Predicting The Effectiveness of Monetary and Fiscal Policies Under Financial Sanction: A Machine Learning and SHAP Analysis

Financial sanctions restrict access to international financial markets and disrupt trade flows, thereby exerting persistent pressure on macroeconomic performance in sanction targeted economies such as Iran. This study develops a data-driven framework to evaluate the effectiveness of monetary and fiscal policies in mitigating these adverse effects under prolonged financial sanctions. To capture nonlinear, regime-dependent, and potentially predictable relationships among key macroeconomic variables, we employ advanced machine learning techniques-including Logistic Regression, Random Forest, XGBoost, and LightGBM-combined with SHAP-based interpretability analysis. Using monthly data covering the period 2006-2024 and a time-based validation strategy, the models are trained to predict policy effectiveness as reflected in three core macroeconomic indicators: real GDP growth, inflation, and the exchange rate. The empirical results show that tree-based machine learning models substantially outperform the linear benchmark. XGBoost delivers the strongest predictive performance, achieving an accuracy of 91% and an ROC-AUC of 0.94, followed by LightGBM (88% accuracy; 0.92 ROC-AUC), Random Forest (85% accuracy), and Logistic Regression (75% accuracy). These findings indicate that policy-outcome relationships exhibit systematic and predictable nonlinear patterns that are better captured by flexible algorithms. SHAP-based analysis identifies the nominal interest rate, the government spending-to GDP ratio, the tax-to-GDP ratio, money supply growth, and a sanctions regime indicator as the most influential determinants of policy outcomes. Financial sanctions are associated with contractionary effects on output while simultaneously intensifying inflationary and exchange rate pressures, thereby altering the transmission mechanisms of monetary and fiscal policies. Although the analysis is predictive rather than causal, the results demonstrate the clear advantage of machine learning methods in capturing complex interactions in sanction-constrained macroeconomic environments. Overall, this study highlights the practical relevance of integrating machine learning with explainable artificial intelligence tools to support transparent, data-driven macroeconomic policy design. The findings provide actionable insights for policymakers by identifying which policy instruments retain relative effectiveness under sanctions and how policy transmission mechanisms adjust across sanction regimes.