MODELLING AND FORECASTING ELECTRICITY CONSUMPTION OF ETHIOPIA: GRANGER CAUSALITY AND BAYESIAN VECTOR AUTOREGRESSIVE APPROACHES

Abstract

Ethiopia is aggressively working on production of electricity from different energy mix setting to have installed capacity of more than 17, 000 MW electricity after 2020 to trigger the economy. However, currently the electricity consumption demand is unparalleled with supply. Besides this, the empirical studies of causality between electricity consumption and economic growth in Ethiopia found mixed results both in direction and magnitude of impact. Therefore, the objective of this study was to model and forecast electricity consumption of Ethiopia using Granger causality and BVAR approaches. Bivariate and VECM Granger causality analysis were employed to investigate the causality between electricity consumption and economic growth of Ethiopia over the period of 1981 to 2015. Both results revealed the presence of a bidirectional causality between electricity consumption demand and economic growth. Besides this, Granger causality between electricity consumption and economic growth were decomposed in to different time horizons. Then, predictive power among variables were evaluated using single univariate model and nested models. The result revealed that inclusion of lagged economic growth in a nested model including long run relationship increases predictive power in making forecast of electricity consumption and vice-versa in Granger causality framework. On the other hand, the different versions of BVAR models were derived from various combination of the overall tightness and relative weights of other variables between stated ranges of hyperparametres and fixing decay parameter as 0.5. Based on out of sample forecast, BVAR1, Minnesota (0.5, 0.9 and 0.5) is selected as best performing BVAR model. Then, the forecasting power of BVAR1 model was compared with the standard UVAR and Univariate VAR models, the one for which the relative weight of other variables collapses to 1 and 0.0001 respectively using RMSE and RMSE based DM test statistic. The results show that BVAR1 outperform both models. This implies that adding proper macroeconomic information, that is, inclusion of appropriate prior information to the VAR model significantly improves the forecasting ability of electricity consumption. This encourages that for an electricity sector like Ethiopia, where data are not available for longer periods, BVAR approach could provide a competitive alternative approach. Finally, the results of forecasting using BVAR suggests that electricity consumption per capita were expected to be increasing in the coming five years, which is slightly higher than in the past three decades. This calls for greater investment from either the national fiscal budget and/or through private sector participation on electricity generation, transmission and distribution.