The impacts of AI on new quality productive forces: an empirical study
Abstract
Artificial intelligence is widely viewed as an engine of new quality productive forces, yet firm-level evidence on how the two are linked—and causal evaluations of the associated industrial policies—remains scarce. Using data on Chinese A-share listed companies from 2011 to 2022, this study measures AI adoption through annual-report keyword counts and constructs firm-level new quality productive forces (NPRO) using an established entropy-weighted index system. Three findings emerge. First, AI disclosure is robustly associated with higher NPRO: the association survives firm and year fixed effects, industry-by-year and province-by-year shocks, alternative index weights, and five TFP-based substitute measures, amounting to about 0.06 standard deviations under the most demanding specification. Second, decomposition evidence indicates that the association travels mainly with R&D human capital rather than with funding alone; this conclusion also holds for outcomes free of index-construction overlap, including TFP. Third, the staggered difference-in-differences analysis of the pilot zones yields positive point estimates, but the formal group-time pretrend test rejects parallel trends. In the balanced short window, where the pretrend test does not reject, the estimate is close to zero, and the Rambachan–Roth sensitivity intervals also include zero. The 2019 cohort is not point-identified because covariate overlap fails, and simple subgroup comparisons substantially overstate policy differences. Theoretically, these findings help clarify how AI contributes to firm productivity: the association between AI and NPRO appears to work mainly through the growth of R&D human capital, rather than through financial investment alone. This suggests that the “new quality” aspect of productivity has more to do with upgrading labor quality and recombining knowledge than with simply deepening capital. The sample ends in 2022, so the results speak to pre-LLM artificial intelligence technologies.