非参数计量经济学中的伪回归诊断

Diagnose the spurious regressions in nonparametric econometrics

研究非参数模型的伪回归诊断问题.在Phillips局部诊断思想的基础上,考察了局部Durbin-Watson(DW)统计量的设计思路和理论基础,研究了检验的可靠性,给出了检验统计量的参考临界值.文章指出:若时间序列存在趋势,使用非参数模型分析时间序列时,可能会发生虚假回归,给研究带来误导;诊断回归残差的序列相关特性是十分重要的,伪回归下的模型残差会表现出一种非线性严重序列相关的特征,它既是序列趋势导致虚假回归的必然结果,同时也是推断估计存在失真风险的充分条件;局部DW统计量可以通过检验该特征来诊断伪回归.研究表明,局部DW检验的性质良好,可有效识别时间序列分析中非参数模型的伪回归.

This paper studies the diagnosis on spurious regression in the kernel estimation model. Local Durbin-Watson(DW) test is analyzed based on the idea of local diagnosis testing mentioned by Phillips, and its test reliability is also studied, and critical values of test statistics are also given. As a result of the influence of trend, nonparametric models may have the spurious-regression problem, which is always accompanied by serious residual serial correlation. On the other hand, estimation is proved to be not credible when residuals have serious serial correlation. To judge whether there is this kind of serial correlation is important. By testing whether there is nonlinear serial correlation, local DW test can be used to diagnose spurious regressions in the nonparametric model. The results of Monte Carlo experiments show that the diagnosis have satisfactory finite sample performances.