On the accuracy of forecasting telephone usage demand

This paper examines the accuracy of telephone-usage-demand forecasts produced by three state-of-the-art forecasting techniques: Box-Jenkins, Akaike state space, and an autoregressive spectrum estimation. The study considers 35 actual monthly demand time series and 300 simulated realizations. Principal results are that: (1) correct identification of the nonstationary behavior of telephone demand is crucial to forecast performance, (2) overparameterization or underparameterization of the stationary aspects of a process has little or no impact on the accuracy of the forecast, and (3) forecasts based on the naive random-walk method compare favorably to those produced by sophisticated techniques. Also, strengths and weaknesses of the investigated techniques are revealed through data analysis. It is further argued that the traditional method for assessing the accuracy of the usage forecast based on average busy season quantities is biased towards underforecasting.