[Cynthia Hourizene]

this is very clear. Again thank you M. Clyde S. for you time and help

[Cynthia Hourizene]

M. Joao Santos S. ,

I have two small questions about PPML estimation model:

I would like to know about the dependent variable. my experience with PPML I never took the log of the dependent variable on the Right Hand Side (RHS); I have tried today a double log PPML and it seems to give better results. So I would like to know

if it is ok to or, why or why not take the log of dependent variable?

my second question is about clustering the log of distance.
I am sure a thread where you explained the use of "cluster(lndistance)" before but I cannot find it again.
could you please explain the importance of clustering the log of distance in the ppml syntax?


my code below:

. ppml lnTrade_resc lnLaggedTrade_resc NAFTAmm EUm ASEANm MERCOSURm lnExpGDP lnImpGDP_re
> sc lnExpPop_resc lnImpPop_resc lnExpProd_resc lnImpProd lnExpFarmInc_resc lnDistance B
> orderDummy qdatedum* Exporter_FE* Importer_FE*, cluster(lnDistance)

thanks

[Joao Santos Silva]

Dear Cynthia,

The reason to use PPML is not to log the dependent variable because that ruins the interpretation of the estimates. So, do not use log trade as the dependent variable. Also, it looks as if you are including the lag of log trade as a regressor; that is not standard and I would not do it.

The reason why we need to cluster is to allow for possible correlation between the errors for the same pair of countries.

Best wishes,

Joao

[Cynthia Hourizene]

Thank you very much sir.