MODELLING THE TREND OF FLOWS WITH RESPECT TO RAINFALL VARIABILITY USING VECTOR AUTOREGRESSION.

Vector Autoregressive (VAR) models for multivariate time series have been extensively developed and used in econometrics. However, use of VAR models outside of these areas is rather limited. This paper models the trend of flows at Bui with respect to rainfall variability and investigates whether there is any causality (Granger or instantaneous) between rainfall and flow in relation to forecasting. Monthly data on rainfall and flows from January 1954 to December 2005 was obtained from Bui Power Authority in Ghana. Both unrestricted and Bayesian VARs were estimated and compared in order to select the best VAR model for forecasting and structural analysis. The results showed that the unrestricted VAR model outperforms the Bayesian VAR model in terms of forecasting flows and rainfall at Bui. Results from the structural analysis also revealed a two-way causality from rainfall to flow and vice-versa. We conclude that modelling flows and rainfall together at Bui improves the forecasting of both of them.