Possible applications of the Box-Jenkins method and other time series procedures for forecasting
Laun, H.
Anwendungsmoglichkeiten der Box Jenkins Methode und anderer Zeitreihenprognoseverfahren: 322
1984
Document Number: 394727
The Box-Jenkins method of identifying autoregressive integrated moving average models is applied to a simulated time series and is shown to be unsuitable for producing acceptable forecasts from series met in real conditions. A model identification procedure is then developed more strictly directed to forecasting quality. Autocorrelation coefficient factors provide information on which parameters necessary to improve forecast quality are identified, as do the Langbehn and Mohr procedures. A procedure is developed for introducing these parameters into the model in turn until a sufficiently comprehensive model is obtained to provide a satisfactory quality for forecast. The empirical data used to test the procedures are GFR pig production statistics for 1969-78.