as one of the previous respondets said, your problem arises because of the dependent variable, which is a first difference, estimated with the full sample (in terms of time dimension). By using the first difference, you are, in a sense 'loosing one year' for which no difference can be calculated any more (the year 2008, as you have no 2009).

You avoid this problem by 1) using a time trend, which is a continuous variable. Thus, it is not collinear with the 'superfluous' time dummy of 2008/the constant. by 2) estimating for a subsample for which the succedding year is availabe, e.g. for 2001- 2005, with data for 2006 available, so that the first difference 2005-2006 can be caculated.

I will try the estimation methods you suggested the (2sls and Ivregress) and will share the result.

For you first clarification about the year dummies, Yes, I have excluded y_1 =2000 as usually recommended in dummy variable coding. I have 8years( 2000-2007) I code them as y_1 =2000 , y_2=2001 ...y_8=2007. In regression I excluded y_1 as base year.

For your last question, regarding my use of the first difference d.cbf and d.rwar. I have the following additional question:

I originally coded the time variable (time trend) as yid (y=year, id= Identifier). In my model when I used this Id= yid as the time or trend variable the constant term is included in reg3 output (not dropped), but when I code the yid as individual year dummies and included the full year dummies -1year (I mean I excluded y_1, and included the remaining y_2 to y_8 in the model) then reg3 will dropped the constant term (with the full 8year-1year dummies).

Note also that when I use only 4years then the constant term is not dropped in the model ( I mean when I split my 8year sample period into 2 4years sub-sample e,g when I use y_2, y_3,y_4, y_5 in this case the constant term is not dropped from reg3 output).

My question is: if there is a collinearity in my model is it between the year dummies and the constant term or with my other variables too? What I should do then in any of this cases?

I hope I am able to explain my problem clearly, I can explain more if necessary. Thank you all for your valuable answers.

Happy New year.

---------------------------------------- > From: [email protected] > Subject: re:RE: st: why reg3 dropped constant term ? > Date: Sun, 2 Jan 2011 08:50:54 -0500 > To: [email protected] > > <> > Here is what I have typed in Stata and the Stata estimated output. > > Manually (using hand to type in Stata command window, > > reg3 (Leverage: d.lvr=l.lvr d.rwar roa llpta lta liq hhi y_2 y_3 y_4 y_5 y_6 y_7 y_8) (Prisk: d.rwar = l.rwar d.lvr llpta liq lta hhi y_2 y_3 y_4 y_5 y _6 y_7 y_8), small > > > Note that: > D.cbf and D.rwar are the only two endogenous variables in this model, and that all other variables are considered to be exogenous in this model. > > > > Try estimating each of these equations via single-equation methods such as -ivregress 2sls- to better diagnose the issues with collinearity. Do y_2-y_8 comprise a complete set of year dummies (or is there an excluded y_1)? > > Keep in mind that with each equation estimated with a first difference as the response variable, a constant term would constitute a linear trend coefficient in the level of the response variable. Do you really mean to test for the existence of such a linear trend in levels? > > Kit > > > > Kit Baum | Boston College Economics & DIW Berlin | http://ideas.repec.org/e/pba1.html > An Introduction to Stata Programming | http://www.stata-press.com/books/isp.html > An Introduction to Modern Econometrics Using Stata | http://www.stata-press.com/books/imeus.html > > > * > * For searches and help try: > * http://www.stata.com/help.cgi?search > * http://www.stata.com/support/statalist/faq > * http://www.ats.ucla.edu/stat/stata/