Abstract
Since the 1980s, two severe global waves of sovereign defaults have occurred in less developed countries (LDCs): the LDC defaults in the 1980s and the LDC defaults at the turn of the 21st century. To date, the topic is contemporary, while the forecasting and monitoring results of debt crises are still at a preliminary stage. This chapter explores whether the application of the Self-Organizing Map (SOM), a neural network-based visualization tool, facilitates the monitoring of multidimensional financial data. Thus, this chapter presents a SOM model for visualizing the evolution of sovereign debt crises’ indicators. The results of this chapter indicate that the SOM is a feasible tool for visualization of early warning signals of sovereign defaults.
TopIntroduction
Throughout the entire monetary history, sovereigns have repeatedly defaulted on their external debt. Ever since the famous serial defaults during the reign of Philip II of Spain, 1556–1598, governments have regularly repudiated on their external debt. In the wake of the massive increase in lending in the 1970s, a wave of sovereign debt crises washed over the developing world in the 1980s. More recently, starting from Russia’s default in August 1998, a new wave of sovereign defaults in less developed countries (LDC) has occurred. The latest wave of sovereign defaults was in fact preceded by the Mexican Tequila crisis in 1994. Furthermore, following the Asian crises from 1997–98, the Russian default triggered a severe series of debt defaults, including Turkey in 2001 and Argentina in 2002. Moreover, the history of serial defaults shows that countries, e.g., Spain and Portugal in the 19th century, have been allowed to become serial defaulters without suffering too much in terms of borrowing costs. The occurrence of sovereign debt crises correlates highly with the amount of research on the phenomena, thus causing a recent comeback in this particular field.
However, although most countries have recently shifted to floating exchange rates, resulting in infrequent currency crises, the early warning signal analyses have still mostly concentrated on currency crises. Moreover, the huge government deficits accumulated during the expansionary stimulus preceding the Sub prime crisis from 2007–2009 have arisen questions regarding the sustainability of some countries’ government debts. Examples of recent concerns are the Icelandic referendum on paying external debt obligations, the rescheduling of small parts of the Dubai debt and the huge fiscal deficits and the total amount of sovereign debt in the Euro area. Thus, the fact that sovereigns run into series of defaults may currently indicate an imminent serial wave of sovereign defaults. This being the case, finding effective tools for monitoring early warning signals of debt crises remains an important issue for both private financial market participants and policy decision-makers.
To downsize financial volatility and instability, the broad scope of preventive monetary policy options should be introduced in an early stage of the pre-crisis period. Discovering the optimal timing of the policies has been attempted by systematic monitoring of indicators of debt crises. The empirical literature has been mostly based on early warning systems (EWSs) using conventional statistical modeling methods, such as logit and probit models (e.g., Detragiache and Spilimbergo (2001) and Ciarlone and Trebeschi (2005)). The occurrence of financial crises is, however, explained by complex, non-linear interactions between non-normally distributed economic and financial variables (Arciniegas Rueda and Arciniegas, 2009). These types of non-linearities derive, for example, from the fact that crises become more likely as the number of fragilities increase. Thus, because of distributional assumptions, conventional statistical modeling techniques may fail in explaining these events. The novel predictive models attempt, however, to apply artificial intelligence for the prediction of financial crises (e.g., Fioramanti (2008)). However, Peltonen (2006) shows that independent of which type of the earlier mentioned models one utilizes, the results of a priori predictions of financial crises are still disappointing. Thus, although the utilization of intelligent techniques has increased a posteriori prediction accuracies to a minor extent, the interpretability of the monitoring systems has not been addressed appropriately. Thus, rather than building highly complex and mathematical models, this motivates the development of monitoring systems with clear visual capabilities and intuitive interpretability, contributing instead to visual tools enabling real human perception.
Key Terms in this Chapter
Moral Hazard: The lender is predisposed to the risk that the borrower can alter their behavior after the transaction has taken place, and thus the borrower’s probability of default might increase. Moral hazard is mostly a problem of asymmetric information and appears after money has been lent. However, high costs for recovering a debt by an enforcement order might make it, even though the lender is fully informed about the borrower’s activities, too costly to prevent moral hazard. Hence, asymmetric information is not necessarily the source of moral hazard.
Artificial Neural Network (ANN): An artificial neural network is a data analysis method which operation resembles a network of biological neurons. ANNs are composed of a system of nodes (equivalent to neurons of a human brain) which are interconnected by weighted links (equivalent to synapses between neurons). The outcome of the ANN is altered by changes of the links’ weights. The data is fed to the input layer and the result of the network is displayed by the output layer. The input nodes represent the independent or predictor variables that are used for predicting the dependent variables, i.e., the output neurons.
Exchange Rate Overvaluation: An exchange rate overvaluation may be defined in different ways depending on the addressed research question. A real exchange rate overvaluation is, in general, the deviation of the actual rate from an equilibrium value that would be appropriate. It may be calculated, for example, as the deviation of the actual real exchange rate from the trend, measured as the percent deviation of the exchange rate level consistent with PPP equilibrium from the nominal exchange rate.
Data Visualization: Data visualization techniques create visual representations, often in lower dimensions, of multidimensional data. In other words, data sets consisting of many analyzed variables can be monitored effectively and intuitively by graphical means. This is extremely important in exploratory analyses, since their main task is to improve the users’ comprehension of the research problem addressed.
Financial Crisis: A financial crisis is defined as a collapse of the financial system. A financial system refers to all financial institutions that perform intermediation of resources between lenders and borrowers, stock exchange institutions and central banks acting as lender of last resorts. A financial crisis often implies a sudden fall in the value of assets and/or financial institutions. The decrease in these values may cause significant threats to the general economic stability.
Debt Crisis: A debt default may be defined in several different ways depending on the research question addressed. In general, a sovereign debt default occurs when a borrowing country fails to meet the terms of an external debt payment obligation. The debt payment obligations consist of short-term debt, measured by debt liabilities with maturity of one year or less, and long-term debt, measured by debt liabilities with maturity of one year or more.
Early Warning System (EWS): An early warning system is a system that gives warning signals of an unwanted event. In this chapter, in order to avoid sovereign defaults, it helps identifying weak macroeconomic variable values. EWSs are most often based on conventional statistical modeling methods, which give a probability of an imminent unwanted event. In order to classify the probability as a warning signal, a threshold is put on the probability of this particular event. All probabilities above this threshold indicate an occurrence of this unwanted event.
Exchange Rate: The exchange rate between two currencies is the market price for one currency in terms of the other, i.e., the rate at which one currency may be converted into the other. Exchange rates may be either fixed (e.g., the gold standard) or floating. A floating exchange rate moves with the demand and supply on the international financial markets. In general, an increase in demand of the domestic currency causes an increase in the value of that currency, and vice versa.
Neuron: See the term Artificial Neural Network.
The Self-Organizing Map (SOM): The Self-organizing map is a non-parametric and non-linear neural network that explores data using unsupervised learning. The SOM can produce output that maps multidimensional data onto a two-dimensional topological map. Moreover, since the SOM requires little a priori knowledge of the data, it is an extremely useful tool for exploratory analyses. Thus, the SOM is an ideal visualization tool for analyzing complex time-series data.
Currency Crisis: A currency crisis, or balance-of-payments crisis, occurs when a country with a fixed exchange rate is force to either devalue or float its exchange rate because of massive capital outflows, and is often caused by speculative attacks. The fixed exchange rate is supported by international foreign reserves, and collapses when the pressure on the currency cannot be eliminated by central bank intervention, i.e., the central bank buying domestic currency.