The Cross Covariogram#
Suppose we have two wide-sense stationary stochastic processes \(y_t\) and \(x_t\). The processes are said to be jointly wide-sense stationary if the cross covariance \(E(y_t - Ey_t)(x_{t-k} - Ex_{t-k})\) depends only on \(k\) and not on \(t\). The cross covariogram is this list of covariances as a function of \(k\):
Now suppose \(y_t\) and \(x_t\) can be expressed as (possibly two-sided) distributed lags of a single white-noise process \(\epsilon_t\):
where
Since \(E\epsilon_t = 0\),
The cross-covariance generating function \(g_{yx}(z)\) is defined by
In the present case, we have
Letting \(h = j-k\) so that \(k = j-h\), we have
The last equation gives
This is a counterpart to (159) and includes it as a special case.
A More General Bivariate System#
Suppose instead we have the system
where \(\epsilon_t\) and \(u_t\) are two mutually uncorrelated white-noise processes with variances \(\sigma_\epsilon^2\) and \(\sigma_u^2\), and \(Eu_t\epsilon_{t-k} = 0\) for all \(k\). By calculations analogous to those above, the cross-covariance generating function is
The representation (171) is in fact very general—it includes all jointly wide-sense stationary, indeterministic bivariate processes.[1]
Symmetry Relations#
We also define the cross-covariance going the other way:
Correspondingly, we have
where the second equality follows by substituting \(k = -h\) and using \(c_{xy}(h) = c_{yx}(-h)\).
The particular system (171) implies the symmetric counterpart: