I was reading a paper about Free Trade Agreements impact on trade, here is the link http://www.aseancenter.org.tw/upload/files/OUTLOOK002_02.pdf. I have been contacting the author, but he wouldn't explain in detail.
For instance, he use gravity model:
Log(Yit) = β0 + β1log(SGDPit) + β2log(RFACit) + β3log(SIMit) + β4log(Distancei) + β5log(Areai) + β6log(REERit) + αi + λt + FTA1it + FTA21it + FTA22it
Yit = real import from country i to j in year t
SGDPit = the sum of real GDP of country i and j in year t
RFACit = relative factor price between country i and j in year t
SIMit = degree of similarity (in terms of GDP) between county i and j in year t
REERit = Real Effective Exchange Rate between country i and j in year t
Distancei = distance from country i capital city to country j capital city (km).
Areai = area country i (km2)
FTA1it, FTA21it, FTA22it = Free Trade Agreements (FTA) dummy variables
αi = country effect, it captures country i's characteristics that affect trade between country i and j other than accounted for by other regressors.
λt = time effect, it captures other factors that affect country j trade with any country in period t.
µit = error term
The estimation is Feasible Generalized Least Square using fixed effects for country variable and random effects for time variable. Please, could anyone explain to me how to do this on any statistical software, esp. STATA or E-views? I was confused since it is a mix of fixed and random effects.
To run GLS in STATA, you need to use the xtreg function (random effect and fixed effect). But before that, set first the period and sample using tsset function.
E.g. syntax
Random effect
xtreg Y X1 X2 X3, i(id) mle
Fixed effect
xtreg Y X1 X2 X3, fe i(id)
For population average you can use xtgls function
Any problem just type findit xtreg in the command centre.
cheers
khairul
Hi Anandita,
To expand a little on Khairul's answer, the analysis in Stata was _probably_ done using random effects (with the "re" option, "mle" uses maximum likelihood and not GLS), with dummy codes estimated for the country effects.
Although I have not looked at the paper you refer to, such a model could be estimated as:
xtset _timevar_
xtreg _depvar_ _indepvars_ i._countryvar_, re
Where _depvar_ and _indepvars_ are the dependent and independent variables (following from Stata notation) and _timevar_ is the time code, and _countryvar_ is the country code where "i." indicates the variable should be treated as a factor (if you have a version of Stata before 11, you'll have to use "xi:" to expand the "i." indicator).
The above model will implement the GLS random effects method for estimating the time-specific intercepts (as outlined in the Stata User's Manual) and will have fixed effects for each country. Other methods are, of course, possible (see (xt)mixed in the User's Manual which also allows for more complex combinations of random and fixed effects) this is just my guess based on the fact that you mention the author used GLS.
- joe
- Badges
- Science topic
- Similar topics
- Geoscience
- Asia
- India