Evaluating Exchange Rate Value at Risks Models for Fourteen African Currencies

• Main Authors: Nausheen Jandoo, Preethee Nunkoo Gonpot
• Format: Article
• Language: English
• Published: Danubius University 2018-11-01
• Series: Acta Universitatis Danubius: Oeconomica
• Subjects: volatility; value at risk; exchange rate; africa; garch
• Online Access: http://journals.univ-danubius.ro/index.php/oeconomica/article/view/4997/4681

The global foreign exchange market is undoubtedly the world's biggest market with huge trading volume, surpassing other markets including equities and commodities. This study focuses on exchange rate modelling where we perform an empirical study to evaluate models which can be used to identify a common Value at Risk (VaR) model for fourteen African currencies. The descriptive statistics of our data reveal the salient features common to financial time series which are nonnormality, high kurtosis, skewness and presence of heteroscedasticity except for one currency, the central African CFA Franc. The latter is excluded from the modelling exercise. We make use of GARCH, GJR-GARCH and FIGARCH to model volatility using four distributions: normal, student-t, GED and skew-t. Unconditional EVT and dynamic GARCH-EVT methodologies are also used for volatility modelling; both with static (S) and rolling windows (R). Results show that static window shows a better performance than rolling window. Unconditional EVT is seen to overpredict VaR and dynamic EVT is not among the best models. The GARCH (33.3%) and GJR-GARCH (38.5%) models produce better forecasts with a dominance for GJR-GARCH models. Despite the data being skewed, the normal distribution gives better forecast. We also observe that GARCH-S-Normal is suitable for Southern African Development Community (SADC) and FIGARCH for East African Community (EAC) countries. A geographical combination reveals the use of GJR-GARCH for Northern and Western African regions and GARCH-S-Normal for South African region. Despite not finding a unique model for all countries, it is interesting to note that different regions/communities can adopt a common Value at Risk model for forecasting purposes. Our results provide a full validation of the models under the different backtesting methods and thus could be implemented at the practitioner’s level.