Estimating Impact of Membership of Free Trade Agreement on Trade using Panel Data

[Alifan Darul]

Good afternoon, everyone.

Recently I work on my research about the impact of membership of free trade agreement on Indonesia's export performance. I use panel data, consist of 35 Indonesia's main export destination countries (9 of them are ASEAN's member and the rest are not). This research is using export demand approach so the model will be like this:

ln exportijt = b0 + b1 ln gdprealjt + b2 realexchangeratejt + b3 cpiratioijt + b4 AFTA + eijt

i denotes for Indonesia (exporter)
j denotes for export destination countries (9 ASEAN countries and 26 non-ASEAN countries, or so called as importer in this case)
t denotes for year

Variable AFTA is a dichotomy (dummy) variable which takes value of 1 if the export destination countries is member of AFTA (ASEAN Free Trade Area) and 0 otherwise. Thus, I create dummy variable with this command:
gen AFTA=0
replace AFTA=1 if (importer=="BRN")
replace AFTA=1 if (importer=="KHM")
replace AFTA=1 if (importer=="LAO")
replace AFTA=1 if (importer=="MYS")
replace AFTA=1 if (importer=="MMR")
replace AFTA=1 if (importer=="PHL")
replace AFTA=1 if (importer=="SGP")
replace AFTA=1 if (importer=="THA")
replace AFTA=1 if (importer=="VNM")

Here I have 3 questions:
1. When I run regression with Fixed Effect command (xtreg lnexport lnrealGDP lnrealexchangerate lncpiratio AFTA, fe), the AFTA dummy variable is omitted of collinearity. Is there any solution to solve that problem?
2. When estimates about trade performance in panel data, I found some literature that use country-pair fixed effect and time fixed effect. Is that necessary? Because in that literature I read, they did not explain why they use that. If so, how to interpretate those effect?
3. Based on the nature of the data I used and the purpose of the research, is Fixed Effect the best method? Because, again, in the literature I read, they always use Fixed Effect Model to estimate the structural model.

Thank you for your advance.

Sincerely,
Alifan Darul

[Phil Bromiley]

You'll increase your chances of a useful answer by following the FAQ on asking questions - provide Stata code in code delimiters, readable Stata output, and sample data using dataex. For example, you don't tell us what your panel variable is or how many years of observations you have.

1. A fixed effect estimation cannot include variables that do not vary within panels - they're colinear with the panel effects. You can look into xthtaylor or a Mundlak estimator.
2. If you only have one exporter, then country fixed effects are the same as country-pair fixed effects. Whether a time fixed effect is necessary is a substantive issue. Does average behavior change over time? Time is normally just a control.
3. As I said in (1), if you want AFTA or even country characteristics that don't vary over time,you can't estimate them in a fixed effect model with the panel being countries.

[Alifan Darul]

Thank you, Phil for your advance

Specifically, my research title is "Estimate the Impact of ASEAN FTA Membership to Indonesia's Export". Most literature, as commonly used when there are time invariant variable(s), suggest that Fixed Effect estimator is the best method, rather than OLS and random effect. Then, again, the dummy of AFTA membership (which takes the same value over the period I used: example I use 2003-2016 as sample period and for Malaysia it will take value 1 overtime, and Australia will take value 0 overtime) is omitted (and it is the same as you said above that FE model will take out all time invariant variable(s) and omit that variable due to collinearity with country ID/countrypair ID)

I read some reference in this forum that xthtaylor does not perform well when I only has 1 time invariant variable. Furthermore, I only use 2 time varying variables (those are real GDP of importing countries and import price index of importing countries). And estimate the model by using hausman-taylor needs independent variables w that treated as endogenous (and I think that neither real GDP nor import price index is suitable to be treated as endogenous). Other than hausman taylor, is there any suggestion what method(s) should I use if I want to estimate the impact of the time invariant variable (in this case is AFTA dummy) on dependent variable?

I really appreciate your help.

Best
Alifan