Chinese private listed firms that report higher digital transformation show a small but significant rise in headcount, driven by revenue expansion and productivity gains; findings are associational and may reflect reporting or selection biases rather than clean causal effects.

Main Finding

Digital transformation in China’s private A‑share listed firms (2013–2023) is associated with a statistically significant expansion of firm employment. The paper finds that digitalization increases employment primarily by (1) enlarging business scale and (2) raising firm productivity. The effect is stronger in firms with higher worker education, more mid-to-high skilled labor, and those located in larger cities.

Key Points

• Scope and motivation: Private enterprises account for >80% of urban employment in China and are central to employment stabilization; the paper studies whether their digital transformation consolidates that role.
• Primary hypothesis (H1): Digital transformation expands employment scale in private firms. Supported by empirical results.
• Mechanisms (H2a, H2b): Two mediating channels are tested and supported:
  • Productivity (TFP) growth: digital adoption raises efficiency and stimulates demand, increasing labor needs.
  • Business scale expansion: digitalization lowers costs, widens markets, and creates new business lines, requiring more workers.
• Heterogeneity: Positive employment effects are more pronounced for firms with:
  • Higher employee education levels,
  • Larger shares of mid-to-high skilled labor,
  • Location in larger/city-tier regions.
• Economic magnitude (baseline): In the fully controlled model, the coefficient on the digitalization index is 0.025 — interpreted in the paper as a 1% increase in the digital transformation measure being associated with a 0.025% increase in employment (small but statistically significant).

Data & Methods

• Sample: Private A‑share listed companies in China, 2013–2023. Final sample: 21,947 firm‑year observations.
• Dependent variable: employ = ln(number of employees).
• Independent variable (digital transformation, dig): Textual analysis of firms’ annual reports. Keywords across five categories (artificial intelligence, blockchain, cloud computing, big data, digital application) counted using Python + Jieba segmentation; total keyword frequency log‑transformed to form dig.
• Mediators:
  • business = ln(operating revenue) (business scale),
  • tfp = firm total factor productivity (estimated via OLS).
• Controls: Firm-level (size, age, top shareholder share, growth, leverage) and regional-level (provincial per-capita GDP, secondary/tertiary industry shares, average wages).
• Econometric specification: Panel regressions with firm fixed effects and year fixed effects. Baseline regressions show dig positively and significantly related to ln(employees) at the 1% level. Mediation analysis supports the productivity and business-scale channels.
• Descriptive stats (selected): employ mean = 7.4838 (sd 1.101); dig mean = 1.7349 (sd 1.462).

Implications for AI Economics

• Labor demand effects of digital/AI adoption are nuanced: this firm‑level evidence suggests net employment expansion via demand and scale channels, not simple displacement — but gains are concentrated in higher-skill segments and larger urban firms, implying distributional shifts.
• Role of productivity and scale: Models of AI/digital adoption should include endogenous effects on market size and firm expansion (demand amplification) as well as productivity gains that reallocate labor across tasks and occupations.
• Measurement: Textual analysis of corporate disclosures is a practical proxy for firm digitalization; useful for empirical AI-economics work but subject to disclosure bias (see limitations).
• Policy relevance:
  • Skills and training: Positive aggregate effects coexists with skill-biased gains — active upskilling and reskilling policies are needed to avoid rising inequality and to help workers transition into new roles.
  • Support for SMEs and less-developed regions: Benefits are concentrated in larger firms and cities; targeted policies can help diffuse digital adoption and job-creation potential more widely.
  • Labor market matching and information platforms: Digital platforms can reduce frictions and improve matching, reinforcing employment effects — policy can support such infrastructure.
• Research design lessons: Future AI-economics studies should pursue stronger causal identification (IVs, natural experiments) to address potential reverse causality (larger firms may both hire more and disclose more about digital initiatives) and measurement error in textual proxies.

Limitations to keep in mind - Sample restricted to listed private firms — results may not generalize to unlisted SMEs, informal firms, or other countries. - Digitalization proxy is based on disclosure keyword frequency (possible reporting/disclosure bias). - Potential endogeneity (reverse causality or omitted variables) is not ruled out by fixed effects alone; causal claims should be interpreted cautiously.

Suggestions for follow-up research - Use quasi‑experimental designs (e.g., local digital infrastructure rollouts, policy shocks) for causal identification. - Extend to unlisted firms and cross‑country samples to test external validity. - Disaggregate effects by occupation/task to map job creation vs. displacement more precisely (routine vs. non‑routine, cognitive vs. manual).