Forecasting Meteorological Time Series Data with a Reduced Form Vector Autoregressive (Var) Model and Three Univariate Time Series Techniques: A Comparative Study

The prime advantages of reduced form multivariate time series modeling such as Vector Autoregression (VAR) over the univariate method is their applications to forecasting and policy analysis. Additional objectives of the multivariate time series analysis includes: To understand the dynamic relationship among time series data, and to improve accuracy for the future. This present work focused on the forecasting performances of the VAR model alongside with other three univariate time series techniques (Decomposition, Holt-Winter and SARIMA models) with application to meteorological data. To achieve this, monthly data on maximum temperature and relative humidity were collected from the meteorological unit of the NCRI, Badeggi. The characteristics of the data revealed that the time series data are stationary, and there is bi-directional causality between temperature and relative humidity. The result revealed a strong negative correlation between temperature and relative humidity. Furthermore, among the competing forecasting methods, the VAR forecasting technique revealed as the best. We concluded that the VAR forecasting technique has the possibility of improving forecast especially in the context of meteorological time series data.