User’s Guide : Information Criteria : Using Information Criteria as a Guide to Model Selection
  
Using Information Criteria as a Guide to Model Selection
As a user of these information criteria as a model selection guide, you select the model with the smallest information criterion.
The information criterion has been widely used in time series analysis to determine the appropriate length of the distributed lag. Lütkepohl (1991, Chapter 4) presents a number of results regarding consistent lag order selection in VAR models.
You should note, however, that the criteria depend on the unit of measurement of the dependent variable . For example, you cannot use information criteria to select between a model with dependent variable and one with log().