Advances and Applications in Statistics
Volume 40, Issue 2, Pages 93 - 107
(June 2014)
|
|
A COMPARISON OF LINEARITY TESTS BASED ON WILD BOOTSTRAP
Daiki Maki
|
Abstract: This paper compares the linearity tests based on wild bootstrap under heteroskedastic variances. We investigate the null hypothesis of linearity against logistic smooth transition autoregressive (LSTAR) models. Monte Carlo simulations show that conventional tests have size distortions when the data generating process (DGP) is highly persistent and/or has heteroskedastic variances. A test using the fixed-design wild bootstrap method performs well when the DGP shows quick convergence, whereas it shows under-rejection and has less power when the DGP is highly persistent. In contrast, a test using the recursive-design wild bootstrap method with Rademacher distribution has a reasonable size and power regardless of whether or not the DGP is highly persistent and/or has heteroskedastic variances. |
Keywords and phrases: linearity tests, LSTAR, wild bootstrap, heteroskedastic variance. |
|
Number of Downloads: 344 | Number of Views: 1088 |
|