PPML gravity estimation for cross-section data -> question about the constant

[George Baker]

Dear all,

I have a question regarding the outcomes of my research. I am writing my thesis on the impact of migrant networks on bilateral FDI. My dependent variable is bilateral inward FDI stock in 2010 (provided by the UNCTAD). As a proxy for migrant networks I took the product of 2 countries' 2010 Chinese migrant stock (provided by the UNDESA). I have a total of 1646 observations of bilateral FDI between 64 countries.

My input in stata looks like this:
ppml FDIstock lnDist lnCHIN colony contiguity fta language $Partnerdummies $Hostdummies, nocons cluster(LnDist)

Where distance and the proxy for Chinese migrant networks are in logs, and the FDI stock is in millions. The other variables are control dummies and dummies for host and partner countries,

What I am quite puzzled about is the fact that whenever I run this regression with a constant, the estimated coefficient for lnCHIN turns out to be very different from the result without a constant, while all other estimates remain similar. There is a warning, stating that lnCHIN has very large values, and I should consider rescaling or recentering. However, from all the posts that I have read on this forum, I have noticed that it is advised to rescale the dependent variable, but I havent seen it being advised for independent variables. Does anyone perhaps know what the reason for this issue could be?

Kind regards,
George

[Joao Santos Silva]

Dear George,

There are two different issues here.

a) You should not estimate without the constant unless you do not omit the base category for the dummies. The fact that the results change shows that you are not doing it, so just include the constant.

b) Sometimes Stata struggles when the regressors have very different scales; the warning about re-scaling suggests that lnCHIN can be much larger that other regressors. However, if the estimator converges, you can safely ignore the warning.

Best wishes,

Joao

[George Baker]

Dear Mr. Santos Silva,

Thank you very much for your rapid reply! I indeed forgot to omit a base catagory for the host and partner dummies, I would furthermore like to thank you for your support on this forum regarding gravity equations and ppml in stata. I´ve been going through a lot of old forum posts and your answers are always very instructive.

Kind regards,
George

[ZHIDI LIU]

Hi George,

Could you please show me where I can find detailed data for FDI (by partner and sector)? I cannot find global bilateral FDI data in UNCTAD.

I look forward to your reply!

Thank you for your time.

Zhidi Liu

[Bjorn Gudmundsson]

Hi Zhidi,

I am also trying to find global bilateral FDI data but have been unable to locate it.
Did you manage to find it elsewhere or are you still looking for it?

Best wishes,
Bjorn Gudmundsson