“…In view of (8.4) and jointly with (8.8), these inequalities, which will be useful also later on, prove our point concerning C~)~),zr and C(k{)~),r if we make # = f. The variance of C(~{)~),~ v can be "computed in the same way as that of C~(~),iv, but with the substitution of the second and fourth eumulants of ~t, r for those of st, so that lim N var C(kd~ = 0 and, therefore, ,dt) of a degree having a given upper bound is superimposed on a process {xt} satisfying the assumptions of the preceding theorem, the trend can be eliminated from a sample by the method of least squares, and the usual asymptotic formula for the covariances of the covariance estimators still applies to the estimators based on the residuals (cf. [13]; the whole paper was written on the assumption that E(et s) was finite, but the proofs of the propositions referred to remain valid when only E(e~) is assumed to be finite). The estimator of R k based on a sample of x t q-]~(t) can be decomposed into two parts: one formed with the "true" process {x~}, and another, denoted by X k and representing the errors in trend elimination.…”