DOI: https://doi.org/10.36347/sjebm.2026.v13i07.001

Pages: 358-372

There has been an urgent need for alternative mechanisms on which to test policies and models that fail to predict systemic crises and to capture emergent economic phenomena, such as Dynamic Stochastic General Equilibrium (DSGE) models. The 2008 global financial crisis and the macroeconomic shocks of the Covid-19 pandemic have highlighted the many fragilities of the representative-agent assumptions, such as the fact that economic systems are really systems of interactions between heterogeneous actors, with non-linearities and decentralisation at double level. We suggest, design, and try out a new test-bed for fiscal and monetary policy interventions "Synthetic Economies" before users make them on this planet: Artificial Economies in which boundedly-rational and heterogeneous agents populate virtual nations and economies, where we test the guidelines for economic policy perfectly. The design used was sequential explanatory mixed methods design. The quantitative phase employed a multi-agent reinforcement learning (MARL) simulation platform, run over a simulated sample of 612 economies, 47 country calibration periods, and spanning the period 2010-2024, where the input data consisted of OECD and World Bank macroeconomic panel data. Systematic tests of fiscal policy interventions, such as progressive tax, government spending multipliers, and countercyclical expenditure rules were carried out. Macroeconomic policymakers and computational economists were the population for the qualitative phase, which involved 38 semi-structured interviews. A total of 11 hypotheses were tested in Structural Equation Modelling (SEM) using R-lavaan and PLS-SEM using SmartPLS. A post-hoc G*Power analysis was performed to make sure that the study was statistically powered (1 - β = 0.95; α = 0.05). AI-based fiscal policy simulations earned a higher score than Saez optimal taxation policies on the equality-productivity trade-off index, by adding 18.3% (β = 0.61, p < 0.001). The likelihood of sim