Abstract
In this paper, size and power comparison of five LM tests (two BLTs, the EART, the START, and the ARCHT) will be performed for different nonlinear models (BL, EAR, TAR, ESTAR, LSTAR and ARCH models) using asymptotic and bootstrap methods. Aiming to find out which method is more suitable. For comparison the general and specific linearity tests, the general linearity test (BDS test) is also included. We found that we should consider using bootstrap critical values rather than asymptotic critical values when : Sample is less than 200; the autoregressive coefficient is near the unit circle; the linearity test is ARCHT or BDS. In addition, we found that the only advantage of the BDS test over an LM test is its generality.