- Title
- Detection of early warning signs of currency crises in South Africa
- Creator
- Gondoza, Gladys Nicola Fernandes
- Subject
- Financial crises -- 21st century
- Date Issued
- 2018
- Date
- 2018
- Type
- Thesis
- Type
- Doctoral
- Type
- PhD
- Identifier
- http://hdl.handle.net/10948/30238
- Identifier
- vital:30905
- Description
- In a world characterised by globalisation, particularly increased financial integration and capital mobility, international economic theory stipulates that countries rather maintain a floating exchange rate system than a fixed exchange rate system in order to have less susceptibility to currency crises (Glick & Hutchison, 2011). South Africa, the economic powerhouse of Africa, is an interesting case to examine. It has a floating exchange rate and should thus be more resistant to currency crises due to market adjustment expectations that limit the build-up of pressure in its foreign exchange markets. South Africa’s foreign exchange market is characterised as volatile with recurring turbulent periods with currency crises observed in 1996, 1998, 2001 and 2008, of which the 2007/2008 global financial crisis was the worst the world had experienced since the Great Depression of the 1930s and it had a significant, negative impact on the South African economy and certainly exposed the country’s vulnerably (South African Reserve Bank, 2012). Having experienced these periods of currency crisis in South Africa and with no specific tool adequately tested and developed for the South African economy to accurately detect such an event before its occurrence, this research was an attempt to fill this gap within the economics discipline. The purpose of this thesis was to examine and make use of Early Warning System (EWS) models to ascertain which one best identifies potential early warning signs of a currency crisis in South Africa. To achieve this, the study tested two standard and commonly used EWS models, namely the Signals and probit models. Added to these approaches, two newer EWS models, namely the Markov regime switching model and the artificial neural networks model were tested. To date only two studies on EWS models for currency crises have been conducted in South Africa. Knedlik (2006) used the signals approach and Knedlik and Scheufele (2007) used the signals, probit/logit and Markov regime switching approaches. Both studies recommended that further research was needed. With this in mind, this thesis built on these studies by extending the sample period under observation from 1993/02 to 2017/03 to fully capture the probability of the global financial crisis of 2007/2008. This study separated the sample period into two parts, a first period (1993/02 – 2004/12) catering for the July 1998 and December 2001 crises and a second period (2005/01 – 2017/03) catering for the October 2008 crisis. This was done to separately observe how well the models detected early warning signs of the October 2008 crisis due to its global nature. By exploring the potential of artificial intelligence by employing the non-parametric approach of artificial neural networks, which has not yet been applied in the South African context for the probability prediction of currency crises, and comparing its prediction performance to the signals, the probit and the Markov regime switching EWS models, this thesis fills an existing information gap. This study found that of these four EWS models for predicting the probabilities of currency crises within the 24-month crisis window, the signals model performed better than the other models for the period 1993/02 – 2004/12. However, the final-outcome of the best model in probability prediction of South African currency crises is not straightforward for this period, as the artificial neural network model and Markov regime switching model performed almost as well as the signals model. During the period 2005/01 – 2017/03, the artificial neural networks model outperformed the other three models in capturing the global financial crisis of 2007/2008, specifically with regard to the evaluations of the percentage of pre-crisis periods called correctly and the percentage of tranquil periods called correctly. As the cut-off probability increases, the artificial neural networks model is the superior model and is not closely followed by the other models. The artificial neural network model also indicated a stable / tranquil economy during the period following the global financial crisis (from about 2009 – 2017), which is a true reflection of that period. The findings of this study suggest that the artificial neural network model is a powerful tool in the probability prediction of early warning signs of currency crises in South Africa.
- Format
- vi, 242 leaves
- Format
- Publisher
- Nelson Mandela University
- Publisher
- Faculty of Business and Economics Sciences
- Language
- English
- Rights
- Nelson Mandela University
- Hits: 809
- Visitors: 760
- Downloads: 108
Collections
NMU School for Economics, Development and Tourism
| Thumbnail | File | Description | Size | Format | |||
|---|---|---|---|---|---|---|---|
| View Details Download | SOURCE1 | GLADYS NICOLA FERNANDES.pdf | 3 MB | Adobe Acrobat PDF | View Details Download |