THE APPLICATION OF MULTIPLICATIVE GARCH-MIDAS TWO COMPONENT MODEL FOR SELECTED DAILY ECX COMMODITIES PRICE RETURN VOLATILITY

Abstract

Recently, modelling and forecasting of high frequency data (such as daily commodity price volatility) using GARCH-MIDAS component attracts the attention of many researchers. Following the same line, the objective of the present study is to apply Multiplicative GARCH MIDAS two component model for selected daily ECX commodities (Harar coffee, Jimma coffee and sesame) price return volatility over the period of 4-11-2009 to 30-12-2016. The GARCH-MIDAS component model decomposes the conditional variance as short run component which follows a mean-reverting GARCH(1,1) process and long run component which considers different frequencies of macroeconomic variables (in this study REER, fuel oil price and inflation rate) via Mixed Interval Data Sampling (MIDAS) specification using beta weight function. The results of ARCH effect tests on the residuals from the mean models revealed the existence of time varying conditional variance. Among the conditional variance models, GARCH (1, 1), GARCH (2, 1) and GARCH (1, 2) were identified for Harar coffee, Jimma coffee and sesame price return volatility, respectively. Engle and Ng tests show the insignificance of the asymmetric term, while Lundbergh & Terasvirta LM and the Li-Mak portmanteau tests from the residuals of GARCH models show the existence of time varying unconditional variance. From the result of the estimated GARCH-MIDAS component models, REER, fuel oil price and inflation rate were found to be the best drivers of Harar coffee, Jimma coffee and sesame price return volatility, respectively. Moreover, the estimated GARCH-MIDAS component models were used for out-of-sample forecasts by incorporating relevant macroeconomic variables. Finally, the MSE, MAE and DM test were used for evaluating and comparing the forecasting ability of GARCH-MIDAS component models against standard GARCH models. The results show that including low-frequency macroeconomic variables improves the forecasting ability of volatility component models.