China’s ageing drag on growth can be blunted where industrial-robot adoption is high; provinces that cross a robot-penetration threshold see much smaller negative effects of ageing on GDP growth, suggesting automation can partly offset demographic headwinds.
Nonlinear effects of ageing population and AI on China’s GDP growth: a threshold analysis
At the provincial level in China, population ageing initially depresses GDP growth, but this negative effect is significantly mitigated in provinces where industrial robot penetration (used as an AI proxy) exceeds a critical threshold.
This research empirically explores the influences of ageing on China’s GDP growth, incorporating Artificial Intelligence (AI) as a moderating factor. Specifically, industrial robot penetration was used as a proxy for AI adoption. This research selects panel data in 31 provinces of China (2000–2022). The nonlinear association between ageing population and GDP growth is examined using panel threshold regression models, while threshold variables are ageing and AI adoption, respectively. To verify the robustnes, the old-age dependency ratio is utilized as a proxy of ageing population. According to the findings, GDP growth is initially negatively affected by ageing population. However, when AI adoption surpasses a critical threshold, this negative effect is significantly mitigated. This finding highlights the importance of AI adoption in managing the economic challenges brought by ageing. Therefore, some valuable recommendations have been put forward to support inclusive and sustainable economic development. These include greater investment in research and expansion concerning AI, promoting AI-driven robotics in key sectors, and offering targeted skilling programs for elderly employees. Further suggestions are to invest in digital infrastructures and the industry of ageing, as well as to leverage and develop elderly human capital.
Summary
Main Finding
China’s ageing population has a negative effect on provincial GDP growth, but that adverse effect is non-linear: once AI adoption (proxied by industrial robot penetration) exceeds a critical threshold, the negative impact of ageing on GDP growth is substantially mitigated. Results are robust when using the old-age dependency ratio as an alternative ageing measure.
Key Points
- Relationship is non-linear: ageing depresses growth on average, but the magnitude depends on regime—identified via panel threshold regressions.
- AI (industrial robot density) moderates the ageing effect: below an AI threshold, ageing’s negative effect is stronger; above the threshold, the negative effect is significantly reduced (i.e., AI can offset some demographic drag).
- The study tests thresholds in two ways: (1) ageing as the threshold variable and (2) AI as the threshold variable, using single- and double-threshold specifications.
- Controls include regional gross capital formation, labour productivity, and human capital; findings are robust to an alternative ageing proxy (old-age dependency ratio).
- Policy recommendations: expand AI R&D and robot adoption in productivity-critical sectors, invest in digital infrastructure, promote targeted reskilling/upskilling (including for older workers), and develop the “ageing industry” and policies to mobilize elderly human capital.
Data & Methods
- Data: balanced provincial panel for 31 Chinese provinces (excl. HK, Macao, Taiwan), 2000–2022. Main sources: National Bureau of Statistics of China (NBSC), World Bank, International Federation of Robotics (IFR). Variables were log-transformed; some missing values interpolated.
- Key variables:
- Dependent: GDP per capita growth (province-level).
- Main regressors: proportion aged 65+ (ageing); AI proxied by industrial robot penetration/density.
- Alternative ageing proxy: old-age dependency ratio (65+/15–64).
- Controls: regional gross capital formation (RGCF), labour productivity (LP), human capital (HC = years of schooling growth).
- Econometric approach:
- Panel threshold regression following Hansen (1999) with grid search (grid step 0.0025) to locate thresholds.
- Models estimated with ageing or AI as threshold variables, in single- and double-threshold forms to capture regime-dependent (non-linear) effects.
- Robustness checks include alternative ageing proxy and standard controls.
- Theoretical framing: endogenous growth / Cobb–Douglas expanded to include human capital and a role for R (research/AI) as the technological input that raises productivity.
Implications for AI Economics
- AI as a macroeconomic moderator: The paper provides evidence that AI adoption can change how demographic shocks (ageing) transmit to aggregate output — implying that technological adoption levels matter qualitatively, not just quantitatively.
- Threshold dynamics matter for policy targets: Because benefits appear only after surpassing an AI-adoption threshold, gradual or uneven AI rollouts may be insufficient to offset demographic headwinds; policymakers should aim for sufficiently high AI diffusion in key sectors/regions.
- Trade-offs and distributional considerations: While AI can raise aggregate growth and partially offset labour-supply declines, the paper reinforces the need to pair automation with reskilling and inclusion policies to avoid displacement and inequality, especially among older and lower-skilled workers.
- Modeling guidance for researchers: Studies of demographic or technological impacts on growth should allow for non-linear and regime-dependent effects (threshold models), use regional heterogeneity, and test alternative measures of AI (robot density is a useful proxy but not comprehensive).
- Directions for future work in AI economics: causal identification of AI’s offsetting effect (e.g., instruments, difference-in-differences at firm or plant level), micro-to-macro channels (TFP, employment composition, wages), heterogenous effects across industries and skill groups, and long-run distributional and fiscal impacts of ageing combined with automation.
Assessment
Paper Typecorrelational
Evidence Strengthlow — Findings are based on observational provincial panel correlations and threshold regressions without clear exogenous identification (no IV, diff-in-diff, or natural experiment), so estimated mitigation of ageing effects by 'AI' may reflect omitted variables, reverse causality, or measurement error in the robot-as-AI proxy.
Methods Rigormedium — The study exploits a long panel (31 provinces × ~23 years), employs nonlinear panel threshold models and robustness checks (alternative ageing measure), which are appropriate and reasonably sophisticated for exploring heterogeneous associations; however, rigor is limited by the lack of causal identification and potential omitted confounders at the provincial level.
SampleProvincial-level panel data for 31 Chinese provinces from 2000 to 2022 (roughly 31×23 ≈ 713 province-year observations), with variables including GDP growth, industrial robot penetration (proxy for AI adoption), elderly population share and old-age dependency ratio, and unspecified control covariates.
Themesproductivity adoption
IdentificationUses panel threshold regression on province-level panel (31 Chinese provinces, 2000–2022) with industrial robot penetration as a proxy for AI adoption and ageing (share of elderly and old-age dependency ratio) as threshold variables; reports robustness checks substituting alternative ageing measures. No exogenous variation, instruments, or natural experiment are reported.
GeneralizabilityUses industrial robot penetration as a proxy for 'AI' — excludes software AI and service-sector AI, so results mainly reflect manufacturing automation., Province-level aggregation in China — may not generalize to firm/worker-level effects or to other countries with different labor markets and institutions., China-specific policy, demographic, and industrial structure may limit applicability to OECD or low-income countries., Possible measurement error and omitted-variable bias (e.g., simultaneous policy changes, capital deepening, or sectoral compositional shifts) reduce external validity., Threshold estimates (critical levels of robot penetration) may not transfer across contexts or time periods as technology and adoption patterns evolve.
Claims (6)
| Claim | Direction | Confidence | Outcome | Details |
|---|---|---|---|---|
| The study uses panel data on 31 Chinese provinces for the period 2000–2022 and employs panel threshold regression models with ageing and AI adoption as threshold variables. Other | null_result | high | methodological approach (panel threshold regression) |
n=713
0.5
|
| Industrial robot penetration is used as a proxy measure for AI adoption in Chinese provinces. Adoption Rate | null_result | high | AI adoption (proxied by industrial robot penetration) |
n=713
0.15
|
| GDP growth is initially negatively affected by the ageing population. Fiscal And Macroeconomic | negative | high | GDP growth |
n=713
0.3
|
| When AI adoption (industrial robot penetration) surpasses a critical threshold, the negative effect of ageing on GDP growth is significantly mitigated. Fiscal And Macroeconomic | positive | high | GDP growth (mitigation of negative ageing effect by AI adoption) |
n=713
0.3
|
| Robustness checks using the old-age dependency ratio as the proxy for ageing deliver consistent results. Fiscal And Macroeconomic | positive | high | GDP growth (robustness of ageing effect and AI mitigation) |
n=713
0.3
|
| Policy recommendations: increase investment in AI research and expansion; promote AI-driven robotics in key sectors; provide targeted skilling programs for elderly workers; invest in digital infrastructure and the ageing industry; and leverage and develop elderly human capital to support inclusive and sustainable economic development. Governance And Regulation | positive | high | policy actions to manage ageing-related economic challenges |
0.05
|