Bridging AIC and BIC: A New Criterion for Autoregression

  title={Bridging AIC and BIC: A New Criterion for Autoregression},
  author={Jie Ding and Vahid Tarokh and Yuhong Yang},
  journal={IEEE Transactions on Information Theory},
To address order selection for an autoregressive model fitted to time series data, we propose a new information criterion. It has the benefits of the two well-known model selection techniques: the Akaike information criterion and the Bayesian information criterion. When the data are generated from a finite-order autoregression, the Bayesian information criterion is known to be consistent, and so is the new criterion. When the true order is infinity or suitably high with respect to the sample… 

Figures and Tables from this paper

Consistent model selection criteria and goodness-of-fit test for affine causal processes

This paper studies the model selection problem in a large class of causal time series models, which includes both the ARMA or AR(∞) processes, as well as the GARCH or ARCH(∞), APARCH, ARMA-GARCH and

Order selection for possibly infinite-order non-stationary time series

This study proposes a two-stage information criterion (TSIC), and shows that TSIC is asymptotically efficient in predicting integrated AR models when the underlying AR coefficients satisfy a wide range of conditions.

Controlling the error probabilities of model selection information criteria using bootstrapping

The Error Control for Information Criteria (ECIC) method is presented, a bootstrap approach to controlling Type-I error using Difference of Goodness of Fit (DGOF) distributions.

On Statistical Efficiency in Learning

A generalized notion of Takeuchi’s information criterion is proposed and it is proved that the proposed method can asymptotically achieve the optimal out-sample prediction loss under reasonable assumptions.

Performance Evaluation of AIC and BIC in Time Series Clustering with Piccolo Method

Results shows that Bayesian information Criterion (BIC) is better than Akaike’s information Criteria (AIC) in time series clustering with Piccolo method.

Sensitivity and Specificity of Information Criteria

In some cases the comparison of two models using ICs can be viewed as equivalent to a likelihood ratio test, with the different criteria representing different alpha levels and BIC being a more conservative test than AIC.

A Penalized Method for the Predictive Limit of Learning

This paper studies a penalized model selection technique that asymptotically achieves the optimal expected prediction loss (referred to as the limit of learning) offered by a set of candidate models.

Analysis of Multistate Autoregressive Models

This paper proposes an inference strategy that enables reliable and efficient offline analysis of this class of time series, referred to as multistate autoregressive models, and provides theoretical results and algorithms in order to facilitate the inference procedure.

Regression model selection via log-likelihood ratio and constrained minimum criterion

Although the log-likelihood is widely used in model selection, the log-likelihood ratio has had few applications in this area. We develop a log-likelihood ratio based method for selecting regression



Order selection for same-realization predictions in autoregressive processes

This paper presents the first theoretical verification that AIC and its variants are still asymptotically elficient (in the sense defined in Section 4) for same-realization predictions, and shows that A IC also yields a satisfactory saute- realization prediction in finite samples.

Parametric or nonparametric? A parametricness index for model selection

A measure, parametricness index (PI), is developed to assess whether a model selected by a potentially consistent procedure can be practically treated as the true model, which also hints on AIC or BIC is better suited for the data for the goal of estimating the regression function.


It turns out that it is not possible to share the pointwise adaptationproperty of BIC and the minimax-rate adaptation property of AIC by any model selection method, and when model selection methods have difficulty in selection, model combining is a better alternative in terms of estimation accuracy.


This work proposes an extension of the focused information criterion (FIC) for model‐order selection, with emphasis on a high predictive accuracy (i.e. the mean squared forecast error is low) and illustrates the possibility of using the FIC for purposes other than forecasting, and explores its use in an extended model.

Finite sample criteria for autoregressive order selection

The special finite sample information criterion and combined information criterion are necessary because of the increase of the variance of the residual energy for higher model orders that has not been accounted for in other criteria.

Catching up faster by switching sooner: a predictive approach to adaptive estimation with an application to the AIC–BIC dilemma

The catch‐up phenomenon is identified as a novel explanation for the slow convergence of Bayesian methods, which inspires a modification of the Bayesian predictive distribution, called the switch distribution, which solves the AIC–BIC dilemma for cumulative risk.

Regression and time series model selection in small samples

SUMMARY A bias correction to the Akaike information criterion, AIC, is derived for regression and autoregressive time series models. The correction is of particular use when the sample size is small,

The Focused Information Criterion

A variety of model selection criteria have been developed, of general and specific types. Most of these aim at selecting a single model with good overall properties, for example, formulated via

Can the Strengths of AIC and BIC Be Shared

It is shown that in a rigorous sense, even in the setting that the true model is included in the candidates, the above mentioned main strengths of AIC and BIC cannot be shared.

Some recent advances in time series modeling

The aim of this paper is to describe some of the important concepts and techniques which seem to help provide a solution of the stationary time series problem (prediction and model identification).