New autoregression model selection criterion improves time series prediction accuracy.
A new method for figuring out the order of a model for time series data has been developed. This method combines the strengths of two existing techniques, the Akaike information criterion and the Bayesian information criterion. It can be consistent like the Bayesian criterion or efficient like the Akaike criterion, depending on the true model behind the data. This new method is more flexible and reliable when there is no prior knowledge about the data's model. Tests show that it works well with different datasets.
