|
|
|
|
|
|
Help | Seminars List | Add Seminar | Edit Seminars | Tips for organisers | RSS | ics Calendar | Search | Send comments about this website to seminar-master@cecs.anu.edu.au
Contact: luke.fletcher@anu.edu.au INFOENG SEMINAR SERIES Colloquium series
Automatic Information Criteria - a brief overviewProfessor Barry Quinn (Department of Statistics Division of Economic & Financial Studies Macquarie University, Sydney )DATE: 2006-02-24 TIME: 10:45:00 - 12:00:00 LOCATION: RSISE Seminar Room, ground floor, building 115, cnr. North and Daley Roads, ANU ABSTRACT: For many years the determination of which statistical model `best' fitted a set of data was carried out via hypothesis testing. Typically, at least in the classical (Neyman-Pearson) setting, a family of models was chosen, with model totally specified by some fixed number of parameters p. The alternative hypothesis allowed the parameters to vary over a p-dimensional subset, while the null hypothesis restricted the parameters to a q-dimensional subset of this, where q [lessthan] p. Automatic criteria were developed to turn this problem into an estimation problem. In particular, Akaike (1969, 1970) and Parzen (1974) introduced automatic techniques (AIC, FPE and CAT) for determining the order of an autoregression. Hannan and Quinn (1979) showed how to modify AIC so that the order estimator was (statistically) consistent. AIC-type procedures have become standard wherever statistical system order is determined, and are used often when it is not appropriate to do so, for example, to choose between non-nested models. I, amongst others, have extended their use to choosing the number of sinusoids in a sinusoidal regression, and, more recently, to estimate orders when testing whether two stochastic processes have the same spectral structure (discrete and/or continuous). This seminar will be part of the NICTA "Workshop on Model Selection and Data Fitting".
http://www.stat.mq.edu.au/staff/bquinn/bquinn.htm
|