The Causality Effect on Vector Autoregressive Model: The Case for Rainfall Forecasting

In recent decades, a stochastics modeling of weather prediction is gaining popularity among government and researchers. A devastating weather may have profound environmental, human and/or financial loss. It is a familiar fact that the series process over a fixed time horizon following historical time series model. A Vector Autoregressive (VAR) is one of time series models that explain current and past values of the multivariate variables as a linear model. Moreover, the instantaneous interaction between two or more variables deal with a causal relations between those variables. In particular application, observing the response of one variable to an impulse in the other is interesting, allowing a number of further variables as well. The explanation of impulse response is essential for a structural modeling, we explore a causality concept well known as Granger-Causality (G-causes) test. Thus, we shall examine the causal relations between rainfall and temperature, addressed by constructing and forecasting simultaneously bivariate VAR model. Henceforth, we consider the VAR(2) based on a parsimonious principle and Akaike Information Criteria (AIC). Evidently, the result present an exogenous shock of temperature has an effect for rainfall, we called temperature is "G-causes" for rainfall. Thus, we argue a temperature do Granger-cause for rainfall, and a vice versa is not properly true. A VAR model provide deeper and more accurate insight into rainfall prediction, shown by an RMSE 0.0021.