各位前辈好,
本人是EVIEWS小白,因为做论文的缘故接触到回归分析及EVIEWS,现在遇到麻烦,还请各位前辈能给指导一下.
我的数据是个平衡长面板(BALANCED PANEL),三个国家46年.找到相关数据后,在考虑散点图和曲线拟和工具结果的基础上将回归模型设定为:Y?=C+a*log(x1?)+b*log(x1?)^2+c*log(x1?)^3+d*log(x3?)
按照一面面板数据的操作方式,我先进行了单位根检验,检验通过证明没有单位根,数据平稳.接下来就是模型形式的选择了,先通过EVIEWS自带的Redundant Fixed Effects - Likelihood Ratio在混合模型和固定效应模型间选择, 结果支持使用固定效应.但是在接下来做 Hausman Test for Correlated Random Effects 的时候遇到了麻烦, EVIEWS要求在做这个检验时将Estimation Method设置为Random的,但是我按照这个设置之后, 系统提示如下
"Random effect estimation requires number of cross section > number of coefs between estimator for estimate of RE innovation varaince" 这个提示的大概意思什么?是说截面数量要大于变量系数数量才能进行随机效应估计么? 我的方程形式是不能改变的,即使我舍弃X2?和X3?两个变量我的变量系数依然不小于截面数量啊. 有什么办法能够解决这个问题啊.
虽然现在的主流就是做固定效应,因为随机效应有诸多问题,而且有的教材,譬如说萧正的教材也提到了了这一点,但是我比较担心答辩的时候有***的会问怎么能确定就用固定效应而不用随机效应这样的问题.
另外,面板数据的异方差和序列相关怎么来检测啊, 看残差RESIDUAL的图么?
本人刚上论坛,米不多,但如果前辈们能给予帮助,仍愿倾囊相赠,聊表谢意!
最后结果:
我把遇到的问题发在EVIEWS的论坛上,EVIEWS的客服工程师回复说当截面数小于变量系数数量时,在EVIEWS下只能做固定效应模型的估计.现把提问和回复的内容贴上,供大家参考.
About Fixed/Random Testing

by Du Jianwei on Fri Oct 19, 2012 12:18 am

hi,
About Fixed/Random Testing

by Du Jianwei on Fri Oct 19, 2012 12:18 am

hi,
I am an Eviews 7 user and now I am doing a panel regression with this software. My data set includes 3 countries for 46 years(1963-2008). The form of my function is:
Y?=C+a*log(x1?)+b*log(x1?)^2+c*log(x1?)^3+d*log(x3?), the intercept "C" is expected to be cross section sepicfic.
My procedure goes like this:
Firstly, I run the "Unit root test". Since the result of this test shows that my data does not have a unit root, I turned to the next step of fixed/random effect model selection.Unfortunately, I encountered some problems in using the build-in function of Fixed/Random Testing.
There are two testing funcions for fixed/random selection,"Redundant Fixed Effects - Likelihood Ratio" and "Hausman Test for Correlated Random Effects", to perform the latter test, I need to estimate my model with  random effects specification.But once I select the "random" choice in Estimation method, Eviwes refuses to run estimation and the following warning has been given: "Random effect estimation requires number of cross section > number of coefs between estimator for estimate of RE innovation varaince". How can I solve this problem?
Thanks
Du
Re: About Fixed/Random Testing

by EViews Glenn on Fri Oct 19, 2012 1:07 pm

The problem is that the between estimator which is required for construction of the random effects estimator uses cross-section averages as data. Since you only have three cross sections, the between estimation has only three observations to estimate all of the parameters of your model.

You'll need to stick with the fixed effects estimator