ATesting for the Onset of Trend, Using Wavelets

From MaRDI portal





The author considers the model: \[ Y_t=\mu(t)+u_t,\qquad t=1,\dots,T, \] where \(\mu (t)\) is a constant for a while, followed by a trend: \[ \mu(t)=\begin{cases} \mu_0,& \text{if \(t<\tau_0\)}\\ f(t/T),& \text{if \(t\geq\tau_0\)},\end{cases} \] \(\mu_0\) and \(\tau_0\;(0<\tau_0<1)\) are unknown constants. A stochastic linear process \(u_t\) is of the form \[ u_t=\sum_{j=0}^{\infty}a_j\varepsilon_{t-j} \] with \((a_i)\) a sequence of constants and \((\varepsilon_t)\) a sequence of i.i.d. random variables. The author tests the null hypothesis \[ H_0:\;f(z)=\mu_0 \text{ for all } z\geq\tau_0\quad \text{versus}\quad H_0: f(z)\not = \mu_0 \text{ for some } z\geq \tau_0. \] The test is based on some \(F\)-type statistic. The nonstandard distribution of the test statistic is tabulated. Properties of the test are studied and a simulation study is given.











This page was built for publication: ATesting for the Onset of Trend, Using Wavelets

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4939813)