Announcement

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

  • Extrapolation using epolate function

    I am working with a panel of countries for which I have several missing values, especially for the inequality and distribution variables. I am attempting to extrapolate the values that fall before/after the data coverage period, and reading the different blog posts I have found that I could use the epolate command combined to the ipolate one. However, whenever I do so, the values before the data coverage period are put in ascending order (instead of the reverse, which would make more sense). Here's the code for the variable MC1 (Middle-Class) which I am trying to extrapolate:

    ipolate MC1 year, generate(iMC1) epolate

    I have attached an image of the results I get (with MC1 being the variable to interpolate and iMC1 the end result of this command). I am aware that there are other interpolation methods (which I have tried, such as csipolate, cipolate), but I don't know of any other extrapolation command.. Is there any, or does it have to be a code/loop? I have found a similar question on a blog, which was left unanswered, I hope this is not too repetitive.
    Attached Files
    Last edited by Catherine Brunet; 11 Mar 2015, 07:33.

  • #2
    epolate is an option of ipolate, not a command or function. (This is not just terminology; anything typing help epolate will be puzzled.)

    Screenshots are of limited use here, as the FAQ Advice does tell you (Section 12). With or without your screenshot, I can't understand your point about the order of observations.

    pchipolate (SSC) automatically extrapolates too, which is precisely why no explicit option is required.

    Comment


    • #3
      Hi,

      I would like to extrapolate so called 'depreciation rates'(delta) for which I have data for 1995-2008.

      As I have a panel data set, my command was..

      by id: ipolate delta year, gen(idelta) epolate

      However, this generates TOO many negative values which would (from a theoretical perspective) force me to make very strong assumptions.
      So, how could I create less negative values? Maybe by integrating a squared time trend...??

      Olli

      Comment


      • #4
        Linear extrapolation is just about the crudest time series forecasting method that could be imagined, short of astrology and inspecting entrails. As you have found, it won't even respect the natural range of a response. You should surely be checking out other methods. What might work best for depreciation rates I can't tell you, but I am confident that will be discussed in your literature.

        Comment

        Working...
        X