Chinese cities that tout digital government see their listed firms invest more abroad — a text-based index and pilot-city analysis suggest digital-government development is associated with roughly 12–13% bigger overseas investment by firms, though causal channels and endogeneity remain underexplored.

Main Finding

The paper finds that stronger digital government development in Chinese prefecture-level jurisdictions is associated with larger overseas expansion by local A‑share listed firms. Using firm-year data (2009–2023) and two‑way fixed‑effects regressions, a one‑unit increase in the author's digital government index is associated with roughly a 12–13% increase in firms' overseas investment scale; results are robust to alternative dependent variables, a pilot‑city DID indicator, log transformations, and additional controls.

Key Points

• Research question: Does home‑country digital government construction help Chinese firms "go global," and by what mechanisms?
• Main hypothesis (H1): Digital government promotes firms' overseas expansion by lowering transaction costs.
• Mechanisms proposed:
  • Reducing information search costs via centralized, reliable digital public information (less information asymmetry).
  • Lowering cross‑border compliance/administrative costs through one‑stop online services and process re‑engineering.
  • Decreasing overseas operational and political/ policy risk via data‑driven monitoring and early‑warning tools.
• Heterogeneity noted: the promotional effect is smaller for non‑state‑owned firms (author reports a weaker effect in non‑SOEs).
• Robustness: effects persist when (a) using a DID indicator based on digital governance pilot cities, (b) substituting alternative OFDI measures (log amount, binary OFDI indicator), (c) log‑transforming the digital index, and (d) adding governance controls (TobinQ, growth, board size).

Data & Methods

• Sample: Chinese A‑share listed firms, 2009–2023 (after exclusions and 1% winsorization). Descriptive counts: ~43,199 observations originally; regressions report ~34k firm‑year observations depending on specification.
• Data sources: Wind and CSMAR databases. Software: Stata 18.
• Key variables:
  • Independent variable (diggov): a prefecture‑level digital government index constructed by word‑frequency analysis of local government documents and announcements (frequency of digital/digital‑transformation terms normalized by document length).
  • Dependent variables (ofdi): measures of overseas expansion including scale of overseas investment, log(amount) of foreign investment, and a binary indicator of whether the firm invests abroad.
  • Controls: firm size (ln total assets), leverage (total liabilities/total assets), ROA, firm age, Top1 share (largest shareholder), and in some specs TobinQ, growth, board size.
• Empirical strategy:
  • Two‑way fixed‑effects panel regressions (firm and year fixed effects); some models include industry fixed effects.
  • Robustness checks include DID using pilot cities for digital governance, alternative DV measures, log transformations, and additional corporate governance controls.
• Main reported magnitudes:
  • Baseline FE models: diggov coefficient ≈ 12.23–13.28 (interpreted as ~12–13% expansion in overseas investment per unit increase in diggov), statistically significant (p ≤ 0.10 → p ≤ 0.05 after controls).
  • DID specification (pilot cities) also shows a positive and significant effect (DID ≈ 0.0481*** in one robustness column).

Implications for AI Economics

• Digital public infrastructure as an economic complement: The results imply that government investments in digital public goods (data platforms, AI analytics, one‑stop online procedures) can materially lower firms' cross‑border transaction costs and enable greater internationalization. Empirical work in AI economics should treat digital government and public AI systems as supply‑side factors that can shift firm investment decisions.
• Role of AI in reducing informational frictions and country risk: AI and big‑data systems that aggregate cross‑border legal/policy information and provide predictive risk signals can substitute for costly firm‑level search and risk assessment. This suggests welfare gains from publicly provided AI tools and avenues for measuring their returns.
• Distributional/heterogeneous effects matter: Smaller effects for non‑SOEs indicate that firm ownership, governance, or absorptive capacity shape gains from public digital/AI services. AI‑economics analyses should model heterogeneous firm responses and complementarities with firm capabilities.
• Measurement and identification challenges for AI policy evaluation:
  • Proxying digital government by word‑frequency captures attention/commitment but has limited variance (author notes mean≈0, max≈0.01), raising measurement validity concerns. Future work should use richer usage/transaction data (platform usage logs, service volumes, API calls) to quantify digital public AI impact.
  • Potential endogeneity remains (regions that develop digital government may also pursue other pro‑OFDI policies). Stronger causal designs (instrumental variables, staggered rollout with pre‑trends checks, or randomized pilots) are recommended.
• Policy design considerations:
  • Investing in interoperable, transparent cross‑border data and AI services can be a lever for national industrial strategies to foster outward FDI.
  • Data governance, privacy, and international data‑sharing rules will affect the scope and efficacy of such digital public goods—AI economics should incorporate institutional constraints and regulatory costs.
• Directions for further research in AI economics:
  • Causal evaluation of specific AI/digital services (e.g., automated legal‑policy summarizers, risk‑scoring models) on firm foreign expansion outcomes.
  • Cross‑country comparisons to test generalizability and to analyze how host‑country institutions interact with home digital public infrastructure.
  • Firm‑level studies on adoption of government digital services: which firms use them, how usage mediates the effect, and complementarities with private AI tools.

Limitations to note (from the study): diggov is a proxy based on word frequency (proxy validity/variance issues), potential endogeneity, and limited direct tests of the three proposed channels (information, compliance, risk). The findings are specific to listed Chinese firms and prefecture‑level measures of digital government.