[DN Jay]

That sounds correct to me.

Best wishes,

Joao

Thank you for the clarification, Professor Santos Silva.

[Maisam Amiry]

Joao Santos Silva
Dear Sir
I hope you are doing well. This is Maisam Amiry a student of MA program. I am using gravity model in my thesis and chosen to use ppml method. Could you please suggest me a material that will help to interpret the coefficients of ppml? In addition, using ppml do I need to some test in my estimation and analysis? Thanks in advance.
Regards
Maisam

[Joao Santos Silva]

Please check

https://www.wto.org/english/res_e/pu...uide2016_e.htm

Best wishes,

Joao

[Ivan Sergio]

Dear Prof. Silva,

I am using in my estimation the PPML with binary dipentent variable (0,1), but I have some problems in the interpretation of the coefficients.

For example: I have ln(GDP) as my independent variable and the coefficient is 0.35. Can I interpret the result as follows: a 1% increase in the ln of GDP is associated with a 0.35% increase in the probability for y?

Thank you very much

Best regards,
Ivan Sergio

[Dr. Iqra Yaseen]

Please check

https://www.wto.org/english/res_e/pu...uide2016_e.htm

Best wishes,

Joao

hello sir, Joao Santos Silva
according to the above cited paper, "the interpretation of the estimate of the coefficient on the logarithm of the continuous variable is simply the elasticity of (the value of trade flows) with respect to the continuous variable." and "The volume effects triggered by a change in an indicator gravity variable, such as the presence of RTAs, can be calculated in percentage terms as (e^beta-1)*100". now my question is that my independent variable is a factor variable with value ranging from 0 to 7 and my dependent variable is the imports value. should i consider the ind. variable as a continous variable or an indicator variable. next if i consider it as a continous variable (as stata suggests it so) how should i interpret beta co-efficient of +0.096 for value of 3 (factor variable) on imports volume. i use ppmlhdfe command for my gravity model of trade.
Thank you.

[Dr. Iqra Yaseen]

.

[Joao Santos Silva]

Dear sybil arqi,

If I understand correctly, you independent variables takes integer values from 0 to 7. If that is correct, the interpretation is as for a dummy variable, that is, (e^beta-1)*100 will give you the percentage change resulting from changing the index by 1 unit. You may, however, want to replace that variable by a set of dummies using the i, notation. In that case the interpretation is obvious and you have a more flexible model.

Best wishes,

Joao

[Dr. Iqra Yaseen]

thank you Joao Santos Silva for your reply. yes, i have already replaced the said variable by a set of dummies using the i. notation and in such situation the co-efficient of +0.096 is for the dummy value of 3, but i still don't know how to interpret the co-efficient. kindly help me about the same. thanks

[Joao Santos Silva]

Dear sybil arqi,

So, in that case, setting that dummy to 1 increases the expected value by (e^0.096-1)*100% = 10% relatively to the base category.

Best wishes,

Joao

[Leo Hudson-Foster]

Dear Prof. Joao Santos Silva


I am exploring the effect on EU membership on bilateral trade flows using a Gravity specification. I have used both OLS estimation and PPML estimation techniques but my coefficients for the PPML estimation do not appear to be the most reasonable, so I would really appreciate if you could check my inputs.

This is the code I am using for my main results:
asdoc ppmlhdfe tradeflow_baci eu_1 eu_2 wto_2 gdp_o gdp_d dist i.year, nest append

For context, eu_1 represents trade diversion (only 1 country is a member of the EU) and eu_2 represents intra trade within the EU (both countries are members).

The results are the following:
tradeflow = -0.226***eu_1 + 1.166***eu_2 + 1.092***wto_2 + 2.65e-10***gdp_o + 2.65e-10***gdp_d + -.0001843***dist

When I exclude the distance variable, the results appear much more reasonable however the coefficients are much larger:

tradeflow = -0.008eu_1 + 2.237***eu_2 + 0.937***wto_2 + 1.97e-10***gdp_o + 2.52e-10***gdp_d

I have not logged any variables so would the correct interpretation for eu_2 be an increase in trade by (e^2.237-1)*100= 836.5%.

Have I made any large errors for the coefficient to be this large?

Kind regards,
Leo

[Joao Santos Silva]

Dear Leo Hudson-Foster,

The problem, as far as I see it, is that you did not log the gdp and dist variables. There are other potential issues such as a choice of fixed effects and maybe the kind of data you are using, but not logging those variables is the major issue.

Best wishes,

Joao

[Leo Hudson-Foster]

Dear Joao Santos Silva

Thank you for your reply,
I have now run a PPML without the use of any fixed effects, as I concluded from a paper that a period like mine (between 1996-2020), the effect of EU membership on the countries that were members in every year of the period (France etc) would be included within the country-pair fixed effects coefficients. Am I right that excluding the fixed effects is the correct decision?

The results follow a similar pattern as before, the inclusion of the distance variable drastically changes the results:

tradeflow = -0.5897***eu_1 - 0.3898***eu_2 + 0.2655***wto_2 + 0.7562***ln_gdp_o + 0.7820***ln_gdp_d - 0.7675***ln_dist

tradeflow = -0.5298***eu_1 + 1.036***eu_2 + 0.1286***wto_2 + 0.7445***ln_gdp_o + 0.7800***ln_gdp_d

When distance is included, the coefficient for eu_2 appears to be completely wrong. Do I focus on the results without the distance variable added?

Kind Regards,
Leo

[Joao Santos Silva]

Dear Leo Hudson-Foster,

The standard in these cases is to include origin-time and destination-time fixed effects to control for multilateral resistance and pair-fixed effects to account for some of the endogeneity. If you do not include distance (or pair fixed effects), you cannot say that you are using a gravity equation.

Best wishes,

Joao

[Leo Hudson-Foster]

Dear Joao Santos Silva,

Thank you very much for your responses, I believe I have corrected my mistakes and now have obtained much better results.

Best wishes,
Leo

[Guest]

Dear Professor Santos Silva, may I ask how to interpret the ppml regression? I used the stata code as ppmlhdfe ExportValue lnDist LnGDP.....NTM (non-tariff measures). The coefficients are 0.4587***GDP -0.7543****Distance ....... -0.0128 NTM. Regarding the coefficients, Q.1 Can I interpret as a 1 unit increase in importing countries GDP would increase the export value by 0.4587 % or 45.87%. Q.2 Since NTM is applied not in the log form, is it possible to interpret the same as Q1 (additional NTM would reduce the export value by -0.0128% or 1.28%). Q.3 Are the coefficients percent point as for example GDP 0.4587% or the percent 45.87%?
Thank you very much in advance for your kind response.