Comparison of different statistical models for forecasting exchange rate of Somali shilling against us dollar

Abstract EN

The main goal of thispaper is to model and forecast the daily exchange rate of Somali Shilling (SOS) against United States Dollar (USD) over the period of 1^st January 2009 to 31^st December 2018 using Box-Jenkins models and Autoregressive Conditional Hetroskedasticity (ARCH) family models to compare between them and selected an appropriate one.

Box-Jenkins models are employed for modeling and forecasting data using the steps of Box-Jenkins methodology.

Additionally, non-normality, skewness, leptokurtosis, volatility clustering, and existence of ARCH effects in the residuals are observed in the data, therefore, ARCH family models which include ARCH, Generalized ARCH (GARCH), Exponential(EGARCH), and Threshold (TGARCH) are developed under three error distributions namely normal distribution, t-student distribution andGeneralized Error Distribution(GED).

The empirical analysis has shown that ARMA(0, 6) is the most appropriate model for the estimated models using Akaike Information Criteria (AIC) and Schwarz Information Criteria (SIC) as a selection criteria and also for the forecasted models usingRoot Mean Square Error (RMSE), Mean olute Error (MAE) andMean olute Percentage Error (MAPE) as forecasting accuracywhile estimating and forecasting the conditional variance of volatility models, it was found that ARCH(6) under t-student is the best model.

After comparing between the models, the result declared that ARCH family models are superior to Box-Jenkins.

Moreover, Diebold Mariano(1995) test is applied and revealed that the ARMA models and ARCH family models have same predictive ability whichimplies that the DM (1995) test does not prefer any model over the other.