| At present, the sovereign debt crisis happens frequently. The impact of the previous round of sovereign debt has not disappeared, another round of sovereign debt crisis is going to ourbreak. The Iceland debt crisis in October2009, Dubai debt crisis in November, Greek debt crisis in December,and intensified European debt crisis. All above incidents indicate that the sovereign debt crisis has become a major threat to the stability of the world economy. Therefore,the research of sovereign debt crisis early warning system is becoming very urgent and important.But when we review the existing papers,we will find the most theoretical and empirical of the early warning system were about currency crisis and banking crisis,there were only a few papers related to the sovereign debt crisis early warning model. In contrast, the reality is that the possibility of currency crisis has been significantly reduced because of the flexible exchange rate regime,and the possibility of bank crisis becomes negligible because of the implementation of the Basel agreement and the strengthening of banking supervision. The sovereign debt crisis become the biggest threat for global economic stability. So we do the research of sovereign debt crisis is of great significance.This paper consists of six chapters.The first chapter is the introduction to the paper, mainly including the background and significance,the main research ideas and methods, as well as innovation of this paper.The second chapter is about theoretical analysis of the sovereign debt crisis. We define the sovereign debt is Sovereign nations use their own sovereignty as collateral to borrow money, the creditors can be countries, also can be the international institutions (such as the IMF and the World Bank,etc). When we define sovereign debt crisis,we find scholars have different definitions,scholars make different definitions to the debt crisis based on different research methods. This paper draws on Manasse (2003) definition of a sovereign debt crisis:A country is defined to be in a debt crisis if it is classified as being in default by Standard&Poor’s, or if it has access to nonconcessional IMF financing in excess of100percent of quota. Next the paper simply introduces the theory of the three generations of the sovereign debt crisis.The three generations of the sovereign debt crisis are Owen Fisher’s "Debt-Deflation" theory, Willfenshen "Declines in asset prices" theory and Suter "Comprehensive international debt" theory.The third chapter is the introduction of the existing crisis early warning model. We first introduced the classic currency crisis early warning model,Classical currency crisis early warning model mainly includes non-parametric Signal Approach(or KLR),FR modle,STV cross-country model,Simple Logit hysteresis model based on macroeconomic and financial data and Artificial Neural Networks modle.Next we introduce a few common sovereign debt crisis early warning model,we summarized the main methods used for the early warning system and compare the advantages and disadvantages of these methods. We found that the ANN model in the comparative advantage of the early warning system.The fourth chapter is the data description and construction of the early warning indicator system. We selected31developing countries which frequent outbreak of the sovereign debt crisis, Data period selection is from the1980s to the early of this century. This paper draws on Kanminsky(1998) definition of a data window:If the sample country within the following24months has been a sovereign debt crisis, the value of the indicator is l,if not0. According to the database, we collected651samples,of which290sent a warning signal,another361did not. In the construction of the early warning indicator system,we innovative divided the early warning indicators into four categories,they are Monetary crisis early warning indicators, Macroeconomic indicators, Debt indicators and Income indicators.These four categories indicators explained the possibility of a debt crisis from different aspects.We use the Significance test and Multicollinearity test to the indicators screening. After screening, We selected10indicators,they are percentage deviation of the exchange rate from its trend,current account balance in percent of GDP, exports in billions of US dollars, london interbank overnight rate(LIBOR),US treasury bill,percent change in CPI (year-on-year),real GDP growth (year-on-year, in percent),FDI inflows in percent of GDP,public interest payments on short-term external debt in percent of GDP,public debt service on long-term external debt in percent of GDP.The fifth chapter is to build a sovereign debt crisis early warning model based on BP neural network.This chapter is the core part of the paper. we will first do a simple introduction of BP neural network,then we will design and create a BP neural network early warning model. The debt crisis early warning model we established is use of Matlab2011b in neural network toolbox. The main parameters of the BP neural network model we designed are:an input layer,a hidden layer,a output layer;10input nodes,16hidden layer nodes,1output nodes; the transfer function of the input layer to the hidden layer is logsig function, the transfer function of the hidden layer to the output layer is logsig function;learning function is learngdm function;training function is trainlm function;the maximum number of training is3000,the training objectives accuracy is0.01and the learning rate is0.01.Considering the lack of data has a great influence on the accuracy of training,so we excluded the samples with data missing. Finally the number of sample whice can be used for the training and testing is only568.And we divided these samples into456training samples and112testing samples. Next we write programs, and run the programs in Matlab2011b. First we take456training samples into model for training,when the network model training to37step, the model achieve the accuracy requirements of the training objectives, the training is over. Second we take112testing samples into model for testing. The test results show the network model achieves91times correct judgment and21times error judgment,the correct rate is81.25%.We using binary logistic model for comparison, the binary logistic model achieves88times correct judgment and24times error judgment,the correct rate is78.57%.The correct rate of network model is2.68%higher than binary logistic model.The sixth chapter is the conclusion of the paper.The contribution of the paper:1. The theme of the paper is the sovereign debt crisis. The possibility of currency crisis has been significantly reduced because of the flexible exchange rate regime.and the possibility of bank crisis becomes negligible because of the implementation of the Basel agreement and the strengthening of banking supervision. So the research of sovereign debt crisis is becoming very urgent and important.2. We selected early warning indicators based on scientific, comprehensive and operability. In the construction of the early warning indicator system,we innovative divided the early warning indicators into four categories,they are Monetary crisis early warning indicators, Macroeconomic indicators, Debt indicators and Income indicators.3. We use BP neural network method into warning the sovereign debt crisis,and use empirical research method to prove the validity of the model.We using binary logistic model for comparison.The correct rate of network model is2.68%higher than binary logistic model.The deficiency of the paper:1. Limitations of the sample, Data period selection is from the1980s to the early of this century in which the sovereign debt crisis frequent outbreak,but we ignored the outbreak of the sovereign debt crisis in the past decade.The data missing is very serious,and it may reduce the accuracy of the training.2. We design the BP neural network model withour a certain rules to follow.The only way to construct the model is performing a large number of experiments.3.The correct rate of BP network model is81.25%, it not achieve our expected height. |
PDF Full Text Request