Abstract
The frequency of currency crises has increased drastically in the last 30 years, and the scale and impact of these episodes has greatly renewed interest in the existing literature and stimulated a growing volume of new literature on both theoretical and empirical sides. These offer quite different explanations for currency crises which the thesis attempts to assess. Firstly, this project looks at the experiences of countries in Latin America, Europe and East Asia during 1970-2001. The aim is to examine how good the so-called 'three generation' crisis theoretical models are in explaining the origins of different currency crises where they were originally inspired. Instead of imposing an arbitrary threshold a priori, we employ the non-linear SETAR (Self Exciting Threshold Autoregression) model and Markov-switching model to identify the crisis incidents. The Limited Dependent Variable Model-Logit Model is adopted to link the defined crisis variable to a wide range of selected variables based on crisis theories and empirical works. The estimated results show limited evidence that the three generation models efficiently address the different features of currency crises in three regions. The overall imperfect picture lays down the task for our further research. In the following chapter, we focus on the issue of credibility within the approach of the second-generation models. In an extended model, we show that whether a strenuous defence of a parity would boost its credibility is ambiguous and depends on the importance of the 'debt-burden' effect relative to the 'signalling' effect. In other words, resisting a crisis could show the authority's resolution not to devalue, and thus enhance the expectation that the regime will be maintained in the future. But at the same time, refraining from inflationary financing increases the debt burden and hence the likelihood of a forced future devaluation. Two testable predictions are derived regarding the impact of the debt maturity on the interest rates in two different types of countries-the debt-burden countries and signalling countries. We distinguish between them in our sample by observing different patterns in the interest rates when countries experience actual crises and/or successful defences. The actual devaluations are identified by the regime changes in the Reinhart-Rogoff index, while the successful defences are periods that are under severe attack (defined by both the SETAR and MRS model) but do not lead to actual crises. By using the GMM (General Method of Moments) for the dynamic panel data model, the estimated results show supportive evidence for the predictions. Finally, we re-examine the crisis transmission mechanism during the 1997/98 East Asia Crises. It is argued that the tests for contagion based on the conditional cross-market correlation coefficient are inaccurate and biased due to heteroscedasticity. We adopt the proposed formula to adjust for the bias and conduct our tests, for the first time, in the framework of a Markov-switching Vector Autoregressive (MS-VAR) model. It is shown that the adjustment has a significant impact on the tests, and the evidence of contagion that is found in the initial tests based on the conditional correlation coefficients soon disappears, after we adjust for heteroscedasticity. Instead, the highly correlated comovements amongst the East Asian markets during the 1997/98 crises is only a continuation of the strong economic linkages- interdependence between them-which exist in all states of the world.