should poi2hdfe results and ppml results be indentical?

[willwang china]

Originally, I use ppml to run my model with multiple fixed effects in a loop, however, due to large datasets, ppml is very slow. So I tried another method: poi2hdfe. If I understands correctly, in theory, poi2hdfe results and ppml results should be identical.

In my results, the results between these two methods are same for most of the time, however, I encountered a problem as below:

	ppml	Poi2hdfe
VARIABLES	lnuv	lnuv
lntariff	0.0734***	0.0734***
	(0.0205)	(0.0205)
intcounp	-0.828***	0
	(0.321)	(0)
mp	0.698***	0.698***
	(0.151)	(0.151)
edds	0.823	0.823
	(0.616)	(0.617)
labord	0.368*	0.368*
	(0.202)	(0.202)
capitald	-0.192*	-0.192*
	(0.112)	(0.112)
agrid	-0.336*	-0.336*
	(0.186)	(0.187)
fta_wto	0.0884	0.0884
	(0.0845)	(0.0846)
eud	-2.021*	0
	(1.140)	(0)
wtod	3.916***	0
	(0.873)	(0)
QR	-0.0612	-0.0612
	(0.0526)	(0.0526)
Constant	-19.90***	0
	(4.510)	(0)
Observations	3,430	3,430
R-squared	0.469

the main variable is "intcounp", however, in ppml, the coefficient of intcounp is significant, while in poi2hdfe, "intcounp" is omitted. Also, "eud", "wtod" are also omitted in poi2hdfe.
my code is:

for ppml: cap noisily ppml lnuv lntariff intcounp mp edds labord capitald agrid fta_wto eud wtod ADP CV QR SG SPS SSG STE TRQ XS countryproduct* year*, cluster(countryproduct)
for poi2hdfe: cap noisily poi2hdfe lnuv lntariff intcounp mp edds labord capitald agrid fta_wto eud wtod ADP CV QR SG SPS SSG STE TRQ XS, id1(countryproduct) id2(year) cluster(countryproduct)

I'm wondering what would be the problem for the above results and reasons behind them.

Thank you very much!

[willwang china]

should I upload my datasets?

[Joao Santos Silva]

You get the same results with both commands, but it looks as if you are including in your model variables that are collinear with the fixed effects; poi2hdfe drops these variables whereas the way you are using it ppml drops the corresponding fixed effects.

Best wishes,

Joao

[willwang china]

Thank you so much is there any way for poi2hdfe to keep the variable "intcounp"? Because this variable is the main variable in my model.

[willwang china]

Thank you so much is there any way for poi2hdfe to keep the variable "intcounp"? Because this variable is the main variable in my model.

[Joao Santos Silva]

I do not think so, but how is that variable defined?

Joao

[willwang china]

It's a trade policy variable times country-pair dummy. the trade policy variable is defined using the number of trade policies levied on one trade product.

[Joao Santos Silva]

I am not sure to understand it, but if it being dropped it is collinear. Are you sure you are including the right fixed effects? maybe you can try ppml_panel_sg and see what happens.

Best wishes,

Joao
PS: Your dependent variable is not in logs, is it?