| Summary: | 博士 === 國立政治大學 === 國際貿易學系 === 88 === This dissertation analyses asymptotic properties of test statistics when the data generating process is nonstationary.And proposes a test procedure to improve size distortions of the tests.This dissertation includes three articles.
The first article invesigates the asymptotic properties of the Phillips-Perron tests and some of their variants when the error is MA with coefficient close to -1.Asymptotic distributions of the tests described above constructed by IV and OLS estimators diverge with rates of square root of T and T,respectively.
So,the sizes of "IV-type" Phillips-Perron tests are better than those of "OLS-type".And increase sample size can not resolve the size problems. Meanwhile,I propose modified Phillips-Perron tests based on IV estimator.In large samples,the modified tests have robust size.Simulations reveal that when mdified tests are bootstrapped, their sizes are very close to the nominal size
even in small samples.
The second article uses localized parameteralizations to invesigate large sample properties of stationary tests when the DGP is AR.Asymptotic distributions explain
why stationary tests would reject null hypothesis too often when AR root is close to unit.Finally,I discuss the possibility of improving size distrotion
by bootstrap.
The last article investigates the asymptotic property of the test proposed by Kuo (1998).When intercept is I(1), the distribution of the above test is different from that of Kuo (1998). This distribution not only explains the simulation results of Kuo which reveal serious size distrotions of the test but also can infer
the size pattern simultaneously.
|
|---|
