Title: Vector autoregressive models with multivariate skew innovations
Authors: Yetkin Tuac - Ankara University (Turkey) [presenting]
Yesim Guney - Ankara University (Turkey)
Senay Ozdemir - Afyon Kocatepe University (Turkey)
Olcay Arslan - Ankara University (Turkey)
Abstract: Multiple time series analysis can be done with the help of vector autoregressive (VAR) models and the parameter estimation of these models are usually done under normality assumption. Since normality assumption is too restrictive for real data analysis, some heavy tailed alternatives, such as the $t$ distribution, have been proposed in literature as the distribution for innovations. However, real data examples show that if the skewness is present, then the parameter estimation from normal or heavy tailed symmetric distributions will be produced inefficient estimators for the parameters of a VAR model. Therefore, if skewness and/or heavy taildness is a concern asymmetric and/or heavy tailed distributed innovation of a VAR model should be considered as alternatives to the symmetric distributed innovations.