LSDV estimator in Eviews 6.0

[pk296]

Hi all,

I would like to run a dynamic panel model on eviews using fixed effects and the Least Squares Dummy Variable (LSDV) estimator.

I have arranged my dataset in balanced panel form and click on the 'Fixed' tab in the Panel Tab for Cross section. I then estimated the equation using Least Squares (LS and AR) method. Is this estimation therefore using the LSDV estimator?

Also, in the estimation output, there appears to be no information for the R squared etc of the dynamic panel model itself. Rather, the only descriptive stats provided are for the cross section fixed variables.

Can someone please kindly clarify these points.

[EViews Gareth]

Yes, you are doing the LSDV estimator.

I think you are just mis-reading the output of the equation; the R-Squared etc.. are for the whole estimation, not just the cross-section fixed effects (Although I can understand why you would think that - the output does look a little misleading).

[pk296]

Thank you for your kind reply.

When running a dynamic panel model with cross-sectional fixed effects, do I need to include a constant (c) in the estimation?

My understanding is that the fixed effects allows us to correct for the endogeneity (and therefore bias) and therefore, it assumes that each country etc has a different intercept. Is this correct?

[EViews Gareth]

If you estimate with fixed effects, EViews will include a constant for you. Note this means that the fixed effects are deviations from the mean (across cross-sections) intercept.

[pk296]

Thanks for your reply.

I am trying to run a dynamic panel model with FE. I have 16 countries which I am looking at and have a range of variables. When I try to run the regression, I keep getting an error message saying "near singular matrix". Do you know why I am getting this message and what I can do.

[EViews Gareth]

Assuming that your equation is linear, the most likely cause of this is that your regressors are co-linear.

[startz]

Thanks for your reply.

I am trying to run a dynamic panel model with FE. I have 16 countries which I am looking at and have a range of variables. When I try to run the regression, I keep getting an error message saying "near singular matrix". Do you know why I am getting this message and what I can do.

Do you have a variable which is always the same for a given country?

[pk296]

I basically want to run the following dynamic panel regression as I am investigating convergence.

Yit = aYi,t-1 + bXit +Nt +Vi + Uit

Where Nt is time fixed effect
Vi is cross section fixed effect

In this context, do i need to use LSDV or system of GMM estimator ?

[pk296]

I should point out that the Yi,t-1 is a moving average process.

Is it only correct to use the LSDV estimator when the lagged dependent variable is constant ? When the lagged dependent variable is changing , as it is in my study, would it be more appropriate to use the system GMM estimator ?

[EViews Glenn]

Almost any time you have a lagged endogenous on the RHS of a panel estimator and you estimate using a fixed effect in the cross-section dimension you'll have problems. Almost every panel book will describe the nature of the bias in some detail and will offer recommendations.

[Greentea]

Could you be so kind to tell me if the fixed effect estimation within Eviews is LSDV or within effect or between effect, please? There is some information: http://www.indiana.edu/~statmath/stat/a ... anel3.html
According to the link above, I think that cannot use LSDV for panel data.
Thank you very much for your help!

[EViews Gareth]

As mentioned above (in this thread) and in the Manual, EViews does the LSDV estimator.

[Greentea]

Sorry again that I haven't made myself clear enough last time and I didn't check if I can open the link above. However, I think that I understand it now. Thanks!

[TheRocK]

Hi,

I also have a question concerning the LSDV regression. Consider a model in which the stock market return is regressed on - among other variables, 3 dummy variables which refer to 1 boom and 2 bust phases, with the second boom phase left out as the base case. In a "normal" cross-section regression, the intercept of that regression would be the average stock market return of the base case time period. To derive the average return of the other time period, one would have to sum the respective coefficients up.

However, in a LSDV regression, the reported constant is the average unobserved effect which has been estimated. It therefore does not present the average stock market return of the base case, does it ? In that case, what is the interpretation of the other three market phase coefficients? Is it still possible to conclude - in some way - that average stock market returns differed accross the specific market phases ?

Cheers !

Kai

[EViews Glenn]

As you point out, when you use fixed effects, there is always a normalization of the effects. One way to do this is to omit one of the dummy variables and to define the other coefficients as deviations from this base case. An alternative is to leave in all of the dummies but to normalize the coefficients so that they sum to zero. In this interpretation, the base case is a hypothetical "average case" and each of the dummies represents the deviation from the average. If you'd like to compute the deviation of any of the categories from one of the other, you can simply look at the difference of the coefficients. So the equivalent difference to the leave-one-out approach is to subtract off the latter coefficient from the one of interest...