The Box-Jenkins Methodology

Abstract

The Box-Jenkins approach to time series modelling consists of extracting predictable movements (or patterns) from the observed data through a series of iterations. The univariate Box-Jenkins method is purely a forecasting tool; no explanation is offered in that there are no regressor-type variables. The Box-Jenkins approach follows a three phase procedure:

• Model identification: a particular category of Box-Jenkins (B-J) model is identified by using various statistics computed from an analysis of the historical data.

• Model estimation and verification: once identified, the “best model” is estimated such that the fitted values come as close as possible to capturing the pattern exhibited by the actual data.

• Forecasting: the final model is used to forecast the time series and to develop confidence intervals that measure the uncertainty associated with the forecast.