Announcement

Collapse
No announcement yet.
X
  • Filter
  • Time
  • Show
Clear All
new posts

  • Two-step estimation problem when the second regression is teffects, bootstrap is the only choice?

    Dear all,
    I have an issue that the first step estimation predicts the residual (actually is the last residual) but the second stage is the teffects estimators.
    To get a consistent standard error, the only way I know so far is using bootstrap. Recently, I heard that the new Cameron (2022) book claims that we can use a GMM method instead, but the author didn't say it very clearly.
    Does anyone know how to solve this issue? Actually I am facing a three-step estimation problem...


    Here is the example:

    global $xlistreg2 are house characteristics
    global $xlistreg1 are independent variables I am interested
    mbsmoke is the dummy variable like in birthweight data used by Cattaneo (2010)

    regress logPrice_Sale_ $xlistreg, robust
    predict y_hat,
    gen residual=logPrice_Sale_-y_hat
    sort houseid Close_year Closetime Closeday
    by houseid: gen last_residual=residual[_n-1]

    The second regression is the teffects doubly robust estimator:

    teffects aipw (logPrice_Sale_ last_residual $xlistreg1 $xlistreg2 ) (mbsmoke last_residual $xlistreg1 $xlistreg2, logit), aequations
Working...
X