A COMPARATIVE STUDY OF ARIMA AND ARFIMA IN THE PRESENCE OF AUTOCORRELATED ERROR TERMS

This research explored the ARIMA and ARFIMA modeling under autocorrelated error terms with two distributions namely, normal and uniform distribution. The study revealed that between ARFIMA and ARIMA, ARFIMA is best for long memory in this regard. The study use data of Inflation rate and Exchange rate of Nigeria from 1987 to 2016 collected from Central Bank of Nigeria statistical bulletin and it further revealed that ARFIMA model has the lowest AIC and BIC values which align with the simulation analysis. More so, the study shows that ARFMA (1, 0.14, 0) can account for long memory for the inflation rate with normal distribution compare to ARIMA (1, 1, 0) whereas, ARFIMA (1, 0.15, 0) model can account for Long memory for uniform distribution compare to ARIMA (1, 1, 0) model of the same distribution. ARFIMA (1, 4.5 X 10-5, 0) model can account for long memory of Exchange rate with normal distribution compare to ARIMA (1, 1, 0) model of the same distribution and ARFIMA (1, 4.5 X 10-5, 0), can account for long memory with uniform distribution compare to ARIMA (1, 1, 0) model of the same distribution.