Forecasting demand for special telephone services

Future demand for services or goods is usually forecasted by fitting ARIMA models and using the optimal rules based on the squared error criterion. When analyzing a large number of time series describing Special Services in the telephone business, we found that a model with independent increments with stable distributions was more suitable and led to better predictions. It also described forecast errors adequately. This paper discusses the model, compares it with a state space model which is currently used for the problem, and applies several data analytic procedures to assess how well the model fits the data. A few remarks on the use of estimated forcast error size conclude the paper.