STATISTICAL ANALYSIS OF MACROECONOMIC DETERMINANTS OF COFFEE PRICE VOLATILITY IN ETHIOPIA: APPLICATION OF GARCH-MIDAS MODEL

Abstract

Modeling and forecasting high frequency data such as daily commodity price volatility using GARCH type model attracts the attention of many researchers. Following the same framework, the objective of the present study is to apply the multiplicative GARCH-MIDAS model for daily exported coffee price as proxy of daily total coffee price of Ethiopia over the period of 1-1-2008 to 7-17-2018 with the purpose of fitting and forecasting coffee price return volatility. The GARCH-MIDAS model decomposes the conditional variance as short-term component, which follows the mean reverting GARCH (1,1) process, and longterm component, which consider different frequencies of macroeconomic variables. In this study exchange rate (nominal exchange rate), inflation rate (general inflation), interest rate (lending interest rate), fuel oil price (price of imported petroleum and petroleum production), total consumption and money supply (broad money) macroeconomic variables were employed through MIDAS specification using beta-weighting scheme to analyze impact of the variables on the long-term volatility component. The result of ARCH effect test on the residual from the mean model revealed the existence of time varying conditional variance for the selected mean model. A conditional variance model GARCH (1,1) was selected and used to model the conditional variance of coffee price return with Quasi Maximum Likelihood along with Bayesian estimation methods and both estimation procedures indicated the persistence of conditional variance observed even for small sample under Bayesian estimation framework. Engle and Ng test show the insignificance of the asymmetric term, while Lundbergh and Terasvirta LM and the Li-Mak portmanteau test from the residual of GARCH model shows the existence of time varying unconditional variance and made call for GARCH-MIDAS model. From the result of estimated GARCHMIDAS model, inflation rate and exchange rat were found to be the best drivers of coffee price volatility in Ethiopia. Moreover, the estimated GARCH-MIDAS component was used for in and out of sample forecast under classical estimation by incorporating the best driver macroeconomic variables. Finally, the MAE, RMSE and DM test were used for evaluating and comparing the forecasting ability of GARCH-MIDAS component model against standard GARCH (1,1) model. The forecasting result shows that including macroeconomic variables improves the forecasting ability of volatility model. From the empirical finding, exchange rate and inflation rate were positively influence the long-term volatility component, as result appropriate fiscal and monetary policy should be imposed, as correction measures to lighten inflation effect and stabilize exchange rate in the country.