Thank you. Yes, we have used PPML for Gravity model in ECOWAS. It is quite relevant 'In addition to solving the problem of heteroskedasticity, the PPML estimator also addresses the problem that sometimes takes the value of zero, in which case is not defined'. Here is the recent paper we published using it.  'Osabuohien, E., Efobi, U., Odebiyi, J and Fayomi, F. (2017). Financial Development, Trade Costs and Bilateral Trade Flows: Connecting the Nexus in ECOWAS. In, Seck. D (Ed.) Investment and Competitiveness in Africa (pp. 153-176), Geneva: Springer International Publishing. DOI: 10.1007/978-3-319-44787-2_8. You can also get it in researchgate. Best wishes

Calculating Gravity model zero trade by using OLS, Tobit and MMPL ( how to run Park test to check for heteroscedasticity )

I am running regression to calculate gravity model

Dependent variable= log exports in (OLS, Tobit) Level Eports in mn US$ (PPML)

Independent variable= GDP-i, GDP-j, Distance-ij, Abs-difrential-pcgdp-ij, population-j, weighted distance-ij, Dummies (border, religion, language, regional agreements, landlocks, WTO)

Expected negative results= Distance-ij, Weighted Distance-ij, landlocks

**OLS resutls**

 **Tobit Results**

**PPML resutls**

why most of the variable gone negative in PPML especially GDPi and GDPj. Please assist me

Regards

Saqib

Dear Loures!

Thank you so much for the detailed reply. I hope your suggestion will work. So that means there is no way to estimate (1xN) countries data by using fixed effects. 

Alexandre Loures how can one estimate a 1xN gravity model? In my case I use US and 30 trading partners (only imports to the US). Do you have any suggestions? If not fixed effects, what regression can (should) I apply?

Alexandre Loures thanks a lot, that makes perfect sense!

Dear Alexandre Loures ,

My analaytical framework is similar with Member One , I wiil estimate 3 models. Firstly, the export of Turkey to 30 main trade partners, secondly the import of Turkey from 30 main trade partners and lastly total trade (export+import) for robustness. So my samples is 1xN. I will use PPML as estimator. In this contexti i have some questions:

1. Should i use the variables in level form (except distance). I dont have zero trade flows. So, if i use log form, is it problem with regards to PPML estimator?

2. I want to add multilateral resistance terms to my models. Which one should i use (exporter-time fixed effects, importer-time fixed effects....).

3. When i use multilateral resistance terms, can't GDP variables estimate? They are so important variables with regards to Gravity model.

Thanks Alexandre Loures , actually i saw some papers used export+import as dependet variable. But i will follow your idea. and i decided to change my analaytical fremework. I will estimate firstly the export of Turkey to all partners, secondly the import of Turkey from all partners. Because also i would like to include zero trade flows. What about you?

Besides, you mean that in the context of 1XN sample we should use just year dummy for exporter-time fixed effect. But, for importer-time fixed effect we should use both country and year dummies. Is it correct? Thanks in advance.

Dear Alexandre Loures

Very thanks for your kind replies

Best regards,

I have for Arab World for ten years, panel data ,

I used stata, but the results of GDP's exporter are negative coefficients for disag .products in PPML?

how can I solve this problem?

Lastly in Yotov at all (2016), there is code for export-importer fixed effects. How can i modify them for my analysis form.

Dear Alexandre Loures ,

Actually i have just solved problem about getting exporter-time and importer-time fixed effects by using stata codes. But in my sample importer time fixed effects constitute from 450 dummies. So in this case PPML doesn't estimate these effects. That is my problem. I didn't understand what is the problem.

Best regards,

Besides, i read some papers which states that you can divide your trade data with a fixed figure such as one billiion, now that we use trade data in level form in context of PPML estimator and some trade data fairly high. In this case our estimation findings don't change. Do you agree with that?

Best regards,

Lot of things changed in all this years. If someone see this question in 2021, here are the newest recommendations:

Preprint International trade and the gravity model: recent evidence i...

Best regards!