Rwanda exchange rate time series analysis using Garch model

Abstract

Volatility modeling and forecasts are essential tools to all financial sectors. This work focuses on weekly exchange rate data of the Rwf referred to the USD for a period of seven years spanning from 2012 to 2018. Through transformation of data from daily to weekly, we intended to reduce the size and adjust the outliers in data. Data obtained from the National Bank of Rwanda is analyzed using a non linear time series model. The aim of this Thesis is to apply a GARCH(1,1) model to describe the volatility of the data using visual inspections and statistic results comparison. Diagnostic check ensured the model accuracy. The main approach was to apply a Bayesian inference that uses MCMC method in unknown parameters estimation. Both visual inspection and basic statistics illustrate a good compatibility between simulated and observed data. The results obtained from the LSQ and MCMC methods are compared and found to be almost similar. An agreement between the model and actual data is obtained since the estimated model matched the real data . The estimated model is trusted for forecasting and is recommended to be applied by researchers and financial institutions. The model is used to predict exchange rate for the next 52 weeks. To find the results through data treatment software like MATLAB, R, Eviews and SPSS were used.