Rolling Correlation Problem

Hello Masters of R,

I was wondering if somebody can help with a problem I am facing. I have a
zoo object consisting of 500 vectors representing constituents of a
benchmark. I want to calculate the rolling correlation on 22 window period.
The observations are daily return series. However, when I run  my code 

x22=rollapply(spx,width = 22, FUN = function(x) {y <- cor(x);
 round(mean(y[lower.tri(y)]),3); }, by.column = FALSE, align='right')

I get warning message such as the underneath one due to too many 0000s in
some constituents. 

Warning messages:
1: In cor(x) : the standard deviation is zero

Do you know how to calculate the rolling correlations in the presence of too
many zeros?