Online Summer School on International Trade and Gravity Models: 18-20 September 2023

[Joao Santos Silva]

Dear all,

This course may be of interest to some of those who frequently ask questions about PPML and gravity equations. Basic knowledge of Stata is required.

Best wishes,

Joao

[Joao Santos Silva]

Dear all,

Just a quick reminder that the deadline to register for the Summer School on International Trade and Gravity Models is approaching fast (Sepember 3rd). Some of you may also be interested in the Macroeconomics/DSGE part of the Summer School. For more details, please see here.

Best wishes,

Joao

[Takesure Tozooneyi]

I have a question on the use of PPML estimation of gravity models. I am trying to understand the effect of regional trade agreements (RTAs) on poultry exports during NewCastle (ND) disease events. So my dependent variable is Poultry exports (Xijt) from country i to country j in time t, RTAijt t is a dummy which takes the value of 1 if country i and j have a trade agreement in time t, NDit is a dummy which takes the value of 1 if the exporter i has a ND event in time t. One of my dependent variable is the interaction of RTAijt and NDit which is how I intend to capture what I want to measure. I am yet to estimate the equation but I am looking for help on how to interpret the coefficient of the interaction term.

I will also be grateful if I can get reference materials on how to interpret the coefficients of interaction terms of dummy variable in PPML estimations of gravity models.

[Joao Santos Silva]

Dear Takesure Tozooneyi,

The interaction gives you the change in the semi-elasticity in periods with ND = 1. More precisely, the difference between the effect of a RTA with and without ND is given by exp(B_rta) - exp(B_rta + B_rta*nd), where B_rta is the coefficient on RTA and B_rta*nd is the coefficient on the interaction. For an example of a paper doing similar things, see https://doi.org/10.1016/j.jdeveco.2018.08.014

Best wishes,

Joao

[Takesure Tozooneyi]

Dear Prof Joao Santos Silva

I am really grateful for this. I have been looking for such information for a long time.

[Takesure Tozooneyi]

Dear Prof Joao Santos Silva

So building on the same question, how would I capture the effect of ND with and without RTA?. Last time, I tried to re-derive how we measure the effect of RTA with and without ND. I failed to reach your answer. hahaha. I would be grateful if you may show your steps as you respond to this current question..

Thank you in advance and looking forward to hearing from you,

[Joao Santos Silva]

Dear Takesure Tozooneyi,

Suppose your model is

y = exp(b0 + b1*x + b2*z + b3*x*z)

where x and z are dummies. The relative effect of turning x from 0 to 1 when z=0 is

exp(b0 + b1)/exp(b0) = exp(b1)

The relative effect of turning both x and z to 1 is

exp(b0 + b1 + b2)/exp(b0) = exp(b1 + b2)

So, the difference between the two is

exp(b1 + b2) - exp(b1)

Hope this helps.

Joao

[Takesure Tozooneyi]

Joao Santos Silva \

Thank you very much Prof. I really appreciate.

[Takesure Tozooneyi]

Joao Santos Silva

Just a follow-up question. When we turn both x and z to 1, what happened to b3?

[Joao Santos Silva]

Dear Takesure Tozooneyi

My mistake! Of course b3 should be included. Thanks for spotting that.

Best wishes,

Joao

[Takesure Tozooneyi]

Joao Santos Silva

Thank you so much

[Takesure Tozooneyi]

Hie Prof Joao Santos Silva

I have a question. When dealing with bilateral trade data for gravity modelling, does it necessarily mean that every exporter should also appear as an importer such that we have the same frequencies for all exporter and importers? I have a data set with more importers than exporters and have been struggling on how best to make it usable for gravity modelling.

Looking forward to hearing from you,

[Joao Santos Silva]

Dear Takesure Tozooneyi,

Ideally, we should have all countries of the world as importers and exporters, but in practice we often do not have data for all the countries and do not have the same number of importers and exporters. Notice also that some datasets do not report zeros and we have to fill in those observations.

Best wishes,

Joao

[Takesure Tozooneyi]

Joao Santos Silva

This is clear. Thank you so much.

[Takesure Tozooneyi]

Joao Santos Silva

Hie Prof, I have a question building on my first question. I am realizing that, given that I am using exporter-fixed effects, I may not be able to identify HPAI since its exporter-specific and time variant. How can I go about this? Can I drop the exporter-time fixed effects but keep importer-time fixed effects and country pair fixed effects and use other control variable to cater to the exporter-time fixed effects? Are there better ways of dealing with this?