Pool/Panel Estimation with Fixed Time Effect

[Penlo]

Greetings!

I’m currently having troubles with fixed time effect using pool estimation and any help would be appreciated. So here goes…I would like to pool estimate the following model

Yt = c + Ft + B*dowjonest

(looking at the attached work file might help)

Where Ft = Fixed time effect, Yi,t = a panel of returns for 3 financial indexes.

To obtain the panel of returns, I went to new object > pool and typed
_ORD
_SP
_FTSE

Then I clicked on estimate and typed:

dependent variable = close?
common coefficient = dowjonest
and chose fixed time(period) effect

Unfortunately, I get “NEAR SINGULAR MATRIX” and yes, I know it’s something to do with perfect multicollinearity but I don’t understand why.

Any helps’ appreciated, thanks.

[EViews Gareth]

You don't have enough variability in your data to include period fixed effects. Your regressors have no variation between cross-sections (since Dowjones is the same for all). Thus there is no way to identify period effects.

[Penlo]

You don't have enough variability in your data to include period fixed effects. Your regressors have no variation between cross-sections (since Dowjones is the same for all). Thus there is no way to identify period effects.

Hi Gareth,

Thanks for the quick reply. Could you provide your opinion on how to regress the following?

I’m trying to regress a similar panel/pool model from a journal article which had:

Yi,t = c + Ft + B1*dowjonest + B2*nasdaqt

Where;

Ft = Fixed time effect for a sub sample within the full sample. For example suppose the full sample dates as 1/1/2009 – 1/12/2009, then the fix time effect was only applied to date 1/06/09 – 1/12/2009.

Yi,t = a panel of returns for 8 financial indexes.

Is there a way to regress this in Eviews 6 using either pool or panel estimations?

I initially tried panel estimations but because of the way the data was stacked, the variables left on the workfile was;

C
date
market
resid

As a result I got stuck since I didn’t know how to use those variables to regress the equation above.

Once again, any help is appreciated, thank you.

[EViews Gareth]

They're not doing traditional period fixed effects (where you have a dummy variable for each time period). They're doing it whereby they have a dummy for groups of time periods. That's perfectly fine.

To do that in EViews you'll have to create the dummies yourself, rather than using the simple fixed effects option.

[katrin]

Hi,

I'm on the search for something similiar.
I have my pool data for 36 periods and need "period fixed effects" for only 35 periods - the first period is excluded.
How can I do this?
By using @seas I got the error: near singular matrix...

THANKS

[EViews Glenn]

Try @expand with the @dropfirst option.

[katrin]

Hi Glenn,

thanks for your answer.
I've already used @expand in combination with @dropfirst. It works but I'm not quite sure, if it makes what I want... Therefore I had asked the question if there are other ways of modelling. I have to take a closer look, but I think it is the solution of my problem :)

May I ask you something different? Is it normal that one time I use @expand I get no error message and a few minutes later I get "near singular matrix" - and if I try it again, it works again... It's only a little bit confusing if I make the same regression with everything equal and one time I get a error message and another time it works without problems.

Thanks

[EViews Glenn]

That is not expected behavior.