On Win32 platform when input data vector is sufficiently long there is an issue with final calculation of autocovariances. The proposed patch fixes it.
A patch to ql/experimental/math/autocovariance.hpp
The patch was applied to the code repository.
It will be included in next release.