0

This must be a fairly standard question: I have some return data that has errors (they are actual errors, not just large returns). I am thinking of the best way to correct this so it doesn't influence my regressions. One idea is to simply set returns that are in the extreme quantiles to mean return. Another solution: have lm ignore these extreme values. Is there a built in way in lm to make it ignore extreme values? I know matlab has something called roust regression which does just this.

Alex
  • 19,533
  • 37
  • 126
  • 195
  • 1
    [The CRAN Task View](http://cran.r-project.org/web/views/Robust.html) on robust statistical methods should get you started. – Josh O'Brien Mar 09 '13 at 04:50
  • Robustness does not necessarily mean to remove outliers, but might also be based on estimators that are not as easily influenced by them (the most trivial being mean vs. median). Thus, I see no duplicate there. – Thilo Mar 09 '13 at 12:56

1 Answers1

4

Is there a built in way in lm to make it ignore extreme values?

Yes. You need to look at rlm.

For more reading material, look at the CRAN Task for robust methods. (Josh already gave this link)

jackStinger
  • 2,035
  • 5
  • 23
  • 36
  • Here is an example for robust regression http://r-statistics.co/Robust-Regression-With-R.html – Selva Jan 01 '16 at 17:33