Volume 2,Issue 8
人工智能政策对企业绩效的影响研究
以1998—2024年绝大部分的中国上市公司为研究对象，构建了35,883个企业年度观测的非平衡面板数据集，采用双向固定效应差分模型（TWFE-DID）、倾向得分匹配（PSM）及XGBoost 机器学习三种方法，估计人工智能政策支持对企业净利润的影响。TWFE 基准估计显示，AI 政策支持使企业对数净利润提升0.578个单位（聚类稳健标准误，p=0.004），对应净利润水平提升约78.2%；事件研究法验证政策前期平行趋势假设成立；PSM 稳健性检验与Heckman 选择修正进一步印证上述结论。XGBoost 模型引入滞后利润、资产增速等动态特征后，5折交叉验证R² 达0.796，SHAP 近似分析揭示滞后利润与企业规模是净利润的首要预测变量，研发投入强度具有独立正向效应，AI 政策支持的模型反事实处置效应贡献1.8%，三类方法结论相互印证。
[1] 科技部. 新一代人工智能发展规划推进办公室工作规则 [R]. 北京: 科技部, 2018.
[2]Aghion P, Howitt P. A model of growth through creative destruction [J]. Econometrica, 1992, 60(2): 323-351.
[3]Bloom N, Griffith R, Van Reenen J. Do R&D tax credits work? Evidence from a panel of countries 1979-1997 [J]. Journal of Public Economics, 2002, 85(1): 1-31.
[4]Callaway B, Sant’Anna P H C. Difference-in-differences with multiple time periods [J]. Journal of Econometrics, 2021, 225(2): 200-230.
[5]Rosenbaum P R, Rubin D B. Constructing a control group using multivariate matched sampling methods that incorporate the propensity score [J]. American Statistician, 1985, 39(1): 33-38.
[6]Heckman J J. Sample selection bias as a specification error [J]. Econometrica, 1979, 47(1): 153-161.
[7]Chen T, Guestrin C. XGBoost: A scalable tree boosting system [C]. Proceedings of the 22nd ACM SIGKDD, 2016: 785-794.
[8]Lundberg S M, Lee S I. A unified approach to interpreting model predictions [J]. NeurIPS, 2017, 30: 4765-4774.
[9]Goodman-Bacon A. Difference-in-differences with variation in treatment timing [J]. Journal of Econometrics, 2021, 225(2): 254-277.
[10]Wooldridge J M. Econometric Analysis of Cross Section and Panel Data [M]. 2nd ed. Cambridge: MIT Press, 2010.