Since the earlier contribution of Box and Jenkins , identification and estimation of linear models have become standard statistical tools for time series analysis . By means of a real application, it is seen how ARMA forecasts can be improved when nonlinearities are present . However, it is widely recognized that the class of ARMA models may fail to capture fully the dynamics of real phenomena since these are often characterized by strong nonlinear components. For this reason, it is important that any preliminary analysis (or evaluation of model adequacy) includes a check on the linearity of the generating process. One of the challenges of modern time series analysis is to develop tests that are capable of detecting nonlinear structure.