On one-side cross-validation in nonparametric regression /
Time-Series Cross-Validation, an automatic bandwidth
| Main Author: | |
|---|---|
| Format: | Thesis Book |
| Language: | English |
| Published: |
[Place of publication not identified] :
[publisher not identified] ;
1996.
|
| Subjects: | |
| Online Access: | http://proxy.library.tamu.edu/login?url=http://proquest.umi.com/pqdweb?did=739668171&sid=1&Fmt=2&clientId=2945&RQT=309&VName=PQD |
| Summary: | Time-Series Cross-Validation, an automatic bandwidth selection method in the problem of time series trend estimation, was suggested by Hart (1994). It predicts the future observation Yj by the past data through time j - 1. We adapt the concept of TSCV to independent data and suggest a new data driven bandwidth selector, called one-sided cross- validation (OSCV). This method yields an automatic bandwidth selector by using different kernels at the estimation stage and the bandwidth selection stage. It is shown that the bandwidths chosen by this method are less variable than those chosen by the ordinary cross-validation. OSCV is also completely automatic, i.e., does not require extra parameter estimation, while the plug-in rule needs to estimate some extra unknown functions. |
|---|---|
| Item Description: | Vita. "Major Subject: Statistics". |
| Physical Description: | viii, 84 leaves : illustrations ; 28 cm. Issued also on microfiche from University Microfilms Inc. |
| Bibliography: | Includes bibliographical references: pages 81-83. |