Zhangjun80 同学,有关VAR定阶的问题.我有这样的看法. 1) 在确定各变量为I(1)时, 才可以进行协整检验. 2) 在进行协整检验之前先要确定阶数 (在水平变量下),否则协整检验一定有误. 3) 虽然各种标准(如AIC BIC FPE 等)可以参考, 但有文献认为进阶检验(LR)并进行样本调整 (SIMS TEST) 最好. 而其他标准倾向于选择过少的阶数. 当然还要进行残差检验. 4) 确定阶数就可以进行协整检验. 5) 如果发现有协整,就可以进行VECM模型, 否则就进行差分, 然后进行VAR. 6) 如果进行差分VAR,不应该有截距项和时间趋势项, 因为在差分时, 已经被消去. 7) 进行差分VAR,最好再重新定阶, 原理同前. 这是本人的一些浅见, 还望各位同仁指正. 谢谢!
Zhangjun, as per the lag selection, I have following coments for your consideration 1) Cointegration test only can be done after you confir that the varaibles under consideration is not stationary, i.e. they are all I(1). 2) doing Cointegration test before correctly determine number of lag may have misleading result. 3) Hamilton and Herrera (2000) argues that the number of lags selected by the AIC criterion is too small whiel Ivanov and Kilian (2003) compare the performances of five criteria in selecting lag-lengths in monthly and quarterly VARs using the mean-squared error as their metric, conclude that with a sample size of less than 120 observations, quarterly data, the Schwartz information criterion (SIC) performs the best among the class of criteria that they select. 4) after the determined the number of lag to be included, then you can proceed to the cointegration test. 5) Should you find that the variables are cointegrated then its appropriate to do VECM, otherwise cary on VAR under first differene. the result of VAR is valid only under stationary condition. 6) the constant and trend may not be included in difference VAR since both of them were eliminated during difference process. 6) It is better to do lag selection procee again with the differeced cariables. Your comments and suggestion are highly appreciated. Thanks.
另外, 我认为,本论坛为大家提供了一个交流学习的平台, 这平台能发挥多大效用, 取决于大家呵护与培养. .....
扫码加好友，拉您进群
全部版块
我的主页
收藏
