xcov sample covariance / dividing by N?

Dear matlab community,

I looked into using xcov to calculate auto-covariances of a Tx1 time series vector. Using xcov in the first instance, it seemed as if this function would not devide by T as opposed to what the general formula for sample autocovariance would suggest: ?

Can someone confirm whether I need to devide by T to get a vector of sample auto-covariances? If I do not devide by T, the covariances seem to explode, as more and more terms are added. It is also not clear from the documentation of xcov.

Best

Marcel

Answer 1

[Pratyush Roy]

Hi,

xcov in MATLAB computes raw covariances with no normalization. Hence after obtaining the results, one can divide by T to obtain the results suggested by the formula in the question.

The "More About" section in this documentation page might be helpful.

Hope this helps!