Thanks Maria. I know e-views can be used. But you know e-views is black box type. You just input the data and you get the output. I have tried through excel. But what happens is you get the coefficients correct but t- value wrong. Obviously some adjustment is required to get the standard error correct. So I am unable to understand how to get the standard error correctly. Can anybody suggest please ?
what do you mean that t-value is wrong? I guess you mean that t-value is less than the absolute value of 2. And this an evidence of not statistical significance. Or not? Did you think to include any dummy variable?Do many combinations. add or subtract variables in order to see when the model gets better... And sth else, check your inputs...Are all ok?
No, No you do not get what I say. When I compare the output of e-views with that of excel, I see that coefficients of variables are equal but t - values are different. Please note that e-views do not use the fixed effect least - square dummy variable model (LSDV), but the fixed-effect within group estimator by expressing the values of the dependent and explanatory variables as deviation from their mean values. The resulting values (de-meaned or mean corrected values) are regressed.
Yes you can say that because level of significance is getting affected leading to wrong conclusion. Obviously excel is making some problem in computing the standard error.
The arguments are not making any sense now, frankly speaking. If different programs produce different kind of output on the same data set then which one should be taken as correct ? I want to do fixed effect panel regression through excel because one should get correct result by following the logical steps like any other parameter. When I go for pooled regression through excel I get perfect result comparable with e-views. But in case of fixed effect panel, as pointed out, when I go for mean- corrected value I must be missing something and that is why t-values are different. One must understand that panel regression is nothing but cross-sectional regression after adjusting fixed firm and fixed year effect. If that logic is correct then some problem is happening not due to excel but I am missing some step. I pointed out earlier that coefficients are consistent with eviews result but t-values are not. So the denominator factor - that is- computation of standard error is creating the problem. Can anybody with good conceptual background on econemetrics help ?
Its possible to use excell to arrange your data in a panel format, then import it to the relevant software especially Stata or Eviews. Then you can proceed with appropriate regressions and tests.
Me2. When I was younger... (hein!) I used to think I could do everything with MS Excel. Now, I've learnt to use the saw, the spanner, and other tools, I don't consider everything as a nail since I only had a hammer.
Depending on the data I use MySQL, LibreOffice.org Calc, or even Gnumeric to sort out the clutter. Then, who knows?, R, Gretl, SPSS... exactly, depends on the target. ;)
think the problem is how excel calculates the variance...compare the variance calculated in eviews with that calculated in any software package, what I remember, they differ...
If you use a fixed effects model, it could be estimated thru Excel. Simply add dummies for fixed effects and run OLS with Excel. But, anyway, you will get very poor information regrding statistics, etc. Specialized econometric software is much more helpful.
I think number of problems you are pointing out such as t value and standard error, t shows spurious correlation , you can put the lags against the independant variable. Better would be to use e-views. All problems will be solved .
You can use GRETL, you can download it from: http://gretl.sourceforge.net/gretl_espanol.html in the spanish version, but you can have the english version too.
If U have enough time U and if U know all theoretical aspects U can make it by hand but I don't recommend U. If U want something good and free I strongly recommend U Eassy Reg International
I compared results of Excel Panel Data analysis and Stata (done on Robust Standard error). The resulting R squared, the coefficients are either exactly the same or very similar but the p-values differ. Further, Excel only shows overall R squared while Stata shows 3 R-squared (within/between/overall) - and the results seems identical for the overall of Stata.
Can somebody help explain the reason behind the difference on "p values" between the two regression runs?
Does Stata robust totally resolved heteroscedaticity and excel cannot ? Why ?
The model is : dependent variable is market concentration of the Big 3 oil companies while the independent variables are P-price, GAE -admn expenses, R -revenue, FA fixed assets, ROA-return on assets, ROE-return on equity, GDP, CP-crude price, FX -foreign exchange rate, D- dummy variable deregulation.