Abstract:
|
Traditional methods of fitting time series models are often based on various information criteria, such as the AIC, BIC and their bias corrected versions. In this talk, we consider some popular penalization methods that have been recently proposed in the statistical machine learning literature as alternatives for model fitting . We derive conditions that ensure selection of the correct model with high probability and investigate large sample properties of the resulting estimators. Results from a moderate simulation study will be used to illustrate finite sample performance of the proposed methods.
|