Regression and time series model selection using variants of the schwarz information criterion

Neath, Andrew A.; Cavanaugh, Joseph E.

doi:10.1080/03610929708831934

Cited by 62 publications

(33 citation statements)

References 9 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…1, it has been previously studied under various names [14,16,20,28]. Additionally, the first term in Eq.…”

Section: Best Harmony Best Matching and Related Approachesmentioning

confidence: 99%

“…The first two levels are the ML learning for unknown parameter learning and model selection shown in the ML row of Table 1, which has been widely studied from various perspective as previously discussed in Sect. 1 [14,16,20,28]. The third level is evaluating or selecting an appropriate meta structure ℵ via q(X N |ℵ), i.e., the second term in Eq.…”

Section: Q(x| ) Q( )/Q(x |S) and Q(x |S) = Q(x| )Q( )μ(D )mentioning

confidence: 99%

“…Differences on q( ) and on techniques for approximating the integral result in different specific forms. Typical efforts include those under the names of Bayesian Information Criterion [20,28], Bayes Factors [14], the evidence or the marginal likelihood [16], etc. Also, the Akaike Information Criterion (AIC) can be obtained as a special case though it was originally derived from a different perspective [1,2].…”

Section: Type 3: Optimizations For Model Selectionmentioning

confidence: 99%

See 2 more Smart Citations

Machine learning problems from optimization perspective

2008

J Glob Optim

View full text Add to dashboard Cite

Both optimization and learning play important roles in a system for intelligent tasks. On one hand, we introduce three types of optimization tasks studied in the machine learning literature, corresponding to the three levels of inverse problems in an intelligent system. Also, we discuss three major roles of convexity in machine learning, either directly towards a convex programming or approximately transferring a difficult problem into a tractable one in help of local convexity and convex duality. No doubly, a good optimization algorithm takes an essential role in a learning process and new developments in the literature of optimization may thrust the advances of machine learning. On the other hand, we also interpret that the key task of learning is not simply optimization, as sometimes misunderstood in the optimization literature. We introduce the key challenges of learning and the current status of efforts towards the challenges. Furthermore, learning versus optimization has also been examined from a unified perspective under the name of Bayesian Ying-Yang learning, with combinatorial optimization made more effectively in help of learning.

show abstract

“…1, it has been previously studied under various names [14,16,20,28]. Additionally, the first term in Eq.…”

Section: Best Harmony Best Matching and Related Approachesmentioning

confidence: 99%

Section: Q(x| ) Q( )/Q(x |S) and Q(x |S) = Q(x| )Q( )μ(D )mentioning

confidence: 99%

Section: Type 3: Optimizations For Model Selectionmentioning

confidence: 99%

See 1 more Smart Citation

Machine learning problems from optimization perspective

2008

J Glob Optim

View full text Add to dashboard Cite

show abstract

“…The difference in BIC values for the two models serves as a large-sample approximation to −2 times the log of the Bayes factor (see [21] or [22]). BIC for spatial data is as follows…”

Section: Criteriamentioning

confidence: 99%

Model selection with misspecified spatial covariance structure

Wang

Zheng

et al. 2014

Journal of Statistical Computation and Simulation

View full text Add to dashboard Cite

The adaptive classification of the interference co-variance matrix structure for radar signal processing applications is addressed in this paper. This represents a key issue because many detection architectures are synthesized assuming a specific covariance structure which may not necessarily coincide with the actual one due to the joint action of the system and environment uncertainties. The considered classification problem is cast in terms of a multiple hypotheses test with some nested alternatives and the theory of Model Order Selection (MOS) is exploited to devise suitable decision rules. Several MOS techniques, such as the Akaike, Takeuchi, and Bayesian information criteria are adopted and the corresponding merits and drawbacks are discussed. At the analysis stage, illustrating examples for the probability of correct model selection are presented showing the effectiveness of the proposed rules.

show abstract

“…12 McQuarrie and Tsai (1998) indicate that "AIC is probably the most commonly used model selection criterion for time series data" (p. 2). On the other hand, Neath and Cavanaugh (1997) note that BIC "is often preferred over AIC by practitioners who find appeal in either its Bayesian justification or its tendency to choose more parsimonious models than AIC" (p. 559) For an interesting discussion on this topic from an online forum for "statisticians, data analysts, data miners and data visualization experts," see http://stats.stackexchange.com/questions/577/ is-there-any-reason-to-prefer-the-aic-or-bic-over-the-other. of 0.5.…”

Section: Statistical Perspectivementioning

confidence: 99%

Natural Expectations, Macroeconomic Dynamics, and Asset Pricing

Fuster¹,

Hébert²,

Laibson³

2012

NBER Macroeconomics Annual

View full text Add to dashboard Cite

How does an economy behave if (1) fundamentals are truly hump-shaped, exhibiting momentum in the short run and partial mean reversion in the long run, and (2) agents do not know that fundamentals are hump-shaped and base their beliefs on parsimonious models that they fit to the available data? A class of parsimonious models leads to qualitatively similar biases and generates empirically observed patterns in asset prices and macroeconomic dynamics. First, parsimonious models will robustly pick up the short-term momentum in fundamentals but will generally fail to fully capture the long-run mean reversion. Beliefs will therefore be characterized by endogenous extrapolation bias and pro-cyclical excess optimism. Second, asset prices will be highly volatile and exhibit partial mean reversion-i.e., overreaction. Excess returns will be negatively predicted by lagged excess returns, P/E ratios, and consumption growth. Third, real economic activity will have amplified cycles. For example, consumption growth will be negatively auto-correlated in the medium run. Fourth, the equity premium will be large. Agents will perceive that equities are very risky when in fact long-run equity returns will co-vary only weakly with long-run consumption growth. If agents had rational expectations, the equity premium would be close to zero. Fifth, sophisticated agents-i.e., those who are assumed to know the true model-will hold far more equity than investors who use parsimonious models. Moreover, sophisticated agents will follow a counter-cyclical asset allocation policy. These predicted effects are qualitatively confirmed in U.S. data.

show abstract

Regression and time series model selection using variants of the schwarz information criterion

Cited by 62 publications

References 9 publications

Machine learning problems from optimization perspective

Machine learning problems from optimization perspective

Model selection with misspecified spatial covariance structure

Natural Expectations, Macroeconomic Dynamics, and Asset Pricing

Contact Info

Product

Resources

About