Forecasting

Chapter 4 downplays forecasting’s role in the design and analysis of time series experiments and emphasizes its potential abuses. While the “best” ARIMA model will outperform other forecasting models in the short and medium-run, long-horizon ARIMA forecasts grow increasingly inaccurate with diminished utility to the forecaster. Although the principles of forecasting help provide deeper insight into the nature of ARIMA models and modeling, the forecasts themselves are ordinarily of limited practical value. Forecasting can provide useful guidance to analysts choosing between two competing univariate models. While forecasting accuracy is only one of many criteria that might be considered, other things being equal, it is fair to say that a statistically adequate model of a process should provide reasonable forecasts of the future. Forecast accuracy depends on a host of factors, many of which lie outside the grasp of model adequacy. More important, forecast accuracy has no universally accepted metric.