我个人理解应该是解释变量矩阵为宽平稳过程,也称为二阶矩过程,一般在大样本理论下用到,不知你看到的是不是在大样本下关于解释变量的假设条件满足这条要求.

If F(x) is a cumulative probability distribution function of any probability distribution,
then the nth moment of the probability distribution is given by the following integral

    E(x^n) = integal  x^n * dF(x) from negative infinty to positive infinity,

when n=2, it is second order moment.

if x is center at 0, then it is the variance in an univariate distribution.  Likewise in such a case, the second order moment matrix is covariance under multvariates distribution.

Often in ML under certain regulities, the estimation covariance matrix is equal to inverse of negative  hessian matrix.

HTH