I have $m$ weakly stationary observations $X_1,X_2,\cdots,X_m$ from a Markov chain. I want to estimate the variance of the mean. My idea was to use nonparametric bootstrapping to make $n$ bootstrap samples from the $m$ observations and compute $n$ bootstrap means $\mu^*_i$. Then estimate the variance using $$ \widehat{\mathrm{Var}(\overline X)} = \frac{1}{n-1}\sum_{i=1}^n ( \mu^*_i - \overline \mu^*)^2 $$ Is it a problem that the observations are correlated, or is this fine?

  • $\begingroup$ By "unparameterised bootstrap" do you mean nonparametric bootstrap or are you referring to something else? $\endgroup$
    – Glen_b
    Jul 9 '17 at 13:58

No you can't use standard bootstrap. There are many dependent bootstraps.

Jens-Peter Kreiss, Efstathios Paparoditis, Bootstrap methods for dependent data: A review, Journal of the Korean Statistical Society, Volume 40, Issue 4, 2011, Pages 357-378, ISSN 1226-3192, Keywords: Stochastic processes; Bootstrap methods; Time series

Here is a review , Here it is not behind paywall.


Your Answer

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

Not the answer you're looking for? Browse other questions tagged or ask your own question.