Testing Composite Hypotheses for Locally Stationary Processes

From MaRDI portal





For Dahlhaus' locally stationary processes the authors consider the problem of testing in a parametric form. A sequence of stochastic processes \(X_{t,T}\) \((t=1,\ldots,T)\) is called locally stationary with transfer function \(A_{t,T}(\lambda)\) if the following conditions hold. There exists a representation \(X_{t,T}=\int_{-\pi}^{\pi}\exp(i\lambda t)A_{t,T} (\lambda) \,d\xi(\lambda)\), where \(\xi(\lambda)\) is a stochastic process on \([-\pi,\pi]\) with \(\overline{\xi(\lambda)}= \xi(- \lambda)\), \(E\xi (\lambda)=0\) and \[ E[d\xi(\lambda) d\xi(\mu)]= \eta(\lambda+ \mu)\,d\lambda \,d\mu, \] where \(\eta(\lambda)= \sum_{j=- \infty}^{\infty}\delta(\lambda+2\pi j)\) is the period \(2\pi\) extension of the Dirac delta function. There exist a constant \(K\) and a \(2\pi\)-periodic function \(A(u,\lambda): [0,1] \times R\to C\) with \(A(u,-\lambda)=\overline{A(u,\lambda)}\) and \[ \sup\limits _{t,\lambda}| A_{t,T}(\lambda)- A(t/T,\lambda)|\leq KT^{-1} \] for all \(T\). The authors study testing a composite hypothesis for the process \(\{X_{t,T}\}\): \(H:\;\theta^{(2)}= \theta^{(2)}_0\) against \(A: \;\theta^{(2)} \neq\theta^{(2)} _0\), where \(\theta=(\theta ^{(1)}, \theta^ {(2)})\), \(\theta^ {(1)}=(\theta_{1},\ldots,\theta_{l})\), \(\theta ^{(2)}=(\theta _{l+1}, \ldots,\theta_{q})\) and \(\theta ^{(2)} _0\) is a specified vector. For the Gaussian likelihood ratio test, the Wald test and the Lagrange multiplier test the authors derive the limiting distributions under the composite hypothesis in parametric form. It is shown that the distribution of these test statistics tends to the \(\chi^2\) distribution under the hypothesis. The authors study the asymptotics of tests under a sequence of local alternatives and the local optimality of tests, is discussed. For a sequence of nonparametric spectral densities the authors show the local asymptotic normality and apply the result to the testing problem. The asymptotics for the log-likelihood ratio are examined by numerical integration.











This page was built for publication: Testing Composite Hypotheses for Locally Stationary Processes

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4455662)