2
$\begingroup$

I have a few questions regarding quantile regression and how to interpret the results. I have several independent variables and I want to figure out which combination of variables have the highest predictive value. I have preformed quantile regression, in Eviews, on all possible combinations of the 10 independent variables I have.

I would like to know which statistical measure provides the best information with regards to evaluate which combinations of independent variables I should use?

The t-statistics seems to be more signification at the extreme quantiles (90, 80, 20, 10%), than the quantiles more centered in the middle (40, 50 ,60%). Why is it so?

$\endgroup$
  • $\begingroup$ The regression at each quantile is a separate analysis. Sometimes IVs are more important at the extremes, sometimes near the middle, sometimes it's some of each. $\endgroup$ – Peter Flom Apr 18 '13 at 10:06
  • $\begingroup$ Thanks for the feedback. I get that the IVs that are important will vary depending on the quantile. What I'm not sure about is if I can interpret the t-stats for the various quantiles in the same way I would for a normal linear regression. For each quantile I run 10k quantile regressions with different combinations of the IVs. At the extremes I find multiple combinations of IVs that are significant at the 5%-level(t-stat), while at the middle there are very few IVs that are significant. Is this due to may data or how the t-stats are evaluated at the quantiles? $\endgroup$ – NTH.RA Apr 18 '13 at 11:53

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.