Copyright © 2026 Authors retain the copyright of this article. This article is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

@article{184369,
        author = {Nirmala talape and AARYAN PRASHANT PATIL},
        title = {AI-Driven Policy Stress Testing in Synthetic Economies},
        journal = {International Journal of Innovative Research in Technology},
        year = {2025},
        volume = {12},
        number = {4},
        pages = {1435-1439},
        issn = {2349-6002},
        url = {https://ijirt.org/article?manuscript=184369},
        abstract = {Policy failures are costly, slow to detect, and politically difficult to reverse. This paper proposes a rigorous framework for AI driven policy stress testing using synthetic economies. large scale, agent based, data conditioned simulations populated by heterogeneous households, firms, intermediaries, and a learning government. We introduce a modular architecture in which (I) micro behavior is calibrated to empirical moments; (ii) markets clear through adaptive price discovery; and (iii) policy instruments are controlled by a reinforcement learning (RL) policy maker that is constrained by legal, fiscal, and equity guardrails. We define a family of stress regimes (supply shocks, demand collapses, climate events, banking runs, geopolitical trade disruptions) and propose counterfactual rollouts that quantify distributional impacts, macro stability, and long run resilience. Key contributions include: (1) a Policy Wind Tunnel Protocol (PWP) for ex ante evaluation under extreme yet plausible scenarios; (2) a Welfare Stability Frontier (WSF) that trades off social welfare and systemic volatility; (3) Bias Aware Learning (BAL) to prevent the RL agent from exploiting spurious correlations; and (4) a Civic Oversight Layer (COL) that makes model behavior auditable and democratically steerable. We illustrate the framework on three stylized policies. VAT adjustment, targeted UBI, and carbon dividends. and show how stress testing reveals nonlinear regime vulnerabilities that would be invisible to static models. The result is a policy design pipeline that is fast, transparent, and safer than trial and error in the real world.},
        keywords = {— policy stress testing, synthetic economies, agent based modeling, reinforcement learning, distributional effects, macroprudential design, welfare optimization, transparency},
        month = {September},
        }