非线性时序模型LM检验的两类临界值检验统计功效比较

本文基于模拟方法比较了不同非线性时序模型的LM检验的功效和规模，同时也考虑一般化线性检验BDS检验参与比较，目的在于探讨蒙特一卡洛渐近法检验与自举法（bootstrap）检验的两类临界值的统计功效何者更为有效。通过实证与对比分析，结果表明，当样本小于200或自回归系数接近单位根，或者线性性检验是ARCHT或BDS时，就可以考虑应用自举法临界值而非渐近临界值。而且还发现，BDS检验仅在一般性上优于LM检验。

The Comparison of Statistical Power of Two Types Critical Values for the LM Test of Nonlinear Time - series Models

For studying which one is more effective for the statistical power between the Monte - Carlo Asymptotic Test and the Bootstrap Test, the power and scale of different nonlinear models are compared in this paper based on simulation, and the generalized linear test BDS, is also included. The empirical results show that, one should choose BCV rather than ACV when the sample is less than 200, or the auto - regression coefficient closes to unit root, or the linearity test is ARCHT or BDS. And, the results show that BDS test is better than LM test only on the generality.