Forecasting twin and triple financial crises: global evidence on macro-financial vulnerabilities

Purpose This study develops an integrated forward-looking framework for forecasting compound twin and triple financial crises and identifying optimal probability thresholds for macroprudential surveillance. Using forward-looking logit models estimated on annual panel data for 196 countries (1980–2023), the study examines how macro-financial indicators predict compound crisis risk across different income groups. The objective is to improve early-warning systems by providing a practical decision-support framework for central banks and international institutions. Design/methodology/approach The study employs forward-looking logit models estimated on annual panel data for 196 countries over the period 1980–2023 to forecast compound banking, sovereign debt, and currency crises two years ahead. Standardized macro-financial indicators are used as predictors, including private credit growth, government debt growth, international reserves, inflation, real GDP growth, and real exchange rate misalignment. Predictive performance is evaluated using AUROC analysis, while optimal probability thresholds are identified under alternative misclassification costs and compared across advanced, emerging market, and low-income developing economies. Findings The findings reveal significant nonlinearities in the relationship between macro-financial imbalances and compound financial crises. Macro-financial indicators exhibit different predictive power across advanced, emerging market, and low-income developing economies, highlighting substantial cross-country heterogeneity. The proposed threshold-optimized framework improves the identification of crisis risk by combining AUROC-based validation with asymmetric misclassification costs. The results suggest that compound crisis prediction can strengthen macroprudential surveillance by providing practical decision-support tools for monitoring systemic vulnerabilities and supporting timely policy interventions. Research limitations/implications The study is subject to several limitations. Although it provides forward-looking two-year-ahead forecasts, it does not perform formal out-of-sample validation, and the rare-event nature of triple crises presents estimation challenges. Future research could incorporate global financial factors, alternative forecasting horizons, rare-event estimation techniques, and machine learning methods to further improve predictive accuracy and strengthen the operational usefulness of early-warning systems for compound financial crises. Practical implications The proposed framework provides an operational decision-support tool for central banks, finance ministries, the IMF, and the World Bank to strengthen macroprudential surveillance. By combining estimated crisis probabilities with optimal probability thresholds, it can support financial stability monitoring, stress testing, reserve adequacy assessments, and supervisory decision-making. The framework complements existing surveillance systems by helping policymakers identify rising macro-financial vulnerabilities and implement timely, country-specific preventive measures before compound financial crises materialize. Social implications By improving the early identification of compound financial crises, the proposed framework can support more timely policy responses that enhance financial stability and economic resilience. Earlier detection of systemic vulnerabilities may help reduce the economic and social costs of financial crises, including unemployment, poverty, fiscal stress, and financial exclusion. More effective macroprudential surveillance can contribute to sustainable economic development and strengthen public confidence in financial institutions and economic policymaking. Originality/value This study develops a unified, forward-looking framework for forecasting compound twin and triple financial crises across 196 economies. Unlike previous research that typically examines banking, sovereign debt, or currency crises separately, it jointly models compound crises, identifies operational probability thresholds using AUROC-based validation and asymmetric misclassification costs, and compares predictive performance across advanced, emerging, and low-income economies. By linking statistical forecasting with macroprudential surveillance, the study provides a practical early-warning framework for central banks and international financial institutions.