PPML with fixed effects (exporter + importer) for gravity equation - How to get the values of fixed effects for the dropped variables?

[Fatma F-B]

Hi everyone,

I want to estimate a gravity model while accounting for Multilateral resistance for 2009 (Cross-section, 84 countries so 84*83 observations). I therefore want to add exporter and importer fixed effects. This will not permit to consider unilateral traditional variables (GDP, landlocked, RTA...). I have two questions:

1- Shall I use -ppml- while generating dummies for exporters and importers OR use -xtpoisson- with FE option ?

2- Since i want to evaluate the effects of UNILATERAL variables, I follow Melitz (2005) and regress the exporter/ importer fixed effects (become the dependent variable) over UNILATERAL variables.
Here, my question is : since STATA drops a dummy in order to prevent collinearity and as I want to have all fixed effects as observations for the NEW dependent variable, how to get all the values of country (exporter/ importer) fixed effects ?
Do the values associated to the exporter/ importer dummies presented in the results correspond to the needed country fixed effects ? IF yes, i want to make sure that for the NEW variable (Exporter OR importer country fixed effect), I shall add the value of ZERO for the dropped dummy to have 84 observations!

Thank you very much!

[Fatma F-B]

I have seen different posts about this subject but didn't find the answers I am looking for!
Dear Joao Santos Silva, could you help me dealing with this subject ?

Thank you in advance

[Joao Santos Silva]

Dear Fatma F-B,

1 - Please use ppmlhdfe; check the help file before using it.
2 - I am afraid I do not really understand this question.

Best wishes,

Joao

[Fatma F-B]

Dear Joao Santos Silva , thank you for your reply.
I have tested PPML while generating exporter and importer fixed effects (dummies generation). I am estimating bilateral exports ij Xij = f ( distance (not in log), contiguity, rta, common language) and other bilateral variables + exporter dummies + importer dummies) on Cross section (year=2009) and for a sample of 83*82 countries.

* I have three questions/points:
- Warning messages about large values for exports, I therefore rescaled exports (/ 1000000). Is this the needed action?
- Warning message about large values of distance (the value in KM2 i think from Geo Dist database of CEPII)
--> Using original values of exports and distance (without rescaling), the coefficient for DISTANCE is negative and significant but extremely weak : -0.01. This value is extremely different from the values in the literature. Can this be driven by the fact that I consider 83 and not 140 countries?
- I therefore rescale distance (distance/ 1000): the coefficient becomes now negative and significant and around -1. Is that normal?
- I also find significant and positive values for contiguity and colonial links BUT the coefficients are high (2 an 5 !!)

I don't know if the problem is caused by the scale of the variables or is normal !


* Besides, is PPML while generating exporter and importer dummies different from using directly PPMLHDFE ?


Could you help me verify the robustness of these estimation and results ?

Thank you !

Fatma

[Joao Santos Silva]

Dear Fatma F-B,

If the estimator is converging, you can ignore those warnings. However, you must be doing something wrong because rescaling the variables should not change their significance. Also, ppml and ppmlhdfe should give you the same estimates for the parameters that can can be identified with your data.

Best wishes,

Joao

[Fatma F-B]

Dear Joao Santos Silva, thank you for your reply!
One more question: when using PPML, I should consider exports in value. But for distance, shall i use Distance in value or log (distance) ? Or, are both correct ?

Thank you very much!
Fatma

[Joao Santos Silva]

Dear Fatma F-B,

When using ppml and other commands implementing a Poisson regression estimator, the dependent variable should be in levels (i.e., not logged). In gravity equations, regressors like distance should be logged.

Best wishes,

Joao