The gravity model

[Ella Imogirie]

Good day
I am estimating trade flows from nigeria to 20 other countries. I am using sectoral data, thus, the issue of zero trade occurs. I use PPML approach to combat this issue. Please how do i interprete the results. For instance my language dummy under ppml has a coefficient of 0.960. does this mean that if two countries share a common language they trade 9.6% more,or i have to find exp(0.960) - 1? Thanks

[Paul Mueller]

Hi Ella, your description is very short but that all depends on your model. I'm not familiar with the implementation of gravity models in Stata, it might help if you post your full output. The regression itself seems to work via a log-log model, right?

However, even if you don't need to use the exp(coefficient): a coefficient is not a % value but depends on the units of your variables.

But as long as there is nobody around who's experienced with gravity models in Stata you need to give us more information (i. e. your Stata in- and output).

[Joao Santos Silva]

Dear Ella,

For regressors that are in logs (e.g., distance), the coefficient is an elasticity. For other regressors (e.g., dummies) the coefficients are approximately semi-elasticities and the approximation is good for small values of the estimates (say, below 0.1 in absolute value). For large values of the estimates you should compute the exact semi-elasticity using the expression you mentioned: exp(b) - 1.

Joao