3

I'm wondering how to correct for the presence of autocorrelation/heteroscedasticity in estimating sample covariance matrices from data.

I know there have been some questions on this before:

However, these answers seem to be relevant when estimating regressions $y = Xb + a$. In my case I have no such regression - I only want to estimate the sample covariance $\sigma(X)$. However I know that my data $X$ is autocorrelated because I'm using overlapping data. Is there any literature on this?

Firebug
  • 15,262
  • 5
  • 60
  • 127
Michael
  • 2,180
  • 4
  • 23
  • 32
  • If you're not in a regression scenario then why is autocorrelation an issue? – Digio Sep 08 '17 at 14:00
  • 1
    @Digio, why would you think so? A naive estimate of the covariance under overlapping data will be biased, so that is a legitimate problem to be concerned about. – Richard Hardy Sep 08 '17 at 14:21
  • Maybe I have misunderstood the question but I can't think of ways of correcting autocorrelation without using a time series model. – Digio Sep 08 '17 at 15:07

0 Answers0