Nonlinear wavelet estimation of time-varying autoregressive processes (Q1962757): Difference between revisions

From MaRDI portal
RedirectionBot (talk | contribs)
Changed an Item
Set OpenAlex properties.
 
(One intermediate revision by one other user not shown)
Property / MaRDI profile type
 
Property / MaRDI profile type: MaRDI publication profile / rank
 
Normal rank
Property / OpenAlex ID
 
Property / OpenAlex ID: W3125392599 / rank
 
Normal rank

Latest revision as of 19:04, 19 March 2024

scientific article
Language Label Description Also known as
English
Nonlinear wavelet estimation of time-varying autoregressive processes
scientific article

    Statements

    Nonlinear wavelet estimation of time-varying autoregressive processes (English)
    0 references
    0 references
    0 references
    0 references
    0 references
    29 January 2001
    0 references
    Let \(\{X_1,\dots,X_T\}\) be a zero mean AR\((p)\) process with time-varying coefficients \(a_1(.),\dots,a_p(.)\). The functions \(a_i(.)\) are defined on \([0,1]\) and the considered model is \[ X_{t,T}+\sum_{i=1}^p a_i (t/T) X_{t-i,T} = \sigma(t/T)\varepsilon_t, \] where \(\varepsilon_t\) are i.i.d.~\((0,1)\) variables. Define \(\phi_{l, k}=2^{l/2}\phi(2^l x-k)\), \(\psi_{jk}=2^{j/2}\psi(2^j x-k)\), where \(\phi\) and \(\psi\) are a given scaling function and a given wavelet, respectively. Assume that \(\{\phi_{lk}(.)\}_{k\in\mathbb Z} \cup \{\psi_{jk}(.)\}_{j\geq l;k\in \mathbb Z}\) forms an orthonormal basis of \(L_2(\mathbb R)\). Then \(a_i\) can be expanded in an orthogonal series \[ a_i=\sum_k a_{lk}^{(i)} \phi_{lk}+\sum_{j\geq l}\sum_k \beta_{jk}^{(i)} \psi_{jk}. \] The empirical wavelet coefficients are derived using a least-squares method. Nonlinear thresholding is applied to the empirical wavelet coefficients \(\beta_{jk}^{(i)}\). Hard thresholding is given by \(\delta^{(\text{h})} (\tilde{\beta}_{jk}^{(i)}, \lambda)= \tilde{\beta}_{jk}^{(i)} I(|\tilde{\beta}_{jk}^{(i)}|\geq \lambda)\) and soft thresholding is defined by \(\delta^{(\text{s})} (\tilde{\beta}_{jk}^{(i)}, \lambda)=\) \(=(|\tilde{\beta}_{jk}^{(i)}|-\lambda)_{+}\text{sgn}(\tilde{\beta}_{jk}^{(i)})\). The resulting estimators attain the usual minimax \(L_2\) rates up to a logarithmic factor. Finite-sample properties of the estimators are demonstrated on two simulated examples.
    0 references
    nonlinear thresholding
    0 references
    non-stationary processes
    0 references
    time-varying autoregression
    0 references
    wavelet estimators
    0 references
    0 references

    Identifiers

    0 references
    0 references
    0 references
    0 references
    0 references
    0 references