Announcement

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

  • ADF test and data transformation

    hello all,

    I have some variables (dependent as well as independent) that appear to perform better after taking the log (ln) of the return ln(pt/pt-1).
    Before the transformation, it were just prices. Dickey fuller test (ADF) indicated that most of them were non-stationary.
    Since I performed the ADF test before transformation, I believe I have made a mistake here. What is common? Doing ADF before or after transformation of the variables?

    The thing is, when one transform a variable (in this case the natural log of returns) the data can become stationary. this is not problematic I believe (?), but:
    1. in situation 1 (no transformation) --> ADF --> stationary as well as non stationary variables --> one might use the ARDL model;
    2. In situation 2 (transformation) --> ADF --> everything is stationary due to transformation --> use standard OLS.

    So what is a good approach according to you?

    Second question: can you still use the ARDL model in case of stationary variables? or just only in case of a mixture of I(0) and I(1)?

  • #2
    Do you just have single time series or do you have a panel? Because stationarity it much more a concern in the first than in the latter.

    Comment


    • #3
      I have time series.
      Slight edit of my question:
      "Before the transformation, it were just prices. Dickey fuller test (ADF) indicated that most of the prices were non-stationary."
      And
      The thing is, when one transform a variable (in this case the natural log of returns) the data can become stationary. this is not problematic I believe (?), but:
      1. in situation 1 (no transformation) --> ADF --> stationary as well as non stationary variables --> one might use the ARDL model;
      2. In situation 2 (transformation) --> ADF --> everything is stationary due to transformation --> use standard OLS. ""
      Hence: the model I end up using depends on the ADF test which in turn depends on whether I perform it before or after the transformation.


      Sorry for being unclear.
      TL;DR: 1: Should one perform an ADF test before or after data transformation (time series data)
      2: can you still use the ARDL model in case of stationary variables? or just only in case of a mixture of I(0) and I(1)?

      Last edited by Robbert Henk; 20 Apr 2018, 07:20.

      Comment


      • #4
        You validate your regression based on tests of the variables used in those regressions. So if you use the transformed variable in a particular regression, then you should test the transformed variable. And vice versa. You can use ARDL models with stationary variables.

        Comment


        • #5
          Thanks jesse have a nice day!

          Comment

          Working...
          X