Anderson T W
Department of Statistics, Stanford University, Stanford, CA 94305-4065, USA.
Proc Natl Acad Sci U S A. 2000 Jun 20;97(13):7068-73. doi: 10.1073/pnas.97.13.7068.
The cointegrated model considered here is a nonstationary vector autoregressive process in which some linear functions are stationary and others are random walks. The first difference of the process (the "error-correction form") is stationary. Statistical inference, such as reduced rank regression estimation of the coefficients of the process and tests of hypotheses of dimensionality of the stationary part, involves the canonical correlations between the difference vector and the relevant vector of the past of the process. The asymptotic distributions of the canonical correlations and the canonical vectors under the assumption that the process is Gaussian are found.
这里所考虑的协整模型是一个非平稳向量自回归过程，其中一些线性函数是平稳的，而另一些是随机游走。该过程的一阶差分（“误差修正形式”）是平稳的。统计推断，例如该过程系数的降秩回归估计以及平稳部分维度的假设检验，涉及差分向量与该过程过去相关向量之间的典型相关性。在该过程为高斯分布的假设下，找到了典型相关性和典型向量的渐近分布。
