非线性Granger因果检验方法的检验功效及有限样本性质的模拟分析

Simulation Analysis for Power and Finite-sample Properties of Nonlinear Granger Causality Tests

为了避免传统Granger因果检验方法因忽略经济变量的非线性特征而导致结论出现显著偏差的局限性,非线性Granger因果检验方法正逐步成为经济学研究领域的重要分析工具。然而,迄今为止,学术界仍较少对非线性Granger因果检验方法在不同非线性模型中的有限样本性质展开系统性的比较与分析。因此,本文通过数据生成过程(DGP),结合Monte Carlo模拟对Diks和Panchenko(2006)等主流的非线性Granger因果检验方法的检验功效、过度拒绝等问题展开比较研究,并对共同滞后阶数、带宽参数的不同设置可能引发的结论敏感性变化进行深入分析。在此基础上,我们从动态非线性滚动分析的角度对其有限样本性质展开进一步的讨论,并提出对未来非线性应用研究具有实际指导意义的若干建议。

Since traditional Granger causality tests do not take account of the nonlinear character of economic variables, conclusions based on traditional tests may be biased. To overcome these limitations, nonlinear Granger causality tests are proposed in the literature and become important analysis tools in economic research. However, so far in the in the literature there is lack of systemic and comprehensive investigation on finite-sample properties of nonlinear Granger causality tests within different nonlinear models. In light of this, this paper employs Monte Carlo simulations to systemically and comprehensively examine the power and over-rejection properties, and analyze the sensitivity of test results to the changes of common lag lengths and bandwidth parameters. Based on the perspective of nonlinear dynamic analysis of rolling to do further discussion on the finite sample properties, and this paper puts forward some suggestions has practical value of application of nonlinear on future research.