Can Regulation Curb AI Adoption Narrative–Practice Decoupling? An Agent-Based Model of the A-Share Market (Research Square)

Summary

• The study builds an agent-based simulation model using annual-report text and financial panel data on 3,133 A-share listed firms from 2021–2024.p.4
• In a baseline market with no regulation (λ = 0), natural pricing discipline against narrative decoupling is weak, resulting in a response coefficient of βD ≈ − 0.37.p.2
• A Pareto search across 300 candidate combinations shows that combining repeat-offender escalation and targeted enforcement achieves βD = − 2.125 and appears in 85.2% of non-dominated solutions.p.12
• Component ablation demonstrates that the escalating multiplier 2s−1 contributes approximately 56% of the strength in repeat-offender escalation.p.11

Key insights

1. Market pricing discipline alone is insufficient to punish AI narrative decoupling, showing a baseline response coefficient βD of − 0.369.p.10
2. Repeat-offender escalation achieves the highest enforcement efficacy among single mechanisms, increasing βD to − 2.065 at intensity λ = 0.3.p.10
3. Structural mechanism design influences deterrent effectiveness more than raw penalty size, as targeted enforcement penalizing only positive dissonance avoids signal contamination.p.12
4. Short selling and label-only disclosure coincide exactly with the no-regulation baseline, generating zero additional deterrent effect under simulated A-share institutional conditions.p.2
5. Regulatory deterrence is non-monotonic in the number of combined mechanisms, with 2 to 4 mechanisms performing best and six or more creating regulatory congestion.p.12
6. The intertemporal escalating multiplier 2s−1 provides about 56% of the deterrent power in repeat-offender escalation.p.11
7. Excluding 312 software and information-technology service firm entries leaves mechanism rankings unchanged, confirming conclusions apply to operational AI adopters rather than AI producers.p.14