Designing an early warning system of sovereign debt crises using Bp neural networks

This paper has developed an early warning system of sovereign debt crises based on Bp neural networks. Using data related to macroeconomic, debt burden and debt repayment history of 54 developing countries or less developed countries from 1991 to 2006, empirical result reveals that the system can predict sovereign debt crises in next three years effectively and its overall result is 86.7%. At same time, predictive comparison with binary Logistic finds that using BP neural network to predict sovereign debt crisis is relatively advanced to binary Logistic method.