Panel data, fixed or random effects estimation?

For econometric discussions not necessarily related to EViews.

Moderators: EViews Gareth, EViews Moderator

jasperq
Posts: 8
Joined: Thu May 10, 2012 3:24 am

Panel data, fixed or random effects estimation?

Postby jasperq » Sun May 19, 2013 6:45 am

Hi,

I am quite new to using panel data unfortunately. I am using a dataset that consists of 8 countries over 8 years (more data is not available on this matter). I already ran an OLS regression which came up with quite significant results. However, I was told that I should compensate for cyclical movements in the economy that are likely to influence my explained variable. Now I begin to understand that this can be achieved by including fixed/random effects. The problem is that I do not really know what I am doing when including these effects. If I include random period effects for instance, what is the rationale behind this? I am actually not interested in specific effects of the explanatory variables in a specific year or for a specific country, but mainly I try to find the impact of my explanatory variables in general, i.e.: my estimated coefficients should give an insight in how they generally influence my explained variable. Also, I was wondering what GLS weights do exactly.

Thanks in advance!

NicV
Posts: 7
Joined: Fri May 10, 2013 5:58 am

Re: Panel data, fixed or random effects estimation?

Postby NicV » Wed May 22, 2013 3:00 am

With fixed effects you can control for variables that are unobservable or for which you don't have data. It's a way to avoid omitted variable bias in your estimation.

You use cross-section fixed effects if you think there are omitted variables that vary across entities (countries, companies, persons) but do not change over time. Think of climate, geographical location and such for a country and e.g. intelligence or innate ability for a person.
You can use period fixed effects, on the other hand, if you think there are omitted variables that are the same for each entity in your sample, but vary over time. If you're for example looking at a specific business sector this might be laws and regulations.
If you suspect both types of omitted variables to be present you can use both period and cross-section fixed effects.
For more information on the pros and cons of fixed effects and random effects and whether to check which one is best for your analysis I highly recommend one of the books mentioned below.

As for GLS weights, I thought that they could help keep heteroskedasticity and serial correlation in check (depending on the weight used).

I found the following two textbooks to be very helpful in panel data analysis:
- Introduction to Econometrics, by Stock and Watson
- EViews Illustrated, by Startz

Please, someone correct me if I made mistake!

mfb
Posts: 28
Joined: Sun Apr 14, 2013 8:16 am

Re: Panel data, fixed or random effects estimation?

Postby mfb » Thu May 23, 2013 8:41 am

Hi,
I am playing with a data set of 21 countries over 5 years. And I am also learning about panel data econometrics.
I gather that one can do cross-section fixed or random effects. Or one can do period fixed or random effects. Or any combination of those (for example: cross-section fixed effects with period random effects).
For example, if you want to measure the impact of education (independent variable x, which varies over time) on GDP growth (dependent variable y, which varies over time), you want to control for things like endowments of natural resources, which vary from country to country and may also influence growth but may be considered constant in each country over several years).
If you just do a cross-section regression of y on x, the coefficient of x may be turn out to be biased because of the different natural endowments in each country that you are not controlling for.
A panel data regression of y on x with cross-section fixed or random effects will measure the effect of x on y separately from the effect of different endowments of natural resources on y.
The logic is the same for period fixed or random effects: there may be omitted (not controlled for) variables varying over time but affecting growth to the same degree in every country. A panel data regression with period fixed or random effects will control for these effects, making sure you get an unbiased coefficient of x as a measure of its specific impact on y.
As for fixed or random effects, I gather that FE is much more often used. But if your independent variable x is time-invariant, then FE is useless. Also, RE may be more efficient (more precise) in some situations. However, RE has some special requirements of its own (for example, the 'effects' must be independent from x, otherwise results will be biased). There is a test (the so-called Hausman Test) in Eviews to check whether FE or RE should be used (not to be confused with the Hausman endogeneity test).

...correct me if I am wrong!
MFB

shaliza
Posts: 14
Joined: Wed Mar 03, 2021 10:17 am

Re: Panel data, fixed or random effects estimation?

Postby shaliza » Fri Mar 19, 2021 12:34 pm

Could anyone please help me, if I select Period Random instead of cross-section Random ,will it be still considered as Random effects model? please I need help urgently


Return to “Econometric Discussions”

Who is online

Users browsing this forum: No registered users and 23 guests