[Andy londale]

Hi there!

I am am new to the stata forum, but I came here to ask a similar question: when I use the ppml FE command, I do not get any coefficient for the "constant" (xtpoisson...fe)
Is this normal?

Thanks

Bets Regards,

Andy

[Joao Santos Silva]

Hi Andy,

Welcome to the forum. I do not believe there is a FE option for -ppml-; can you please clarify.

Joao

[Andy londale]

Hi joao!

I did "xtpoison......, robust fe" in stata also I do not get a coefficient for the "constant" so I was wondering if this was normal?


Thanks a lot!


Andy

[Joao Santos Silva]

Dear Andy,

Thanks for clarifying. Yes, that is normal but annoyingly this is not consistent with what Stata does in the linear model. Essentially, without further restrictions, the constant is not identified when we include a full set of fixed effects. In the case of linear models Stata reports a constant by normalizing the fixed effects in a different way.

Best regards,

Joao

[Andy londale]

Hi again Joao,

I just had another specific question.

I want to take into account exchange rate volatility. If I subtract the exchange rate of one country (in US dollars) form another country (in US dollars). And I kept the absolute value of this calculation and plot it into my regression (hence if |2-3|= 1). Would this be a good measure for that? Or I am doing something wrong?

Thanks a lot!

Best Regards,

Andy

[Joao Santos Silva]

Sorry Andy, this is out of my area of expertise :-(

Best wishes,

Joao

[Gabriel Arrieta]

Hi Joao,

Nice to meet you.

I am doing a dynamic-panel data model to represent the effects of FTAs (Trade Creation and Trade Diversion) in Trade Flows (Log Exports and Imports) for Peru and the 90% of its trading partners (from 1995 to 2015). Nowadays, I have problems with the estimator, I don't know if i have to use FE or RE, or as it say in the new literature I have to use PPML. Is there any characteristics of this kind of models to be sure that what estimator I have to use? I know that PPML estimator has the characteristic of consider the zero trade flow, but is this the unique advantage?

Another question Mr. Santos, if I introduce dummies for measuring trade creation and trade diversion, can I use the PPML estimator? Is there any difference about the way that I have to analyze the output? Could you please give me an example between the diference of intrepretation of FE, RE and PPML estimator?


Thank you very much, Mr. Santos.

[Joao Santos Silva]

Dear Gabriel,

Any model where the dependent variable is the log of trade is likely to give you very misleading results. The main advantage of PPML is that it allows us to estimate models for trade, rather than for log of trade. There is no problem at all in including trade creation and trade diversion dummies in models estimated by PPML. Have a look at our "Log of Gravity" paper for further details, OK?

Best regards,

Joao

[Gabriel Arrieta]

Thanks Mr. Santos!

[Andy londale]

Hi Joao!

I wanted to ask you a quick question!

If my dependent variable is migration flows instead of trade, should I log the dependent variable in this case? And should I use ppml even if I have no zero-values on the dependent variable?

Thank you very much!

Best Regards

Andy

[Joao Santos Silva]

Hi Andy,

Do not log it even if you do not have zeros; ppml is much safer.

All the best,

Joao

[dupont john]

Hi Joao,

I also get different results for my PPML and FE regression while analysis trade in south America. Could I ask you why PPML is safer?
I also do not have any zero-values on the dependent variable hence I am a bit confused why my results differ.

Thanks a lot!

Best,

JD

[Joao Santos Silva]

Dear JD,

When you say FE regression you mean linear FE regression and not Poisson FE regression, right?

Anyway, the zeros are not the issue; the problem is that a log-log model will not estimate the elasticities you have in mind unless the data has very special characteristics. So, estimating a model where the dependent variable is the log of trade is always a bad idea. If you want to include Fixed Effects, you can do Poisson regression with FE (see -xtpoisson-).

All the best,

Joao

[dupont john]

Hi again joao!

Thanks for you reply! Very useful!

Could I just ask you how to interpret the results as a percentage? as with PPML we do not have a log on the dependent variable hence I am confused how to analyse this

Thank you so much!

Best Regards,

JD

[Joao Santos Silva]

Dear JD,

You do not have logs in the dependent variable but you have an exponential function on the RHS. So, the interpretation is exactly as in the usual gravity equation. Actually, it has to be like that because in both cases you are trying to estimate the same parameters (the coefficients in the gravity equations), so the interpretation has to be the same.

All the best,

Joao