Computing Reviews
Today's Issue Hot Topics Search Browse Recommended My Account Log In
Review Help
Search
Bayesian inference and model comparison for asymmetric smooth transition heteroskedastic models
Gerlach R., Chen C. Statistics and Computing18 (4):391-408,2008.Type:Article
Date Reviewed: Jun 4 2009

How do you create a viable forecast model when the data you are dealing with has two different levels and variances? Such data is termed heteroskedastic.

You could fit the data with classic regression--ordinary least squares (OLS)--but the data at the left end of the scatter would pull down the predicted value(s) just beyond the range of the scatter. And in the case of the data discussed in this paper, that could mean that your predicted values were underestimated by several million dollars.

You could split the data into two groups, say, halfway between the minimum and maximum values on the x-axis. Then, you could simultaneously model the data in the left-hand region (regime) with the data in the right-hand region. Each region would have its own mean and variance for predictive purposes. But what happens to predictions when you cross from the lower to upper region? There is a jump or step when you cross region boundaries; it would be better if the predictive equation moved smoothly from one region to the other. This is the type of predictive equation that Gerlach and Chen describe in this paper.

The discussion describes a smooth-transition heteroskedastic model that employs an adaptive sampling scheme to produce initial parameter estimates. It cites numerous other papers on the subject. The authors demonstrate how the smooth transition model does better at prediction than two other limiting cases--DTX GARCH, primary smoothing parameter infinite, and ARX GARCH, primary smoothing parameter zero.

This paper is written by PhDs in econometric modeling for PhDs in econometric modeling. An intimate knowledge of the subject matter is required, especially with regard to the mathematics. That said, if you can locate some software that applies these techniques to a forecasting problem, that software’s predictions might be significantly better than the status quo.

Reviewer:  Dick Brodine Review #: CR136911 (1002-0183)
Bookmark and Share
  Reviewer Selected
 
 
Probabilistic Algorithms (Including Monte Carlo) (G.3 ... )
 
 
Inference Engines (I.2.3 ... )
 
 
Markov Processes (G.3 ... )
 
 
Modeling Methodologies (I.6.5 ... )
 
 
Deduction And Theorem Proving (I.2.3 )
 
 
Model Validation And Analysis (I.6.4 )
 
  more  
Would you recommend this review?
yes
no
Other reviews under "Probabilistic Algorithms (Including Monte Carlo)": Date

Type: Article
Jul 1 1987
A probabilistic lower bound for checking disjointness of sets
Manber U. (ed) Information Processing Letters 19(1): 51-53, 1984. Type: Article
Feb 1 1985
On iterative methods in Markov modelling
Schatte P. Journal of Information Processing and Cybernetics 23(1): 49-51, 1987. Type: Article
Oct 1 1987
more...

E-Mail This Printer-Friendly
Send Your Comments
Contact Us
Reproduction in whole or in part without permission is prohibited.   Copyright 1999-2024 ThinkLoud®
Terms of Use
| Privacy Policy