Forecasting performance of switching models in hydrological time series,

The non-linear structure of river flow time series can be adequately explained by regime switching models, however good fitting results do not guarantee an equal good forecasting performance. Aim of this paper is to evaluating an comparing different approaches to compute forecasts from regime switching models with respect to their predictive accuracy. To this purpose, different nonlinear prediction techniques are applied to a time series of hydrological data and their forecasting ability in predicting the future values of the observed series at different lead times are assess by means of a wide range of loss functions.