Modelling and Forecasting Currency Exchange Rates Using Hybrid ARIMA and ANN Models in High-Frequency Data

Introduction: Exchange rates are central to the economic as well as the financial systems and structures because they determine trade, investment and policy. This study focuses on the statistical analysis and forecasting of exchange rates for the Kenyan Shilling (KES) against regional currencies: The currencies include; Burundian Franc (BIF), Rwandan Franc (RWF), Tanzanian Shilling (TSHS) and Ugandan Shilling (USHS).

Methods: Significance tests and models of time series were used in the analysis with regard to trend and volatility of exchange rates. Descriptive measures which include mean, median, standard deviation and range were calculated for each of the currency pairs. Furthermore, to determine cycles and rate predictions, a more complex ARIMA (1,1,0) with drift model was applied. Various statistics namely AIC, AIC c and BIC were used to assess model performance.

Results: It was seen that the mean rate of KES/USHS was the highest at 27.0 and KES/RWF was the least at 8.99. The standard deviations suggested moderate fluctuation in the pairs with the KES/USHS exchanging at 2.26. The results of the ARIMA analysis were highly significant for coefficients (AR1 = 0.5359, drift = 0.0412), and feasible for Fit statistics (AIC = -123.57, BIC = -116.07). It also pointed out that there were high positive correlations between all the currency pairs analyzed; the values being above 0.98.

Discussions: The results also confirm narrow fixed relationships between KES and the regional currencies with fluctuation in KES/USHS. The applied ARIMA model allows for capturing the dynamics of exchange rate and provides accurate forecast. The result of this study can be useful for policymakers and financial analyst in order to comprehend behaviors of regional currencies.