Long-range dependence in third order and bispectrum singularity (Q653800): Difference between revisions

The paper under review is to define a third order long-range dependence (LRD) for non-Gaussian time series in terms of the bispectrum and bicovariances. The bispectrum has been used to examine nonlinear interactions among various measurements, and to apply in oceanography, acoustics, seismics, plasma physics, economies and neurology. A stationary time series \(X_t,\;t=0, \pm 1, \pm 2, \dots\), is in third order if the covariance function of \(X_t\) and the third order cumulants \(\text{Cum}(X_{t+s_1}, X_{t+s_2}, X_t) = C_3(s_1, s_2)\) are invariant under the time shift. The bispectrum \(S_3\) is a complex valued integrable function on \([-1/2, 1/2]\times [-1/2, 1/2]\) with Fourier coefficients \(C_3(s_1, s_2)\). Similar to the long-range dependence (LRD) as the spectrum behaves like \(| \omega| ^{-{2h}}\) at zero, a time series \(X_t\) is long-range dependent in third order if \[ \lim_{a\to \infty} \frac{S_3(w_1/a, w_2/a)}{\tau (a)} = S_3^0(w_1, w_2) \in L_1, \] \[ \lim_{a\to \infty} \frac{C_3(as_1, as_2)}{\tau (a)/a^2}=C_3^0(s_1,s_2),\quad | C_3^0(s_1, s_2)| <\infty, \] where \(Y\) is the interior of the triangle with vertices \((0, 0),\;(1/2, 0)\) and \((1/3, 1/3)\), \((w_1, w_2)\in Y\), \(\tau (\cdot )\) is a regularly varying function of index \(3g_0\) and \(g_0\) is the radial parameter of the bispectrum. If \(0< g_0< 2/3\), the bispectrum is singular at zero and the decay of bicovariances is hyperbolic. An example of fractional noise processes is given in the Section 3. Under the assumption that the third order cumulant \(\text{Cum}(Z_0, Z_0, Z_0)\) exists and is nonzero, the author shows that the third order cumulant \(C_3(s_1, s_2)\) is given by a hypergeometric series. Hence the bispectrum can be calculated by the transfer function of a linear series. Another example is the Rosenblatt process which is the subject of a non-central limit theorem.