I am making some simulation in order to show the performance of the Spline Regression vs other methods like Smoothing Splines, Hodric Prescott filter, etc., I would like to knot if it's better to sample the dependent variable $X$ out of a uniform distribution rather than fixing them to be equidistant. I know that it depends on what I want to do, if I am trying to approximate a signal that is sampled each 1 second, then the observations will be equidistant, but is it better to test the performance in a more general case to sample $X$ ?

  • $\begingroup$ That's going to depend entirely upon your application area. The key question is: what does your actual data look like? If it's sampled each second, as you say, then you'll learn more from your simulation study by duplicating that than by constructing $X$ in some other way. $\endgroup$
    – jbowman
    Mar 12, 2019 at 19:15
  • $\begingroup$ Just a comment, I would not use HP filter with a non-equidistant X. $\endgroup$
    – Gi_F.
    Mar 14, 2019 at 12:20


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.