Announcement

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

  • F-Test for strength of instruments in GMM model

    Dear Statalisters,

    I am currently using a system GMM to estimate the relation between independence of the board and firm value and want to test the strength of the used instruments. However, I am still struggling to understand how to correctly specify the F-test for the first stage.

    I am using the xtabond2 command in Stata 14.2:

    xtabond2 L(0/3).value indep n o industry* i.year, ///
    gmmstyle(L(1/3).value n , lag(4 6)) ///
    ivstyle( o industry*) ///
    ivstyle( i.year, eq(level)) ///
    twostep robust orthogonal

    where n stands for the other variables I have included that are endogenous and o for the other exogenous variables I have included.
    industry* stands for the industry dumies and i.year for the year dummies. Both my Hansen and AR(2) tests returned favourable values and I wanted now to test the strength of my instruments by assessing the F-test of the first stage regression. As I am using three lags, I used for the difference equation the following first stage estimation:

    quietly reg d.indep l(3/5).(indep n l.value l2.value l3.value) D.(o industry*) i.year
    test l3.indep l4.indep l5.indep l3.n l4.n l5.n l4.value l5.value l6.value l7.value l8.value

    but for the estimation in levels I am less sure as I am not sure how many differenced variables I have to include.
    Am I using only the differenced variables between the periods (so two differences between t-3 and t-5) or differences d.3 to d.5?
    So the two option I am considering are:

    quietly reg indep d(4/5).(indep n l.value l2.value l3.value) l.(o industry*) i.year
    test d4.indep d5.indep d4.n d5.n d4.l.value d5.l.value d5.l2.value d5.l3.vlue

    quietly reg indep d(3/5).(indep n l.value l2.value l3.value) l.(o industry*) i.year
    test d3.indep d4.indep d5.indep d3.n d4.n d5.n d3.l.value d4.l.value d5.l.value d5.l2.value d5.l3.vlue

    I would be very thankful for help or advise as I am currently a bit stuck.

  • #2
    Hi,
    I have the same problem in my work. Did you find any answer to this issue???

    Comment

    Working...
    X