Thank you for your feedback. regarding your first question, the intercept of regression is the same as the fixed effect for USA. That is so because we left out USA to avoid perfect multicolinearity. In the LSDV based implementation of the FE model, there isn't a separate intercept of regression.

Regarding your second question, unfortunately, I have not looked at applying FE models to the problem of estimating phenomena that are modeled using the translog functional form. It would be an interesting research exercise, given a suitable dataset.

'best