Speed up Block Stationary Bootstrap

Hello,

I am currently working with Kevin Sheppard's block bootstrap code for my project, but I've noticed that it becomes slow for bootstraps exceeding 10^6. I'm looking for suggestions on how to optimize and speed up the code, possibly through parallelization or other techniques. Your expertise and insights are highly appreciated. Thank you!

https://www.kevinsheppard.com/files/teaching/mfe/advanced-econometrics/AFE_Week_1_markup.pdf

Slide 34.

n = 300; y = randn(n,1);
T=size(y,1);
% Average block size
m = 10;
%Nbootstraps
B = 10^6;

% Loop over B bootstraps
yRepl = [y;y];
u = zeros(T,1);
tic
for b=1:B
    u(1) = ceil(T*rand);
    for t=2:T
        if rand<1/m
            u(t) = ceil(T*rand);
        else
            u(t) = u(t-1) + 1;
        end
    end
    % y-star sample simulation
    yStar = yRepl(u);
    % Mean of y-star
    muStar(b) = mean(yStar);
end
toc

I've tried a parfor-loop but I am getting the following error on u.

Unable to classify the variable 'u' in the body of the parfor-loop. For more information, see Parallel for Loops in MATLAB, "Solve Variable Classification Issues in parfor-Loops". 

Answer 1

[Keshav]

Hi,

You can look at the below MathWorks documentation to resolve the error you are getting while using the ‘parfor-loop’.

https://in.mathworks.com/help/parallel-computing/troubleshoot-variables-in-parfor-loops.html#bvjfik_