Robustness properties of time series cross-validation /
It is very important to choose an appropriate bandwidth for a kernel estimator. An ideal bandwidth would be one that is determined only by the observed data, makes an estimator close to the true function, and is robust to mild deviations com the model assumptions. The TSW bandwidth proposed by Hart...
| Main Author: | |
|---|---|
| Format: | Thesis Book |
| Language: | English |
| Published: |
[Place of publication not identified] :
[publisher not identified] ;
1999.
|
| Subjects: | |
| Online Access: | http://proxy.library.tamu.edu/login?url=http://search.proquest.com/docview/304561968?accountid=7082 |
| Summary: | It is very important to choose an appropriate bandwidth for a kernel estimator. An ideal bandwidth would be one that is determined only by the observed data, makes an estimator close to the true function, and is robust to mild deviations com the model assumptions. The TSW bandwidth proposed by Hart (1994) is proved to possess these good properties. A special case of TSW, called OSW by Hart and Yi (1998), also possesses these good properties. We show that the TSW curve is asymptotically equal (in probability) to the ASE curve. Simulation studies are carried out which are in good agreement with our theoretical results. We use two data sets, wheat prices in Europe from 1500 to 1869 and yearly records in the mile race from 1860 to 1982, to demonstrate the usefulness of the TSW bandwidth. |
|---|---|
| Item Description: | Vita. "Major Subject: Statistics". |
| Physical Description: | xi, 94 leaves : illustrations ; 28 cm. Issued also on microfiche from University Microfilm Inc. |
| Bibliography: | Includes bibliographical references (leaves 91-93). |