Bootstrap-basedmodel selection has been shown inmany practical instances to be superior to classical methods such the AIC and MDL. This is particularly noticeable when the distribution of modelling noise is unknown and/or when the available data samples are small. One of the main problems of...