Estimating time invariant variables with fixed effects

[Nile]

Hi everybody

I have a problem with entering time invariant variables to the panel model in order not to have near singular matrix. I wonder if you can help me. Hope that you can. The following is my problem in details.

My model tries to test the hypothesis: Logistics performance works as a barrier to trade in The case of GAFTA (Great Arab Free Trade Area) countries.

In order to test this hypothesis, I used a gravity model for trade between arab world. The model consists of:
The dependent variable is exports between 16 arab countries. In each observation Xij= exports from country i to country j from 2005-2009. So I used a panel data approach with 240 panel for country pairs (Alg_Bah, Alg_Egy, Alg_Jor, Alg_Kuw, Alg_Leb, ...etc).
The independent variables are
GDPi*GDPj
GDPPC (the GDP per capita for each pair of countries)
DIST (geographic distance between each pair of countries)
ECC (exporting costs in country i)
ICC (importing costs in country j)
DAGR (dummy variable for a bilateral agreements betwen pairs)
...... etc.

By adding data to eviews some variables (like distance between each pair and bilateral agreements dummy variable) are time invariant. I added them to eviews as the rest of the variant variables across time and cross section (by adding the same value for each subseries (dis_alg_bah for instance) along the time series (2005 to 2009).

Tring to estimate the equation from the pool object using fixed effext (that the Hausman test suggest), eviews gave me a statement of "Near singular matrix".

I thought that there is a dummy trap, but the variable distance is not a dummy. The problem as I think is because these variables are invariant (constant across time). Are there another way other than poolgenr from the pool object to tell eviews that these are invariant variables to avoid the singular matrix and can estimate?
Would you mind helping me please by telling how to enter such invariant variables across time like distance?
Thank you.

[startz]

Fixed effects in the presence of time invariant variables is meaningless.

Think of it this way. For each country you have a separate intercept. You also have time invariant variables. Both the intercept and time invariant variables are constants. You can't have two constants in the equation.

[Nile]

Dear startz
First Thank you so much for your quick responce. OK I appreciate your help startz. I understand what you said about the existance of 2 time invariant regressors (for instance distance and the intercept). And as you said fixed effects in the presence of time invariant variables is meaningless.

Regarding what you said what is the appropriate way of estimating such model knowing that random effect method is inconsistant in this case according to Hausman test.

[startz]

Drop the time invariant variables. They're effects will be absorbed in the fixed effect.

[Nile]

Dear startz

Thanks a milion, Iam realy appreciate your help. But forgive me I still have a question.

Yes I got your point, you mean that any time invariant effects and only vary across the cross sections will be included in the intercept by definition using the fixed effect model.

But all the previous studies regarding the gravity model use GDP and distance as major variables in their model and then they were adding other independant variables subject to their target. In your opinion are their a way to include such time invariant variables other than fixed effects.

[startz]

That's a well put question. In general, you can include all sorts of time invariant variables so long as you don't also include fixed effects.

[Nile]

Sorry just to make sure that I got what you mean. Do you mean just pooling the data without both fixed or random effects?

[Nile]

Please I konw that I am asking alot, but I want to make sure about what I understand from your answer. Would you mind answering my question please?

[startz]

No fixed effects. Off the top of my head, I can't remember about random effects.

[samijo]

With Random Effects you can model time invariant variables. It's only in Fixed Effects models you can't do so.