When its autocovariance is absolutely summable, a weakly stationary process has spectral density . The inverse relation is . A linear filter multiplies this density by the squared modulus of its frequency response.
Under angular-frequency normalization, the displayed time-series spectral density gives and for nonzero integer . This follows from the even integral . All nonzero even-lag covariances vanish while the odd-lag covariances are positive.
For two real centered time series , require joint weak stationarity: all auto- and cross-covariances depend only on the time lag. Individual weak stationarity alone does not impose this condition on their joint behavior. With the convention , define the cross-spectrum byAssume absolute summability of the auto- and cross-covariance sequences, so the displayed Fourier series and the two individual spectral densities are continuous. Reality and joint weak stationarity giveConsequently the spectral matrix is a Hermitian matrix. It is positive semidefinite, because its quadratic forms are nonnegative spectral densities of complex linear combinations of the jointly stationary series. Its nonnegative determinant therefore yields .
The imaginary part records directional lag information. For white noise of variance and , only is nonzero. Hence , although both individual spectral densities equal . The cross-spectrum is generally complex, rather than real and even like a real series' own spectral density.
For real stationary processes, the one-sided density on is twice the usual two-sided density. Its integral is the variance and . When covariances are absolutely summable, .
For a real weakly stationary process, restricting its even angular-frequency density to gives . Folding both frequency halves into one density gives and removes that factor two. Consequently white noise of variance has restricted density and folded density .
A weakly stationary process has a time-series spectral density precisely when its spectral measure of a stationary time series is absolutely continuous with respect to Lebesgue measure. Absolute summability of its autocovariances is sufficient and yields a continuous Fourier-series density. It is not necessary for existence; for a general integrable density, Fejér sums give an recovery formula.
In cycles per observation, frequency lies in and . Compared with angular-frequency density , one has . In particular, white noise of variance has .
A centered weakly stationary process has a representation with orthogonal random increments. Their variance measure is the spectral measure. When it has density , the autocovariance is .
For a centered record of length , the periodogram is the squared modulus of its discrete Fourier transform, with normalizationUnder suitable short-memory assumptions its expectation approaches the spectral density of a stationary process, but its variance generally does not vanish. Smoothing nearby frequencies can provide an estimate with statistical consistency.
Let be a centered real stationary Gaussian time series with absolutely summable autocovariances and two-sided angular-frequency spectral density . The parent periodogram uses . At an interior positive frequency define the one-sided quantitiesThus the doubling applies to both the statistic and its target spectral density.
For Fourier-grid frequencies with , setThen . Summability makes their variances tend to and their covariance tend to zero. Joint Gaussianity therefore givesUnder these Gaussian assumptions the fourth moments also converge, so . An unsmoothed periodogram is therefore not a consistent estimator of a positive spectral density. At zero and the Nyquist frequency , the sine coordinate vanishes and the real-only calibration differs; the interior doubling and two-coordinate argument do not apply there.[ref]
Introduction to Time Series Analysis, Lecture 20 by Peter Bartlett. Slides 2–9; converted from cycles per observation to angular frequency
For a centered stationary real Gaussian sequence, let and be its cosine and sine coefficients with normalization at a nonzero Fourier frequency. Grouping the double covariance sum by lag and extending each lag sum to a whole record leaves at most omitted bounded products. The full cosine-sine sum is zero, giving the bound. If , the two coefficients become asymptotically uncorrelated, and their limiting joint normal distribution consequently has independent components.
Average neighboring periodogram ordinates with nonnegative weights summing to one. Let the number of averaged ordinates tend to infinity while their frequency span, of order , tends to zero. For a smooth spectral density and suitable short-memory or fourth-order dependence conditions, the bias tends to zero and variance is of order . Smooth second-order spectral density alone does not ensure this: a time-constant random scale multiplying independent white noise has a smooth unconditional spectrum but a nonergodic random limiting local power.
Articles by others on the same topic
There are currently no matching articles.