Sample Size

[lnp3]

dear forum,

Do you know if I can adjust the sample "size" in Eviews in panel format (cross sectional units of observation, not the sample period)?

Let's say I want to get rid of units that have large missing values. which command should I use? I really appreciate any feedback on this issue.

best regards,

[EViews Gareth]

Could you provide more details? Your question is open to interpretation.

[lnp3]

dear Eviews Gareth,

in my data set, I have 39 units of observations (firms) across a range of variables.

When I look at the file in the excel spread sheet, I see that some firms have lots of missing values (n.a). I beleive that is creating a problem for my analysis. I was told that data has to be as complete as possible (balanced and few missing values)

I want to get rid of those units of analysis with lots of na's. In other words, i want to reduce sample size? How can I reduce sample size?

Hope this is clear. I really need your help.

as always, appreciated.

lnp3
M.D

[EViews Gareth]

For 99% of statistical analysis in EViews, it will drop the NAs for you. You don't need to change the sample.

[lnp3]

that is good to know indeed..

[lnp3]

dear Eviews Gareth,

on a related topic, I tried to run dynamic OLS in a panel framework. However, I receive this error:

It says " no valid observations after removing cross-sectons with estimation errors"

There is no problem running Fully Modified analysis (FMOLS). It gives problems with Dynamic OLS.

What can be the reason of this problem?

I really appreciate your help.

[EViews Gareth]

You probably don't have enough observations.

[lnp3]

dear Eviews,

For dynamic OLS, I think I need longer time period and larger data set. My period is 1997-2012 (with a total of 624 observations), 39 units of observation.

I beleive that missing values are complicating the problem. Some cross sections have insufficient observations.

Do you think I should use the method of "interpolation"... Eviews says that this procedure " fills in missing values or na's within a series by interpolating them from values that are non-missing"

Do you think it is necessary and if so will you help me interpolate my series? I opened a series but could not find the command in Eviews 8 Proc/Interpolate..

[Kwong]

You probably don't have enough observations.

I include the 1 dummy variable and 1 interaction term while estimating the fmols and it show he same statement. Any suggestion to estimate the model which is cointegrated in the panel pedroni cointegration test?
Thanks.