Counterfactual with panel data

[Boris Godounov]

Hi everyone,

I try to build a counterfactual model with panel data from 1995 to 2014 for Germany. I can easily show the evolution of unemployment in this country during this period, using twoway.
But I would like to add in the twoway graph a line showing the evolution of unemployment if the unit labor cost would have remained the same as in 1995.

I have then two variables (let's simplify my model): U (unemployment) and ulc (unit labor cost).

The commands I used are the following (I used a fixed-effect model):
xtset country year
xtreg U ulc, fe

Then I've tried to build my counterfactual but it does not work and I'm not sure of what I am doing actually...

estimates store unemployment
xtreg ulc
estimates store unit
forecast estimates unemployment
forecast solve, begin(1996)
forecast adjust ulc=ulc_1995
forecast solve, begin(1996) prefix(new_)
twoway connected U ulc

Could someone help me please ?

Thanks a lot for reading.

BG

[Phil Bromiley]

Does your data go all the way to 2014? If so, I'm not sure why you need forecast rather than just predict or margins. I note that you did fe on the first model and re on the second so they're not comparable. The dv's are also different. This seems odd if you want to compare the two.

[Boris Godounov]

Thanks for you answer.
The data go all the wat to 2014.
I'm sorry for confusing re and fe, in my do-file I just used re because Hausman gives me no difference and I'm just focusing on Germany for my counterfactual (so fixed effects between countries are not relevant here).

In fact, I think I succeeded to do something:
1) I run my model
xtset country year
xtreg U ulc
estimates store U
xtreg ulc
estimates store ulc

2) I remove every country but Germany
keep if country=="Germany"
keep if year >=1995&year<=2014
forecast create U, replace
forecast estimates U
forecast estimates ulc
forecast solve, begin(1996)
forecast adjust ulc=ulc_1995
forecast solve, begin(1996) prefix(new_)
twoway connected U year||connected new_U year

It looks better because the model takes into account every country, but I just apply to Germany.
Am I bullshiting or is it something consistent ?