Hi to everyone,
In a course I am attending, I am asked to reproduce a study presented in a paper and related to the psychological costs of a new currency adoption. In the paper - through a difference-in-difference approach - they try to estimante the change in personal economic utility perceived by individuals after the adoption of euro in Slovakia. The control group is represented by Czech Republic.
So far, I estimated the DD estimator through different approaches (OLS, 2x2 table, etc). In addition, I am asked to conduct the analysis through another method (propensity score matching). I wrote the coding in Stata as follows:

logit stfeco country agea eisced
predict propscore
teffects psmatch (stfeco) (country agea eisced) if essround == 1, gen(matchnr) nneighbor(3)
teffects psmatch (stfeco) (country agea eisced) if essround == 1, atet nneighbor(3)
tebalance density
tebalance summarize

The result (ATE) I am getting does not refelect the one I got with the previous DID estimation, i.e., a decrease in perceived economic utility (-1.025277). What I get is a positive ATE showing an increase in perceived economic utility after the treatment was introduced, thus, contrasting the general results of the paper I am trying to replicate.
One reason why this happen may be given by the fact that for 2169 first observations, when generating matchnr, there are no matches and thus the size of our sample is reduced (from 7859 from the DID estimation to 4122 for PSM) and our estimate of ATE is biased. It is the only explanation I can give myself given my limited Stata abilities.
Is there a way to cope with the missing macthnr or I just run PS matching and explain in my paper why that estimation is biased?
In any case, you can find attached my dataset, the Do-file I have written so far and the paper I am referring to - if that can help somehow.

Thank you.