(1)N=2

(2)协整关系反映同期关系，没有滞后项，否则是ADL模型。因为如果解释变量包含Yt-1，如文中所指，设定x的系数为零，Yt-1的系数为1，则残差一定平稳。

（3）johansen检验更不能包含Yt-1，看看其原理就十分清楚，它利用了Yt-1对所有解释变量的回归的残差，如果解释变量含有Yt-1，则该残差为零，因而出现"Near singular"问题

（4）协整系统的t值没有什么含义，不必考虑

（5）其他方法确定误差修正项也行，但它们通常不一致，相对而言，Johanson利用的信息最为充分

I found your questions are intersting. I think you may:

Question 1: N=2 this is for sure.

question 2: you can't put yt-1 in the estimation of cointegration. What you have found in the other article is definetly wrong.

Question 3: No!

Question 4: Please make sure that y1 x1 x2 are integrated in same order or I(1). You need to check whether there are any possible structural break in the time series. I can say that most of the macro-sereis are bound to struancrual breaks. Pease check with Perron (1997), or Zivot and Andrews (1992) or Perron And V. (1992). If there is structural break, you need to incoparate the sturcturla break into consideration. you may use Gregory and Hansen (1996) test.

question 5: Of course you may any other methods. such as EGtwo steps, Hansan FILM. ARDL etc.

I hope this will assist you in some sense. (P.S. I really want to write in Chinese but my computer have some problem in keyin Chinese charaters. Sorry for that.)

终于茅塞顿开。衷心感谢两位大哥的指教！

本人不是学计量出生的，很多内容是自己学的。看来书一定要选对，否则有可能被误导。 比如把y 、x1、x2和y（-1）弄在一起作协整检验是某教材上举的例子，某知名刊物上有篇文章也是这样作的。害人不浅啊！和为贵，没必要说出是谁了。

预祝二位新春愉快！