Abstract
For panels of possible cross-sectional and serial dependency, stationary bootstrapping is applied to construct unit root tests that are valid regardless of the nuisance parameters of such dependency. The tests are semiparametric in that no model structure is imposed on the serial correlation and the cross-sectional correlation. The statistics are Wald tests and t-bar type tests based on the OLSE (ordinary least squares estimator). Residual-based and difference-based stationary bootstrapping are applied to obtain valid critical values of the tests. Both ordinary and recursive mean adjustments are considered. Large sample validity of the bootstrap tests is established for a large time series dimension. A Monte-Carlo simulation compares the proposed tests, yielding some promising tests, i.e., the t-bar type tests based on difference-based bootstrapping and recursive adjustment.
| Original language | English |
|---|---|
| Pages (from-to) | 14-25 |
| Number of pages | 12 |
| Journal | Computational Statistics and Data Analysis |
| Volume | 83 |
| DOIs | |
| State | Published - 2014 |
Bibliographical note
Funding Information:The authors are very grateful to two anonymous referees for valuable comments which improved the paper considerably. This work was supported by the National Research Foundation of Korea (NRF) grant funded by the Korea government, Basic Research Program (2012-001361) and SRC Program (2011-0030811).
Publisher Copyright:
© 2014 Elsevier B.V. All rights reserved.
Keywords
- Cross-sectional dependence
- Difference-based bootstrapping
- Recursive mean adjustment
- Residual-based bootstrapping