STOCHASTIC MODELS FOR WIND SPEED TIME SERIES: A CASE STUDY

The idea of using a mathematical model to describe the behaviour of a physical phenomenon is well established, but in many problems we have to consider a time-dependent phenomenon for which it is not possible to write a deterministic model; nevertheless, it may be possible to derive a stochastic model. The models for time series that are needed for example to achieve optimal forecasting and control are in fact stochastic models, but the choice of a proper model is never straightforward. In particular, this paper is concerned with the problem of forecasting a time series that possibly exhibits long-memory features. It results that the fractionally integrated ARMA processes may provide an adequate representation of the actual process, but do not yield a satisfactory forecasting performance.