A Note on Spurious Inference in a Linearly Detrended Vector Autoregression
通过模拟研究发现，对趋势平稳变量进行线性去势后，向量自回归中的推断会出现严重偏差，随机游走变量在5%显著性水平下最多有60%的概率被误判为格兰杰原因，拒绝次数比未去势时高出2到5倍。
A simulation study is designed to evaluate the sensitivity of inference in a Vector Autoregression in which the variables of interest (GNP, the money stock, the price level, and a short-term interest rate) are treated as trend stationary processes. Using the normal asymptotic theory, the authors find that an artificially generated random walk Granger-causes the genuine variables in the model as often as 60% of the time for a 5% level test. They also observe substantial bias when other persistent stochastic processes are included in the autoregressions. The number of rejections are two to five times greater than if the variables are not linearly detrended prior to analysis. Copyright 1991 by MIT Press.