Announcement

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

  • is using lagged independent variables the same as using a lead dependent variable?

    hello ...

    i have a very simple question that i couldnt find an answer four. my question: is using lagged independent variables the same as using a lead dependent variable? to use the lead/lag variable in a unbalanced panel data setting should i use "xterg f.Y varlist, fe or Xtreg Y L.varlist, fe) directly or i need to add something ( e.g. "by: id" ) to the command?

    thanks
    sultan

  • #2
    I don't know what -by: id- would mean in this context or what it would accomplish. In any case, no, these two things are not the same. The simple reason is this: with -any_command F.y varlist-, the last observation in each panel will be omitted from the estimation sample because it's lead, F.y is undefined. By contrast with -any_command y L.varlist-, the first observation in each panel will be omitted because it's lag is undefined. So these two approaches are actually operating on different subsets of the data.

    Comment


    • #3
      I'm not sure I agree with Clyde's conclusion.
      Code:
      reg y L.x
      will have no L.x to match y[1], and will pair y[2] with x[1], ..., y[_N] with x[_N-1]
      Code:
      reg F.y x
      will pair y[2] with x[1], ..., y[_N] with x[_N-1], and will have no F.Y to match x[_N]

      So the same _N-1 data points are included in the estimation sample.

      Added in edit: I originally wrote that the same _N-1 observations are included in the estimation sample. This technically isn't correct, because as Clyde noted, one version "omits" the first observation of the dataset and the other "omits" the final observation of the dataset. But the sample "observations" are defined differently in the two versions, so the same data points are included both ways, with the same results. At least in the example I tried, when I thought - incorrectly - that F.y was not allowable syntax.

      So I haven't been particularly insightful, just made a dumb mistake that had the side effect of showing me the two versions yield the same results.
      Last edited by William Lisowski; 19 Aug 2018, 13:31.

      Comment


      • #4
        Yes, William is right. Sorry.

        Comment


        • #5
          Thanks clyde and Willian

          i have run the regression using both commands and the results are identical. That is, using lead dependent variable with contemporaneous independent variable is equivalent to using contemporaneous dependent variable with lagged independent variables.

          thanks

          Comment

          Working...
          X