies”, American Economic Journal: Macroeconomics, 13(1), pp.333~372. [23]Cadman, B., M. S. Heinle and D. Macciocchi, 2023, “Insider Horizon and Disclosure Policies”,The Accounting Review, 98(4), pp.55~88. [24]Chu, J., Y. He, K. W. Hui and R. Lehavy, 2025, “New Product Announcements, Innovation Disclosure, and Future Firm Performance”,Review of Accounting Studies, 30(1), pp.352~383. [25]Darendeli, A., K. K. Law and M. Shen, 2022, “Green New Hiring”,Review of Accounting Studies, 27(3), pp.986~1037. [26]Duarte, J., E. Hu and L. Young, 2020, “A Comparison of Some Structural Models of Private Information Arrival”,Journal of Financial Economics, 135(3), pp.795~815. [27]Hirshleifer, D. and J. Sheng, 2022, “Macro News and Micro News: Complements or Substitutes?”,Journal of Financial Economics, 145(3), pp.1006~1024. [28]Hunt, W., S. Sarkar and C. Warhurst, 2022, “Measuring the Impact of AI on Jobs at the Organization Level: Lessons from a Survey of UK Business Leaders”,Research Policy, 51(2), 104425. [29]Jiang, W., J. Park, R. J. Xiao and S. Zhang, 2025, “AI and the Extended Workday: Productivity, Contracting Efficiency, and Distribution of Rents”,NBER Working Paper, No. 33536. [30]LeCun, Y., Y. Bengio and G. Hinton, 2015, “Deep Learning”, Nature, 521(7553), pp.436~444. [31]Loh, R. K., 2010, “Investor Inattention and the Underreaction to Stock Recommendations”,Financial Management, 39(3), pp.1223~1252. [32]Müller, O., M. Fay and J. Vom Brocke, 2018, “The Effect of Big Data and Analytics on Firm Performance: An Econometric Analysis Considering Industry Characteristics”,Journal of Management Information Systems, 35(2), pp.488~509. [33]Pástor,. and P. Veronesi, 2006, “Was There a Nasdaq Bubble in the Late 1990s?”,Journal of Financial Economics, 81(1), pp.61~100. [34]Rock, D., 2022, “Engineering Value: The Returns to Technological Talent and Investments in Artificial Intelligence”, Brookings Institution Center on Regulation and Markets Working Paper. [35]Sharma, S. S. and P. K. Narayan, 2022, “Technology Shocks and Stock Returns: A Long-Term Perspective”,Journal of Empirical Finance, 68, pp.67~83.
