Announcement

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

  • Be control variables as cause of treatment effect

    Good afternoon everyone, I am in the final stages of my thesis and am currently questioning my choice of control variables.

    I am investigating the impact of an existence of tax loss carryforwards on deal year returns of target companies using a DiD approach.
    The problem with my control variables such as firm size or e.g. leverage ratio is that all these variables could change after the treatment, especially in the treatment group, due to the treatment (as firms turn these "screws" to maximize the exploitation of loss carryforwards).

    Generally speaking: Is it always a mistake to include control variables that could be affected by the treatment? In my setting, I should not include control variables, because it is uncertain which measures are taken by the companies to increase the returns. It could be actions that increase the balance sheet (firm size) by raising new capital for example. Unfortunately, for orientation, no papers exist that address my topic in particular. Should I rather omit all control variables and work exclusively with fixed effects? I have never seen a study that does not use any control variables at all....


    My regression equation is just as follows:

    Yield =ß0+ß1year + ß2group + ß3*i.year##i.group+ß*Controls+FE+ε

    Controls: firm size, fixed asset ratio, intangible ratio, leverage



    Thank you very much and kind regards,
    Lenny

  • #2
    Hey Lennard,

    I feel these two things that people frequently do in their papers could be of help in your case:

    1) Use one period lags of your control variables to control for pre-treatment characteristics.

    2) If you decide to include control variables show explicitly the results both with and without controls. They should ideally point in the same direction.

    Best
    Sebastian

    Comment


    • #3
      How many treated units are there?

      Comment


      • #4
        Thank you for your answers.

        @Sebastian: Do you mean with the one-period-lags that I shift the values of the control variables by one period? So that STATA uses the control variables from t-1 in the deal year?
        Going further, I will show the results with and without control variables - thanks for the input.

        @Jared: Around 100 in the treatment group and around 200 in the control group. 4 observations per unit in each case.

        Comment


        • #5
          Lennard Maxim exactly that was what I was intending to say.

          Best
          Sebastian

          Comment

          Working...
          X