China’s national digital‑economy pilot zones materially raise the quality and digital intensity of listed firms’ productive capacity, boosting high‑tech productivity and green innovation; gains are concentrated among non‑state, high‑tech, and eastern companies.
The Impact of Digital Economy Pilot Zones on Corporate New Quality Productive Forces: Evidence from Double Machine Learning
Establishing National Digital Economy Innovation and Development Pilot Zones significantly strengthened listed firms’ New Quality Productive Forces (NQPF) between 2015–2023, operating through improved factor allocation, deeper digital technology adoption, and increased green innovation, with largest effects for non‑state, high‑tech, and eastern firms.
As a transformative force, the digital economy serves as a critical engine for driving high-quality economic development and fostering New Quality Productive Forces (NQPF)—characterized by high technology, high efficiency, and high quality. Viewing the establishment of China’s National Digital Economy Innovation and Development Pilot Zones as a quasi-natural experiment in economic system management, this study employs a Double Machine Learning (DML) framework to evaluate its systemic impact on A-share listed companies from 2015 to 2023. Unlike traditional linear models, the DML approach flexibly controls for high-dimensional confounding variables and functional form misspecification, thereby ensuring highly rigorous causal inference. The empirical results demonstrate that these pilot zones create an optimized “digital environment” that significantly enhances corporate NQPF, a conclusion that remains highly robust across a comprehensive battery of robustness and endogeneity tests. Mechanism analysis reveals three systemic transmission pathways through which the policy operates: optimizing factor allocation, deepening digital technology empowerment, and promoting green innovation and sustainability. Furthermore, heterogeneity analyses indicate that the policy’s efficacy varies significantly across corporate profiles, manifesting most prominently in non-state-owned enterprises, high-tech firms, and those located in eastern regions. These findings provide robust micro-level evidence for policymakers aiming to optimize digital economic systems and accelerate the systemic formation of advanced productive forces.
Summary
Main Finding
China’s National Digital Economy Innovation and Development Pilot Zones causally and robustly strengthen firms’ New Quality Productive Forces (NQPF) — i.e., higher technology, efficiency, and product/service quality — for A‑share listed companies (2015–2023). Using a Double Machine Learning (DML) causal framework, the study shows the pilot zones create an optimized “digital environment” that raises firm-level technological upgrading and productivity, with effects concentrated in non‑state‑owned firms, high‑tech firms, and firms in eastern regions.
Key Points
- Research design: Treats the establishment of pilot zones as a quasi‑natural experiment and estimates the causal effect on firm‑level NQPF.
- Methodology strength: Uses Double Machine Learning to flexibly control for high‑dimensional confounders and avoid functional‑form misspecification, improving credibility of causal inference.
- Primary result: Pilot zones significantly improve firm NQPF (technology intensity, efficiency, and quality measures).
- Mechanisms identified:
- Optimizing factor allocation — better allocation of capital, labor, and other inputs toward higher‑productivity uses.
- Deepening digital technology empowerment — increased digital adoption, R&D and digital tool use that raise firm capabilities.
- Promoting green innovation and sustainability — higher green patenting, eco‑innovation and sustainability practices.
- Heterogeneity: Effects are stronger for non‑state‑owned enterprises (NSOEs), firms classified as high‑tech, and firms located in eastern China.
- Robustness: Findings hold across a comprehensive set of robustness and endogeneity checks reported by the authors.
Data & Methods
- Sample: A‑share listed companies in China, 2015–2023.
- Treatment: Firm exposure to locations designated as National Digital Economy Innovation and Development Pilot Zones.
- Outcome(s): Firm‑level indicators capturing NQPF — technological upgrading, productivity/efficiency metrics, product/innovation quality, and green innovation measures.
- Econometric approach:
- Double Machine Learning (DML): orthogonalized (debiased) estimation with machine‑learning models for nuisance functions and cross‑fitting to mitigate overfitting and bias from high‑dimensional controls.
- Advantage over linear models: permits flexible, nonparametric control of many covariates and reduces sensitivity to functional‑form misspecification.
- Robustness and endogeneity strategy: the paper reports a battery of tests (placebo checks, alternative specifications, sample splits, and other endogeneity adjustments) to validate causal interpretation.
- Mechanism tests: empirical mediation analyses linking treatment to intermediate channels (factor allocation, digital adoption/R&D, green innovation) and then to outcomes.
- Heterogeneity analysis: stratified regressions/subsample analyses by ownership type, technology intensity, and geographic region.
Implications for AI Economics
- Policy evaluation methods: Demonstrates the value of Double Machine Learning and other ML‑based causal tools for credible policy evaluation in digital/AI economics where confounders are high‑dimensional and relationships nonlinear.
- Digital ecosystems and AI adoption: Policy‑driven digital environments (pilot zones) materially accelerate firm‑level technology adoption and productivity—likely including adoption of AI and complementary digital tools—highlighting the importance of ecosystem policies (infrastructure, data governance, skills).
- Factor reallocation and labor markets: Improved factor allocation implies firms shift resources toward higher‑value activities; AI economics research should account for endogenous reallocations across firms and occupations when assessing welfare and distributional effects.
- Green AI and sustainability: The link between digital policy and green innovation suggests policy design can simultaneously promote AI/digitalization and environmental objectives; analyses of AI’s environmental footprint should incorporate inducement effects on green innovation.
- Heterogeneity and targeting: Stronger impacts for NSOEs, high‑tech, and eastern firms indicate uneven benefits — policymakers should consider targeted support (capacity building, financing, regionally balanced rollout) to avoid widening gaps.
- Future research directions: Use DML and related ML causal methods to evaluate AI‑specific policies (e.g., AI testing zones, data‑sharing regimes), study dynamic/general equilibrium effects of digitalization on markets and wages, and measure firm‑level AI adoption with granular microdata to trace causal chains from policy → AI adoption → productivity and welfare.
Assessment
Paper Typequasi_experimental
Evidence Strengthmedium — The study uses a modern causal‑ML estimator (DML) and a policy shock as a quasi‑experiment, with extensive robustness and mechanism tests, which supports credible inference; however, treatment assignment to pilot zones may be endogenous (selection by local governments), spillovers across firms/regions are possible, the NQPF outcome appears composite and may contain measurement choices, and causal claims rely on conditional unconfoundedness assumptions that are hard to fully verify.
Methods Rigormedium — The authors apply a state‑of‑the‑art DML framework to handle high‑dimensional confounding and functional form misspecification and report robustness/endogeneity checks and heterogeneity and mechanism analyses; nevertheless, key implementation details that affect rigor (e.g., evidence on pre‑trends or parallel trends, how sample splitting/cross‑fitting was done, clustering of standard errors, addressing spatial spillovers, and measurement construction of NQPF) are not specified here, leaving some practical identification risks.
SampleFirm‑level panel of Chinese A‑share listed companies from 2015 through 2023, with treatment defined by firm headquarters/location inside designated National Digital Economy Innovation and Development Pilot Zones; data likely include firm financials, industry and region controls, patent/innovation and environmental/green innovation indicators, and other high‑dimensional covariates used in DML.
Themesproductivity innovation
IdentificationTreated firms are those located in China’s National Digital Economy Innovation and Development Pilot Zones (policy rollout treated as a quasi-natural experiment); the paper compares treated vs. control A‑share listed firms over 2015–2023 and implements Double Machine Learning to flexibly control for high‑dimensional covariates and nonlinear functional forms, supplemented by robustness and endogeneity checks (e.g., alternative controls, placebo tests, heterogeneity analyses).
GeneralizabilityLimited to publicly listed (A‑share) firms — excludes private, small, and informal firms, China‑specific policy and institutional context may not generalize to other countries, Pilot zone selection may be targeted to more advanced/eastern regions, limiting geographic representativeness, Findings concern digital‑economy pilot policy broadly rather than AI‑specific interventions, so transfer to AI‑only contexts is imperfect, Timeframe 2015–2023 may miss longer‑run or delayed effects
Claims (7)
| Claim | Direction | Confidence | Outcome | Details |
|---|---|---|---|---|
| Establishment of China’s National Digital Economy Innovation and Development Pilot Zones significantly enhances corporate New Quality Productive Forces (NQPF). Firm Productivity | positive | high | corporate New Quality Productive Forces (NQPF) |
0.48
|
| The DML approach flexibly controls for high-dimensional confounding variables and functional form misspecification, enabling highly rigorous causal inference compared with traditional linear models. Other | positive | high | quality of causal inference / methodological rigor |
0.48
|
| The pilot zones create an optimized 'digital environment' that underlies the positive impact on corporate NQPF. Organizational Efficiency | positive | high | presence/quality of digital environment / organizational digital infrastructure |
0.48
|
| Mechanism analysis identifies three systemic transmission pathways for the policy: optimizing factor allocation, deepening digital technology empowerment, and promoting green innovation and sustainability. Innovation Output | positive | high | mechanistic channels: factor allocation, digital technology empowerment, green innovation |
0.48
|
| The estimated positive effect of the pilot zones on corporate NQPF is robust across a comprehensive battery of robustness and endogeneity tests. Firm Productivity | positive | high | robustness of estimated policy effect on NQPF |
0.48
|
| Policy efficacy varies significantly across corporate profiles, with the strongest effects observed in non-state-owned enterprises, high-tech firms, and firms located in eastern regions. Firm Productivity | positive | high | heterogeneous policy impact on corporate NQPF across firm subgroups |
0.48
|
| The empirical analysis is based on A-share listed companies from 2015 to 2023. Other | null_result | high | study sample/timeframe |
0.8
|