Multivariate Time Series Modelling with Seasonal Univariate Components; Evidence from Nigeria GDP

Abstract

The patterns of GDP variables are graphically examined using time plot presented the time plot for the GDP variables concerning time presented a combined single time plot for all the considered GDP variables. The relationship, as well as the degree of relationship between/among the GDP variables, was further revealed by computing the pairwise correlation. Based on the output, each variable when crossed classified with itself have a strong positive correlation with an output of (1), while pairwise correlation reveals a positive figure with the least estimate being (0.3149), this implies that for all the variables there exist a positive correlation. All the pairwise relationship reveals a strong positive association with all the estimates revealing a value between (0.8-0.9) except ‘trade and industry' that shows a positive relationship but not strong with an estimate of (0.3149). The initial test in fitting a time series model is to examine the series for stationarity. The Augmented Dickey-Fuller test revealed that ‘Agriculture’, ‘Construction’, and Services’” satisfies the requirement of stationarity while the series ‘industry and “Trade” are non-stationary initially but later became stationary after the application of the first difference transformation which was confirmed after the application of the ADF test to the first differenced series. The Johansen co-integration's Trace test was employed to determine the order of co-integration and it was revealed that the series are cointegrated hence presentation of the equation of integration. We presented the lag length estimation criteria which revealed that the lag length of order 5 is appropriate for the VAR model as suggested by Akaike Information Criteria (AIC), Hannan-Quinn (HQ) Information Criteria, Schwarz Information Criteria (SC). The VAR(5) model was fitted for all the considered GDP variables.