Hello everyone, this is my first time posting on this forum.

I'm a master student in economics writing my thesis on determinants of bilateral trade of OECD countries using augmented gravity model estimation with linear and non-linear methods. For the linear method, I'm using OLS and FE model. and for non-linear I'm using PPML, FGLS, NLS, GPML. I have a few questions concerning the estimation and the dataset. I'm using a sample of all OECD countries from 1995-2018, I aggregate product-level trade from country i to j and my variables of interest are the MRT proxy (Head,2003), Bilateral exchange rates, Infrastructure (Road density and Infrastructure investment), Institutions(size of the government and openness to trade, I'm also considering Democracy index),
  1. I also focus on the importance of controlling for MRT through weighted GDPs of exporter and importer divided by the bilateral distance. I've seen that you said that this method is not perfect. However, I want to evaluate what would be the best way to control for MRT
    1. I'm testing first by using OLS pair FE + year FE to control for MRT however, time-invariant variables get omitted.
    2. Thus I do OLS country-year FE to control for MRT with robust errors clustered on pairid.
  2. Then I test this on PPML without FE + I do the same FE I mentioned above for PPML.
    1. What is your advice on PPML FE, which is the best FE model using OLS and PPML to estimate the time-variant and invariant determinants of bilateral trade?
  3. Fortunately, my results are significant but I still don't know which one to put on my thesis paper.
  4. What estimation methods would you recommend for me?
  5. Is aggregating trade by product level by bilateral trade flows per country pair the right way to represent bilateral trade or should I take the mean of individual product flows?
  6. Is imputing trade flows even allowed and okay approach to handle missing values? or should I only use the original dataset? for example for bilateral exchange rates, I only have until 2014, However, my dataset is until 2018, those 4 years are valuable for me, what would you recommend?
I would like to thank you for your time in reading this post. I wish you a great day!

Kind regards,

Tom Scholtes.

P.S.: I'm sorry, dear professor for tagging you, I urgently need your help! Your work has inspired me to choose gravity model topic for my thesis.