An Evaluation of the Forecasting Performance of ARIMA Models for Seasonally Adjusted and Unadjusted Data

Merdi Ahmed Orsud

Abstract


This paper undertakes an evalution of the forecasting performance of univariate ARIMA model when the data are seasonally adjusted and when they are not. The method of adjustment is the U. S. Bureau of Census Method X-II. To attained stationarity, natural logarithm transformation for water flow amount at Eldaim Station was taken before univariate ARIMA models for the adjusted and unadjusted data constructed (during 1990 to 2006) and their forecasting performance upon the process of updating was compared (during 2001 to 2006). The paired t-test for the actual and predicted monthly averages shows no difference between the univariate ARIMA models for the unadjusted and the adjusted data. The comparison of the forecast error statistics obtained reveals that the forecasting performance of the models for the unadjusted logged series is better than that of the adjusted. Therefore, the use of unadjusted monthly data when constructing ARIMA models for forecasting or control of water flow amounts at Eldaim Station is better.

Key Words: ARIMA, Adjusted, Unadjusted, Bureau of Census Method X-II.


Full Text: PDF

Refbacks

  • There are currently no refbacks.