[Joao Santos Silva]

Dear Javier Rua Montes,

Yes, that should not be a problem; just use predict.

Best wishes,

Joao

[Javier Rua Montes]

Greetings Mr. Joao,

I am trying to develop a gravity model for the soybean trade globally (I have 105 importing countries and 98 exporters) for a period of 11 years (2006 - 2016). I have 8798 observations. I would like to know if the models and the commands I am using are the most suitable:

flux: import flow volume
PBIp: per capita PBI
exchange rate: exchange rate in dollars
impTot: Total Import
i.imp: fixed effects of the importer
i.exp: fixed effects of the exporter
i.inter: fixed effect of the importer-exporter interaction


xi: ppml fl_ux i.imp i.exp ln_PBIp_imp ln_PBIp_exp lnimpTot_imp lnimpTot_exp ln_rate_exchange ln_Dist, cluster (inter)

xi: ppml fl_ux i.imp i.exp i.inter ln_PBIp_imp ln_PBIp_exp ln_impTot_imp ln_impTot_exp ln_rate_exchange ln_Dist, cluster (inter)

The first model does not pass the Ramsey test but the second does. However, in the second model the variables that do not vary in time as distance are not eliminated, I read in the forum that by including fixed effects of pairs of countries (import -export) this effect absorbs all the invariant variables over time and I fear that the model is not good either.


I thank you very much for the previous answers and I thank you for what you can give me now.


Cordially,
Javier Rua

[Joao Santos Silva]

Dear Javier Rua Montes,

Only you can decide which model to use, so I am not going to comment on that. About your second question, my guess is that distance is not dropped because Stata is instead dropping a fixed effect. You can estimate the same mode using xtpoisson with the FE option and that will avoid that issue.

Best wishes,

Joao

[Susan Jain]

Hello prof joao

[Susan Jain]

Hello Prof Joao
I WANT to incorporate time fixed effect and host country fixed effect in PPML model. Earlier I thought by including time dummy and country dummy it’s possible. But after reading your threads I realise it’s not so. Pl guide how to account for time fixed effect and country fixed effect in PPML model.
please help 🙏🏻🙏🏻

[Joao Santos Silva]

Dear Susan Jain,

You can do that, but the estimation would be very slow. It is better to use ppmlhdfe which will give you the same results much faster.

Best wishes,

Joao

[Susan Jain]

Dear Joao,
Thanks for the reply indeed.
I have seen papers using time fixed effect and country fixed effect in Tobit and negative binomial models. To my understanding they have random model specifications , then how come they have fixed effects … or just by using time and country dummy in these models we can say it has fixed time and country effect.

Can we estimate time fixed effect in Tobit without stating ‘ fe ‘ in the end .. we
Also, can you suggest some model specification to check robustness of PPML estimates , most papers ask for robustness check

[Joao Santos Silva]

Dear Susan Jain,

Please read about the incidental parameter problem; maybe you can start by reading a good textbook. You will find that there are no real fixed effects versions of the Tobit and NB models. However, PPML can be estimated with fixed effects and has excellent properties, as discussed in

Wooldridge, J. M., “Distribution-Free Estimation of Some Nonlinear Panel Data Models,” Journal of Econometrics 90 (1999), 77–97.

To check the model, you can use a RESET test.

Best wishes,

Joao

[Susan Jain]

Thanks again Professor! I will surely read the mentioned paper.

[Ronan Moore]

Dear Joao Santos Silva ,

I would really appreciate your help. I aim to find how a single country's exports (Ireland) have been impacted by the Brexit referendum, and I am looking to use the ppmlhdfe command. I have created a Brexit referendum dummy for years >= 2016. However, as suggested by numerous posts on various Stata gravity forums before, I need to account for multilateral resistances. I understand from previous statalisters' posts that as I will only have one exporter (Ireland), importer-time fixed effects are not needed.

My goal as previously mentioned is to find how Irish trade has changed pre- and post-Brexit referendum (2016). I have also read that remoteness indexes have been used in the past to proxy multilateral resistances for Nx1 papers, but Head and Mayer (2014) argue against their use. How would I account for multilateral resistances instead? Yotov et al. (2016) and many papers strongly suggest to include these resistance terms, so it would be great if I could include them in my model if possible. Would I need outwards and inwards multilateral resistances for a study on a single exporting country too?

Furthermore, if there are no other workarounds for controlling for multilateral resistances besides remoteness indexes and I do end up including these in the model, should I still include the exporter-time fixed effects in the model as well?

Apologies for the lengthy message. Any advice would be great! If you have any possible regression ideas and codes that would apply to the above context please let me know.

Kind regards,

Ronan

[Joao Santos Silva]

Dear Ronan Moore,

My suggestion is that you use the standard importer and exporter time-varying fixed effects but, because you only have an exporter, the exporter fixed effect is just a time effect.

Best wishes,

Joao

[Ronan Moore]

Thank you very much Joao Santos Silva for your help.

Kind regards,

Ronan

[diana amaya]

Hi Professor Santos,

I have a question similar to Ronan Moore's, but in my case, I am working with one importing country and 63 exporting countries. I attempted to implement the suggestion you provided, but I encountered multicollinearity issues.

I am working with quarterly panel data, and I am trying to establish the OLS model before running a PPML model. Here is the code I am currently using: xi: reg ImpVol GDP_Exp GDP_US Dist RealImpPIndex RealTariff Prod_US Real_Exchange_Rate Refusal FTA_USA covid_cases US_cases AD i.exp_time i.time i.pair_id

Do you have any suggestions on how to address the multicollinearity issue or any other recommendations for improving this model?

Thank you,

Diana Amaya

[Joao Santos Silva]

Dear diana amaya,

I suggest that you use the user-written command reghdfe and absorb the fixed effects (check the examples in the help file); similarly, for PPML you should use ppmlhdfe. From what I can see, you are introducing in your model variables that are perfectly collinear with the fixed effects, and therefore they will drop out. You need to study carefully which variables can be included with the set of fixed effects you need.

Best wishes,

Joao

[diana amaya]

Thank you,

I have run the model including only the fixed effects, but collinearity persists, particularly with the exporter-time fixed effects. Here is the model I used: This is the model: xi: reg ImpVol i.exp_time_ i.Quarter i.panel_id

The best solution I have found so far is to exclude the exporter-time fixed effects. However, I am concerned about the consistency of this approach since Yotov et al. (2016) suggest including these resistance terms in the model.

Diana