Inference for Time Series Regression Models With Weakly Dependent and Heteroscedastic Errors
针对误差存在非平稳异方差的时间序列回归，采用自归一化方法避免估计渐近方差，结合野自助法进行参数推断，模拟和实际数据（美国名义工资）验证了方法的有效性。
Motivated by the need to assess the significance of the trend in some macroeconomic series, this article considers inference of a parameter in parametric trend functions when the errors exhibit certain degrees of nonstationarity with changing unconditional variances. We adopt the recently developed self-normalized approach to avoid the difficulty involved in the estimation of the asymptotic variance of the ordinary least-square estimator. The limiting distribution of the self-normalized quantity is nonpivotal but can be consistently approximated by using the wild bootstrap, which is not consistent in general without studentization. Numerical simulation demonstrates favorable coverage properties of the proposed method in comparison with alternative ones. The U.S. nominal wages series is analyzed to illustrate the finite sample performance. Some technical details are included in the online supplemental material.