Please use this identifier to cite or link to this item:
|Title||ON SELECTION OF AUTOREGRESSIVE ORDER IN CASE OF INCORRECTLY MODEL SPECIFICATION|
The purpose of this paper is to compare different autoregressive models performance in case of incorrectly model order determination. The autoregressive models are selected based on four important information criteria: the Estimated Mean Square Prediction Error (EMSPE), Akaike's Information Criterion (AIC), Schwarz's Bayesian Information Criterion (BIC), and the estimated white noise variance 2 ˆ w σ. We generate different models with a second order autoregressive model, and consider the robustness of model selection based upon incorrect order. Combining a statistical interpretation of the four important information criteria and the principle of model parsimony, a model selection strategy is proposed. The main finding from the simulation is that over specification outperforms in selecting the most appropriate model, especially when the sample size is small compared with the number of parameters to be estimated.
|Published in||Islamic Countries Society of Statistical Sciences|
|Item link||Item Link|
|Files in this item|
|Safi, Samir K._37.pdf||21.44Mb|