Forecasting electricity demand using univariate time series volatility forecasting models : a case study of Uganda and South Africa.

Master of Science in Statistics. University of KwaZulu-Natal, Pietermaritzburg 2016.Different sectors of economies are significantly affected by the supply of electricity. However, with the available limited resources, supply and demand of electricity in Africa are strongly correlated. In order to efficiently improve electricity supply, its demand has to be accurately predicted. In this research, we analyse electricity demand in two cases; peak monthly electricity demand in Uganda from January 2008 to December 2013, and daily electricity demand for South Africa from 1st January 2004 to 30th June 2008, using ARIMA and ARCH/GARCH models. We use this data to forecast future demand for both countries in order to help policy makers in the electricity sector make decisions for sustainable development of both countries. GARCH models are introduced to correct the volatility found in South Africa's daily demand data. Results from the study show that; for Uganda, a seasonal ARIMA(0,0,0)(1,1,1)[12] model describes the data better, with RMSE of 4.872027 and MAPE of 2.347028, and gives better forecasts which display a continued increase in electricity demand for months ahead. For South African data, a seasonal ARIMA(1,0,1)(0,1,0)[365] describes the data better but a standard GARCH(1,1) with normally distributed error terms accommodates volatility. Therefore, a combination of the two models produces better forecast accuracy