Original
The Answer Was There All Along: Worry About the Dynamics!
Abstract
In an influential article in 1974, Clive Granger and Paul Newbold demonstrated that regressions with variables that have unit roots often lead to spurious inferences. However, researchers have overlooked the fact that Granger and Newbold’s results were from a static time series model plagued with residual autocorrelation. We demonstrate, analytically and through simulations, that, under most circumstances, adding lags of both the outcome and predictor(s) can account for serial correlation and avoid spurious inferences. We thus recommend the use of a general modeling strategy, an autoregressive distributed lag model with multiple lags of the outcome and predictor(s), as one possible alternative of model choice that is less likely than other commonly used models to reach incorrect conclusions about true effects. Furthermore, our recommended approach performs well relative to the Webb, Linn, and Lebo bounds approach, the latest in a series of techniques designed to avoid the problematic nature of unit root testing.
中文
答案始终就在那里：关注动态！
摘要
在1974年一篇颇具影响力的文章中，Clive Granger和Paul Newbold表明，包含单位根变量的回归往往会导致虚假推断。然而，研究者们忽视了一个事实：Granger和Newbold的结果来自一个饱受残差自相关困扰的静态时间序列模型。我们通过解析推导和模拟表明，在大多数情况下，同时加入结果变量和预测变量的滞后项可以解释序列相关，并避免虚假推断。因此，我们推荐采用一种一般性建模策略，即包含结果变量和预测变量多个滞后项的自回归分布滞后模型，作为一种可能的模型选择替代方案，它比其他常用模型更不可能就真实效应得出错误结论。此外，与Webb、Linn和Lebo的界限方法——一系列旨在规避单位根检验问题性质的最新技术中的最新一种——相比，我们推荐的方法表现良好。
关键词
时间序列分析、虚假回归、单位根、自回归分布滞后模型、序列相关、政治方法论