CFE 2015: Start Registration
View Submission - CFE
A1176
Title: Markov switching autoregressive forecasts: Symmetric or asymmetric loss functions Authors:  Marcella Niglio - University of Salerno (Italy)
Fabio Forte - University of Salerno (Italy) [presenting]
Abstract: In nonlinear time series domain the forecast accuracy has been largely debated. It has been shown that, in some cases, the complex structure of nonlinear models does not guarantee more accurate predictions with respect to those obtained from the linear models of the ARMA class. The reasons can be differently attributed. Among them: the selected model is not able to catch the features of the series under analysis; the loss function selected to generate the predictors needs to be revised. We focus the attention on the second problem. In more detail, we investigate the forecast ability of some nonlinear models with switching structure. After showing how their predictors can be obtained by using symmetric and asymmetric loss functions (of the LinEx class) we investigate and discuss if (and when) the introduction of asymmetric loss functions can be of help to increase the forecast accuracy.