1: \begin{abstract}

2: We investigate the accuracy of two general non-parametric methods for

3: estimating optimal block lengths for block bootstraps with time series --

4: the first proposed

5: in the seminal paper of Hall, Horowitz and Jing

6: (\textit{Biometrika} \textbf{82} (1995) 561--574) and the second from

7: Lahiri \textit{et al.}

8: (\textit{Stat. Methodol.} \textbf{4} (2007) 292--321).

9: The relative performances of these general methods

10: have been unknown and, to provide a comparison,

11: we focus on rates of convergence for these

12: block length selectors

13: for the moving

14: block bootstrap (MBB) with variance estimation problems

15: under the smooth function model.

16: It is shown that,

17: with suitable choice of tuning parameters,

18: the optimal convergence rate of

19: the first method is $O_p(n^{-1/6})$

20: %as $n\rightarrow\infty$,

21: where $n$ denotes the sample size.

22: The optimal convergence rate of the second method,

23: with

24: the same number of tuning parameters,

25: is shown to be $O_p(n^{-2/7})$, suggesting that the second method

26: may generally have better large-sample properties for block selection

27: in block bootstrap applications beyond variance estimation. We also

28: compare the two general methods

29: with other plug-in methods specifically designed for block selection in

30: variance estimation,

31: where the best possible convergence rate is shown to be $O_p(n^{-1/3})$

32: and achieved by

33: a method from Politis and White

34: (\textit{Econometric Rev.} \textbf{23} (2004) 53--70).

35: %A key result is a uniform approximation for the

36: %mean square error function of the MBB variance estimators

37: %over a range of block sizes that may be of independent

38: %interest.

39: \end{abstract}